Handling of Intra-UE Uplink Overlapping Grants for Heterogeneous Traffic in 5G Wireless Systems

(1)

INOM TEKNIKOMRÅDET EXAMENSARBETE

TEKNIK OCH LÄRANDE OCH HUVUDOMRÅDET

INFORMATIONS- OCH KOMMUNIKATIONSTEKNIK, AVANCERAD NIVÅ, 30 HP

,

STOCKHOLM SVERIGE 2020

Handling of Intra-UE Uplink

Overlapping Grants for

Heterogeneous Traffic in 5G

Wireless Systems

JIACHENG ZHENG

KTH

(2)

iii

Abstract

5G system is considered as a wireless communication platform that could sup-port a variety of scenarios including industrial Internet of Things (IoT). Sev-eral challenges are to be addressed in order for 5G New Radio (NR) to enable industrial automation, such as efficient multiplexing among heterogeneous in-dustrial traffic types. The heterogeneity of the inin-dustrial traffic is due to the simultaneous existence of several critical and non-critical traffic types even within a single UE, i.e., 1) Periodic deterministic, 2) Aperiodic deterministic, and 3) Non-deterministic. There have been several studies on achieving criti-cal traffic requirements. However, this study focuses on maintaining QoS for critical traffic while enhancing efficiency. The method chosen to achieve such a target in this study is for gNB to send overlapping grants with different ro-bustness. Furthermore, we focus on handling overlapping grants in the context of the Intra-UE uplink scenario, from the media access control (MAC) layer perspective. The main advantage of our solutions is that it requires very low complexity (comparing to other solutions, e.g., puncturing, spectrum sharing) resulting in minimal changes to existing standardization specifications. Using a Java-based system-level simulator, our results show that the proposed solu-tion could improve UE’s throughput by up to 25.9%, under specific simulasolu-tion settings, while maintaining latency requirements.

(3)

iv

Sammanfattning

5G-system betraktas som en trådlös kommunikationsplattform som kan stödja olika scenarier inklusive industriell IoT. Flera utmaningar måste hanteras för att 5G NR ska möjliggöra industriell automatisering, till exempel effektiv mul-tiplexering bland heterogena industritrafiktyper. Den industriella trafikens he-terogenitet beror på den samtidiga existensen av flera kritiska och icke-kritiska trafiktyper även inom en enda UE, dvs 1) Periodisk deterministisk, 2) Aperio-disk deterministisk, och 3) Icke-deterministisk. Det har gjorts flera studier för att uppnå kritiska trafikkrav. Emellertid fokuserar denna studie på att upprätt-hålla QoS för kritisk trafik och samtidigt öka effektiviteten. Metoden som valts för att uppnå ett sådant mål i denna studie är för gNB att skicka överlappande bidrag med olika robusthet. Dessutom fokuserar vi på att hantera överlappande bidrag inom ramen för Intra-UE-upplänkscenariot, från MAC-lagers perspek-tiv. Huvudfördelen med våra lösningar är att det kräver mycket låg komplexi-tet (jämförelse med andra lösningar, t.ex. punktering, spektrumdelning) vilket resulterar i minimala ändringar av befintliga standardiseringsspecifikationer. Med hjälp av en Java-baserad simulator på systemnivå visar våra resultat att den föreslagna lösningen kan förbättra UE: s kapacitet med upp till 25,9% un-der specifika simuleringsinställningar, samtidigt som latenskraven bibehålls.

(4)

List of Tables

2.1 Subcarrier Spacing Supported by NR . . . 6

4.1 System Setup . . . 14 4.2 Propagation Model . . . 15 4.3 System Parameters . . . 16 4.4 Scenario 1 . . . 17 4.5 Scenario 2 . . . 17 vii

(7)

List of Figures

2.1 Grant-based uplink accessing . . . 5

2.2 Grant-free uplink accessing . . . 6

3.1 Benchmark 2 . . . 9

3.2 Proposal . . . 9

4.1 Seven Hexagonal Cells . . . 13

4.2 Definition for some Symbols . . . 14

5.1 Latency Percentile at 3ms . . . 19

5.2 Average Throughput . . . 20

5.3 Percentage of Users Fulfilling Requirement . . . 20

5.4 Percentage of Users Fulfilling QoS Requirement with Variable cmtc Request Intensity, under Different cmtc Data Unit Size . 22 5.5 Average Throughput vs Critical Traffic Intensity, under Differ-ent cmtc Data Unit Size . . . 22

5.6 Percentage of Users Fulfilling QoS Requirement with Variable cmtc Request Intensity, under Different CG Periodicities . . . 24

5.7 Average Throughput vs Critical Traffic Intensity, under Differ-ent CG Periodicity . . . 24

5.8 Percentage of Users Fulfilling QoS Requirement with Variable eMBB Data Unit Size, under Different CG Periodicities . . . . 25

5.9 Average Throughput vs eMBB Data Unit Size, under Different CG Periodicities . . . 26

(8)

Acronyms

BLER Block Error Rate 15

CDF cumulative distribution function 11

CG configured grant 2, 5, 8, 9, 10, 11, 12, 15, 16, 18, 19, 21, 23, 25, 26, 27 CQI channel quality indicator 7

DG dynamic grant 2, 3, 8,

9, 15, 18, 19, 23, 26 eMBB enhanced Mobile Broadband vi, 1, 2,

7, 11, 12, 17, 21, 23, 25, 26 FDD frequency division duplex 16 FTP File Transfer Protocol 15, 17 HARQ hybrid Automatic Repeat Request 5

(9)

x Acronyms

IoT Internet of Things iii, iv, 1, 2, 4, 27

ISI intersymbol interference 16

LTE Long Term Evolution 1, 4, 6

MAC media access control iii, iv, 2, 11, 23 MCS modulation & coding scheme 2, 4, 15 MMSE-IRC Minimum Mean Square Error – Interference

Re-jection Combining

16

NR New Radio iii, iv, 4,

6 OFDM Orthogonal Frequency-division Multiplexing 6, 7, 16

PDU Protocol Data Unit 23, 27

PRB physical resource block 8, 15, 16, 18, 19, 21

QoS quality of service 1, 2, 3,

4, 12, 21, 23, 27

RAN radio access networks 1, 4

RRM radio resource management 1, 2, 4 SINR signal-to-interference-plus-noise ratio 2, 16 SPS semi-persistent scheduling 2, 5 TTI transmission time interval 6, 16

(10)

Acronyms xi UE user equipment 1, 2, 5, 8, 9, 10, 13, 14, 15, 18, 23

(11)

xii Acronyms

Acknowledgement

I would like to thank my supervisor Abdulrahman Alabbasi and manager Jonas Kronander for giving me such a chance to do this thesis at Ericsson. I truly appreciate Dr. Alabbasi’s guidance and his devotion to work really impresses me. I really learned a lot from him. I also appreciate the chance to work at Ericsson Research, where there are a lot of excellent colleagues and great research atmosphere. I would like to thank my colleagues for their help during this thesis. Thanks David Sandberg, Jonas Olsson M, Alexey Shapin, and Torsten Dudda. Without their help, I might need even more time to code for the simulator. Thanks Milad Ganjalizadeh for using his resource for me to run simulations. It really saves a lot of time for me. I also appreciate Dr. Liang Hu, Yanpeng Yang and Kittipong Kittichokechai’s advice during this thesis!

I thank my academic supervisor Ki Won Sung and examiner Gabor Fodor for helping me from academic perspectives. Thanks for their help and guid-ance!

Thanks my smart girlfriend Jiaojiao Zheng for her support during this pro-cess. Her sincere support and help inspire me so much and I appreciate it from the bottom of my heart.

Finally, I would like to thank my father, mother, uncles, aunts and grand-mother for their continuous help during my whole life. They always give their best to me and never care about return. Without them, I could not have achieved anything.

(12)

Chapter 1 Introduction

The next generation of radio access networks (RAN) is expected to support more complex traffic demands such as time-critical communication and en-hanced Mobile Broadband (eMBB) traffics. Those traffic types require bet-ter quality of service (QoS) from the perspective of latency, reliability and throughput [1]. The 4G (Long Term Evolution (LTE)) RAN most likely can-not align with the growing demand of these QoS requirements[2]. Thus, the next generation of wireless networks (5G) is developed to handle those re-quirements.

An important scenario the 5G networks are expected to support is indus-trial Internet of Things (IoT), where traffic with low latency and high through-put QoS coexists [3]. Satisfying the latency requirement usually comes with the cost of spectral efficiency and vice versa [4]. Thus, research from differ-ent perspectives has been conducted to solve this problem, among which new protocols and algorithms of radio resource management (RRM) have attracted great attention [1].

This thesis aims at maintaining latency performance while improving the spectral efficiency of 5G RAN systems by means of improving RRM schemes, so that the spectral efficiency of 5G systems is improved to support industrial IoT efficiently. More specifically, the effectiveness of letting the user equip-ment (UE) choose the uplink transmission format is studied.

(13)

2 CHAPTER 1. INTRODUCTION

1.1 Background

1.1.1 Traffic in Industrial IoT

Various traffic types exist in industrial IoT. They can be periodic or aperiodic. Some traffic is time-sensitive and requires stringent reliability and latency per-formance, while others are less time-critical but may require high throughput.

1.1.2 Radio Resource Management

In order to use radio resources efficiently and satisfy the QoS for different traffic types, 5G uses contention-free media access control (MAC) protocol. In current 5G systems, the base stations are responsible for RRM for all the UEs in a system, i.e. the base stations allocate time-frequency resources, transmission power, and modulation & coding scheme (MCS) for all UEs in a system.

There are two kinds of scheduling schemes in 5G RRM, one is grant-based scheduling in which the base station should send a dynamic grant (DG) to a UE for each transmission in the uplink. The other type is grant-free (semi-persistent scheduling (SPS)) scheduling in which the UE could transmit data using the pre-allocated resource and format without requesting grants from the base station before each tranmission [1]. In neither of these two scheduling schemes can the UE choose the transmission format by itself.

However, due to the unpredictable arrival of aperiodic traffic and time-varying signal-to-interference-plus-noise ratio (SINR), the transmission for-mat decided by the base station might not be optimal. Neither of grant-based scheduling and grant-free scheduling could well handle this problem, which will be discussed in Section 1.2.

1.2 Problem Statement

Under the scenario of industrial IoT, time-critical traffic and eMBB traffic co-exists.

SPS is seen as a solution for fast uplink access. It works well for peri-odic traffic. But it leads to low spectral efficiency and flexibility for aperiperi-odic traffic due to the unpredictable arrival of the aperiodic traffic [5]. In order to address this issue, 3GPP allows the base station to send a DG to override peri-odic resource allocated by the configured grant (CG), so that the flexibility is increased [6]. Under such condition, if the critical traffic uses the resource and

(14)

CHAPTER 1. INTRODUCTION 3

format allocated by a less robust grant, there comes potential latency perfor-mance loss. If the critical traffic does not use the DG-allocated resource and format, it has to wait for the next physical resource block that is not overridden, leading to higher delay.

1.3 Benefit, Ethics, and Sustainability

This thesis proposed a new way to handle heterogeneous traffic. By giving the UE the flexibility to choose a grant, the waste and inefficient use of spectrum is reduced, leading to an increase in the spectral efficiency. And this enhance-ment is achieved with almost no loss in QoS, meaning that the same service could be provided with less equipment or spectrum.

Less equipment means less pollution, less power and resource consump-tion, and less cost of money. Thus, it is both environment-friendly and eco-nomic.

By saving the spectrum, we could have spectrum resources reserved for future use. That makes social development more sustainable.

1.4 Outline

The content of this thesis is organized as follows: Chapter 2 introduces the the-oretical background and some related work; Chapter 3 explains the methodol-ogy of this study; Chapter 4 presents the detailed parameters and models used in this study; Chapter 5 includes the results of this study and the analysis of the results; In Chapter 6 we draw our conclusions and propose some suggestions for future work.

(15)

Chapter 2 Background

This section described the related background knowledge of this thesis project. Chapter 2.1 explains what RRM is and details of two scheduling schemes in 5G systems. Chapter 2.2 is a brief introduction of the frame structure used in 5G. Chapter 2.3 is about some related work to support heterogeneous traffic in industrial IoT scenarios.

2.1 Concept of Uplink Scheduling/ RRM

In-troduction

A base station serves multiple users. However, if the users are transmitting at the same time using the same time-frequency resource, the collision might happen. So proper radio resource management schemes are needed, so that there is an appropriate arrangement of time-frequency resources, antennas, power level, and MCS, etc.

The reservation-based protocols of RRM could be used to avoid collisions, and there are two phases in the reservation-based protocols, reservation and data phase. The reservation phase could be divided into two types, scheduling and contention based [7]. Since spectrum is a kind of scarce natural resources, a complex scheduling-based protocol is chosen for modern RAN, e.g. LTE and New Radio (NR) [7, 8]. Meanwhile, the resource allocation schemes also need to consider the QoS requirement of different traffic [9]. In this thesis, only uplink scheduling for heterogeneous traffic is studied.

(16)

CHAPTER 2. BACKGROUND 5

Figure 2.1: Grant-based uplink accessing

2.1.1 Uplink Schduling in 5G

The scheduler of the base station (interchangeable with gNodeB in this thesis) determines the transmission parameters and physical resources for the UEs. Uplink transmission-related information including modulation and coding for-mat, resource allocation, and hybrid Automatic Repeat Request (HARQ) in-formation are included in an uplink grant and the grant should be sent to the UE in order to conduct uplink transmissions [10]. According to the commu-nication procedures between the UE and the base station, scheduling could be divided into grant-based (dynamic) scheduling and SPS (grant-free schedul-ing).

Dynamic Scheduling

Dynamic scheduling is the basic operation mode for 5G.

As shown in Figure 2.1, each time the UE wants to transmit data, it has to send a scheduling request (SR) or buffer status report to the gNodeB, then the gNodeB informs the UE what resource and transmitting parameters it should use by sending a grant to the UE. This type of scheduling provides the wireless communication systems with good flexibility. Its drawback is that it causes additional control signaling overhead, whose impact could be magnified under frequent traffic arrivals.

Grant-free Scheduling

Grant-free scheduling (SPS) in 5G shown in figure 2.2 features that the gN-odeB allocates periodic time-frequency resources to a UE by issuing a CG only at the beginning of a set of transmissions. Thus, the overhead of consecutive scheduling requests and grants transmissions is greatly reduced in grant-free scheduling, leading to a reduction in latency. However, when using CG, the transmission format of a UE is fixed, leading to a low flexibility of resource

(17)

6 CHAPTER 2. BACKGROUND

Figure 2.2: Grant-free uplink accessing

Table 2.1: Subcarrier Spacing Supported by NR

allocation. To overcome this shortage, 3GPP Release 15 allows the gNodeB to send another grant to override the configured grant [11].

2.2 Frame Structure

Orthogonal Frequency-division Multiplexing (OFDM) is chosen as the basic transmission waveform in both the downlink and uplink transmission direc-tions of 5G [12]. By making the sub-channels narrow-band, inter-symbol in-terference is greatly reduced [13]. In LTE, 15 kHz sub-carrier spacing is used while in NR this sub-carrier spacing value is scaled by powers of two.

2.3 Related Work

To meet the low latency and high reliability requirement of critical traffic, academia and industries have paid efforts from different perspectives. Flex-ible numerology and frame structure made it possFlex-ible for 5G to utilize shorter transmission time intervals (TTIs) than LTE, which is essential to achieve low

(18)

CHAPTER 2. BACKGROUND 7

latency [10, 14]. On top of that, link adaptation enhancement by including time-filtered interference information in channel quality indicator (CQI) is proposed in [15] and achieves good latency performance at low critical traffic load. Adaptive OFDM-subcarrier selection is studied in [16], it could enhance reliability by using an adaptive channel-based subcarrier selection mechanism compared to the standard OFDM scheme. [17] shows that grant-free transmis-sions could achieve lower latency even at considerable load. However, high reliability and low latency come with low spectral efficiency [4], the afore-mentioned studies do not consider the proper handling of co-existing eMBB traffic in 5G, which is gaining increasing attention.

A joint scheduler of both eMBB and critical traffic is proposed in [18], dif-ferent eMBB rate loss models are evaluated when critical traffic overlays with eMBB traffic. In [19, 20], the spatial pre-emption schemes are proposed and the system capacity is improved significantly. All schemes in [18, 19, 20] are about dynamic scheduling in the downlink. In [21], power control settings of multiplexing of eMBB and grant-free critical traffic in the uplink is studied, but it is concluded that the overlaying of the traffic is only feasible at low critical traffic load. In the aforementioned researches of this paragraph, great inter-ests have been put to the superposition/puncturing of the traffic. The resource conflicts-free scheme has not been studied thoroughly, which we believe is a very interesting topic, especially at high critical traffic load.

(19)

Chapter 3 Methodology

Experiment is the main method of this study because system-level analysis of wireless communication systems is quite complex and hard to build mathe-matical models to represent. However, the mathemathe-matical models of each part including but not limited to fading, modulation, noise, and interference will be inserted in our Java-based simulation tool. The simulator is developed by Er-icsson, and data is collected through simulations. Computational mathematics (MATLAB) is used to analyze the results.

We compare our proposed scheme with two benchmark schemes to present the effectiveness of our proposal under different systems settings and traffic conditions. In section 3.1 we give a brief introduction of two benchmarks and our proposal. Some basic information of the simulator is given in section 3.2. In section 3.3, how the simulations are conducted (i.e. traffic conditions, pa-rameters to be varied for each simulation, number of simulations, etc) is ex-plained. In section 3.4, what kinds of data we will analyze and how to analyze them are discussed.

3.1 Benchmark and Proposal

We will compare our proposal with two benchmarks to demonstrate the per-formance of our proposed scheme.

In all these three schemes, we have configured grants that allocate periodic physical resource blocks (PRBs) to the UEs. And CG is assumed to be more robust than DG. Besides, critical traffic will not use DG in any of these three schemes.

In Benchmark 1, there is no DG that has overlapping PRBs with the CG. While in Benchmark 2, once a UE receives a DG, it always uses the newly

(20)

CHAPTER 3. METHODOLOGY 9

Figure 3.1: Benchmark 2

Figure 3.2: Proposal

received DG, i.e. CG will be overridden by the DG as shown in figure 3.1, assuming all boxes in this figure use the same frequency resource.

Under our proposed scheme, the DG will not always be prioritized. The UE will select a proper grant depending on if it has critical traffic. If the UE has critical data to transmit, it will choose the transmitting format configured by the CG. Otherwise, the DG will be selected as shown in figure 3.2.

3.2 Simulator and Implementation

The simulator used in this thesis is a system-level simulator developed by Er-icsson. However, for some reason, there is no support for re-transmission and repetition when CG is enabled. And before our implementation, it only sup-ports Benchmark 1 at CG periodicity of 0.5ms. Proposal and Benchmark 2 are supported after our implementation.

(21)

10 CHAPTER 3. METHODOLOGY

3.3 Simulation Design

Simulations will be run under different system settings and traffic conditions, because we do not expect our proposal to outperform the benchmarks in all conditions in terms of latency and throughput performance. For each param-eter (e.g. traffic request intensity, data unit size, number of users), we iterate over multiple values to explore the effectiveness under different conditions. In order to avoid the results being dominated by the outliers, ten random seeds are run for each simulation. Each simulation lasts for 36 seconds.

3.3.1 The Impact of Configured Grant Periodicity on

the Effectiveness of the Proposal

In the first simulation series, the traffic load is fixed. The CG periodicity is varied for all these three schemes to study the effectiveness under different CG periodicities, i.e. 0.5ms, 2.5ms, and 5ms.

3.3.2 The Impact of User Amount on the Effectiveness

of the Proposal

In this experiment, how the number of UEs affects these schemes’ performance will be evaluated. Thus, we will run simulations when the number of users varies from 21 to 147 when other settings are fixed. Latency performance will be the focus of this study.

3.3.3 The Impact of Critical Traffic on the

Effective-ness of the Proposal

Critical request intensity is varied in this simulation series. Both the bench-marks and the proposal will be evaluated at different CG periodicities (0.5ms, 2.5ms, 5ms) and critical traffic data unit size (320 bits, 744 bits). When sim-ulating for different CG periodicities, critical traffic request intensity ranges from 20 requests/s to 1000 requests/s. Otherwise, it ranges from 20 requests/s to 800 requests/s.

(22)

CHAPTER 3. METHODOLOGY 11

3.3.4 The Impact of non-critical traffic load on the

Ef-fectiveness of the Proposal

Since the aim of this thesis is to enhance the spectral efficiency and over-provisioning might occur, we would like to vary the non-critical (eMBB) traffic data unit size so that the allocated resource goes from being more than enough to being less than enough. And we will analyze how this variation affects these schemes.

3.4 Data collection and Analysis Method

Data is collected by selecting corresponding fields of the simulator and will be exported as ’.mat’ format files. Note that in the simulations, critical traffic is named ’cmtc’ traffic and non-critical traffic is written as eMBB traffic. Below is an introduction to each log field. Both delay and duration are latency for one hop on MAC layer.

• Duration

This is one field in the simulator to measure latency. The simulator logs duration for each data unit of each user in the simulation process. This field logs all the data units generated during the simulations. However, values that are larger than an upper bound will be logged as the value of the upper bound. The upper bound is configurable.

• Delay

This field also serves to measure latency. The difference from duration is that it only logs for successfully received data units and there is no upper value bound as the duration.

• Throughput

T hroughput is not logged for each data unit as the other fields above. It is the throughput for each user in each second, meaning that each user has m values of throughput if the simulation lasts for m seconds.

3.4.1 Analysis

Latency Percentile at 3ms vs CG Periodicity

The latency percentile at 3ms means the percentage of data units of all users that has a latency value of less than 3ms. It could be obtained after the

(23)

cumula-12 CHAPTER 3. METHODOLOGY

tive distribution function (CDF) of delay has been calculated. It is important to know if our proposal could maintain the QoS, thus, we observe the perfor-mance of our proposal under different CG periodicities and compare it with the benchmarks.

Average Throughput

The average throughput is calculated by averaging the throughput value of each second of all the users. Figures are plotted versus CG periodicity, cmtc request intensity, eMBB data unit size.

Percentage of Users Reaching 99.99% Reliability at 5ms Latency (cmtc Traffic)

For each user, we use MATLAB to calculate that if it fulfills the QoS require-ment of reaching 99.99% reliability at 5ms latency. And then calculate the percentage of users that meets the QoS requirement. Just the same as

Aver-age Throughput, it will also be plotted versus CG periodicity, cmtc request

(24)

Chapter 4 System Model and Parameters

This chapter provides detailed information about the system settings and the simulator.

4.1 Simulation Scenarios and Setup

The scenario studied in this thesis is Urban Macro (UMa) defined in 3GPP 38.900 specification. And only outdoor users are considered. Table 4.1 is some basic information of the system, which is modified based on Table 7.2-1 in 3GPP 38.900 specification. The cell layout is a hexagonal grid and the number of base station sites is 7 as shown in Figure 4.1. Each site has 3 sectors. The inter-site distance is 166.66 meters. UEs are uniformly distributed.

Figure 4.1: Seven Hexagonal Cells

(25)

14 CHAPTER 4. SYSTEM MODEL AND PARAMETERS

Parameters Value (UMa)

Cell layout Hexagonal grid, 7 macro sites, 3 sectors

per site (Inter Site Distance = 166.66m)

BS antenna height 25m

UE Location

Outdoor/indoor Outdoor Line of Sight (LOS) LOS & Non-LOS

Height Same as 3D-Urban Macro in TR36.873

Indoor UE ratio 25m

UE mobility (horizontal plane only) 3km/h UE distribution (horizontal) Uniform

Table 4.1: System Setup

4.1.1 Propagation Model

The path loss model used in this thesis can also be found in 3GPP specfica-tion[22]. It is also listed in Table 4.2 below. Distance definitions are indicated in Figure 4.2. d0BP is called breakpoint distance, and d

0 BP = 4h 0 BSh 0 U Tfc/c , where fc is the center carrier frequency in Hz, c is light propagation ve-locity in free space. h0BS and h

0

U T are the effective antenna heights at the base station and UE side respectively. They could be calculated as follows: h0_{U T} = hU T − hE, h0BS = hBS − HE, in which hE is a function of the link between a BS and a UT. It depends on d2Dand hU T. Thus, it should be deter-mined for each link between a base station and a UE respectively.

(26)

CHAPTER 4. SYSTEM MODEL AND PARAMETERS 15

Table 4.2: Propagation Model

4.1.2 FTP Model 3

File Transfer Protocol (FTP) model 3 introduced in 3GPP document 36.889 is developed from FTP model 2 specified in 3GPP 36.814 [23, 24]. The packet arrival follows a poisson process with arrival rate λ. The transmission time is counted from the time a packet arrives in the queue.

4.1.3 Scheduling Algorithm

The scheduling algorithm for CG users is called sequential. Basically, each UE is allocated the same amount of PRBs following their arrival order. If there are not enough PRBs available, the rest PRBs will all be allocated to one UE. Other UEs which has no allocation will not be never be scheduled. In my implementation, each UE will get fifteen PRBs until there are not enough PRBs. For DG, the algorithm used is proportional fair. However, DG for a user can only use the same frequency resource as CG due to the specific implementation of the simulator.

Block Error Rate (BLER) target is used by the scheduler to determine the MCS for each grant. For CG, the value of target BLER is 10−6while it is set to 10−4 for the DG. However, when the base station thinks it could not empty a UE’s buffer, it will allocate a fixed value to DGs which is 2800 bits.

There is a mechanism in the simulator to trigger the base station issuing a DG, i.e. when the buffer estimation at the base station side is not 0, the base station will make a DG and send it to the UE.

(27)

16 CHAPTER 4. SYSTEM MODEL AND PARAMETERS

4.2 Parameters

The carrier frequency used in this thesis is 4 Ghz, and 40M hz bandwidth is used. frequency division duplex (FDD) is used as the duplexing scheme. The OFDM subcarrier spacing is set to 30 kHz. This will provide less latency than 15 kHz, and suffer less from intersymbol interference (ISI) compared to larger subcarrier spacing. As a result, each TTI last for 0.5 ms and have 14 OFDM symbols. The uplink receiver used in this thesis is the Minimum Mean Square Error – Interference Rejection Combining (MMSE-IRC) receiver, which could suppress the inter-cell interference and is expected to greatly improve 5G net-works’ performance[25]. Uplink link adaptation outer loop is disabled in our experiments. The targeted power control SINR is 20 dB.

Table 4.3: System Parameters

4.3 Traffic and Number of Users

Our proposal is aimed at dealing with the unpredictable arrival of traffic. Be-cause if CG is used to reduce latency for sporadic traffic, over-provisioning might happen and many PRBs might be wasted. Intuitively, for periodic traf-fic, such effect will be reduced if PRBs and traffic are well-aligned. Thus, the focus of this thesis will be a mixture of aperiodic critical (interchangeable with

(28)

CHAPTER 4. SYSTEM MODEL AND PARAMETERS 17

cmtc in this thesis) and non-critical (interchangeable with eMBB in this the-sis) traffic. Both types of traffic will use FTP model 3. The difference is their traffic intensity and data unit size. Two groups of common parameters will be used, as listed in Table 4.4 and Table 4.5.

Table 4.4: Scenario 1

(29)

Chapter 5 Results and Analysis

As stated in section 4.3, simulations are run under 2 different scenarios. How effective and how different traffic conditions affect the effectiveness of our proposal will be evaluated.

5.1 Impact of Configured Grant Periodicity

Figure 5.1 and Figure 5.2 are a group of results. Scenario 1 in Table 4.4 is used here. Their horizontal axes are both CG periodicity of 0.5 ms, 2.5 ms, and 5 ms.

The y-axis of Figure 5.1 is the percentage of critical packets from all users that have less than 3ms latency. In this figure, we can see that our proposal has pretty close latency performance to Benchmark 1 and is much better than Benchmark 2. Benchmark 2 is a scheme in which DG will directly override CG while critical traffic will not use DG. Thus, the PRBs allocated for critical traffic might become inadequate. The transmission rate for critical traffic is less than the demanded rate, leading to a lot of packets waiting in the buffer for longer than 3 ms. And that is why the latency percentile at 3ms for Benchmark 2 is very low. This problem does not exist for Benchmark 1 since there are no overlapping grants for Benchmark 1. As for our proposal, although overlapping grants exist, the UE will choose CG which is assumed to be more robust in this thesis as long as there are critical traffic packets in the buffer. This means that the behavior of the UE in our proposal is similar to that of Benchmark 1 when critical traffic presents. Intuitively, they will have close performance. Comparing the latency performance at different CG periodicities, we can see that short periodicity is desired for sporadic traffic.

The y-axis of Figure 5.2 is the average throughput of all users. We can

(30)

CHAPTER 5. RESULTS AND ANALYSIS 19

Figure 5.1: Latency Percentile at 3ms

see that our proposal becomes effective when CG periodicity is 0.5 ms. Under this periodicity, our proposal could improve the system throughput by 25% compared to Benchmark 1. Benchmark 2 can also achieve better throughput than Benchmark 1 because it could use the higher-efficiency grant, however, it does not make any sense since Benchmark 2 has poor latency performance. In other conditions regarding throughput performance, our proposal has similar performance as Benchmark 1. That is because the throughput in-crease happens in PRBs where overlapping grants co-exist and DG is selected. Under the periodicities of 2.5 ms and 5 ms, only 20% and 10% of the PRBs have overlapping grants respectively and not always will the DG be selected. Thus, the throughput increase of our proposal compared to Benchmark 1 hap-pens less than 20% and 10% of the PRBs when the CG periodicities are 2.5 ms and 5 ms respectively. Moreover, when only the DGs exist, how the base stations issue DGs will also affect the throughput, which can not be known pre-cisely. This also helps to explain the throughput trend when CG periodicities are 2.5 ms and 5 ms respectively in Figure 5.2.

(31)

20 CHAPTER 5. RESULTS AND ANALYSIS

Figure 5.2: Average Throughput

(32)

5.2 Impact of Number of Users

Figure 5.3 is simulated under Scenario 1 in Table 4.4. How the number of users would affect our results is evaluated here. The y-axis is no longer com-puted based on all packets in the system. Instead, it is based on the concept introduced in Subsection 3.4.1, i.e. the percentage of users reaching 99.99% reliability at 5 ms latency.

Our proposal and Benchmark 1 have similar performance. At CG peri-odicities of 0.5 ms and 2.5 ms, both schemes’ performance is decreasing as the number of users increases. One reason for the decrease is that more interfer-ence is generated as the number of users grows. Another reason is due to the scheduling algorithm adopted in the simulation, which is called sequential in Subsection 4.1.3. Due to this algorithm, some users’ critical traffic will never be served. When CG periodicity is 5ms, none of the schemes could achieve 99.99% reliability at 5ms latency. Intuitively, this is because CG periodicity is too long.

5.3 Impact of Critical Traffic Intensity

The impact of the critical traffic intensity is studied in two simulation groups.

Under Different Critical Data Unit Size

The simulations in this subsection are conducted under Scenario 1 in Table 4.4. Figure 5.4 presents the percentage of users fulfilling the 99.99% reliability at 5 ms latency QoS requirement when critical (cmtc) traffic request intensity varies under different critical traffic data unit size. Figure 5.5 shows how the average throughput changes with critical (cmtc) traffic request intensity under different cmtc data unit sizes.

From Figure 5.4, we can see that our proposal and Benchmark 1 have acceptable performance. When the cmtc request intensity exceeds 100 re-quests/s, both schemes could maintain their performance (higher than 85%) even when cmtc request intensity reaches 1000 requests/s, no matter whether the eMBB data unit size is 320 bits or 744 bits. Thus, when PCG/Ic = 1/2, the PRBs are sufficient to carry the critical traffic under the CG format in this thesis, where PCG is the periodicity of configured grant and Ic is the request intensity of the critical traffic. This, on the other hand, illustrates that in Fig-ure 5.3, the poor performance at the CG periodicity of 2.5 ms is mainly due to the long periodicity rather than the amount of PRBs.

(33)

Figure 5.4: Percentage of Users Fulfilling QoS Requirement with Variable cmtc Request Intensity, under Different cmtc Data Unit Size

Figure 5.5: Average Throughput vs Critical Traffic Intensity, under Different cmtc Data Unit Size

(34)

Benchmark 2 performs much worse than the other two schemes. It is simply because CGs are frequently overridden by DGs.

Moving on to Figure 5.5, the average throughput of Benchmark 1 in-creases with critical (cmtc) traffic intensity while our proposal appears to be a concave shape. The capacity of Benchmark 1 does not increase since its re-source amount and transmission format are fixed. The throughput increase of Benchmark 1 is because more critical packets are multiplexed into MAC Pro-tocol Data Units (PDUs). The critical traffic packets have a smaller size than eMBB packets, leading to less padding bits in a MAC PDU. As a consequence, the system throughput is slightly improved. The reason for the throughput in-crease of our proposal is dominated by the adoption of DG. When traffic load increases, the base station allocates more overlapping grants to a UE. As more DGs are used, the spectral efficiency increases. However, as critical traffic be-comes more and more intense, the UE has to use more CGs to guarantee the QoS of critical traffic, which sacrifices the spectral efficiency and also explains the drop when cmtc intensity becomes larger than 400 requests/s.

Overall, our proposal could lead to a throughput increase of 21% to 25% compare to Benchamrk 1. Our proposal’s throughput performance is not compared with Benchmark 2 as it could not satisfy the QoS requirement at all.

Under Different CG Periodicity

The simulations in this subsection are conducted under Scenario 2 in Table 4.5. The axes of Figure 5.6 and Figure 5.7 are the same as Figure 5.4 and Figure 5.5 respectively. The difference is that simulations in this subsection are run for different CG periodicities.

From Figure 5.6, we can see that when CG periodicity is 0.5 ms, 93% to 95% of UEs in both of our proposal and Benchmark1 could maintain 99.99% reliability at 5 ms when critical (cmtc) traffic request intensity varies. At the periodicity of 2.5 ms, the percentage of UEs fulfilling QoS requirement of both our proposal and Benchmark 1 decreases significantly as the critical traffic request intensity grows. This illustrates that in order to achieve good reliability and low latency, over-provisioning is needed. At the periodicity of 5 ms, the resource allocated to a UE is inadequate, thus the performance is very bad.

Moving on to Figure 5.7, we could see that our proposal shows noticeable improvement only at the CG periodicity of 0.5 ms. When CG periodicity is larger than 0.5 ms, the difference is very small. The reason is the same as that in Figure 5.2.

(35)

Figure 5.6: Percentage of Users Fulfilling QoS Requirement with Variable cmtc Request Intensity, under Different CG Periodicities

Figure 5.7: Average Throughput vs Critical Traffic Intensity, under Different CG Periodicity

(36)

Figure 5.8: Percentage of Users Fulfilling QoS Requirement with Variable eMBB Data Unit Size, under Different CG Periodicities

5.4 Impact of eMBB Traffic Load

Our proposal is not superior to the benchmarks under all conditions. In this section, the impact of the eMBB traffic load on the effectiveness of our pro-posal is studied. Figure 5.8 presents the percentage of users fulfilling 99.99% reliability at 5 ms under different eMBB data unit size. Figure 5.9 shows how system throughput changes with the eMBB data unit size. The simulations are conducted under Scenario 2 in Table 4.5.

In Figure 5.9, both our proposal and Benchmark 2 appear that the shorter the CG periodicity, the higher the throughput. Comparing Benchmark 1 with our proposal, we could see that there is almost no difference when the eMBB data unit size is smaller than 120000 bits. When data unit size becomes bigger than 120000 bits, our proposal starts to outperform Benchmark 1. That is because when the eMBB data unit size is smaller than 120000 bits, both the proposal and Benchmark 1 can handle the traffic load. Thus, the through-put we see is almost the same as the traffic load. However, when the eMBB data unit size becomes larger, Benchmark 1 then reaches its bottleneck. The throughput almost can not increase as the traffic load becomes heavier. Due

(37)

Figure 5.9: Average Throughput vs eMBB Data Unit Size, under Different CG Periodicities

to the adoption of DGs, our proposal could handle more traffic. In Figure 5.8, we can see that the user percentage of Benchmark 2 decreases quickly as the eMBB data unit size grows while Benchmark 1 and our proposal could maintain the latency performance when CG periodicity is 0.5 ms.

(38)

Chapter 6 Conclusions and Future Work

In order to support critical traffic in industrial IoT scenarios, CG is proposed in 3GPP standard. But due to the unpredictable arrival of sporadic traffic, the current scheme either sacrifices spectral efficiency to guarantee QoS or lost its reliability for high throughput. Thus, we proposed a new scheme for intra-UE overlapping grants handling and evaluated it under different scenarios using a system-level simulator. Latency and throughput performance are paid great attention.

In order to achieve good latency performance for aperiodic traffic, over-provisioning is needed. No matter if the critical traffic is intense, short CG periodicity is essential to achieve good latency performance. However, over-provisioning leads to low spectral efficiency. When the traffic load is low, both Benchmark1 and our proposal performs well. However, when traffic load become high, our proposal shows better throughput performance than Benchmark 1.

This thesis can be further extended, i.e. more factors could be taken into account when choosing a grant and multiplexing data into a PDU.

In this thesis, as long as the critical traffic exists, CG will be used. Possible future research questions could are:

• When critical traffic exists, under what channel quality can DG be used

for critical traffic?

• Under such a scheme, what would the latency and throughput

perfor-mance be like?

Since we do not consider re-transmission and repetition schemes in this thesis, another problem is:

• How would re-transmission and repetition affect the system performance?

(39)

Bibliography

[1] António Morgado et al. “A survey of 5G technologies: Regulatory, stan-dardization and industrial perspectives”. In: Digital Communications

and Networks 4.2 (2018), pp. 87–97.

[2] Zsolt Marcell Temesvári, Dóra Maros, and Péter Kádár. “Review of Mobile Communication and the 5G in Manufacturing”. In: Procedia

Manufacturing 32 (2019), pp. 600–612.

[3] Shancang Li, Li Da Xu, and Shanshan Zhao. “5G Internet of Things: A survey”. In: Journal of Industrial Information Integration 10 (2018), pp. 1–9.

[4] Beatriz Soret et al. “Fundamental tradeoffs among reliability, latency and throughput in cellular networks”. In: 2014 IEEE Globecom

Work-shops (GC Wkshps). IEEE, pp. 1391–1396. isbn: 1479974706.

[5] Dajie Jiang et al. “Principle and performance of semi-persistent schedul-ing for VoIP in LTE system”. In: 2007 International Conference on

Wireless Communications, Networking and Mobile Computing. IEEE.

2007, pp. 2861–2864.

[6] 3GPP. NR; Medium Access Control (MAC) protocol specification. Tech-nical Specification (TS) 38.321. Version 15.0.0. 3rd Generation Part-nership Project (3GPP), Jan. 2018. url: http://www.3gpp.org/ %5C-DynaReport/%5C-38321.htm.

[7] Guowang Miao et al. Fundamentals of mobile data networks. Cam-bridge University Press, 2016.

[8] John Chapin and William Lehr. “Mobile broadband growth, spectrum scarcity, and sustainable competition”. In: TPRC. 2011.

(40)

BIBLIOGRAPHY 29

[9] I. Siomina, A. Furuskär, and G. Fodor. “A mathematical framework for statistical QoS and capacity studies in OFDM networks”. In: 2009

IEEE 20th International Symposium on Personal, Indoor and Mobile Radio Communications. Sept. 2009, pp. 2772–2776. doi: 10.1109/

PIMRC.2009.5450287.

[10] 3GPP. NR; Overall description; Stage-2. Technical Specification (TS) 38.300. Version 15.0.0. 3rd Generation Partnership Project (3GPP), Jan. 2018. url: http://www.3gpp.org/%5C-DynaReport/%5C-38300.htm.

[11] Evolved Universal Terrestrial Radio Access. “Medium Access Control (MAC) Protocol Specification”. In: Release 8 (2008), p. 30.

[12] Erik Dahlman, Stefan Parkvall, and Johan Skold. 5G NR: The next

gen-eration wireless access technology. Academic Press, 2018.

[13] Robert W Chang. “Synthesis of band-limited orthogonal signals for multichannel data transmission”. In: Bell System Technical Journal 45.10 (1966), pp. 1775–1796.

[14] 3GPP. Study on latency reduction techniques for LTE. Technical Re-port (TR) 36.881. Version 14.0.0. 3rd Generation Partnership Project

(3GPP), July 2016. url: https://portal.3gpp.org/desktopmodules/ Specifications/SpecificationDetails.aspx?specificationId= 2901.

[15] Guillermo Pocovi et al. “MAC layer enhancements for ultra-reliable low-latency communications in cellular networks”. In: 2017 IEEE

In-ternational Conference on Communications Workshops (ICC Workshops).

IEEE. 2017, pp. 1005–1010.

[16] J. M. Hamamreh, E. Basar, and H. Arslan. “OFDM-Subcarrier Index Selection for Enhancing Security and Reliability of 5G URLLC Ser-vices”. In: IEEE Access 5 (2017), pp. 25863–25875. issn: 2169-3536. doi: 10.1109/ACCESS.2017.2768558.

[17] Thomas Jacobsen et al. “System level analysis of uplink grant-free trans-mission for URLLC”. In: 2017 IEEE Globecom Workshops (GC

Wk-shps). IEEE. 2017, pp. 1–6.

[18] A. Anand, G. De Veciana, and S. Shakkottai. “Joint Scheduling of URLLC and eMBB Traffic in 5G Wireless Networks”. In: IEEE INFOCOM

2018 - IEEE Conference on Computer Communications. Apr. 2018,

(41)

30 BIBLIOGRAPHY

[19] Ali A. Esswie and Klaus I. Pedersen. “Opportunistic Spatial Preemp-tive Scheduling for URLLC and eMBB Coexistence in Multi-User 5G Networks”. In: IEEE Access 6 (2018), pp. 38451–38463. issn: 2169-3536. doi: 10 . 1109 / access . 2018 . 2854292. url: https : //dx.doi.org/10.1109/access.2018.2854292.

[20] A. A. Esswie and K. I. Pedersen. “Capacity Optimization of Spatial Pre-emptive Scheduling for Joint URLLC-eMBB Traffic in 5G New Radio”. In: 2018 IEEE Globecom Workshops (GC Wkshps). Dec. 2018, pp. 1–6. doi: 10.1109/GLOCOMW.2018.8644070.

[21] Renato Abreu et al. “System Level Analysis of eMBB and Grant-Free URLLC Multiplexing in Uplink”. In: IEEE. doi: 10.1109/vtcspring. 2019 . 8746557. url: https : / / dx . doi . org / 10 . 1109 / VTCSpring.2019.8746557.

[22] 3GPP. Study on 3D channel model for LTE. Technical Report (TR) 36.873. Version 12.4.0. 3rd Generation Partnership Project (3GPP), Mar. 2017. url: https://portal.3gpp.org/desktopmodules/

Specifications/SpecificationDetails.aspx?specificationId= 2574.

[23] 3GPP. Feasibility Study on Licensed-Assisted Access to Unlicensed

Spec-trum. Technical Report (TR) 36.889. Version 13.0.0. 3rd Generation

Partnership Project (3GPP), July 2015. url: https : / / portal .

3gpp.org/desktopmodules/Specifications/SpecificationDetails. aspx?specificationId=2579.

[24] 3GPP. Evolved Universal Terrestrial Radio Access (E-UTRA); Further

advancements for E-UTRA physical layer aspects. Technical Report

(TR) 36.814. Version 9.2.0. 3rd Generation Partnership Project (3GPP), Mar. 2017. url: https://portal.3gpp.org/desktopmodules/

Specifications/SpecificationDetails.aspx?specificationId= 2493.

[25] Fernando ML Tavares et al. “On the potential of interference rejection combining in B4G networks”. In: 2013 IEEE 78th Vehicular

(42)

TRITA TRITA-EECS-EX-2020:18

Handling of Intra-UE Uplink Overlapping Grants for Heterogeneous Traffic in 5G Wireless Systems

Handling of Intra-UE Uplink

Overlapping Grants for

Heterogeneous Traffic in 5G

Wireless Systems

JIACHENG ZHENG

Abstract

Sammanfattning

Contents

List of Tables

List of Figures

Acronyms

Acknowledgement

Chapter 1

Introduction

1.1

Background

1.1.1

Traffic in Industrial IoT

1.1.2

Radio Resource Management

1.2

Problem Statement

1.3

Benefit, Ethics, and Sustainability

1.4

Outline

Chapter 2

Background

2.1

Concept of Uplink Scheduling/ RRM

In-troduction

2.1.1

Uplink Schduling in 5G

2.2

Frame Structure

2.3

Related Work

Chapter 3

Methodology

3.1

Benchmark and Proposal

3.2

Simulator and Implementation

3.3

Simulation Design

3.3.1

The Impact of Configured Grant Periodicity on

the Effectiveness of the Proposal

3.3.2

The Impact of User Amount on the Effectiveness

of the Proposal

3.3.3

The Impact of Critical Traffic on the

Effective-ness of the Proposal

3.3.4

The Impact of non-critical traffic load on the

Ef-fectiveness of the Proposal

3.4

Data collection and Analysis Method

3.4.1

Analysis

Chapter 4

System Model and Parameters

4.1

Simulation Scenarios and Setup

4.1.1

Propagation Model

4.1.2

FTP Model 3

4.1.3

Scheduling Algorithm

4.2

Parameters

4.3

Traffic and Number of Users

Chapter 5

Results and Analysis

5.1

Impact of Configured Grant Periodicity