LTE coverage improvement by TTI bundling

Transcription

1 LTE coverage improvement by TTI bundling Riikka Susitaival Ericsson Research, NomadicLab, Finland Michael Meyer Ericsson Research, Aachen, Germany Abstract Compared to WCDMA, the LTE radio access has a significantly shorter Transmission Time Interval (TTI) in order to reduce end-to-end delays. However, if a User Equipment (UE) at the cell edge is limited by its available transmission power, it may not be able to transmit an entire VoIP packet during one TTI, since the instantaneous source data rate is too high. Thus TTI bundling has been recently introduced as a feature of LTE Rel. 8 to improve the uplink coverage. In TTI bundling, a VoIP packet is transmitted as a single PDU during a bundle of subsequent TTIs without waiting for the HARQ feedback. HARQ feedback is only expected for the last transmission of the bundle. This paper studies TTI bundling and compares it to the conventional RLC segmentation. The simulation results indicate that TTI bundling provides a gain of more than 4 db in terms of the sustainable path loss. Keywords- LTE, VoIP, coverage, HARQ I. INTRODUCTION The Long Term Evolution (LTE) radio access has been standardized by 3GPP. The requirement has been to provide high data rates for end users with low latency and good QoS [1][2]. The key facilitator for low delays has been the introduction of a short TTI of only 1 ms, which enables short Hybrid Automatic Repeat Request (HARQ) round trip times (RTT) of 8 ms. However, having such a short TTI can be problematic when the user is at the cell border and has limited transmission power: Even for a small Voice Over IP (VoIP) packet it cannot spend enough energy during one TTI in order to achieve a successful (with a certain probability) transmission. For example, transmitting a VoIP packet of 33 bytes within 1 ms results (including L1/L2 overhead) in a data rate of 312 kbit/s which might be difficult to sustain at the cell edge. Thus, it is likely that HARQ retransmissions are required inducing additional 8 ms delay per retransmission. Many retransmissions lead to long delays, which can be intolerable as delay sensitive applications as VoIP are considered. The conventional solution to reduce delays and improve the coverage is to segment RLC SDUs and transmit those segments in consecutive TTIs. However, segmentation increases overhead, control signaling and vulnerability to packet loss due to HARQ feedback errors. An alternative solution to segmentation is to bundle a set of TTIs for a certain UE. The idea is that a fixed number of transmissions of a given transport block is done in consecutive TTIs without waiting for the HARQ feedback. Only when the receiver has decoded the whole bundle of transmissions, it sends the corresponding HARQ feedback. This approach reduces the number of required HARQ feedback messages significantly. In addition, the header overhead as well as the signaling overhead resulting from uplink grants is reduced. In this paper we first discuss the problems of the two traditional concepts: solely relying on HARQ retransmissions or using RLC segmentation. Then we introduce TTI bundling as an approach to improve the coverage of the LTE radio access. The TTI bundling method in uplink was originally presented to 3GPP in [3] and has in the meanwhile been included as a feature of LTE Rel. 8. In this paper the concept is evaluated in more detail. By dynamic radio network simulations we compare the performance of the TTI bundling and segmentation approaches for VoIP traffic. The VoIP capacity of the LTE uplink is earlier studied in many papers, such as [6][7], but the coverage issue discussed in this paper has received little attention. The discussion in this paper concentrates on the LTE FDD operation, while TTI bundling has been defined for TDD operation as well. This paper is organized as follows: In Section II we discuss the two conventional approaches based on HARQ retransmissions and segmentation. Also the limitations of those concepts for UEs at the cell edge are provided. As a solution to improve the coverage we describe TTI bundling in Section III. In Section IV we study different scheduling strategies of the bundled users. Section V compares the performance of segmentation and bundling by simulations. Finally, Section VI provides a short conclusion. II. SOLUTIONS TO IMPROVE UPLINK COVERAGE As mentioned in the introduction, during the TTI of 1 ms the UE may not be able to accumulate enough energy so that even a small VoIP packet can be received correctly by the evolved NodeB (enb). The conventional solutions to the problem are HARQ retransmissions or segmentation of higher layer Protocol Data Units (PDUs) into smaller units or a combination of these two concepts. These approaches and drawbacks of them are discussed next. A. HARQ Retransmissions HARQ is an error correction mechanism and applied at the LTE Medium Access Control (MAC) layer. The idea is that each user has multiple parallel stop-and-wait processes: When waiting for an ACK/NACK feedback of one process, other transport blocks are sent using other processes. By this approach the throughput of the link is maximized. In addition, only a single bit HARQ feedback signal is needed instead of feedback messages containing sequence numbers as in case of a window-based protocol /09/$ IEEE 1

2 The number of parallel HARQ processes depends on the HARQ RTT. Taking transmission, reception, and processing delays into account, it can be calculated that the retransmission of the transport block is possible 8 ms after the previous transmission. Thus, the number of parallel HARQ processes is also set to 8 in LTE. In the uplink direction of LTE, synchronous HARQ is used, meaning that if there is a retransmission of the transport block, it occurs exactly 8 ms after the previous transmission. segmented into 4 RLC PDUs and transmitted in parallel in four different HARQ processes. By this approach the delay is shorter than when having only pure HARQ retransmissions as depicted in the previous subsection. Figure 2. Segmenting an RLC SDU into RLC PDUs and transmission of corresponding transport blocks with hybrid ARQ. Figure 1. Transmitting an RLC SDU as a single RLC PDU and corresponding transport block with hybrid ARQ. In Figure 1 the transmission of a VoIP packet as a single RLC PDU is illustrated. The first transmission of the transport block resulting from the considered VoIP packet occurs at TTI # 0. The HARQ feedback occurs 4 ms and the retransmission 8 ms after the initial transmission. In the example three HARQ transmissions are required before the decoding of the transport block is successful. If the UE is at the edge of the cell, the number of retransmissions can be even higher leading to too long delays as regard to VoIP traffic. The QoS requirements for VoIP traffic over a one-way radio link can be assumed to be the followings: The delay should not exceed 50 ms and residual packet error rate should be under 1% [8]. B. RLC segmentation The LTE Radio Link Control (RLC) protocol is responsible for segmentation and concatenation of higher layer PDUs, i.e., Packet Data Convergence Protocol (PDCP) PDUs, to suitably sized RLC PDUs. It is also responsible for in-sequence delivery of RLC SDUs to upper layers. There are different operation modes in RLC; in Acknowledged Mode (AM) retransmissions of missing PDUs are done by a second ARQ protocol in RLC, but in Unacknowledged Mode (UM) only insequence delivery is provided without retransmissions. The latter approach is suitable for VoIP traffic where short delays are more stringent compared to the packet loss requirement. A straightforward concept to improve the LTE uplink coverage is to segment RLC SDUs into several smaller units. Due to stronger coding, it is more likely that smaller transport blocks can be decoded correctly. Thus, retransmissions are required less frequently, resulting in less delay. As an example, in Figure 2 the RLC SDU resulting from a VoIP packet is However, the segmentation approach has some drawbacks especially when used in poor radio conditions: 1. Overhead: Let us assume a typical VoIP packet including 36 bytes for the voice frame and the compressed RTP/UDP/IP header [5]. The combined RLC/MAC header is assumed to be 3 bytes and the CRC overhead also 3 bytes [2]. As shown in Table 2, if the number of RLC PDUs per VoIP packet is increased from 1 to 8, the L2/L1 overhead increases from 14% to 55%. 2. Control signalling load: Each transmitted RLC PDU requires a separate control message carried on the Physical Data Control Channel (PDCCH) to grant the uplink resources. In addition, possible HARQ retransmissions need to be granted, too. A high number of segments as well as retransmissions lead to high load on the PDCCH. 3. HARQ feedback errors: Each HARQ transmission is followed by a downlink HARQ feedback signal. Assuming RLC UM, a VoIP packet loss can occur when a HARQ NACK is interpreted as an ACK. This may result in a significant packet loss rate. For example, if 12 (re- )transmissions for a transport block are required and the NACK-to-ACK error probability is 0.1%, the probability to stop retransmissions too early due to a HARQ feedback error is 1-( ) %, which is too high for VoIP. TABLE I. SIZE OF VOIP PACKETS WITH DIFFERENT SEGMENT SIZES Segments Transport block size CRC bits All bytes Amount of overhead % % % % 2

3 III. TTI BUNDLING An alternative approach to pure HARQ retransmissions or RLC segmentation is a method named TTI Bundling. Instead of segmenting the RLC SDU into smaller units at the RLC layer, in this approach several redundancy versions (RVs) corresponding to the entire RLC SDU are transmitted in consecutive TTIs. Only when the last redundancy version of the transport block is received by the enb, the HARQ feedback is sent. the UE should transmit a single TTI or a TTI bundle, is configured by the Radio Resource Control (RRC) protocol. A UE can be configured to use either normal HARQ operation or TTI bundling. 2. There is only one HARQ feedback message per bundle. This decreases control signaling and the vulnerability to NACKto-ACK errors that can lead to data loss. 3. Because RLC SDUs are not segmented, the overhead resulting from RLC and MAC headers as well as CRC is decreased leading to a more efficient use of radio resources. A side effect of TTI bundling is that the granularity to match the number of transmissions exactly to required amount (i.e., the early termination gain) decreases. With segmentation, when a single segment is decoded, retransmissions of that segment can be stopped. For TTI bundling always a bundle of retransmissions is done. For that reason TTI bundling should be used only for the UEs that are at the cell edge. IV. SCHEDULING OF BUNDLES Figure 3. TTI Bundling of 4 TTIs Figure 3 shows the TTI bundling solution adopted in 3GPP Specifications [4] and [5] for FDD. In the standard, the bundle size has been fixed to 4 transmissions. That means, 4 redundancy versions of the transport block resulting from a single RLC SDU are transmitted in consecutive TTIs with HARQ process number 0. After all 4 transmissions have been received and decoded by the enb, the HARQ feedback is sent. Assuming 1 ms delay for transmission and 3 ms delay for decoding and processing, this feedback is possible in TTI # 7. After the UE has received the feedback, assuming again 3 ms for processing, the earliest retransmission by the UE is possible in TTI # 11. Thus, the shortest HARQ RTT with the bundle size of 4 would thus be 11 ms. However, the HARQ RTT of 11 ms cannot be synchronized with the 8 ms normal LTE HARQ RTT (compare Figure 1), because the retransmissions of the normal processes and the bundled processes using the same physical resource blocks would collide when non-adaptive retransmissions are applied. For that reason, in 3GPP it has been agreed that the retransmission of the bundle is delayed until the 16 th TTI. The time slots in between can be used for some other bundled HARQ processes from the given UE or other UEs using TTI bundling. Also part of TTIs can be used for normal HARQ processes of other UEs as will be explained in Section IV. TTI bundling has several advantages compared to the RLC segmentation approach: 1. Uplink resources over multiple TTIs can be assigned with a single grant, which decreases the signaling overhead. To trigger a transmission of a TTI bundle, the same grant format can be used as for ordinary HARQ transmissions. Whether In normal FDD operation of LTE, allocation and granting of physical resources are done once per TTI. When TTI bundling is configured for a given user, the transmission grant is valid during the whole bundle of TTIs. For this reason, every time when the transmission of a new transport block is scheduled, it has to be checked that any transmission of the new bundle does not collide with the retransmissions of the existing bundle. Considering a single user, it is obvious that transmission of only one transport block per TTI is possible and collision occurs if the retransmissions of an old process occur at the same time that the transmission of a new bundle. This is illustrated in Figure 4: A new bundle of the considered user can not be scheduled to start from TTI #13 to TTI #15, because the transmission of the new bundle would collide with the retransmission of the existing bundle 1. When there are multiple users in the system, a collision can occur if different users are allocated to the same frequencies (i.e., subbands) at the same time. Again in Figure 4, another bundled user cannot be scheduled from TTI #13 to TTI #15 to the same frequencies as the existing process already uses. Similarly, another UE using the normal HARQ operation with 8 ms HARQ RTT cannot be scheduled during TTI #8 - TTI #11 on the frequencies of bundle 1 because the transmissions and retransmissions of these two UEs would collide. Despite of the above mentioned restrictions, there still remains some flexibility in scheduling in the time domain of a single user or in scheduling in the time and frequency domain of all users. Here two of approaches are explained: 3

4 Figure 4. Scheduling of the resource blocks for a single or many users Synchronous scheduling. In this strategy the transmission times of all bundles are synchronized. As depicted by dotted blocks in Figure 4 this would mean that a new bundle could be scheduled to start every 4 th TTI, that is, in TTIs #4, #8 or #12 in the figure. The benefit of this approach is that scheduling is more straightforward and we can fill the time domain with bundles if the traffic load of the network is high. Non-synchronous scheduling: In this strategy we can schedule new bundles or normal processes whenever there will not be any collision as explained below. As a result, the bundled processes and normal processes are arbitrary mixed. This approach decreases the access delay but efficient allocation of resources is more complicated as compared to the synchronous strategy. V. SIMULATION RESULTS In this section we compare the performance of pure HARQ retransmissions, segmentation and TTI bundling for VoIP traffic at the cell edge. The simulations are done by a dynamic network simulator developed by Ericsson. In the simulation scenario all generated users are located at the cell border and have a constant path loss. On top of the fixed path loss a fast fading model reflecting an ITU3 channel is used. We assume a simple VoIP traffic model where only uplink traffic from a UE to the enb is generated. Our assumption is that the client in the terminal side talks all time during its call and SID frames are ignored. The transport block sizes used for VoIP packets with different segmentation approaches are given in Table 1. The QoS requirements for the VoIP application are set as follows: the user is satisfied if the delay does not exceed 60 ms and the residual packet loss is below 1 %. All packets exceeding the delay bound are considered as lost. There are some simplifications used in the simulation study. First, scheduling requests are not used. Instead, it is assumed that the enb knows perfectly the buffer status of the UE. In addition, the scheduling is kept as simple as possible: If there is data available, the number of allocated subbands, i.e. 15 khz sub-carriers, is fixed to 3. Although in the LTE standard the bundle size is fixed to 4, in this section we study also alternative options such as bundle size of 8. However, the bundle size does not change dynamically but remains fixed in each simulation scenario. A. User satisfaction with a perfect feedback channel In Figure 5 the share of satisfied VoIP users is plotted as a function of the normalized path loss for the different segmentation and bundling approaches. In the scenario there is a single user at a time in the system but the results are averaged over 200 users. We can see that the sustainable path loss difference between pure HARQ transmissions and other approaches is significant, 2-3 db. When segmentation of a VoIP packet into 4 and 8 segments are compared, these two approaches give almost the same results. This is explained by the fact that the benefit from the delay reduction when using 8 segments is lost due to an increase in overhead. When 4 segments and bundling over 4 ms are compared, also in this case the performance is similar. An explanation for this is that due to tight delay requirements for VoIP traffic, the maximum number of retransmissions without bundling remains as 8. If a VoIP packet arrives every 20th ms, 12 free TTIs still exist before the UE is power-limited. Those TTIs can be used to increase the code rate by segmentation. Finally, when transmissions are bundled over 8 ms, the gain of 1 db is achieved when compared to other approaches. Figure 5. User satisfaction as a function of the normalized average path loss, error-free PHICH. B. User satisfaction with NACK-ACK errors In LTE, HARQ feedback messages of uplink data traffic are carried in the Physical Hybrid-ARQ Indicator Channel (PHICH). The 3GPP target for the probability that an ACK is 4

5 misinterpreted as NACK is 10-2 and that a NACK is misinterpreted as ACK is in the range of 10-3 to In practice, if there is an ACK-to-NACK error, the UE sends some unnecessary retransmissions, which consumes some extra resources. However, a NACK-to-ACK error is more critical, since the HARQ transmitter terminates the (re-)transmissions. In the case of RLC UM the whole VoIP packet will be lost. One option to recover from NACK-to-ACK error is that the enb tries to detect if the UE has not transmitted anything. If not, the enb invokes a retransmission. However, this kind of recovery mechanism brings additional complexity to the enb and the scheduler and is not assumed in this paper. Figure 7. Number of used resource blocks per VoIP packet, error-free PHICH. Figure 6. User satisfaction as a function of the normalized average path loss of the link, PHICH NACK-to-ACK error probability 0.1%. In Figure 6 the user satisfaction is plotted when the ACKto-NACK error probability is set to 0.1%. We can see that the user satisfaction is much lower than with the perfect feedback channel. When TTI bundling is used, the overhead and the number of HARQ feedback messages are independent of the bundle size. Thus, with bundling over 4 ms we can achieve 4 db and bundling over 8 ms even 6 db improvement in coverage as compared to segmentation. C. Resource usage with different approaches Although the uplink coverage improvement has been the main objective when developing the TTI bundling concept, it is also important to minimize the used resources. Thus, in Figure 7 the average number of used resource blocks per transmitted VoIP packet as a function of the path loss is shown. When the link quality is reasonably good, segmentation and bundling over 4 or 8 TTIs use the same amount of resources. However, when the path loss increases, the resource usage of TTI bundling is remarkably lower, which can mainly be explained by the smaller overhead. On the other hand, the resource usage of the bundle over 4 ms is significantly smaller than that of the bundle over 8 ms providing thus a good tradeoff between the coverage improvement and the capacity usage. VI. CONCLUSIONS In this paper we discussed and evaluated the LTE TTI bundling feature that aims at improving the uplink coverage. The idea is that, instead of segmentation, bigger transport blocks are used. By relying on incremental redundancy, HARQ transmissions are performed in consecutive TTIs without waiting for HARQ feedback. The HARQ receiver accumulates the received energy of all transmissions and responds with HARQ feedback only once after the entire bundle has been received and evaluated. The simulation results indicate that there is only a small benefit of TTI bundling for VoIP traffic if the HARQ feedback channel is perfect. The gain achieved by decreasing the overhead of segmentation is lost due to a longer delay of TTI bundling. However, when the NACK-to-ACK errors are taken into account, the bundling over 4 ms gives more than 4 db better results in terms of the sustainable path loss with reasonable resource usage. This encourages the use of TTI bundling. REFERENCES [1] H. Ekström, et al., Technical solutions for the 3G long-term evolution. IEEE Communications Magazine, vol. 44, no. 3, pp , March [2] 3GPP TS , Evolved Universal Terrestrial Radio Access (E- UTRA) and Evolved Universal Terrestrial Radio Access Network (E- UTRAN); Overall description; Stage 2. [3] 3GPP TDoc R , HARQ operation in case of UL Power Limitation, Ericsson, June [4] 3GPP TS , Evolved Universal Terrestrial Radio Access (E- UTRA); Medium Access Control (MAC) protocol specification, Rel- 8, V [5] 3GPP TS , Evolved Universal Terrestrial Radio Access (E- UTRA); Physical layer procedures, Rel-8, V [6] D. Jiang, H. Wang, and X. Che, Uplink VoIP Performance in E-UTRAN TDD Mode, IEEE VTC Spring, [7] Y. Kim, An efficient Scheduling Scheme to Enhance the capacity of VoIP service in Evolved UTRA uplink, EURASIP Journal on Wireless Communications and Networking, Vol [8] 3GPP TS , Technical Specification Group Services and System Aspects, Quality of Service (QoS) concept and architecture, Rel 8. 5