Link Adaptation for Unicast Device-to-Device Communication

Link Adaptation for Unicast Device-to-Device Communication Yu-Ngok Ruyue Li, Haigang He, Yifei Yuan

Shuanshuan Wu

Wireless Product R&D Institute, ZTE Corporation Email: {li.ruyue, he.haigang}@zte.com.cn, [email protected]

Department of Electrical Engineering, University at Buffalo, Buffalo, New York [email protected]

Abstract — Device-to-Device (D2D) communication underlaying cellular network enables local services that can enhance energy efficiency and provide new business opportunities for wireless network operators. In this paper, the study is focused on link adaptation mechanisms for unicast D2D. In particular, we first investigate the benefits of having hybrid automatic repeat request (HARQ) and channel state information (CSI) feedback for unicast D2D and discuss implementation of link adaptation based on these conventional feedback mechanisms. Next, we propose a new joint feedback-based link adaptation scheme called soft HARQ feedback which provides a good solution for high reliable and low latency feedback for unicast D2D applications. This scheme has potential to simplify the implementation of fast link adaptation for unicast D2D compared to conventional HARQ and CSI feedback enabled link adaptation. With lower overhead and complexity, our evaluation shows that this scheme provides comparable performance compared to conventional CSI feedback scheme. Keywords—Device-to-Device (D2D); link adaptation; soft HARQ; Unicast; radio resource management; Vehicle-to-Vehicle (V2V);

I. INTRODUCTION Proliferation of smart devices has been leading to increased need for more powerful cellular communication systems to fulfill the ever increasing wireless communication requirements such as higher energy efficiency, better coverage and new diversified services. The industry is looking for new innovation in architecture and technologies that will address the capacity and service demands in both evolving 4G and 5G wireless communication systems. Direct device-to-device (D2D) communication underlaying cellular network, which enables direct traffic exchange between devices in cellular infrastructure, shows tremendous potential to meet the evolving requirements of wireless communication networks [1][2]. D2D integrated in cellular networks offers the following application scenarios. • Social networking- D2D enables users to discover interested users/apps in surrounding areas, and offers direct communication between them. Such ProSe-based (Proximity Service) social application can improve user experience and diversify cellular services and thus has the potential to provide operators a new source of income. • Traffic offloading - Cellular communication between proximity devices can be switched to direct D2D mode

and the two-hop uplink and downlink cellular communication can be replaced by D2D. Not only the resource usage can be more efficient, but also device battery life can potentially be prolonged due to the decreased power consumption. Moreover, another traffic offloading use case is that users could acquire interested contents via D2D, from other nearby users who own the contents [3] or from a media server which is set up to offer media contents [4]. The cellular downlink traffic load is thus relieved. • Relaying - D2D can be used to extend the cellular network coverage, i.e. devices in coverage hole is able to access cellular network via D2D, from another device within cellular coverage. Relaying can also be used to improve throughput of devices e.g. devices at the cell edge and save their power consumption [5]. • Public safety applications- In some countries e.g. the United States, spectrum has been reserved for networks to incorporate national security and public safety services [6]. According to public safety requirements, cellular services should be available even when network nodes become dysfunctional in case of natural disasters or emergency situations. D2D has the potential to meet the public safety requirements. For instance, D2D enables devices to form an Ad-hoc network and devices in the network can communicate with each other or even access to cellular network. • Ultra-Reliable Low latency Communications – Applications like Vehicle to Everything (V2X) communication including Vehicle-to-Vehicle have stringent requirements on both relability and latency. Standardization of vehicular communication has been developed based on the framework of D2D standard. Though D2D has already been specified in 3GPP (Third Generation Partnership Project) LTE (Long Term Evolution) Release-12 [7], only broadcast communication is supported: packet is transmitted four times and no feedback, i.e. acknowledgement/negative acknowledgement (ACK/NACK) or channel state information (CSI) feedback is enabled. That means, transmitting devices (user equipment or UE in term of LTE terminology) or base station do not have the knowledge about channel condition of D2D link (sidelink or SL in LTE), or even do not know whether the packet is correctly transmitted or received. In fact, the major application scenario of D2D communication in Releases 13 and 14 are public safety and

978-1-5386-3920-7/17/$31.00 ©2017 IEEE

relaying, and one-to-many broadcast communication is main use case for current public safety requirements. In other words, the potential gain of D2D is not fully explored in LTE. As we know, highly efficient unicast (one-to-one transmission) D2D communication plays a considerably important role if we want to fully explore D2D performance gains in typical use cases such as traffic offloading, relaying and V2X communications. Compared to one-to-many broadcast communication, the major advantage of unicast D2D is link adaptation based on HARQ (Hybrid Automatic Repeat Request) transmission. In HARQ transmission, HARQ-ACK feedback helps terminating unnecessary retransmissions (when transmission is completed or expired) and guarantees the traffic to be successfully delivered. CSI feedback enables time/frequency selective scheduling and adaptive modulation and coding (AMC) which helps improving spectral efficiency. However, D2D is quite different from traditional cellular as additional direct sidelink transmission is introduced. Scheduling, radio resource management, as well as channel and interference conditions may be thus largely different compared to those in cellular networks [8]. Some of applications need high reliability and low latency which makes fast link adaptation more important for D2D communication. Therefore, link adaptation for unicast D2D should be carefully studied and designed to maximize D2D communication performance and minimize the impact to cellular communication at the meantime. Our work focuses on the link adaptation-based HARQ transmission mechanisms for unicast D2D transmission performance. Specifically, we investigate the solutions of HARQ-ACK feedback and CSI feedback for unicast D2D and the implementation of link adaptation for D2D based on conventional link adaptation mechanisms in section II. In Section III, we discuss a new link adaptation mechanism called soft HARQ feedback scheme which has the potential to improve the performance of unicast D2D transmission while keep implementation complexity low. In section IV, systemlevel simulation is carried out to study numerical performance of the referred link adaptation solutions for D2D transmission. Finally, we draw our conclusions in Section V. II. HARQ-ACK AND CSI FEEDBACK FOR UNICAST D2D Feedback-based link adaptation mechanisms have been widely used in current wireless communication systems to improve link performance. Feedback referred here comprises HARQ-ACK and CSI. ACK/NACK is essential to unicast communication as discussed earlier. CSI is very important supplementary information as it offers a reference for scheduling and MCS selection, and hence makes transmission more efficient. A. HARQ-ACK feedback Unlike transmitter-receiver cellular communication, 3 kinds of node can be involved in D2D communication: transmitting UE, receiving UE and base station. The HARQ-ACK feedback mechanism is thus different from cellular communication. For example, HARQ-ACK can either be fed back to base station or to transmitting UE. 1) Option-1: transmitting HARQ-ACK to base station

This is more aligned with D2D scheduling performed by base station which can be called centralized D2D scheduling. A typical centralized scheduling process for D2D can be summarized as shown in Fig. 1. 1) Option-2: transmitting HARQ-ACK to D2D transmitter HARQ-ACK can also be directly transmitted to transmitting UE of D2D traffic. If NACK is detected, the transmitting UE performs D2D traffic retransmission. A typical scenario is the so-called semi-centralized D2D scheduling: base station allocates resources for D2D semi-statically or periodically, and the transmission and retransmission of D2D traffic are performed by D2D UEs in the allocated resources. An example is shown in Fig. 2.

Fig. 1. Centralized scheduling model for unicast D2D. In centralized scheduling, HARQ-ACK information is fed back to base station and base station dynamically schedules packet transmission and retransmission.

(1) Scheduling request (2) DL control information (3) SL control information/data (4) ACK/NACK SL control information/data ACK/NACK

Fig. 2. Semi-centralized scheduling model for Unicast D2D. Resources are semi-statically or persistently allocated by base station, and then D2D transmitter performs scheduling, e.g. dynamic link adaptation based on HARQACK feedback and/or channel state feedback.

Intuitively, Option-1 shown in Fig.1 is more flexible and efficient due to more dynamic scheduling. However, flexibility means frequent signaling, and hence high signaling overhead. In particular, the scheduling delay in addition to the feedback delay causes prolonged HARQ RTT (Round Trip Time), which in turn results in extra service delay and more required buffer size in UE as number of HARQ processes have to be increased according to the characteristic of stop-wait HARQ protocol. While for Option-2, the major problem is inflexibility of scheduling, and this may cause performance degradation sometimes, for example when the semi-statically or periodically allocated band suffers from a deep fade. B. Channel State Information Channel state information, such as channel quality, rank and precoder information, is very important for link adaptationbased wireless communication as they act as references for determining the resources and MCS, as well as references for multiple antenna technologies for transmission. Specifically, two aspects are needed for obtaining channel state information: CSI measurement and CSI feedback. 1) CSI measurement Generally speaking, CSI can be measured based on the received/detected reference signal (RS) which are dedicated to channel state measurement. For LTE D2D, sounding mechanisms in LTE uplink can be reused if D2D works on uplink resource. For example, a part of sounding resources (time/frequency resource, sounding reference signal sequences, etc) can be allocated to D2D link. Of course, though careful design and implementation can be carried out to minimize the impact to cellular communication, the impact can’t be thoroughly eliminated as the sounding resources are shared by D2D, e.g. cellular sounding performance may suffer from degradation due to the decreased available sounding resources and the potential interference caused by D2D sounding. D2D-specific dedicated measurement reference signal is another option. However, introducing additional new measurement reference signal not only needs extra specification effort, but also may lead to a very complicated D2D implementation. Generally speaking, D2D discovery and/or D2D synchronization may accompany D2D communication in an integrated D2D solution, discovery signal (also known as beacon) or synchronization signal/channel may consequently act as alternative measurement references. In particular, demodulation reference signals accompanying the traffic packet can be a more appropriate solution for channel measurement as it is often more frequently transmitted. 2) CSI feedback The general channel state information comprises rank indication, precoding matrix indication and channel quality indication. Since devices supporting D2D are required to transmit and receive in the same frequency, implementation of D2D system is similar to a TDD system. In particular, D2D can be done in uplink carrier or uplink resource of cellular system considering the interference level and under-utilized uplink resources. As a consequence, channel reciprocity in TDD systems can be explored in D2D as well. For instance multiantenna related channel state information such as channel rank and precoding matrix can be calculated by channel reciprocity

theoretically. However, channel quality measured in transmitter and receiver may be different due to different interference observed at different sides. In other words, CSI feedback is necessary if we want to fully explore benefits for unicast D2D from link adaptation, at least considering the feedback of channel quality which is usually used to determine MCS. Similar to ACK/NACK feedback, channel state information can be transmitted to D2D transmitter or base station. If the CSI is transmitted to base station, extra scheduling delay will be introduced and D2D transmission performance may be degraded since instantaneity is very important to channel state information. Therefore, feeding back CSI to D2D transmitter makes more sense. C. Conventional Link Adaptation Mechanisms It can be seen from the previous discussion that HARQACK is considerably important to link adaptation of unicast D2D, while CSI act as additional mechanism to improve link performance. Based on the related discussion, we have the following two link adaptation schemes for unicast D2D based on conventional feedback mechanisms. a) Scheme-1: Outer Loop Link Adaptation (OLLA) [9], only HARQ-ACK/NACK is fed back from UEs and the transmission MCS is determined based on the statistical block error rate (BLER) with a predefined target BLER. b) Scheme-2: Inner Loop Link Adaptation (ILLA), both HARQ-ACK and CSI feedback are enabled and the transmission MCS can be determined based on the CSI as well as the statistical BLER. Scheme-1 is much simpler as no channel measurement and CSI feedback mechanism is needed. However, the following two challenges may make Scheme-1 problematic for D2D. • Channel coherence time: Since OLLA uses binary ACK-NACK information to adapt to the channel, it takes considerable time for OLLA to be stable and converge to the MCS corresponding to the target BLER. It is impossible to guarantee that OLLA can catch up the channel variation if the coherence time is not long (e.g. tens of milliseconds). The channel becomes independent out of the coherence time window which means the channel may have a deep fade which can have a big difference (e.g. 10dB) on the received power. It is hard to catch up the channel variation in a coherence time under the assumption of LTE HARQ feedback delay provided that the step-size of link adaptation is not large (typically it is around 0.1dB). Although CSI also has feedback delay, it can measure the instantaneous channel and let the transmitter utilize the CSI within the coherence time window. • Burstiness of traffic: Bursty traffic characteristic is the most problematic part for OLLA. If the traffic is bursty, OLLA can't track the channel because the transmission may end before the OLLA converges. Bursty traffic also makes the interference bursty. If it is a low-medium load which there is usually only one active UE at a given time, it needs tens of milliseconds to transfer a file,

similar level of interference can be kept and CSI can track the interference. In other words, Scheme-2 will outperform Scheme-1 in D2D communication theoretically from the perspective of performance. However, as discussed in previous section, conventional channel state measurement and feedback mechanisms need more effort on implementation as well as standardization. Cellular network performance may be impacted as well. III. SOFT HARQ FEEDBACK SCHEME According to the discussion on the necessity as well as preliminary implementation of HARQ-ACK and CSI feedback for link adaptation-based unicast D2D in Section II, it can be observed that both OLLA and ILLA based on conventional feedback mechanisms have potential issues and hence leaves rooms for improvement on unicast D2D. In this Section, we discuss a new link adaptation mechanism based on joint feedback scheme called soft HARQ scheme, which can achieve good performance while keeping implementation complexity and impact to cellular low. For HARQ-ACK feedback, 1 bit is used to indicate whether a traffic packet is correctly received. The motivation of soft HARQ scheme is to slightly expand the payload of HARQACK bits to accommodate more information. For example, if the HARQ-ACK corresponding to each transport block is 2 bits, then the additional 1 bit can be used to indicate information related to channel state of sidelink or power control (power adjustment) for D2D transmission. Take D2D in LTE as an example, introducing additional bits in HARQ-ACK information does not cause any implementation issue since transmission of very large HARQ-ACK payload (up to 20bits) is already supported in LTE. A. Solution-1: additional bit(s) used for power control / power adjustment One advantage of D2D is its lower transmission power due to the very proximity transmission range. Nevertheless, power control for D2D is difficult to perform since distribution of D2D devices is often arbitrary. If power control is performed according to base station signaling, base station needs to obtain the knowledge of sidelink channel state, D2D traffic status, etc, which leads to an inefficient solution; while if power control is performed by UE, it may be difficult to figure out a suitable reference to adjust transmission power as it may be difficult for UE to have the full knowledge of interference between D2D and cellular. If UE always transmits in the allowed maximum transmission power, not only the solution is inefficient, but also the interference to cellular may become severe. Therefore, the motivation of Solution-1 is to introduce power adjustment to link adaptation for unicast D2D, i.e. D2D receiver will give power adjustment information to transmitter based on its reception status. For instance, two bits are used for soft HARQ feedback: if the parity check result of data packet is ACK, the additional 1 bit is to indicate whether the transmission power should be stepped down; if the check result is NACK, the 1 bit can be used to indicate whether the transmission power should be stepped up, or to indicate different increment. The value of the 1 bit can be decided by the demodulation SINR,

and/or the difference between real transmission power and the allowed maximum transmission power. Table I is an example of soft HARQ bits definition with one additional HARQ bit used for power adjustment. TABLE I. AN EXAMPLE OF SOFT HARQ BITS DEFINITION (SOLUTION-1)` Soft HARQ bits

Explanation

00 NACK, power up and power step = X1 01 NACK, power up and power step = Y1 10 ACK, power step = 0 11 ACK, power down and power step = Z1 1 Parameters X, Y and Z are predefined by empirical values and X > Y.

B. Solution-2: additional bit(s) used for MCS indication / adjustment As discussed earlier, D2D is likely to be implemented based on TDD in uplink carrier or uplink resource, and then multiantenna transmission related channel state information such as channel rank and precoding matrix can be calculated according to channel reciprocity. In other words, channel state information related to determining MCS (e.g. channel quality indication in LTE) plays a more important role for D2D CSI feedback. Therefore, the motivation of soft HARQ Solution-2 is to introduce additional bit(s) to accommodate the indication of variation of CQI or MCS. For example, additional 1 bit is used to indicate the adjustment of MCS accompanying with HARQ-ACK. If D2D data packet is correctly decoded, the additional 1 bit is to indicate whether a higher MCS is allowed based on the measured SINR to achieve the target BLER; while receiver fails decoding D2D traffic block, the additional 1 bit can indicate whether a lower MCS is needed, or it can indicate two different MCS decrement. A pseudo-code algorithm of Solution-2 is shown in Table II. TABLE II. ALGORITHM OF SOFT HARQ FEEDBACK (SOLUTION-2) Start Perform parity bits check of received data packet Compute the demodulation SINR of received data packet and/or reference signal If ACK, BIT1=1; If demodulation SINR – stepdown1> required SINR A higher MCS is allowed, BIT2=1; else MCS should remain the same, BIT2=0; end; else BIT1=0; If demodulation SINR + stepup_11 < required SINR A much lower MCS is needed, BIT2=1; else if demodulation SINR + stepup_21 < required SINR A lower MCS is needed, BIT2=0; end end end 1 Parameters stepdown, stepup_1 and stepup_2 are predefined by empirical values, stepup_1 > stepup_2.

Alternatively, two joint soft HARQ bits can be defined as follows: use 11 to indicate ACK, 10/01/00 to indicate NACK, as well as three different MCS decrement. This is more helpful to make the link adaptation converged as soon as possible. The major advantage of the joint soft HARQ feedback method is that feedback-based link adaptation is enabled for unicast D2D with limited signaling overhead and implementation complexity of link adaptation for unicast D2D can be largely simplified. Moreover, since channel state related information is transmitted together with HARQ-ACK, the instantaneity of CSI is guaranteed, which is more meaningful for D2D as interference fluctuation is much larger in D2D communication and thus CSI can be utilized within channel coherent time.

UEs taken into account as shown in Fig.3. The parameter Pmax is the maximum allowed transmission power of a UE, PRB is the transmission power of each resource block (RB) calculated by power control parameters, M is the maximum available RBs determined by transmission power, NRB is the number of RBs haven’t been allocated yet in sector, NUE is the number of UEs that haven’t been scheduled. START

For each UE, calculate M: M =

IV. PERFORMANCE ASPECTS In this section, system-level simulation results are shown based on the performance evaluation conducted for different link adaptation mechanisms discussed in the previous sections. The following four HARQ schemes are evaluated: 1) Blind transmission 2) HARQ feedback 3) CSI feedback 4) Soft HARQ feedback Blind transmission is used as a baseline for comparison. This is the scheme without any HARQ/CSI feedback adopted in LTE Release-12 D2D communication. A fixed code rate 772/1024 with 64QAM modulation [10] is chosen for blind transmission. HARQ feedback and CSI feedback are the conventional feedback schemes, Scheme-1 and Scheme-2 discussed in Section II respectively. For CSI feedback, 4 bit CQI is used and feedback is performed per 5ms. The forth scheme evaluated here is the soft HARQ scheme which is aligned with what described in Table II, i.e. one additional soft HARQ bit is used to perform MCS adjustment.

N >0 && N >0 No Yes Schedule a UE with the highest round robin scheduling metric and the minimum value of M

Allocate min(M，

) resource blocks for the

scheduled UE

N N

=N

= N -1 - min (M，

)

Simulation assumptions of are described as follows: • Deployment layout A typical network topology used within 3GPP D2D study [11] is used here, i.e. hexagonal grid, 3 sectors per site with 19 macro sites with 500 m inter-site distance and one indoor hotspot per cell. • UE dropping 150 UEs are dropped in each sector and details of UE dropping is also described in [11]. On the average, 10 UEs are paired up for D2D communication in each sector with association threshold of -80dBm. • Power control Open loop power control is used for D2D link based on measured pathloss between D2D UEs with the target received power of -75dBm. • Resource allocation Semi-persistent resource allocation is used for D2D, and the resource allocation cycle is 160 milliseconds. The detailed resource allocation is a revised Round-Robin scheduling with maximum transmission power of D2D

END Fig. 3. Per-subframe resource allocation algorithm

The cell average and 5-percentile cell edge spectral efficiency comparison is shown in Fig.4. Note that the traffic model is full-buffer and the maximum transmission number of each data packet is 4 in the simulation. From the observed performance evaluation results, it proves that feedback for unicast D2D communication is essential as it outperforms blind transmission significantly. Regarding the link adaptation mechanisms, though the conventional ILLA with CSI feedback has the best performance of both cell average and cell edge spectrum efficiency, we note that soft HARQ feedback based link adaptation achieves similar spectrum performance as CSI feedback. Both schemes outperform hard HARQ scheme especially for cell edge performance. Moreover, normalized user throughput comparison is shown in Fig.5 for reference. It can be observed that all the three enhancement schemes clearly outperform the baseline scheme. Both CSI feedback and Soft HARQ schemes have the benefit on cell edge performance

compared to HARQ feedback. Unicast D2D could have very small benefit from CSI feedback of sidelink channel state information which causes higher feedback overhead and complexity. Considering less signaling overhead and implementation complexity, soft HARQ scheme is proven to be the best choice for unicast D2D.

Fig. 4. Comparison of cell average and 5% cell edge spectral efficiency.

Fig. 5. Normalized user throughput of different feedback schemes

V.

CONCLUSIONS

We analyze the usage of conventional link adaptation mechanisms in D2D underlaying cellular network and propose a new link adaptation solution for D2D unicast communication. In the new link adaptation solution, a soft HARQ feedback mechanism is discussed and recommended, which enables channel state information feedback with only marginal signaling overhead. According to our performance evaluation based on system-level simulation, the soft HARQ link adaptation scheme achieves similar performance compared to conventional inner loop link adaptation with HARQ-ACK and CSI feedback enabled, while with a much simpler implementation. Therefore, it provides a good solution particularly for the unicast D2D applications which require high reliablility and low latency.

REFERENCES [1]

3GPP TR 38.913, Study on Scenarios and Requirements for Next Generation Access Technologies [2] S. Talwar, D. Choudhury, K. Dimou, E. Aryafar, B. Bangerter and K. Stewart. “Enabling technologies and architectures for 5G wireless Microwave Symposium (IMS),” 2014 IEEE MTT-S International, pp. 14, June 2014. [3] F. Malandrino, C. Casetti and C.-F. Chiasserini, “Toward D2D-Enhanced Heterogeneous Networks,” IEEE Communications Magazine, vol. 52(11), 2014, pp. 94-100. [4] K. Doppler, M. Rinne, C. Wijting, C. B. Ribeiro and K. Hugl. “Deviceto-Device Communication as an Underlay to LTE-Advanced Networks,” IEEE Communications Magazine, vol. 47(12), 2009, pp. 42-49. [5] T. C.-Y Ng and W. Yu. “Joint optimization of relay strategies and resource allocations in cooperative cellular networks,” Selected Areas in Communications, IEEE Journal on, vol. 25(2), 2007, pp. 328-339. [6] Delivering Public Safety Communications with LTE, 3GPP white paper, available: http://3gpp.org/Public-Safety. [7] Qualcomm Inc.. “RP-122009: Study on LTE Device to Device Proximity Services,” 3GPP TSG RAN Meeting #58, December 2012. [8] G. Foder, E. Dahlman, G. Mildh, S. Parkvall, N. Reider, G. Miklós, et al. “Design Aspects of Network Assisted Device-to-Device Communications”. IEEE Communications Magazine, vol. 50(3), 2012, pp. 170-177. [9] K. Pedersen et al., “Frequency domain scheduling for OFDMA with limited and noisy channel feedback,” in Proc. 66th IEEE Trans. Veh. Technol. Conf., 2007, pp. 1792–1796 [10] 3GPP Technical Specification 36.213 V12.4.0. “Evolved Universal Terrestrial Radio Access (E-UTRA); Physical layer procedures (Release 12).” January 2015. [11] 3GPP Technical Report 36.843 V12.0.1. “Study on LTE Device to Device Proximity services; Radio Aspects (Release 12),” March 2014