See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/343435055

LoRaquatica: Studying Range and Location Estimation using LoRa and IoT in
Aquatic Sensing

Conference Paper · March 2020

DOI: 10.1109/PerComWorkshops48775.2020.9156088

CITATIONS READS
14 149

5 authors, including:

Marko Radeta Miguel Ribeiro

Wave Labs, MARE, ARDITI, University of Madeira, University of Belgrade Instituto Superior Técnico
39 PUBLICATIONS 178 CITATIONS 24 PUBLICATIONS 264 CITATIONS

SEE PROFILE SEE PROFILE

Dinarte Vasconcelos Nuno Jardim Nunes

Instituto Superior Técnico University of Lisbon
10 PUBLICATIONS 107 CITATIONS 265 PUBLICATIONS 3,292 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Marko Radeta on 02 November 2021.

The user has requested enhancement of the downloaded file.

 LoRaquatica: Studying Range and Location
Estimation using LoRa and IoT in Aquatic Sensing
Marko Radeta Miguel Ribeiro Dinarte Vasconcelos
ITI/LARSyS, University of Madeira ITI/LARSyS, Tecnico - U.Lisbon ITI/LARSyS, Tecnico - U.Lisbon
Wave Lab, Tigerwhale Lisbon, Portugal Lisbon, Portugal
Funchal, Portugal jose.miguel.ribeiro@tecnico.ulisboa.pt dinarte.vasconcelos@tecnico.ulisboa.pt
marko.radeta@m-iti.org

Hildegardo Noronha Nuno Jardim Nunes

ITI/LARSyS, Tecnico - U.Lisbon ITI/LARSyS, Tecnico - U.Lisbon
Lisbon, Portugal Lisbon, Portugal
hildnoronha@gmail.com nunojnunes@tecnico.ulisboa.pt

Abstract—While ubiquitous computing remains vastly applied existing technologies. These existing devices indeed, can collect
in urban environments, their applications in ocean environment valuable parameters which are crucial for studying the marine
remain scarce due to the limitations in range and cost of current flora and fauna, their habitats and ecosystems. However, all
radio technology. This hinders environmental telemetry in the
oceans and other remote areas. In this study, we explore the of these technologies have a high cost, facing challenges and
usage of IoT and Long Range Radio Communication (LoRa) in risks when applied in harsh ocean settings. These challenges
ocean environments. We study the maximum distance for LoRa include dealing with poor signal propagation, salt corrosion,
and a potential location estimation based on the same technology water and pressure proofing, battery autonomy, etc, even though
using the passive RSSI analysis. Using three coastal based nodes some devices can use renewable energies, such as solar panels,
and a node mounted on a sea vessel, we report a maximum range
of 83.6km. We also achieve a location error within a radius wind turbines, or Wave Energy Converters (WEC). [Rynne
of 3.4km (4% of maximum distance) in the sea. These results and Von EllenriederRynne and Von Ellenrieder2008]. Another
support marine biologist expeditions, allowing them to use low- big challenge is obtaining the geolocation of the collected
cost, long-lasting and easy to deploy solutions for tracking marine data. Traditionally, geolocation is acquired using satellite-based
objects and species in open ocean, providing them data in near- systems, which is then stored locally with other relevant data.
real-time. We discuss the findings from used models, outlining
limitations, and providing a scenario for future ubiquitous IoT However, satellite-based location systems are both expensive
applications for tracking sea objects. and energy consuming, and the data can only be retrieved
Index Terms—location services, sensor networks, embedded later, when marine biologists recapture the taxa [Cermeño,
systems, LoRa, IoT, sea Vessels, environmental telemetry Quílez-Badia, Ospina-Alvarez, and BlockCermeño et al.2015].

I. I NTRODUCTION A. Application Scenario

Interest for ocean exploration has been growing evermore In most cases, marine biologists study species by gathering
during the last years, both in terms of natural resource data from animal tags either by: (i) physically recovering
scavenging, as well as in its protection and conservation. the tag, or by (ii) using radio (usually VHF) and satellite
While most of the oceans remain greatly unexplored, current communication (typically GPS). The former solution is long
applications of technology in aquatic settings allow sea vessels with repetitive tasks which require the recapture of the animals
to accomplish numerous tasks. For instance, present tools on taking months or even years to locate again. As opposed to
market such as sonars1 facilitate the detection of fish and other land animals, marine animals do not have as many physical
marine taxa using sound waves and Wi-Fi for communication to constraints, and the ability to dive makes it very difficult
a mobile application. Moreover, emergency position indicating to relocate them, increasing fuel costs. On other hand, the
radio beacons [Joo, Lee, Lee, Sin, Lee, and KimJoo et al.2008] latter solution is an improvement, using radio signals for
(EPIRB) using 406 MHz radio in combination with GPS, the tag recovery, providing a rough estimate of direction and
facilitate the rescue of those in need. Nevertheless, marine range from a receiving antenna. It still requires the tags to be
biologists are of crucial importance regarding oceanic studies. physically recovered to obtain back the data. Satellites also
As they explore marine species, they focus on understanding provide both remote data recovery and accurate geolocation,
the impact of human activities on their natural habitat. However, however, by using the GPS, the battery autonomy of bio loggers
they often find themselves limited by the high costs of current is short and the data transfer fees are high (e.g. ARGUS). Our
study explores LoRa as a low-cost and long range solution
1 https://deepersonar.com for real-time remote environmental telemetry and location
estimation for future scientists, while also studying the longest than in the ocean environments [Guegan, Murad, Lebreton,
LoRa range. and BonhommeauGuegan et al.2017], [Gogendeau, Murad,
Bernard, Kerzerho, and BonhommeauGogendeau et al.2018]
B. Research Questions and Contributions which are the focus of our study. Furthermore, they only focus
While other studies focus on small distances, or city on small distances (under 10 km), where in this paper, we
environments, this study, explores the issues of using long range focus on large open areas. When dealing with geolocation
radio (LoRa) in ocean environments, focusing on low altitudes estimation, the Lora Alliance can provide solutions using LoRa
for data collection using sea vessels, where the curvature of transceivers [CommitteeCommittee[n. d.]]. Their solutions can
the Earth makes a great impact on communications range use two different technologies to achieve the geolocation: (i)
[P. M. HallP. M. Hall1980]. We explore LoRa as a mean for One using the Received Signal Strength Indication (RSSI) for
oceanic environmental telemetry as well as to approximate a coarser geolocation (1 - 2 km) and; (ii) the other which uses
location without the usage of high energy devices. To achieve Time Difference of Arrival (TDoA) for a finer (20 - 200 m)
this, we focus on the following research questions: geolocation.
[RQ1]. Which is the maximum LoRa distance in ocean SmartParks2 is an example of an initiative which uses
environments? We explore the maximum range of LoRa signal, LoRaWAN geolocation technology to help with nature study
emitted from the sea vessel reaching the coastal nodes. and conservation. In a presentation3 during the The Things
[RQ2]. How does the RSSI-based distance and location Network (TTN) 2017 conference, Tim Van Dam4 , explained
estimation behaves in ocean setting? Using several distances the usage of their system to cover, track and protect endangered
obtained from land and the nodes, we explore the feasibility species in natural parks. Their biggest example is the Akagera
of a generic model to estimate the distance from sender to National Park with an area of 1 122 km2 . Even though they
receiver, applied in ocean setting. managed to keep their costs relatively low, the solution is still
The contribution of this study is therefore the maximum based on large stationary gateways using additional expensive
range in ocean environments and location estimation techniques hardware. There are several projects run by SmartParks that
using LoRa and low-cost Internet of Things (IoT). use a similar system to protect wildlife, one of which tries to
protect the black rhinos from poachers in Tanzania.
II. R ELATED W ORK
Recent studies use low-cost controllers to detect and classify
In this IoT era, sensing and communicating is becoming cetacean vocal calls [Radeta, Nunes, Vasconcelos, and NisiR-
inexpensive, and is a favorable occasion to explore low-cost adeta et al.2018]. Also, Nikita and colleges used Raspberry
sensing and location estimation. This opens an opportunity Pi and Arduino UNO to build a simple ROV prototype
to obtain geotagged environment data and empower regular for a surveillance application [Pinjare, Chaitra, Shraavan,
citizens to use these technologies, previously only available to Naveen, et al.Pinjare et al.2017]. The Parrticle Photon, an
corporations or researchers. The areas of previous work that Arduino based integrated IoT platform, has been successfully
primarily drive this research are: (1) long range radio data used in biodiversity monitoring, and simple signal processing
communication; and (2) location estimation techniques without [Vasconcelos, Nunes, Ribeiro, Prandi, and RogersVasconcelos
satellite-assisted systems. et al.2019].
A following experiment, conducted by [Gogendeau, Murad,
Bernard, Kerzerho, and BonhommeauGogendeau et al.2018], B. Location Estimation Techniques without satellite-assistance
explored the different configurations of LoRa (spreading factor,
In general, several techniques can be used to estimate the
bandwidth, coding rate) in sea environments. Trying to obtain
position of ubiquitous devices. These techniques commonly use
the location of endpoints (using RSSI) which were fixed in
what has become a standard of GPS. However, GPS cannot be
8 coastal locations. They claim a maximum location error of
used in some applications due to hardware, power or location
100m at a maximum distance of 1.6km. Our study, builds upon
(e.g. indoor) constraints.
these findings, focusing in expanding the range much further.
Distance Estimation based on RSSI - Several research
A. Long Range Radio Communication has been done by using the signal strength in the form of
Most modern communication systems use either electricity RSSI [Elnahrawy, Li, and MartinElnahrawy et al.2004]. RSSI
or electromagnetism as a way to carry information. Several represents the relationship between a transmited and a received
technologies exist with ranges that go from a few meters power, used to calculate the distance between a transmitter and
(Infra-Red Transmitters, Bluetooth, Wi-Fi) to thousands of km a receiver when most of the signal propagates in a line-of-sight.
(Satellite), passing through those with a range of a few km It has the disadvantage of depending on the transmitted power,
(Mobile Phones, TV and Radio). Usually, the longer the range, thus not being applied to all hardware, however, it has the
the more restricted and expensive it becomes, greatly limiting advantage of being less costly and not requiring additional
its usage in IoT low-cost solutions. hardware.
Many studies have used LoRa or similar technologies in 2 https://www.smartparks.org/
urban and other land environments [Fargas and PetersenFargas 3 TimVanDam rotectingWildLi f eWithLoRaWAN
P
and Petersen2017] where it behaves significantly different 4 https://www.wildlabs.net/users/tim-van-dam/
 III. M ETHODOLOGY
We deployed 5 coastal based nodes (2 failed) and 2 sea
vessel nodes on the same vessel, for the duration of 3 days,
allowing us to test the range and location of the sea vessel.
A. System Apparatus
The system apparatus was based on 3 coastal nodes and 1
sea vessel node. Each node used a LoPy microcontroller, which
was placed into a casing. These LoPys were equipped with the
PySense expansion board granting us access to several sensors. Figure 1. Left: Bilateration theory [Cota-Ruiz, Rosiles, Sifuentes, and Rivas-
Three coastal nodes were deployed within an average distance PereaCota-Ruiz et al.2012]; Right: Bilateration example with the two solutions,
and an error of 1823 (excluding the solution located on land).
of 30km, at static locations facing the south of the Madeira
island, Portugal. Each node has been placed on top of a 3m
pole at an altitude higher than 50m from the sea level. Finally, And finally, P3 = (x3 , y3 ) in terms of P0 = (x0 , y0 ), P1 =
one node was mounted on top of the sea vessel, capturing (x1 , y1 ) and P2 = (x2 , y2 ), is:
the GPS location using a PyTrack. We used the following
settings for LoRa: Region: EU868; Transmitted power: 14 x3 = x2 + −h(y1 − y0 )/d
(3)
dBm; Bandwidth: 125 KHz Spreading Factor: 7. y3 = y2 − +h(x1 − x0 )/d
B. Sensory Input When the two circles do not intercept, we only know that the
From this apparatus, we gathered a total of 4 366 data points solution is along the line perpendicular to P0P1 with its center
starting at 18:00 hours and the following 40 hours, spanning to in the point P2. A relaxation can be made, to estimate that the
3 days, including a vessel stationary period between the hour solutions satisfy the Ri j = q j − qk − Ri k. Then averaging those
14 to 24. two solutions, a single solution would fall in the equivalent
of P2 when the circles do not intersect. The same principle
C. Location Estimation applies when one circle is contained inside the other.
We explored a basic location estimation using the RSSI. Although a higher degree multilateration would usually result
Since the RSSI is in a logarithmic scale, we can either derive in better solutions, in this case bilateration was chosen for 2
the linear equation for the data by: 1) turning the RSSI into a reasons: 1) the land-nodes are aligned in an almost straight line
linear scale or 2) turning the distance into a logarithmic scale. with a small curvature in-land. This causes the multilateration
We used the first approach, using a common formula (eq. 1) equations to near a singularity where it is highly unstable and
[Al AlawiAl Alawi2011] for calculating the RSSI: also tends to give false results inland. 2) the nodes’ RSSI values
are unstable and don’t provide coherent values throughout time,
RSSI = −(10 × n)log10 (d) − A (1) which translates in oscillating calculated radius from each node.
Issue 2) further exaggerates issue 1) leading to the usage of a
and reversing it to get the distance: d = 10RSSI/10
more stable, although less accurate solution: bilateration.
These equations use the RSSI in dBm, and the distance d
in meters and have tunning parameters such as n, the signal IV. R ESULTS
propagation constant and A being a reference received signal
In this section, we present our results, namely the maximum
strength in dBm (the RSSI value measured at 1m distance).
range obtained when sensing data from the sea (subsection
Figure 1 shows this geometrically, and the solution points are
4.1), the distance estimation based on RSSI error using the
defined as the following [Cota-Ruiz, Rosiles, Sifuentes, and
different data sources (subsection 4.2), location calculation and
Rivas-PereaCota-Ruiz et al.2012]:
errors from the bilateration (subsection 4.3), as well as the
• d > r0 + r1 → no solutions - circles are separated.
environmental telemetry (subsection 4.4).
• d < |r0 − r1 | → no solutions - one circle is contained
within the other. A. LoRa maximum distance
• d = 0 and r0 = r1 → the circles are coincident and there The maximum sustained distance captured by all 3 nodes
are an infinite number of solutions. was of 54.9km away from the shore with a minimum RSSI of
Figure 1 (left) shows this geometrically, and the solution -127. And the peak distance was captured by node Green with
points are defined as the following: a2 + h2 = r02 and b2 + h2 = a distance of 83.6km and an RSSI of -126. These results were
r12 Using d = a + b we can solve for a, and it can be readily captured when the vessel was going away from the coast in a
shown that this reduces to r0 when the two circles touch at straight line.
one point, i.e.: d = r0 r1 .
Solving for h by replacing a into the first equation, we get B. Distance Estimation from RSSI
h = r02 − a2 . Thus,
2
We modeled the data using both the raw logarithmic RSSI
P0 + a(P1 − P0 ) and distance values and applying the linear regression to them,
P2 = (2) and we also transformed the RSSI into a linear scale.
d
 Figure 3. Location estimation from RSSI: GPS ground truth (green); location
Figure 2. RSSI evolution of the 3 receivers by colors over the trip different estimation (yellow - the darker, the more overlapped points).
distances for the same data point. Original RSSI as a dashed line and the
moving average as a solid line.

Due the presence of outliers we use the RANSAC (RANdom

SAmple Consensus) method [DerpanisDerpanis2010] iteratively
using the minimum number of observations and generating
candidate solutions where the maximum residual/threshold for a
data sample to be classified as an inlier was the MAD (Median
Absolute Deviation). This threshold is a robust measure of
how spread out a dataset is. It uses the variance and standard
deviation, also measuring spread, however they are more
affected by extremely high or low values and non normality.
[Leys, Ley, Klein, Bernard, and LicataLeys et al.2013].
Figure 4. Estimated location errors (in meters) along the distance (as the
In figure 2 we can see the variations of the different RSSI vessel got further away)
signals corresponding to the same vessel location at the different
distances that the land nodes were located. While we can
observe a steady progression for the orange line, we also
notice many oscillations in the green and yellow even in the pattern of the location estimation with some estimations overlap
smoothed signal, which affect modeling. These oscillations are (represented by the transparency).
possibly resultant from the placement of the land nodesdue to
The location estimation resulted in the errors presented in
some mountains and land nearby hindering the line of sight
table II. We can observe that the minimal errors come from the
and the capture of the first signals.
Orange and Orange and Yellow using the LS method, followed
Table I shows the residual errors comparison for the different
by the Yellow. As expected from the previous section, any
combinations of datasets and models created, combining them
combinations that involved the green receiver, resulted in large
in pairs for the different models. We observe that the oscillations
errors, due to its model and noise.
from the Green and Yellow receivers (figure 2) influence
the inter-dataset data modeling, as the errors increase. In the Seeing the error in a relative perspective of the maximum
LoRa context and this oceanic setting, the mean errors have distance observed in section IV-A, when comparing to 83.6km
a relatively low impact on the system if we look at them as and 54.9km, for instance the Orange and Yellow models
a percentage of the maximum range of 83.6km and 54.9 km, have average error (Orange=5 657; Yellow=6 207, µ=5 932),
the average of the combined Orange && Yellow for instance which represents 7% and 10.8% respectively of the maximum
(Orange=1 578 m; Yellow=3 997 m; µ=2 788 m) represents distances possible we achieve.
only 3.3% and 5% respectively of the maximum distances. Figure 3 shows the estimated locations in comparison to the
GPS ground truth. While many points are very close to the
C. Modeling Location Estimation from IoT input GPS line, we can also see the deviations that occur along the
For the bilateration, we needed to choose two of the receivers, way, due to the inconsistency of the Yellow, in comparison to
and, as we noted in the previous section, the green receiver the Orange. In figure 4 we can see the progression of the error
has a large error which influences its model and any other over the data points, where with the bigger the distance, the
model that pairs with it. Hence, we decided to perform the bigger the error becomes. This comes from oscillations in the
bilateration using only the estimated distance from the receivers lower end of the RSSI range, which being logarithmic, small
Orange and Yellow. The location estimation was modeled using oscillations produce larges distance in the estimation. This
bilateration with an average RSSI of 4 points. The data in figure error could be diminished by obtaining more points for each
3 shows the estimation errors. Due to the low resolution of location, instead of just the four used in the moving average
RSSI, it creates an aliasing effect which results in a grid-like due to the sea vessel being in constant movement.
 V. D ISCUSSION not have a low-cost way to find the angle, but do for range,
we used multilateration with RSSI values, to find the distance
A. Research Findings
between the nodes. Since the RSSI is inherently sensitive to the
1) Maximum Distance using LoRa: During this study, we environment, even after post-processing, some noise remained,
achieved a maximum range of 83.6km while using LoRa as can be seen in figure 2. Despite that, our results show an
overseas. This range linked a land endpoint (at an altitude average distance error (compared to the GPS ground truth)
of 281m) to a boat’s endpoint in the middle of the ocean. This for the individual models that ranges from 1 359m (with a
is above the manufacturers’ range of 10 to 40 km but bellow standard deviation of 1 183 m) up to 5 749 m (with a standard
a record that used a helium balloon to rise the endpoint up deviation of 4 754 m) which represents 3.5% to 7% of the
to 38 km of altitude before transmitting a packet to 702 km maximum range measured. The results also show individual
away with a transmit power similar to ours5 . The maximum errors points with a minimum error of 1 m (that are very close
simultaneous range from all the land endpoints to the boat to the estimated regression line) and a maximum error of 25
endpoint was of 54.9 km (at altitudes ranging from 57 m to 070 m from a very noisy endpoint. Excluding this endpoint, the
281 m), within predicted ranges, achieved using LoPy devices maximum error is less than 10km. By combining the dataset
coupled with a 1/4 length 868MHz LoRa monopoles. No high- from two endpoints, we managed to slightly improve the worst
end and expensive gateways or antennas were used in the setup. endpoint range estimation at the cost of the other endpoints.
We can all but speculate that the achieved higher ranges were This evidentiates how sensitive the model is to noise, but it also
due to any combination of the bellow as well as any other suggests that solutions such as a higher number of endpoints
unforeseen factor: or a moving average, do reduce the error.
- Having perfect line-of-sight without obstacles or reflections.
The land points were mounted facing the sea, with no obstacles; 3) Location Estimation using Bilateration: In our case
- The altitude of the land endpoint vs the ocean endpoint. The we were forced to use bilateration. As we explained in
longest range was achieved from the highest endpoint (83.6km chapter 3, the two reasons are: 1) the land endpoints are
range from 281m high) and there is a trend of declining range aligned in an almost straight line with a small curvature in-
with the altitude (63.8km range from 185m high and 54.9 km land. And 2) the nodes’ RSSI values are unstable and don’t
range from 57m high). At 281 m high, the distance to the provide coherent values through the time, resulting in different
equator, in direct line of sight, is about 60km, which is the estimated distances for each node. We can fix 1) in the future by
majority of whole range, meaning that the signal still has to better placing or adding more nodes making a better geometry.
travel 23km over the horizon to get to the 83.6km range. As for 2), would require a more expensive setup or better
- An improved build quality that was achieved over time. The hardware. Using the post processed RSSI range values and
LoRa technology was patented in 2008 and has had time to bilateration from two land endpoints, we calculated the boat
mature and improve since then. endpoint’s location. We then compared those values to GPS
2) Distance Estimation with LoRa: One of our focus on this derived values to understand the accuracy and quality of our
study was to use a low tech, low energy solution to find location. results as seen in figure II. The geolocation error ranges from
This ruled out the power hungry satellite based technology and an average of 5 657m for the model that uses just the orange
the expensive ToF based technologies. We were left out with endpoint to an average geolocation error of 21 531 when using
multilateration or multiangulation based technologies. We do the very noisy green model. For individual points in the dataset,
the results have a minimum geolocation error of 218 m and a
5 https://www.thethingsnetwork.org/article/ground-breaking-world-record- maximum geolocation error of 33 461m, again, for the green
lorawan-packet-received-at-702-km-436-miles-distance model. If we exclude the green model, the maximum error

Table I
E RROR COMPARISON ( IN METERS ) FOR THE DIFFERENT MODELS USED AFTER LINEARIZING THE RSSI

Error
Modeled Data Nr Error LS
Dataset RANSAC
Sources samples
µ σ min max µ σ min max
Orange 251 1 359 1 183 1 7 156 1 359 1 183 1 7 156
Individual Model Yellow 221 3 595 2 135 115 9 542 3 616 2 591 7 10 635
Green 363 5 749 4 754 3 25 070 5 718 4 861 61 25 820
Orange 251 1 578 1 361 6 7 635 1 578 1 361 6 7 635
Orange & Yellow
Yellow 221 3 997 2 553 25 11 300 3 997 2 553 25 11 300
Yellow 251 9 722 4 218 2 023 19 526 14 566 5 633 110 25 247
Yellow & Green
Green 363 8 979 4 785 8 19 718 7 143 5 110 25 23 876
Orange 251 7 793 4 660 151 19 703 9 169 6 270 39 24 258
Orange & Green
Green 363 7 288 3 937 76 16 963 6 514 4 935 8 23 106
 Table II R EFERENCES
C OMPARISON OF LOCATION ESTIMATION ERRORS ( IN METERS ) FOR THE
DIFFERENT MODELS USED FOR BILATERATION [Al AlawiAl Alawi2011] Raida Al Alawi. 2011. RSSI based location
estimation in wireless sensors networks. In 2011 17th IEEE International
Conference on Networks. IEEE, 118–122.
LS RANSAC [Cermeño, Quílez-Badia, Ospina-Alvarez, and BlockCermeño et al.2015]
µ σ µ σ Pablo Cermeño, Gemma Quílez-Badia, Andrés Ospina-Alvarez, and
Barbara A Block. 2015. Electronic tagging of Atlantic bluefin
Ora. 3 432 2 013 3 432 2 013 tuna (Thunnus thynnus, L.) reveals habitat use and behaviors in the
Yel. 6 213 3 696 7 513 4 356 Mediterranean Sea. PLoS One 10, 2 (2015), e0116638.
Green 21 313 3 812 22 046 3 848 [CommitteeCommittee[n. d.]] LoRa Alliance Strategy Committee. [n. d.].
Orange && Yellow 4 067 2 170 4 067 2 170 LoRaWAN Geolocation Whitepaper.
Yellow && Green 12 324 2 422 20 279 2 422 [Cota-Ruiz, Rosiles, Sifuentes, and Rivas-PereaCota-Ruiz et al.2012] Juan
Orange && Green 16 343 3 724 21 650 5 989 Cota-Ruiz, Jose-Gerardo Rosiles, Ernesto Sifuentes, and Pablo
Orange && Yellow && Green 10 631 2 399 4 550 1 206 Rivas-Perea. 2012. A low-complexity geometric bilateration method
for localization in wireless sensor networks and its comparison with
least-squares methods. Sensors 12, 1 (2012), 839–862.
[DerpanisDerpanis2010] Konstantinos G Derpanis. 2010. Overview of the
is around 15km. Since our bilateration derives, directly, from RANSAC Algorithm. Image Rochester NY 4, 1 (2010), 2–3.
the distance estimated using RSSI, all the improvements and [Elnahrawy, Li, and MartinElnahrawy et al.2004] Eiman Elnahrawy,
Xiaoyan Li, and Richard P Martin. 2004. The limits of localization
pitfalls are shared. Although in absolute values these may seem using signal strength: A comparative study. In 2004 First Annual
large, in open ocean, with an average error of around 3.4km, IEEE Communications Society Conference on Sensor and Ad Hoc
it is possible to locate objects with the naked eye, thus this Communications and Networks, 2004. IEEE SECON 2004. IEEE,
406–414.
location estimation proving to be useful in such scenarios. [Fargas and PetersenFargas and Petersen2017] Bernat Carbonés Fargas and
Improvements may be done to remove the outliers from the Martin Nordal Petersen. 2017. GPS-free geolocation using LoRa in
dataset when testing against the models, reducing the error in low-power WANs. In 2017 Global Internet of Things Summit (GIoTS).
IEEE, 1–6.
30-40%, however, as in real scenarios, we would not know in [Gogendeau, Murad, Bernard, Kerzerho, and BonhommeauGogendeau et al.2018]
advance which values were outliers. Pierre Gogendeau, Nour Murad, Serge Bernard, Vincent Kerzerho,
and Sylvain Bonhommeau. 2018. Oversea Radio Measurements
B. Contributions and Channel Characterization with LoRa Technology. 1–2.
https://doi.org/10.23919/RADIO.2018.8572461
We contribute with nearly 84km of LoRa maximum distance [Guegan, Murad, Lebreton, and BonhommeauGuegan et al.2017] L.
and a location error of 3.4km (4 % of maximum distance) Guegan, N. M. Murad, J. M. Lebreton, and S. Bonhommeau.
for the long range tests. Nevertheless, other short range 2017. Integrating over sea radio channel for sea turtles localization in
the Indian Ocean. In 2017 IEEE Radio and Antenna Days of the Indian
studies (max 2km) claim an accuracy of 100m (5% of Ocean (RADIO). 1–2. https://doi.org/10.23919/RADIO.2017.8242222
distance) [Guegan, Murad, Lebreton, and BonhommeauGuegan [Joo, Lee, Lee, Sin, Lee, and KimJoo et al.2008] Inone Joo, Jeom-Hun Lee,
et al.2017]. When using the real-time location of marine Young-Min Lee, Cheon Sig Sin, Sang-Uk Lee, and Jae Hoon Kim. 2008.
Development and performance analysis of the second generation 406
species, 3.4km is adequate when studying migration flows. MHz EPIRB. In 2008 4th Advanced Satellite Mobile Systems. IEEE,
Our main focus in this study was to use an IoT low cost 333–336.
and energy efficient LoRa solution that could be used in sea [Leys, Ley, Klein, Bernard, and LicataLeys et al.2013] Christophe Leys,
Christophe Ley, Olivier Klein, Philippe Bernard, and Laurent Licata.
environments, to aid the research and conservation of marine 2013. Detecting outliers: Do not use standard deviation around the mean,
life. We explored LoRa, how it behaves over the ocean and the use absolute deviation around the median. Journal of Experimental
extraction of geolocation of nodes using the RSSI that comes Social Psychology 49, 4 (2013), 764–766.
[P. M. HallP. M. Hall1980] M P. M. Hall. 1980. Effects of the troposphere
at no cost in any kind of radio communication. We delivered on radio communication. Stevenage Herts England Peter Peregrinus Ltd
LoRa packets at a maximum distance of 83.6km, much more IEE Electromagnetic Waves Series 8 (01 1980).
than the manufacturer claimed range (40km for Pycom LoPy4) [Pinjare, Chaitra, Shraavan, Naveen, et al.Pinjare et al.2017] Nikita S Pin-
jare, S Chaitra, S Shraavan, IG Naveen, et al. 2017. Underwater remotely
as well as most literature. operated vehicle for surveillance and marine study. In 2017 International
We also modeled a bilateration and RSSI based geolocation Conference on Electrical, Electronics, Communication, Computer, and
that, even though is far from the accuracy of the modern Optimization Techniques (ICEECCOT). IEEE, 330–334.
satellite based technology, is low-cost both in terms of [Radeta, Nunes, Vasconcelos, and NisiRadeta et al.2018] Marko Radeta,
Nuno J Nunes, Dinarte Vasconcelos, and Valentina Nisi. 2018.
hardware/software as in terms of energy usage. The geolocation POSEIDON-Passive-acoustic Ocean Sensor for Entertainment and
had an average error of approximately 5% of the maximum Interactive Data-gathering in Opportunistic Nautical-activities. In
range, or, about 3.4 km. This error is adequate for study of Proceedings of the 2018 on Designing Interactive Systems Conference
2018. ACM, 999–1011.
migration patterns, the general location of animals and other [Rynne and Von EllenriederRynne and Von Ellenrieder2008] PF Rynne and
situations where a pin-point location is not needed. KD Von Ellenrieder. 2008. A wind and solar-powered autonomous
surface vehicle for sea surface measurements. In OCEANS 2008. IEEE,
ACKNOWLEDGEMENTS 1–6.
[Vasconcelos, Nunes, Ribeiro, Prandi, and RogersVasconcelos et al.2019]
Authors are thankful to the Funchal Marina and NDR Douro Dinarte Vasconcelos, Nuno Nunes, Miguel Ribeiro, Catia Prandi, and
sea vessel crew allowing the study. Several grants supported Alex Rogers. 2019. LOCOMOBIS: a low-cost acoustic-based sensing
the study: (i) LARGESCALE, 32474 by FCT and PIDDAC; (ii) system to monitor and classify mosquitoes. In 2019 16th IEEE Annual
Consumer Communications & Networking Conference (CCNC). IEEE,
LARSyS, UID/EEA/50009/2019 by FCT; (iii) INTERTAGUA, 1–6.
MAC2/1.1.a /385 MAC INTER-REG 2014-2020; and (iv)
MITIExcell, M1420-01-01450FEDER0000002 by RAM.

View publication stats