GTI 5G Device Power Consumption Whitepaper

GTI 5G Device
Power Consumption
White Paper

1
http://gtigroup.org/
 GTI 5G Device Power Consumption Whitepaper

GTI 5G Device Power Consumption

White Paper

Version: V4.0

Deliverable Type □ Procedural Document

√ Working Document

Confidential Level √ Open to GTI Operator Members

√ Open to GTI Partners

□ Open to Public

Program 5G eMBB

Working Group Terminal WG

Project Project 3: New Device

Task 5G device Power Consumption Task Force

Source members China Mobile, VIVO, MTK, Smarter micro, Qorvo, OPPO,
Anritsu, Keysight, XIAOMI, Intel

http://gtigrou
2
 GTI 5G Device Power Consumption Whitepaper

Support members Quectel, Unisoc, SIMCom,Starpoint,Datang Mobile, Taiyo

Yuden, Qualcomm, Rohde-Schwarz.

Editor Jianwei Zhang, Nader Atteya ,Weide Wu,ZhangYuan, Simon

George-Kelso, Francesc Boixadera, Hu Hao, Jinming Zhao,
Guiying Wang, YangyangPeng, Lawrence.Tao, Xiaowei Jiang,
Luting Kong, Jiang Bo, Zhihua Shi, TT Chiang, Shaoliang Tang.

Last Edit Date 22-10-2020

Approval Date 22-10-2020

Confidentiality: This document may contain information that is confidential and access to this
document is restricted to the persons listed in the Confidential Level. This document may not be
used, disclosed or reproduced, in whole or in part, without the prior written authorization of GTI,
and those so authorized may only use this document for the purpose consistent with the
authorization. GTI disclaims any liability for the accuracy or completeness or timeliness of the
information contained in this document. The information contained in this document may be
subject to change without prior notice.

Document History

3
 GTI 5G Device Power Consumption Whitepaper

Date Meeting # Version # Revision Contents

21-01-2019 24th GTI V1.0 The first version of GTI 5G Device Power
Workshop Consumption Whitepaper.

03-06-2019 25th GTI V2.0 Update 5.3 parameter configuration

Workshop recommendation

19-06-2019 25th GTI V2.01 Formatting edits

Workshop

06-11-2019 26th GTI V3.0 5.4 Enhancement of Power Saving in Release 16

Workshop

22-10-2020 29th GTI V4.0 5.4 Enhancement of Power Saving in Release 16

Workshop
5.5 Enhancement of Power Saving in Release 17

4
 GTI 5G Device Power Consumption Whitepaper

Table of Contents
GTI 5G Device Power Consumption White Paper................................................................................2

Document History..............................................................................................................................3

Table of Contents................................................................................................................................. 5

1 Executive Summary.......................................................................................................................... 8

2 Abbreviations....................................................................................................................................8

3 Introduction....................................................................................................................................11

4 Key components of power consumption....................................................................................... 11

4.1 RFFE..................................................................................................................................... 12

4.1.1 Basic Information of PA............................................................................................15

4.1.2 Methods of improving the PA efficiency................................................................. 17

4.2 BP.........................................................................................................................................20

4.3 AP.........................................................................................................................................21

4.4 RFIC......................................................................................................................................21

4.5 Screen.................................................................................................................................. 22

5 Factors of power consumption - 5G features................................................................................ 24

5.1 5.1 RRC idle state and RRC inactive state............................................................................24

5.1.1 Bandwidth................................................................................................................ 24

5.1.2 Paging....................................................................................................................... 24

5.1.3 Measurement...........................................................................................................25

5.1.4 System information acquisition (including synchronisation).................................. 25

5.1.5 RNA update.............................................................................................................. 27

5.2 RRC connected state........................................................................................................... 27

5.2.1 Bandwidth................................................................................................................ 27

5.2.2 UL-MIMO..................................................................................................................29

5.2.3 HPUE.........................................................................................................................30

5.2.4 BWP.......................................................................................................................... 31

5
 GTI 5G Device Power Consumption Whitepaper

5.2.5 DRX........................................................................................................................... 34

5.2.6 Cross-slot scheduling............................................................................................... 36

5.2.7 Periodic PDCCH monitoring..................................................................................... 37

5.3 Parameter configuration recommendation........................................................................ 37

5.3.1 3GPP power model.................................................................................................. 38

5.3.2 Applying the 3GPP power model to NSA.................................................................41

5.3.3 Modelling scenarios................................................................................................. 42

5.3.4 Standalone baseline scenarios.................................................................................43

5.3.5 Standalone power saving evaluation.......................................................................47

5.3.6 Nonstandalone comparison.....................................................................................60

5.4 Enhancements of Power Saving in Release 16................................................................... 63

5.4.1 BWP adaptation framework.................................................................................... 64

5.4.1.1 PDCCH Monitoring Reduction...............................................................................64

5.4.1.2 SCell dormancy......................................................................................................64

5.4.1.3 MIMO layer adaptation.........................................................................................66

5.4.1.4 Cross-slot scheduling............................................................................................ 66

5.4.2 Wakeup signal for cDRX optimization......................................................................67

5.4.3 RRM measurements relaxation............................................................................... 68

5.4.4 Extended UE Assistance information.......................................................................69

5.4.4.1 DRX Preference..................................................................................................... 69

5.4.4.2 Max BW and Max CC Preference.......................................................................... 69

5.4.4.3 Minimum Scheduling Offset Preference...............................................................70

5.4.4.4 Release Preference for efficient RRC state transitions......................................... 70

5.4.5 SECONDAY DRX.........................................................................................................70

5.4.6 Proposed Features Combination for different traffic profiles:................................71

5.5 Enhancements of Power Saving in Release 17................................................................... 74

5.5.1 NR Idle-Mode Power Consumption Issue................................................................74

6
 GTI 5G Device Power Consumption Whitepaper

5.5.2 Paging Enhancements for UE Power Saving in NR...................................................75

5.5.3 TRS/CSI-RS for idle/inactive UE................................................................................77

5.5.4 Extension to Rel-16 DCI-based power saving adaptation....................................... 78

5.5.5 Relaxing UE Measurements for RLM/BFD............................................................... 80

5.5.6 Sending Data in RRC inactive state.......................................................................... 81

6 Requirement of 5G terminal power consumption........................................................................ 82

6.1 NSA...................................................................................................................................... 82

6.1.1 RRC idle state and RRC inactive state...................................................................... 82

6.1.2 RRC connected state................................................................................................ 84

6.2 SA.........................................................................................................................................87

6.2.1 RRC idle state and RRC inactive state...................................................................... 87

6.2.2 RRC connected state................................................................................................ 88

6.3 Estimating power consumption in NR UEs......................................................................... 90

7 Service............................................................................................................................................ 92

7.1 Service type and Parameter configuration......................................................................... 92

7.1.1 FTP model 3..............................................................................................................92

7.1.2 Game........................................................................................................................ 93

7.1.3 Video........................................................................................................................ 94

7.1.4 Voice......................................................................................................................... 95

7.1.5 Other Applications................................................................................................... 95

7.2 User model.......................................................................................................................... 98

7.3 5G industry application power analysis.............................................................................. 98

7.3.1 7.3.1 Laptop..............................................................................................................98

8 Power consumption test................................................................................................................ 98

8.1 Test instrumentation .......................................................................................................... 98

8.2 Test method.......................................................................................................................101

9 References.................................................................................................................................... 104

7
 GTI 5G Device Power Consumption Whitepaper

1 Executive Summary
5G brings wider bandwidth, higher data rate and lower delay, but it also brings new challenges to
the power consumption of 5G devices. This whitepaper provides the analysis of the factors of
power consumption, the test solution and the performance requirements of power consumption
for 5G device.

First of all, the power consumption of the components that have the greatest impact on terminal
power consumption is introduced, including RFFE (such as PA, AD/DA, filter), RFIC, BP, AP and
screen parts, and analyze the power consumption performance of the main solutions in the
current 5G component industry. Secondly, based on the 5G technical features, the possible power
consumption problems (such as wider bandwidth, uplink dual transmission, etc.) are discussed,
and the 5G power saving schemes (such as BWP, cross-slot scheduling, DRX, periodical PDCCH
monitoring). At the same time, based on the analysis of the power consumption performance of
the components, the influence of 5G technical features on power consumption is deduced
preliminarily. Eliminate the unnecessary technical requirements which seriously affect the device
power consumption and have no obvious benefit. Summarize the key technical characteristics
that affect the 5G device power consumption, pending further test and verification. Thirdly,
based on the analysis results of typical service types and user models, as well as the key technical
characteristics affecting 5G terminal power consumption, we will delimit the test scope and
complete the test cases. Discuss the specific test scheme with the instrument manufacturer. The
instrument manufacturer is responsible for the test case development and recommends the
power test scheme suitable for 5G terminal. Fourthly, to build the 5G terminal test environment.
Based on the previous 5G terminal test scheme, the key technical features affecting the power
consumption of 5G terminal are further tested and verified. Finally, the key technologies suitable
for 5G terminal are selected. Finally, based on the test results, the power performance
requirements of 5G terminal are finalized, including the power requirements of typical service
types in RRC connection state and the power requirements of terminal in RRC idle state and RRC
inactive state.

In the early stage, the analysis of power consumption for sub6G eMBB terminal and the vertical
terminal is the main part, the start-up time of the research of power consumption for millimeter
wave device depends on the industry maturity.

2 Abbreviations
Abbreviation Explanation
3GPP 3rd Generation Partnership Project

AR/VR augmented reality /virtual reality

AP applications processor

BP baseband processing

BWP Bandwidth part

DL Downlink

8
 GTI 5G Device Power Consumption Whitepaper

DRX Discontinuous Reception

eMBB Enhanced Mobile Broadband

eMBMS Enhanced Multimedia broadcast multicast services

ETSI European Telecommunication Standardisation Organisation

FCC Federal Communications Commission

GTI Global TD-LTE Initiative

HPUE High power user equipment

IMT International Mobile Telecommunication

ITU International Telecommunication Union

ITU-R International Telecommunication Union - Radio

LTE Long Term Evolution

MNO Mobile Network Operator

MWC Mobile World Congress

MOOD MBMS Operation On Demand

MIMO Multiple input – Multiple output

NSA Non-standalone

NR New Radio

OAM Operation, Administration and Maintenance

ODU LTE outdoor CPE

OEM Original Equipment Manufacturer

ODM Original Device Manufacturer

QoS Quality of Service

PDCCH Physical downlink control channel

RAN Radio Access Network

RRM Radio Resource Management

RFFE RF Front End

RFIC Radio frequency integrated circuit

SA Standalone

TD-LTE Time Division Long Term Evolution

9
 GTI 5G Device Power Consumption Whitepaper

TDD Time Division Duplex

TTI Transmission time interval

UE User equipment

UL Uplink

10
 GTI 5G Device Power Consumption Whitepaper

3 Introduction
One of the 5G challenges is the power consumption of 5G devices. This whitepaper provides the
analysis of the factors of power consumption, such as the key components /the 5G feature and
the service type/the test solution and the performance requirements of power consumption for
5G device. In the early stage, the analysis of power consumption for sub6G eMBB terminal and
the vertical terminal is the main part, the start-up time of the research of power consumption for
millimeter wave device depends on the industry maturity.

4 Key components of power consumption

Comparing 5G and 4G, the 5G device needs to support new bands, wider bandwidth, more
antennas, higher transmit power, new RAT and new services. The block diagram of the 5G device
is shown in Figure 4- 1. The new features might affect the power consumption of the blocks, such
as antenna tuning unit, RFFE, RFIC, BP and AP. The factors of power consumption should be
considered in the design and the development of the key components. The factors of power
consumption of key components are analyzed and the trends of the solutions are discussed.

Figure 4- 1 5G Device Block Diagram

The factors of power consumption for each component are listed in following in the aspect of 5G
features. The change of the components power consumption is analyzed from the point of 5G
feature.

Table 4- 1 Factors of Power Consumption - 5G Features vs. Key Component

5G Features ATU RFFE RFIC BP AP

Bandwidth √ √ √ √ √

UL-MIMO √ √ √ √ √

11
 GTI 5G Device Power Consumption Whitepaper

HPUE √ √ √ √ √

BWP √ √ √ √

DRX √

Cross-slot scheduling √

Multiple PDCCH monitoring √

periodicities

Measurement √

System information acquisition √

Paging √

RNA update √

4.1 RFFE
RF Front End Integration can reduce loss and improve power consumption

Integrated RF Front End as below is the best approach for 5G to save power consumption

12
 GTI 5G Device Power Consumption Whitepaper

n77 PA efficacy inside the module.

Digital Predistortion Technology can further save power consumption.

13
 GTI 5G Device Power Consumption Whitepaper

Simple memoryless DPD.

DPD enables dramatic power reduction

- 30% current reduction

- 30% PAE for 5G Waveform in APT mode.

DPD increases linear power range by 1 dB.

Digital Predistortion Technology can also further improve linearity.

Digital Predistortion Technology improve both power consumption and linearity.

14
 GTI 5G Device Power Consumption Whitepaper

4.1.1 Basic Information of PA

The basic topology of power amplifier in handset devices can be shown as below.

Figure 4- 2 Basic topology of power amplifier

For the power amplifier, the efficiency (always noted as Power Add Efficiency or PAE) is always
used to represent the power consumption character. The relationship between the efficiency and
power consumption can be calculated as:

15
 GTI 5G Device Power Consumption Whitepaper

OUT OUT
PAE = =

There are many parameter related with the efficiency of the power amplifier, include back-off
power, loadline loss, topology, etc. Before talking about how to improve the efficiency, we need
to analyze the relationship between efficiency and other parameters.

For the loadline loss, the loss of the loadline consumes the power. The power consumed by the
loadline in dBm equals to the number of the loadline loss in dB. The following table shows the
relationship between the potential of the transmitted power in percentage and the loadline loss
in dB.

Table 4- 2 the relationship between transmitted power efficiency (%) and loadline loss (dB)

Load line loss (dB) Transmitted Power Efficiency Lost Power Efficiency

-0.1 98% 2%

-0.2 95% 5%

-0.3 93% 7%

-0.4 91% 9%

-0.5 89% 11%

-0.6 87% 13%

-0.7 85% 15%

-0.8 83% 17%

-0.9 81% 19%

For a typical loadline, the loss is between 0.3dB to 0.7dB, which means will degrade the PAE by 7
to 15%.
16
 GTI 5G Device Power Consumption Whitepaper

Another factor effects the efficiency of the power amplifier is the back-off power. For the
topology of traditional linear power amplifiers like Class-A/Class-AB/Class-F, the relationship
between the PAE and the output power can be shown in the following picture:

Figure 4- 3 Typical relationship between efficiency and output power

As the picture shows, the PAE decreases with the power back-off from the peak power.

4.1.2 Methods of improving the PA efficiency

Based on the previous consideration, there are several methods of improving the efficiency.

 Average Power Tracking

Average Power Tracking (APT) is a technique that can be utilized for vary the supply voltage to a
power amplifier on a timeslot basis in order to reduce power consumption of the PA.

With APT method, the supplied voltage of the power amplifier is adjusted according to output
power level so that linearity of the power amplifier is maintained while the efficiency is
improved.

But limited by the speed of the DC-DC converter, the APT voltage must be kept constant in the
same time slot. Which means the efficiency cannot be improved if the power amplifier works at
the highest power constantly.

 Envelope Tracking

The envelope tracking approach is one recommended power supply technique that maximizes
the energy efficiency of the PA by keeping it in compression over the whole modulation cycle,
instead of just at the peaks, by dynamically adjusting the supply voltage to the PA.

The ET technique was first developed in 1930s to handle the excessive energy consumption of
high-power amplitude modulation broadcast radio transmitters. Since constant amplitude ns and
frequency modulation techniques displaced AM in the 1950s, ET became marginalized and
17
 GTI 5G Device Power Consumption Whitepaper

irrelevant to engineering, where it languished as an academic curiosity. However, the increasing

PAPR of modern signals with advanced digital modulation schemes such as 4G reinvigorate ET
technique to achieve considerable energy saving in high-PAPR digital transmitters. More
commercial ET Pas have been developed for modern 4G/5G wireless communications and
beyond.

Figure 4- 4 Envelope Tracking RF PA system

 High Voltage Supply

With higher supply voltage, the portion of knee voltage of the transistors can be minimized,
which means the power amplifier will get higher efficiency.

The other advantage of high supply voltage is the loadline can also be increased. With the
increasing of the loadline, the loadline loss will also be lower. This also can improve the efficiency.

But because the supply voltage of the handset is 3.8V at nominal condition, another DC-DC boost
is needed to get a high voltage. It should be noted that the DC-DC’s efficiency is not 100%.

 Power Combination Techniques

The Power amplifiers usually need to support very large output power. So, power combination
techniques are dispensable techniques to combine small power cells together.

Because the combiners are always at the last stage of the power amplifier, the efficiency of the
combiners will affect the power amplifier’s efficiency directly. There are many power
combination techniques, includes voltage combination, current combination. No matter which
combination techniques, low loss, high Q devices must be used.

With the combination techniques, the push-pull combiner has the advantage of doubling the
impedance which as to be matched. This will lower the loss of the impedance transformation.

 The Doherty Amplifier

The Doherty amplifier was first proposed in 1963. The Doherty amplifier uses a configuration
called active load-pull technique, which can modify the RF load of the power amplifier in
different RF power levels. When the power back-offs from the peak power, the efficiency won’t
drop too much. Comparing with the traditional power amplifier topology, the power back-off
efficiency can be improved.
18
 GTI 5G Device Power Consumption Whitepaper

In the Doherty amplifier includes two devices, which can be called “main” and “auxiliary” devices.
The final maximum RF output power is the combined power of both devices. As the input drive
level is reduced, both devices contribute to the output power until a certain point is reached,
typically 6dB down from the maximum composite power, where the auxiliary amplifier shuts
down and generates no more RF power; it is assumed that it will also cease to draw DC as well.

Figure 4- 5 Basic topology of Doherty PA

In the Doherty amplifier, typical relationship between the power and efficiency can be shown as
below. With the picture shows, the power back-off efficiency can be improved obviously.

Figure 4- 6 Efficiency of Doherty PA

The Doherty amplifier also has some disadvantages. First, the AM-AM and AM-PM of the
amplifier cannot maintain constant as the auxiliary amplifier turning on and off with the changes
of the input power. So, digital pre-distortion method is always adopted in the Doherty power
amplifier. Second, the performance of the Doherty amplifier is very sensitive to the load. Based
on those characters, Doherty amplifier topology is always adopted in the base station power
amplifier design.
19
 GTI 5G Device Power Consumption Whitepaper

 The Reconfigurable Technique

The reconfigurable technique allows the power amplifier to adapt to different configuration in
different bands and different modes. Smarter Micro has applied this technique into power
amplifier design. With this technique, the loadline, bias point, and other configurations can be
adjusted. Comparing with wide-band design, this technique will improve the efficiency at
specified modes.

Figure 4- 7 Topology of reconfigurable amplifier

4.2 BP
The baseband processing in an NR UE modem has to handle a significantly increased volume of
data compared to its LTE counterpart, and this is reflected in increased power consumption at the
highest data rates. In an earlier study [1], estimates were made of the likely baseband
contribution to UE power consumption at peak throughput. These are reproduced below for
reference.

Table 4- 3 Estimated baseband processing power for different UE configurations

UE configuration Power at peak Power at peak throughput

throughput (FDD) (TDD)

2x100MHz, 4x4 4500mW 2970mW

DL, 2x2 UL

1x100MHz, 4x4 2250mW 1485mW

DL, 2x2 UL

That study was based on a very simple set of basic assumptions – in practice a 2 carrier baseband
does not require double the power of a single carrier baseband, as not all resources are
duplicated. More recent results suggest that these estimates were reasonably accurate, being
pessimistic for the 2 carrier UE, but optimistic for the 1 carrier UE.
20
 GTI 5G Device Power Consumption Whitepaper

However, power consumption at peak throughput is far from the whole story. Even in an active
connection, a UE will spend a large proportion of its time monitoring the downlink control
channel in TTIs which do not carry any data for it. It is therefore important that in this
PDCCH-only state, power consumption is kept to a minimum.

This can be achieved by a combination of cross-slot scheduling, bandwidth part adaptation and
MIMO restriction.

The UE can further reduce its average power consumption in connected mode (at the expense of
higher latency) by entering a DRX cycle to reduce the time spent monitoring PDCCH. These
techniques are discussed further in sections 5 and 6

4.3 AP
The applications processor in a smart phone comprises a number of processing cores and
graphical processors which cooperate to support the computational requirements of the active
applications. Multiple cores support a mixture of clock rates and processing capabilities, so
power consumption can vary significantly depending on the number of cores that are active and
the set of applications that are running. The highest processing loads are usually associated with
display-intensive applications, and the combined power consumption of the applications
processor and display can exceed 1000mW in some scenarios, even in flight mode.

AP power consumption can in many cases be considered independently from the UE modem
power. Present day application data requirements are relatively modest in NR terms – even
streaming of high definition video to the UE uses a comparatively small proportion of the data
bandwidth that is available. However, the applications themselves can have a significant impact
on modem power consumption. Interactive gaming applications can involve frequent transfer of
small packets of data to give a real-time response, and if the update frequency falls within the
inactivity timer period of the DRX cycle the modem will spend most of its time awake with
increased power consumption. Updates from different applications, if their timing is not
coordinated, increase the proportion of DRX cycles in which data is active and thereby increase
power consumption.

With the increased bandwidth offered by NR it seems inevitable that new applications such as
augmented reality gaming will combine intensive application processing with simultaneously high
on-air transfer rates. Providing a good experience to mobile users who have to rely on finite
battery capacity will require cooperation between application developers, UE manufacturers and
network providers.

4.4 RFIC
The receivers and transmitters in the NR RFIC must operate over a wide range of frequencies, and
a single implementation that will cover more than 2 decades (600MHz – 86GHz) of the RF
21
 GTI 5G Device Power Consumption Whitepaper

spectrum is an unrealistic objective. Individual receivers and transmitters in the RFIC will be
optimised for different sub-bands, and enabled as necessary for cell search and selection.

Power requirements will vary according to the frequency band and the number of active paths. In
the earlier study [1], estimates of power consumption were made for a 100MHz FR1 carrier, and
these are reproduced below in Table 4- 4.

Table 4- 4 Estimated transceiver power for a 100MHz FR1 carrier

MIMO configuration Active power

100MHz downlink 4x4 200-250mW

100MHz uplink 2x2 150-200mW

There is very little information so far in the public domain on FR2 power consumption, but higher
losses at mmWave frequencies, higher bandwidth in the transceiver and more complex
beamforming will all contribute to increased power consumption in this region of the spectrum.

Reducing the number of active receive or transmit paths will reduce power consumption,
although the reduction is generally less than pro-rata – halving the number of paths might reduce
power by around 30%.

Reducing the bandwidth and sampling rate can also reduce power in the transceiver, but the
savings are generally more noticeable in the downlink, where they are accompanied by a
reduction in baseband processing power, than in the uplink where the PA power contribution is
often dominant.

The most significant savings in transceiver power come from duty cycle reductions resulting from
a good DRX configuration.

4.5 Screen
 Screen market status and trends

For electronics products, the material of the screen determines the display effect of the product
to a large extent. If the screens are categorized by their material, the major screens of the
smartphones can be divided by their material into two categories: one is LCD (Liquid Crystal
Display) and the other is OLED (Organic Light-Emitting Diode). TFT and SLCD, which are more
common in the market, belong to the category of LCD. Lightweight, flexible, higher resolution,
larger size and other requirements will affect the future direction of mobile display industry.

The 4K resolution is defined as 4096 pixels in horizontal and the aspect ratio is still 16:9. There
are two main types of 4K resolution: 4096 * 2160, which is widely used in the field of digital
movies; 3840 * 2160, which is basically the 4K resolution we see in TV, displays, mobile phones
and other daily devices. There are twice as many pixels for 3840 * 2160 in both horizontal and
vertical directions comparing with the common FHD (1920 * 1080, 1080p). So the total pixel
number is four times that of FHD. Since the number of pixels is four times that of FHD, driving a
4K screen can be a huge resource drain (especially GPU). When the screen's PPI (pixels per inch)
is greater than 300, the human eye can't distinguish the pixels on the screen, such as retina

22
 GTI 5G Device Power Consumption Whitepaper

screen (326 ppi). The 1080p screen (over 400 ppi) is clearer than the retina screen, so is 1440p
(2560 * 1440, 2K). For 5.5 inch 4K, the number of ppi has reached an exaggerated 806. The
display screen of VR application is much closer to the eye coupled with the lens effect and the
left and right eye images are rendered separately. Even if the screen ppi reaches 400 or even 500,
users can still feel the obvious sense of granularity. Thus, 4K device is a better choice. At a
distance of 20 cm, the accuracy difference is negligible between 1080p and 2K. It is difficult to tell
exactly what resolution it is by simply giving a small 1080p or 2K screen. Furthermore, high
resolution means high power consumption, high processor load, which causes high power
consumption and heat problem. Thus, many manufacturers are reluctant to use 2K screens. The
higher screen resolution is only suitable for the mobile phone which supports VR function. For
the general users, the accuracy of 1080p is enough.

Flexible screen refers to flexible OLED. It not only greatly benefits the manufacture of a new
generation of smart phones, but also has a far-reaching impact on the application of wearable
devices due to its low power consumption and flexibility. In the future, flexible screens will be
widely used with the continuous penetration of personal smart devices. OLED uses plastic
substrates, rather than common glass substrates, with the help of film packaging technology, and
paste a protective film on the back of the panel, so that the panel becomes flexible, not easy to
break. Flexible screens can be curled, but they cannot be folded. Currently on the market are
ordinary rolled-up mobile phones since the shape of other parts of the mobile phone cannot be
changed. In February 2016, Canadian researchers at Queens University's Human Media
Laboratory unveiled the exciting invention of flexible screen technology, creating the world's first
flexible-screen smartphone equipped with full-color, high-resolution displays and wireless
technology. The phone, called ReFlex, comes with a 720p flexible OLED screen from GDisplay and
Android 4.4 KitKat system. Because the battery and PCB are not yet bent, it only has a flexible
screen with PCB and battery built below the screen. Many smart phone manufacturers show
their interests in flexible screen for larger screens and smaller size.

 Factors of 5G device power consumption

The larger size and the higher resolution and brightness of the screen will cause more power
consumption. The screen power consumption is calculated by its power, which is the product of
current and voltage. When the dot pitch and the brightness of the screen remain unchanged, the
larger the size and the higher the resolution. The screen power consumption increases with the
number of luminous points. Similarly, when the brightness and the size of the screen main
unchanged, the screen power consumption increases with the resolution. For 5G devices with
larger screen sizes and 5G devices supporting VR, the problem of screen power consumption
needs to be solved.

The power consumption depends on the content of video. When broadcasting advertisements or
live video, white and color dominate, power consumption might be large. When playing simple
graphics such as text, most of the screen is black, and the average power consumption is the
lowest. As 5G terminal products support higher data rate and lower latency, and have broad
prospects of video application, the proportion of video services of the typical user model will
increase. We need to build a new typical service model for 5G user and further analyze the
impact on 5G power consumption.

23
 GTI 5G Device Power Consumption Whitepaper

5 Factors of power consumption - 5G features

5.1 5.1 RRC idle state and RRC inactive state

5.1.1 Bandwidth

In NR, UE in RRC idle state or RRC inactive state can operate on an initial active DL BWP, which
can be, e.g., 24RB, 48RB or 96RB configured by MIB for different SS/PBCH block and PDCCH
subcarrier spacing [Section 13 in TS 38.213], to receive SSB, RMSI, paging, etc. The number of RBs
for initial active DL BWP will impact on UE power consumption. From power saving perspective, it
is better to configure a smaller initial active DL BWP by MIB. However, smaller initial active DL
BWP may have lower capacity, e.g., lower paging capacity and lower system information capacity,
and thus may impact the system performance. So there is a trade-off between UE power
consumption and system performance, when configuring initial active DL BWP.

5.1.2 Paging

Paging reception consumes the main power consumption for NR UEs in RRC idle state and RRC
inactive states. For each paging cycle, it may takes the UE up to tens of ms to receive the paging
message and whole procedure may include UE’s hardware ramp up procedure,
re-synchronization procedure and paging PDCCH monitoring and paging PDSCH reception and
UE’s hardware ramp down procedure. The main power consumption for paging reception
includes:

 (Re-)synchronization

Before paging reception, the UE needs to acquire time and frequency synchronization to the gNB.
Since there is no “always on” reference signals like LTE CRS available as in LTE, the UE needs to
utilize SS/PBCH to get time and frequency synchronization reference. For UEs in bad channels
conditions, one-shot time and frequency synchronization using just one SS/PBCH block is not
enough, the UE needs to wake up tens of ms before the paging PDCCH monitoring occasion to
accumulate several SS/PBCH samples to achieve good enough time and frequency
synchronization performance.

The length of paging DRX cycle may have influence on the (Re-)synchronization operation. For
example, for short paging cycle, the UE may maintain the synchronization to the network for one
or more paging cycles, therefore the UE only needs to perform synchronization every several
paging cycles. But for long paging cycle, the UE may need to perform synchronization every
paging cycle.

24
 GTI 5G Device Power Consumption Whitepaper

 Paging PDCCH monitoring and paging reception

After (Re-)synchronization to the network, the UE can monitor PDCCH within its paging
monitoring occasions. In NR, paging DCI would be often transmitted in beam-sweeping manner.
The UE can determine the best PDCCH monitoring occasions to receive the paging DCI for power
saving purpose based on the occasion relationships between paging monitoring occasions and
SS/PBCH blocks.

For one UE, it would be paged in only very few of paging cycles when there is traffic for the UE or
in the case of system information update or ETWS or CMAS, therefore, for most of the paging
cycles, paging PDCCH monitoring would result in no valid PDCCH for the UE.

5.1.3 Measurement

RRM measurement is another important power-consuming behaviour for UEs in RRC idle states
and RRC inactive states. The UE needs to continuously to perform RRM measurement for cell
re-selection.

RRM measurement in RRC idle states and RRC inactive states includes intra-frequency
measurement and inter-frequency measurement. Within the SMTC window configured by system
information, the UE can perform RRM measurement periodically for each frequency layer. The
essential procedures for measurements include：

 Cell detection

For each frequency layer, before performing RRM measurements, the UE needs to detect the
target measurement cells. Cell detection would consume much UE’s power since the UE needs to
search the SS/PBCH block within the SMTC window and at the meantime try to blindly detect the
PSS/SSS with at most 1008 hypothetical physical cell IDs.

It is noted that for synchronous network and with the high layer indicated SS/PBCH block
measurement locations, the cell detection complexity for the UE in the time domain would be
reduced.

 Measurement

The RRM measurement include RSRP measurements and RSRQ measurements. Since
multiple-beam deployment would be the typical operation mode for NR, the UE needs to
generate the measure results based on the measurements on multiple SS/PBCH blocks which
would increase the UE’s power consumption.

5.1.4 System information acquisition (including synchronisation)

In LTE, the position and periodicities for PSS, SSS and PBCH are definite and known to a UE.
Besides, the SCS for PSS, SSS and PBCH is also definite, i.e., 15KHz. LTE UE may use this
information to complete system information acquisition including synchronization. Compared
with LTE, there are uncertainties for configurations of SS/PBCH in NR. This may cause complexity
and additional power consumption when NR UE performs initial access.
25
 GTI 5G Device Power Consumption Whitepaper

NR UE should support synchronization in time and frequency and the detection of the physical
cell IDs from 1008 candidates. Besides, NR UE should support the detection of the SS/ PBCH block
under different numerologies and time locations in various frequency carriers and bandwidth
configurations. For Sub-6GHz, 15kHz and 30kHz are allowed for SCS of SS/ PBCH block. For
above-6GHz, 120kHz and 240kHz are allowed for SCS of SS/ PBCH block. There are several
challenging aspects in detecting SS/ PBCH block as mentioned below, and NR UE should be able
to handle these challenges.

- In NR, while SS burst set is periodic whose period is assumed to be 20ms by a UE during
initial access, there can be multiple SS/ PBCH blocks (up to 8 in sub-6GHz) within each burst
set which are not necessarily periodic. As seen in the figure below for 14 (28, respectively)
OFDM symbols for 15kHz (30kHz, respectively) SCS, SS/ PBCH block which comprises 4
OFDM symbols is not periodic especially for 30kHz case. In addition to this, not all SS/ PBCH
block candidate locations are guaranteed to transmit valid signal.

Figure 5- 1 Illustration of SS/PBCH block allocation in time-domain

- 3 or 2 LSB’s of SS/ PBCH block within each burst set which a UE needs to identify during
initial access is carried by PBCH DMRS scrambling sequence, and this requires the
corresponding blind detection for NR UE.

- For some bands, there can be uncertainty on SS/PBCH SCS between 15kHz and 30kHz. UE
need perform blind detection.

- Mapping of SS/ PBCH block is described in the figure below. It can be seen that PSS is
mapped in the first OFDM symbol, and this creates challenge on AGC operation for PSS
detection since there may not be useful signal before PSS which can be used to set proper
gain level. However, in LTE, UE can use CRS before PSS.

Figure 5- 2 Illustration of SS/PBCH block mapping

- In frequency-domain, unlike LTE whose SS/PBCH is located at the center of system

26
 GTI 5G Device Power Consumption Whitepaper

bandwidth, SS/ PBCH block in NR can be flexibly located within each channel. Hence, NR UE
would need to be careful if it makes any assumption on spectral shape around SS/ PBCH
block.

NR UE should support obtaining the essential minimum system information, including at least
SFN, SS block time index and configuration information of PDCCH for RMSI (Remaining minimum
System Information) from PBCH.

In NR, detected SS/ PBCH block may not have an associated RMSI, i.e., the SS/PBCH block is not
the cell-defining one, and NR UE should be prepared to continue SS/ PBCH block detection at
different frequency location with the corresponding indication on PBCH.

5.1.5 RNA update

Compared to LTE, 5G inactive state UEs need to additionally perform RNAU. The size of RAN
paging area is a factor contributing to UE power consumption. RAN paging area size can be
either the same as or smaller than the Tracking Area. If RAN paging area equals Tracking Area, UE
performs combined TAU/RNAU when across TA boundary, no additional signaling/power
consumption is expected. However, if RAN paging area is smaller than Tracking Area, more
signaling/power consumption is expected due to extra RNAUs.

Besides, UE needs to perform periodic RNAU. The periodicity of RNAU will also impact UE power
consumption. Generally, UE with less activity can be configured with a smaller RNAU periodicity.
Also, a decent match of the periodicity of RNAU and TAU can help reduce UE power consumption.
Ideally, if they are to have the same length, UE can always perform RNAU and TAU together,
power consumption can be minimized. A lesser option is to ensure that the periodicity of TAU is
multiple of the periodicity of RNAU, thus TAU can be performed together with RNAU.

UE speed is also a factor that impacts power consumption. The faster the UE speed, more
frequent RNAU could be envisioned. One way to counteract this is to make sure one RAN paging
area covers UE routes as much as possible by adjusting the RAN paging area at each RNAU based
on the history RNAUs. This also applies to adjust the periodicity of RNAU.

5.2 RRC connected state

5.2.1 Bandwidth

The UE channel bandwidth supports a single NR RF carrier in the uplink or downlink at the UE.
For Frequency Range 1 (FR1, 450 MHz – 6000 MHz), maximum UE Channel bandwidths for
different SCS (sub-carrier spacing) for some typical operating bands in Table 5- 1 are supported by
NR UE.

Table 5- 1 Maximum UE Channel bandwidths per operating band for FR1

FR Band 15kHz SCS 30kHz SCS 60kHz SCS

FR1 Band n77 (3.3 GHz ~4.2GHz) 50MHz 100MHz 100MHz

Band n78 (3.3 GHz ~3.8GHz) 50MHz 100MHz 100MHz

27
 GTI 5G Device Power Consumption Whitepaper

Band n79 (4.4GHz~5GHz) 50MHz 100MHz 100MHz

Band n1 20MHz 20MHz 20MHz
Band n3 30MHz 30MHz 30MHz
Band n8 20MHz 20MHz NA
Band n41 50MHz 100MHz 100MHz

For Frequency Range 2 (FR2, 24250 MHz – 52600 MHz), maximum UE Channel bandwidths for
different SCS (sub-carrier spacing) for some typical operating bands in Table 5- 2 are supported by
NR UE.

Table 5- 2 Maximum UE Channel bandwidths per operating band for FR2

FR Band 60kHz SCS 120kHz SCS

Band n257 200MHz 400MHz

FR2 Band n258 200MHz 400MHz

Band n260 200MHz 400MHz
Band n261 200MHz 400MHz

28
 GTI 5G Device Power Consumption Whitepaper

The maximum UE Channel bandwidths for NR FR1 and FR2 is 100MHz and 400MHz, respectively,
which is much wider than that of LTE (i.e., 20MHz per carrier). Larger bandwidths will cause more
power consumed for NR UE. Not all UEs support the maximum Channel bandwidths. UE can
report its supported UL and DL bandwidths to network, by parameters channelBWs-DL and
channelBWs-UL, as shown in Table 5- 3, which is narrower than the maximum UE channel
bandwidth.

Table 5- 3 UE Channel bandwidth capabilities

Definitions for parameters

channelBWs-DL

Indicates for each subcarrier spacing whether the UE supports channel bandwidths lower than
the maximum channel bandwidth as defined in TS 38.101-1 and TS 38.101-2. If this parameter is
not included, the UE supports all channel bandwidths.

For FR1, the bits starting from the leading / leftmost bit indicate 5, 10, 15, 20, 25, 30, 40, 50, 60
and 80MHz. For FR2, the bits starting from the leading / leftmost bit indicate 50, 100 and
200MHz.

channelBWs-UL

Indicates for each subcarrier spacing whether the UE supports channel bandwidths lower than
the maximum channel bandwidth as defined in TS 38.101-1 and TS 38.101-2. If this parameter is
not included, the UE supports all channel bandwidths.

For FR1, the bits starting from the leading / leftmost bit indicate 5, 10, 15, 20, 25, 30, 40, 50, 60
and 80MHz.

For FR2, the bits starting from the leading / leftmost bit indicate 50, 100 and 200MHz.

The BWP operation, by which UE in RRC connected state can operate on a narrower bandwidth
than the maximum bandwidth, can also reduce the NR UE power. The BWP concept is descried in
Section 5.2.4.

5.2.2 UL-MIMO

UL-MIMO can improve spectrum efficiency compared to single UL transmission. Depending on

different deployment scenarios, BS configurations, transmission schemes and SNR distribution
etc., an approximately 0%~40% throughput gain could be expected from 2Tx UL MIMO according
to simulation.

However, UL-MIMO would require UE to equip multiple RF Tx chains in the uplink. This means
multiple sets of PAs, antenna and other components. Compared to single UL, even with the same

29
 GTI 5G Device Power Consumption Whitepaper

output power, the power consumption introduced by multiple RF chains would be greater,
since activating an extra RF chain would cause more power consumption. According to the
power model assumption agreed in 3GPP RAN1 #95 meeting, power consumption of 2Tx is
1.4 times of 1Tx at 0dBm total transmission power, and 1.2 times at 23dBm total
transmission power.

The additional power consumption of 2Tx is more significant in lower total transmission
power.

In order to reduce the power consumption, fall back to single Tx is a straightforward way,
particularly when the scenario is not favorable for 2Tx, i.e. small packet scenario.

5.2.3 HPUE

UE Power Classes define the maximum output power for any transmission bandwidth within the
channel bandwidth. Power class 3 of 23dBm, which is reused from LTE, is also the default
power class for NR. HPUE (High power UE) means UE supporting higher power class
compared to default, e.g. Power class 2 which has a 26dBm maximum output power had
been defined for NR bands n41, n77, n78 and n79. Currently, Power class 2 UE was defined for
1Tx (26dBm) and UL-MIMO (2Tx 23+23dBm) SA case. For NSA HPUE, similar definition is being
studied.

Table 5- 4 NR Band list

NR Class 2 Class 3
(dBm) (dBm)
band

n41 26 23

n77 26 23

n78 26 23

n79 26 23

The purpose of introducing HPUE feature is to improve the uplink coverage. There is an
imbalance between uplink and downlink coverage and generally UL coverage is weaker than DL,
and this became more apparent for 5G, since the spectrum available for 5G is usually higher
compared to LTE. As an example, the following NR link budget result was referenced from
[R1-1809266 IMT-2020 Self-Evaluation: NR Link Budgets] in which a lower coupling loss means a
weaker coverage. It can be seen that the coupling loss gap between UL and DL is relatively high

30
 GTI 5G Device Power Consumption Whitepaper

especially for 3.5GHz.

Figure 5- 3 Preliminary results of maximum coupling loss for 15kHz numerology at 800MHz and
30kHz numerology at 3500MHz.

It should be noted that, in order to control emission to satisfy SAR requirements, only TDD band
has the definition for HPUE and there is also a restriction in UL duty cycle to control the actual
transmission time.

In the case of identical uplink path-loss, a PC2 HPUE higher peak power delivers higher uplink
SNR, which may allow uplink scheduler to allocate higher MCS, which may translate into shorter
transmission time, and in turns may reduce power consumption especially for large uplink packet
transmissions. This effect is more notice-able at the PC3 UE cell edge location and for large uplink
data payloads. As the PC3 UE reaches cell edge, the lower peak power of a PC3 UE forces the
uplink scheduler to decrease DTX rate, i.e. the PC3 UE transmission time might be longer than
that of a PC2 UE (assuming same uplink path loss) resulting into a higher battery power
consumption. Nevertheless, the relationship between peak power capability and UE battery life is
not straightforward as it depends on many factors, such as UE quality of service, packet size,
scheduler algorithm, cell loading, etc… etc. Initial field trial measurements [1] tend to confirm
these assumptions, but considering the large number of variable, further studies would be
needed to evaluate user experience for a variety of use-cases. Alternatively, the higher peak
power capability of PC2 could be used by operators as a way to enhance cell coverage, making
this feature very attractive. However the extended coverage may come at the expense of a higher
battery power consumption.

5.2.4 BWP

The Bandwidth Part (BWP) controls the size and location of the bandwidth used by the DL/UL
links. The size of the BWPs used for DL/UL transmissions are configured independently for the
DL/UL. The UL bandwidth size has a relatively small impact on the power consumption of the UE
which is mainly controlled by the transmit power (see [4]). On the other hand, the bandwidth
used for DL reception has a large impact on the power consumption as it controls the RF, the ADC
sampling rate, the FFT size and the baseband processing as well as the peak throughput. Based
on the NR UE power model recently agreed in 3GPP [3], lowering the bandwidth from 100MHz to
31
 GTI 5G Device Power Consumption Whitepaper

20MHz reduces UE downlink power consumption by ~60% (see Figure 5- 4).

Figure 5- 4 UE power consumption scaling with the DL bandwidth for 4Rx MIMO reception

In addition to the BWP size, the BWP configuration can also specify several other parameters
with impact on the power consumption (check Table 5- 5 for summary):

- Cross-slot scheduling: DL cross-slot scheduling is enabled or disabled according to the k0

values specified in the TimeDomainResourceAllocationList in PDSCH-ConfigCommon in
BWP-DownlinkCommon [11]. The power consumption saving from cross-slot scheduling is
estimated at ~30% (see [3]), however this can be only achieved if the UE is guaranteed that
same slot scheduling cannot be used by the gNodeB which requires (K0≥1) for the applicable
PDSCH time domain resource allocation. To allow an effective use of cross-slot scheduling in
Rel-16, a change of the framework to support an explicit signaling of the minimum K0 per power
profile (Figure 5- 5) will allow more flexibility and a more direct mechanism to control the power
consumption

- Number of MIMO layers: rank restriction for CSI reporting per BWP configuration are
possible in Rel-15. This allows the UE to reduce the CSI reporting complexity and can be
used to limit the maximum number of MIMO layers that the UE has to report and expect to
receive. However, in Rel-15 this mechanism does not prevent the gNodeB from using a
higher number of MIMO layers or requiring performance with the maximum number of Rx
antennas supported by the UE, and this limits the power savings that the UE can achieve.
Therefore to improve power saving, further mechanisms of adaptation of the maximum
number of MIMO layers and number of Rx antennas should be supported in Rel-16 as part
of the power profiles (Figure 5- 5).

- PDCCH monitoring: PDCCH monitoring is controlled through the DL BWP configuration.

When no data is transmitted (PDCCH-only), the power consumption is dominated by the
PDCCH processing, hence reducing PDCCH monitoring improves the power consumption
significantly. The main mechanism for PDCCH monitoring reduction is DRX (see section
5.2.5), but switching between BWPs with different bandwidths and PDCCH monitoring

32
 GTI 5G Device Power Consumption Whitepaper

periodicities also enables adaptation of the UE configuration to the traffic and saving power
when latency requirements are relaxed.

Table 5- 5 Main parameters controlled by the BWP with impact on power consumption.

Parameters controlled by the BWP Impact on power consumption

Bandwidth size UE power consumptions scales with the DL bandwidth

(see Figure 5- 4)

Cross-slot scheduling Cross slot scheduling (K0 ≥ 1) reduces power

consumption (see section 5.2.4)

Number of MIMO layers In Rel-15, partial control of the number of MIMO

layers through the CSI configuration per BWP
Controlled by the DL BWP

PDCCH monitoring Adaptation of the PDCCH monitoring to the traffic

allow to reduce power consumption.

Figure 5- 5 shows an example of how the BWP can be used to build different power profiles that
can then be used to adapt the UE configuration to the traffic characteristics, therefore achieving
savings on power consumption without impact the quality of service. Some of building blocks for
realizing power profiles similar to those of Figure 5- 5 are already present in Rel-15 BWP
configurations, however, improvements to the framework such as minimum K0 and maximum
MIMO configurations (#layers and ~Rx antennas) are needed in Rel-16 to improve effectiveness.

Figure 5- 5 Examples of power profiles for different traffic characteristics

Rel-15 supports up to 4 BWPs for each of the DL/UL and the switching between these BWPs is
either RRC-based or DCI-based. For Rel-16, the switching between power profiles (Figure 5- 5)
can be based on the same or similar mechanisms to the BWP switching, however given that the
RRC-based switching is typically order of magnitude slower than DCI-based switching (tens of ms
compared to few ms), therefore to achieve fast adaptation and lower traffic disruption when
switching between the power profiles the support of a DCI-based switching is essential.

33
 GTI 5G Device Power Consumption Whitepaper

5.2.5 DRX

DRX is one scheme to save the UE’s power consumption for UE in RRC connected state. The
power consumption factors for DRX procedure are as follows:

 (Re-)synchronization procedure

For DRX, time and frequency (Re-)synchronization as discussed in the paging section 5.1.2 is a key
power consumption procedure for DRX. When the DRX cycle is long, the UE needs to utilize
SS/PBCH blocks to acquire time and frequency (Re-)synchronization to the network.

 DRX parameters setting

Proper DRX parameter setting is very important for UE’s power consumption. It would be
beneficial to configure the DRX cycle, the length of on-duration and DRX related timers to match
the UE’s traffic characteristics to save the UE’s power.

Discontinuous Reception (DRX) design for Rel-15 NR, is very similar to E-UTRA DRX. Using DRX
enables power saving by reducing the time UE is awake to monitor the PDCCH.

The UE is only required to monitor the PDCCH during the DRX-OnDuration. If the UE detects a DCI
with a DL assignment or an UL grant, then the UE restarts the DRX-inactivity timer during which
UE keeps monitoring the PDCCH, otherwise the UE may stop monitoring the PDCCH at the end of
the DRX-OnDuration and go back to sleep.

The power reduction achieved is proportional to the time the UE spends in sleep mode, and so
the OnDuration and Inactivity timers should be short in relation to the DRX cycle time. TTI
duration is inversely proportional to subcarrier spacing, so for a given timer setting, an NR UE will
process 2μ times as many TTIs as its E-UTRA equivalent. This means that unless traffic patterns
dictate otherwise, shorter timer settings should be the norm in NR. If this cannot be achieved,
additional signaling can be used to provide early termination of the DRX cycle when possible.

In addition to the PDCCH monitoring, in order for the UE to maintain connection and for the
network to optimize resources utilization, the network needs to configure a number of reference
signals and control channel transmissions, while the UE needs to perform the corresponding
Background Activity (BA) tasks in connected DRX.

The background tasks that the UE has to perform in C-DRX can be divided into 2 groups:

- Background Activity during the OnDuration:

o This group of background activities includes CSI-RS acquisition, Beam

Management, CSI reporting and SRS transmission tasks. These tasks can only
take place during the OnDuration.

- Background Activity outside of the OnDuration:

34
 GTI 5G Device Power Consumption Whitepaper

o This group of background activities includes the SSB acquisition/processing, TRS

acquisition/processing and RRM measurement. While the SSB/TRS processing
has to take place before the OnDuration for the UE to be ready to start receiving
the PDCCH/PDSCH, the RRM measurement timing depends on the SMTC
configuration, although the UE maintains some flexibility on which SMTC
occasion to use.

The setting of Figure 5- 6 shows that the power consumed by the BA tasks outside of the
OnDuration is of the same order of magnitude as the power consumed during the OnDuration.
The BA tasks during the OnDuration lead to a relatively small power consumption increase, but
still occupy a significant portion of the OnDuration hence putting a limit on how small the
OnDuration could be made.

To achieve better power consumption, optimization of the BA tasks should be targetted in NR

Rel-16. Power saving could be realized by grouping the CSI acquisition/reporting and potentially
other BA activities with SSB/TRS processing in a pre-wakeup window. Such a mechanism would
allow reduction of the DRX-OnDuration length, which would improve performance and save
power.

35
 GTI 5G Device Power Consumption Whitepaper

Figure 5- 6 Processing Timeline including the Background activity.

5.2.6 Cross-slot scheduling

NR support flexible timings between PDCCH and PDSCH and the timing K0 is informed in the DCI
from one RRC configured set named PDSCH-TimeDomainResourceAllocationList. If the minimum
value of k0 within the PDSCH-TimeDomainResourceAllocationList is 0, the UE can be scheduled
with same-slot PDSCH scheduling by indicating that k0 equals to 0 or with cross-slot scheduling
by indicating that k0 is larger than 0 from that set. In this case, the UE needs to receive and buffer
all the PRBs in the current active BWP in order to be ready to receive PDCCH and PDSCH within
the same slot. Otherwise, if the minimum value of k0 within the
PDSCH-TimeDomainResourceAllocationList is larger than one K_threshold, only cross-slot
scheduling would be allocated for the UE. In this case, the UE can tune its RF with a narrow
bandwidth that is wide enough to receive and buffer the PDCCH CORESET, then if one DCI is
detected, the UE can tune its RF to the cover the bandwidth of PDSCH resource. By such
operation, the UE can save its power when there is no PDSCH scheduling. It is noted that
36
 GTI 5G Device Power Consumption Whitepaper

K_threshold shall be large enough to guarantee that the UE has enough time to switch its RF.

Therefore, for UE’s services that are not latency sensitive, cross-slot scheduling is beneficial for
UE’s power saving. And the key point to achieve the power saving target is that configure that
PDSCH-TimeDomainResourceAllocationList with all the elements in the list are larger than one
K_threshold.

5.2.7 Periodic PDCCH monitoring

The most power consuming procedure for one NR UE would be PDCCH monitoring in RRC
connected state, as reported by several chipset vendors, for LTE, more than half of the PDCCH
monitoring occasion corresponds to no PDSCH schedule. NR support much more flexible PDCCH
search space sets configuration to match the UE’s traffic characteristic. However, the UE’s traffic
may fluctuates time by time, it is hardly possible for the gNB to use only one set of PDCCH search
space configuration for one UE all the time. Furthermore, the network often needs to schedule
many UEs at the same time thus it cannot be guaranteed that one UE could be scheduled when
the system load is high. Therefore, reducing power waste for PDCCH monitoring is one important
direction for NR UE’ power saving. There are several factors affect the power consumption for
PDCCH monitoring.

 Match to the UE’s traffic characteristic

The more the PDCCH search space configuration matches the UE’s traffic characteristic, the less
the power waste. Therefore, it is important for the network to have a clear knowledge of the UE’s
traffic characteristic for different services.

 Fast switching of PDCCH search space configuration

In order to adapt to the fluctuation of the UE’s traffic, it is important to switch the UE’s PDCCH
search space configuration as fast as possible to reduce the chances of PDCCH monitoring
without corresponding PDSCH scheduling. The current PDCCH search space reconfiguration
mechanism based on RRC reconfiguration may not meet such requirements and needs to be
enhanced.

5.3 Parameter configuration recommendation

In this section we consider how parameter configuration affects UE power consumption in
different scenarios, and show using examples that the best configuration to minimise UE power
consumption depends on the type of data traffic that the UE is experiencing. The evaluation is
based on work done in support of the 3GPP study item on NR UE power consumption.

In this study, consensus was reached on a UE power model that can be used to compare the
relative benefits of different NR power saving strategies. Full details of the model can be
found in [13], but the key features are summarized below.

37
 GTI 5G Device Power Consumption Whitepaper

5.3.1 3GPP power model

The model reference configuration for FR1 is based on a single 100MHz TDD carrier with
30kHz SCS. In the downlink the UE has 4 Rx antennas. PDCCH occupies 2 symbols at the start
of a slot, using same-slot scheduling (k0=0), and PDSCH operates at maximum data rate
(256QAM). The UE uplink has 1 Tx antenna transmitting at 0dBm or 23dBm.

Power values are averaged over 1 slot, and the model specifies the average power
consumption in each of several power states, listed in Table 5- 6. The model is based on an
uncalibrated power unit representing average UE power in the deep sleep state, which
makes it independent of any specific UE implementation. Although different UE
implementations are expected to depart from the model ratios for the relative power
consumption of each state, simulations using the model provide a platform-independent
reference point for evaluation and comparison of different power saving strategies.

Power State Characteristics Relative Power

Deep Sleep Time interval for the sleep should be larger than the total 1
transition time entering and leaving this state. Accurate timing (Optional: 0.5)
may not be maintained.

Light Sleep Time interval for the sleep should be larger than the total 20
transition time entering and leaving this state.

Micro sleep Immediate transition is assumed for power saving study purpose
45
from or to a non-sleep state

PDCCH-only No PDSCH and same-slot scheduling; this includes time for PDCCH
100
decoding and any micro-sleep within the slot.

SSB or SSB can be used for fine time-frequency sync. and RSRP
CSI-RS processing measurement of the serving/camping cell. TRS is the considered
100
CSI-RS for sync. FFS the power scaling for processing other
configurations of CSI-RS.

PDCCH + PDSCH PDCCH + PDSCH. ACK/NACK in long PUCCH is modeled by UL

300
power state.

UL Long PUCCH or PUSCH. 250 (0 dBm)

700 (23 dBm)

Table 5- 6 3GPP model power states

The model also includes transitional energies for the ramp down + ramp up energy associated
38
 GTI 5G Device Power Consumption Whitepaper

with entering and leaving a sleep state, listed in Table 5- 7.

Additional transition energy:

Sleep type Total transition time
(Relative power x ms)

Deep sleep 450 20 ms

Light sleep 100 6 ms

Micro sleep 0 0 ms*

* Immediate transition is assumed for power saving study purpose from or to a non-sleep state

Table 5- 7 3GPP model transition energies

By modelling the time spent in each state and the number of sleep transitions in a given period,
using system level simulation, the average power consumption for the reference configuration
can be calculated. When the reference configuration is modified by enabling cross-slot scheduling
or by changing BWP, the number of carriers or the number of antennas, the model provides a set
of scaling factors to be applied to the Table 5- 6 parameters. These are listed in Table 5- 8.

39
 GTI 5G Device Power Consumption Whitepaper

Scaling for FR1 Proposal Comment

BWP Bandwidth (DL) Scaling of X MHz = 0.4 + 0.6 * (X - 20) / 80. For 10MHz BW, only AL up to 8 can be used for
Linear interpolation for intermediate PDCCH
bandwidths. Valid only for X = 10, 20, 40, 80,
The transition time is the same as DCI-based BWP
and 100.
switching delay for Rel-15.
Above scaling is applicable for FR1 only.
If the power after scaling is smaller than the BWP
In case scaling is needed for FR2, companies transition power, assume the BWP transition power
can report the assumed scaling factor. as the output of scaling unless otherwise justified.

BWP Bandwidth (UL) No scaling at 0dBm or 23dBm

Above scaling is applicable for FR1 only.

In case scaling is needed for FR2, companies

can report the assumed scaling factor.

CA (DL) 2CC is 1.7x1CC Activation/deactivation delay follows RAN4

specification; FFS transition energy
4CC is 3.4x1CC (i.e. 2x 2CC)
Applicable for FR1 and FR2
Above refers to the worst case CA
configuration in terms of power consumption.

CA (UL) Same as downlink at 0dBm. No scaling at Applicable for FR1 and FR2
23dBm

2CC is 1.2x1CC at 23dBm

Limit scaling up to 2CC.

Antenna scaling (DL) 2Rx power is 0.7x 4Rx power for FR1 Assume same number of antenna elements per Rx
chain
1Rx power is 0.7x 2Rx power for FR2

Antenna scaling (UL) 2Tx power is 1.4x 1Tx power at 0dBm. 1.2x.at
23dBm FR1 only
2Tx support is not considered for FR2.

PDCCH-only Power of cross-slot scheduling is 0.7x Applicable for FR1 and FR2
same-slot scheduling

SSB One SSB power is 0.75 of two SSB power, i.e.

75 power units

PDSCH-only slot This assumes the same number of PDSCH symbols

280 for FR1
as in the PDCCH+PDSCH case.

40
 GTI 5G Device Power Consumption Whitepaper

325 for FR2

CSI-RS FFS for scaling w.r.t. # of symbols for CSI-RS

Short PUCCH Short PUCCH power = 0.3 x uplink power Applicable for FR1 and FR2.

Reference config consists of 1-symbol PUCCH

SRS SRS power = 0.3 x uplink power Applicable for FR1 and FR2.

Table 5- 8 3GPP model scaling parameters

It should be noted that for BWP scaling in the downlink, the final comment modifies the proposal.
This imposes a minimum value of 50 units on the UE power after BWP reduction is applied, which
affects PDCCH-only and SSB/CSI-RS power when the BWP is 20MHz or less.

5.3.2 Applying the 3GPP power model to NSA

In its currently agreed form the 3GPP power model is only applicable to standalone
configurations. Most initial 5G deployments will be non-standalone, and in this paper we also
want to discuss power consumption for these configurations. It is therefore necessary to make
some assumptions about the power model that will be applied to the NSA configuration.

For the analysis presented here, we have used the following assumptions to align the LTE
component modelling with the 3GPP NR model, while avoiding adding new power states to the
model.

 The LTE modem and the NR modem operate independently in the UE (they may in fact
be separate devices)

 Sleep states and transition energies use the same parameters in both modems

 The LTE modem in active states uses the same parameters as an NR modem operating in
a 20MHz BWP.

 LTE sync and measurements use the same power as NR SSB/CSI-RS for sync/reference
symbol processing

 When LTE combines PDCCH and uplink transmission in the same slot, the power is the
sum of the two component powers

 Because the LTE modem only has 2 Rx paths instead of 4, the 50 unit floor is not applied
to the BWP scaling for PDCCH-only and Sync/RS monitoring

41
 GTI 5G Device Power Consumption Whitepaper

5.3.3 Modelling scenarios

The analysis in [13] is based around 3 reference traffic scenarios (Table 5- 9). Results are
presented from many companies, firstly using the baseline model configuration and then
applying different power saving configurations.

FTP traffic Instant messaging VoIP

Model FTP model 3 FTP model 3 As defined in R1-070674.

Packet size 0.5 Mbytes 0.1 Mbytes Assume max two packets
bundled.
Mean inter-arrival time 200 ms 2 sec

DRX setting Period = 160 ms Period = 320 ms Period = 40 ms

Inactivity timer = 100 ms Inactivity timer = 80 ms Inactivity timer = 10 ms

On duration = 8 ms On duration = 10 ms On duration = 4 ms

Table 5- 9 Traffic scenarios for modelling

NR allows considerable flexibility in configuration, and the savings reported vary widely between
companies, but all of the results indicate that by adapting the network configuration to UE traffic
patterns, significant savings can be achieved.

In this paper the same reference scenarios are used as the baseline for power comparisons.
Standalone results are based on the simulation timings in [14], using a simplified power
calculation to examine the savings available from DRX cycle configuration, BWP adaptation,
cross-slot scheduling and MIMO reduction.

Power saving techniques applied to standalone cases would give reductions of similar magnitude
if applied to the same traffic on the NR carrier of an NSA network, but the percentage saving
would be smaller in the NSA case due to the added power contribution of the LTE component.

For an NSA network the configuration flexibility is even greater than for SA, and to avoid a
multiplicity of scenarios we have considered here an example of mixed IM and FTP traffic, with
IM traffic (mostly small packets) routed over the 20MHz LTE carrier, and the FTP traffic (mostly
large packets) routed over the 100MHz NR carrier. This is compared with an SA configuration
where both sets of traffic are routed over a single 100MHz NR carrier.
42
 GTI 5G Device Power Consumption Whitepaper

The results from system level simulation are presented in subsequent sections.

5.3.4 Standalone baseline scenarios

The first of the reference scenarios is FTP traffic, which is representative of applications such as
file download or streaming video. The simulation output is summarised in Table 5- 10 below.

The simulation averages results from multiple UEs in a cell, so the number of sleep transition
instances in a 10 second period can be fractional.

FTP baseline (0.5MB packets, 200ms arrival, DRX 160, 100, 8)

Power state Model power Time percentage Average power % power

Microsleep 45 0.48% 0.22 0.42%

Light sleep 20 5.17% 1.03 1.99%

Deep sleep 1 55.65% 0.56 1.07%

PDCCH-only 100 33.13% 33.13 63.73%

PDCCH+PDSCH 300 4.09% 12.27 23.60%

SSB/CSI-RS 100 1.31% 1.31 2.52%

UL 250 0.16% 0.40 0.77%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 38.99 0.39 0.75%

DS transitions 450 59.55 2.68 5.15%

Total Power 51.99 100%

Table 5- 10 FTP traffic model baseline power consumption

The most power-hungry states are not necessarily the biggest contributors to average power
consumption. 63.7% of average power consumption results from monitoring PDCCH waiting for
new packets to arrive. Downlink data traffic (PDCCH+PDSCH) has the highest power consumption
at 300 units, but because it accounts for just over 4% of the time allocation it only contributes
24% of total power consumption. Uplink traffic power is slightly lower (250 units) and accounts

43
 GTI 5G Device Power Consumption Whitepaper

for only 0.8% of total power consumption, but even at 23dBm (power level 700 units) this would
only rise to 2.1%. Sleep states, although they occupy over 60% of the time allocation, account for
only 3.5% of average power consumption, but the sleep transitions contribute an additional
5.9%.

This pattern is typical of many representative use cases. The highest throughput cases that result
in UE overheating will be relatively infrequent for most users.

In this scenario a large proportion of the PDCCH monitoring power is spent waiting for the DRX
inactivity timer to expire. It will be shown later that the DRX cycle is suboptimal for this type of
traffic, and by adapting the DRX cycle to the traffic patterns significant reductions in power
consumption can be achieved.

The next reference scenario is Instant messaging traffic, which is characterised by a smaller
packet size and a longer mean delay between successive packet arrivals. The simulation results
are shown in Table 5- 11.

IM baseline (0.1MB packets, 2000ms arrival, DRX 320, 80, 10)

Power state Model power Time percentage Average power % power

Microsleep 45 0.23% 0.10 0.96%

Light sleep 20 2.88% 0.58 5.34%

Deep sleep 1 89.79% 0.90 8.32%

PDCCH-only 100 6.26% 6.26 58.04%

PDCCH+PDSCH 300 0.06% 0.18 1.67%

SSB/CSI-RS 100 0.70% 0.70 6.49%

UL 250 0.08% 0.20 1.85%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 21.34 0.21 1.98%

DS transitions 450 36.78 1.66 15.35%

Total Power 10.79 100%

Table 5- 11 Instant messaging traffic model baseline power consumption

The reduced data packet frequency means that power consumption in this case is significantly
lower – just over 20% of the FTP case. PDCCH monitoring is still the biggest contributor to the

44
 GTI 5G Device Power Consumption Whitepaper

average, accounting for 58% of the total, and sleep/sleep transitions account for a further 32%.
Of the remaining 10%, 6.5% comes from synchronisation and measurement activities, with only
3.5% resulting from user data transfer.

The power from sleep transitions is again higher than the power in sleep states - reducing the
number of wakeup occasions (for example by lengthening the DRX cycle, or by grouping activities
closer together in time) would make a significant reduction in the sleep contribution.

The last reference scenario is voice traffic, characterised by very small data packets closely
spaced in time, requiring frequent wakeups to satisfy latency requirements. Results for this case
are shown in Table 5- 12 below.

45
 GTI 5G Device Power Consumption Whitepaper

VoIP baseline (R1-070674, DRX 40, 10, 4)

Power state Model power Time percentage Average power % power

Microsleep 45 2.99% 1.35 2.68%

Light sleep 20 43.53% 8.71 17.32%

Deep sleep 1 24.11% 0.24 0.48%

PDCCH-only 100 22.85% 22.85 45.45%

PDCCH+PDSCH 300 1.25% 3.75 7.46%

SSB/CSI-RS 100 5.10% 5.10 10.14%

UL 250 0.16% 0.40 0.80%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 322.26 3.22 6.41%

DS transitions 450 103.51 4.66 9.27%

Total Power 50.27 100%

Table 5- 12 VoIP traffic model baseline power consumption

Although the average data rate is only kilobits per second, power consumption in this case is
almost as high as for FTP traffic. PDCCH-only power is still the largest contributor at 45%, but the
short DRX cycle means that the number of wakeups increases dramatically, and the short sleep
duration prevents deep sleep in many cases. Sleep and sleep transitions account for just over
36% of the total.

When the traffic data rate is very low it is wasteful for the UE to process the entire 100MHz
bandwidth of the carrier. Savings can be made by using a narrower BWP and/or fewer active
antennas to reduce the processing load, or by reducing the duration of the active portion of the
DRX cycle.

The effectiveness of different power saving techniques is examined in the next section

46
 GTI 5G Device Power Consumption Whitepaper

5.3.5 Standalone power saving evaluation

In each of the examples that follow, a single feature of one of the baseline scenarios is changed,
and the resulting power saving is evaluated. By combining multiple power saving techniques in a
single configuration, greater savings are possible, but the 3GPP model is likely to overestimate
the combined saving in such cases since some savings may be duplicated in different scaling
factors – if two techniques each deliver a power saving of 40% when applied individually, the
saving when both are combined may be less than the expected 64%.

DRX cycle adaptation

The DRX cycle applied to FTP traffic in the baseline case is not well matched to the data patterns.
The cycle duration is close to the mean packet arrival time, so a large proportion of cycles will
contain data, triggering the inactivity timer and forcing the UE to be active for 100ms of the
160ms cycle. Reducing the inactivity timer duration will increase the proportion of time that the
UE can spend asleep, and reduce the time spent monitoring the control channel. Table 5- 13
shows the result of reducing the inactivity timer from 100ms to 20ms.

FTP with DRX adaptation (DRX 160, 20, 8)

Power state Model power Time percentage Average power % power

Microsleep 45 0.52% 0.23 0.78%

Light sleep 20 5.45% 1.09 3.62%

Deep sleep 1 78.10% 0.78 2.60%

PDCCH-only 100 10.32% 10.32 34.30%

PDCCH+PDSCH 300 4.04% 12.12 40.28%

SSB/CSI-RS 100 1.41% 1.41 4.69%

UL 250 0.16% 0.40 1.33%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 41.16 0.41 1.37%

DS transitions 450 73.79 3.32 11.04%

Reduction 42.13% Total Power 30.09 100%

47
 GTI 5G Device Power Consumption Whitepaper

Table 5- 13 Reducing FTP traffic power consumption with DRX adaptation

This single change reduces average power consumption by 42%. The power required for uplink
and downlink user data transfer (12.5units) has not changed significantly, but the time that the
UE spends asleep has increased from 61% to 84%, and the power due to PDCCH-only monitoring
has fallen from 33.1 units to 10.32 units so that it is no longer the most significant user of power.
The only performance penalty is a small increase in latency for those data packets that arrive
while the UE is asleep.

More aggressive reductions in the inactivity and on-duration timer settings would result in
further savings.

For instant messaging traffic, the DRX cycle is significantly shorter than the mean interval
between consecutive packets. If the application can tolerate increased latency, a longer DRX
cycle will save power. A longer DRX cycle increases the probability that data will arrive during the
DRX cycle to trigger the inactivity timer, so reducing the inactivity timer setting will provide
further savings. Table 5- 14 shows the result of increasing the DRX cycle from 320ms to 1280ms,
and at the same time reducing the inactivity timer from 80ms to 20ms.

48
 GTI 5G Device Power Consumption Whitepaper

IM with DRX adaptation (DRX 1280, 20, 10)

Power state Model power Time percentage Average power % power

Microsleep 45 0.06% 0.03 0.84%

Light sleep 20 0.72% 0.14 4.46%

Deep sleep 1 97.75% 0.98 30.25%

PDCCH-only 100 1.22% 1.22 37.75%

PDCCH+PDSCH 300 0.05% 0.15 4.64%

SSB/CSI-RS 100 0.18% 0.18 5.57%

UL 250 0.02% 0.05 1.55%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 5.32 0.05 1.65%

DS transitions 450 9.56 0.43 13.31%

Reduction 70.04% Total Power 3.23 100%

Table 5- 14 Reducing IM traffic power consumption with DRX adaptation

This change reduces instant messaging power consumption by 70%. The time that the UE is
asleep increases from 93% to 98.5%, and PDCCH-only power consumption reduced from 6.3 units
to 1.2 units.

These two examples show that large reductions in UE power consumption can be obtained by
adapting DRX parameters to traffic patterns. The power saving is usually accompanied by an
increase in latency, so the longest acceptable DRX cycle is usually constrained by application
performance requirements. Once the DRX cycle has been set, if there is a high probability that
data will arrive in one DRX period then reductions in the inactivity timer setting will be most
effective in reducing power. If the probability is low, then reductions in the on-duration timer
setting will be more effective.

The short DRX cycle required for VoIP traffic means that there are fewer opportunities for
significant savings from DRX adaptation alone. For this case, semipersistent scheduling can
reduce signaling traffic, and bandwidth and MIMO reductions will reduce the processing load.

49
 GTI 5G Device Power Consumption Whitepaper

Locating measurement resources near to the start of the DRX period will minimize the duration
of the active portion of the DRX cycle, and this may allow more efficient use of sleep states. After
these options are in place, DRX adaptation may offer further savings.

Cross-slot scheduling

If PDCCH can schedule PDSCH resources in the same slot that it is received (same-slot
scheduling) , then downlink symbol reception and capture must continue until PDCCH decoding is
complete, in case the captured symbols contain data for the UE. If the decoded PDCCH does not
schedule data for the UE in that slot, then the energy spent doing this has been wasted.

In NR, PDCCH can schedule PDSCH resources in the next slot, or a later slot, by setting parameter
k0 to a value greater than zero (cross-slot scheduling). If cross-slot scheduling is enabled, the UE
can disable its receive path as soon as the last sample of the last symbol of PDCCH has been
received, and when PDCCH decoding is complete it can enter microsleep until the start of the
next slot. In the 3GPP model this reduces power consumption in PDCCH-only slots by 30%.

Simulations with cross-slot scheduling enabled generate the same set of time percentages as the
baseline scenarios, but PDCCH-only power consumption is reduced from 100 units to 70 units.
Since PDCCH-only power is the largest component in each of the baseline scenarios, this
produces significant savings. Table 5- 15 - Table 5- 17 show the results from each of the three
reference scenarios.

FTP traffic with Cross-slot scheduling

Power state Model power Time percentage Average power % power

Microsleep 45 0.48% 0.22 0.51%

Light sleep 20 5.17% 1.03 2.46%

Deep sleep 1 55.65% 0.56 1.32%

PDCCH-only 70 33.13% 23.19 55.15%

PDCCH+PDSCH 300 4.09% 12.27 29.18%

SSB/CSI-RS 100 1.31% 1.31 3.12%

UL (0dBm) 250 0.16% 0.40 0.95%

50
 GTI 5G Device Power Consumption Whitepaper

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 38.99 0.39 0.93%

DS transitions 450 59.55 2.68 6.37%

Saving 19.12% Total power 42.05 100.00%

Table 5- 15 Reducing FTP traffic power consumption with cross-slot scheduling

IM traffic with Cross-slot scheduling

Power state Model power Time percentage Average power % power

Microsleep 45 0.23% 0.10 1.16%

Light sleep 20 2.88% 0.58 6.47%

Deep sleep 1 89.79% 0.90 10.08%

PDCCH-only 70 6.26% 4.38 49.19%

PDCCH+PDSCH 300 0.06% 0.18 2.02%

SSB/CSI-RS 100 0.70% 0.70 7.86%

UL (0dBm) 250 0.08% 0.20 2.25%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 21.34 0.21 2.40%

DS transitions 450 36.78 1.66 18.58%

Saving 17.41% Total power 8.91 100.00%

Table 5- 16 Reducing IM traffic power consumption with cross-slot scheduling

51
 GTI 5G Device Power Consumption Whitepaper

VoIP traffic with Cross-slot scheduling

Power state Model power Time percentage Average power % power

Microsleep 45 2.99% 1.35 3.10%

Light sleep 20 43.53% 8.71 20.05%

Deep sleep 1 24.11% 0.24 0.56%

PDCCH-only 70 22.85% 16.00 36.84%

PDCCH+PDSCH 300 1.25% 3.75 8.64%

SSB/CSI-RS 100 5.10% 5.10 11.75%

UL (0dBm) 250 0.16% 0.40 0.92%

Transitions Model energy Instances/10sec Average power % power

LS transitions 100 322.26 3.22 7.42%

DS transitions 450 103.51 4.66 10.73%

Saving 13.64% Total power 43.42 100.00%

Table 5- 17 Reducing VoIP traffic power consumption with cross-slot scheduling

The 30% reduction in PDCCH monitoring power gives power reductions of between 13% and 19%
for these reference scenarios. The 3GPP model calibration results in appendix 8.2 of [13] show
clearly that if C-DRX is not enabled, the time percentage for PDCCH-only slots is close to 100%, so
for non- DRX scenarios cases the cross-slot savings could approach 30%.

Bandwidth part adaptation

The processing requirement in NR UEs is related to the number of subcarriers that have to be
processed, and therefore to the BWP bandwidth. A smaller bandwidth means a lower sampling
rate and a smaller FFT, and this has a direct relationship to power consumption. However, it also
reduces the maximum transport block size, which makes data transfer less efficient.

NR allows for rapid switching between up to four BWPs using the control channel DCI. BWP
switching incurs a small delay for retuning, filter settling and channel adaptation during which no
data transfer is performed.

The analysis presented here assumes that 2 BWPs are configured. The first is a 20MHz BWP
(power saving) that is used for control channel monitoring and the transfer of small data packets.
52
 GTI 5G Device Power Consumption Whitepaper

The second is a 100MHz BWP (full power) which is used for large data packets and (while it is
active) for all other purposes. When the network has large data to transmit to the UE, it uses the
DCI to switch the UE to the full power BWP, where it remains while there is data to transfer. After
a timeout of 20ms in which there has been no data transfer it reverts to the power saving BWP.
BWP switching is assumed to add 7 slots delay when switching from a low bandwidth to a high
bandwidth BWP, to allow for channel adaptation.

In the power model, a 20MHz BWP requires 40% of the power for a 100MHz BWP, subject to a
floor of 50 model units.

Sleep and uplink power are not affected by BWP switching. The remaining full power BWP
parameters are the same as the baseline case. For the power saving BWP, power for
PDCCH+PDSCH is 120 units (0.4*300) and power for PDCCH-only, SSB/CSI-RS and BWP switching
is 50 units.

Results are presented below in Table 5- 18 – Table 5- 20.

FTP traffic with BWP adaptation

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 0.13% 0.06 0.15%

Light sleep 20 5.43% 1.09 2.74%

Deep sleep 1 55.04% 0.55 1.39%

Power saving 50 23.77% 11.89 29.95%

PDCCH-only
Full power 100 8.02% 8.02 20.21%

Power saving 120 0.17% 0.20 0.51%

PDCCH+PDSCH
Full power 300 4.05% 12.15 30.62%

Power saving 50 1.28% 0.64 1.61%

SSB/CSI-RS
Full power 100 0.10% 0.10 0.25%

UL (0dBm) 250 0.34% 0.85 2.14%

BWP switching 50 1.67% 0.84 2.10%

Transitions Substate Model energyInstances/10sec Average power% power

53
 GTI 5G Device Power Consumption Whitepaper

LS transitions 100 40.60 0.41 1.02%

DS transitions 450 64.32 2.89 7.29%

Saving 23.67% Total power 39.68 100.00%

Table 5- 18 FTP traffic power reduction using BWP switching

For FTP traffic, the total time spent in each power state has not changed significantly from the
baseline, but a large proportion of the downlink activity now takes place in the power saving BWP.
This saves 23.7% of the power. With a shorter timeout, or if the network switches the UE to the
power saving BWP when it has no more data, the power savings would be higher.

54
 GTI 5G Device Power Consumption Whitepaper

IM traffic with BWP adaptation

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 0.03% 0.01 0.17%

Light sleep 20 2.88% 0.58 7.11%

Deep sleep 1 89.85% 0.90 11.09%

Power saving 50 5.15% 2.58 31.79%

PDCCH-only
Full power 100 0.91% 0.91 11.24%

Power saving 120 0.02% 0.02 0.30%

PDCCH+PDSCH
Full power 300 0.04% 0.12 1.48%

Power saving 50 0.70% 0.35 4.32%

SSB/CSI-RS
Full power 100 0.01% 0.01 0.12%

UL (0dBm) 250 0.18% 0.45 5.56%

BWP switching 50 0.22% 0.11 1.36%

Transitions Substate Model energyInstances/10sec Average power% power

LS transitions 100 21.35 0.21 2.64%

DS transitions 450 41.08 1.85 22.83%

Saving 24.91% Total power 8.10 100.00%

Table 5- 19 IM traffic reduction using BWP adaptation

In the IM case there is very little activity in the full power BWP, and the power saving is slightly
higher than FTP at 24.9%. There is still significant power consumption for PDCCH-only in the full
power BWP, because of the timeout. As before, a shorter timeout or an active switch back to the
low power BWP would increase the power saving

VoIP traffic with BWP adaptation

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 2.99% 1.35 3.91%

55
 GTI 5G Device Power Consumption Whitepaper

Light sleep 20 43.52% 8.70 25.32%

Deep sleep 1 24.11% 0.24 0.70%

Power saving 50 22.70% 11.35 33.02%

PDCCH-only
Full power 100 0.00% 0.00 0.00%

Power saving 120 1.25% 1.50 4.36%

PDCCH+PDSCH
Full power 300 0.00% 0.00 0.00%

Power saving 50 5.10% 2.55 7.42%

SSB/CSI-RS
Full power 100 0.00% 0.00 0.00%

UL (0dBm) 250 0.32% 0.80 2.33%

BWP switching 50 0.00% 0.00 0.00%

Transitions Substate Model energyInstances/10sec Average power% power

5.322 100 322.24 3.22 9.38%

9.5589 450 103.52 4.66 13.55%

Saving 31.63% Total power 34.37 100.00%

Table 5- 20 VoIP power saving using reduced bandwidth BWP

For VoIP traffic, there are no large data packets to trigger a switch to the full power BWP, so all
downlink activity takes place in the power saving BWP. This gives the best power saving of the
three scenarios at 31.6%.

Antenna adaptation

Each active transceiver/antenna makes a direct contribution the UE power consumption from its
own power requirement, and also an indirect contribution through the increase in processing
workload. In the 3GPP model, 2 active antennas require 30% less power than 4 antennas.

The simulations for these cases use a BWP based approach with DCI based switching, as in the
previous section, with a power saving 2 layer BWP for PDCCH monitoring and small data packets,
and a full power 4 layer BWP for large data. Simulation results are listed in Table 5- 21 - Table
5- 23 VoIP traffic power saving with antenna adaptation

FTP traffic with antenna adaptation

56
 GTI 5G Device Power Consumption Whitepaper

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 0.13% 0.06 0.13%

Light sleep 20 5.40% 1.08 2.40%

Deep sleep 1 55.15% 0.55 1.23%

Power saving 70 24.61% 17.23 38.29%

PDCCH-only
Full power 100 8.95% 8.95 19.89%

Power saving 210 0.17% 0.36 0.79%

PDCCH+PDSCH
Full power 300 3.86% 11.58 25.74%

Power saving 70 1.29% 0.90 2.01%

SSB/CSI-RS
Full power 100 0.10% 0.10 0.22%

UL (0dBm) 250 0.35% 0.88 1.94%

Transitions Substate Model energyInstances/10sec Average power% power

LS transitions 100 40.42 0.40 0.90%

DS transitions 450 64.63 2.91 6.46%

Saving 13.45% Total power 44.99 100.00%

Table 5- 21 FTP traffic power saving with antenna adaptation

The power reduction for the 2 antenna case is smaller than the power reduction from 100MHz to
20MHz bandwidth, but the behaviour characteristics are similar. Downlink activity spends more
time in the 2 antenna power saving configuration than it does in the 4 antenna full power
configuration, achieving a 13.4% power saving. PDCCH activity at full power is still significant, but
could be reduced by applying a shorter timeout, or by dynamically switching back to the 2
antenna case when there is no more data for the user.

57
 GTI 5G Device Power Consumption Whitepaper

IM traffic with antenna adaptation

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 0.03% 0.01 0.14%

Light sleep 20 2.88% 0.58 6.15%

Deep sleep 1 89.85% 0.90 9.60%

Power saving 70 5.26% 3.68 39.32%

PDCCH-only
Full power 100 1.02% 1.02 10.89%

Power saving 210 0.02% 0.04 0.45%

PDCCH+PDSCH
Full power 300 0.04% 0.12 1.28%

Power saving 70 0.70% 0.49 5.23%

SSB/CSI-RS
Full power 100 0.01% 0.01 0.11%

UL (0dBm) 250 0.18% 0.45 4.81%

Transitions Substate Model energyInstances/10sec Average power% power

LS transitions 100 21.35 0.21 2.28%

DS transitions 450 41.08 1.85 19.74%

Saving 13.18% Total power 9.36 100.00%

Table 5- 22 IM traffic power saving with antenna adaptation

The saving in the IM case is similar at 13.1%, and the FTP conclusions are also applicable to this
case

VoIP traffic with antenna adaptation

Power state Substate Model power Time percentageAverage power% power

Microsleep 45 2.94% 1.32 3.25%

Light sleep 20 43.55% 8.71 21.41%

Deep sleep 1 24.15% 0.24 0.59%

Power saving 70 22.90% 16.03 39.40%

PDCCH-only
Full power 100 0.00% 0.00 0.00%

PDCCH+PDSCHPower saving 210 1.20% 2.52 6.19%

58
 GTI 5G Device Power Consumption Whitepaper

Full power 300 0.00% 0.00 0.00%

Power saving 70 5.10% 3.57 8.78%

SSB/CSI-RS
Full power 100 0.00% 0.00 0.00%

UL (0dBm) 250 0.16% 0.40 0.98%

Transitions Substate Model energyInstances/10sec Average power% power

LS transitions 100 322.35 3.22 7.92%

DS transitions 450 103.68 4.67 11.47%

Saving 19.08% Total power 40.68 100.00%

Table 5- 23 VoIP traffic power saving with antenna adaptation

The VoIP simulation never enters the full power substates, and the power reduction is
correspondingly higher at 19%

59
 GTI 5G Device Power Consumption Whitepaper

5.3.6 Nonstandalone comparison

In this comparison the traffic model used combines the FTP and IM traffic scenarios used in
previous sections, and in the NSA case routes the IM traffic through the LTE modem and the FTP
traffic through the NR modem. For the SA case both sets of data traffic are routed through the NR
modem

Simulation results for the LTE modem in the NSA configuration are listed in Table 5- 24.

The power model parameters are based on 20MHz BWP scaling, using the assumptions of section
5.3.2

LTE modem in NSA configuration, mixed data traffic

Power state Model power Time percentage Average power % power

Microsleep 45 0.00% 0 0.00%

Light sleep 20 2.87% 0.57 7.15%

Deep sleep 1 89.24% 0.89 11.11%

PDCCH-only 40 6.02% 2.41 29.98%

PDCCH+PDSCH 120 0.63% 0.76 9.41%

Sync/meas 40 0.63% 0.25 3.14%

PDCCH+sync/meas 68 0.31% 0.21 2.62%

PDCCH+UL 290 0.31% 0.90 11.19%

Model energy Instances/10sec Average power % power

LS transitions 100 22.34 0.22344 2.78%

DS transitions 450 40.37 1.816713 22.62%

Total Power 8.03 100%

Table 5- 24 Power consumption for the LTE modem in a nonstandalone NR configuration,

mixed data traffic

The resulting power level of 8.03 is lower than the baseline standalone NR modem with IM traffic,
and slightly lower than the same modem with 20MHz BWP reduction.

60
 GTI 5G Device Power Consumption Whitepaper

Simulation results for the NR modem in the NSA configuration are listed in5-25. The power model
parameters are essentially the same as for the baseline case.

61
 GTI 5G Device Power Consumption Whitepaper

NR modem in NSA configuration, mixed data traffic

Power state Model power Time percentage Average power % power

Microsleep 45 0.49% 0.22 0.43%

Light sleep 20 5.16% 1.03 2.02%

Deep sleep 1 55.41% 0.55 1.08%

PDCCH-only 100 33.95% 33.95 66.47%

PDCCH+PDSCH 300 3.51% 10.53 20.62%

SSB/CSI-RS 100 1.32% 1.32 2.58%

UL 250 0.16% 0.40 0.78%

Model energy Instances/10sec Average power % power

LS transitions 100 38.93 0.39 0.76%

DS transitions 450 59.60 2.68 5.25%

Total Power 51.08 100%

Table 5- 25 Power consumption for the NR modem in a nonstandalone NR configuration, mixed

data traffic

In this case the power level of 51.08 units is lower than for the baseline standalone NR modem
with FTP traffic (51.99 units). The power distribution is very similar to the baseline case, but there
is a slight reduction in PDCCH+PDSCH traffic for the NSA case, which may be due to signaling
differences

Results for the standalone NR modem are listed in Table 5- 26.

NR modem in standalone configuration, mixed data traffic

Power state Model power Time percentage Average power % power

Microsleep 45 0.51% 0.23 0.43%

Light sleep 20 5.03% 1.006 1.88%

Deep sleep 1 53.42% 0.5342 1.00%

PDCCH-only 100 35.91% 35.91 67.15%

PDCCH+PDSCH 300 3.67% 11.01 20.59%

SSB/CSI-RS 100 1.30% 1.3 2.43%

62
 GTI 5G Device Power Consumption Whitepaper

UL 300 0.16% 0.48 0.90%

Model energy Instances/10sec Average power % power

LS transitions 100 37.99 0.379857 0.71%

DS transitions 450 58.31 2.6239275 4.91%

Total Power 53.47 100%

Table 5- 26 Power consumption for a standalone NR modem, mixed data traffic

Compared with the baseline FTP case (51.99 units), the additional IM traffic means that power
for the mixed traffic case increases to 53.47 units.

The combined power for the nonstandalone mixed traffic case is (8.03+51.08) = 59.11 model
units, compared with the standalone mixed traffic power of 53.47 model units. For this traffic
pattern the UE in the NSA configuration requires 10.55% more power from its battery than in an
SA configuration.

5.4 Enhancements of Power Saving in Release 16

Release 15 included baseline features in NR that enable UE power saving. Power savings of 80%
or more can be achieved using DRX in RRC_CONNECTED, with a further 20%+ power
consumption reduction when BWP is configured to reduce the UE processing bandwidth when
high throughput demand is not needed.
Release 16 further enhances UE power efficiency with a focus on reducing background activity
whenever possible, and simplifying the ways in which UE power management can adapt quickly
to dynamically varying traffic conditions. A summary of those features is illustrated in Figure 5-7,
and described in more detail in subsequent sections.

 A power saving signal is introduced to regulate the DRX cycle. The DRX OnDuration
timer (❶) will not be activated if the preceding power saving signal indicates “do not
wake up”, which reduces power consumption in periods of low data activity.
 L1 signaling can provide fast adaptation to changes in data activity of the cell
configuration (# MIMO layers, cross-slot/same slot scheduling), to reduce power
consumption during active time (❷)
 Activated SCells can be dormant or non-dormant. A UE does not monitor PDCCH in a
dormant SCell until instructed to do so by signaling in the PCell, but will perform
occasional measurements as needed to maintain the link. This saves power while the
SCell is dormant (❸), and allows a quick transition to non-dormancy when higher data
volume is about to be scheduled.

63
 GTI 5G Device Power Consumption Whitepaper

Figure 5- 7 Release 16 power saving enhancements

5.4.1 BWP adaptation framework

The bandwidth part in NR is more than simply a way of partitioning the carrier bandwidth
efficiently. Each bandwidth part that is specified (up to four for each carrier/UE) has its own
configuration for each of the physical channels, for dedicated resource allocations and for link
management. A UE in RRC_CONNECTED can be dynamically switched between BWPs using the
DCI, allowing a shorter switching time between BWP configurations. In Release 16 BWP
framework was further extended, to include some layer-1 configurations that were defined for
carrier level only in earlier release, like number of MIMO layers.

5.4.1.1 PDCCH Monitoring Reduction

One of the enhancements on the BWP framework in R16 is the inclusion of Periodic PDCCH
monitoring reduction, and while the concept was already defined in R15 (refer to section 5.2.7),
it was based on RRC reconfiguration which is not fast enough to adapt to fluctuating traffic
behavior aside its signaling overhead increase. So in R16, for example the UE could simply be
switched from PDCCH monitoring 100% i.e. every slot (data efficiency) to 25% i.e. each 4 slots
(power saving), with DCI based BWP switching.

5.4.1.2 SCell dormancy

When a SCell is first activated, the UE needs around 30ms to synchronize for link adaptation
before user data transfer can take place. This time delay makes it difficult to match the number of
active SCells to dynamically changing traffic patterns, and it is inefficient to perform PDCCH
monitoring on a SCell if no data is expected.

Release 16 introduces the concept of SCell dormancy to the active state. If an active SCell is
dormant, the UE will not monitor PDCCH on the cell, but it will perform CSI measurements, AGC

64
 GTI 5G Device Power Consumption Whitepaper

adjustments and beam management as necessary to maintain the link to the cell. This
background activities will typically employ a longer periodicity than what is required for PDCCH
monitoring, but it enables a rapid transition to non-dormancy when user data transfer is needed.

Figure 5- 8 SCell State diagram for Rel-16

The transition from dormancy can be triggered by the power saving signal (see section 5.4.2) at
the start of an active DRX period, or by DCI signaling in the PCell or PSCell if the UE is in the Active
Time. Return to dormancy can be triggered by DCI signaling in the SCell itself, or simply with the
bwp-InactivityTimer expiration.

Figure 5- 9 SCell transition from and to dormancy

Power Saving gains up to 49% can be achieved for CA FR1 (1 x 20 MHz CC, 4-layer MIMO) +FR2 (4
x 100 MHz CC, 2-layer MIMO), with C-DRX configuration (Cycle, On-duration, Inactivity timer) =
(40 ms, 10 ms, 20 ms) and BWP Switching Timer of 8ms. Latency evaluation on the other hand
showed about 17% increase.
65
 GTI 5G Device Power Consumption Whitepaper

5.4.1.3 MIMO layer adaptation

In release 15 the maximum number of downlink MIMO layers supported is carrier specific, and is
not normally changed (although the UE can request a reduction to mitigate overheating). Release
16 allows the maximum number of MIMO layers to be configured for different BWPs on a carrier,
provided that at least one BWP on the carrier is configured with the maximum layers that the UE
can support.

This means DCI-based BWP switching can be used to dynamically configure the maximum
number of active MIMO layers in a downlink transmission, allowing the UE to reduce its power
consumption when fewer receive paths are used.

Evaluation using the FTP/Video traffic model gives power savings of around 20%, when the
maximum number of MIMO layers is reduced (e.g. from 4 layers to 2 layers).

5.4.1.4 Cross-slot scheduling

Although cross-slot scheduling was considered for power saving in release 15, and while it could
be used for data scheduling, it had a limitation as the slot offset for aperiodic CSI-RS was fixed at
K0=0 for sub-6GHz, which means that cross-slot scheduling could not be used if aperiodic CSI-RS
was configured, as the UE needs to continuously process and decode the PDSCH in anticipation
for A-CSI-RS.

This restriction has been resolved in release 16, and enabling cross-slot scheduling to be used in
all configurations. Power reductions of around 20% are observed from cross-slot scheduling, with
further reductions if this is combined with MIMO layer reductions.

Figure 5- 10 Same-slot (K0=0) vs. Cross-Slot (K0>0) Scheduling

66
 GTI 5G Device Power Consumption Whitepaper

The minimum applicable k0 and k2 values which determine whether cross-slot scheduling is
active are part of the physical channel configuration, which means that cross-slot scheduling can
be dynamically activated and deactivated using the BWP adaptation framework. However, if the
number of BWP is limited, cross-slot scheduling can also be activated and deactivated within the
context of the active BWP using PDCCH.

5.4.2 Wakeup signal for cDRX optimization

Release 16 introduces a power saving signal (channel) to reduce power consumption in UEs that
support the feature. The signal uses a new DCI format 2_6, scrambled by PS-RNTI (Power
Saving-RNTI), to provide a “UE wake up or not” indication for one or more UEs sharing the same
PS-RNTI. The format can also be used to transition active SCells out of the dormant state. If the
power saving signal is not decoded successfully during the monitoring period, the network can
configure the UE default behavior to be either “wake up” or “do not wake up” using higher layer
signaling.

If a UE in the RRC_CONNECTED state is configured with a DRX cycle and has the power saving
signal configured on the active BWP in an active cell, it monitors the power saving signal at a time
PS-offset (higher layer parameter) before the start of DRX ON in the long DRX cycle. If the
outcome of monitoring (decoded or default) is “do not wake up”, the UE enters DRX OFF and
does not start the On-duration timer for the next DRX cycle. The power saving signal is not
monitored during active portions of the long DRX cycle, or at any time when the short DRX cycle
is active.

Figure 5- 11 Wake UP Signal

Generally the probability of data allocation during c-DRX on-duration is very low under normal
DoU cases, as well the ctrl-signal allocation in PDCCH is also relatively rare. So WUS could provide
considerable power saving gains since it allows UE to skip reception in these parts. The rationale

67
 GTI 5G Device Power Consumption Whitepaper

behind the power saving signal is that with appropriate configuration it can be much more energy
efficient to decode a simple power saving signal and return to the inactive state than to perform
a full PDCCH decode over multiple search spaces.

The PS-offset delay between the wakeup signal and the start of DRX ON may be used (when
wake-up is indicated) to update CSI information or perform fine time/frequency tracking as
necessary.

Initial evaluation suggests that for sparse data traffic (e.g. instant messaging), reductions of the
order of 20-30% can be achieved relative to a release 15 DRX cycle. For more dense traffic (e.g.
FTP, video) the active portion of the DRX cycle is more dominant in determining the UE average
power consumption, and power saving gains are correspondingly lower.

5.4.3 RRM measurements relaxation

RRM (Radio Resource Management) measurements are an essential part of link management in
high mobility scenarios, or poor RF conditions for users in RRC_IDLE or RRC_INACTIVE and this
measurement overhead can be a significant component of energy consumption.

This overhead could be reduced then for users with low mobility and/or not at cell edge by
allowing those UEs to apply RRM relaxation for intra/inter-frequency neighboring cells whenever
they meet any or both of the low mobility and not at cell edge criterions with a relaxation factor
applicable for fulfilled case.

The relaxed measurement criterion for UE with low mobility is fulfilled when:

- (SrxlevRef – Srxlev) < SSearchDeltaP,

Where Srxlev = current Srxlev value of the serving cell (dB) and SrxlevRef = reference Srxlev value
of the serving cell (dB).
And the relaxed measurement criterion for UE not at cell edge is fulfilled when:
- Srxlev > SSearchThresholdP, and,
- Squal > SSearchThresholdQ, if SSearchThresholdQ is configured,
Where Srxlev = current Srxlev value of the serving cell (dB) and Squal = current Squal value of the
serving cell (dB).

RRM neighboring measurements relaxation in connected mode has also been studied and
evaluated in Release 16 but considering potential impacts on UE mobility performance it has not
been specified, although it is still possible from network control point of view. And RRM
measurement relaxation for serving cell was not specified for UE in any RRC state.

68
 GTI 5G Device Power Consumption Whitepaper

5.4.4 Extended UE Assistance information

The UEAssistanceInformation message provides a mechanism for the UE to signal configuration

preferences to the network. In release 15 two optional IEs are specified for this message:
 DelayBudgetReport: contains the UE preference for adjustment of the DRX cycle
duration (positive, negative or zero) in RRC_CONNECTED.
 OverheatingAssistance: contains the UE preference for temporary capability reduction
in the event of overheating. The assistance request can set upper limits on any or all of
the number of SCells, the aggregate bandwidth, and the number of MIMO layers in the
uplink and downlink in FR1 and FR2. If the UE no longer experiences overheating, the
indication is sent with no limits specified.

Triggers for sending the message are primarily a function of UE implementation, but the network
can restrict the maximum message frequency by configuring prohibit timers (one for each IE)
which start when the first IE is sent and prevent the sending of second IE of the same type before
timer expiry.

In release 16, extensions to this mechanism have been put in place for multiple purposes,
including further reductions in UE power consumption. A new power saving assistance IE, will
allow the UE to report its preference for the connected mode DRX parameters, the maximum
aggregated bandwidth, the maximum number of secondary component carriers, the maximum
number of MIMO layers, and/or the minimum scheduling offset for cross-slot scheduling cycle
length for power saving purpose.

In all cases, it is under the network control whether to accommodate the UE preference.

5.4.4.1 DRX Preference

Release 15 UE assistance already makes provision for adjustments to the length of the connected
mode DRX cycle in the DelayBudgetReport IE. In release 16 the UE will be able to express
preferences for a larger set of DRX parameters, based on its traffic patterns and latency
requirements. The parameters for which a UE could indicate a preference are the long and short
DRX cycle durations and the inactivity and short DRX cycle timers. The OnDuration timer and
start offset are defined by the network, and do not form part of the assistance message

As with other forms of UE assistance, the network has the final decision on the configuration
selected.

5.4.4.2 Max BW and Max CC Preference

The number of active SCells, their total aggregated bandwidth and their distribution among
frequency bands, influence UE power consumption, as do issues of coexistence. Consequently
69
 GTI 5G Device Power Consumption Whitepaper

the best power configuration for a particular UE is implementation dependent. SCell dormancy
(section 5.4.1.2) allows the network more flexibility to allocate SCells to a particular UE based on
traffic patterns, but does not provide a feedback mechanism for the UE to express a configuration
preference.

Release 16 UAI extension allows for a UE to indicate to the network its preference for Aggregated
BW for DL and UL and for both FR1 and FR2 components, as well as its preference to the number
of Carriers Components (CC) for DL and UL to better save power.

UE can implicitly indicate a preference for NR SCG release by indicating zero number of carriers
and zero aggregated maximum bandwidth in both FR1 and FR2.

5.4.4.3 Minimum Scheduling Offset Preference

A reduction in operating frequency can allow circuits to operate at a lower voltage, with a
consequent reduction in power consumption. Real time constraints generally determine the
minimum operating frequency, so if real time requirements can be relaxed this may allow
additional power savings. The achievable savings will depend on UE implementation, and this
makes a case for UEs signaling their preferred minimum values for K0 (DCI to PDSCH delay) and
K2 (DCI to PUSCH delay) to the network if changes to the default configuration will permit
additional power savings.

This signaling can be viewed as an extended version of cross-slot scheduling (see section 5.4.1.4).

Preferred minimum values will be reported for each supported subcarrier spacing, with {1,2,4,6}
slots possible for FR1 with SCS 15KHz and 30 KHz, and {2,4,8,12} slots possible for FR2 with SCS
60KHz and 120KHz.

5.4.4.4 Release Preference for efficient RRC state transitions

If a UE in RRC_CONNECTED does not expect to send or receive data in the near future (for
example, if an ongoing file transfer is completed), it may be able to reduce its power
consumption by returning quickly to RRC_INACTIVE or RRC_IDLE. Release 16 introduces a new IE
for connection release assistance which fulfils this purpose.

The network is the final arbiter of whether and which state the UE should transition to (i.e.
autonomous release by the UE is not supported), but the UE could express a preference with
PreferredRRC-State-r16 IE being set to {idle, inactive or connected}.

5.4.5 SECONDAY DRX

In release 15, whenever FR1 and FR2 cells are configured through Carrier Aggregation (CA) then
all the cells share the same cDRX configuration and operation. A c-DRX enhancement was
70
 GTI 5G Device Power Consumption Whitepaper

proposed allowing configuration of a secondary c-DRX Cycle applied to second group of cells (e.g.
FR2 cells) to go to sleep earlier, and thus reduce power consumption.

In order to minimize the impact on Network scheduler, cells are expected to become available for
scheduling at the same time, and since FR2 cells operate with a different numerology compared
to FR1, therefore the OnDuration can be shorter compared to FR1, while providing the same
number of PDCCH occasions.

Figure 5- 12 Legacy and Secondary DRX Cycle for FR1 and FR2

In Release 16, only the basic functionality for secondary DRX was specified – only OnDuration
could be different between 1st and 2nd DRX groups- , with no specification for joint configuration
with DCP or joint configuration with SCell dormancy during active time, also the combination of
cross-carrier scheduling and secondary DRX group is not supported. The inter-operations
between DRX groups are not introduced in Rel-16, e.g. the related behaviour for
drx-ShortCycleTimer/ drx-InactivityTImer is handled per-DRX group or for both DRX groups.

So while the feature could provide more promising UE power saving, it is still not very mature in
R16 and could use further enhancements for flexibility in R17.

for CA FR1 (1 x 20 MHz CC, 4-layer MIMO) +FR2 (4 x 100 MHz CC, 2-layer MIMO), with legacy
C-DRX configuration (Cycle, On-duration, Inactivity timer) = (40 ms, 10 ms, 20 ms) for FR1 and
2nd DRX configuration for FR2 (Cycle, On-duration, Inactivity timer) = (40 ms, 5 ms, 10 ms ), 26%
power gain can be achieved, while latency showed 38% increase.

5.4.6 Proposed Features Combination for different traffic profiles:

While each individual one of the features presented earlier provide different power saving gain,
that might differ based on the network configuration and the traffic patterns it is applied to, it is
expected that in real deployment several features could be applied together or on top of each
71
 GTI 5G Device Power Consumption Whitepaper

other to achieve considerable total power saving gain to the UEs.

In this section, and for different network frequency allocations (FR1 and FR2), and different traffic
profiles, power saving gains are being measured starting from a baseline features combination,
and then other features are being added on top of the baseline combination one by one with an
evaluation of each feature contribution to the gains.

So starting with the simulation assumptions:

Simulation parameters FR1 FR2

Carrier center frequency 4 GHz 30 GHz
Subcarrier spacing 30 kHz 120 kHz
Bandwidth (per CC) 100MHz 100MHz
# of CC 1, 2 4, 8
Channel model IMT2020 3D UMa IMT2020 3D UMa
Deployment Dense Urban Dense Urban
ISD 200 m 200 m
# of BS Tx antennas 32; (M, N, P, Mg, Ng) = 2
(8, 8, 2, 1, 1)
# of UE Rx antennas 4 2

Table 5- 27 Model used for Power Saving assessment

For FR1, BW scaling is assumed to be from 100MHz (data efficiency) to 20MHz (power saving),
and PDCCH monitoring reduction from 100% i.e. every slot (data efficiency) to 25% i.e. each 4
slots (power saving).

The FR1 simulation evaluates power saving for 5 configurations with different combinations:

 Config. A = R15 DRX baseline + BW scaling + SCell dormancy for 2CC

 Config. B = Config. A + PDCCH monitoring reduction (duty cycle: 100%  25%)
 Config. C = Config. B + WUS/DCP
 Config. D = Config. B + MIMO Rx path reduction
 Config. E = Config. B + Cross-slot scheduling

FTP IM (Instant Message)

Power Saving Features
Power saving gain (%) Power saving gain (%)
A: R15 DRX baseline + BW scaling 11.25 % (1 CC) 12.06 % (1 CC)
+ SCell dormancy for 2CC 26.75 % (2 CC) 25.72 % (2 CC)
B: A + PDCCH monitoring reduction 18.71 % (1 CC) 19.93 % (1 CC)
(Duty cycle: 100% 25%) 32.92 % (2 CC) 32.21 % (2 CC)
21.79 % (1 CC) 27.66 % (1 CC)
C: B + WUS/DCP
35.52 % (2 CC) 38.61 % (2 CC)
19.73 % (1 CC) 21.14 % (1 CC)
D: B + MIMO Rx path reduction
33.76 % (2 CC) 33.22 % (2 CC)
72
 GTI 5G Device Power Consumption Whitepaper

19.15 % (1 CC) 20.33 % (1 CC)

E: B + Cross-slot scheduling
33.28 % (2 CC) 32.54 % (2 CC)

Table 5- 28 Power Saving Gains for different features combinations for FR1

From the above results, it could be seen that for FTP Traffic type Config. B provides majority of
the power saving gains, with marginal gains on top of it with R16 MIMO path reduction to 2Rx,
R15/R16 Cross-slot scheduling or R16 WUS.

While for IM traffic Type, Config. C which adds R16 WUS on top of Config. B provides the highest
possible power saving gain.

So for FR1 the following features combinations could be recommended:

 For FTP Traffic type: R15 DRX baseline + BW scaling + PDCCH monitoring reduction+
SCell dormancy for 2CC
 For IM traffic Type: R15 DRX baseline + BW scaling + PDCCH monitoring reduction+
SCell dormancy for 2CC + R16 WUS
Moving to FR2, The simulation evaluates power saving for 4 configurations with different
combinations:

 Config. A = DRX + SCell dormancy (4 CC or 8 CC  1 CC)

 Config. B = Config. A + PDCCH monitoring reduction (duty cycle: 100%  25%)
 Config. C = Config. B + WUS/DCP
 Config. D = Config. B + Cross-slot scheduling

FTP IM (Instant Message)

Power Saving Features
Power saving gain (%) Power saving gain (%)
A: DRX + SCell dormancy 21.90 % (4 CC) 18.05 % (4 CC)
(4 CC or 8 CC 1 CC) 32.78 % (8 CC) 34.04 % (8 CC)
B: A + PDCCH monitoring reduction 39.03 % (4 CC) 33.29 % (4 CC)
(Duty cycle: 100% 25%) 46.40 % (8 CC) 45.94 % (8 CC)
40.31 % (4 CC) 37.63 % (4 CC)
C: B + WUS/DCP
47.52 % (8 CC) 49.33 % (8 CC)
D: B + Cross-slot scheduling 39.49 % (4 CC) 33.70 % (4 CC)
46.77 % (8 CC) 46.26 % (8 CC)

Table 5- 29 Power Saving Gains for different features combinations for FR2

In general BW reduction delivers negligible power scaling for FR2, and from the results presented
above it could be seen that a similar recommendation to the one provided for FR1 could apply
here for FR2, where R15 Baseline DRX + R16 Scell dormancy + R15 PDCCH reduction could be
recommended for FTP traffic type and R16 WUS could be considered on top of them for IM and
sporadic traffic types.

73
 GTI 5G Device Power Consumption Whitepaper

5.5 Enhancements of Power Saving in Release 17

Release 17 introduces new 5G device types with reduced capabilities, namely REDCAP devices,
targeting reducing device complexity and applicable to various use cases including industrial
wireless sensors, smart wearables and video surveillance.

And while release 16 focused on power saving optimization for RRC connected mode, for Release
17 Idle mode/RRC Inactive power saving optimization was put into focus - along with further
extensions to R16/15 features for connected mode.

Rel-17 Power Saving will mainly focus on:

 Enhancements for idle/inactive mode power saving:

o Paging enhancement(s) to reduce unnecessary UE paging receptions

o Providing potential TRS/CSI-RS occasion(s) available in connected mode to

idle/inactive-mode UEs

 Enhancements for connected-mode power saving:

o Extension(s) to Rel-16 DCI-based power saving adaptation during DRX Active Time
for an active BWP, including PDCCH monitoring reduction when C-DRX is
configured

o Relaxing UE measurements for RLM and/or BFD, particularly for low mobility UE
with short DRX periodicity/cycle

5.5.1 NR Idle-Mode Power Consumption Issue

The RRC idle/inactive state UE is required to monitor one paging occasion per DRX cycle to detect
the scheduling of paging and system information update.

But one of the challenges of the ultra-lean physical layer design in NR compared to LTE, especially
in poor coverage/low SINR areas is that, in order for the UE to decode paging PDSCH successfully,
it needs to receive multiple SS bursts for pre-synchronization before the paging occasion (PO).
And due to the SS burst periodicity, it is not possible for a UE to enter deep sleep before PO
which results in UE power consumption waste, especially if the UE is not paged. For LTE, the UE
can go out of deep sleep state to synchronize, receive and process the PO, and carry out
neighboring measurements consecutively and then go back to deep sleep again.

In Figure 5- 8, there compare UE idle-mode processing timelines in LTE and NR SA networks,

respectively. As shown, more power state transitions are needed for NR UEs, this is because LTE
has CRS in every sub-frame, and even if UE needs to monitor multiple CRSs before paging
74
 GTI 5G Device Power Consumption Whitepaper

reception, it needs not to wake up several time as in NR.

Figure 5- 13: Idle-mode power consumption comparison between LTE and NR

Initial analysis shows up to 73% power consumption increase for NR UE compared to LTE UE in
Low SINR. But even for high SINR condition, NR idle-mode power consumption is still 26% higher
than LTE, where the additional wake-up energy overhead for a SS burst apart from PO is the main
cause.

5.5.2 Paging Enhancements for UE Power Saving in NR

The above results highlight the need for UE idle mode behaviour optimization in 5G SA, targeting
a power consumption performance in par with LTE. Rel. 17 paging enhancements framework
aims for minimizing UE wake-up energy overhead, particularly when there is no paging to the UE,
which could be achieved through the following proposals.

Paging Early Indication (PEI)

The idea is to let the network send an indication before a PO, whether any UE monitoring the PO
is to be paged and the UE does not need to receive PO if negative indication is received. For
which case the UE can skip not only paging PDCCH/PDSCH but also unnecessary SSB processing

Figure 5- 14: Paging Early Indication (PEI)

for pre-synchronization and all related power state transitions and go to deep sleep to save
power.

75
 GTI 5G Device Power Consumption Whitepaper

UE Grouping

The idea is to further divide UEs monitoring the same PO into subgroups, so as to reduce the
“false alarm rate”, i.e. the chances that UE decodes paging message and finds that it is for
another UE in the group rather than itself.

Figure 5- 15: UE grouping

In the figure below, there illustrate the corresponding UE processing timelines with PEI and UE
grouping, respectively. With PEI, UE can skip SS burst and PO processing after early indication of
no paging. On the other hand, UE grouping cannot reduce any SS burst processing and the
corresponding wake-up energy overhead; otherwise, performance is degraded if UE is paged.

Figure 5- 16: UE processing timelines: (a) Rel-15/16, (b) paging early indication, (c) UE grouping

UE Group PEI (GPEI)

The idea is to combine both enhancements discussed above to early indicate whether the
corresponding sub-group of UEs need to monitor PO. This reduces the probability UE is falsely
paged and improves the power saving gain, particularly when original group paging rate is high.

76
 GTI 5G Device Power Consumption Whitepaper

Figure 5- 17: UE Group PEI

Initial evaluation over the three schemes shows that UE Group PEI can achieve the best power
saving gain. When original group paging rate (GPR) is 10%, there can realize 25% power saving.

Figure 5- 18: Power saving gains for different paging enhancements in low/high SINR

5.5.3 TRS/CSI-RS for idle/inactive UE

One more solution to minimize the number of times an idle/inactive UE in low SINR needs to
wake up to synchronize, is to consider additional reference signals (RS).

TRS/CSI-RS already configured to UEs in RRC Connected mode, could be leveraged for
idle/inactive mode UEs also, with the following gains in mind:

 As UE needs to monitor one SSB burst and one TRS before PO, instead of 3 SSB
bursts in Low SINR, leading to less power states transition and UE power saving.

 Reduce the impact on network power efficiency and resource overhead.

Figure 5- 19: Paging monitoring with TRS information for low-SINR UE

The configuration of TRS/CSI-RS occasion(s) for idle/inactive mode UE(s) is provided by higher
layer signaling, and a UE may apply both PEI and TRS/CSI-RS:

 PEI helps save power when no UE in the group not paged; UE can even skip paging
77
 GTI 5G Device Power Consumption Whitepaper

PDCCH monitoring;

 TRS/CSI-RS information helps save power when some UE in the group is paged, i.e.,
when UE needs to decode paging PDSCH and thus requires higher accuracy.

5.5.4 Extension to Rel-16 DCI-based power saving adaptation

As discussed in previous section, In Rel-15 and Rel-16 NR, DCI-based power saving adaptations
are developed in addition to common DRX feature. Via DCI indication, network can indicate UE to
apply reduced reception bandwidth, reduced number of non-dormant SCells, reduced number of
MIMO layers, cross-slot scheduling and longer PDCCH monitoring periodicity when there is no
data for the UE.

One area with possible improvement is that in Release 16, whenever a UE wakes up for data
transmission, the DRX Inactivity Timer would be started and UE has to monitor PDCCH before the
timer expiration. The PDCCH monitoring configuration within the on-duration is based on the
search space set configuration same as in non-CDRX state. Although the DRX on-duration is
configurable, it would be beneficial if PDCCH monitoring periodicity within the on-duration could
be configured to provide more power saving.

Dynamic Search Space (SS) Set Group Switching

In Rel-16 and for NR-U a UE could be configured to switch between SS-set groups with sparse and
frequent PDCCH monitoring - with only one of them active at a certain time by explicit or implicit
SS set switching indication. Where the explicit switching functionality is supported using DCI 2-0.

Compared with the existing BWP switching, the SS set switching mechanism can achieve dynamic
PDCCH monitoring adaptation within the same BWP, reducing PDCCH monitoring and providing
power saving.

Figure 5- 20: Example for SS-set group switching between sparse and dense PDCCH monitoring

Rel-17 could consider the extensions to Rel-16 SS set group switching mechanism and the use of

78
 GTI 5G Device Power Consumption Whitepaper

explicit field in scheduling DCI formats (e.g. DCI 0-1/1-1) to indicate SS set group switching.

PDCCH Skipping

Figure 5- 21: PDCCH skipping mechanism

Another possible solution for PDCCH reduced monitoring is PDCCH skipping where the UE could
receive a DCI indication to skip PDCCH monitoring. If the DCI indicates PDCCH skipping, UE does
not monitor PDCCH for a number of slots after an application delay. The number of slots in which
UE does not monitor PDCCH is denoted as PDCCH skipping period, where the longer the PDCCH
skipping period is, the more power saving gain can be achieved. However, the latency is also
impacted by the PDCCH skipping period. Since PDCCH skipping can be combined with dynamic
BWP adaptation and DRX operation, the configuration of BWP or DRX should be considered when
determining the range of PDCCH skipping period.

Pre-indication for power saving adaptation

Mainly targeting the traffic types with frequent data arrival, with short inter-packet arrival time.
Conventionally the network tends to switch UE to power saving settings after receiving ACK
information from UE.

A faster adaptation allowing UE to stay longer in power saving period, will be to allow the
network to send the indication before HARQ-ACK information reception, with some condition
configuration like checking whether all HARQ processes are ACKed, in order to have minimum
impact on retransmission efficiency.

For example, network sends the adaptation indication in the scheduling DCI for the last TB of a
packet. If PDSCH is decoded successfully, UE switches to power saving setting; otherwise, UE shall
stay in data-efficiency setting for retransmission(s).

This principle of “pre-indication” is to balance between gNodeB preference of retransmission

with data-efficiency setting and UE preference of power saving after correct PDSCH
decoding.

79
 GTI 5G Device Power Consumption Whitepaper

Figure 5- 22: Power saving adaptation: (a) legacy and (b) R17 enhancement with pre-indication

Initial analysis for both VoIP (FR1, 1 CC) and Real-Time-Video (FR2, 4CC) showed considerable
gains as shown below, while minor latency degradation were seen specially in Video case

VoIP: FR1, 1CC Real-Time Video: FR2, 4CC

Legacy Pre-indication Legacy Pre-indication

Latency 0.00% 6.96% 0.00% 1.17%

Increment (+0.98 ms) (+0.11 ms)

Power 0.00% 9.49% 0.00% 38.09%

Saving Gain

Table 5- 30: Power saving gain and latency comparison between legacy and pre-indication
scheme

5.5.5 Relaxing UE Measurements for RLM/BFD

The NR UE shall monitor the downlink link quality based on the reference signal in the configured
RLM-RS resource(s) in order to detect the downlink radio link quality of the PCell and PSCell as
specified in TS 38.213. To get the qualities, the samples of the RS within a period of TEvaluate are
filtered and processed, then compared with two thresholds Qout and Qin and finally based on the
result of the comparisons of all RS for RLM, PHY layer may send L1 indication to higher layer to
indicate OOS or IS. PHY layer get the result of the comparisons once for every TIndication_interval.
Same process is used for BFD.

80
 GTI 5G Device Power Consumption Whitepaper

Figure 5- 23: The UE procedures for RLM

It can be observed that the power consumption of RLM/BFD is impacted mainly by two
parameters, TEvaluate and TIndication_interval. Currently, the values of the two parameters are
determined by the configuration of RS and DRX cycle applied.

So comparing R15 baseline to proposed relaxed measurements, R15 baseline averages the SINR
of 10 samples, where the time interval between each sample is 1 DRX cycle for FR1 and 8 DRX
cycle for FR2.

While relaxed method averages the SINR of 10 samples, where the time interval between each
sample is K*DRX cycle for FR1, K*8 DRX cycle for FR2, where K is a scaling factor.

Figure 5- 24: R15 Baseline vs Relaxed Measurements with K=2

Relaxing RLM/BFD measurement will obviously provide power saving gains, especially for the
cases when PDCCH WUS is configured, but these gains might come at the cost of performance
loss whenever the Radio link/beam quality is degrading quickly as UE might not realize it in time.
Some SINR and mobility criteria could be considered so that when UE’s channel condition
changes, the UE can have the SINR or time margin to fall back to the normal RLM measurement
frequency and capture the possible channel deterioration timely.

5.5.6 Sending Data in RRC inactive state

NR Release 15/16 does not allow for the transmission of user data in inactive state. As a
consequence, even for the transmission of very small amount of data /heartbeats the device has
to resume the connection, which has a negative impact on signaling overhead as well as device
energy consumption. Release 17 consider the possibility for transmission of small data payloads -
used only for control signaling - in inactive state including an extension of the payload size
supported for these. 2 Step-RACH and configured grant frameworks are main pillars for this
capability.
81
 GTI 5G Device Power Consumption Whitepaper

6 Requirement of 5G terminal power consumption

6.1 NSA
In a non-standalone deployment the UE is connected to a master node which is an eNB (EN-DC)
or an ng-eNB (NGEN-DC), and communicates with it using E-UTRA protocols [2]. When the UE is
in the connected state, NR connectivity is provided by a secondary node which is a gNB. Other
forms of dual connectivity exist, but are not considered here. The UE RRC state is derived from
the master node RRC state.

Dual connectivity inevitably increases power consumption compared to single connectivity, so for
the lowest UE power consumption in an NSA deployment the secondary node connection should
be activated only during periods of high data traffic.

6.1.1 RRC idle state and RRC inactive state

RRC idle state

The UE is in the idle state when there is no RRC connection established. In RRC_IDLE:

A UE specific DRX cycle may be configured by upper layers

UE monitors a paging channel for core network paging using 5G-S-TMSI or IMSI

UE performs neighbour cell measurements and cell reselection

UE acquires system information (SI)

The UE uses neighbour cell measurements to manage its mobility. If cell reselection to a different
tracking area is required it notifies the network using a RACH procedure. SI acquisition and cell
reselection are assumed to occur sufficiently infrequently that their impact on UE average power
consumption can be neglected.

The DRX cycle defines how often the UE needs to wake up to check for paging (320, 640, 1280 or
2560ms). During the wakeup it also measures the signal strength of the serving cell, and if this is
below the cell reselection criteria it initiates measurement of the neighbour cells. Since there is
an energy cost to waking and returning to sleep, these activities are usually performed in the
same wakeup as the paging occasion.

The wakeup duration is generally small as a proportion of the DRX cycle, and average power in
this state is dominated by the sleep state power and the energy cost of waking and returning to
sleep.

Measurements on neighbour cells can be scheduled in the same wakeup as the paging occasion
to reduce power consumption, but this may increase the wakeup duration if the symbols needed
82
 GTI 5G Device Power Consumption Whitepaper

for the measurement only occur infrequently, or if measurement of inter-frequency cells requires
RF retuning. For an NSA deployment, it is assumed that idle mode measurements will not include
the NR carrier because reselection to the NR cell is not possible. This means that the UE has no
knowledge of the secondary gNB before entering the connected state, and would need to
complete cell search and beam alignment procedures before it could access the secondary node.

All activities are performed in the master E-UTRA cell, with the exception of neighbour cell
measurements on other E-UTRA cells, so power consumption in this state is expected to be
similar to the equivalent power for an E-UTRA UE.

RRC inactive state

The RRC_INACTIVE state is not present in EN-DC. It only exists in NGEN-DC when the master node
is an ng-eNB.

The UE is in the inactive state when there is an RRC connection which has been suspended. In
inactive mode the UE performs all of the functions that are part of the idle state, but additionally
in RRC_INACTIVE:

A UE specific DRX cycle may be configured by the RRC layer

UE Stores the access stratum context

A RAN-based notification area is configured by the RRC layer

UE performs RAN-based notification area updates periodically

UE performs additional updates when moving outside the configured RNA

UE monitors a paging channel for RAN based paging using I-RNTI

A UE in the inactive state can move between nodes within its notification area without needing
to inform the network. The network is only notified when the UE moves outside its notification
area.

The RAN-based activities are not expected to add significantly to the wakeup energy compared
with idle mode, but in the inactive state a shorter DRX cycle may be configured to give a faster
response, and additional measurements would be performed (including the NR cell(s)). A shorter
DRX cycle would result in higher RRC_INACTIVE power consumption than for RRC_IDLE.

RAN-based paging initiates an RRC Connection Resume procedure which returns the UE to the
RRC_CONNECTED state. CN paging would return the UE to the idle state and inform the NAS that
paging had occurred.

The network has choices to make in how the RRC_INACTIVE state is managed. If a similar DRX
cycle is configured for both RRC_INACTIVE and RRC_IDLE, the difference in power consumption
between the two will be small, and keeping the UE in RRC_INACTIVE is beneficial in giving faster
transitions to the RRC_CONNECTED state.

If alternatively a faster DRX cycle is configured for RRC_INACTIVE to improve response times
83
 GTI 5G Device Power Consumption Whitepaper

there will be a significant increase in power consumption since the UE spends more time
monitoring the control channel. In this case the network should configure a timeout, such that if
the time spent continuously in RRC_INACTIVE without a transition to RRC_CONNECTED exceeds
the timeout, the UE will revert to the RRC_IDLE state

If these principles are followed, time spent in the RRC_INACTIVE state is not expected to make a
large contribution to UE average power consumption, but it offers the advantage of a quicker
transition to the connected state than is available in idle mode, and provides lower power
consumption than connected DRX.

6.1.2 RRC connected state

In the connected state the UE monitors the control channel for indications that user data or
signaling is present. In an NSA deployment this monitoring can take place on both RATs.
Additional measurements are performed to assess channel quality, and the results are reported
to the network periodically. In the connected state mobility is managed by the network based on
the measurements reported by the UE.

No data transmission without DRX or with DRX

If there is no data traffic and DRX is not enabled, the key to lower UE power consumption is to
reduce the power required for PDCCH reception and decoding. Monitoring only on the master
node would be the easiest way to achieve this, but quick activation and deactivation of the
secondary node would be needed to make good use of the NR data capability, and the additional
delays would impact on latency. Dual-RAT PDCCH monitoring would increase monitoring power
consumption significantly, with two independent monitoring paths active instead of one, so there
may be power saving benefits from cross-carrier scheduling.

The power increase from dual-RAT monitoring can be mitigated by using a low-duty cycle DRX
configuration, but in a dual-RAT scenario DRX operation has the potential to be more complicated
and less efficient, as there is no simple mechanism in the specifications for synchronising the DRX
cycles between the master and slave nodes. For this reason it is highly desirable that operators of
NSA deployments ensure DRX cycle boundaries are aligned between the E-UTRA and NR cells to
reduce the number of separate wakeup occasions.

An important power saving provision currently under discussion in 3GPP is the concept of a
dormant state for the SCell. In the dormant state measurements would take place on the SCell for
CSI, RRM and beam management, but there would be no PDCCH monitoring on the cell until it
was activated. This would mean that the SCell could be dormant during no-data periods, but
activated quickly when high data bandwidth is needed.

Uplink data transmission in NR side only (high data rate and low data rate)
When there is uplink data to transmit, the node which has the lowest path loss should ideally be

84
 GTI 5G Device Power Consumption Whitepaper

selected, since this will require the lowest transmitted power per subcarrier, and give the lowest
power consumption at the UE. Uplink allocations can be as little as 1 symbol, but at high data
rates it is likely that multiple symbols will be required for each transmission. In most cases
transmission over NR would use less energy in total, as the duration of the transmission would be
shorter than in E-UTRA.

If the NR SCell is dormant, there will be signaling costs for activation and deactivation to allow
transmission to take place, so before activating the secondary node there should be sufficient
data in the uplink, downlink or both to justify these costs.

At low data rates, if path losses are similar, there is little reason to prefer one over the other for
reasons of power consumption, and if the SCell is not already active, transmission over the
E-UTRA master node would be preferable.

Uplink data transmission in NR side and LTE side (high data rate and low data rate)
In NSA configurations there is generally a disparity in bandwidth between the E-UTRA node and
the NR node such that the NR node has significantly higher capacity than the E-UTRA node in the
uplink. There is also an upper limit on the total power that a UE can transmit, so increasing the
number of uplink carriers limits the power that each can transmit.

This means the uplink coverage is generally better for one carrier transmitted at full power than
for two carriers transmitted at half power. If the data can be transmitted over a single node it is
also more power efficient for the UE to process one grant and one acknowledgement rather than
two, and it means that the wakeup period is only extended in one DRX cycle rather than two.

At high data rates it is obvious that the single node should be the NR node, but at low data rates
the signaling cost of activation and deactivation needs to be taken into account.

There will clearly be occasions where control is transmitted to the master node and data to the
secondary node, and also high data rate transmission scenarios where the network dedicates the
resources of both nodes to a single user to provide faster throughput, but the asymmetry
between the two carriers means that energy consumption per bit at the UE is likely to be higher
in such cases. Carrier aggregation works best if the carriers are of similar bandwidth.

Downlink data transmission in NR side only (high data rate and low data rate)
UE downlink efficiency (energy consumed per bit received) improves as the bandwidth and
MIMO configuration increase, but so does the total power consumption. Unlike the uplink, data is
received across the entire Rx bandwidth rather than just the resource block allocation, so data
transfer efficiency falls when the RB allocation is a small fraction of the available bandwidth due
to the energy spent processing the unallocated RBs. It is therefore desirable to adapt the Rx
bandwidth to the volume of data traffic.
85
 GTI 5G Device Power Consumption Whitepaper

In an NSA deployment, there is already a low bandwidth E-UTRA channel and a high bandwidth
NR channel, so UE power consumption benefits if the network routes small transport blocks via
the master node and larger transport blocks via the secondary node.

BWP for the NR SCell can be configured to adapt the NR Rx bandwidth to the data size, but if
BWP selection is done via the DCI to improve speed it is also necessary to configure cross-slot
scheduling. For NSA, if the SCell is activated only during periods of high data traffic the
bandwidth occupancy should be high, and the savings from this type of adaptation may therefore
be small as a proportion of total UE energy usage. The effect would be greater in a standalone
deployment.

If high data rates (in either the downlink or the uplink, or both) are sustained for prolonged
periods, the UE may experience overheating. If this occurs, the UE should request a temporary
capability reduction from the network. This will allow it to disable some resources, or operate
them at lower power to maintain a safe temperature. Details of signaling for the request are still
under discussion.

Downlink data transmission in NR side and LTE side (high data rate and low data
rate)
Because there is a power penalty for dual connectivity, UE downlink power consumption in NSA
deployments is also minimised by activating the NR connection only during periods of high
bandwidth traffic and using the E-UTRA connection when data rates are lower.

Some control signaling has to take place via the master node, but simultaneous routing of data
traffic to both nodes should be avoided where possible, as this will extend the active period of
DRX in both of the DRX cycles, whether or not they are synchronised. This will increase the total
wakeup time, and therefore the power consumption.

Voice
To deliver good quality bidirectional voice traffic in real time, as required for voice telephony, the
UE needs to send and receive speech packets each representing 20ms of audio with an
end-to-end delay of less than 100ms. To achieve this in a power efficient way, the UE needs a DRX
cycle of 20 or 40ms, with short OnDuration and inactivity timers. The data rate is low (kilobits per
second), and can easily be carried by the master or the slave node.

Semi-persistent scheduling of resources can help to reduce the need for PDCCH monitoring – for
voice traffic the uplink and downlink speech packets and their respective ACK/NACK
transmissions should be scheduled close together in time to minimise the wakeup duration

86
 GTI 5G Device Power Consumption Whitepaper

If voice traffic is the only UE activity, it is difficult in an NSA deployment to justify activation of the
secondary node, as this will increase power consumption on the UE. It may be necessary if there
is congestion at the E-UTRA node, but if higher data rate traffic is routed preferentially over NR
this should not be a frequent occurrence.

6.2 SA
In a standalone deployment the UE is connected to a master node which is a gnB, and
communicates with it using NR protocols [2]. When the UE is in the connected state, there may
additionally be one or more secondary nodes which can be gNB or en-gNB. The UE RRC state is
derived from the master node RRC state.

An SA deployment provides a high data bandwidth connection without carrier aggregation or

dual connectivity, although additional SCells can be activated to increase peak throughput.

For reasons already discussed, this should only be done when data requirements are particularly
intense.

A UE with low data activity connected to an SA network is operating in a high bandwidth

environment, and potentially has to discard a high proportion of the downlink data that it
processes. This could be bad for power consumption, but NR has introduced a number of new
features which can be configured by the network to improve UE power efficiency when data rates
are lower.

These features can also be applied in non-standalone networks, but are discussed in more detail
here because the UE in an NSA network is assumed to be connected to the NR secondary node
only when high bandwidth is needed.

6.2.1 RRC idle state and RRC inactive state

RRC idle state

In an SA deployment the idle state functions of the UE are very similar to those listed in 6.1.1.1,
but paging is only by 5G-S-TMSI, and the UE may send SI requests to the network. Neither of
these differences has a significant effect on power consumption.

However, in an SA configuration measurements for cell reselection can include potential

secondary nodes, and if this is the case then transitions to the RRC_CONNECTED state can be
much faster than for the NSA case.

It should be noted that a paging occasion in an SA master node may have a longer duration than
for an NSA master, because in an NR cell the paging duration corresponds to the beam sweeping
period [4]. The selection of beams for paging and measurement is up to the UE implementation,
87
 GTI 5G Device Power Consumption Whitepaper

but it is reasonable to suppose that if the configuration of measurement resources means a

longer wakeup for the UE there may be some increase in the RRC_IDLE average power.

Initial modelling suggests that the average power in RRC_IDLE is likely to be dominated by the
sleep power and the energy cost of the sleep-wakeup transitions, so for similar measurement
configurations any net increase compared to NSA is likely to be small.

RRC inactive state

The RRC_INACTIVE state is a new feature in NR, and is not present in E_UTRA. The discussion in
section 6.1.1.2 is therefore equally applicable to SA deployments.

6.2.2 RRC connected state

No data transmission without DRX or with DRX

In an SA deployment there are opportunities for PDCCH-only power reduction that are only
available when the master node is NR (similar techniques can be employed in the secondary
node in NSA, but are less effective if the secondary node is dormant most of the time).

One important technique for reducing power in no-data scenarios (new in NR) is cross-slot
scheduling. In E-UTRA, the control channel indicates the presence or absence of data in the same
slot in which it is received, which means that the receiver must remain active while the control
channel is being decoded in case data is present. In NR, the control channel in slot N indicates the
presence of data in slot N+K0, where K0 is a configuration parameter.

K0 is 0 by default, but if the network configures a larger value the UE can disable its receiver as
soon as PDCCH reception is complete, thereby reducing power consumption. PDSCH reception
then only takes place in slots where there is data present for the UE to decode.

The NR power model recently agreed in 3GPP TR38.840 [13] suggests that PDCCH-only power will
be approximately 1/3 of the power for peak downlink throughput with K0=0, and that cross-slot
scheduling can reduce this by a further 30%.

Reducing the frequency of PDCCH monitoring (while observing latency requirements) would also
reduce UE power consumption. This is currently under discussion for Rel-16, and would allow
multiple slots of PDSCH to be scheduled from a single PDCCH slot, increasing resource utilisation.

Cross-slot scheduling and PDCCH monitoring reduction both increase latency in data traffic, but
trading latency for power reduction will be acceptable for most users. The default configuration
should favour low power consumption, with applications that require low latency from the
88
 GTI 5G Device Power Consumption Whitepaper

network requesting it explicitly.

Bandwidth part adaptation is another power saving feature that is new to NR. By reducing the
bandwidth that is captured and processed, the baseband throughput can be significantly reduced,
giving further power reductions of up to 60% [3] Reducing the number of receive paths provides
additional opportunities for power saving.

The reduction in throughput with BWP adaptation and Rx path reduction applies to the data
channel as well as the control channel – switching to a higher bandwidth may be needed when
there is more data present, and rapid switching of BWP via the DCI is desirable to avoid
unnecessary increases in latency.

Uplink data transmission (high data rate and low data rate)
In the uplink, Tx power control determines the transmitted power in each subcarrier, and the RB
allocation is matched to the size of the transport block. The network will normally configure the
highest order modulation scheme that the channel can support, so there are few opportunities
for reducing UE power consumption in uplink transmissions.

Reducing the number of transmitting antennas from 2 to 1 can save UE power in some scenarios,
but this decision depends on channel conditions, and should take into account the throughput
benefits of diversity gain or additional MIMO layers.

In range limited scenarios there is an upper limit on the size of RB allocation that can be
supported, due to the maximum power that the UE can transmit. Reducing the uplink duty cycle
can allow the UE to increase its transmit power provided that the average power remains within
the limit, and if necessary, repetition can be used to increase the range of coverage.

Downlink data transmission (high data rate and low data rate)
In the downlink, for a given subcarrier spacing, processing requirements increase with bandwidth
and the number of receive paths. A larger subcarrier spacing reduces the processing requirement,
but shortens the TTI so that processing must be completed more quickly.

Higher bandwidth can deliver data faster with lower energy per bit to more users, but from a UE
perspective decoding of resource blocks that are not part of its allocation constitute wasted
energy. Energy consumption is reduced if data transmission occupies many resource blocks and
few TTIs compared to occupying few resource blocks in many TTIs in a narrow bandwidth. This
principle also applies when transport block sizes are small, but if the processed bandwidth is
substantially larger than the UE throughput, energy is wasted.

89
 GTI 5G Device Power Consumption Whitepaper

These fundamental principles become more and more important as the bandwidth of the
channel increases. The ultimate limit is set by thermal noise, and as bandwidth grows so does the
noise within the channel. Path delay spread leads to an increase in subcarrier spacing, which
increases the noise per subcarrier. In order to design a high bandwidth network that is power
efficient, it is essential that no node in the network has to process significant quantities of
irrelevant data.

This implies that capture of the data channel should be avoided unless the control channel
indicates that relevant data is physically present. This in turn implies that there should be a delay
between reception of the control channel and reception of the data channel to which it refers.
This delay should be sufficient for the UE to configure the bandwidth of data reception to match
the transport block size.

The success of NR (and its successors in mobile communication technology) will depend on how
well these principles are embodied in the network structure. Cross–slot scheduling, bandwidth
adaptation, transceiver path reduction, and DRX cycles that are adapted to traffic patterns and
latency requirements will be key elements in ensuring that user devices in a mobile
communications network operate in the most power efficient manner.

UE overheating remains a possibility if the highest data rates are maintained for long periods, and
signaling will be necessary to reduce the UE throughput if this occurs

Voice
Voice telephony requires only a fraction of the bandwidth offered by NR, so in an SA deployment
a narrow bandwidth part is a basic requirement for low UE power consumption. Reducing the
number of antennas for uplink and downlink traffic would allow further savings. Data traffic in
both directions is predictable, with speech packets generated every 20ms, so semipersistent
scheduling reduces the need for control signaling. A DRX cycle of 20 or 40 ms matches the data
traffic pattern and meets the end-to-end latency requirement.

For best power consumption the on duration in the DRX cycle should be short – combining the
uplink speech packet with the downlink acknowledgement (or vice-versa) can reduce the number
of data transactions per cycle to 3, and mini-slot would allow these to be scheduled in close
proximity to each other, but provision must also be made for background measurements and
possible retransmissions. The distribution of measurement resources may limit how short the
wakeup duration can be - it is not possible to schedule every user adjacent to an SSB
transmission.

6.3 Estimating power consumption in NR UEs

The NR power model agreed in 3GPP [3] describes how UE power consumption scales in relative
terms for different configurations, but because it represents a consensus between several UE
suppliers there is no absolute calibration of the power unit in the model. The model does not
90
 GTI 5G Device Power Consumption Whitepaper

include E-UTRA operation, so it is only applicable to the standalone configuration.

CMCC have produced a requirements specification for the sub-6GHz NR UE, which includes
targets for UE power consumption in different configurations. UEs that are approved for use on
its network are expected to meet target performance, so these targets can be viewed as
providing a pessimistic estimate of UE power consumption. Table 6- 1 below compares some key
model parameters to the power consumption requirement of the nearest related configuration in
the CMCC specification

Table 6- 1 Comparison between the 3GPP model and CMCC requirement

3GPP model [3] CMCC requirement

Slot Configuration Model power Power (mW) Configuration (TDD 7D1S2U)

PDCCH+PDSCH 300 2400 SA 100MHz 1CC, 1Gbps DL, 0dBm Tx

UL transmission 0dBm 250 1880 SA 100MHz 1CC, 100Mbps UL, 0dBm Tx

UL transmission 750 3600 SA 100MHz 1CC, 100Mbps UL,23dBm Tx

23dBm

It is clear from the table that the power scaling is different in the comparison cases, but the
comparison is not direct, because data transfer in either the uplink or the downlink requires a
mixture of transmissions in both directions. Noting that the CMCC configurations allocate only
20% of slots to uplink transmission, reasonable alignment between the two sets of data can be
obtained if we assume a conversion factor of 1 model unit approximately equal to 8mW.

Based on the CMCC specification, we can expect a UE in an NSA deployment to require 15-25%
more power when active and transferring data in the NR cell only. If both RATs are used for data
transfer the power penalty would be higher, but so would the data rate.

At maximum throughput and maximum transmit power, the allowable power consumption is at a
level where UE overheating becomes a possibility, and for the HPUE case even higher
instantaneous power levels are possible, although these cannot be sustained on a continuous
basis. For most use cases these high throughput conditions only occur in short bursts, but if
substantial temperature rises occur, UE assistance signaling may be needed to request a reduced
power configuration.

91
 GTI 5G Device Power Consumption Whitepaper

However, these are extreme cases – more typically, UEs will spend a high proportion of their time
in the idle or inactive state, where power consumption is much lower, and even in the connected
state a UE will spend much of its time monitoring the control channel waiting for data to arrive.
In the 3GPP model this PDCCH-only state (operating continuously at maximum bandwidth)
consumes only 1/3 of the power required at peak throughput.

PDCCH-only power (estimated at 800mW using the previously derived conversion factor) can be
reduced even further by network configuration. Based on 3GPP model assumptions, reducing the
control channel monitoring bandwidth from 100MHz to 20MHz can deliver a 60% reduction in
monitoring power, with a further 30% saving (combined saving of 72%) if cross-slot scheduling is
enabled. In delay-tolerant applications DRX can provide additional power reductions by reducing
the duty cycle for monitoring activities, bringing average power consumption down to very low
levels.

This brief overview suggests that instantaneous power consumption in an NR UE can vary over a
wide range in normal use, but if the network configuration allows the UE to perform
demand-based switching between efficient high bandwidth data transfer, reduced bandwidth low
power control channel monitoring and periods of inactivity then NR can deliver a user experience
that combines rapid responsivity and throughput with good battery life.

7 Service

7.1 Service type and Parameter configuration

 Note：Model analysis, parameter configuration and power consumption effect

To evaluate the metric of different power saving schemes, diverse traffic with different
characteristic need to be considered. According to the RAN1#94bis agreement [3], FTP model 3
should be included in the evaluation for at least FTP application. Other bursty traffic arrival
models can also be considered. In addition, other applications including web-browsing, video
streaming, instant messaging, VoIP, gaming and background app sync can be considered for traffic
modelling for power saving proposal evaluation.

7.1.1 FTP model 3

FTP models already defined in in 3GPP TR 36.814 can be at least a starting point for traffic
modelling. It is characterised as large packets size, e.g., Mbytes and less frequent packet arrival,
e.g., hundreds of ms or several seconds. FTP model 3 (use 0.1 Mbytes packet size, mean
inter-arrival time 200msec) is used for the purpose of basic calibration of traffic modeling.
92
 GTI 5G Device Power Consumption Whitepaper

7.1.2 Game

Gaming is characterised as smaller packets size, e.g., hundreds of bytes, frequent and random
packet arrival, e.g., tens of ms. The relevant parameters of the gaming traffic are as follows.

93
 GTI 5G Device Power Consumption Whitepaper

Table 7- 1 Parameters for gaming

Parameter Statistical Characterization

Packet For packet arrival of <60ms, fixed probability of 0.4%, i.e., packet arrival for any
arrival value from 0 to 59ms has fixed 0.4% probability;

For packet arrival of >= 60ms, Largest Extreme Value Distribution (also known as
Fisher-Tippett distribution),

, x is the packet arrival (in ms) and fx is the probability of

x.

a= 66 and b= 3.

Without loss of generality, a= [30-80] and b= [2-5] can be considered.

Values for Fisher-Tippett distribution can be generated by the following procedure:

x = a – b ln (− ln y ), where y is drawn from a uniform distribution in the range

[0,1]

Packet size Largest Extreme Value Distribution (also known as Fisher-Tippett distribution):

, x is the packet size (in Bytes) and fx is the probability of x.

a= 220 and b= 25.

Without loss of generality, a= [200-300] and b=[20-30]

Values for Fisher-Tippett distribution can be generated by the following procedure:

x = a – b ln (− ln y ), where y is drawn from a uniform distribution in the range

[0,1]

7.1.3 Video

The typical video streaming traffic (i.e. YouTube) modelling is described with two phases, an
initial burst phase followed by a throttling phase (Figure 7- 1). In the initial phase, the video
streaming progressive download commences by transferring an initial burst of data with several
seconds. And then in the throttling phase, the video server throttles down the traffic generation
rate during which the traffic pattern alternates between the reception of data chunks and short
periods without packets. It is illustrated in the following figure.

94
 GTI 5G Device Power Consumption Whitepaper

Figure 7- 1 Initial burst phase and throttling phase in video streaming traffic modelling

Since the initial phase of video streaming more looks like the full buffer traffic, it is sensible to
only model the throttling phase for power saving traffic.

7.1.4 Voice

Data generated by voice calls are highly deterministic, as the corresponding codecs generate data
at known intervals and of a known size. The relevant parameters of the VoIP traffic are as follows
from R1-070674.

Table 7- 2 Parameters for VoIP

Parameter Characterization

Codec RTP AMR 12.2, Source rate 12.2 kbps

Encoder frame length 20 ms

Voice activity factor (VAF) 50% (c=0.01, d=0.99)

SID payload Modelled

15 bytes (5Bytes + header)

SID packet every 160ms during silence

Total voice payload on air interface 40bytes (AMR 12.2)

7.1.5 Other Applications

Additionally, some other representative applications can be selected. The following applications

95
 GTI 5G Device Power Consumption Whitepaper

have been mentioned such as web-browsing, instant messaging, and background app sync.

For web-browsing application, it is usually modelled to request several web pages with different
inter-arrival time for user reading the web page. A web page (web session) consists of a main
object following some other inline objects. After the downloading process of one web session, a
session inter-arrival period takes place. Web-browsing can use HTTP to model.

Table 7- 3 Web Browsing traffic modelling parameters

Model Parameters Description Distribution

Session inter-arrival time Gaps between sessions (User reading time) Weibull

Number of Objects per session Number of object per session Gamma

Object inter-arrival time Delay between the arrival of two objects Gamma

Object size Size of each object Weibull

For instant messaging, the traffic tends to be sporadic. For simplicity, the inter-arrival time also
can be modelled as fixed value. FTP Model 3 with some modifications can be used to model
instant messaging applications.

For background app sync application, for power consumption evaluation purpose, it can be
assumed that idle mode operations (inclusive of page detection, RRM, deep sleep and transition
overhead) contributes to X% of the use case power. The remaining portion is contributed by
intermittent RRC connections due to background activities (FFS: value of X). In summary, the
traffic characteristics are captured in the following table.

Table 7- 4 Proposed traffic models and related characteristics

Application Characteristics

FTP model 3 Large packets size, e.g., Mbytes

Less frequent packet arrival (exponential distribution), e.g., hundreds of ms or

several seconds

Gaming Smaller packets size, e.g., hundreds of bytes

Frequent and random packet arrival, e.g., tens of ms

Video Initial burst size, throttled data rate and chunk size.
streaming

VoIP Small packets size, e.g., tens of bytes

Shorter and fixed packet arrival, e.g., tens of ms

Asymmetric in UL and DL traffic

96
 GTI 5G Device Power Consumption Whitepaper

Web-browsing Session inter-arrival time, Number of objects per session, Object inter-arrival
time, Object size

Instant The size of the instant message is determined by the Pareto distribution.
messaging
The inter-arrival time between two messages is modelled as the Lognormal
distribution

Background RRC connections every few minutes. May be modelled as a fixed power cost
app sync relative to I-DRX

According to the RAN1#94bis agreement, percentage power consumption reduction from the
baseline scheme will be used to express the power saving gain. And Latency of packet or
scheduling delay, user throughput should also be reported as the result of the evaluation, in
addition to power saving gain. On UE side, it means support of good power saving capabilities
while at the same also provide for good user experience in terms of delay. On network side it
includes support of flexible scheduling possibilities of UE’s in Active mode while being able to
provide good power saving possibilities to UE’s in Active mode.

DRX has been used for UE power saving since Rel-99 UMTS and continued in LTE and NR by
allowing UE to get into deep/light sleep state during the DRX OFF period. UE would autonomous
wake up before DRX ON cycle in preparation for the signal processing. However, most of time, UE
wakes up at the DRX ON period and gets no grant from PDCCH and no data from PDSCH
especially for the sporadic services like instant messaging traffic. PDCCH-only state is the highest
contributor.

Possible solution is the introduction of WUS (Wake up signal) which allows UE to skip the
upcoming on duration period and go back to sleep. Go-to-sleep signaling can be used to stop
drx-onDurationTimer or drx-InactivityTimer for faster skipping of monitoring.

Another solution to improve this condition is to enhance the current DRX, i.e., more dynamic and
flexible C-DRX configuration can be supported. gNB can make decisions and change the C-DRX
configuration more dynamically by some Layer1 signaling. In addition, UE can give a
recommendation of the DRX configuration according to its traffic characteristics and power
saving requirements.

97
 GTI 5G Device Power Consumption Whitepaper

7.2 User model

Based on different user model to analysis the device last time

7.3 5G industry application power analysis

7.3.1 7.3.1 Laptop

- Need define power consumption target per running status :

o Power off
o Standby
o Idle
o Connected standby (w/ WiFi or WWAN)
o Local Video playback (720p, 1080p, 2k, 4k)
o Video streaming (both WiFi and WWAN).
o 3DMark – full performance Power.
- Need define power target per system configure :
o Storage type and size (SSD or HDD, 128GB to 1TB, etc.
o Memory type and size (DDR3, DDR4, LPDDR3/4/5, 2GB – 32GB, etc.).
o Discrete display card or integrated display card
o CPU model and power rate
o Other factors ……
- Need have enough range to accommodate different type of laptop with different
configurations.
- Typical power consumption requirement (Subject to PC configurations – Display, CPU,
memory, etc.) :
o Local video playback (1080p): 8 hours.
o Web browsing: 5 hours.
o Video streaming (WIFI / 5G-WWAN): 5 hours / TBD
o MM(Mobile Mark) battery life : > 5 hours

8 Power consumption test

8.1 Test instrumentation

The specification relates to test instruments including,

98
 GTI 5G Device Power Consumption Whitepaper

1. Common Test System Architecture

Figure 8- 1 Common Test System Architecture

99
 GTI 5G Device Power Consumption Whitepaper

2. System Simulator including Application Server and Base Station Emulator. Following function
should be considered.

 Frequency bands and System bandwidth:

Table 8- 1 5G NR Band list

Max BW for Max BW for Max BW for

RAT Band
15kHz 30kHz 60kHz

Band n77(3.3GHz ~ 4.2 GHz) 50MHz 100MHz 100MHz

Band n78(3.3GHz ~ 3.8 GHz) 50MHz 100MHz 100MHz

Band n79(4.4GHz ~ 5GHz) 50MHz 100MHz 100MHz

NR Band n1 50MHz 20MHz 20MHz

Band n3 50MHz 30MHz 30MHz

Band n8 50MHz 20MHz NA

Band n41 50MHz 100MHz 100MHz

 MIMO configurations:

o UL: 2 layers required, 4 layers recommended

o DL: 4 layers required, 8 layers recommended

 NSA(Option 3X) and SA(Option 2).

 Full stack including PHY/L2/RRC/NAS compatible with 3GPP Release 15.

 UDP/IP data transmission service.

 Voice over 5G NR(TBD).

3. DC Power Analyzer includes following functions:

 High accuracy (detailed requirement is TBD).

 Data logger function.

 Multiple outputs.

4. Channel Emulator( detailed requirement is TBD)

100
 GTI 5G Device Power Consumption Whitepaper

8.2 Test method

1. Test Environment: Refer to the Test System Architecture in 8-1, the test environment

would be simulated by based station emulator and Channel Emulator. Following

parameter and scenarios should be considered:

 Test Frequency: should refer to 3GPP Test Specs, such as TS38.521-1.

 DL power level: it emulates the network coverage situation. A user on the edge of

a cell will inevitably experience higher power consumption than one who is near

a gNB.

 UL power level: The uplink power will vary to simulate the distance between the

UE and the gNB. And uplink power is dominated by the UE transceiver

contribution.

 Bandwidth part adaptation: It can be used to reduce UE power consumption

when the full data bandwidth of NR is not required.

 Setting for Cross-slot scheduling and DRX: Cross-slot scheduling can be used to

eliminate unnecessary capture of the data channel in slots where no data is

present for the UE. While DRX provides a means of trading latency for further

power reductions. Used in combination, these two features can reduce the

energy cost of control channel monitoring In many common low data rate use

cases control channel monitoring can consume more power than data transfer,

so this can lead to significant power reductions.

 Multi-carriers and MIMO setting: The UE transceiver power is a function of the

number of active carriers and the number of MIMO layers supported on each.

 User data throughput: The simple power model presented suggests that NR UE

power consumption at peak throughput may be higher than a LTE UE offering

lower throughput, but the energy per bit at maximum performance will be

significantly better.
101
 GTI 5G Device Power Consumption Whitepaper

 User patterns: Actually, most of real use case for NR mobile phone would be lower

throughputs. So, the key to competitive NR UE performance will be how well the

power consumption scales with data rate at lower throughputs. This will require

cooperation between networks and UE to match UE resource allocation to data

traffic patterns in ways that allow the UE to reduce the data bandwidth that it

has to process, and the time for which it is active, so that it can optimize its

power consumption.

 Channel condition: If the increase in throughput is proportionately greater than

the increase in battery power then there will be a net energy saving, but this

tradeoff will depend on channel conditions.

 NSA/SA network configuration: In principle a standalone UE should consume less

power than a non-standalone UE, requiring fewer receive and transmit

resources for equivalent bandwidth. However, in some case, Synchronizing DRX

wakeups between LTE and NR will make NSA operation more power efficient.

 Voice service on NR: As one of low data rate services, the key to low power

consumption for NR voice will be to keep the data that the UE has to process for

a voice call to a minimum.

2. Test Procedure: Refer to the Test System Architecture in 8-1, the general test procedure

would be followings:

 Connect RF ports of DUT and based station emulator.

 Connect Power ports of DUT and power analyzer.

 Initial based station emulator and channel emulator to pre-defined parameter and

settings.

 DUT power on and perform pre-defined scenarios.

102
 GTI 5G Device Power Consumption Whitepaper

 Record test results by power analyzer.

3. Real network real battery DUT power consumption test

To maximally simulate the real working situation of DUT in real network, and evaluate
the DUT efficient work time, it is important to test power consumption in real network
with standalone battery.

DUT working with standalone battery without DC power supply.

Current/Voltage probe, for battery-powered devices, probing can be done at the battery
connectors.

Controller PC could control and read/record the current/voltage value.

The beginning condition of battery should be defined as precondition like 100%, 50%,
20%.

The test location should include near point, middle point and far point from gNB

The test situation should include data throughput, voice and so on.

4. Real battery lab power consumption test

To make it more easily and test the real battery in lab, the following test system could be
used to test real battery power consumption in lab.

103
 GTI 5G Device Power Consumption Whitepaper

 Network simulator parameter configuration refer to the 8-2-1, same with DC

power supply test methods.

 DUT working with standalone battery without DC power supply.

 Current/Voltage probe, for battery-powered devices, probing can be done at the

battery connectors.

 Controller PC could control and read/record the current/voltage value.

 The beginning condition of battery should be defined as preconditions like 100%,

50%, 20%.

 The test situation should include data throughput, voice and so on.

9 References
[1] “GTI Sub-6GHz 5G Device White Paper V2.0” GTI Group, February 2018

http://gtigroup.org/Resources/rep/2018-02-22/11888.html

[2] 3GPP TS37.340 V15.3.0 “Evolved Universal Terrestrial Radio Access (E-UTRA) and NR;
Multi-connectivity; Stage 2

[3] Chairman’s notes section 7.2.9, RAN WG1 Meeting #94bis, Chengdu, China

[4] 3GPP TS38.304 “User Equipment (UE) procedures in Idle mode and RRC Inactive state”

[5] R1-1810447 “Evaluation Methodology for NR UE Power Saving Study”, MediaTek, RAN WG1
Meeting #94bis, Chengdu, China
104
 GTI 5G Device Power Consumption Whitepaper

[6] R1-1810794 “Evaluation methodology for power saving techniques”, Intel, RAN WG1 Meeting
#94bis, Chengdu, China

[7] R1-1811281 UE Power Saving Evaluation Methodology Qualcomm Incorporated

[8] R1-070674, “LTE physical layer framework for performance verification”, Orange, China
Mobile, KPN, NTT DoCoMo

[9] R1-1812048 Summary on UE Power Saving evaluation methodology Qualcomm

[10]Chairman’s notes section 7.2.9, RAN WG1 Meeting #95, Spokane, USA

[11]3GPP TS38.331 “Radio Resource Control (RRC) protocol specification”

[12]GTI 5G Sub-6GHz UE Feature List，GTI #23 workshop

[13]TR38.840, “NR; Study on UE Power saving (Release 16)

[14]R1-1901803 Evaluation Results of NR Power Saving Designs, MediaTek RAN WG1 Meeting #96,
Athens, Greece

105