Recent Advances in LoRa: A Comprehensive Survey

ZEHUA SUN, HUANQI YANG, KAI LIU, ZHIMENG YIN, ZHENJIANG LI, and WEITAO XU∗ , City
University of Hong Kong Shenzhen Research Institute, City University of Hong Kong, China
The vast demand for diverse applications raises new networking challenges, which have encouraged the development of a
new paradigm of Internet of Things (IoT), e.g., LoRa. LoRa is a proprietary spread spectrum modulation technique, which
provides a solution for long-range and ultra-low power-consumption transmission. Due to promising prospects of LoRa,
signiicant efort has been made on this compelling technology since its emergence. In this paper, we provide a comprehensive
survey of LoRa from a systematic perspective: LoRa analysis, communication, security, and its enabled applications. First,
we summarize works focusing on analyzing the performance of LoRa networks. Then, we review studies enhancing the
performance of LoRa networks in communication. Afterward, we analyze the security vulnerabilities and countermeasures.
Finally, we survey the various LoRa-enabled applications. We also present comparisons of existing methods, together with
insightful observations and inspiring future research directions.
CCS Concepts: · Networks; · General and reference → Surveys and overviews;
Additional Key Words and Phrases: LoRa, Analysis, Communication, Security, Application

1 INTRODUCTION
The rapid growth of IoT in past decades has witnessed the explosion of applications in a wide range of ields with
respect to smart city [176], industry [34], agriculture [19], etc. IoT thrives on a variety of wireless communication
technologies, such as short-range wireless standards (e.g., Zigbee, Bluetooth) and cellular technologies (e.g.,
4G, 5G). However, in the face of the future demand for tens of billions of IoT access, these legacy wireless
technologies are limited by communication range and energy consumption. In this context, such long-range
and energy-eicient communication demands have inspired the emergence of Low Power Wide Area Networks
(LPWANs) as a new IoT new paradigm, which ills the gap of legacy wireless communication technologies (see
Figure 1). Among which, LoRa, due to its open-source privilege (operating in the unlicensed sub-GHz ISM band)
and low-cost Commercial Of-The-Shelf (COTS) devices compared with other LPWAN technologies such as
NB-IoT and SigFox, shows great potential in industry and research communities recently.
LoRa is a proprietary spread spectrum modulation technique on the basis of Chirp Spread Spectrum (CSS),
which is resilient and robust against interference and noise. Such modulation technique and a high sensitivity
ofered by LoRa, enable receiving the potential weak signals at extremely low energy consumption, which
provides signiicant link budget improvement to support a wide coverage [94]. LoRaWAN is a data link layer
speciication built on top of LoRa, deining the typical stat-topology network architecture and its bi-directional
communication protocol. To date, LoRa networks have been deployed by 163 LoRaWAN network operators
∗ Weitao Xu is the corresponding author.

Authors’ address: Zehua Sun, zehua.sun@my.cityu.edu.hk; Huanqi Yang, huanqi.yang@my.cityu.edu.hk; Kai Liu, kailiu@cityu.edu.hk;
Zhimeng Yin, zhimeyin@cityu.edu.hk; Zhenjiang Li, zhenjiang.li@cityu.edu.hk; Weitao Xu, weitaoxu@cityu.edu.hk, City University of Hong
Kong Shenzhen Research Institute, City University of Hong Kong, China.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that
copies are not made or distributed for proit or commercial advantage and that copies bear this notice and the full citation on the irst
page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy
otherwise, or republish, to post on servers or to redistribute to lists, requires prior speciic permission and/or a fee. Request permissions from
permissions@acm.org.
© 2022 Association for Computing Machinery.
1550-4859/2022/8-ART111 $15.00
https://doi.org/10.1145/3543856

ACM Trans. Sensor Netw.

111:2 • Sun et al.

Fig. 1. Comparison with legacy wireless

communication technologies. Fig. 2. Number of papers with years through Google Scholar hits for
łLoRaž and łLPWANž keywords.

across 177 countries globally [76], widely distributed in various applications scenarios that require large-scale
and delay-insensitive deployment [19, 21, 34, 176]. In the research community, an intuitive fact is that nearly
1,000 papers involving LoRa in their contents were published during 2021 through Google Scholar hits for the
łLoRaž keyword (see Figure 2)1 .
Due to large-scale deployments and promising prospects of LoRa, extensive works on LoRa have been pre-
sented since its appearance. Accordingly, such fact motivated several survey papers [56, 85] on LoRa in the
recent six years. In particular, some early surveys [22, 62, 146] gave an overview of LoRa with an emphasis of
preliminary background and principle introduction, which laid a foundation for follow-up research. A small
portion targeted speciic areas such as testbed [103], simulator [32], security [114], and mesh topology [29].
Recently, several comprehensive surveys of LoRa [56, 85] were proposed. For example, Gkotsiopoulos et al. [56]
focused on the network capacity from ive main aspects encompassing PHY layer characteristics, deployment
and hardware features, transmission settings, MAC protocols, and application requirements. Li and Cao [85]
gave a comprehensive and structured survey of LoRa from a two-dimensional taxonomy: networking layers (i.e.,
PHY, MAC, Link, and Application layer) and performance metrics (i.e., range, throughput, energy, and security).
However, many LoRa methods typically do not show a clear boundary on the networking layer level but span
multiple ones.
In comparison with prior works (see Table 1), our survey presents two novel contributions. First, our survey
covers various eforts made to LoRa comprehensively and up to date, which complements the previously published
ones. Second, our survey summarizes and compares LoRa works from a new perspective. Figure 3 gives a taxonomy
of our survey. Speciically, our paper provides a comprehensive survey on LoRa from four-fold: LoRa analysis
works and tools, LoRa communication studies in terms of Physical (PHY) and Media Access Control (MAC) layer,
LoRa security vulnerabilities and countermeasures, and LoRa-enabled applications. The rationale behind the
organization is that since the advent of LoRa, early-stage studies focus on understanding and analyzing LoRa
performance through various ield studies or simulation tools. Afterward, as a networking technology, a large
quantity of research eforts have been made to improve the performance of LoRa networks in communication.
Meanwhile, with massive deployments of LoRa networks, security is receiving much attention. As LoRa tends to
mature in a variety of common IoT applications, many works contribute ones beyond the scope of LoRa radio.
Thus, such taxonomy provides a good it for the cognition of the evolution of LoRa technology, which provides
1 Dueto the huge number of references about LoRa, this survey only focuses on the papers published at top conferences or journals, coupled
with highly inluential ones.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:3

Table 1. Summary and comparison with prior LoRa surveys/reviews (❍śnone, ◗śmoderate, ●ścomprehensive; Perf. Meas.:
Performance Measurement, Anal.: Analytical, Config.: Configuration Seting, V. & S.: Vulnerabilities and Countermeasures,
PLS: Physical Layer Security).
LoRa Analysis LoRa Communication LoRa Security LoRa-Enabled Applications
Survey Year
Perf. Meas. Anal. Models Simulators Testbeds Modem MAC Protocol Conig. V. & C. PLS Backscatter Sensing WCE & CTC Others
[22] 2016 ◗ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍
[103] 2017 ● ❍ ◗ ● ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍
[62] 2018 ● ● ● ◗ ❍ ◗ ◗ ◗ ❍ ❍ ❍ ❍ ●
[146] 2019 ● ◗ ◗ ◗ ◗ ● ● ● ❍ ● ❍ ❍ ●
[114] 2020 ❍ ❍ ◗ ❍ ❍ ❍ ❍ ● ❍ ❍ ❍ ❍ ●
[81] 2020 ❍ ● ● ◗ ❍ ◗ ● ❍ ❍ ❍ ❍ ❍ ❍
[29] 2020 ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ◗
[32] 2021 ◗ ❍ ● ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍ ❍
[56] 2021 ● ◗ ● ❍ ❍ ● ● ❍ ❍ ❍ ❍ ❍ ●
[85] 2022 ● ● ❍ ◗ ● ● ● ◗ ❍ ● ● ◗ ●
Ours 2022 ● ● ● ● ● ● ● ● ● ● ● ● ●

an aggregation of LoRa methodologies in diferent aspects and a new perspective for the research community. In
particular, the four parts are speciied as follows:
• LoRa Analysis. LoRa performance analysis works aim to investigate and interpret LoRa network perfor-
mance in various environments, which can also serve the further exploration in various LoRa communica-
tion, security, and its enabled applications works. Speciically, these works include early-stage performance
measurements, and three types of conducted tools: analytical models, simulators, and testbeds.
• LoRa Communication. Many eforts have been made to target improving the performance of LoRa
networks in communication, in terms of throughput, communication range, scalability, and energy con-
sumption. These works give a focus on LoRa PHY and MAC layer, which can be divided into three categories:
LoRa (de)modulation techniques, MAC protocols, and coniguration settings.
• LoRa Security. Security is fragile but critical for any wireless communication technologies, receiving
signiicant attention in vulnerabilities and countermeasures, coupled with PHY layer security methods.
• LoRa-Enabled Applications. The wide deployment of LoRa networks has inspired a wide range of
applications, including backscatter, sensing, integration with heterogeneous wireless technologies, and
other applications.
Based on that, comparisons of existing LoRa works (e.g., in Tables 6, 8, 11, 12, 13), with brief summaries and
insightful discussions, are also given.
The remainder of the survey is organized as follows. LoRa preliminary, inclusive of its PHY and MAC layer, is
introduced in Section 2. LoRa analysis works and diferent types of tools are reviewed in Section 3. Various LoRa
communication studies are surveyed and compared in Section 4. LoRa security vulnerabilities and countermeasures
are discussed in Section 5. LoRa-enabled applications in diferent ields are reviewed in Section 6. The challenges
and potential future development of LoRa are discussed in Section 7, followed by the Conclusion in Section 8.

2 LORA PRELIMINARY
Although LoRa has been extensively introduced in numerous papers, to make this paper self-contained, we
provide a brief preliminary of LoRa, together with LoRa PHY layer and LoRaWAN MAC layer.

2.1 LoRa Overview

LoRa, standing for łLong Rangež, is a proprietary spread spectrum modulation technique, while LoRaWAN is
a data link layer speciication built on top of LoRa. LoRa operates in the unlicensed sub-GHz ISM radio band,
depending on the deployed regions (e.g., 863ś870 MHz in Europe, 902ś928 MHz in the USA), but also obliges
to the duty cycle regulations (e.g., 1% in Europe). Compared with short-range wireless standards and cellular

ACM Trans. Sensor Netw.

111:4 • Sun et al.

Fig. 3. Taxonomy of this survey.

technologies, LoRa shows remarkable results in long-range transmission and ultra-low power consumption.
Speciically, its coverage range is up to 15 km (rural areas) and 5 km (urban areas), its device battery life is
up to 10 years, and its data rate ranges from 0.3 to 37.5 Kbps [22]. Some other properties, such as low cost of
devices, concurrent reception capacity of gateways, and the resiliency of modulation attribute against fading,
multipath, and Doppler efect compared to other wireless signals [89], also make LoRa compelling. It is noted
that real-life deployed LoRa networks typically cannot obtain the optimal performance that LoRa promises, due
to the deployment complexity and various interference.

2.2 LoRa PHY Layer

LoRa PHY layer adopts a proprietary spread spectrum modulation technique derived from CSS with an integrated
Forward Error Correction (FEC) mechanism [128], which ofers substantial processing gains for link budget
improvements and resiliency to multipath and interference. Figure 4 shows a LoRa packet structure, which
consists of a preamble with 8 base up-chirps, a synchronization word with 2 base up-chirps, a Start of Frame
Delimiter (SFD) with 2.25 base down-chirps, and payload data followed by Cyclic Redundancy Check (CRC).
In LoRa modulation, the frequency of chirps constantly varies linearly with time within the pre-deined
available narrow bandwidth. Speciically, a base up-chirp is deined as one that linear sweeps from its initial
k BW
frequency −BW /2 to BW /2, represented as C (t ) = e j2π ( 2 t − 2 ) t . LoRa modulation is performed by representing
eachbit of payload
 by multiple chirps with a shifted initial frequency fsym , thus the signal of a symbol is denoted
as S t, fsym = C (t ) · e j2π fsym t [68].
As for the demodulation, the receiver performs the łde-chirpž operation where  each received symbol is
multiplied with a base down-chirp (i.e., the conjugate of a base up-chirp), denoted as S t, fsym ·C −1 (t ) = e j2π fsym t
[68], resulting in a single frequency tone. Then the receiver applies the Fast Fourier Transform (FFT) on the
multiplication result, where the resulted one FFT peak indicates the initial frequency fsym .
The coniguration for a LoRa device is mainly speciied by such parameters: Spreading Factor (SF), BandWidth
(BW), Code Rate (CR), Carrier Frequency (CF), and Transmission Power (TP). Speciically, SF, ranging from 7 to
12, represents the number of symbols per bit of payload. A higher applied SF incurs a lower data rate, longer
propagation time, and more energy consumption, but provides better sensitivity and a wider transmission distance.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:5

Hence, devices located farther from the gateway require a higher SF due to the more link budget need, which
provides increased processing gain but at the cost of a lower data rate. Besides, SFs are orthogonal (essentially
quasi-orthogonal [128]), which allows transmission of signals modulated with diferent SF in the same channel.
The BW is typically 125 kHz, 250 kHz, and 500 kHz. And the CR, denoting the rate of the FEC code, can be set
to 4/5, 4/6, 4/7, or 4/8, where a higher one ofers more protection at the cost of increasing the air time. Then,
the LoRa modulation bit rate Rb can be calculated through Rb = SF × BW 2S F
× CR. Besides, LoRa band in distinct
regions deines diferent multiple frequency channels, and CF is the center frequency of these channels. TP on a
LoRa radio is restricted by the speciic hardware, generally ranging from 2 to 20 dBm [14]. Tuning the parameters
above can achieve a trade-of between communication range, data rate, and power consumption, which inspires
plenty of coniguration setting methods (see Section 4.3).

Fig. 4. LoRa PHY packet structure.

Fig. 5. LoRa network architecture.

2.3 LoRaWAN MAC Layer

As LoRa only deines the lower PHY layer in the communication stack, resulting in several available upper
network protocols, among which LoRaWAN is the most popular networking protocol. It is noted that LoRaWAN is
extensively regarded as a MAC protocol for LoRa; however, the LoRaWAN stack operates at the data link layer and
does not deine any channel access control scheme. LoRaWAN protocol mainly deines LoRa network architecture
and its bi-directional communication protocol. As shown in Figure 5, the LoRaWAN network architecture is a
typical star topology, mainly including end devices, gateways, network servers, and application servers. End
devices (also called nodes, clients, points) equipped with various sensors are placed in the scenarios as data
sources. Gateways (also called base stations) are the bridge between end devices and network servers. Network
servers are the LoRa network controller taking charge of the management of devices and applications, while
application servers provide the operation interface with users. Additionally, LoRaWAN deines three types of
operation classes for end devices, i.e., class A(ll), B(eacon), and C(ontinue). Class A is the default one that all
devices must implement, while B and C are the optional ones. Class A, where two receive downlink windows
are created shortly after an uplink transmission from devices, is the most energy-eicient among these three
ones. Apart from two receive slots, end devices add scheduled downlink ping slots in class B, triggered by
synchronization beacons broadcasting from gateways. Class C, where end devices continuously keep listening
when unless are transmitting (i.e., half-duplex), thereby is the most energy-consuming one. Choosing the optimal
operation class can reduce energy consumption and meet the response time requirements of diferent application
scenarios.
Additionally, LoRaWAN provides media access, Adaptive Data Rate (ADR), and security services. LoRaWAN
MAC layer employs a default primitive ALOHA MAC protocol [49] which allows end devices to transmit as soon
as they wake up, without channel detection and time synchronization. Besides, ADR is a vital mechanism of
LoRaWAN, allowing end devices to be conigured with diferent data rates dynamically according to network
conditions. However, LoRaWAN does not give a deinition of the ADR algorithm, hence resulting in many studies
[81]. LoRaWAN provides security functions by relying on Advanced Encryption Standard (AES) algorithms and
two 128-bit unique session keys (i.e., NwkSKey and AppSKey) to complete data encryption, message integrity

ACM Trans. Sensor Netw.

111:6 • Sun et al.

checking, and node authentication. The session keys can be obtained through two activation processes, i.e.,
Activation By Personalization (ABP) and Over-The-Air Activation (OTAA).

3 LORA ANALYSIS
Since the advent of LoRa technology, various LoRa performance analysis works have been ongoing, aiming
to investigate and interpret the performance of LoRa networks in terms of throughput, communication range,
scalability, and energy consumption. Speciically, these works include early-stage performance measurements, and
three types of conducted tools (i.e., analytical models, simulators, and testbeds). Analytical models are designed
to give a mathematical explanation process for some speciic tasks, such as link and energy analysis. Simulators
are popular in theoretical method evaluation due to their convenience and low cost. Testbeds are utilized for
the performance evaluation of the real LoRa network under diferent scenarios, to explore its capabilities and
limitations to provide a benchmark standard. To this end, we irst give a summary of performance measurement
works, then review corresponding analysis tools in this section.

3.1 Performance Measurements

Some early performance measurement works conduct testbed experiments, perform simulations, build analytical
models, or adopt their collaborations to explore LoRa-related performances in terms of LoRa radio [8], link
condition [44], node energy consumption [89], network communication range [120], scalability [108], etc. It is
noted that analytical models are generally inevitable parts serving for testbed experiments and simulations, but
testbed measurement is the most favored choice owing to its real and objective experimental results. Table 2
summarizes the classical performance measurement works, along with their corresponding conclusions.

Table 2. Summary of LoRa performance measurement works (PRR: Packet Reception Ratio, RSSI: Received Signal Strength
Indication, SIR: Signal-to-Interference Ratio, SNR: Signal-to-Noise Ratio).
Reference Year Experimental Setup Performance Metric & Conclusion
Modulation: resistance to interference
Augustin et al. [8] 2016 Testbed and simulator
Coverage: 2.8 km in an urban area
Bor et al. [15] 2016 LoRaSim simulator Coverage: 120 nodes in the area of 3.8 ha in a city scenario
1-node and Doppler robustness: getting worse when speed is greater than 40 km h−1
Petäjäjärvi et al. [120] 2017
1-gateway testbed Range: 30 km on water with PPR 62%, SF12, and TP14 dBm
RSSI: decreases from -80 (2 km) to −100 dBm (10 km)
Feltrin et al. [44] 2018 2-node outdoor testbed
Inter-SF: quasi-orthogonal, co-SF: SIR varies from 0.3 to 1.7 dB
Range: >10 km (LOS), <3 km (NLOS)
3-gateway and Node lifetime: 1.19ś4.54 years
Liando et al. [89] 2019
over 50-node testbed Multiple access: gateway capacity is 6,249 nodes with PRR 70%
LoRa is resilient against Doppler efect, etc.
10-node indoor testbed Large-scale fading: inluenced by many factors (e.g., materials, layout)
Xu et al. [177] 2019
(4 types buildings) Temporal fading: follows Rician distribution K-factors (12ś18 dB)
High temperature gradient across nodes in diferent deployments
Tian et al. [152] 2021 21-node outdoor testbed Temperature and RSS: high correlation
Impact of temperature is greater than weather conditions

Liando et al. [89] conducted various LoRa network performance measurements in a 3-gateway and over 50-node
testbed. Speciically, they revealed: 1) the communication range in the line-of-sight (LOS) scenario is 10 km,
while drops sharply to 2 km in obstacle blocking non-line-of-sight (NLOS) one under the settings of SF12 and
PRR 70%; 2) the predicted node lifetime under diferent coniguration settings ranges from 1.19 to 4.54 years; 3)
multiple access performances in terms of SF and single-channel capacity, where the gateway capacity is 6,249
nodes with PRR 70%. Besides, they provided some insights on the enhancement of parameter optimization, MAC

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:7

protocol, concurrent reception, and PHY layer based on their measurements. Xu et al. [177] investigated the
LoRaWAN network performance in 4 types of multi-loor buildings, including LoRa large-scale and temporal
fading characteristics, coverage, and energy consumption. They conclude that many factors, such as building
materials and layout, inluence the path loss greatly, and the temporal fading follows Rician distribution with
its K-factors falls between 12 and 18 dB. Tian et al. [152] released a large-scale dataset focusing on the network
and link-level performance in a 21-node outdoor LoRa network for over 4 months. Speciically, the data features
with three types of attributes, i.e., basic (time, SF, TP, etc.), connectivity and link quality (PRR, RSS, SNR), and
environmental (temperature measured by nodes, weather condition, etc.) attributes. Besides, they provided
evaluation scripts for data analysis and visualization.
LoRa preliminary performance measurement works aim to understand the capabilities and limitations of LoRa
networks, which have made a fundamental contribution to the follow-up research. Liando et al. [89] conducted
large-scale deployment and measurement to comprehensively investigate the performance of LoRa networks in
terms of range, energy, and capacity. Radio fading [177] and inter-SF transmission [44] also received considerable
attention. Additionally, public datasets [152] provide a novel and labor-saving solution for research. However, the
type of hardware and the diversity of deployment environments bring complexity to such measurement works.
Furthermore, quantifying LoRa performance with the corresponding factors is a crucial and popular research
trend.

3.2 Analytical Models

An analytical model, i.e., a mathematical model with closed-form solutions, quantitatively represents the interre-
lationship of a set of parameters to describe a speciic problem. A large number of methods pursue capturing
the complexity of real-world deployment of LoRa networks to derive an accurate, interpretative, and general
analytical model, before solving various research problems such as optimal parameter assignments [53]. The
existing analytical models mainly concentrate on the link and energy models, where network deployment, device
coniguration, and environmental factors are integrated into these models for LoRa link and energy analysis.
Table 3 summarizes current LoRa analytical models.

Table 3. Summary of LoRa analytical models (Para.: Parameter, Envir.: Environment).

Reference Year Model Features
Para. Multiple Channel Collision SF Quasi- Duty Envir.
Conig. Gateways Variation Probability orthogonality Cycle Factor
SNR-based link-outage
Georgiou and Raza [55] 2017 " % " " % " %
SF-based link-outage
Link Model

Throughput
Waret et al. [166] 2018 " % " " " " %
(bit-rate×success probability)
Mahmood et al. [102] 2018 SNR-based success probability " % " " " " %
Empirical measurement-based
Chall et al. [40] 2019 % % " % % % "
path loss
Demetri et al. [37] 2019 Learning-based path loss % " " % % % "
Liu et al. [94] 2021 Learning-based path loss % " " % % % "
Toro-Betancur et al. [157] 2021 Node-level delivery ratio " " " " " " %
Para. Multiple Inner Collision Multiple Duty Envir.
Conig. Chipsets Unit Probability Modes Cycle Factor
Current consumption, lifetime,
" % % " " " %
Energy Model

Casals et al. [20] 2017

and energy eiciency
Bouguera et al. [16] 2018 Node Energy Consumption " % " % " % %
Liando et al. [89] 2019 Node lifetime " " " " " " %
Battery-free LoRa node
Delgado et al. [36] 2020 " % " " " % "
Markov-based uplink and downlink
Finnegan et al. [46] 2020 Ambient RF energy harvesting " % " % " " "

ACM Trans. Sensor Netw.

111:8 • Sun et al.

Link Models. A large body of work aims to understand the channel quality by analyzing the link conditions
of LoRa networks, such as path loss [94], delivery ratio [157], and interference. In a single gateway scenario,
Georgiou and Raza [55] leveraged a stochastic geometry framework to study link-outage conditions concerned
with SNR and co-SF, and revealed that the network scalability is susceptible to the latter one owing to the
exponential degrading performance. Likewise, SNR threshold, co-, and inter-SF interference were studied in
[102, 166]. Several models [55, 166] only considered simple network topologies or make strict assumptions about
the spatial network distribution, resulting in poor generality. Toro-Betancur et al. [157] proposed a general
node-level delivery ratio model without any restrictions on network deployment or device coniguration, which
characterizes quasi-orthogonal transmission under many considerations such as capture efect, duty cycle,
multiple gateways, and channel variation.
Besides, several methods [12, 40] focus on the path loss modeling of LoRa networks. The path loss (also
called path attenuation) refers to the power attenuation during the transmission, varying along with diferent
land-covers on the path due to the radio relection, difraction, etc. Chall et al. [40] adapted the most widely
used free-space, log-distance, and multiwall-and-loor path loss models based on the empirical measurement
results to derive their models in indoor and outdoor environments, respectively. Remote sensing techniques are
utilized for land-covers analysis in [37, 94]. Demetri et al. [37] proposed an automated link quality estimation
method without on-site measurements, which can also be used for gateway deployment planning. Speciically,
they irst proposed a toolchain to recognize seven types of land-covers based on freely multispectral images
from satellites, and then presented an Okumura-Hata [107] model-based framework for expected received
power estimation. Rather than the physical path loss model [40], Liu et al. [94] proposed a deep learning-based
long-distance path loss estimation framework termed DeepLoRa, with an emphasis on the types and order of
land-covers. Speciically, DeepLoRa divides the link into an ordered sequence of micro links with equal length,
and utilizes remote sensing images to identify the detailed land-covers of each micro-link. Then, DeepLoRa
adopts a Bidirectional Long-Short-Term-Memory (Bi-LSTM) network to learn the path loss model.
Energy Models. Energy consumption of LoRa nodes to complete routine data collection and transmission
process, is a key indicator for constrained LoRa networks. Thus, several studies [20, 89] proposed models to
characterize the energy consumption of transmission or the nodes’ lifetime. Casals et al. [20] irst deined the
diferent states of the nodes from waking up to sleep in one transmission, then derived a series of energy models
under various considerations concerning data rate, collision, un- and acknowledged transmission. Speciically,
they modeled the node average current consumption of these diferent states, lifetime through the battery capacity
divided by the current, and energy eiciency of data delivery through the real divided by the expected one.
Bouguera et al. [16] deduced an energy model allowing for diferent settings of parameters and IoT scenarios, by
calculating and summing the energy consumption of inner sensor node elements in sleep and active operating
modes. Liando et al. [89] modeled the energy consumption to quantify the lifetime of LoRa nodes via the testbed
measurement across multiple chipsets. Speciically, they irst captured the energy proile of the microcontroller
unit (MCU) and LoRa transceiver using a monsoon power monitor under diferent parameter settings, then
calculated the node lifetime by multiplying the single transmission cycle time duration and the supporting
transmission cycle number of a speciic battery.
Besides, several models [36, 96] explored the possibility of energy harvesting of LoRa. Delgado et al. [36]
irst derived a battery-free LoRa node model involving energy harvesting system, circuit, and load models, then
proposed a Markov model to characterize the uplink packet delivery ratio (PDR) and probability of receiving
downlink packets, deined by parameters with respect to device coniguration, application behavior, etc. Finnegan
et al. [46] explored the boundaries of the feasibility of ambient Radio Frequency (RF) energy harvesting for
LoRa devices. Speciically, they irst recombined the energy model of nodes from the sensing, networking, data
processing, and other system tasks parts, then deduced the aggregated energy model by quantifying ambient RF

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:9

power level and integrating the impacts of harvesting components including rectenna, power management unit,
and storage device.
In general, a reliable analytical model integrates principle characterization and interrelationship of parameters
to relect the internal law, which signiicantly vital for further research in terms of quantitative analysis and
performance evaluation. For link models, network coniguration, SF quasi-orthogonality, and radio fading are
typically common considerations. Remote sensing techniques [37, 94] provide a solution for measurement-free
link modelling, which can be used for localization [91]. For energy models, the energy proiles under diferent
inner units, states, and parameter settings are captured. Additionally, energy harvesting models shed light on
some harvester designs [100]. However, strict assumptions, complex and long-term environmental variance will
limit the deduction of general models.

3.3 Simulators
A network simulator is a virtual tool to explore the system-level or link-level performance through the re-
production of communication interactions in the networks. It is typically composed of network coniguration
deinition (e.g., topology, parameter, and propagation model), event simulation (e.g., uplink and downlink), and
performance evaluation. Simulators are widely used in theoretical model evaluation when large-scale testbeds
are not deployed or inconvenient to operate. Therefore, several open-source LoRa simulators have been proposed
in recent years, especially used for LoRa network scalability and throughput evaluation. Table 4 presents current
popular simulators designed for LoRa networks, along with their included features.

Table 4. Summary of LoRa network simulators (Topol.: Topology).

Features
Simulator Year Envir.
Topol. & Para. Class C MAC Bi-directional Propagation SF Quasi- Duty Node’s Energy
Conig. Support Command Traic Model orthogonality Cycle Consumption
LoRaSim [15] 2016 SimPy " % % % " % % %
Magrin et al. [101] 2017 ns-3 " % " % " " " "
Abeele et al. [159] 2017 ns-3 " % " " " % " "
Croce et al. [30] 2018 Matlab " % % % " " % %
FLoRa [140] 2018 OMNeT++ " % " " " % " "
LoRaFREE [2] 2019 SimPy " % " " " " " "
Finnegan et al. [45] 2020 ns-3 " % " " " " " "
LoRaWANSim [106] 2021 Matlab " " " " " " " "

Bor et al. [15] proposed a simulator termed LoRaSim based on their derived communication behavior model
involving range and collision behavior considerations. Speciically, the range considers the coniguration settings
(i.e., received signal power, path loss, and sensitivity) to determine whether a packet is received or not, while the
collision behavior determines the decoding under diferent conditions (i.e., CF, SF, timing, power). Magrin et al.
[101] designed a ns-3 module for the LoRaWAN network performance simulation under the assumptions at both
the link and the system level. Speciically, the link model contains a link measurement model abstracting the efect
of such as propagation loss and fading, and a performance model determining transmission and interference. The
system model includes SF and CF allocation tasks. Abeele et al. [159] built a physical error model through the
baseband simulations over an Additive White Gaussian Noise (AWGN) channel, together with LoRaWAN MAC
protocol, for the proposed ns-3 simulator. Croce et al. [30] focused on the superposition of multiple LoRa radios
from diferent SFs. Besides, several proposed simulators were designed with a emphasis on speciic methods,
such as ADR [45, 140] and MAC protocol [2]. Marini et al. [106] proposed LoRaWANSim, a to-date simulator for
the sake of completeness, which characterizes the LoRaWAN network behavior with respect to PHY, MAC, and
network aspects.

ACM Trans. Sensor Netw.

111:10 • Sun et al.

In general, a simulator is an eicient, convenient, and low-cost tool to evaluate the performance of LoRa
networks. Many classic simulators have been proposed so far, including [15, 30, 101], and LoRaWANSim [106]
is the most comprehensive one to date. However, current simulators are not full-featured enough owing to the
focus on some speciic kinds of tasks. Hence, developing full-featured simulators still deserves further eforts.
In addition, the characteristics of the simulator indicate that it is generally suitable for the theoretical method
evaluation, so the testbed can make up for it with regard to the evaluation objectivity and reality.

3.4 Testbeds
A testbed is a real-life deployment platform containing complete network components, aiming at conducting
controllable and reproducible experiments to obtain actual results via real-life deployments. Apart from private
LoRa networks [49, 92], research communities have contributed to many public open-source testbeds for evaluation
works. Current public open-source testbeds primarily include testbeds platforms [54] and endpoint prototypes
[65]. Speciically, remotely accessible testbeds platforms typically contain a complete network architecture for
stand-lone instance spawning, including device management, programming, and web services. They enable
researchers for remote application test and development, which resolve the problem of high-cost, time-consuming,
and labor-consuming deployment and maintenance of a LoRaWAN testbed. Testbed endpoint prototypes provide
a lexible and easily deployable solution in terms of heterogeneous networking protocol, inconvenient network
(e.g., Ethernet, cellular), and power conditions. It is noted that both of them rely on a framework for device and
service proiles management. Table 5 lists current LoRa public testbed platforms and endpoint prototypes.

Table 5. Summary of LoRa public testbed platforms and endpoint prototypes.

Testbed Year Architecture Networking Node Case Study
Softwarte (user, back-end), Wi-Fi, LoRa, 2,728 IoT nodes, Impact of Wi-Fi traic visualizing,
FIT IoT-Lab [3] 2015
Hardware (open node, gateway, control node) BLE, etc. 117 mobile robots smart loor demonstration
LoRaWAN wtih user management, API, etc.,
Platform

OpenChirp [38] 2017 LoRa - -

LoRaBug hardware
Synchronous transmission protocol,
Testbed server (data services, web interface),
FlockLab 2 [158] 2020 Wi-Fi, LoRa, etc. 15 observers worklow testing,
observers, target devices
protocol debugging and analysis
Web IDE, online compilers, Wi-Fi, LoRa, Indoor environment monitoring,
LinkLab [54] 2020 156 IoT nodes
device management components Buletooth, Zigbee IoT education
WiSHFUL uniied interface,
Prototype

WiSH-WalT [97] 2018 LoRa - -

WalT single-board computer
Software radio, OTA programmer, 900 MHz, LoRa: modem, resource allocation, MAC
TinySDR [65] 2020 -
power management system 2.4 GHz BLE: transmission, latency
LoRaDisC protocol, Benchmarking protocol performance,
ChirpBox [151] 2021 LoRa -
target nodes, control nodes impact of temperature

Testbed Platforms. Dongare et al. [38] proposed an open-source LoRaWAN network testbed platform termed
OpenChirp, built upon LoRaWAN with the user management framework, application programming interfaces
(APIs), and core services. In particular, OpenChirp is currently powering a hosted campus network, enabling
users to spawn stand-alone instances with various services concerning device registration, data serialization and
storage, and web access. Gao et al. [54] proposed LinkLab, a scalable and heterogeneous testbed with support for
multi-user and multi-site remotely experimenting and web-based developing. Speciically, LinkLab is composed of
the WebIDE, online compilers featuring incremental compilation and multi-user caching, and device management
components with a uniied naming mechanism. Likewise, FIT IoT-Lab [3] and FlockLab 2 [158] large-scale testeds
also support heterogeneous wireless nodes including LoRa, but they possess a limited number of nodes and are
deployed mainly in indoor environments [151].
Testbed Endpoint Prototypes. Lone et al. [97] presented a framework for controllable and reproducible
LoRa testbeds termed WiSH-WalT, which consists of a WiSHFUL uniied interface for lexible radio and network

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:11

settings and a WalT single-board computer for easy device deployment and operation monitoring. Hessar et al.
[65] proposed the irst Software-Deined Radio (SDR) platform named TinySDR to enable large-scale deployment
and over-the-air programming. Speciically, TinySDR is composed of software radio, OTA programmer, and a
power management system, supporting various research of PHY/MAC protocols, IoT localization, backscatter
readers, etc. Besides, TinySDR outperforms existing SDR platforms in terms of low power consumption and
cost. The complexity of assembling the backbone infrastructure limits the development and availability of
existing LoRa outdoor testbeds. To this end, Tian et al. [151] proposed an infrastructure-less LoRa testbed termed
ChirpBox, which could be deployed in areas without the need of cellular or backbone infrastructure providing
communication and power conditions. Speciically, all LoRa target nodes are equipped with a daemon used for
orchestrating the node’s activities (e.g., execution of a test run, the collection of log trace) and a Firmware Under
Test (FUT) for running tests. Target nodes are connected via a multi-hop network to the control node that serves
as an interface with the user. Besides, they designed an eicient all-to-all multi-channel protocol based concurrent
transmission termed LoRaDisC to disseminate the FUT and test run conigurations.
In general, current public open-source LoRa testbeds provide a lexible and low-cost solution for researchers to
conduct controllable and reproducible experiments for real-life evaluation, presenting signiicantly important
to the research community. The existing testbed platforms [3, 38, 54, 158] all built a well-deined framework,
among which LinkLab [54] can support multi-user remote development featuring Web-based development and
incremental compilation. However, the existing testbed platforms only possess a limited number of LoRa nodes
and type of sensors (mainly including temperature, humidity, sound, acceleration, and light sensors), whereas
FlockLab 2 [158] is also capable of achieving high-dynamic range power and logic timing measurements for the
debug and trace. FIT IoT-Lab [3], FlockLab 2 [158], and LinkLab [54] testbeds support various heterogeneous IoT
nodes in various places, while OpenChirp [38] is a LoRa-speciic testbed, but is deployed in restricted campus
areas with an unknown number of LoRa nodes. Early testbed platforms [3, 158] are generally more popular than
the emerging ones [38, 54], due to their stable and mature operations. Besides, as for the programmable endpoint
prototypes, TinySDR [65] provides large-scale deployments with ultra-low energy consumption manner and
OTA solutions, which is also satiable for research with respect to backscatter, PHY/MAC protocol, sensing, etc.
Likewise, ChirpBox [151] ofers an infrastructure-less solution, featuring irmware dissemination and log trace,
which is hence recommended to be deployed in remote areas without communication and power conditions.
Additionally, both possess a low cost of construction and the capability of concurrent reception, but require a
battery supply where energy harvesting can be further considered [151]. However, current testbeds have some
limitations with regard to hardware (limited number of nodes and sensors, single network topology), framework
(limited functionalities), and user experience (complex operation). Hence, providing a large-scale, comprehensive,
and human-machine friendly testbed is in great need.

4 LORA COMMUNICATION
After fully understanding the performance of LoRa, a large number of studies have been devoted to enhancing
its performance in communication. In this section, we review these studies from three aspects: modulation and
demodulation schemes, MAC protocols, and coniguration setting methods. Speciically, various LoRa modulation
and demodulation schemes were proposed for the network throughput improvement on the basis of LoRa
PHY layer CSS modulation and de-chirp demodulation. In addition, modiied MAC protocols aim to handle
multiple access problems without incurring collisions. Coniguration settings refer to optimal determination of
transmission parameter sets to rationalize the use of channel resources and achieve energy fairness across nodes.

ACM Trans. Sensor Netw.

111:12 • Sun et al.

4.1 Modulation & Demodulation

As stated in Section 2.2, LoRa PHY layer adopts a spread spectrum modulation technique derived from CSS,
possessing substantial processing gains and link budget improvements. With considerable modulation beneits,
LoRa demodulation based on the de-chirp operation enables low-SNR [86] and collision [167] signals decoding,
which also receives great attention. Consequently, extensive methods [61, 167] were proposed to expand the
beneits of LoRa PHY modulation technique.

4.1.1 Modulation. Recently, some modiied modulation methods [27, 42] were proposed to leverage LoRa CSS
properties or adopt a combination with other modulation techniques, aiming at increasing the modulation data
rate without compromising its original performance.
Traditional digital modulation are primarily based on the amplitude, frequency, and phase information of the
carrier to form these particular parameters shift keying, including conventional ASK, FSK, PSK, along with their
improvements or combinations. The essence of LoRa modulation is the frequency shift of the chirp, thus some
LoRa modiied modulation methods [13, 112] mainly try to carry extra data information in the LoRa symbols by
combining with other modulation techniques to increase the data rate. Bomin et al. [13] proposed a modulation
method termed PSK-LoRa, which adopts PSK modulation to embed additional data in the phase shift of the
transmitted chirps. Speciically, the information bits are divided into two groups, which determine the initial
frequency and the initial phase of the LoRa symbol, respectively. Similarly in [112], LoRa symbols can start from
any phase value with the help of the pulse shaping ilter, such that extra data information can be encoded to the
initial stage of each symbol, which is then recovered using a coherent receiver. It is noted that PSK can only be
decoded under coherent demodulation, as no carrier component is in the PSK signal’s power spectrum. Such
additional receiver design, along with timing and phase synchronization requirements, will increase the cost and
degrade the battery life of nodes.
Besides, some methods [42, 61] focus on elaborate LoRa signal orthogonal sets. In LoRa conventional CSS
modulation, the linearly varying-frequency up-chirp spans the entire bandwidth, whose cyclic shift serves as
a multi-dimensional space for the orthogonal signals. Elshabrawy and Robert [42] proposed an Interleaved
Chirp Spreading LoRa (ICS-LoRa) modulation method, which creates a new multi-dimensional space from the
interleaved version of nominal LoRa signals. Speciically, the proposed ICS interleaver block subdivides the
given signal containing 2S F samples into four subintervals and simply swaps the samples of the two middle
intervals of each signal. Such that ICS-LoRa can deploy all 2S F possible cyclic shifts of the interleaved base chirp
signal to achieve interleaved chirp signal that carries one extra bit within each symbol. In addition, ICS-LoRa
demodulation can also share this ICS interleaver block for simpliication without any coherent demodulation
requirements. To reduce the cross-correlation between generated and base chirps in ICS-LoRa, Hanif et al. [61]
proposed a slope-shift-keying LoRa (SSK-LoRa) modulation method, which utilizes the linearly varying-frequency
down-chirp to generate the second orthogonal basis set to increase the data rate.
The improved modulation technique is promising since it can carry more data in one transmitted symbol.
Some studies [61, 112] have achieved the throughput gain from 11% to 33% of LoRa networks. However, the
complexity of the transceiver design, the compatibility with COTS devices, and energy consumption issues still
make this work challenging.

4.1.2 Demodulation. Recently, plenty of LoRa demodulation studies were proposed to push the limit of LoRa
capabilities in terms of single-source decoding and collision disambiguation.
Single-Source Decoding Methods. Single-source decoding methods aim to improve the performance of
decoding signals from one speciic source, mainly including low-SNR decoding [86], Cloud Radio Access Network
(C-RAN) [39], error recovery methods [104], etc.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:13

Low-SNR decoding refers to the correct payload reduction of the received signals with an ultra-low SNR
or RSSI. Such methods can break the decoding threshold of conventional LoRa and signiicantly improve the
link budget to provide a wider coverage. The standard LoRa demodulation method performs the de-chirp
operation to determine the SNR threshold, which is generally sub-optimal. To this end, Li et al. [86] proposed
NELoRa, a Deep Neural Network (DNN) demodulator to decode ultra-low SNR LoRa signals by extracting
ine-grained information embedded in LoRa chirps. Speciically, NELoRa transforms the extracted chirp symbols
to dual-channel spectrograms containing phase and amplitude, which are then fed into the dual-stream DNN
for demodulation. The irst DNN stream performs the noise ilter operation to obtain a masked spectrogram,
while the second one is used for packet decoding. Tong et al. [154] proposed Falcon, which provides an available
link for unreachable LoRa devices via making them selectively interfering transmissions of other devices (base
signals) on the same channel. Speciically, a Falcon device overhears the channel using a CAD-based detection
approach, and synchronizes its time and frequency ofset by a reception-based algorithm, then maximizes the
base signal deformation by an adaptive frequency adjusting strategy.
Additionally, several emerging methods [39, 93, 134] focused on the C-RAN design. Speciically, C-RAN is a
centralized radio access network architecture based on cloud computing, whose core is to process the LPWAN
PHY layer in the cloud. Hence, LoRa C-RAN architecture mainly transfers the incompetent PHY layer decoding
tasks at the gateway to the cloud and exploits the spatial diversity gain for weak signal reconstruction. For
example, Dongare et al. [39] proposed Charm, a LoRa C-RAN that pools and jointly decodes weak signals that
cannot decode in any individual gateway, into the cloud. In particular, the gateway introduces a hardware and
software design to detect those signals much weaker than the noise loor by transforming the structure of the
LoRa packet data symbol as resistant as the preamble. The cloud irst adopts a phase-based time synchronization
method, then exploits the timing and geographical location of the received signals to identify their sources
(i.e., the combination of gateways), and collates such information to recover the transmitted data inally. Jointly
decoding the multi-channel LoRa PHY layers in the cloud can improve the SNR of the radio signal but requires a
higher bandwidth, which will then cause severe network congestion. To reduce the bandwidth, Liu et al. [93]
proposed a compressive sensing-based C-RAN termed Nephalai, which leverages the sparsity of the PHY layer for
signal compression and joint reconstruction. Speciically, Nephalai utilizes tailored dictionaries and measurement
matrices for LoRa PHY layer compression, with which the adaptive compression ratios are chosen by considering
SF and SNR.
In all, LoRa ofers substantial processing gains and link budgets to decode low-SNR signals. Thus, many
LoRa low-SNR signal decoding methods adopt diferent methodologies to expand these advantages and improve
network throughput. Among these, learning methods [86] can achieve a high upper limit of the SNR gain based
on the representation learning of data samples, but possess a poor generalization for distinct settings. Besides,
C-RAN architectures [39, 93] mitigate the disadvantages of LPWAN in terms of narrow bandwidth and loose
latency bounds. A full-scale distributed multiple-input multiple-output (MIMO) system embedded in C-RAN can
provide a solution for handling collisions from a large number of nodes [39]. However, C-RAN architectures have
a high operating cost, short transmission distance, and insuicient security, thus inspiring further developments.
Packets may be lost or destructed during transmission due to the varying channel and environment conditions.
Thus, a payload recovery scheme [9, 104, 143] is hence considered to provide the increasing gain and enhance
network throughput. Marcelis et al. [104] proposed a data recovery coding scheme at the application layer termed
DaRe, which applies fountain codes on a sliding window with a inite window. DaRe extends the frame with the
parity check of randomly selected previous frames as the redundancy data, such that the receiver can decode all
payload data when only subsets are received. However, this comes at the cost of increased transmission time.
Balanuta et al. [9] proposed an Opportunistic Packet Recovery (OPR) approach, which exploits gateway spatial
diversity and tracks packet RSSI value for error detection, and leverages the Message Integrity Check (MIC)
ield information for error correction. Speciically, OPR collects and groups the corrupt packets across gateways

ACM Trans. Sensor Netw.

111:14 • Sun et al.

instead of discarding them, then generates a candidate set of corrupt bits based on the spatial diversity and
reception time. The generated candidate set is leveraged by the cloud to search through the possible valid CRC
combinations, which results in a few packets for inal iltering via MIC check.
Additionally, decreasing clock rates can reduce the energy consumption of LoRa radio, but it is limited by
the Nyquist sampling theorem that requires the clock rate to be at least twice the LoRa chirp bandwidth. To
reliably decode packets sampled at the sub-Nyquist rate, Xia et al. [169] proposed LiteNap, which introduces
downclocked operating mode for LoRa. They had two-fold observations, irst, two under-sampled LoRa chirps
sufer from frequency aliasing that causes demodulation ambiguity; second, radio hardware induces a constant
phase shift to all modulated chirp after jitters, resulting in frequency leakage in the time domain. Based on these,
LiteNap leverages such frequency leakage as a ingerprint to uniquely identify a LoRa chirp and extract the
timing information. Besides, they proposed novel packet detection and synchronization strategies, along with the
integration with LoRaWAN protocol.
Collision Disambiguation Methods. The LoRa gateway has an extremely wide coverage of nodes, thus
LoRa networks may be subject to pervasive and severe intra- and inter-network interference, especially when
in a dense deployment scenario. Interference adversely afects the signal reception, coupled with the capture
efect, which is a waste of air time and spectrum. This problem results in the demand to support concurrent
transmissions or avoid collisions. Consequently, plenty of interference solutions were proposed, including PHY
layer collision disambiguation algorithms [167, 168] to recover packet information and MAC protocols [49, 173]
(reviewed in Section 4.2) to maximize the use of channel resources, which both do not conlict but collaborate
with each other.
Collision disambiguation refers to the separation and correct decoding of multiple collided signals at the receive
side. As stated in Section 2.2, in demodulation, the LoRa receiver detects the packet preamble and SFD, and divides
the received signals into a series of windows. Then it performs the de-chirp operation in each window, where each
symbol is multiplied with a base down-chirp. The resulted a single frequency tone (i.e., a FFT peak in FFT bins)
indicates the initial frequency. While in the case of collisions, signals from multiple nodes are superimposed at
the gateway, which induces a distorted one, i.e., multiple peaks are obtained in FFT bins. To this end, the existing
collision disambiguation methods mainly separate and divide the collided symbols to the transmitters based
on the unique features (in time [162, 168], frequency [41], phase [167], and power [68, 165] domains) of each
transmitter symbol, and decode them individually. Table 6 summarizes the existing LoRa collision disambiguation
methods.
As the transmitted signal has a shift in time, frequency, and phase resulting from natural hardware ofsets,
Eletreby et al. [41] proposed Choir, which exploits such subtle frequency shifts to separate diferent signals,
then track users to decode data inally. However, such hardware-induced frequency ofset is hard to capture
due to the background noise and frequency leakage. Hu et al. [71] proposed SCLoRa with adaptation to the
dynamic environment, which exploits cumulative spectral coeicient integrating multi-dimensional information
(frequency and power) under the considerations of channel fading and spectrum leakage. Xia et al. [168] revealed
that the frequency of the LoRa symbol increases periodically, while the symbol edges of diferent interference
transmissions are misaligned in the time domain. They presented FTrack, which separates collisions and recovers
frames by leveraging such time-domain misaligned edges and signal frequency continuity. FTrack requires a
sliding window per sample, resulting in a large computational overhead. In their follow-up work [167], besides
using time and frequency features, they designed a Phase-based Parallel Packet decoder termed PCube for
concurrent transmissions by leveraging the reception diversities of MIMO hardware. Speciically, Pcube irst
calibrates the frequency ofset of the received signal to extract the correct frame timing of each packet based on
the frame structure of the LoRa preamble and SFD. Then it measures the phases of all concurrent symbols, and
mitigates the impact of hardware-induced phase variance. Furthermore, Pcube extracts the air-channel phase of
each symbol and groups symbols to corresponding packets.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:15

Table 6. Summary of LoRa disambiguation decoding methods.

Performance
Method Year Methodology
Low-SNR Real-Time #Collision Throughput
Choir [41] 2017 Hardware imperfection-induced ofset % % 10 6.84× the conventional LoRa
mLoRa [162] 2019 Successive interference cancellation % % 3 3× conventional LoRa
CoLoRa [156] 2020 Identical peak ratio of chirps in the same packet " % 20 3.4× Choir [41]
FlipLoRa [180] 2020 Encoding with interleaved up-down chirp " % 18 3.84× conventional LoRa
Time-domain misaligned edges,
Ftrack [168] 2020 % % 10 3× conventional LoRa
and signal frequency continuity
Nscale [155] 2020 Non-stationary amplitude scaling down-chirp " % 30 3.3× Choir [41]
OCT [165] 2020 Decoding with time and power ofsets % " 3 3× conventional LoRa
SCLoRa [71] 2020 Multi-dimensional cumulative spectral coeicient " % 20 3× Ftrack [168]
Temim et al. [150] 2020 Successive interference cancellation % % 3 -
1.68× Nscale [155]
AlignTrack [26] 2021 Collided chirps alignment " % 12
3× CoLoRa [156]
Reception diversities of MIMO,
Pcube [167] 2021 % % 40 4.9× Ftrack [168]
and air-channel phase diference measurement
Variation of FFT peak heights
Pyramid [181] 2021 " " 6 2.11× conventional LoRa
in multiple windows
Shahid et al. [135] 2021 Concurrent Interference cancellation " % 20 4× Ftrack [168]

Some collision disambiguation methods [156, 180] are designed for low SNR situations. For example, Chen and
Wang [26] proposed AlignTrack for low-SNR collision decoding, which aligns a moving window with diferent
collided chirps to ind the peak of the aligned chirps, and thus separates these frequency peaks to their belonging
packets. Instead of chirp partition in [155, 156], AlignTrack leverages the entire chirps, such that the aligned
chirp can induce the highest peak in the frequency and lowest SNR loss. Some methods [162, 168] focused on the
time-domain feature that is vulnerable to noise, falling short of decoding low SNR signals. To this end, more
robust frequency features [155, 156, 181] were utilized. Tong et al. [156] proposed CoLoRa for collision decoding
based on the peak ratio of chirps. Speciically, CoLoRa irst demodulates the collided chirps in multiple (usually
misaligned) reception windows, thus each chirp can result in two peaks in two windows. Then CoLoRa leverages
the inding that the peak ratio of chirps (the height of the latter divided by that of the former) in the same packet
is identical, to disentangle collided signals. To relax the restrictions of the reception window, Tong et al. [155]
proposed NScale. Speciically, NScale leverages the non-stationary amplitude scaling down-chirp to translate the
packet time misalignment into frequency features, thus can determines the distribution characteristics of the
symbol segments in the window according to the peak height variation before and after scaling. Besides, they
proposed an iterative peak recovery algorithm to resolve peak distortion. With the reception window shifting,
the FFT peak heights of one chirp in multiple windows will be presented in a pyramid shape. So Xu et al. [181]
proposed Pyramid to achieve low-overhead real-time collision decoding, which separates packets via exploiting
each łtopž of the pyramid.
Rather than matching symbols to transmitters, interference cancellation algorithms [135, 150, 162] are also
applied. Wang et al. [162] proposed mLoRa, which iteratively decodes and then cancels part of the collision-free
symbols. Similarly, Temim et al. [150] identiied and decoded the strongest received signal, then reproduced
its complex envelope and removes it from the received signal utilizing a conventional Successive Interference
Cancellation (SIC) algorithm. Shahid et al. [135] proposed a Concurrent Interference Cancellation (CIC) method.
In particular, CIC irst determines these symbol boundaries via preamble detection, then selects an Interference
Cancelling Sub-Symbols Set (ICSS) with no common interfering symbol across all symbols. Finally, CIC adopts a
spectral intersection operation to demodulate symbols via canceling out all interfering symbols.
Current collision disambiguation methods have signiicantly ameliorated the concurrent transmission capability
of LoRa gateways and alleviated serious collisions. Most of these methods [41, 156] achieve more than 3× increment

ACM Trans. Sensor Netw.

111:16 • Sun et al.

in throughput, and some [155, 156] give a focus on weak decoding of signals below −5 dB. Fundamentally diferent
with collided signals separation methods by leveraging signal features, SIC algorithms provide a novel solution.
However, devising an efective and energy-eicient collision decoding method remains challenging in terms
of accurate packet detection, optimal peak height search, and algorithm complexity. Speciically, the Carrier
Frequency Ofset (CFO) and inter-packet interference need to be combated irst to avoid the wrong packet
estimation [26]. Demodulating a low-SNR signal results in a poor peak height, coupled with accompanied
sidelobes around. Besides, the reception windows (e.g., alignment, length) also afect the peak height. Heavy
algorithm complexity is impractical for resource-constrained LoRa networks, as a semi- or of-line decoding way
will cause great delay, memory cost, and computational overhead at the gateway side.
Overall, various demodulation methods leveraging LoRa PHY CSS properties and adopting advanced techniques
have improved the network throughput and scalability to a large extent, which is a primary research focus in the
research community. Irrespective of the remarkable results achieved, the complexity of the algorithm, cost of
hardware and actual deployment, and the inluence of multiple external factors (e.g., dynamic channel conditions,
various environments) still requires further study.

4.2 MAC Protocols

MAC protocols refer to the methodology that allows multiple nodes in the networks to access a shared transmission
medium. LoRaWAN employs a default pure ALOHA MAC protocol, whose basic idea is to send data without
detection. It is simple and has no time synchronization requirements, making it suitable for resource-constrained
LoRa networks. However, the network performance will degrade dramatically due to the inevitable interference
restrictions when the number of transmissions increases greatly, thereby sufering limited scalability. To avoid
severe collisions, plenty of MAC protocols were proposed recently, which can be divided into two classes: the
contention [49] and schedule-based [2] ones. The contention-based MAC protocols are mainly those random
access protocols such as Slotted-ALOHA and Carrier Sense Multiple Access (CSMA), where all nodes keep
listening to and compete for the shared medium for transmission. In the schedule-based ones, multiple nodes
access the predetermined collision-free medium that is divided according to time (TDMA) or frequency (FDMA).
Table 7 summarizes the existing MAC protocols.
4.2.1 Contention-Based MAC Protocols. Recently, plenty of LoRa contention-based MAC protocols were proposed,
where the sender listens to the shared medium before transmission and transmits until the medium is free. The
existing methods mainly include Slotted-ALOHA and CSMA ones.
Slotted-ALOHA Methods. Slotted-ALOHA, a variant of ALOHA, aligns the transmissions to slot boundaries
relying on the relative synchronization of nodes, which can reduce the interference and increase channel capacity.
The general process is to divide the time into several identical time slices, where all senders access the channel
synchronously at the beginning of the time slice, and have to wait until the beginning of the next time slice
before transmission in the case of an occurring conlict. As Real-Time Clocks (RTCs) have signiicant clock drift
problems over time, Polonelli et al. [122] introduced a lightweight time synchronization method based on the
timestamp and ofset during one transmission, and achieved a Slotted-ALOHA mechanism with random backof
to increase the channel capacity. Ihirri et al. [72] proposed a scalable Slotted-Aloha method by introducing a
contention sequence code in the option ield of the LoRa frame. Such code determines the priority of transmission
and decreases iteratively. Beltramelli et al. [11] proposed a Slotted-ALOHA MAC protocol using an out-of-band
synchronization based on FM-radio data system (FM-RDS) broadcasting, which could save energy and improve
the capacity unlike an in-band synchronization strategy.
Carrier Sense Multiple Access Methods. Slotted-ALOHA and TDMA protocols require clock synchroniza-
tion that brings a burden to LoRa networks, while CSMA protocols provide an alternative solution without
synchronization operation. Several studies [49, 173] have exploited CSMA or its variants (i.e., Listen-Before-Talk

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:17

Table 7. Summary of LoRa MAC protocols.

Method Year Methodology Performance
Slotted-ALOHA with synchronization leveraging
Ihirri et al. [72] 2019 -
FM-RDS broadcasting
Slotted-ALOHA with synchronization relying on
Contention-Based

Polonelli et al. [122] 2019 Throughput: 5.8× conventional LoRaWAN

the timestamp and ofset
Three LMAC versions based on CAD: LMAC-1 using
Throughput: 2.2× conventional LoRaWAN
LMAC [49] 2020 DIFS with BO, LMAC-2 by indirect channel probing,
Energy eiciency: 2.4× conventional LoRaWAN
LMAC-3 relying on beacon broadcasting
Throughput: 3ś20× conventional LoRaWAN
p-CARMA [80] 2020 p persistent-CAD multiple access
Energy eiciency: 1.42ś1.63× conventional LoRaWAN
Residual energy-based CAD, CSMA-CA or Lantency: 0.79× conventional LoRaWAN
Xu and Zhao [173] 2020
dynamic adjustment duty cycle according to traic load Energy eiciency: 1.19× conventional LoRaWAN
Scalable Slotted-ALOHA based on the divided Throughput: 2× conventional LoRaWAN
Beltramelli et al. [11] 2021
contention sequence code from LoRa frame Energy eiciency: 2× conventional LoRaWAN
Rizzi et al. [131] 2017 Integration with TSCH and TDMA -
Haxhibeqiri et al. [63] 2018 Centralized scheduling Throughput: 1.3× conventional LoRaWAN
Piyare et al. [121] 2018 On-demand TDMA using wake-up radios PDR: almost 100%
Schedule-Based

RS-LoRa [129] 2018 Two-step lightweight MAC scheduling Throughput: 1.2× conventional LoRaWAN
EF-LoRa [52] 2019 Energy fairness-enabled FDMA Energy fairness: 177.8%
FREE [2] 2019 Bulk transmission and parameter assignment PDR: almost 100%
LoRaCP [58] 2019 TDMA featuring an urgent ALOHA channel and negative ACK Throughput: 65% → 90%
RT-LoRa [84] 2019 Novel MAC layer design -
S-MAC [179] 2020 Sending time prediction and frequency channel assignment Throughput: 4.01× conventional LoRaWAN
TS-LoRa [190] 2020 Distributed time-slotted approach PDR: >99%
PolarTracker [164] 2021 Node attitude tracking and transmission scheduling Throughput: 1.49× conventional LoRaWAN

mechanism [116]) on LoRa networks based on the Channel Activity Detection (CAD) feature. To address the
issue of false negatives introduced by CAD during the transmission of the payload, Kouvelas et al. [80] proposed
a CSMA variant termed p-CARMA, where nodes adaptively select an appropriate probability value of p for
transmission using a heuristic approach when the channel is idle. In the system proposed by Xu et al.[173],
the LoRa nodes exploit the residual energy to determine the efective preamble from the noise to avoid false
awakening, and access to the channel via selecting CSMA-CA or dynamic adjustment duty cycle mechanism in
the case of low or high traic load, respectively. Gamage et al. [49] designed three modiied LMAC versions to
enable CSMA for LoRa networks based on the CAD feature. Speciically, LMAC-1 achieves the basic functionality
of CSMA based on a Distributed Inter-Frame Space (DIFS) mechanism with a ixed number of CADs and a random
back-of (BO) strategy. LMAC-2 achieves a balanced resource load through an indirect channel probing approach,
which updates the knowledge regarding the channels’ crowdedness based on the CAD results during DIFS and
BO processes. LMAC-3 receives a global view of channel loads via broadcasting periodic beacons by gateways.

4.2.2 Schedule-Based MAC Protocols. Recently, LoRa schedule-based MAC protocols are also popular, where
the sender transmits according to pre-assigned link resources. The existing schedule-based ones mainly focus
on Time Division Multiple Access (TDMA) along with time slot scheduling strategies, and Frequency Division
Multiple Access (FDMA) methods.
Time Division Multiple Access Methods. TDMA protocols allow multiple nodes to use the same frequency
for transmission in diferent time slots, thereby sharing the same transmission medium without incurring
collisions. They also avoid the energy waste of over-listening and idle-listening to the channel, hence receiving
much popularity. For example, Rizzi et al. [131] integrated Time Slotted Channel Hopping (TSCH) strategy with
TDMA to enhance the network throughput and reliability. However, the synchronization strategy is missing.
Piyare et al. [121] proposed an on-demand TDMA approach using low-energy wake-up radios, which respectively
provides unicast and broadcast modes for node triggering and time slots allocation. Gu et al. [58] designed

ACM Trans. Sensor Netw.

111:18 • Sun et al.

a TDMA-based LoRa multi-channel transmission control featuring an urgent ALOHA channel and negative
acknowledgment (ACK), to achieve the one-hop out-of-band control plane for wireless sensor networks.
For TDMA MAC studies, the critical problem is time slot scheduling/allocation. Haxhibeqiri et al. [63] relied on
the network synchronization and scheduling entity (NSSE) as the central scheduler for the LoRaWAN network to
schedule transmissions. Speciically, the node sends a request containing the traic periodicity to the NSSE, and
receives a reply about the allocated time slots encoded in a probabilistic space-eicient data structure. However,
some nodes may share the same slot with a certain probability, incurring collisions. Abdelfadeel et al. [2] proposed
a ine-grained scheduling scheme termed FREE. Speciically, nodes are assigned with corresponding transmission
parameters inclusive of SF, TP, and time slot, then perform bulk data transmission in the predetermined time
slot. However, FREE resolves the collision problem but falls short of real-time transmission. Leonardi et al. [84]
proposed RT-LoRa, a novel LoRa MAC protocol as an alternative of LoRaWAN, which can support real-time low
transmission. In RT-LoRa, the time slot duration is limited by minimal packet size as a lower bound, and varies
according to diferent SFs. Rather than relying on the centralized scheduler to allocate separate time slots for all
nodes, Zorbas et al. [190] proposed TS-LoRa, a self-organizing time-slotted communication approach based on
computing a hash algorithm mapping the nodes’ assigned addresses into unique slot numbers. Additionally, the
dynamic attitude of loating nodes incurs signal strength losses and packet errors compared with those deployed
statically on the ground due to the polarization and directivity of the antenna. To this end, Wang et al. [164]
proposed an attitude-aware link model along with a channel access method termed PolarTracker, which leverages
the attitude alignment state of the node then schedules the transmissions into best-aligned periods for better link
quality.
Frequency Division Multiple Access Methods. Similarly, FDMA protocols allow multiple nodes to transmit
in diferent frequency channels of the shared medium simultaneously. As LoRa band in distinct regions deines
diferent multiple frequency channels, FDMA-based studies are CF allocation methods in essence. Plenty of CF
parameter allocation methods [52, 129] were proposed recently. For example, Reynders [129] proposed a MAC
layer protocol termed RS-LoRa, which employs a two-step lightweight scheduling based on the RSS at both
nodes and gateways. Speciically, the gateway speciies the allowed TPs and SFs for each frequency channel via
beacon broadcasting, while nodes specify their own ones. Gao et al. [52] proposed EF-LoRa to allocate frequency
channels under the consideration of the randomness of LoRa MAC protocol. Xu et al. [179] proposed an adaptive
MAC layer scheduler termed S-MAC, which predicts the sending times based on the periodic transmission
characteristics of nodes and allocates frequency channels according to the SF parameter of the packet.
In general, the existing MAC protocols have signiicantly ameliorated the severe occurrence of collisions to
improve the network performance in capacity and scalability. However, it is also challenging to devise adaptive,
efective, and energy-eicient MAC protocols. Classic Slotted-ALOHA methods adopt synchronization strategies
mainly based on timestamp [122] and out-of-band radio. Nevertheless, the number of time slots required is ixed
and cannot be adjusted arbitrarily, which can induce collisions (too many) or time slot waste (too few). The
synchronization and slot scheduling will bring the cost of calculation and propagation transmission time. The
CAD features of LoRa facilitate the CSMA-based methods, among which LMAC [49] achieves communication
fairness across nodes on the basis a 2.2× and 2.4× improvement in throughput and energy eiciency, respectively.
CSMA-based methods have no requirements for synchronization, but impose additional energy for the over-
listening and long-time continuous transmission detection operation. TDMA-based methods mainly focus on the
eicient time slot allocation in centralized [2] and distributed [190] manners. They are inlexible, especially when
the data packet size varies and network topology changes dynamically. Energy-eicient synchronization and
feedback ACK problems require to be tackled. FDMA-based methods are CF allocation methods in essence, along
with other parameters together. They lower the interference and are easy to implement, but may cause spectrum
waste due to the existence of the guard-band.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:19

4.3 Configuration Setings

Coniguration settings typically refer to the optimal determination of transmission parameter sets in terms of
SF, TP, CF, and CR assigned to LoRa nodes or networks. As diferent transmission parameter sets resulting in
diferent data rates and air time, have a distinct efect over the network performance or functionality, such as
throughput and energy consumption unfairness across nodes. For example, nodes far away from the gateway
usually are tuned a larger SF to trade low data rate for long range, but resulting in longer transmission time
and higher energy consumption. Hence, an efective network coniguration setting method is crucial for the
appropriate division of channel resources and enhancing the throughput and energy eiciency of LoRa networks.
Speciically, the existing coniguration settings methods [52, 87] mainly achieve the optimal parameter sets
determination of LoRa nodes according to the deployment topology, communication behavior, and channel
conditions. The main process consist of three steps: irst, taking the throughput and energy consumption/fairness
as the optimization goal; second, formulating a system or link model based on the optimization problem; third,
adopting speciic optimization methods based on current system model. It is noted that the data rate of LoRa
radio is deined by the combination of SF, BW, and CR, thus the ADR method is also regarded as the coniguration
setting category. Table 8 summarizes the existing LoRa coniguration setting methods.

Table 8. Summary of LoRa configuration seting methods (S: Simulation, T: Testbed).

Method Year Parameter Problem Formulation Methodology Real-Time Dynamic Evaluation
EXPLoRA [31] 2017 Data extraction rate optimization Heuristic algorithm % % S
CA-ADR [105] 2020 Minimizing the collision probability Choosing the smallest SF based on the power % " T, S
SF Throughput optimization
Loubany et al. [98] 2020 Adjusting SNR thresholds % % S
(traic load×probability of success)
Maximizing throughput based on
Mu et al. [109] 2020 K-Nearest Neighbors " " T
the dataset containing RSS, SNR, SF, etc.
FADR [1] 2018 Fair data rate distribution based on DER Allocation based on the average RSSI values % % S
Allocation based on the distance and
Liando et al. [89] 2019 Node lifetime model % " T
lifetime threshold
SF, TP Matching theory for SF,
Amichi et al. [5] 2020 Short-term average rate modeling % % S
constrain approximation for TP
Traversing and comparing all combinations
DyLoRa [87] 2020 Symbol error and energy eiciency model " " T
in the prediction model for optimal settings
Allocation based on path loss values
Reynders et al. [127] 2017 Packet error rate fairness % % S
and SF distribution under TP
RSS value restriction,
RS-LoRa [129] 2018 Two-step lightweight MAC scheduling % % S
and packet error rate fairness
EF-LoRa [52] 2019 CF, SF, TP Max-min energy fairness optimization Greedy heuristics algorithm % % S
Matching theory for channel,
Maximizing system
Su et al. [142] 2020 heuristic algorithm for SF, % % S
and node energy eiciency
and optimization algorithm for TP
Bor et al. [14] 2017 SF, BW, CR, TP PRR measurement Traversing and comparing all combinations % " T
MDP model characterizing harvested energy ECAA for channel,
Liu et al. [96] 2019 CF, TP " " S
and channel conditions dynamic programming for TP
Symbol-level network model Allocation adaption acceptance depending on
AdapLoRa [53] 2020 CF, SF, CR, TP % " T
for network lifetime optimization the comparison with the lifetime threshold
Optimal frequency estimation Multipath signal disentangling
Chime [47] 2020 CF % " T
across base stations collaboratively and recombination
Trade-of between the delivery ratio Allocation based on
EARN [117] 2020 SF, TP, CR % " S
and energy consumption the aggregated load status for SF and SNR

SF Allocation Methods. Loubany et al. [98] formulated the throughput optimization problem under consid-
eration of the capture efect, and proposed an adaptive SF allocation algorithm via adjusting the SNR thresholds.
Mu et al. [109] proposed a K-Nearest Neighbors (KNN)-based SF allocation method, containing an initialization
and operation period. In the initialization period, nodes are set all SF conigurations in a round-robin fashion,
and the gateway creates a dataset containing link characteristics (i.e., RSS, SNR), SF conigurations, and packet
reception results. In the operation one, a KNN algorithm is adopted to select optimal SF under the given link
conditions, or to meet the application reliability requirements via voting threshold adjustment. Cuomo et al. [31]
proposed a heuristic SF allocation algorithm termed EXPLoRA to improve the Data Extraction Rate (DER) and

ACM Trans. Sensor Netw.

111:20 • Sun et al.

achieve air time fairness. Speciically, EXPLoRA irst calculates the available SF list of each node based on the RSSI
and the sensitivity of each SF, and then utilizes an łordered waterillingž strategy to allocate the SF to balance
the air time. Marini et al. [105] proposed a collision-aware ADR (CA-ADR) algorithm. CA-ADR irst calculates
the maximum number of nodes under speciic allocated SFs based on the given packet success probability that
considers the link-level performance and the collision probability, and then determines the smallest available SF
for each node based on the average received power.
SF and TP Allocation Methods. Liando et al. [89] modeled the energy consumption of LoRa nodes for SF
and TP allocation. Amichi et al. [5] formulated a joint SF and power allocation problem as the uplink short-term
average rate modeling featuring SF quasi-orthogonality. Speciically, they achieved SF assignment sub-question
under ixed TP via a low-complexity many-to-one matching algorithm, and TP allocation under ixed SF via two
types of constraints’ approximation. Abdelfadeel et al. [1] proposed a Fair Adaptive Data Rate (FADR) algorithm.
Speciically, FADR derives a fair data rate distribution on the basis of Data Extraction Rate (DER) across nodes,
then adopts a genetic algorithm for the optimal SF distribution and adjusted TP within a safe margin. Li et al.
[87] proposed a Dynamic LoRa transmission control system termed DyLoRa, which derives an energy eiciency
characterization model based on transmission parameters including TP, SF, and SNR. In particular, DyLoRa
gateway irst extracts the average SNR of the last pre-deined number of data packets as an indicator of link
quality, then traverses and inputs all combinations SF and TP with this average SNR to the prediction model for
the optimal setting.
CF, SF, and TP Allocation Methods. Reynders et al. [127] derived the optimal SF distribution under the
constrained TP to minimize collision probability, and also assigned nodes from long distances that possess large
path loss values to diferent frequency channels to avoid near-far problems for fairness. In their further work [129],
they proposed RS-LoRa, which employs a two-step lightweight scheduling. Speciically, the gateway speciies the
allowed TPs and SFs for each channel, which are restricted by the RSS value and collision probability, respectively.
Then, the nodes traverse to ind a proper channel, among which nodes divided into one group choose similar
parameters to alleviate the capture efect. Gao et al. [52] proposed EF-LoRa, formulating energy fairness issues as
a max-min optimization problem and utilizing a greedy heuristics algorithm to allocate frequency channel, SF,
and TP parameters. Speciically, EF-LoRa adopts a multiple-gateway system model considering various LoRa
network features such as interference, the randomness of MAC protocol, and capacity limitation of gateways, to
serve energy fairness optimization. However, EF-LoRa runs only once at the irst time of network deployment.
In [142], they formulated user scheduling as a two-sided many-to-one matching problem with peer efects to
achieve channel allocation, and allocated SF via a heuristic algorithm. Then, they performed TP allocations under
maximizing system and nodes’ minimal energy eiciency via lower bound approximation and sequential convex
programming, respectively.
Other Parameters Allocation Methods. Gadre et al. [47] proposed Chime, where the node sends one packet
across multiple base stations at one frequency, and the stations can collaboratively determine the optimal
frequency. Speciically, Chime irst requires synchronizing distributed base stations to avoid the time-varying
and long-lasting phase errors, then models the signals and disentangles diferent signal multipath. Finally, Chime
recombines these separated signal components to estimate the optimal one. Bor et al. [14] proposed a simple link
probing regime which traverses and approaches the optimal parameter coniguration based on the measured
PRR. Liu et al. [96] adopted a Markov Decision Process (MDP) model to formulate the harvested energy and
channel conditions for energy harvesting-based LoRa networks. Besides, they proposed an eicient channel
allocation algorithm (ECAA) based on a many-to-one matching game by enabling users to self-match properest
ones, and perform optimal TP allocation via solving the dynamic programming problem. Current methods mainly
focus on static resource allocation, Gao et al. [53] proposed AdapLoRa, an adaptive allocation system for CF, SF,
TP, and CR parameters based on the dynamic link conditions on the contrary. Speciically, AdapLoRa adopts
a symbol-level ine-grained network model featuring the properties of enhanced error correction scheme and

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:21

packet reception by multiple gateways to periodically estimate the network lifetime under diferent resource
allocations, and determines whether to perform this adaption through comparison with the threshold rather than
always seeking the optimal setting. Park et al. [117] formulated the link performance as an energy per packet
(EPP) model, and then proposed an enhanced greedy ADR mechanism with CR adaptation termed EARN based
on the aggregated load status for each SF and SNR. Besides, EARN exploits adaptive SNR margin to withstand
the dynamic link changes.
Apart from the aforementioned coniguration setting studies at the node level, several methods aim at the
network topologies and deployments at the network level, such as gateway planning [123, 144] and packet
oloading [43]. Rady et al. [123] irst adopted a K-means clustering-like method or a grid and spatial method to
determine the optimal gateway location under the network-aware or network-agnostic gateway deployment,
respectively. Then, they performed link allocation tasks, i.e., many-to-one mappings between node and gateway.
Speciically, such tasks are conducted based on the minimal distance or the corresponding RSSI value under the
unconstrained gateway capacity, while using the integer linear programming (NP-complete) approach under the
constrained one. Sun et al. [144] proposed a deleted greedy algorithm for optimal gateway number and location
search for heterogeneous LoRa nodes, and also considered SF allocations.
In general, the existing coniguration setting methods utilize network resources to the greatest extent based on
the link conditions and network deployments, which greatly improved the network throughput, energy eiciency,
and fairness. Such methods generally take the throughput and energy performance indicator as a starting point
for problem formulation and modeling. Then, comprehensive and careful considerations of various efect factors
are required, such as dynamic link characteristics and signal interference, along with their representation margin.
Among these methods, SF, TP, and CF are the most chosen parameters, which are mainly allocated based on the
packet SNR and node distance information. However, the algorithm complexity, convergence time, and dynamic
adaption deserve to be tackled for further improvements.
Apart from the aforementioned methods on LoRa communication, other works have made contributions
with respect to the network synchronization strategies [124], data forwarding schemes [25], diferent network
topologies such as mesh [83, 115] and tree [149].

5 LORA SECURITY
With massive deployments of LoRa networks recently, the security and privacy issues are receiving great attention.
Security requires the hardware, software, and data low in LoRa networks are protected, and the whole system
can operate regularly and continually. However, maintaining the conidentiality, integrity, and availability (CIA)
of LoRa networks faces severe challenges due to the openness of the transmission medium and the instability of
the network structure. To this end, we survey LoRa security-related works, including vulnerability analysis and
corresponding countermeasures, coupled with the emerging PHY layer security methods.

5.1 Vulnerabilities and Countermeasures

The vulnerability is the weakness that an attacker can exploit to perform unauthorized actions, modify data,
or make a system unavailable, while the countermeasure is the defense taken against such attacks. The self-
explanatory importance of cyber security and the wide popularity of LoRa networks have stimulated the rapid
development of the attacks and defenses designed for LoRa networks. The irst step in conducting LoRa security
is to understand the network security requirements. CIA Triad is typically used to describe the security of IoT
applications:
• Conidentiality. Data is only authorized to legitimate parties, while not leaked to others.
• Integrity. Data is ensured not to be tampered or destructed during the period of transmission and storage,
or can be discovered quickly after malicious use.

ACM Trans. Sensor Netw.

111:22 • Sun et al.

• Availability. Data can meet the standards for use when needed.
Table 9 lists the common attacks to LoRa networks from the perspective of the violation of the CIA Triad, along
with the proposed corresponding countermeasures.

Table 9. Summary of LoRa vulnerabilities and countermeasures.

Vulnerability Operation Countermeasure
Eavesdropping Retrieving private information stealthily Key generation [50]
Integrity Conidentiality

MITM Attack Creating separate links to both legitimate parties Authentication [163]
Key-related solution [79, 147, 153],
Replay Attack Sending a former eavesdropped packet
random token [110], authentication [163]
Recovering secret key information from additional
Side-Channel Attack key management [60, 78]
signal features (e.g., power, electromagnetic leaks)
Spooing Attack Impersonating legitimate parties to access and tamper data Authentication [163], RFFI [132]
Covert Channel Embedding hidden information Demodulation examination,
(e.g., CloakLoRa [69], EMLoRa [137]) into the covert channel gateway collaboration [68]
Creating a high-speed link tunnel Beating jammer reaction time [7],
Wormhole Attack
between two malicious parties message relation [66]
Beacon authentication key [17],
Beacon Synchronization Attack Sending fake beacons
cryptographic signature [184]
Availability

Delay Attack Malicious frame collision and delayed replay RFFI [57]
Causing the link overload
(Distributed) DoS Attack Blockchain [67, 90]
or triggering network crash (from multiple sources)
Disrupting the legitimate communication Traic analysis [7], intrusion detection [33],
Jamming Attack
using powerful interference radio collision decoding [68], passive packet sniing (LoRadar) [28]

5.1.1 Vulnerabilities. Studying network vulnerabilities shows signiicant importance, which can better serve
the corresponding defenses. Thus, plenty of studies [6, 17, 18, 66, 184] have investigated the common attacks
of networks, inclusive of replay, jamming, spooing attacks, etc. Besides, some vulnerabilities speciic to LoRa
networks [69, 137] are discovered. Hou et al. [69] revealed the existence of a covert channel using a modulation
scheme orthogonal to CSS over LoRa PHY layer, which is transparent and covert to current security mechanisms.
Speciically, they proposed CloakLoRa to embed hidden information into the covert channel utilizing Amplitude
Modulation (AM), where the malicious attacker could decode the hidden data based on the RSS while the
transmission between legitimate parties is not afected. There is a common belief that electromagnetic (EMG)
covert channel is a common short-range attack, as EMG radiation is easily attenuated. However, Shen et al.
[137] proposed a resilient EMG covert channel termed EMLoRa, which reshapes EMG radiation into LoRa-like
chirps through AM, hence the receiver can decode and steal the sensitive data from a long distance. Speciically,
EMLoRa enables three attacks in terms of wide-area data exiltration, penetrating Faraday cage, and localization
of air-gapped devices.
5.1.2 Countermeasures. LoRa networks are vulnerable to kinds of attacks as investigated in [6, 184], hence plenty
of attack defense and prevention methods are proposed accordingly.
Countermeasures Against Replay Attacks. The replay attack refers that the attacker sends a former
eavesdropped packet intact (the receiver node has received before) to deceive the receiver. The attacker does not
need to obtain the explicit raw data but replays some data packets to destroy the correctness of authentication,
which burdens the link load and induces some speciic former-packet efects. Traditional solutions include adding
nounces, timestamps, session ID [178]. For LoRa, replay attack defense methods mainly focus on DevNonce and
NwkSKey of the LoRaWAN packet header in the OTAA join procedure. DevNonce is a random number generated
from nodes, while NwkSKey is the session key that changes every time the joining process is completed. Na
et al. [110] proposed a replay attack scenario occurring in the join request transfer process, and a token-based
countermeasure against it accordingly. The random token is the irst six bits of NwkSkey, which is used to be
XOR-ed with the DevNonce and MIC ields of the join request packet. However, the problem of missing NwkSkey

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:23

is ignored when the device resets. The network server needs to store all DevNonce values used in the previous
joining process, such that requests from benign nodes will not be mistaken as replay attacks [153]. To this
end, Kim and Song [79] deined the initial and non-initial join requests, and checked the validity of NwkSKey
of non-initial join requests and DevNonce of initial ones to prevent from replay attacks. Since checking the
DevNonce value only is not reliable [153], Sung et al. [147] also utilized RSSI and a hand-shaking technique to
protect networks.
Countermeasures Against Jamming Attacks. Jamming attack, a subset of Denial of Service (DoS) attack,
refers to deliberately disrupting or preventing legitimate communication based on malicious interference. Such
attacks can be addressed by eicient intrusion detection/iltering, traic analysis, and veriication. Several methods
[28, 68] were proposed for anti-jamming attacks for LoRa networks. Aras et al. [7] proposed three jamming
attack techniques, and a series of complementary countermeasures such as maximum use of channel hopping
and real-time traic analysis. Danish et al. [33] proposed a novel LoRaWAN-based Intrusion Detection System
(LIDS) involving two LIDS algorithms, namely, Kullback Leibler Divergence and Hamming Distance, deployed on
gateways to monitor the real-time traic distribution for the comparison with those from baseline to prevent from
jamming attacks. Synchronized jamming chirps will make packets not be decoded in the time domain, hence Hou
et al. [68] proposed a prevention and error recovery method by leveraging the signal strength diference. With
massive deployments of LoRa networks, Choi et al. [28] proposed a passive packet sniing framework for MAC
layer termed LoRadar. LoRadar cannot decode payload data but extracts a large number of parameter information
and deployment statistics related to the link quality, which is utilized for jamming detection, RSSI-based device
localization, and so on.
Countermeasures Against Other Attacks. Apart from the aforementioned studies, several other attack
defense methods [57, 163] were proposed. Wang et al. [163] proposed a lightweight node authentication method
termed SLoRa, using ine-grained CFO resulting from hardware imperfections and spatial-temporal link signature
relying on positions of nodes, to prevent from various attacks such as spooing, MITM, DoS attacks. Speciically,
they proposed a CFO compensation algorithm adopting linear itting for received up-chirps to mitigate the
noise’s randomness, and derived a conventional de-convolution operation-less link signature extraction scheme.
However, SLoRa is insensitive to drift caused by weather and environmental conditions. Gu et al. [57] proposed a
synchronization-free data timestamping approach based on the signal arrival time at the gateway due to the star
topology of LoRa networks rather than multi-hop and high time accuracy requirements (µs level). This approach is
vulnerable to the frame delay attack, so they designed a LoRaTS gateway to track the natural frequency deviation
of the nodes based on the linear regression and least-squares methods. Further, they proposed a Pseudorandom
Interval Hopping scheme to prevent from zero frequency bias attacks to maintain security. Besides, several
key management schemes [60, 78] focus on the LoRaWAN key derivation, distribution, update, and destruction
process to prevent from side-channel attacks.
Overall, the existing vulnerabilities and countermeasures complement each other to make a signiicant contri-
bution to the security of LoRa networks. In addition to attacks common to networks, attacks speciic to LoRa
speciic networks are an open and trending research area, coupled with the combination with cross-technology.
Covert channels [69, 137] have been proven as threatening attacks to LoRa networks, where there are no efec-
tive countermeasures against EMLoRa [137]. The existing countermeasures mainly focus on the packet header
information [79, 110, 153], passive traic analysis [7, 28], or authentication system [57, 163] to improve the safety
of LoRa networks. However, the efectiveness and energy eiciency of such security defense mechanisms are key
factors that need attention.

ACM Trans. Sensor Netw.

111:24 • Sun et al.

5.2 PHY Layer Security

PHY layer security methods essentially exploit the physical characteristics of the wireless channel to achieve
secure transmission. The traditional secure key establishment between two parties can be completed by Public
Key Cryptography (PKC), but PKC schemes require a Public Key Infrastructure (PKI) and are computationally
expensive. Another solution is the Pre-Shared Key (PSK) scheme, but it lacks scalability [178]. Thus, plenty of
methods [50, 138] consider PHY layer features as a supplement or replacement of the upper layer cryptography
method for the consideration of network security. We divided them into two categories: PHY layer key generation
and Radio Frequency Fingerprinting Identiication (RFFI) methods. Table 10 summarizes the existing LoRa PHY
layer security methods.
Table 10. Summary of LoRa PHY layer security methods
(BCH: BoseśChaudhuriśHocquenghem, SHA: Secure Hash Algorithm).
Method Year Methodology Performance
Feature Quantization Reconciliation Privacy Ampliication KGR (bit/s) KAR(%)
Key Generation

LoRa-Key [175] 2018 RSSI Multilevel Compressive sensing SHA 18ś31 98ś100
Zhang et al. [187] 2018 RSSI Diferential Secure sketch Hash function - 95ś96
Ruotsalainen et al. [133] 2019 RSSI Threshold Secure sketch with BCH code SHA-256 - 71ś85
Gao et al. [50] 2021 RSSIr Multilevel Compressive sensing SHA-256 13.8 86
LoRa-LiSK [76] 2021 RSSI Multilevel BCH code SHA-256 - 80ś90
Vehicle-Key [182] 2022 arRSSI Multilevel Autoencoder SHA-256 14ś16 98ś99
Input Learning Model Supervision Manner Accuracy(%)
59ś99 (identical vendors),
Robyns et al. [132] 2017 Signal sample MLP, CNN Supervised, zero-shot
99ś100 (distinct vendors)
RFFI

Jiang et al. [75] 2019 Diferential constellation trace Clustering of Euclidean distance Unsupervised 63ś99
SLoRa [163] 2020 CFO and spatial-temporal link signature SVM Supervised 97 (indoor), 90(outdoor)
DeepLoRa [4] 2021 IQ, amplitude-phase, and spectrogram CNN, RNN-LSTM Supervised 89(RNN), 99 (2D CNN)
Shen et al. [138] 2021 IQ, amplitude-phase, and spectrogram CNN Supervised 98

5.2.1 Key Generation Methods. Key generation, also called key agreement or establishment, refers to the process
of generating the same cryptographic key based on the PHY layer characteristic parameters (e.g., RSSI, CSI, phase)
of the wireless channel through common channels between two legitimate parties that have no prior secret. Its
feasibility is mainly due to the characteristics of channel reciprocity, spatial variation, and temporal variation,
which ensure the uniqueness and randomness of the generating keys. Channel reciprocity means that the channel
characteristics between two communication nodes are almost identical. Spatial and temporal variation mean that
the radio channels between two nodes vary with the location and the environment across time.
Key generation is a promising technique to maintain secure communications for LoRa nodes recently [76,
133]. It generally includes the following four stages: channel probing, quantization, reconciliation, and privacy
ampliication. Two legitimate nodes send messages end-to-end, and measure some kinds of channel features,
mainly RSSI information for LoRa. The measured value is then converted into a string of key bits using diferent
quantization methods. Reconciliation is for discarding or correcting the bit diferences, and privacy ampliication
is designed for handling information leakage issues to the attacker. The feasibility of key generation of LoRa was
irst veriied in [174]. Zhang et al. [187] proposed a diferential quantization method, which discards the RSSI
value whose variation with the adjacent value is smaller than the set RSSI resolution against the measurement
imperfection. LoRa-Key [175] is the irst RSSI-based key generation method for LoRa, which employs several
signal processing techniques (e.g., outlier detection, linear interpolation) to improve the key generation rate, and
a novel compressive sensing-based reconciliation approach to reduce the key disagreement rate. Rather than RSSI,
Gao et al. [50] employed RegRssiValue (RSSIr ) provided by LoRa transceivers for LoRa key generation. Speciically,
RSSIr is the raw instantaneous strength estimation, which provides better channel estimation compared with
RSSI and whose distributions produced by the legitimate parties are extremely similar. Furthermore, they adopted
a random waypoint model to derive an optimal window size, which can balance the channel reciprocity and

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:25

entropy. To further improve the correlation of channel measurements, Yang et al. [182] exploited the mean value
of adjacent RSSIr (arRSSI) as a novel feature for key generation in Internet of Vehicles (IoV) scenarios. They
also proposed a Bi-LSTM model for prediction and an autoencoder-based reconciliation method for mismatch
correction.
PHY layer key generation for LoRa could maintain the network security, without the need for ixed infras-
tructure or secure communication channels. Also, it requires less storage and computing power compared to
asymmetric cryptographic solutions. However, due to the characteristics of the low data rate and high energy
eiciency of LoRa technology, the issues of low channel reciprocity and probe packet exchanging make key
generation for LoRa challenging. Besides, spatial and temporal variations of the radio channel require multiple
key establishment times for the LoRa nodes, which is not a long-term security mechanism. Additionally, current
key generation methods are almost designed for two legitimate parties [50, 175], group key generation among a
large number of nodes is still an open problem [178].

5.2.2 Radio Frequency Fingerprinting Identification Methods. Device identiication is essential for IoT security
to allow legitimate users to access the network while preventing malicious users. Recently, the emerging PHY
layer RFFI technique uses features extracted from radio signals to uniquely identify devices. Its essence is to
use the inherent tiny defects (e.g., inphase and quadrature imbalance, frequency/sampling ofset) in the analog
circuit of radio device hardware to generate a unique ingerprint of this device, which is impossible to imitate by
adversarial devices. As hardware defects are interrelated and complex, hand-crafted low-dimensional features are
often unable to generate distinguishable and high-level ingerprints efectively, machine learning algorithms are
utilized to make up for this issue. RFFI generally includes two stages, namely training and classiication. In the
training stage, the trainer performs feature extraction (e.g., IQ, CFO, spectrogram) after collecting enough data
packets. Then, these features are fed into the classiier for training. While in classiication one, after receiving the
data packet and feature extraction, the classiier infers the identity of the device.
As the pioneering work of LoRa PHY layer ingerprinting, Robyns et al. [132] proposed two per-symbol
supervised machine learning models, i.e., a Multilayer Perceptron (MLP) and a Convolutional Neural Network
(CNN). These models process the entire signal instead of low-dimensional features of the local one to distinguish
devices from diferent manufacturers. Besides, a zero-shot learning model was proposed to consider the unknown
device cases. Diferently, Jiang et al. [75] adopts the diferential constellation trace igure as the feature, and
utilizes an unsupervised method based on the Euclidean distance comparison. Gu et al. [57] designed a LoRaTS
gateway to track the radio frequency biases of the nodes based on the linear regression and least-squares methods.
Wang et al. [163] proposed SLoRa, an authentication method using ine-grained CFO and spatial-temporal link
signature of diferent nodes’ positions. SLoRa collects and inputs CFO and signature features into a SVM model for
training in the oline stage. Then, such features from the new node are inputted into the model for authentication
in the online one. Al-Shawabka et al. [4] irst collected a dataset containing LoRa waveform data from 100
bit-similar devices, and then proposed a deep learning-based data augmentation technique termed DeepLoRa.
Speciically, DeepLoRa focuses on three diferent representations, i.e., IQ, amplitude-phase, and spectrogram
of the signal preamble or payload, as the input of the deep learning models. Besides, DeepLoRa generated and
applied inite input response (FIR) ilter taps to transform the original dataset, which acts as a data augmentation
technique and increases channel diversity. Shen et al. [138] proposed a CNN taking the IQ, amplitude-phase, and
spectrogram of LoRa signal as input, among which spectrogram achieves the best performance. Besides, they
revealed that the drift of the instantaneous CFO will afect the classiication results and degrade system stability.
To this end, they proposed a CFO estimation and compensation algorithm, where a CFO database was generated
to help the hybrid classiier use the estimated CFO to calibrate CNN’s softmax output.
In general, RFFI methods rely on the hardware imperfection, where the derived ingerprint is unique to
maintain the network security preventing from spooing or counterfeiting attacks. Compared with traditional

ACM Trans. Sensor Netw.

111:26 • Sun et al.

cryptography-based security solutions, RFFI methods do not impose additional computational burdens on devices
for authentication. The existing RFFI methods [75, 138, 163] have achieved more than 90% identiication accuracy
due to the representation learning capability of deep learning models. However, they are not friendly to newly
join-request legitimate devices, resulting in poor scalability. In addition, since most RFFI methods seek stable and
discriminative features from the entire LoRa signal samples, using instantaneous features is still challenging.
Apart from the aforementioned studies, several other studies [67, 90, 113] focused on blockchain-enabled LoRa
networks for trust veriication and security issues. Blockchain is a Peer-to-Peer (P2P) distributed and decentralized
ledger technology. Its essence is a shared database that contains speciic and veriiable records of each transaction,
possessing the characteristics of non-counterfeiting, traceable, and transparent. Lin et al. [90] proposed the irst
conceptual blockchain-enabled LoRaWAN infrastructure design, which is built upon many LoRaWAN network
servers that communicate with each other via P2P. Each network server is added the blockchain management
components of packaging transaction, hashing broadcasting, veriication, making and storing blocks, to perform
the message low process. Niya et al. [113] proposed a blockchain-enabled LoRaWAN network based on Ethereum,
an open-source, public blockchain platform supporting smart contracts to store data. Speciically, Ethereum
Light Clients (ELCs) were deployed in the LoRa nodes or gateways for the data transmission to the application
server. Hou et al. [67] proposed HyperLoRa, a blockchain-enabled LoRa system with edge computing ability. In
particular, HyperLoRa possesses two ledgers in the could and getaway to process the delay-tolerant application
data with large size and the time-critical network data with small size, respectively. Besides, HyperLoRa utilizes
edge computing technology to migrate the works of the join procedure and application packages processing
from the network servers. Blockchain-enabled LoRa methods mitigate security risks and solve authentication
problems, but are still immature.

6 LORA-ENABLED APPLICATIONS
The wide deployment of LoRa has inspired a wide range of applications. In this section, we review these
applications from four categories: backscatter, sensing, integration with heterogeneous wireless technologies, and
other applications2 . Speciically, backscatter refers to the passive relection and modulation of incident RF signals
for transmission. LoRa sensing captures the LoRa signal variance during propagation to achieve speciic kinds
of task sensing, such as respiration monitoring and localization. The integration with heterogeneous wireless
technologies aims to explore their interoperability, which involves wireless co-existence (WCE) and cross-
technology communication (CTC). Other applications such as smart city, industry, agriculture, and healthcare,
are reviewed at the end of this section.

6.1 Backscater
Backscatter has been widely used in long-distance, low-cost communication systems such as Radio Frequency
Identiication (RFID) tags and commodity Wi-Fi access points. As the representative example, a RFID system
typically relies on readers and tags, where the reader transmits high-power RF signals as queries and the tag
responds by changing the antennas’ impedance. Another typical backscatter communication system is the
ambient one, which does not require a dedicated signal source but explores the available RF signals nearby (e.g.,
RF or LoRa signals from active nodes), thus becoming the most energy-eicient and the lowest cost solution
among them.
LoRa backscatter [73, 141, 160] is becoming a promising technology recently, due to its high sensitivity and
resilience against both in-band and out-of-band interference, Typically, a LoRa backscatter system consists of
three parts: transmitters, receivers, and backscatter tags. When the transmitter sends excitation LoRa signals, a
2 Inspiredby the LoRa survey [85], we continuously adopt łapplicationž as the classiication criterion of the references about backscatter,
sensing, and integration with heterogeneous wireless technologies.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:27

backscatter tag uses the on-board circuit (e.g., programmable logic units) to modulate RF signal (e.g., amplitude,
frequency, phase) under a modulation mechanism (e.g., OOK), and then relects the signal. The receiver captures
the relected backscatter signal and decodes the information. Table 11 summarizes the existing LoRa backscatters.

Table 11. Summary of LoRa backscaters.

Method Year Excitation Signal Range Consumption Parallel Decoding COTS Tag Compatibility
LoRa Backscatter [148] 2017 Dedicated 2.800 m 9.25 µW % %
LoREA [160] 2017 Dedicated 3.400 m 70 µW % "
PLoRa [119] 2018 Ambient 1.100 m 220 µW % %
NetScatter [64] 2019 Dedicated - 45.2 µW " %
Aloba [59] 2020 Ambient 250 m 0.3 mW " "
PolarScatter [141] 2020 Dedicated 1.500 m 71 µW % %
300 ft (LOS),
FD LoRa Backscatter[77] 2021 Dedicated 3.04 W % "
4.000 ft (NLOS)
P2 LoRa [74] 2021 Ambient 2.200 m 320 µW " "

As the pioneering work, Talla et al. [148] proposed LoRa Backscatter, which receives and utilizes the single
tone transmitted by a single RF source to synthesize the CSS signal for decoding of the receiver. Speciically,
LoRa Backscatter adopts a hybrid digital-analog backscatter design which uses the digital domain to create a
frequency plan of a Voltage-Controlled Oscillator (VCO) for the continuously varying CSS signal, and then map
it to the analog domain via a converter. In addition, a harmonic cancellation mechanism was also proposed to
improve spectral eiciency. Hessar et al. [64] proposed Netscatter to decode large-scale concurrent backscatter
transmissions by using only one single FFT operation. In particular, they introduced a distributed CSS coding
mechanism, where each concurrent node is assigned to a diferent chirp cyclic shift and utilizes OOK to transmit
bits. They also considered the Near-far problem using power-aware and power-adaption methods and leaves gaps
in cyclic shifts to be more robust in time synchronization among nodes. Instead of using dedicated single-tone RF
as external excitation signals in [64, 148], Peng et al. [119] proposed Passive LoRa (PLoRa), which modulates an
ambient LoRa signal into a new chirp signal and shifts it into a diferent channel. Speciically, PLoRa is composed of
a low-power packet detection circuit, a blind chirp modulation algorithm, and an energy management component.
The packet detection circuit aims to reduce the sampling rate of the input signal and perform cross-correlation
operation between input signals and the pre-stored preambles. The modulation algorithm is to generate FSK
modulated baseband signal and multiply it with incoming LoRa chirp. To disentangle and demodulate the weak
backscatter signal from the strong excitation signal, Guo et al. [59] proposed Aloba, which irst detects ambient
LoRa excitation signal using unique RSS pattern from other irrelevant signals or noise, then utilizes OOK to
modulate the data and relects to the receiver. The receiver decodes the carrier signal by leveraging the capture
efect and transforms it into a constant sinusoidal tone, so that the backscatter signal can be demodulated via
tracking the amplitude and phase variation. To achieve ubiquitous backscatter connectivity, Katanbaf et al. [77]
designed the irst Full-Duplex (FD) LoRa backscatter reader. FD LoRa Backscatter consists of a single antenna
hybrid coupler along with a two-stage tunable impedance network for carrier and ofset cancellation, and a
microcontroller for implementing the adaptive tuning algorithm. Since PLoRa and Aloba can decode a small
number of concurrent backsacatter packets, Jiang et al. [74] proposed an ambient Passive and Parallel LoRa
backscatter design termed P2 LoRa. Speciically, P2 LoRa shifts the ambient LoRa signal at a small certain frequency
to modulate the data in the backscatter signal, and concentrates the leaked energy in the frequency and time
domain to improve its SNR. Then it utilizes a two-level parameter estimation method to reconstruct and eliminate
the in-band excitation signal accurately, and adopts a window-based method to eliminate the interference to
perform parallel decoding.

ACM Trans. Sensor Netw.

111:28 • Sun et al.

So far, current LoRa backscatter systems have signiicantly improved the communication range and throughput
of LoRa networks at a low energy cost. Ambient excitation signals [59, 74, 119] get rid of the limitation of dedicated
excitation signal source, low-power packet detection approaches [74, 119] achieve long-term deployment, parallel
decoding [59, 64, 74] enables simultaneous communications among multiple nodes. However, the existing
LoRa backscatter designs still face some challenges in terms of high-range packet detection, simple backscatter
modulation mechanism at the tag side, out-of-band LoRa signal interference, and compatibility with COTS
devices.

6.2 Sensing
Wireless sensing is an emerging technology of acquiring information about a remote object and its characteristics
using ambient wireless signals. The rationale behind wireless sensing is to capture the signal variance (e.g.,
phase, RSSI) of the wireless signal itself (e.g., RFID, Wi-Fi, LoRa) relected from the targets with some speciic
movements. Compared with transitional wireless signals, LoRa possesses strong penetration capability to perform
through-wall sensing tasks and long sensing range to ill the gap of Wi-Fi (3ś6 m), RF (3ś6 m), and acoustic (less
than 1 m) signals [185]. Thus, a large number of LoRa sensing methods [92, 185] were proposed recently. We
review the existing LoRa sensing works, including respiration monitoring and localization.

Table 12. Summary of LoRa respiration monitoring methods (RPM: Respiration Per Minute).
Method Year Sensing Model Methodology Antenna Array Through-wall #target Range Metric
Phase diference of signal ratio Mean absolute error:
Zhang et al. [185] 2020 Modeling and quantifying 2 directional " 1 25 m
of received signals at two antennas 0.1-0.37 RPM (5ś25 m)
Phase diference of signal ratio Average error:
Sen-fence [170] 2020 Virtual fence, search algorithm 2 directional " 1ś3 50 m
of received signals at two antennas 0.2ś0.7 RPM (1ś3 interferer)
Phase diference of signal ratio Search algorithm, Average absolute error:
Xie et al. [171] 2021 3 directional " 1 75 m
of received signals at two antennas time-domain beamforming 0.1ś0.6 RPM (1ś7 wall)
Phase diference of OOK-based modulated LoRa backscatter 1 omni- Median deviation:
Palantir [73] 2021 % 1 100 m
backscatter and direct path signal (signal shaping, clustering) directional 4.39ś11.65% (10ś100 m)
Dividing beamforming signal by Beamforming, direction-frequency Accuracy:
Zhang et al. [186] 2021 beam-nulled one, phase of dynamic spectrogram pre-processing, 4 directional " 2-5 24 m 98.12ś99.75% (1ś5 targets, 10 m),
path signal mapping distance multi-target detection 96.46ś99.62% (8ś24 m, 2 targets)

6.2.1 Respiration Monitoring. Studies on LoRa-based respiration monitoring generally model the propagation
and relection of the LoRa signal, then adopts a series of signal processing techniques to quantify a mapping
between speciic signal variation (e.g., phase, amplitude) and the sensing goal (e.g., distance). Table 12 summarizes
the existing LoRa respiration monitoring methods.
Zhang et al. [185] achieved real-time LoRa sensing for the irst time, including through-wall ine-grained
respiration and coarse-grained walking monitoring in the range of 25 m. Speciically, they irst derived a basic
LoRa signal propagation model, and utilized the ratio of received signals at two antennas to cancel out noise
and eliminate the random signal phase shift. They further quantiied the phase change of the signal ratio to
capture the relationship with distances corresponding to target movements for sensing. Besides, they have
conducted a comprehensive experiment concerning the impacts of the target, distance, and environment. LoRa
radio sufers surrounding interference along with its long-distance communication, Xie et al. [170] proposed
Sen-fence, which restricts interference from outside the created virtual fence to mitigate it. In particular, Sen-fence
irst forms a beam-shaped or spot-shaped virtual fence depending on the number of receiver, then maximizes
the movement-induced phase variation by adding a newly static signal purely in software. Such static signal
needs to meet the requirements of tuning the phase diference between the static and dynamic vectors to 180°
and minimizing the amplitude of the composite signal. Finally, Sen-fence conines the phase within the virtual
fence based on the search algorithm of the optimal static signal. However, Sen-fence requires the sensing area as
prior knowledge and multiple receivers, which limits sensing in mobile scenarios. In their further study [171],

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:29

they increased the respiration sensing range to 75m, resulting from enlarging the target-induced phase variation
and adopting a time-domain beamforming method combining signals with diferent timestamps to increase SNR.
Jiang et al. [73] provided a cyclist respiration sensing method relying on OOK modulated LoRa backscatter
signal termed Palantir, which consists of four stages of preprocessing, signal shaping, clustering, and sensing.
Speciically, Palantir irst demodulated the received direct path signal and backscatter one by exploiting the
capture efect. Then, Palantir performed signal shaping to obtain the stable state samples, including amplitude
shaping by leveraging a low-pass ilter to resolve the problem of amplitude instability and baseband removal by
conjugate demodulation and curve itting to eliminate ofset and drift. Then Palantir adopts the dual clustering
method to transfer the state samples from the I-Q coordinate system to the logarithm of the amplitude-phase
coordinate system, and performs cluster identiication enabling consistent identiication even if there is a global
mismatch, which resolves the challenge of spectrum leakage. Finally, Palantir achieves sensing based on the
derived phase diference between the vectors of the direct path signal and the backscatter one. To enable multi-
target sensing, Zhang et al. [186] proposed a LoRa multi-antenna beamforming technology to separate the
signals in the space domain. In particular, they constructed a łbeam nullingž signal as a reference and divided
beamforming signal of diferent directions with this beam-nulled signal, which can eliminate impacts of CFO
and Sampling Frequency Ofset (SFO) for synchronization-free between transceivers and not corrupt the signal
amplitude or phase variation information. Moreover, they utilized the amount of phase rotation of the location-
independent dynamic path signal rather than the composite signal to determine the chest displacement, thereby
solving the location-dependent issue of the composite signal for respiration monitoring. Apart from respiration
monitoring, the aforementioned methods also achieved walking [171, 185, 186] and gesture tracking [170].
In general, LoRa has greatly compensated for the short sensing range of conventional wireless signals, and
achieved great improvements. The existing methods [170, 185] mainly focus on the signal modeling and processing
to capture the phase diference among multiple antennas to achieve ine-grained respiration monitoring sensing.
Although backscatter signal [73] and beamforming technique [171, 186] provide new possibilities on range
expansion and multi-target sensing, there still remain some challenges:
• Range. Sensing is more susceptible to channel quality compared with communication [73]. With range
expansion, LoRa sensing systems will inevitably sufer from various and complex interference, remaining
challenging to be resolved.
• Multi-target. Signals from multiple targets will be interleaved and superposed, so a higher complexity
of the receiving antenna array and a more robust sensing model are required for sensing. Besides, target
mobility will bring additional challenges.
6.2.2 Localization. Target localization and tracking techniques primarily focus on four categories of information:
Angle of Arrival (AoA) [92], Time Diference of Arrival (TDoA) [10], RSSI [91], and amplitude [24]. AoA-based
methods leverage the phase diference of a signal arriving at multiple antennas for localization. The multipath
can be efectively separated, but good resolution and accuracy require large-scale antenna arrays at the receiver
side. TDoA-based methods achieved localization using the time diference of the same signal arriving at multiple
gateways. High resolution can be achieved, but sufers from the measurement error and the limited bandwidth.
AoA and TDoA methods generally require clock synchronization between transceivers. RSSI-based methods can
be further divided into ingerprinting-based methods [91] and model-based methods [12, 82]. They are generally
efective, but have poor resolution. Besides, the amplitude is utilized to deal with the multipath efect, which is
simple and low-cost, but has poor accuracy due to signal attenuation. Table 13 summarizes the existing LoRa
localization methods.
Chen et al. [24] proposed a localization prototype termed WIDESEE, which consists of a reconigurable
antenna system, a data collection and antenna control system, and a target detection and localization system.
Speciically, the antenna system integrates horn directional antennas and phased arrays for fast radiation mode

ACM Trans. Sensor Netw.

111:30 • Sun et al.

Table 13. Summary of LoRa localization methods.

Method Year Technique Antenna Array Synchronization Latency Range Error COTS Compatibility
1.5 m (outdoor),
µLocate [111] 2018 Backscatter 3 chip antennas Access Points 25ś70 ms 60 m %
0.3 m (indoor)
53 m (outdoor),
WIDESEE [24] 2019 Amplitude-based 1 reconigurable directional - - 4.6 m %
20×42×85 m3 (indoor)
SateLoc [91] 2021 RSSI-based 1 omnidirectional - 3s 350×650 m2 47.1 m "
OwLL [10] 2021 TDoA-based Virtual multi-antenna arrays Base stations 20.97 s >500 m 9m %
100 m (outdoor), 4.6 m (outdoor),
Seirios [92] 2021 AoA-based 2×2 MIMO Multiple Channels 0.24 s "
25 m (indoor) 2.4 m (indoor)

switching and narrower beamwidth ofering to further reduce interference. The data collection and antenna
control system is built on single LoRa transceiver pair carried by a lying drone. WIDESEE exploits the power
spectrum density (PSD) for target detection after vibration noise elimination using a low-pass ilter, then extracts
direction-related information from amplitude and isolates the target path from the interfering multipath to
achieve localization sensing. Lin et al. [91] proposed SateLoc, a LoRa localization system based on the virtual
ingerprints extracted from satellite images. Speciically, in the oline stage, SateLoc trains a Random Forest (RF)
using satellite images associated with labeled land-cover types, to generate an Expected Signal Power (ESP) map
as a virtual ingerprinting for each gateway. In the online one, SateLoc produces a location likelihood distribution
for each gateway based on its ESP map using the extracted RSSI and SNR from the received packets, and adopts
a weighted combination strategy for joint localization. Besides, µLocate [111] designed a sub-centimeter sized
multi-band backscatter system, together with the extracted phase information for microwatt-level localization.
As the narrow bandwidth of LoRa incurs low-range localization resolution, several methods [10, 92] focused
on the bandwidth expansion. Bansal et al. [10] proposed an Outdoor whitespace-band LoRa Localization (OwLL)
method based on the TDoA transformed by the measured phase diference across antennas. Speciically, OwLL
emulates wide bandwidth in a low-cost manner by frequency hopping over wireless spectrum in TV whitespace
and ISM frequency bands, where an iterative maximum-likelihood algorithm is adopted to determine a small set
of optimal frequencies hopped rather than all possible ones. After ensuring phase synchronization with reference
to Chime [47], OwLL treats diverse spatial base stations as a virtual distributed array to mitigate the impact
of signal multipath, and utilizes a particle ilter as well as the prior measured phase and TDoA to trilaterate
LoRa clients. Liu et al. [92] proposed a super-resolution localization algorithm termed Seirios. In particular,
Seirios utilizes a novel interchannel synchronization algorithm to increase the bandwidth, where ToneTrack [172]
technique is for overlapped channels and a virtual intermediate channel response is generated as a bridge is for
non-overlapped ones. Then, Seirios exploits both the original and the conjugate of the CSI measurements for AoA
estimation based on the ES-PRIT algorithm [70] to further increase the capacities for multipath resolution, where
the synchronized LoRa CSI value of multiple channels is obtained through the amplitude and phase comparison
between the received symbols with the pre-deined training ones.
In general, LoRa localization methods have achieved promising accuracy and resolution results, and greatly
expanded the range compared with Wi-Fi and RFID signals. The existing localization methods have attempted
various features, such as amplitude [24], RSSI [91], angle [92], and time [10]. Among which, RSSI can achieve the
longest range, but amplitude, time, and angle information can obtain higher resolution. Apart from attempting
various features to expand the range, several studies [10, 92] focus on the bandwidth expansion of LoRa to
improve the resolution. However, some possible challenges may include:
• Synchronization-Free and Multipath Disambiguation. The synchronization between transceivers
guarantees the accuracy of AoA and TDoA-based methods, resulting in additional energy consumption.
Multipath disambiguation addresses the problem of signal change in terms of the polarization mode, phase,
and Doppler shift for better accuracy.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:31

• Multi-target Sensing. Multi-target relected signals will be inter-weaved, thus separating these signals
under a limited channel bandwidth is challenging. Besides, concurrent transmission for multi-target
localization deserves further exploration.
• Mobile Localization. Moving targets or gateways induce the complex multipath efect and signal attenu-
ation, which deserves further study.

6.3 Wireless Co-Existence & Cross-Technology Communication

The emerging paradigm of IoT has inspired the exploration of interoperability between LoRa and other heteroge-
neous wireless technologies. These works can be divided into two categories: wireless co-existence (WCE) and
cross-technology communication (CTC). Table 14 summarizes the existing LoRa WCE and CTC methods.
Table 14. Summary of LoRa WCE and CTC methods.
Method Year Heterogeneous Technology Methodology Performance Hardware Design COTS Compatibility
Zhang et al. [188] 2018 LoRa & NB-IoT Multi-module node design Range: 1.6 km " %
WCE

Gao et al. [51] 2019 LoRa & NB-IoT Multi-module node design - " %
LoFi [23] 2021 LoRa & Wi-Fi Spectrum reservation PRR: 98% % "
PSR [143] 2021 LoRa & CT interference Symbol recovery PRR: 45.2% → 82.2% % "
Symphony [88] 2019 BLE and ZigBee → LoRa GFSK and OQPSK modulation Range: >500 m % "
CTC

Accuracy: >97%
XFi [95] 2020 ZigBee, LoRa → Wi-Fi Signal hitchhiking % "
under 8 parallel devices
EMLoRa [137] 2021 EMG → LoRa AM modulation Range: 130 m % "
LoRaBee [139] 2021 LoRa → ZigBee Correlation with payload data and RSS value Throughput: 281.61 bps % "

Wireless Co-Existence Methods. WCE methods mainly rely on interference avoidance, detection, and
cancellation to achieve the co-existence of diferent technologies. Recently, a new LoRa chip called SX1280 was
proposed by Semtech provides a possibility of LoRa radio operating at 2.4 GHz frequency, resulting in a larger
available bandwidth (from 500 to 1600 kHz) and a faster data rate (from 21 to 70 kbps). In this context, LoRa
packets may be severely damaged by Wi-Fi interference, so Chen et al. [23] proposed a weak signal detection
method termed LoFi to achieve the co-existence of LoRa and Wi-Fi. Inspired by the physical phenomenon named
Stochastic Resonance (SR), LoFi adds appropriate white noise that is capable of enhancing weak LoRa signals at
speciic frequencies. Speciically, LoFi irst selectively transforms the frequency of a LoRa chirp into one particular
small frequency, and separates Wi-Fi signal to another frequency range to detect LoRa signals. Then, LoFi adopts
a bandwidth-aware spectrum reservation method to adaptively reserve the spectrum for LoRa collision-free
transmission according to the spectrum occupancy. Sun et al. [143] proposed a Partial Symbol Recovery (PSR)
scheme to combat CT interference, including Wi-Fi, ZigBee, and Bluetooth. In the coarse-grained localization
stage, PSR performs the maximum pooling and calculates the ratio between the dominant frequency component
and the average after Short Time Fourier Transform (STFT) operation. PSR identiies and recovers LoRa symbols
based on this ratio in the ine-grained detection one. Besides, several methods [51, 188] paid attention to the
combination of NB-IoT and LoRa technology. Zhang et al. [188] proposed an information monitoring system
integrating NB-IoT and LoRa, which mainly relies on the designed main-nodes equipped with both radio modules
to receive messages from LoRa sub-nodes and transmit them to the cloud server.
Cross-Technology Communication Methods. CTC mainly refers to exchanging instructions and data low
across two or more diferent technologies (i.e., the carrier and target ones), which exploits both (or all) of their
advantages. Shi et al. [139] proposed LoRaBee to support CTC from LoRa to ZigBee unidirectional. In LoRaBee,
the LoRa signal is adopted as the carrier with elaborate frequency tuning and payload encoding, and Zigbee can
decode the packet by sampling the RSS value. Speciically, the main idea is to correlate (i.e., map) the data bytes
in the LoRa payload with the generated RSS signature between the LoRa transmitter and ZigBee receiver, and
make both sides store this mapping. However, it is computation ineicient and lacks scalability. Li et al. [88]
proposed Symphony to support the CTC from Bluetooth Low Energy (BLE) and ZigBee to LoRa, along with a

ACM Trans. Sensor Netw.

111:32 • Sun et al.

parallel decoding method. Speciically, BLE, ZigBee, and LoRa signals are modulated by Gaussian Frequency
Shift Keying (GFSK), Ofset Quadrature Phase Shift Keying (OQPSK), and CSS to generate single-tone sinusoidal
signals, respectively. The received samples are split, multiplied with correlation templates, then disentangled
and decoded via performing FFT operation at the LoRa receiver side in turn. Shen et al. [137] proposed EMLoRa,
which reshapes EMG radiation into LoRa-like chirps through AM. Liu et al. [95] presented XFi based on signal
hitchhiking. Low-speed IoT (ZigBee, LoRa) data packets collide with (hitchhike) high-speed Wi-Fi ones in the
overlapped spectrum, which can then be decoded by commodity Wi-Fi receivers. Speciically, XFi irst reconstructs
the IoT waveform with erased segments by analyzing the Wi-Fi payload, and then utilizes the enhanced IoT
decoder to reliably decode the reconstructed waveform through incorporating the signal erasure pattern with the
IoT signal redundancy information.
Current emerging WCE and CTC methods adapt to the unprecedented proliferation of heterogeneous wireless
devices, and make up for the shortcomings of another single technology by exploiting the advantages of the
considerable communication range of LoRa. The existing WCE [23, 143] and CTC [88, 137] methods mainly focus
on interference avoidance and signal approximation to the receiver waveform methodologies. Additionally, RSS
mapping [139] and signal hitchhiking [95] provide novel solutions. However, these methods still have a long way
in hardware design, network deployment, generality for other technologies, and security issues.
Apart from the aforementioned LoRa-enabled applications, several methods focus on others, such as network
aggregation [48, 183], C-RAN [39, 93]. For example, LoRa network aggregation [48, 183], information retrieving
on the basis of the selection and analysis of data in the network, receives popularity recently. Yang et al. [183]
proposed an accurate, general, future-proof, and energy-eicient analytic framework for LPWAN termed Joltik.
Joltik can calculate sensor data aggregation, utilize universal sketching for transmission, and support unforeseen
metrics without additional energy overhead. Speciically, Joltik is built on universal sketching, which discreetly
provides a smaller number of counters at the lower level for eicient storage, employs a compression scheme for
reducing the communication cost, and eliminates updates of redundant counters for reducing the computational
cost. Gadre et al. [48] proposed QuAiL, which enables coarse aggregation queries of sensing data across LPWAN
(LoRa, NB-IoT) clients within one packet timespan, inducing spatial distribution, statistics, and machine learning
types. QuAiL mainly relies on encoding the information in the energy of concurrent transmissions across clients,
and leverages this linear addition of powers of phase-asynchronous channels for diferent types of queries. Besides,
QuAiL requires wireless impairment tackling (e.g., timing and frequency ofsets, noise) for synchronization, and
involves security and privacy considerations using random power weights of clients.

6.4 Other Applications

As LoRa networks possess the advantages of low-power consumption and wide coverage, they have been widely
used in various real-world IoT applications, summarized as below:

• Smart City. Smart building and lighting [118, 176], public assets tracking [35], surveillance system [136],
are deployed to better serve for the city life.
• Smart Industry. Industry applications, such as smart grids [34], smart metering [161], and distributed
measurement system [130], also receives much popularity.
• Smart Agriculture. LoRa is widely used in agriculture area, such as soil health monitoring [125], smart
irrigation system [189], and precision agriculture [19].
• Smart Healthcare. Besides, LoRa is utilized for healthcare monitoring [21] and against COVID-19 pan-
demic [126].

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:33

7 CHALLENGES AND FUTURE DIRECTIONS

In previous sections, we have reviewed extensive works developed for LoRa networks. Below, we provide
researchers with challenges of existing works and potential future research directions in this domain.

7.1 Challenges
LoRa Analysis. Extensive analysis of LoRa performance can help understand the capabilities and limitations
of LoRa. Quantifying LoRa performance with the corresponding factors becomes an early and crucial work for
further study. LoRa analysis tools also leave room for further improvements. The existing analytical models are
mostly derived under strict assumptions. The major inluencing factors need to be explicitly represented before
combating various interference for devising accurate and general analytical models. Simulators ofer a convenient
way for experiment test and validation, but they are not full-featured enough. Developing large-scale public
testbeds is challenging concerning network deployment, numbers of LoRa devices, types of sensors, remote
development, and user experience.
LoRa Communication. The performance of LoRa communication primarily relies on its PHY layer modula-
tion and demodulation, MAC protocol, and coniguration settings of nodes. The efectiveness and eiciency are
of the primary concern for LoRa communication studies. In particular, the transceiver modiication, the compati-
bility with COTS devices, and additional energy consumption issues make improvements on LoRa modulation
challenging. For demodulation, the complexity of the algorithm, synchronization requirement, and the impact of
CFO, SFO, and inter-packet interference still require further study. It is also challenging to devise adaptive and
efective MAC protocols, due to the inlexibility of contention-based and energy waste of schedule-based ones.
Besides, diferent network deployment, various intra- and inter-network interference, dynamic link quality, and
algorithm complexity are challenging for coniguration settings methods.
LoRa Security. Security and privacy are vital with the surging growth of LoRa networks, on the premise
that LoRa is susceptible to various vulnerabilities. LoRa PHY properties have shed light on novel and powerful
attacks that are diicult to combat, while the high power eiciency requirement also makes its countermeasures
challenging. Additionally, although PHY layer security methods can maintain theoretically absolute security,
non-robust characteristics limit the practicality. For example, the existing key generation methods generally
allow occurring only two legitimate parties in a long-term probing period but not group ones. Discriminative
instantaneous features and suitability for newly join-request legitimate devices are challenging for RFFI methods.
LoRa-Enabled Applications. Beyond the scope of LoRa networks, plenty of works have attempted various
LoRa-enabled application designs inclusive of backscatter, sensing, heterogeneous technologies, etc. These works
have made promising progress and shown great potential in the research community, but also deserve further
much-room exploration. For backscatter, ambient excitation signals, concurrent transmission capability, and
demodulating weak backscatter signals from the strong superposed excitation signal are valuable and challenging
factors. The major challenges of wireless sensing lie in efective feature extraction, robust signal model derivation,
and comprehensive data collection. The generality is a long-term challenge for integration with heterogeneous
wireless technologies.

7.2 Future Directions

LoRa Protocol Stack. As an emerging LPWAN technology, various eforts have been made on the protocol
stack of LoRa, especially its PHY/MAC layers. The unique characteristics of LoRa PHY signals have been leveraged
for collision disambiguation [167], sensing [186], and backscatter [74], which undoubtedly demonstrates the
capability and potential of LoRa PHY layer. For further directions, novel modulation technology is a new way to
break through the conventional theory for transmission performance optimization. Apart from leveraging LoRa
PHY packet structure and chirp features in time and frequency domains, exploiting spatial [93] and reception

ACM Trans. Sensor Netw.

111:34 • Sun et al.

[167] diversity gain, adopting deep learning networks [86] can break through some inherent limitations of LoRa
conventional PHY decoding while ensuring the advantages of its weak and collision decoding ability. Cross-layer,
cross-device, and cross-sensor sensing can provide multi-domain knowledge fusion for better sensing [99].
Integrated Sensing and Communication (ISAC), focusing on joint-protocol design and time-frequency resource
reuse, has been a hot research topic recently. Additionally, as LoRa only deines the lower PHY layer in the
communication stack, upper network protocols can reine LoRa protocol stack with respect to MAC protocol,
data/control plane, etc.
LoRa Network. In essence, LoRa is a communication technology to form wireless sensor networks featuring
low energy consumption and long transmission distance. However, many network performance indicators,
such as throughput, communication range, energy consumption, capacity, scalability, and security, deserve
further consideration and enhancement. Apart from LoRa networking protocols, the aforementioned ones can be
improved via link coordination and adaptability, network management, etc. Link quality dramatically afects the
extent of signal propagation’s attenuation (e.g., fading, path loss). Hence, channel diversity, adaptive transmission
strategies, and opportunistic spectrum access deserve further improvements. The real-life deployed LoRa networks
often present a large-scale complex architecture, which could be multi-topology, multi-hop, or heterogeneous
networks. Therefore, network management can achieve resource allocation, load balance, device coniguration,
accounting, and security services. Among these, network aggregation [48, 183] on LoRa networks is an emerging
way for information retrieving and accounting. Besides general security mechanisms, regular irmware updating
via over-the-air for remotely deployed LoRa devices sheds light on network security.
AI-Empowered LoRa. Artiicial Intelligence (AI) plays a signiicant role in a wide range of research and
industry ields [145]. The mainstream deep learning methods can learn implicit and ine-grained feature repre-
sentations from the sample instances, render high accuracy and acceptable generalization, and avoid complex
feature engineering. Data-driven deep learning methods on LoRa signals have achieved remarkable results and
incredible purposes in signal demodulating [86], RFFI [138], and sensing [91]. Additionally, few-shot learning,
un-/semi-supervised learning [145], federal learning, and embedded AI can be applied to LoRa technology to
achieve diferent tasks. Thus, LoRa technology integration with AI is a worthy and broad future direction.

8 CONCLUSION
LoRa is a crucial and promising LPWAN technology that gains signiicant research momentum over the past
decades, thus inspiring extensive works. In this paper, we give a comprehensive review of LoRa from four aspects,
i.e., analysis, communication, security, and applications. Besides, several challenges and potential research
directions have also been discussed. Although we have reviewed nearly 200 articles in this survey, the list of
studies is far from exhaustive. Nevertheless, it covers the majority of recent achievements and directions. We hope
that this survey will make it easier for researchers to identify research gaps and discover answers. Additionally,
we would like to welcome all researchers to contribute to this intriguing ield by expanding it and providing
fresh insights.

ACKNOWLEDGMENTS
The work described in this paper was substantially sponsored by the project 62101471 supported by NSFC
and was partially supported by the Shenzhen Research Institute, City University of Hong Kong. The work
described in this paper was partially supported by a grant from the Research Grants Council of the Hong
Kong Special Administrative Region, China (Project No. CityU 21201420). The work described in this paper was
partially supported by Shenzhen Science and Technology Funding Fundamental Research Program (Project No.
2021Szvup126), NSF of Shandong Province (Project No. ZR2021LZH010), and a grant from Chow Sang Sang
Group Research Fund sponsored by Chow Sang Sang Holdings International Limited (Project No. 9229062). The

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:35

work was also partially supported by CityU MFPRC grant 9680333, CityU SIRG grant 7020057, CityU APRC grant
9610485, CityU ARG grant 9667225 and CityU SRG-Fd grant 7005666.

REFERENCES
[1] Khaled Q Abdelfadeel, Victor Cionca, and Dirk Pesch. 2018. Fair adaptive data rate allocation and power control in LoRaWAN. In 2018
IEEE 19th International Symposium on" A World of Wireless, Mobile and Multimedia Networks"(WoWMoM). IEEE, 14ś15.
[2] Khaled Q Abdelfadeel, Dimitrios Zorbas, Victor Cionca, and Dirk Pesch. 2019. F REEÐFine-Grained Scheduling for Reliable and
Energy-Eicient Data Collection in LoRaWAN. IEEE Internet of Things Journal 7, 1 (2019), 669ś683.
[3] Cedric Adjih, Emmanuel Baccelli, Eric Fleury, Gaetan Harter, Nathalie Mitton, Thomas Noel, Roger Pissard-Gibollet, Frederic Saint-
Marcel, Guillaume Schreiner, Julien Vandaele, et al. 2015. FIT IoT-LAB: A large scale open experimental IoT testbed. In 2015 IEEE 2nd
World Forum on Internet of Things (WF-IoT). IEEE, 459ś464.
[4] Amani Al-Shawabka, Philip Pietraski, Sudhir B Pattar, Francesco Restuccia, and Tommaso Melodia. 2021. DeepLoRa: Fingerprinting
LoRa Devices at Scale Through Deep Learning and Data Augmentation. In Proceedings of the Twenty-second International Symposium
on Theory, Algorithmic Foundations, and Protocol Design for Mobile Networks and Mobile Computing. 251ś260.
[5] Licia Amichi, Megumi Kaneko, Ellen Hidemi Fukuda, Nancy El Rachkidy, and Alexandre Guitton. 2020. Joint allocation strategies
of power and spreading factors with imperfect orthogonality in LoRa networks. IEEE Transactions on Communications 68, 6 (2020),
3750ś3765.
[6] Emekcan Aras, Gowri Sankar Ramachandran, Piers Lawrence, and Danny Hughes. 2017. Exploring the security vulnerabilities of LoRa.
In 2017 3rd IEEE International Conference on Cybernetics (CYBCONF). IEEE, 1ś6.
[7] Emekcan Aras, Nicolas Small, Gowri Sankar Ramachandran, Stéphane Delbruel, Wouter Joosen, and Danny Hughes. 2017. Selective
jamming of LoRaWAN using commodity hardware. In Proceedings of the 14th EAI International Conference on Mobile and Ubiquitous
Systems: Computing, Networking and Services. 363ś372.
[8] Aloÿs Augustin, Jiazi Yi, Thomas Clausen, and William Mark Townsley. 2016. A study of LoRa: Long range & low power networks for
the internet of things. Sensors 16, 9 (2016), 1466.
[9] Artur Balanuta, Nuno Pereira, Swarun Kumar, and Anthony Rowe. 2020. A cloud-optimized link layer for low-power wide-area
networks. In Proceedings of the 18th International Conference on Mobile Systems, Applications, and Services. 247ś259.
[10] Atul Bansal, Akshay Gadre, Vaibhav Singh, Anthony Rowe, Bob Iannucci, and Swarun Kumar. 2021. OwLL: Accurate LoRa Localization
using the TV Whitespaces. In Proceedings of the 20th International Conference on Information Processing in Sensor Networks (co-located
with CPS-IoT Week 2021). 148ś162.
[11] Luca Beltramelli, Aamir Mahmood, Patrik Österberg, Mikael Gidlund, Paolo Ferrari, and Emiliano Sisinni. 2021. Energy eiciency of
slotted lorawan communication with out-of-band synchronization. IEEE Transactions on Instrumentation and Measurement 70 (2021),
1ś11.
[12] Giulio Maria Bianco, Romeo Giuliano, Gaetano Marrocco, Franco Mazzenga, and Abraham Mejia-Aguilar. 2020. LoRa System for Search
and Rescue: Path-Loss Models and Procedures in Mountain Scenarios. IEEE Internet of Things Journal 8, 3 (2020), 1985ś1999.
[13] Roberto Bomin, Marwa Chaii, and Gerhard Fettweis. 2019. A novel modulation for IoT: PSK-LoRa. In 2019 IEEE 89th Vehicular
Technology Conference (VTC2019-Spring). IEEE, 1ś5.
[14] Martin Bor and Utz Roedig. 2017. LoRa transmission parameter selection. In 2017 13th International Conference on Distributed Computing
in Sensor Systems (DCOSS). IEEE, 27ś34.
[15] Martin C Bor, Utz Roedig, Thiemo Voigt, and Juan M Alonso. 2016. Do LoRa low-power wide-area networks scale?. In Proceedings of
the 19th ACM International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems. 59ś67.
[16] Taouik Bouguera, Jean-François Diouris, Jean-Jacques Chaillout, Randa Jaouadi, and Guillaume Andrieux. 2018. Energy consumption
model for sensor nodes based on LoRa and LoRaWAN. Sensors 18, 7 (2018), 2104.
[17] Ismail Butun, Nuno Pereira, and Mikael Gidlund. 2018. Analysis of LoRaWAN v1. 1 security. In Proceedings of the 4th ACM MobiHoc
Workshop on Experiences with the Design and Implementation of Smart Objects. 1ś6.
[18] Ismail Butun, Nuno Pereira, and Mikael Gidlund. 2019. Security risk analysis of LoRaWAN and future directions. Future Internet 11, 1
(2019), 3.
[19] Antonio Caruso, Stefano Chessa, Soledad Escolar, Jesús Barba, and Juan Carlos López. 2021. Collection of Data with Drones in Precision
Agriculture: Analytical Model and LoRa Case Study. IEEE Internet of Things Journal (2021).
[20] Lluís Casals, Bernat Mir, Rafael Vidal, and Carles Gomez. 2017. Modeling the energy performance of LoRaWAN. Sensors 17, 10 (2017),
2364.
[21] Philip A Catherwood, David Steele, Mike Little, Stephen McComb, and James McLaughlin. 2018. A community-based IoT personalized
wireless healthcare solution trial. IEEE journal of translational engineering in health and medicine 6 (2018), 1ś13.
[22] Marco Centenaro, Lorenzo Vangelista, Andrea Zanella, and Michele Zorzi. 2016. Long-range communications in unlicensed bands: The
rising stars in the IoT and smart city scenarios. IEEE Wireless Communications 23, 5 (2016), 60ś67.

ACM Trans. Sensor Netw.

111:36 • Sun et al.

[23] Gonglong Chen, Wei Dong, and Jiamei Lv. 2021. Loi: Enabling 2.4 ghz lora and wii coexistence by detecting extremely weak signals.
In IEEE INFOCOM 2021-IEEE Conference on Computer Communications. IEEE, 1ś10.
[24] Lili Chen, Jie Xiong, Xiaojiang Chen, Sunghoon Ivan Lee, Kai Chen, Dianhe Han, Dingyi Fang, Zhanyong Tang, and Zheng Wang. 2019.
WideSee: Towards wide-area contactless wireless sensing. In Proceedings of the 17th Conference on Embedded Networked Sensor Systems.
258ś270.
[25] Po-Yu Chen, Laksh Bhatia, Roman Kolcun, David Boyle, and Julie A McCann. 2020. Contact-Aware Opportunistic Data Forwarding in
Disconnected LoRaWAN Mobile Networks. In 2020 IEEE 40th International Conference on Distributed Computing Systems (ICDCS). IEEE,
574ś583.
[26] Qian Chen and Jiliang Wang. 2021. AlignTrack: Push the Limit of LoRa Collision Decoding. In 2021 IEEE 29th International Conference
on Network Protocols (ICNP). IEEE, 1ś11.
[27] Marco Chiani and Ahmed Elzanaty. 2019. On the LoRa modulation for IoT: Waveform properties and spectral analysis. IEEE Internet of
Things Journal 6, 5 (2019), 8463ś8470.
[28] Kwon Nung Choi, Harini Kolamunna, Akila Uyanwatta, Kanchana Thilakarathna, Suranga Seneviratne, Ralph Holz, Mahbub Hassan,
and Albert Y Zomaya. 2020. LoRadar: LoRa sensor network monitoring through passive packet sniing. ACM SIGCOMM Computer
Communication Review 50, 4 (2020), 10ś24.
[29] Jeferson Rodrigues Cotrim and João Henrique Kleinschmidt. 2020. LoRaWAN Mesh Networks: A Review and Classiication of Multihop
Communication. Sensors 20, 15 (2020), 4273.
[30] Daniele Croce, Michele Gucciardo, Stefano Mangione, Giuseppe Santaromita, and Ilenia Tinnirello. 2018. Impact of LoRa imperfect
orthogonality: Analysis of link-level performance. IEEE Communications Letters 22, 4 (2018), 796ś799.
[31] Francesca Cuomo, Manuel Campo, Alberto Caponi, Giuseppe Bianchi, Giampaolo Rossini, and Patrizio Pisani. 2017. EXPLoRa:
Extending the performance of LoRa by suitable spreading factor allocations. In 2017 IEEE 13th International Conference on Wireless and
Mobile Computing, Networking and Communications (WiMob). IEEE, 1ś8.
[32] Jéssika C da Silva, Daniel de L Flor, Vicente Angelo de Sousa Junior, Níbia Souza Bezerra, and Alvaro AM de Medeiros. 2021. A Survey
of LoRaWAN Simulation Tools in ns-3. Journal of Communication and Information Systems 36, 1 (2021), 17ś30.
[33] Syed Muhammad Danish, Arfa Nasir, Hassaan Khaliq Qureshi, Ayesha Binte Ashfaq, Shahid Mumtaz, and Jonathan Rodriguez.
2018. Network intrusion detection system for jamming attack in lorawan join procedure. In 2018 IEEE International Conference on
Communications (ICC). IEEE, 1ś6.
[34] Mauricio de Castro Tomé, Pedro HJ Nardelli, and Hirley Alves. 2018. Long-range low-power wireless networks and sampling strategies
in electricity metering. IEEE Transactions on Industrial Electronics 66, 2 (2018), 1629ś1637.
[35] Anthony S Deese, Joe Jesson, Thomas Brennan, Steven Hollain, Patrick Stefanacci, Emily Driscoll, Connor Dick, Keith Garcia, Ryan
Mosher, Brian Rentsch, et al. 2020. Long-term monitoring of smart city assets via Internet of Things and low-power wide-area networks.
IEEE Internet of Things Journal 8, 1 (2020), 222ś231.
[36] Carmen Delgado, José María Sanz, Chris Blondia, and Jeroen Famaey. 2020. Batteryless LoRaWAN communications using energy
harvesting: Modeling and characterization. IEEE Internet of Things Journal 8, 4 (2020), 2694ś2711.
[37] Silvia Demetri, Marco Zúñiga, Gian Pietro Picco, Fernando Kuipers, Lorenzo Bruzzone, and Thomas Telkamp. 2019. Automated
estimation of link quality for LoRa: A remote sensing approach. In 2019 18th ACM/IEEE International Conference on Information
Processing in Sensor Networks (IPSN). IEEE, 145ś156.
[38] Adwait Dongare, Craig Hesling, Khushboo Bhatia, Artur Balanuta, Ricardo Lopes Pereira, Bob Iannucci, and Anthony Rowe. 2017.
OpenChirp: A low-power wide-area networking architecture. In 2017 IEEE International Conference on Pervasive Computing and
Communications Workshops (PerCom Workshops). IEEE, 569ś574.
[39] Adwait Dongare, Revathy Narayanan, Akshay Gadre, Anh Luong, Artur Balanuta, Swarun Kumar, Bob Iannucci, and Anthony Rowe.
2018. Charm: exploiting geographical diversity through coherent combining in low-power wide-area networks. In 2018 17th ACM/IEEE
International Conference on Information Processing in Sensor Networks (IPSN). IEEE, 60ś71.
[40] Rida El Chall, Samer Lahoud, and Melhem El Helou. 2019. LoRaWAN network: Radio propagation models and performance evaluation
in various environments in Lebanon. IEEE Internet of Things Journal 6, 2 (2019), 2366ś2378.
[41] Rashad Eletreby, Diana Zhang, Swarun Kumar, and Osman Yağan. 2017. Empowering low-power wide area networks in urban settings.
In Proceedings of the Conference of the ACM Special Interest Group on Data Communication. 309ś321.
[42] Tallal Elshabrawy and Joerg Robert. 2019. Interleaved chirp spreading LoRa-based modulation. IEEE Internet of Things Journal 6, 2
(2019), 3855ś3863.
[43] Sezana Fahmida, Venkata P Modekurthy, Mahbubur Rahman, Abusayeed Saifullah, and Marco Brocanelli. 2020. Long-Lived LoRa:
Prolonging the Lifetime of a LoRa Network. In 2020 IEEE 28th International Conference on Network Protocols (ICNP). IEEE, 1ś12.
[44] Luca Feltrin, Chiara Buratti, Enrico Vinciarelli, Roberto De Bonis, and Roberto Verdone. 2018. LoRaWAN: Evaluation of link-and
system-level performance. IEEE Internet of Things Journal 5, 3 (2018), 2249ś2258.
[45] Joseph Finnegan, Ronan Farrell, and Stephen Brown. 2020. Analysis and enhancement of the LoRaWAN adaptive data rate scheme.
IEEE Internet of Things Journal 7, 8 (2020), 7171ś7180.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:37

[46] Joseph Finnegan, Kyriaki Niotaki, and Stephen Brown. 2020. Exploring the Boundaries of Ambient RF Energy Harvesting With
LoRaWAN. IEEE Internet of Things Journal 8, 7 (2020), 5736ś5743.
[47] Akshay Gadre, Revathy Narayanan, Anh Luong, Anthony Rowe, Bob Iannucci, and Swarun Kumar. 2020. Frequency coniguration for
low-power wide-area networks in a heartbeat. In 17th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI}
20). 339ś352.
[48] Akshay Gadre, Fan Yi, Anthony Rowe, Bob Iannucci, and Swarun Kumar. 2020. Quick (and Dirty) Aggregate Queries on Low-Power
WANs. In 2020 19th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN). IEEE, 277ś288.
[49] Amalinda Gamage, Jansen Christian Liando, Chaojie Gu, Rui Tan, and Mo Li. 2020. Lmac: Eicient carrier-sense multiple access for
lora. In Proceedings of the 26th Annual International Conference on Mobile Computing and Networking. 1ś13.
[50] Jiayao Gao, Weitao Xu, Salil Kanhere, Sanjay Jha, Jun Young Kim, Walter Huang, and Wen Hu. 2021. A Novel Model-Based Security
Scheme for LoRa Key Generation. In Proceedings of the 20th International Conference on Information Processing in Sensor Networks
(co-located with CPS-IoT Week 2021). 47ś61.
[51] Shang Gao, Xuehui Zhang, Cuicui Du, and Qian Ji. 2019. A multichannel low-power wide-area network with high-accuracy synchro-
nization ability for machine vibration monitoring. IEEE Internet of Things Journal 6, 3 (2019), 5040ś5047.
[52] Weifeng Gao, Wan Du, Zhiwei Zhao, Geyong Min, and Mukesh Singhal. 2019. Towards energy-fairness in lora networks. In 2019 IEEE
39th International Conference on Distributed Computing Systems (ICDCS). IEEE, 788ś798.
[53] Weifeng Gao, Zhiwei Zhao, and Geyong Min. 2020. AdapLoRa: Resource Adaptation for Maximizing Network Lifetime in LoRa
networks. In 2020 IEEE 28th International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[54] Yi Gao, Jiadong Zhang, Gaoyang Guan, and Wei Dong. 2020. LinkLab: A Scalable and Heterogeneous Testbed for Remotely Developing
and Experimenting IoT Applications. In 2020 IEEE/ACM Fifth International Conference on Internet-of-Things Design and Implementation
(IoTDI). IEEE, 176ś188.
[55] Orestis Georgiou and Usman Raza. 2017. Low power wide area network analysis: Can LoRa scale? IEEE Wireless Communications
Letters 6, 2 (2017), 162ś165.
[56] Panagiotis Gkotsiopoulos, Dimitrios Zorbas, and Christos Douligeris. 2021. Performance Determinants in LoRa Networks: a Literature
Review. IEEE Communications Surveys & Tutorials (2021).
[57] Chaojie Gu, Linshan Jiang, Rui Tan, Mo Li, and Jun Huang. 2021. Attack-aware synchronization-free data timestamping in lorawan.
ACM Transactions on Sensor Networks (TOSN) 18, 1 (2021), 1ś31.
[58] Chaojie Gu, Rui Tan, and Xin Lou. 2019. One-hop out-of-band control planes for multi-hop wireless sensor networks. ACM Transactions
on Sensor Networks (TOSN) 15, 4 (2019), 1ś29.
[59] Xiuzhen Guo, Longfei Shangguan, Yuan He, Jia Zhang, Haotian Jiang, Awais Ahmad Siddiqi, and Yunhao Liu. 2020. Aloba: rethinking
ON-OFF keying modulation for ambient LoRa backscatter. In Proceedings of the 18th Conference on Embedded Networked Sensor Systems.
192ś204.
[60] Jialuo Han and Jidong Wang. 2018. An enhanced key management scheme for LoRaWAN. Cryptography 2, 4 (2018), 34.
[61] Muhammad Hanif and Ha H Nguyen. 2020. Slope-shift keying LoRa-based modulation. IEEE Internet of Things Journal 8, 1 (2020),
211ś221.
[62] Jetmir Haxhibeqiri, Eli De Poorter, Ingrid Moerman, and Jeroen Hoebeke. 2018. A survey of LoRaWAN for IoT: From technology to
application. Sensors 18, 11 (2018), 3995.
[63] Jetmir Haxhibeqiri, Ingrid Moerman, and Jeroen Hoebeke. 2018. Low overhead scheduling of LoRa transmissions for improved
scalability. IEEE Internet of Things Journal 6, 2 (2018), 3097ś3109.
[64] Mehrdad Hessar, Ali Najai, and Shyamnath Gollakota. 2019. Netscatter: Enabling large-scale backscatter networks. In 16th {USENIX}
Symposium on Networked Systems Design and Implementation ({NSDI} 19). 271ś284.
[65] Mehrdad Hessar, Ali Najai, Vikram Iyer, and Shyamnath Gollakota. 2020. TinySDR: Low-Power {SDR} Platform for Over-the-Air
Programmable IoT Testbeds. In 17th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 20). 1031ś1046.
[66] Frank Hessel, Lars Almon, and Flor Álvarez. 2020. ChirpOTLE: a framework for practical LoRaWAN security evaluation. In Proceedings
of the 13th ACM Conference on Security and Privacy in Wireless and Mobile Networks. 306ś316.
[67] Lu Hou, Kan Zheng, Zhiming Liu, Xiaojun Xu, and Tao Wu. 2020. Design and prototype implementation of a blockchain-enabled LoRa
system with edge computing. IEEE Internet of Things Journal (2020).
[68] Ningning Hou, Xianjin Xia, and Yuanqing Zheng. 2021. Jamming of LoRa PHY and Countermeasure. In IEEE INFOCOM 2021-IEEE
Conference on Computer Communications. IEEE, 1ś10.
[69] Ningning Hou and Yuanqing Zheng. 2020. Cloaklora: A covert channel over LoRa phy. In 2020 IEEE 28th International Conference on
Network Protocols (ICNP). IEEE, 1ś11.
[70] Anzhong Hu, Tiejun Lv, Hui Gao, Zhang Zhang, and Shaoshi Yang. 2014. An ESPRIT-based approach for 2-D localization of incoherently
distributed sources in massive MIMO systems. IEEE Journal of Selected Topics in Signal Processing 8, 5 (2014), 996ś1011.
[71] Bin Hu, Zhimeng Yin, Shuai Wang, Zhuqing Xu, and Tian He. 2020. SCLoRa: Leveraging Multi-Dimensionality in Decoding Collided
LoRa Transmissions. In 2020 IEEE 28th International Conference on Network Protocols (ICNP). IEEE, 1ś11.

ACM Trans. Sensor Netw.

111:38 • Sun et al.

[72] Soukaina Ihirri, Essaid Sabir, Ahmed Errami, and Mohamed Khaldoun. 2019. A Scalable Slotted Aloha for Massive IoT: A Throughput
Analysis. In 2019 15th International Wireless Communications & Mobile Computing Conference (IWCMC). IEEE, 508ś513.
[73] Haotian Jiang, Jiacheng Zhang, Xiuzhen Guo, and Yuan He. 2021. Sense Me on the Ride: Accurate Mobile Sensing over a LoRa
Backscatter Channel. In Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems. 125ś137.
[74] Jinyan Jiang, Zhenqiang Xu, Fan Dang, and Jiliang Wang. 2021. Long-range ambient LoRa backscatter with parallel decoding. In
Proceedings of the 27th Annual International Conference on Mobile Computing and Networking. 684ś696.
[75] Yu Jiang, Linning Peng, Aiqun Hu, Sheng Wang, Yi Huang, and Lu Zhang. 2019. Physical layer identiication of LoRa devices using
constellation trace igure. EURASIP Journal on Wireless Communications and Networking 2019, 1 (2019), 1ś11.
[76] Aisha Kanwal Junejo, Fatma Benkhelifa, Boon Wong, and Julie A McCann. 2021. LoRa-LiSK: A Lightweight Shared Secret Key
Generation Scheme for LoRa Networks. IEEE Internet of Things Journal (2021).
[77] Mohamad Katanbaf, Anthony Weinand, and Vamsi Talla. 2021. Simplifying Backscatter Deployment: Full-Duplex LoRa Backscatter. In
18th {USENIX} Symposium on Networked Systems Design and Implementation ({NSDI} 21). 955ś972.
[78] Jaehyu Kim and JooSeok Song. 2017. A dual key-based activation scheme for secure LoRaWAN. Wireless Communications and Mobile
Computing 2017 (2017).
[79] Jaehyu Kim and JooSeok Song. 2017. A simple and eicient replay attack prevention scheme for LoRaWAN. In Proceedings of the 2017
the 7th International Conference on Communication and Network Security. 32ś36.
[80] Nikolaos Kouvelas, Vijay S Rao, R Venkatesha Prasad, Gauri Tawde, and Koen Langendoen. 2020. p-CARMA: Politely Scaling LoRaWAN..
In EWSN. 25ś36.
[81] Rachel Kufakunesu, Gerhard P Hancke, and Adnan M Abu-Mahfouz. 2020. A survey on adaptive data rate optimization in LoRaWAN:
Recent solutions and major challenges. Sensors 20, 18 (2020), 5044.
[82] Ka-Ho Lam, Chi-Chung Cheung, and Wah-Ching Lee. 2019. Rssi-based lora localization systems for large-scale indoor and outdoor
environments. IEEE Transactions on Vehicular Technology 68, 12 (2019), 11778ś11791.
[83] Huang-Chen Lee and Kai-Hsiang Ke. 2018. Monitoring of large-area IoT sensors using a LoRa wireless mesh network system: Design
and evaluation. IEEE Transactions on Instrumentation and Measurement 67, 9 (2018), 2177ś2187.
[84] Luca Leonardi, Filippo Battaglia, and Lucia Lo Bello. 2019. RT-LoRa: A medium access strategy to support real-time lows over
LoRa-based networks for industrial IoT applications. IEEE Internet of Things Journal 6, 6 (2019), 10812ś10823.
[85] Chenning Li and Zhichao Cao. 2022. LoRa Networking Techniques for Large-scale and Long-term IoT: A Down-to-top Survey. ACM
Computing Surveys (CSUR) 55, 3 (2022), 1ś36.
[86] Chenning Li, Hanqing Guo, Shuai Tong, Xiao Zeng, Zhichao Cao, Mi Zhang, Qiben Yan, Li Xiao, Jiliang Wang, and Yunhao Liu.
2021. NELoRa: Towards Ultra-low SNR LoRa Communication with Neural-enhanced Demodulation. In ACM Conference on Embedded
Networked Sensor Systems (SenSys’ 21).
[87] Yinghui Li, Jing Yang, and Jiliang Wang. 2020. Dylora: Towards energy eicient dynamic lora transmission control. In IEEE INFOCOM
2020-IEEE Conference on Computer Communications. IEEE, 2312ś2320.
[88] Zhijun Li and Yongrui Chen. 2019. Achieving universal low-power wide-area networks on existing wireless devices. In 2019 IEEE 27th
International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[89] Jansen C Liando, Amalinda Gamage, Agustinus W Tengourtius, and Mo Li. 2019. Known and unknown facts of LoRa: Experiences
from a large-scale measurement study. ACM Transactions on Sensor Networks (TOSN) 15, 2 (2019), 1ś35.
[90] Jun Lin, Zhiqi Shen, and Chunyan Miao. 2017. Using blockchain technology to build trust in sharing LoRaWAN IoT. In Proceedings of
the 2nd International Conference on Crowd Science and Engineering. 38ś43.
[91] Yuxiang Lin, Wei Dong, Yi Gao, and Tao Gu. 2021. SateLoc: A virtual ingerprinting approach to outdoor LoRa localization using
satellite images. ACM Transactions on Sensor Networks (TOSN) 17, 4 (2021), 1ś28.
[92] Jun Liu, Jiayao Gao, Sanjay Jha, and Wen Hu. 2021. Seirios: leveraging multiple channels for LoRaWAN indoor and outdoor localization.
In Proceedings of the 27th Annual International Conference on Mobile Computing and Networking. 656ś669.
[93] Jun Liu, Weitao Xu, Sanjay Jha, and Wen Hu. 2020. Nephalai: towards LPWAN C-RAN with physical layer compression. In Proceedings
of the 26th Annual International Conference on Mobile Computing and Networking. 1ś12.
[94] Li Liu, Yuguang Yao, Zhichao Cao, and Mi Zhang. 2021. Deeplora: Learning accurate path loss model for long distance links in lpwan.
In Proceedings of IEEE INFOCOM.
[95] Ruofeng Liu, Zhimeng Yin, Wenchao Jiang, and Tian He. 2020. XFi: Cross-technology IoT Data Collection via Commodity WiFi. In
2020 IEEE 28th International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[96] Xiaolan Liu, Zhijin Qin, Yue Gao, and Julie A McCann. 2019. Resource allocation in wireless powered IoT networks. IEEE Internet of
Things Journal 6, 3 (2019), 4935ś4945.
[97] Qasim Lone, Etienne Dublé, Franck Rousseau, Ingrid Moerman, Spilios Giannoulis, and Andrzej Duda. 2018. WiSH-WalT: a framework
for controllable and reproducible LoRa testbeds. In 2018 IEEE 29th Annual International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC). IEEE, 1ś7.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:39

[98] Ali Loubany, Samer Lahoud, and Rida El Chall. 2020. Adaptive algorithm for spreading factor selection in LoRaWAN networks with
multiple gateways. Computer Networks 182 (2020), 107491.
[99] Yongsen Ma, Gang Zhou, and Shuangquan Wang. 2019. WiFi sensing with channel state information: A survey. ACM Computing
Surveys (CSUR) 52, 3 (2019), 1ś36.
[100] Michele Magno, Faycal Ait Aoudia, Matthieu Gautier, Olivier Berder, and Luca Benini. 2017. WULoRa: An energy eicient IoT end-node
for energy harvesting and heterogeneous communication. In Design, Automation & Test in Europe Conference & Exhibition (DATE), 2017.
IEEE, 1528ś1533.
[101] Davide Magrin, Marco Centenaro, and Lorenzo Vangelista. 2017. Performance evaluation of LoRa networks in a smart city scenario. In
2017 IEEE International Conference on Communications (ICC). ieee, 1ś7.
[102] Aamir Mahmood, Emiliano Sisinni, Lakshmikanth Guntupalli, Raúl Rondón, Syed Ali Hassan, and Mikael Gidlund. 2018. Scalability
analysis of a LoRa network under imperfect orthogonality. IEEE Transactions on Industrial Informatics 15, 3 (2018), 1425ś1436.
[103] Jaco M Marais, Reza Malekian, and Adnan M Abu-Mahfouz. 2017. LoRa and LoRaWAN testbeds: A review. In 2017 Ieee Africon. IEEE,
1496ś1501.
[104] Paul Marcelis, Nikolaos Kouvelas, Vijay S Rao, and Venkatesha Prasad. 2020. DaRe: Data recovery through application layer coding for
LoRaWAN. IEEE Transactions on Mobile Computing (2020).
[105] Riccardo Marini, Walter Cerroni, and Chiara Buratti. 2020. A Novel Collision-Aware Adaptive Data Rate Algorithm for LoRaWAN
Networks. IEEE Internet of Things Journal (2020).
[106] Riccardo Marini, Konstantin Mikhaylov, Gianni Pasolini, and Chiara Buratti. 2021. LoRaWANSim: A Flexible Simulator for LoRaWAN
Networks. Sensors 21, 3 (2021), 695.
[107] Arturas Medeisis and Algimantas Kajackas. 2000. On the use of the universal Okumura-Hata propagation prediction model in rural
areas. In VTC2000-Spring. 2000 IEEE 51st Vehicular Technology Conference Proceedings (Cat. No. 00CH37026), Vol. 3. IEEE, 1815ś1818.
[108] Konstantin Mikhaylov, Juha Petaejaejaervi, and Tuomo Haenninen. 2016. Analysis of capacity and scalability of the LoRa low power
wide area network technology. In European Wireless 2016; 22th European Wireless Conference. VDE, 1ś6.
[109] Di Mu, Yitian Chen, Junyang Shi, and Mo Sha. 2020. Runtime Control of LoRa Spreading Factor for Campus Shuttle Monitoring. In
2020 IEEE 28th International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[110] SeungJae Na, DongYeop Hwang, WoonSeob Shin, and Ki-Hyung Kim. 2017. Scenario and countermeasure for replay attack using join
request messages in LoRaWAN. In 2017 international conference on information networking (ICOIN). IEEE, 718ś720.
[111] Rajalakshmi Nandakumar, Vikram Iyer, and Shyamnath Gollakota. 2018. 3d localization for sub-centimeter sized devices. In Proceedings
of the 16th ACM Conference on Embedded Networked Sensor Systems. 108ś119.
[112] Tung T Nguyen, Ha H Nguyen, Robert Barton, and Patrick Grossetete. 2019. Eicient design of chirp spread spectrum modulation for
low-power wide-area networks. IEEE Internet of Things Journal 6, 6 (2019), 9503ś9515.
[113] Sina Rafati Niya, Sanjiv S Jha, Thomas Bocek, and Burkhard Stiller. 2018. Design and implementation of an automated and decentralized
pollution monitoring system with blockchains, smart contracts, and LoRaWAN. In NOMS 2018-2018 IEEE/IFIP Network Operations and
Management Symposium. IEEE, 1ś4.
[114] Hassan Noura, Tarif Hatoum, Ola Salman, Jean-Paul Yaacoub, and Ali Chehab. 2020. LoRaWAN Security Survey: Issues, Threats and
Possible Mitigation Techniques. Internet of Things (2020), 100303.
[115] Moises Nunez Ochoa, Arturo Guizar, Mickael Maman, and Andrzej Duda. 2017. Evaluating LoRa energy eiciency for adaptive
networks: From star to mesh topologies. In 2017 IEEE 13th International Conference on Wireless and Mobile Computing, Networking and
Communications (WiMob). IEEE, 1ś8.
[116] Jorge Ortín, Matteo Cesana, and Alessandro Redondi. 2019. Augmenting LoRaWAN performance with listen before talk. IEEE
Transactions on Wireless Communications 18, 6 (2019), 3113ś3128.
[117] Junhyun Park, Kunho Park, Hyeongho Bae, and Chong-Kwon Kim. 2020. EARN: Enhanced ADR with coding rate adaptation in
LoRaWAN. IEEE Internet of Things Journal 7, 12 (2020), 11873ś11883.
[118] Gianni Pasolini, Chiara Buratti, Luca Feltrin, Flavio Zabini, Cristina De Castro, Roberto Verdone, and Oreste Andrisano. 2018. Smart
city pilot projects using LoRa and IEEE802. 15.4 technologies. Sensors 18, 4 (2018), 1118.
[119] Yao Peng, Longfei Shangguan, Yue Hu, Yujie Qian, Xianshang Lin, Xiaojiang Chen, Dingyi Fang, and Kyle Jamieson. 2018. PLoRa: A
passive long-range data network from ambient LoRa transmissions. In Proceedings of the 2018 Conference of the ACM Special Interest
Group on Data Communication. 147ś160.
[120] Juha Petäjäjärvi, Konstantin Mikhaylov, Marko Pettissalo, Janne Janhunen, and Jari Iinatti. 2017. Performance of a low-power wide-area
network based on LoRa technology: Doppler robustness, scalability, and coverage. International Journal of Distributed Sensor Networks
13, 3 (2017), 1550147717699412.
[121] Rajeev Piyare, Amy L Murphy, Michele Magno, and Luca Benini. 2018. On-demand LoRa: Asynchronous TDMA for energy eicient
and low latency communication in IoT. Sensors 18, 11 (2018), 3718.
[122] Tommaso Polonelli, Davide Brunelli, Achille Marzocchi, and Luca Benini. 2019. Slotted aloha on lorawan-design, analysis, and
deployment. Sensors 19, 4 (2019), 838.

ACM Trans. Sensor Netw.

111:40 • Sun et al.

[123] Mina Rady, Maryam Hafeez, and Syed Ali Raza Zaidi. 2019. Computational methods for network-aware and network-agnostic IoT low
power wide area networks (LPWANs). IEEE Internet of Things Journal 6, 3 (2019), 5732ś5744.
[124] Ceferino Gabriel Ramirez, Anton Sergeyev, Assya Dyussenova, and Bob Iannucci. 2019. LongShoT: Long-range synchronization of
time. In Proceedings of the 18th International Conference on Information Processing in Sensor Networks. 289ś300.
[125] SR Jino Ramson, Walter D León-Salas, Zachary Brecheisen, Erika J Foster, Clif T Johnston, Darrell G Schulze, Timothy Filley, Rahim
Rahimi, Martín Juan Carlos Villalta Soto, Juan A Lopa Bolivar, et al. 2021. A Self-Powered, Real-Time, LoRaWAN IoT-Based Soil Health
Monitoring System. IEEE Internet of Things Journal 8, 11 (2021), 9278ś9293.
[126] Arvind Singh Rawat, Jagadheswaran Rajendran, Harikrishnan Ramiah, and Arti Rana. 2020. LORA (Long Range) and LORAWAN
technology for IoT applications in Covid-19 pandemic. In 2020 International Conference on Advances in Computing, Communication &
Materials (ICACCM). IEEE, 419ś422.
[127] Brecht Reynders, Wannes Meert, and Soie Pollin. 2017. Power and spreading factor control in low power wide area networks. In 2017
IEEE International Conference on Communications (ICC). IEEE, 1ś6.
[128] Brecht Reynders and Soie Pollin. 2016. Chirp spread spectrum as a modulation technique for long range communication. In 2016
Symposium on Communications and Vehicular Technologies (SCVT). IEEE, 1ś5.
[129] Brecht Reynders, Qing Wang, Pere Tuset-Peiro, Xavier Vilajosana, and Soie Pollin. 2018. Improving reliability and scalability of
lorawans through lightweight scheduling. IEEE Internet of Things Journal 5, 3 (2018), 1830ś1842.
[130] Mattia Rizzi, Paolo Ferrari, Alessandra Flammini, and Emiliano Sisinni. 2017. Evaluation of the IoT LoRaWAN solution for distributed
measurement applications. IEEE Transactions on Instrumentation and Measurement 66, 12 (2017), 3340ś3349.
[131] Mattia Rizzi, Paolo Ferrari, Alessandra Flammini, Emiliano Sisinni, and Mikael Gidlund. 2017. Using LoRa for industrial wireless
networks. In 2017 IEEE 13th international workshop on factory communication systems (WFCS). IEEE, 1ś4.
[132] Pieter Robyns, Eduard Marin, Wim Lamotte, Peter Quax, Dave Singelée, and Bart Preneel. 2017. Physical-layer ingerprinting of LoRa
devices using supervised and zero-shot learning. In Proceedings of the 10th ACM Conference on Security and Privacy in Wireless and
Mobile Networks. 58ś63.
[133] Henri Ruotsalainen, Junqing Zhang, and Stepan Grebeniuk. 2019. Experimental investigation on wireless key generation for low-power
wide-area networks. IEEE Internet of Things Journal 7, 3 (2019), 1745ś1755.
[134] Eryk Schiller, Silas Weber, and Burkhard Stiller. 2020. Design and Evaluation of an SDR-based LoRa Cloud Radio Access Network. In
2020 16th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob)(50308). IEEE, 1ś7.
[135] Muhammad Osama Shahid, Millan Philipose, Krishna Chintalapudi, Suman Banerjee, and Bhuvana Krishnaswamy. 2021. Concurrent
interference cancellation: decoding multi-packet collisions in LoRa. In Proceedings of the 2021 ACM SIGCOMM 2021 Conference. 503ś515.
[136] Vishal Sharma, Ilsun You, Giovanni Pau, Mario Collotta, Jae Deok Lim, and Jeong Nyeo Kim. 2018. LoRaWAN-based energy-eicient
surveillance by drones for intelligent transportation systems. Energies 11, 3 (2018), 573.
[137] Cheng Shen, Tian Liu, Jun Huang, and Rui Tan. 2021. When LoRa Meets EMR: Electromagnetic Covert Channels Can Be Super Resilient.
In 2021 IEEE Symposium on Security and Privacy (SP). IEEE, 1304ś1317.
[138] Guanxiong Shen, Junqing Zhang, Alan Marshall, Linning Peng, and Xianbin Wang. 2021. Radio frequency ingerprint identiication for
LoRa using spectrogram and CNN. In IEEE INFOCOM 2021-IEEE Conference on Computer Communications. IEEE, 1ś10.
[139] Junyang Shi, Di Mu, and Mo Sha. 2021. Enabling Cross-technology Communication from LoRa to ZigBee via Payload Encoding in
Sub-1 GHz Bands. ACM Transactions on Sensor Networks (TOSN) 18, 1 (2021), 1ś26.
[140] Mariusz Slabicki, Gopika Premsankar, and Mario Di Francesco. 2018. Adaptive coniguration of LoRa networks for dense IoT
deployments. In NOMS 2018-2018 IEEE/IFIP Network Operations and Management Symposium. IEEE, 1ś9.
[141] Guochao Song, Hang Yang, Wei Wang, and Tao Jiang. 2020. Reliable wide-area backscatter via channel polarization. In IEEE INFOCOM
2020-IEEE Conference on Computer Communications. IEEE, 1300ś1308.
[142] Binbin Su, Zhijin Qin, and Qiang Ni. 2020. Energy Eicient Uplink Transmissions in LoRa Networks. IEEE Transactions on Communica-
tions 68, 8 (2020), 4960ś4972.
[143] Kai Sun, Zhimeng Yin, Weiwei Chen, Shuai Wang, Zeyu Zhang, and Tian He. 2021. Partial Symbol Recovery for Interference Resilience
in Low-Power Wide Area Networks. In 2021 IEEE 29th International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[144] Yuyi Sun, Jiming Chen, Shibo He, and Zhiguo Shi. 2020. High-Conidence Gateway Planning and Performance Evaluation of a Hybrid
LoRa Network. IEEE Internet of Things Journal 8, 2 (2020), 1071ś1081.
[145] Zehua Sun, Qiuhong Ke, Hossein Rahmani, Mohammed Bennamoun, Gang Wang, and Jun Liu. 2020. Human action recognition from
various data modalities: A review. arXiv preprint arXiv:2012.11866 (2020).
[146] Jothi Prasanna Shanmuga Sundaram, Wan Du, and Zhiwei Zhao. 2019. A survey on lora networking: Research problems, current
solutions, and open issues. IEEE Communications Surveys & Tutorials 22, 1 (2019), 371ś388.
[147] Woo-Jin Sung, Hyeong-Geun Ahn, Jong-Beom Kim, and Seong-Gon Choi. 2018. Protecting end-device from replay attack on LoRaWAN.
In 2018 20th International Conference on Advanced Communication Technology (ICACT). IEEE, 167ś171.
[148] Vamsi Talla, Mehrdad Hessar, Bryce Kellogg, Ali Najai, Joshua R Smith, and Shyamnath Gollakota. 2017. Lora backscatter: Enabling the
vision of ubiquitous connectivity. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 1, 3 (2017), 1ś24.

ACM Trans. Sensor Netw.

 Recent Advances in LoRa: A Comprehensive Survey • 111:41

[149] Yas Hosseini Tehrani, Arash Amini, and Seyed Mojtaba Atarodi. 2020. A Tree-Structured LoRa Network for Energy Eiciency. IEEE
Internet of Things Journal 8, 7 (2020), 6002ś6011.
[150] Mohamed Amine Ben Temim, Guillaume Ferré, Baptiste Laporte-Fauret, Dominique Dallet, Bryce Minger, and Loïc Fuché. 2020. An
enhanced receiver to decode superposed LoRa-like signals. IEEE internet of things journal 7, 8 (2020), 7419ś7431.
[151] Pei Tian, Xiaoyuan Ma, Carlo Alberto Boano, Ye Liu, Fengxu Yang, Xin Tian, Dan Li, and Jianming Wei. 2021. ChirpBox: An
Infrastructure-Less LoRa Testbed. In EWSN. 115ś126.
[152] Pei Tian, Fengxu Yang, Xiaoyuan Ma, Carlo Alberto Boano, Xin Tian, Ye Liu, and Jianming Wei. 2021. Environmental Impact on the
Long-Term Connectivity and Link Quality of an Outdoor LoRa Network. In Proceedings of the 19th ACM Conference on Embedded
Networked Sensor Systems. 565ś568.
[153] Stefano Tomasin, Simone Zulian, and Lorenzo Vangelista. 2017. Security analysis of lorawan join procedure for internet of things
networks. In 2017 IEEE Wireless Communications and Networking Conference Workshops (WCNCW). IEEE, 1ś6.
[154] Shuai Tong, Zilin Shen, Yunhao Liu, and Jiliang Wang. 2021. Combating link dynamics for reliable lora connection in urban settings. In
Proceedings of the 27th Annual International Conference on Mobile Computing and Networking. 642ś655.
[155] Shuai Tong, Jiliang Wang, and Yunhao Liu. 2020. Combating packet collisions using non-stationary signal scaling in LPWANs. In
Proceedings of the 18th International Conference on Mobile Systems, Applications, and Services. 234ś246.
[156] Shuai Tong, Zhenqiang Xu, and Jiliang Wang. 2020. CoLoRa: Enabling Multi-Packet Reception in LoRa. In IEEE INFOCOM 2020-IEEE
Conference on Computer Communications. IEEE, 2303ś2311.
[157] Verónica Toro-Betancur, Gopika Premsankar, Mariusz Slabicki, and Mario Di Francesco. 2021. Modeling Communication Reliability in
LoRa Networks with Device-level Accuracy. In IEEE INFOCOM 2021-IEEE Conference on Computer Communications. IEEE, 1ś10.
[158] Roman Trüb, Reto Da Forno, Lukas Sigrist, Lorin Mühlebach, Andreas Biri, Jan Beutel, and Lothar Thiele. 2020. FlockLab 2: Multi-Modal
Testing and Validation for Wireless IoT. In 3rd Workshop on Benchmarking Cyber-Physical Systems and Internet of Things (CPS-IoTBench
2020). ETH Zurich, Computer Engineering and Networks Laboratory (TIK).
[159] Floris Van den Abeele, Jetmir Haxhibeqiri, Ingrid Moerman, and Jeroen Hoebeke. 2017. Scalability analysis of large-scale LoRaWAN
networks in ns-3. IEEE Internet of Things Journal 4, 6 (2017), 2186ś2198.
[160] Ambuj Varshney, Oliver Harms, Carlos Pérez-Penichet, Christian Rohner, Frederik Hermans, and Thiemo Voigt. 2017. Lorea: A
backscatter architecture that achieves a long communication range. In Proceedings of the 15th ACM Conference on Embedded Network
Sensor Systems. 1ś14.
[161] Nadège Varsier and Jean Schwoerer. 2017. Capacity limits of LoRaWAN technology for smart metering applications. In 2017 IEEE
international conference on communications (ICC). IEEE, 1ś6.
[162] Xiong Wang, Linghe Kong, Liang He, and Guihai Chen. 2019. MLoRa: A multi-packet reception protocol in LoRa networks. In 2019
IEEE 27th International Conference on Network Protocols (ICNP). IEEE, 1ś11.
[163] Xiong Wang, Linghe Kong, Zucheng Wu, Long Cheng, Chenren Xu, and Guihai Chen. 2020. SLoRa: towards secure LoRa communications
with ine-grained physical layer features. In Proceedings of the 18th Conference on Embedded Networked Sensor Systems. 258ś270.
[164] Yuting Wang, Xiaolong Zheng, Liang Liu, and Huadong Ma. 2021. PolarTracker: Attitude-aware channel access for loating low power
wide area networks. In IEEE INFOCOM 2021-IEEE Conference on Computer Communications. IEEE, 1ś10.
[165] Zhe Wang, Linghe Kong, Kangjie Xu, Liang He, Kaishun Wu, and Guihai Chen. 2020. Online concurrent transmissions at LoRa gateway.
In IEEE INFOCOM 2020-IEEE Conference on Computer Communications. IEEE, 2331ś2340.
[166] Antoine Waret, Megumi Kaneko, Alexandre Guitton, and Nancy El Rachkidy. 2018. LoRa throughput analysis with imperfect spreading
factor orthogonality. IEEE wireless communications letters 8, 2 (2018), 408ś411.
[167] Xianjin Xia, Ningning Hou, Yuanqing Zheng, and Tao Gu. 2021. PCube: scaling LoRa concurrent transmissions with reception
diversities. In Proceedings of the 27th Annual International Conference on Mobile Computing and Networking. 670ś683.
[168] Xianjin Xia, Yuanqing Zheng, and Tao Gu. 2020. FTrack: Parallel decoding for LoRa transmissions. IEEE/ACM Transactions on
Networking 28, 6 (2020), 2573ś2586.
[169] Xianjin Xia, Yuanqing Zheng, and Tao Gu. 2021. LiteNap: Downclocking LoRa Reception. IEEE/ACM Transactions on Networking
(2021).
[170] Binbin Xie and Jie Xiong. 2020. Combating interference for long range LoRa sensing. In Proceedings of the 18th Conference on Embedded
Networked Sensor Systems. 69ś81.
[171] Binbin Xie, Yuqing Yin, and Jie Xiong. 2021. Pushing the Limits of Long Range Wireless Sensing with LoRa. Proceedings of the ACM on
Interactive, Mobile, Wearable and Ubiquitous Technologies 5, 3 (2021), 1ś21.
[172] Jie Xiong, Karthikeyan Sundaresan, and Kyle Jamieson. 2015. Tonetrack: Leveraging frequency-agile radios for time-based indoor
wireless localization. In Proceedings of the 21st Annual International Conference on Mobile Computing and Networking. 537ś549.
[173] Ting Xu and Ming Zhao. 2020. A LoRaWAN-MAC Protocol Based on WSN Residual Energy to Adjust Duty Cycle. In 2020 IEEE 40th
International Conference on Distributed Computing Systems (ICDCS). IEEE, 1415ś1420.
[174] Weitao Xu, Sanjay Jha, and Wen Hu. 2018. Exploring the feasibility of physical layer key generation for LoRaWAN. In 2018 17th IEEE
International Conference On Trust, Security And Privacy In Computing And Communications/12th IEEE International Conference On Big

ACM Trans. Sensor Netw.

111:42 • Sun et al.

Data Science And Engineering (TrustCom/BigDataSE). IEEE, 231ś236.

[175] Weitao Xu, Sanjay Jha, and Wen Hu. 2018. LoRa-Key: Secure key generation system for LoRa-based network. IEEE Internet of Things
Journal 6, 4 (2018), 6404ś6416.
[176] Weitao Xu, Jun Young Kim, Walter Huang, Salil Kanhere, Sanjay Jha, and Wen Hu. 2020. EMIoT: A LoRa-enabled Smart Building
Solution Based on Emergency Lights. In Proceedings of the 7th ACM International Conference on Systems for Energy-Eicient Buildings,
Cities, and Transportation. 330ś331.
[177] Weitao Xu, Jun Young Kim, Walter Huang, Salil S Kanhere, Sanjay K Jha, and Wen Hu. 2019. Measurement, characterization, and
modeling of lora technology in multiloor buildings. IEEE Internet of Things Journal 7, 1 (2019), 298ś310.
[178] Weitao Xu, Junqing Zhang, Shunqi Huang, Chengwen Luo, and Wei Li. 2021. Key Generation for Internet of Things: A Contemporary
Survey. ACM Computing Surveys (CSUR) 54, 1 (2021), 1ś37.
[179] Zhuqing Xu, Junzhou Luo, Zhimeng Yin, Tian He, and Fang Dong. 2020. S-MAC: achieving high scalability via adaptive scheduling in
LPWAN. In IEEE INFOCOM 2020-IEEE Conference on Computer Communications. IEEE, 506ś515.
[180] Zhenqiang Xu, Shuai Tong, Pengjin Xie, and Jiliang Wang. 2020. FlipLoRa: Resolving collisions with up-down quasi-orthogonality. In
2020 17th Annual IEEE International Conference on Sensing, Communication, and Networking (SECON). IEEE, 1ś9.
[181] Zhenqiang Xu, Pengjin Xie, and Jiliang Wang. 2021. Pyramid: Real-time lora collision decoding with peak tracking. In IEEE INFOCOM
2021-IEEE Conference on Computer Communications. IEEE, 1ś9.
[182] Huanqi Yang, Hongbo Liu, Chengwen Luo, Yuezhong Wu, Wei Li, Albert Y. Zomaya, Linqi Song, and Weitao Xu. 2022. Vehicle-Key: A
Secret Key Establishment Scheme for LoRa-enabled IoV Communications. In 2022 IEEE 42th International Conference on Distributed
Computing Systems (ICDCS). IEEE.
[183] Mingran Yang, Junbo Zhang, Akshay Gadre, Zaoxing Liu, Swarun Kumar, and Vyas Sekar. 2020. Joltik: enabling energy-eicient"
future-proof" analytics on low-power wide-area networks. In Proceedings of the 26th Annual International Conference on Mobile
Computing and Networking. 1ś14.
[184] Xueying Yang, Evgenios Karampatzakis, Christian Doerr, and Fernando Kuipers. 2018. Security vulnerabilities in LoRaWAN. In 2018
IEEE/ACM Third International Conference on Internet-of-Things Design and Implementation (IoTDI). IEEE, 129ś140.
[185] Fusang Zhang, Zhaoxin Chang, Kai Niu, Jie Xiong, Beihong Jin, Qin Lv, and Daqing Zhang. 2020. Exploring lora for long-range
through-wall sensing. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies 4, 2 (2020), 1ś27.
[186] Fusang Zhang, Zhaoxin Chang, Jie Xiong, Rong Zheng, Junqi Ma, Kai Niu, Beihong Jin, and Daqing Zhang. 2021. Unlocking the
beamforming potential of lora for long-range multi-target respiration sensing. Proceedings of the ACM on Interactive, Mobile, Wearable
and Ubiquitous Technologies 5, 2 (2021), 1ś25.
[187] Junqing Zhang, Alan Marshall, and Lajos Hanzo. 2018. Channel-envelope diferencing eliminates secret key correlation: LoRa-based
key generation in low power wide area networks. IEEE Transactions on Vehicular Technology 67, 12 (2018), 12462ś12466.
[188] Xihai Zhang, Mingming Zhang, Fanfeng Meng, Yue Qiao, Suijia Xu, and Senghout Hour. 2018. A low-power wide-area network
information monitoring system by combining NB-IoT and LoRa. IEEE Internet of things Journal 6, 1 (2018), 590ś598.
[189] Wenju Zhao, Shengwei Lin, Jiwen Han, Rongtao Xu, and Lu Hou. 2017. Design and implementation of smart irrigation system based
on LoRa. In 2017 IEEE Globecom Workshops (GC Wkshps). IEEE, 1ś6.
[190] Dimitrios Zorbas, Khaled Abdelfadeel, Panayiotis Kotzanikolaou, and Dirk Pesch. 2020. TS-LoRa: Time-slotted LoRaWAN for the
industrial Internet of Things. Computer Communications 153 (2020), 1ś10.

ACM Trans. Sensor Netw.