Dynamic Two-Threshold Flow Control Scheme for 3GPP LTE-A Relay Networks Ping-Chen Lin and Ray-Guang Cheng Department of Electronic and Computer Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan, R. O. C. crg@mail.ntust.edu.tw Abstract Relaying technology is a cost-effective solution for achieving throughput enhancement for cell-edge users or extending the cell coverage for the 3GPP LTE-A systems. In this study, we investigate the flow control scheme in the LTE-A relay networks employing the Type-I relay nodes (RNs). However, the buffer-overflow problem [7, 8] caused by high-speed arrival traffic and the handover forwarding problem caused by UE s handover are two main issues that affect the performance of 3GPP LTE-A relay networks. To mitigate these problems, we present a dynamic two-threshold flow control scheme () which can dynamically adust the upper buffer-threshold of RN based on the channel quality. Results showed that can reduce 40% of the total number of forwarding packets during the handover when compared with the two-threshold flow control scheme. Additionally, when the proposed scheme is compared to single-threshold flow control scheme, excels by imposing much less signaling overhead on the expense of a slight increase in the number of forwarding packets. Keywords-dynamic two-threshold flow control scheme (); handover; type-i relay node (RN) I. INTRODUCTION Relaying is one of low-cost solutions for extending the cell coverage or enhancing throughput for cell-edge users in networks without passing through the wired backhaul. In Long Term Evolution-Advanced (LTE-A) relay networks, two types of relay nodes (RNs) have been specified into the third generation partnership proect (3GPP) standard [1]-[3]. The first one is called a Type-I RN and the other one is referred to as Type-II RN. Type-I RN is considered as a small base station for the user equipment (UE) because it has its own cell identity (ID) and broadcast information. It can also perform its radio resource management (RRM) policy based on its own common reference signal, synchronization channels, and scheduling information. Additionally, due to its own broadcast information, Type-I RN can also provide the coverage extension for LTE-A systems. On the other hand, Type-II RN does not have its cell ID and RRM information. It can only forward the broadcast information from a base station, which is named a donor enb (DeNB) in 3GPP. A Type-II RN is transparent to UE, since UE is unaware of its presence. In this scenario, a UE cannot hear any broadcast information when it is out of DeNB s coverage even though it is under the coverage of a Type-II RN. Accordingly, Type-II RN cannot support coverage extension for LTE-A relay networks. However, a number of studies [4]- [6] use Type-II RNs to achieve throughput enhancement via cooperated multi-point (CoMP) transmission. To extend the coverage of relay networks, we study the performance of relay networks with Type-I RN. The rest of this paper is organized as follows: In Section II, the problem statement in relay networks with Type-I RN and literature review for different flow control schemes are addressed. The proposed dynamic two-threshold flow control scheme () is then elaborated in Section III. Section IV shows the simulation results. Conclusions are finally drawn in Section V. II. PROBLEM STATEMENT AND LITERATURE REVIEW As shown in Fig. 1, there are N fixed Type-I RNs and M mobile UEs deployed in a LTE-A relay network. For notation convenience, we use RN instead of Type-I RN. In such networks, the maximum transmission hop-count is specified in 3GPP standard [3] and is limited to two in order to reduce the implementation complexity. According to the cell selection policy, a UE (1 M) can directly be served by a DeNB or served by a RN i (1 i N) passing through two independent radio links. One radio link connecting the DeNB and the RN is referred to as Un interface. The other radio link between the DeNB and UE or between the RN and UE is named the Uu interface. However, due to the two-hop transmission, some critical problems may emerge in such networks to degrade the system performance. In the following we elaborate these problems. Figure 1. System model for LTE-A relay networks. A. Buffer-overflow Problem In LTE-A relay network, the network operator will deploy a number of fixed-location RNs with light-of-sight (LOS) propagation between the RN and the DeNB, so that the channel quality of Un interface is better than Uu interface. Thus, in the downlink transmission of relay networks, the buffer-overflow will occur at RN due to high link rate of Un interface and 978-1-4577-1348-4/13/$26.00 978-1-4577-1348-4/13/$31.00 2013 IEEE 2664
results in severe packet loss. This problem is called bufferoverflow problem. B. Handover Forwarding Problem In LTE-A relay networks, cells (e.g., DeNB or RN) can adopt the hybrid automatic repeat request (ARQ) protocol to retransmit the corrupted packets to guarantee their correctness. After receiving the hybrid ARQ acknowledgement from RN, the DeNB will remove the packets form its buffer and only keep them at RN s buffer. The handover (or handoff) procedure is also necessary to prevent the service interruption due to UE s mobility. However, different to the traditional handover between two DeNBs, no physical (wired) link can directly connect a RN with a DeNB. Thus, all packets buffered at serving RN should be forwarded to target cell via a wireless link (i.e., Un interface). The uplink bandwidth will be occupied by forwarding packets so that the system performance will be degraded, this is referred to as the handover forwarding problem in LTE-A relay networks. C. Literature Review A number of research work has recently been conducted for flow control schemes to deal with the buffer-overflow problem. In [7], the credit-based flow control scheme was adopted to alleviate the buffer-overflow problem in multihop cellular network. In [8], the authors further indicated that the bufferoverflow problem does not merely occur at RN. This problem also emerges at DeNB due to the high-speed wired backhaul. Thus, the flow control scheme should also be extended to the wired link between the DeNB and the wired backhaul. Jagannathan et al. [9] investigated the performance of threshold-based flow control schemes to relieve network congestion. They presented an analytical model to study the trade-off between the congestion probability and the rate of control (or feedback) message. Based on their analytical model, they show that the larger the gap between two bufferthresholds the lower the rate of control message. However, the aforementioned flow control schemes did not consider the handover forwarding problem, so the system performance still degrades when UE performs the handover. The handover forwarding problem can be ignored when the service type of a packet is a real-time (e.g., voice, video) service. This is because the real-time packets buffered at the serving RN can be reduced via the priority-based scheduler [7] or the active queue management scheme (e.g., random early detection) [9]. Packets are also allowed to be dropped while quality of service (QoS) constraint is not met during handover [10]. However, the packets for non-real-time (e.g., webbrowsing, FTP) service are never allowed to be dropped due to the reliable property. In other words, all packets for non-realtime service should be forwarded to the target cell. Thus, large amount of uplink bandwidth for Un interface is wasted by forwarding the handover traffic. In addition, Lin et al. [11] also considered the handover forwarding problem and proposed employing a smaller buffer to solve the issue. However, a small buffer cannot be considered as a general solution. To the best of our knowledge, so far, there is no comprehensive study to fully address both buffer-overflow and handover forwarding problems in LTE-A relay networks. Without solving all these problems, QoS provisioning in such networks is difficult to achieve. The contribution of this paper lies in that it is the first work to propose an effective flow control scheme, named the dynamic two-threshold flow control scheme (), to solve all these problems. III. DYNAMIC TWO-THRESHOLD FLOW CONTROL SCHEME In this section, we will elaborate how to design the. Since is one of the threshold-based flow control schemes, we will first introduce the threshold-based flow control schemes in detail and then explain how to evolve into the. A. Backgrounds of Threshold-based Flow Control Scheme In threshold-based flow control schemes [9], the sender can control its traffic rate via the feedback of the receiver (e.g., DeNB or RN), which is triggered according to the buffer (or queue) state of a receiver. In this section, the well-known twothreshold flow control scheme () and the single-threshold flow control scheme () are addressed. Figure 2. Design concept of. As shown in Fig. 2, there are two distinct thresholds set for the at each queue of the receiver reserved for UE. Let T U and T L be the upper buffer-threshold and the lower bufferthreshold of (unit: packet), respectively. The former is designed to prevent the buffer overflow problem and the latter is designed to guarantee the minimum throughput. Let, and be the ratio of the upper and the lower buffer-thresholds for. Let K i, denotes the total queue length of a receiving cell i (e.g., cell 0 is DeNB, cell 1~N is RN) reserved for UE (unit: packet). We can have T U = K i, and T L = K i,. Note that T L < T U < K i,. It is worth noting that can be downgraded to when T L = T U. Let us denote the arrival rate of threshold-based flow control scheme sent from a cell i for UE by i, (unit: packet/subframe), which is based on the feedback information of the flow control scheme. We assume that the receiver does not have any packets in its queue initially. Let 1 and 2 be the high arrival rate and the low arrival rate of a flow control scheme at the sender (i.e., DeNB or wired backhaul), respectively. Since no buffer-overflow problem occurs in the beginning, the sender is allowed to use the high arrival rate ( i, = 1 ) to transmit large amount of data packets to its receiver. When the queue length at a receiver is larger than T U, it implies that the receiver is in the Congestion State and the 2665
sender should reduce its arrival rate to i, = 2 after it receives the feedback message. When the queue length at a receiver is less than T L, it means that the receiver is in the Non-Congestion State because of less buffered packets at receiver; hence, the sender should increase its arrival rate to i, = 1 to meet its QoS requirement (i.e., minimum throughput) after it receives the feedback message. It is worth noting that T L can be configured by the minimum throughput guarantee of arrival traffic and T U can be set by considering either the signaling cost of feedback messages or the delay constraint of real-time (delay-sensitive) service. In order to reduce the signaling cost of feedback message, T U can be set as a very large value, such as 80% of the total queue size. It implies that the receiver using has a very large gap (i.e., T U T L ) to maintain its state without sending any feedback message. However, under this configuration of, the receiver will keep large amount of data packets in its buffer causing a serve handover forwarding problem. This problem can be alleviated when lower bufferthreshold of is applied. However, it will bring large amount of signaling overhead due to frequent feedback. B. Dynamic Two-Threshold Flow Control Scheme In order to deal with both buffer-overflow and handover forwarding problems, we proposed a dynamic two-threshold flow control scheme () for LTE-A relay networks to dynamically adust the buffer-threshold of RN. Fig. 3 illustrates the design concept of the proposed. Let 1, 2, and be the ratio of the large upper, the small upper, and the lower buffer-thresholds for, respectively. Note that 0 < < 2 < 1 < 1. Let SNR i, and SNR TDFCS be the signal-to-noise ratio of UE served by cell i and threshold of, respectively. Similar to, two buffer-thresholds are employed to control the arrival rate from DeNB. However, while UE has better channel quality (i.e., SNR i, > SNR TDFCS ), as illustrated in the left hand side of Fig. 3, high Uu transmission rate will result in large variation at RN s queue. Hence, a large amount of feedback messages will be sent from RN to DeNB. To prevent this issue, a large gap, constructed by a large upper buffer-threshold (i.e., T U = 1 K i, ) and a lower bufferthreshold (i.e., T L = K i, ), of RN queue is used. Moreover, when UE has poor channel quality (i.e., SNR i, SNR TDFCS ) and is going to perform the handover procedure, RN should not keep large amount of packets in its queue. Consequently, the small upper buffer-threshold (i.e., T U = 2 K i, ) should be applied to prevent serve handover forwarding problem, as illustrated in the right hand side of Fig. 3. Note that the lower buffer-threshold is not changed due to the minimum throughput guarantee and SNR TDFCS is larger than the threshold of handover, SNR HO, since the upper buffer-threshold should be changed before the handover. In the detailed implementation, the algorithm of is based on Algorithm 1. For steps 2 to 6, the upper bufferthresholds are configured herein. When the SNR of UE received at RN i (SNR i, ) is better than configured SNR value (SNR TDFCS ), RN can use the larger upper buffer-threshold for. Otherwise, the smaller upper buffer-threshold is used for. In, the lower buffer-threshold is fixed due to throughput guarantee, as shown in step 7. In steps Figure 3. Design concept of. Algorithm 1 ( Qi,, SNRi, ) Input: the signal to noise ratio SNRi,, the queue length Qi, Output: feedback message fi, {0,1} 1: begin 2: if SNRi, SNR, and i is the serving RN then 3: TU 1 Ki, // larger upper buffer-threshold 4: else 5: TU 2 Ki, // smaller upper buffer-threshold 6: 7: TL Ki, // lower buffer-threshold 8: if Qi, TU then 9: return fi, 1 // feedback fi, and set i, 2 at DeNB 10: 11: else if Qi, TLthen 12: return fi, 0 // feedback fi, and set i, 1 at DeNB 13: 14: 15: end 8 to 14, then decides whether to send the feedback message to DeNB and adust the transmission rate from DeNB to RN. Let f i, be the value of feedback message for UE reported to RN i in. Let Q i, [k] be the queue length of RN i reserved for UE at the subframe k. When the current queue length Q i, [k] is larger than T U, RN will send a feedback message with single-bit value 1 (f i, = 1) to notify that DeNB has entered the Congestion State and should slow down its transmission rate ( i, = 2 ) to avoid the buffer-overflow. When the current queue length Q i, [k] is less than T L, RN will send a feedback message with single-bit value 0 (f i, = 0) to notify that DeNB has got into the Non-Congestion State and then raises its transmission rate ( i, = 1 ) for preventing service interruption due to the lack of packets at RN. Besides these conditions, RN will not report any feedback messages to DeNB to reduce the signaling overhead. 2666
IV. SIMULATION RESULTS In this section, we evaluate the flow control schemes through system level simulation. The simulator of LTE-A relay network was constructed by a C-based platform. In the following figures, each point represented the mean value of 100 simulation samples. Each simulation sample was obtained by averaging data collected within 1,000 seconds (or 100,0000 subframes). We deployed a DeNB at the center of each cell with coverage of 866 meters. Three RNs were located at the edge of each cell. Twenty mobile UEs are randomly dropped in a DeNB s coverage and follow the Random Way Point (RWP) mobility model with speed from 30 to 60 km/hr. The maximum transmission power of a DeNB and a RN are 46 dbm and 30 dbm, respectively. The path-loss models are based on 3GPP TR 36.814 [3]. To ensure the high efficiency of Un interface, DeNB and RN were both equipped with two antenna sets. That is, a 70 directional antenna serving for Un interface and an omni-directional antenna serving for Uu interface. The configuration number 4 of frame structure type 2 (TDD) was chosen for this system [12]. In downlink transmission, we assume the source traffic is generated from wired backhaul and follows the Poisson process with rate i,. The link rate of Un interface for each UE is controlled by the flow control scheme and the link rate of Uu interface for each UE is based on its channel quality (i.e., SNR i,, ), according to the adaptive modulation and coding (AMC) scheme. The parameters for AMC are shown in Table I. The hybrid ARQ protocol and the handover scheme are also employed. The handover threshold SNR HO is 7dB and the handover latency is 60 msec. The total queue length at a cell (e.g., DeNB or RN) reserved for a UE is 100 packets (i.e., K i, = 100). level TABLE I. Modulation and coding AMC PARAMETERS IN LTE-A SYSTEM. Rate (bits/symbol) Required SNR (db) 0 Silent 0 0 1 QPSK (1/2) 1 6 2 16QAM (1/2) 2 11.5 3 16QAM (3/4) 3 15 4 64QAM (2/3) 5 64QAM(5/6) In all simulation results, the and the [9] are chosen as the benchmark to compare with. The,, are all elaborated in Sec. III and implemented at RNs. As mentioned in [8], DeNB only implements the since DeNB doesn t have the handover forwarding problem. The detailed simulation setting for all flow control schemes were configured as in Table II. TABLE II. Description Trigger SNR threshold for SIMULATION SETTING FOR FLOW CONTROL SCHEMES. Value (unit) SNR = 10 db Ratio of buffer-threshold in 1 = 0.8, 2 = 0.4, = 0.2 Ratio of buffer-threshold in = 0.8, = 0.2 Ratio of buffer-threshold in = = 0.2 Arrival rate in non-congestion state Arrival rate in congestion state 1 = 5 packets/subframe 2 = 1 packets/subframe In this paper, four key performance metrics were considered to evaluate the efficiency of the flow control scheme. The first performance metric is the dropping probability. However, we did not show the results of dropping probabilities since all flow control schemes have approximately zero dropping probabilities. The number of forwarding packets, the signaling overhead, and the average spectral efficiency are other key performance metrics, which are illustrated in Figs. 4, 5, and 6, respectively. The number of forwarding packets is defined as the total number of uplink forwarding packets (in packet) during the simulation. The signaling overhead is defined as the average number of feedback messages sent from a RN per second for a flow control scheme (e.g.,,, ). The average spectral efficiency is defined as the average data bits received from a UE in downlink transmission (in bps/hz). # of forwarding packets (packet) signaling overhead (packet/sec) 6000 5500 5000 4500 4000 3500 3000 2500 2000 1500 1000 140 120 100 80 60 40 20 Figure 4. UE speed vs. Number of forwarding packets. 0 Figure 5. UE speed vs. Signaling overhead. Figs. 4 and 5 illustrated the results of the number of forwarding packets and the signaling overhead for three flow control schemes under different UE speed. In Fig. 4, results showed that the number of forwarding packets rises when UE s speed increases. It is because that UE performs the handover more frequently. In this figure, we can find that can 2667
achieve the best performance for uplink transmission since it always keeps the least number of buffered packets at RN (around 20% of the total queue length at the receiver). However, due to the usage of a single buffer-threshold, cannot prevent the frequent feedback. Hence, as shown in Fig. 5, resulted in much higher signaling overhead than the other two schemes. A flow control scheme adopting dual buffer-thresholds (e.g., and ) can largely reduce the signaling overhead. However, results further showed that is more efficient than because it can reduce up to 40% of the total number of forwarding packets and slightly increases the signaling overhead. average apectral efficiency (bps/hz) 4.5 4 3.5 3 2.5 2 1.5 1 0.5 0 Figure 6. UE speed vs. Average spectral efficiency. Fig. 6 showed the results of average spectral efficiency for three flow control schemes under different UE speed. Results showed that all flow control schemes can maintain the similar average spectral efficiency since an appropriate lower bufferthreshold was chosen to guarantee the minimum throughput. That is, no starvation occurs when we applied the thresholdbased flow control scheme in LTE-A relay networks. In addition, due to the handover latency, it was found that each flow control scheme slightly reduces its average spectral efficiency when the UE speed increases. V. CONCLUSIONS In this paper, we investigate the performance for LTE-A relay networks. However, the buffer-overflow and the handover forwarding problems will occur due to the higher arrival rate and UE s handover, respectively. To solve both aforementioned problems, we proposed a to dynamically adust the upper buffer-threshold based on UE s channel quality. Results showed that can consume the smallest uplink bandwidth for data forwarding during UE handover. However, it still wastes large amount of uplink bandwidth to transmit the feedback messages. It is not an efficient way to solve these problems. Results also showed that and have much less signaling overhead than. However, our results further indicated that is a better flow control scheme than since it can reduce up to 40% of the total number of forwarding packets in uplink transmission and maintains a similar low level of signaling overhead. ACKNOWLEDGMENT This work was supported in part by the National Science Council, Taiwan under Contracts, NSC 102-2219-E-011-001 and NSC 102-2221-E-011-003-MY3. REFERENCES [1] P. Bhat et al., LTE-Advanced: An operator perspective, IEEE Commun. Mag., vol. 50, no. 2, pp. 104-114, Feb. 2012. [2] S. Parkvall, A. Furuskar, and E. Dahlman, Evolution of LTE toward IMT-Advanced, IEEE Commun. Mag., vol. 49, no. 2, pp. 84-91, Feb. 2011. [3] 3GPP TR 36.814, Evolved universal terrestrial radio access (E-UTRA): Further advancements for EUTRA physical layer aspects, Mar. 2010. [4] A. S. Ibrahim, A. K. Sadek, W. Su, and K. J. R. Liu, Cooperative communications with relay-selection: when to cooperate and whom to cooperate with? IEEE Trans. Wireless Commun., vol. 7, no. 7, pp. 2814-2827, July 2008. [5] L. Su, D. Chen, T. Wu, and J. Huang, Efficient power allocation schemes for MIMO cooperative relaying systems, in Proc. International Conference on Computer Application and System Modeling (ICCASM), Shanxi, China, Oct. 2010, vol. 14, pp. 179-183. [6] C. I. Kuo, S. H. Wu, C. K. Tseng, Robust linear beamformer designs for coordinated multi-point AF relaying in downlink multi-cell networks, IEEE. Trans. Wireless Commun., vol. 11, no. 9, pp. 3272-3282, Sep. 2012. [7] R. Schoenen, Credit-based flow control for multi-hop wireless networks and stochastic Petri Nets analysis, in Proc. 9th Annu. Commun. Netw. and Services Research Conf. (CNSR), Ottawa, Canada, May 2011, pp. 284-290. [8] R. Schoenen and H. Yanikomeroglu, Wireless hop-by-hop credit-based flow control extended to source for stable best effort traffic, in Proc. Australasian Telecommunication Networks and Applications Conf. (ATNAC), Melbourne, Australia, Nov. 2011, pp. 1-6. [9] K. Jagannathan, E. Modiano, and L. Zheng, On the trade-off between control rate and congestion in single server systems, in Proc. IEEE INFOCOM, Rio de Janeiro, Brazil, Apr. 2009, pp. 271-279. [10] J. Y. Kim and D. H. Cho, Pre-buffering scheme for seamless relay handover in relay based cellular system, in Proc. IEEE VTC-Spring, Barcelona, Spain, April 2009, pp. 1-5. [11] P. C. Lin, R. G. Cheng, and Y. J. Cheng, A dynamic flow control algorithm for LTE-Advanced relay networks, Accepted by IEEE Trans. Veh. Technol., 2013. [12] 3GPP TS 36.216, Evolved universal terrestrial radio access (E-UTRA): Physical layer for relaying operation, Mar. 2011. 2668