Abstract—In this paper, we study Voice-over-IP (VoIP)

performance in UTRA Long Term Evolution (LTE) Downlink (DL). We have utilized fully dynamic system simulations to study the VoIP Adaptive Multi-Rate (AMR) 12.2 codec capacity in four different 3GPP simulation cases. The effects of Link Adaptation (LA), packet bundling, control channel capacity and number of HARQ processes on VoIP capacity have also been considered. The results present the absolute VoIP capacity numbers of LTE DL. We also show that LA together with packet bundling provides clear gain on the VoIP capacity, because more VoIP packets can be scheduled in each TTI. Also, the control channel limitations can be effectively compensated by packet bundling.

Index Terms—VoIP, LTE, system simulations

I. INTRODUCTION

The Evolved UTRAN (E-UTRAN) or the UTRAN Long Term Evolution (LTE) specifications are being finalized in 3GPP. LTE aims at ambitious goals of e.g. peak data rate of 100 Mbps in downlink and 50 Mbps in uplink, increased cell edge user throughput, improved spectral efficiency, scalable bandwidth from 1.25 MHz to 20 MHz, etc. [1].

The main principles of E-UTRA downlink, uplink and the core network have been decided already. LTE supports both time (TDD) and frequency division duplex (FDD) modes, but in this article we concentrate on FDD. Orthogonal Frequency Division Multiple Access (OFDMA) has been selected for the downlink multiple access technology and Single Carrier Frequency Division Multiple Access (SC-FDMA) for uplink [1]. To achieve the objectives set for LTE, advanced Radio Resource Management (RRM) functions have been defined. The algorithms include e.g. Hybrid ARQ (HARQ), Link Adaptation (LA), Channel Quality Indication (CQI) and Packet Scheduling (PS). More on these can be found e.g. from [2].

E-UTRAN is optimized for packet data transfer and the core network is purely packet switched, so speech is transmitted purely with Voice-over-IP (VoIP). VoIP traffic consists of talk-spurts and silent periods, with relatively small

packets transmitted quite rarely. The Adaptive Multi-Rate (AMR) codec provides quite bursty traffic; one VoIP packet at 20 ms intervals during talk spurt and one Silence Indicator (SID) packet at 160 ms intervals during silence period. E-UTRAN is expected to support a very high number of VoIP users and the Quality-of-Service (QoS) for VoIP is determined by maximum End-to-End delay and tolerable packet loss. These facts set challenges to the resource allocation of VoIP users, for both PS and LA algorithms. Also, the capacity of Physical Downlink Control Channel (PDCCH) induces some restrictions, at least with higher system bandwidths. These restrictions become most relevant with dynamic packet scheduling, since each allocation consumes signaling resources from PDCCH. Thus, several persistent resource allocation schemes, such as fully persistent scheduling, talk-spurt based persistent scheduling and semi-persistent scheduling have also been proposed in 3GPP [3]. However, these scheduling types limit the gain from multi-user and frequency domain scheduling. VoIP service in E-UTRAN has been studied e.g. in [4] and [5].

The objective of this article is to provide the baseline VoIP performance results of E-UTRAN FDD downlink with dynamic packet scheduling. The effect of different features, such as system bandwidth, LA, Control Channel (CC) capacity, packet bundling and HARQ processes, on VoIP capacity are studied using simulations. The simulation results are gathered from fully dynamic system simulator, which models the UE mobility, RRM functionalities and their interactions with the system.

The paper is organised as follows: Chapter II discusses the general aspects of VoIP in LTE and related modeling. Chapter III lists the simulation assumptions including a short description of the simulator. Chapter IV presents the simulation results and analysis. Finally, Chapter V reviews the main conclusions.

Voice-over-IP Performance in UTRA Long Term Evolution Downlink

Jani Puttonen1, Tero Henttonen2, Niko Kolehmainen3, Kennett Aschan2, Martti Moisio2 and Petteri Kela3

1Magister Solutions Ltd, c/o Mattilanniemi 6-8,

40101 Jyväskylä, Finland. Email: firstname.lastname@magister.fi

2Nokia, P.O.BOX 45, FIN-00045 Nokia Group, Finland Email: firstname.lastname@nokia.com

3University of Jyväskylä, Dept. of Mathematical Information Technology

P.O. Box 35, 40014 University of Jyväskylä, Finland. Email: firstname.lastname@jyu.fi

978-1-4244-1645-5/08/$25.00 ©2008 IEEE 2502

II. VOICE-OVER-IP IN LTE

VoIP has at least three characteristics that need consideration in LTE (as well as in any wireless system): Bursty low bitrate traffic, strict packet delay-based QoS and high number of simultaneous users). These issues set challenges to the RRM functions. Next, we discuss these characteristics as well as the required RRM functions in more detail.

A. High Capacity Demand

The requirements of E-UTRA and E-UTRAN are described in TR.25.813 [6]. The service related requirements for VoIP are:

• The E-UTRA should efficiently support various types of service. These must include currently available services like web-browsing, FTP, video-streaming or VoIP, and more advanced services (e.g. real-time video or push-to-talk) in the Packet Switched domain.

• VoIP should be supported with at least as good radio backhaul efficiency and latency as voice over UMTS Circuit Switched (CS) networks.

• Voice and other real-time services supported in the CS domain in Release 6 shall be supported by E-UTRAN via the packet switched domain with at least equal quality as supported by UTRAN (e.g. in terms of guaranteed bit rate) - over the whole speed range.

B. Strict packet delay-based QoS

The system capacity for VoIP service is limited by the outage limits defined in TR 25.814 [1] and updated in 3GPP contribution R1-070674 [7]:

• The system capacity is defined as the number of users in a cell when more than 95% of the users are satisfied

• A single VoIP user is in outage if less than 98% of its speech frames are delivered successfully within 50 ms air interface delay.

According to [8], the maximum acceptable mouth-to-ear delay for voice is on the order of 250 ms. Assuming that the delay for Core Network is approximately 100 ms, the tolerable delay for Radio Link Control (RLC) and Medium Access Control (MAC) buffering, scheduling and detection should be strictly lower than 150 ms. Hence, assuming that both end users are E-UTRAN users, tolerable delay for buffering and scheduling is lower than 80 ms. A delay bound of 50 ms (for delay from eNB to UE) has been chosen for the 3GPP performance evaluations to better account for variability in network end-to-end delays.

C. Bursty low bitrate traffic

In the context of this article, we consider VoIP traffic as provided by AMR codec. The AMR VoIP traffic is quite bursty: There’s one VoIP packet at 20 ms intervals during talk spurt and one SID packet at 160 ms intervals during silence period. Thus, for any given TTI, only few of the active users need to be scheduled. At the same time, each unscheduled user contributes to a backlog of scheduling requests for later TTIs.

Since this backlog can start accumulating easily, leading to resource stalling for several users, scheduling should take care that the buffering delay of each VoIP user is taken into account in the scheduling decisions.

Since VoIP packets are relatively small (regardless of the used AMR codec), there are some challenges in allocating the resources; 2-4 symbols of each carrier in each Physical Resource Block (PRB) are reserved for control data (reference symbols, allocation information, HARQ ACK/NACK channels), depending on the need for allocation signaling. With the demand for several users to be scheduled simultaneously, the control channel capacity might become a limit for the VoIP capacity due to lack of signaling bits.

D. Packet scheduling and link adaptation

Since VoIP is strictly delay-restricted service, the PS needs to take the buffering delay of UEs into account. As presented in e.g. [9], dynamic packet scheduling provides good frequency domain and multi-user gain for best effort type traffic. However, because of the VoIP service characteristics discussed before, several persistent type scheduling algorithms (such as fully persistent, talk-spurt based persistent and semi-persistent scheduling) have been proposed in 3GPP. These scheduling mechanisms limit or even lack entirely the gain from multi-user and frequency domain scheduling, but work around a difficult problem of the PDCCH capacity restricting the overall VoIP capacity. However, also dynamic PS may be improved for improving the VoIP capacity with control channel restrictions. With packet bundling the eNb may decide to bundle one or more VoIP packets into one L1 PDU improving the spectral efficiency together with LA due to better resource utilization.

E. Handovers and mobility

E-UTRAN utilizes a UE assisted hard handover algorithm for mobility: UE measures downlink signal quality and sends the measurement reports to eNB either periodically or when an event triggers.(such as another eNB becoming stronger than tha current eNB). The eNB then makes the final handover decisions based on the received measurement reports. Typically, measurement averaging, handover margins and timers are used in order to avoid excess or ping-pong handovers.

During a handover the old serving eNB flushes HARQ Stop-and-Wait (SAW) buffers, which means that VoIP packets still waiting for a retransmission will be discarded permanently. Also, a UE cannot be scheduled while the handover is in progress, which may lead to additional delays for PDUs. After a connection to the new eNB is established both HARQ and PS processed continue normally.

2503

III. SIMULATION ASSUMPTIONS AND MODELING

A. System simulator description

We have used a fully dynamic system simulator for studying the VoIP performance. Both E-UTRAN downlink and uplink are simulated with TTI (1 ms) resolution. Simulator contains detailed modeling of RRM, mobility and handovers as well as traffic models. Exponential Effective SINR Mapping (EESM) interface is used as link-to-system interface [10].

B. Scenario setup and related modeling

The VoIP capacity evaluation is based on the UTRAN LTE downlink parameters and assumptions described in [1]. All the simulation cases were run in a three tier diamond-pattern macro scenario with 19 3-sector sites, i.e. a total amount of 57 cells. Users are uniformly dropped and move within the 21 cells in the middle. The 26 cells at the edge of the scenario are just generating interference at the same magnitude as the average load in the center cells. The VoIP capacities are presented in all 3GPP defined macro simulation cases shown in TABLE I [1]. Note, that Cases 1, 2 and 3 are modified to have only 5 MHz bandwidth.

TABLE I. 3GPP SIMULATION CASE DEFINITIONS.

Case CF (GHz)

ISD (m)

BW (MHz)

PLoss (dB)

Speed (kmph)

1 2.0 500 5.0 20 3 2 2.0 500 5.0 10 30 3 2.0 1732 5.0 20 3 4 0.9 1000 1.25 10 3

A set of common parameters for the simulations is presented in TABLE II. We utilize a de-coupled Time Domain (TD) and Frequency Domain (FD) packet scheduler presented e.g. in [11]. We utilize Round Robin (RR) in the Time Domain and Even Resources (ER) in the Frequency Domain. RR chooses users with longest time since last scheduling time instant for FD-PS scheduling candidates. ER first sorts the candidate users based on the buffering delay. Then, for each user in turn, the PRBs are sorted according to user experienced CQI and each user is allocated enough PRBs to be able to transmit a VoIP packet, or more if the PS decides to bundle more than one packet. LA tries to maximize the spectral efficiency by choosing a best Modulation and Coding Set (MCS) for a scheduled user based on instantaneous radio channel conditions.

VoIP AMR 12.2 traffic model is modeled with both active and silence periods. Packets are modeled to include Real-time Transport Protocol (RTP), Robust Header Compression (ROHC), Packet Data Convergence Protocol (PDCP), RLC and MAC headers in the total packet size. The VoIP traffic model parameters have been presented in TABLE III.

TABLE IV shows the parameters varied in the simulations.

C. Simulation cases

The VoIP capacity depends on several different features, such as:

1. Bandwidth: The used system bandwidth determines the total amount of frequency domain resources

2. Link adaptation: With LA each TB can be optimized in terms of spectral efficiency and BLER.

3. Delay threshold: Since VoIP is delay-critical service determined by the delay threshold.

4. Number of control channels: The number of maximum schedulable users in a TTI depends on the control channel capacity.

5. Packet bundling: The amount of VoIP packets bundled per UE L1 PDU may improve the resource utilization, especially with control channel limitations.

6. HARQ processes: LTE requires small round trip times, which is provided by fast L1 retransmissions by HARQ.

TABLE II. COMMON PARAMETERS.

Parameter description Parameter value Scenario / network / direction 57 cells, Synchronous

reuse 1 network, DL

UE velocity 3 km/h UE receiver type MRC 1x2 Channel model TU 20 Simulation length 1M steps = 72

seconds Symbols per subframe 14 (with 4

control symbols)

Subframe length (TTI) 1 ms Carriers per PRB 12 Duplexing FDD Power control Off HARQ mode Asynchronous, with

Chase combining HARQ max retransmissions 3 ARQ Off CQI measurement interval 5 ms CQI reporting delay 2 ms CQI reporting resolution 2 PRBs CQI error variance 1 dB Initial MCS (LA off) QPSK 2/3 Possible MCSs (LA on) QPSK 1/3, ½, 2/3

16QAM ½, 2/3, 4/5 64QAM ½, 2/3, 4/5

LA Outer Loop LA BLER target 0.2

TD packet scheduler Round Robin FD packet scheduler Even Resources Segmentation Off Hard handover margin 3 dB Hard handover sliding window size 200 ms

2504

TABLE III. VOIP AMR 12.2 PARAMETERS.

Parameter description Parameter value VoIP packet 38 bytes / 20 ms SID packet 14 bytes / 160 ms Voice Activity Factor 50 % Call length Neg.exp. distr, mean 20 s Talk spurt Neg.exp. distr. mean 2 s

TABLE IV. VARIED SIMULATION PARAMETERS.

Parameter description Parameter values Simulation case 1, 2, 3, 4 Delay threshold 40, 50, 60, 80, 100 ms Number of control channels 6, 8, 10 Packet bundling On, off Link adaptation On, off HARQ processes 7, 8, 9

IV. SIMULATION RESULTS

A. VoIP capacity in different 3GPP cases

The VoIP capacities in four different 3GPP cases are shown in Figure 1. The capacity for Case 1 and Case 3 is about 300 UEs/cell with LA, showing that the larger ISD in Case 3 does not lower the VoIP capacity. This indicates simply that the capacity in Case 3 is limited by other factors than transmission power and noise.

When the UE velocity is increased to 30 km/h in Case 2, the VoIP capacity drops by about 42% with LA and by about 35% without LA. There are three main factors contributing to this capacity loss. First, the FD-PS performance is worse due to less accurate CQI information – the PRB allocation becomes more random. Second, the bad CQI affects also LA and less optimal MCS is selected. Third, HO performance is slightly worse with higher speed as the UE is connected to a non-optimal cell more often. Note that the second point is valid only with LA, which explains why the capacity loss without LA is less than with LA.

Figure 1. VoIP capacity in different 3GPP cases.

LA provides about 44% to 78% gain over static MCS of

QPSK 2/3 depending on the 3GPP simulation case. This is because with LA a VoIP packet might fit in fewer PRBs for UEs with good radio conditions, thus more VoIP packets can be sent each TTI in average. On the other hand, UEs with bad radio channel conditions can utilize more robust MCS improving BLER of the transmissions.

When the system bandwidth is decreased by a factor of four (i.e. case 4 with 1.25 MHz bandwidth) the VoIP capacity is decreased by a factor of five. This is due to better packet bundling gain with higher bandwidths. In 3GPP Case 4 LA provides less gain, because with 1.25 MHz bandwidth the system is not control channel limited and packet bundling does not provide any gain.

B. Effect of number of control channels and packet bundling

With 1.25 MHz bandwidth (Case 4) the PDCCH capacity is not limiting the VoIP capacity due to low traffic channel capacity. However, with higher system bandwidths the PDCCH capacity might become a limiting factor, at least if dynamic scheduling is used.

According to Figure 2, with 5 MHz bandwidth and without LA, the number of control channels clearly limits the VoIP capacity without packet bundling. The capacity gain over 6 control channels is about 33% and 80% with 8 and 10 control channels, respectively. On the other hand, with LA the gains are 20% with 8 control channels and 21% with 10 control channels. Smaller the PDCCH capacity, the more relative gain we get from joint packet bundling and LA. With 6 control channels the capacity gain with LA and packet bundling is 106%, with 8 control channels 86% and 10 control channels only about 39%. In Figure 3 the PDU size distribution with 250 UEs per cell is shown. It can be seen that with 6 control channels much more VoIP packets are bundled (PDU size is 2* 38 bytes = 76 bytes) than with 8 or 10 control channels.

Figure 2. Effect of PDCCH capacity, packet bundling

and link adaptation on VoIP capacity.

2505

C. Effect of delay bound

VoIP service is strictly delay critical and the radio interface delay threshold is specified to be 50 ms in 3GPP. However, as can be seen in Figure 4, the delay threshold has only a little impact on the VoIP capacity. Only the 40 ms delay threshold affects the VoIP capacity and with other delay thresholds the effect is only visible after the capacity point in the UE satisfaction curve. This is because the 95% satisfaction rate is reached at a point where the unsatisfied users are unsatisfied due to packet losses, so regardless of the delay bound, the same users are unsatisfied.

Figure 3. L1 PDU size distribution with 250 UEs/cell.

Figure 4. Effect of different delay thresholds.

D. Effect of HARQ processes

The number of HARQ SAW channels for DL is being defined at 3GPP. We simulated the VoIP capacity with 7-9 SAW channels and according to the results the number of SAW channels does not seem to have any effect to the results. This suggests that 7 SAW channel is already enough for VoIP, as expected: the delay bound of 50 ms means that at most 2-3 packets should be waiting retransmission.

V. CONCLUSION

We have presented the basic VoIP downlink capacity results in different 3GPP simulation cases. Also, we have studied the effect of multiple features on VoIP capacity, such as the effect of delay threshold, packet bundling, control channel capacity and number of HARQ processes on VoIP capacity.

The main conclusions from the results are that • VoIP downlink capacity is maximally about 60 UEs per

cell with 1.25 MHz system bandwidth and about 300 UEs per cell with 5 MHz.

• Link adaptation together with packet bundling provides about 44-78% gain over the static MCS of QPSK 2/3 depending a little from the simulated case,

• In general the higher the system bandwidth the higher the gain is from packet bundling and link adaptation,

• VoIP capacity is clearly control channel limited but packet bundling can quite effectively compensate the limitations.

Future work includes studying the VoIP capacity with persistent scheduling algorithms and dynamic PDCCH. Also more realistic mixed traffic scenarios are to be studied.

REFERENCES

[1] “Physical Layer Aspects for Evolved UTRA”, 3GPP Technical Report 25.814, version 7.1.0, September 2006.

[2] H. Ekström, A. Furuskär, J. Karlsson, M. Meyer, S. Parkvall, J. Torsner and M. Wahlqvist, “Technical Solutions for the 3G Long-Term Evolution”, IEEE Communications Magazine, Vol. 44, No. 3, pp. 38-45, March 2006.

[3] D. Jiang, H. Wang, E. Malkamäki and E. Tuomaala, ”Principle and Performance of Semi-Persistent Scheduling for VoIP in LTE System”, in Proceedings of the International Conference on Wireless Communications, Networking and Mobile Computing (WiCom'07), pp. 2861-2864, September 2007.

[4] F. Persson, “Voice over IP Realized for the 3GPP Long Term Evolution”, in Proceedings of the 66th IEEE Vehicular Technology Conference (VTC’F07), September 2007.

[5] S. Choi, K. Jun, Y. Shin, S. Kang and B. Choi, “MAC Scheduling Scheme for VoIP Traffic Service in 3G LTE”, in Proceedings of the 66th IEEE Vehicular Technology Conference (VTC’F07), September 2007.

[6] "Requirements for Evolved UTRA (E-UTRA) and Evolved UTRAN (E-UTRAN)", 3GPP Technical Report 25.813, version 7.3.0

[7] R1-070674 Orange, China Mobile, KPN, NTT DoCoMo, Sprint, T-Mobile, Vodafone, Telecom Italia “Physical Layer Framework for Performance Verification”

[8] “One Way Transmission time”, ITU-T recommendation G.114 [9] A. Pokhariyal, T. E. Kolding and P. E. Mogensen, “Performance of

Downlink Frequency Domain Packet Scheduling for the UTRAN Long Term Evolution”, Proceedings of the 17th Annual IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’06), September 2006.

[10] K. Brueninghaus, D. Astely, T. Salzer, S. Visuri, A. Alexiou, S. Karger, and G.-A. Seraji, “Link Performance Models for System Level Simulations of Broadband Radio Access Systems,” in Proceedings of the Personal, Indoor and Mobile Radio Communications (PIMRC’05), vol. 4, September 2005, pp. 2306–2311.

[11] P. Kela, J. Puttonen, N. Kolehmainen, T. Ristaniemi, T. Henttonen, and M. Moisio, “Dynamic Packet Scheduling Performance in UTRA Long Term Evolution Downlink,” in Proceedings of the International Symposium on Wireless Pervasive Computing (ISWPC’08), May 2008, to be published.

2506