Semi-Distributed Resource Selection for D2D Communication in LTE-A Network

Transcription

1 Semi-Distributed Resource Selection for D2D Communication in LTE-A Network Seungil Park and Sunghyun Choi Department of ECE and INMC Seoul National University, Seoul, Korea Abstract Device-to-Device (D2D) communication underlaying cellular network has received much attention as a means to utilize cellular resources in a more efficient manner. Since D2D communication range is expected to be shorter than the normal cellular communication range, there are potential advantages such as reduced delay, reduced number of hops, improved spectral efficiency, and offloaded cellular traffic. In this paper, a semi-distributed resource selection scheme for D2D communication is proposed to efficiently reuse cellular resources. In addition, we propose interference avoidance scheme and D2D power control scheme to enhance D2D performance and minimize the degradation of cellular network performance, respectively. We demonstrate that the proposed semi-distributed algorithm outperforms the existing scheme by as much as 64% in terms of sum throughput of D2D users. I. INTRODUCTION Cellular traffic volume has been exponentially increasing due to the advent of smartphones as well as many other portable devices. Although Long Term Evolution (LTE)/LTE- Advanced (LTE-A) system developed by the 3rd Generation Partnership Project (3GPP) has been widely deployed, it is expected to be challenging to satisfy users demand in the future if the demand for mobile data traffic continues to grow. In order to meet such demand, research for improving system capacity in cellular network has been conducted in various ways. As one of the solutions, Device-to-Device (D2D) communication has received much attention in recent years. Since devices set up a link and transmit data directly in D2D communication, offloaded cellular traffic and reduced delay are highly anticipated as potential advantages. Normally, D2D communication range is expected to be shorter than the normal cellular communication range. Therefore, many D2D devices can reuse cellular resources simultaneously in order for D2D communication to achieve high spectral efficiency. Instead of using separate resources for D2D and cellular communication modes, reusing cellular resources for D2D communication achieves higher performance [1]. When it comes to reusing cellular resources for D2D communication, there is an issue in whether to reuse downlink resources or uplink resources. It is reported that reusing only uplink resources is more desirable [2] [6]. First of all, downlink traffic tends to be heavier than uplink traffic, leaving less free resource available for D2D communication. Second, downlink resource map in LTE-A is very complicated because of many dedicated resources (for broadcast and control usages). In addition, for D2D devices, it seems to be better to use the Single Carrier-Frequency Division Multiple Access (SC- FDMA) scheme, which is used in uplink LTE-A system due to its lower Peak-to-Average Power Ratio (PAPR). Lastly, reusing uplink resources is more tractable than reusing downlink resources since the potential victims of D2D communication are only enhanced Node Bs (enbs). In order to minimize the negative impact on enb for uplink resources used by D2D communications, any proposed resource allocation schemes for D2D devices must ensure that they do not cause harmful interference on the normal uplink communications (from D2D users to enb). Even though the papers [7] [10] develop their resource allocation schemes under the assumption of reusing downlink resources, they do not have grounds for reusing downlink resources instead of uplink resources. Many schemes have been proposed to deal with resource allocation for D2D communication. However, almost all rely on unrealistic system models that make it difficult for practical deployment. First, the schemes proposed in [1], [2], [7], [8] assume that Channel State Information (CSI) of all links are known to enb or every device. Such an assumption is not practical when there are many links among D2D devices. In reality, it is impractical to acquire all such information due to prohibitively large signaling overhead. Second, the schemes proposed in [1] [3], [7] [11] deal with only intracell problem (i.e., multicell environment is not considered). In D2D communication, Inter Cell Interference (ICI) problem is much severer than that in conventional cellular communication. That is because D2D devices which are closer to cell edge areas become more vulnerable to ICI. Although the authors in [12] consider the ICI problem, their proposed scheme is not suitable for real environments because they assume a multi-cell environment where each base station has an omnidirectional antenna. However, in the system, where base stations have directional antennas (i.e., sectorized cell) and D2D devices have omnidirectional antenna, the performance of D2D communication cannot be guaranteed due to different antenna properties. In contrast, our proposed scheme is based on a system model that closely approximates realistic cellular systems. First, we design a D2D communication procedure so that the signaling overhead is kept small and manageable. In our system, D2D devices can select their resources to reuse in a semi-distributed manner. That leads to substantial reduction of signaling overhead and computational complexity of enb. Second, we consider not only intra-cell interference problem but also ICI problem. In particular, we adopt 3-sector based multicell environments to evaluate the performance of our proposed scheme. The rest of the paper is organized as follows. We present

2 enb C-UE1 C-UE2 D-Pair1 DTx1 DRx C-UEN D-PairM DRxM DTxM Fig. 1: System model. DTx4 DTx3 DTx1 C-UE2 DRxk DTx2 DTx5 C-UE1 C-UE3 NLk D-field C-field Fig. 2: Example of Neighbor List (NL). our system model in Section II, and then our proposed resource selection scheme is described in Section III. In Section IV, the proposed scheme is evaluated via realistic, high-fidelity system-level simulations. Finally, we conclude our paper in Section V. II. SYSTEM MODEL Fig. 1 illustrates the system model in consideration. In a cell, there are N Cellular User Equipments (C-UEs) and M D2D Pairs (D-Pairs) denoted by C-UE 1,, C-UE N and D-Pair 1,, D-Pair M, respectively. Each D-Pair k consists of D2D Transmitter k (DTx k ) and D2D Receiver k (DRx k ). Accordingly, DTx k (DRx k ) is the target DTx (DRx) of DRx k (DTx k ). enb manages the C-UEs and D-Pairs. The roles of enb for D2D communication can be classified into two. The first is helping resource selection of D-Pairs and the second is power control of DTxs. enb allocates resources for cellular communication whereas D-Pairs select their resources in a distributed manner with the help of enb. We call our algorithm semi-distributed due to the help of enb. In other words, enb firstly helps D-Pairs select proper resources for an efficient transmission. Secondly, enb controls DTx power such that a DTx generates interference to enb only up to an acceptable level. Now, we explain how DRxs can learn the existence of neighboring C-UEs (or C-UEs in the proximity). Note that only uplink resources are reused by D-Pairs, and hence, DTxs need not learn the existence of C-UEs. That is because only DRxs will be the potential victim of C-UEs uplink transmission. Since enb allocates only a part of full system bandwidth to a specific C-UE at a given time, enb has to learn the quality of whole uplink resources. For the purpose, LTE defines Sounding Reference Signal (SRS) so that enb can estimate uplink channel quality and allocates proper resources to each C-UE based on the SRS measurement result. The location and period (one-time transmission is also possible) of SRS are also determined by enb. Here, we assume that enb controls the transmission of SRS and informs the location of SRS to DRxs. Then, DRxs can learn the existence of C-UEs by overhearing and measuring the received signal strength of SRS. For a resource selection scheme for D2D communication, we propose two kinds of signaling. enb manages the resources for the signaling. The first is Beacon and the second is Neighbor List (NL). Beacon is a signal that a DTx transmits to DRxs, where the location (i.e., which resource to use) of each Beacon is determined and broadcasted by enb. Therefore, every D2D UE (whether DTx or DRx) is aware of the location of Beacons. When a DRx receives Beacons, it learns the existence of neighboring DTxs including its target DTx. The existence of DTxs is judged based on the received signal strength. If the received signal strength is over a threshold, it is determined to be in the DRx s proximity. NL is a signal that DRx broadcasts to neighboring DTxs. NL k indicates the existence of DTxs and C-UEs in the neighborhood (i.e., proximity) of DRx k using two bitmaps, referred to as D-field and C-field, respectively. For example, as shown in Fig. 2, if DRx k finds out that DTx 1,DTx 2, and DTx 5 are in the proximity while DTx 3 and DTx 4 are not, then the D-field of NL k (i.e., the NL transmitted by DRx k ) will be Similarly, if DRx k finds out that C- UE 2 is in the proximity while C-UE 1 and C-UE 3 are not, then the C-field of NL k will be 010. Among the incoming NLs, DTx k needs to decode NL k received from DRx k to learn the information of DRx k s neighboring UEs. However, in the case of other incoming NLs, DTx k does not decode them, but just measures their received signal strengths to determine the existence of neighboring DRxs. Accordingly, by receiving NLs, DTx k learns not only the existence of neighboring DRxs but also the existence of neighboring DTxs and C-UEs of its target DRx. When enb allocates resources for Beacon and NL, the size of NL can be easily determined since enb is aware of how many D-Pairs and C-UEs exist in the cell. The size of NL mainly depends on the number of cotransmitting D-Pairs and C-UEs, which have data to transmit at the same time. That is because the size of each NL is proportional to the number of co-transmitting D-Pairs and C-UEs, and there are as many NLs as the number of cotransmitting D-Pairs. Since the number of co-transmitting D- Pairs and C-UEs is normally considered under a score (twenty) in a cell, the overhead of NL will not be large. Even when the number is over a score, enb can control the number of cotransmitting D-Pairs and C-UEs so that the overhead of NL is kept being small. In addition, we assume that synchronization is maintained among D-UEs as specified in [13] so that multiple transmissions and receptions can occur simultaneously. To make D-Pairs determine their transmissions in a distributed manner, enb also has to set priority. D-Pairs in a cell have their unique priority. The priority prevents D-Pairs from using the resources disorderly when two or more D- Pairs are located in the neighborhood. If resources are used indiscriminately, no D-Pair will get satisfactory performance.

3 DTxs DRxs C-UEs D-Pair Scheduling Beacon NL SRS Cellular scheduling information D-Pair Scheduling enb Fig. 3: Flow chart of the proposed resource selection scheme. Determining priority is similar to scheduling in conventional cellular systems. When enb determines priority of D-Pairs, it can consider various aspects such as D-Pairs channel condition, accumulated throughput, and minimum guaranteed rate. The rule of transmission is that when two or more D-Pairs are located in the neighborhood, the only D-Pair, which has the highest priority among neighboring D-Pairs, uses resources while the others keep refraining. Priority is a simple and easy method that resolves strong interference among D-Pairs. In addition, it is efficient in that enb need not schedule all the D-Pairs in the cell. Instead, by setting priority, enb can easily achieve an indirect scheduling effect. Furthermore, enb need not use additional resources to inform the priority because the information can be implicitly represented by using the location of Beacon or NL. This is possible if enb and D-Pairs make a prior agreement of the resource allocation pattern such that a specific location of resources represents a specific priority. In this case, enb can update priority as often as the period of Beacon and NL, which are repeatedly transmitted to reflect channel change and trace the existence of D-Pairs and C-UEs. III. STRATEGY FOR D2D COMMUNICATION In this section, the proposed strategy including resource selection and interference avoidance in multicell environments are presented. A. Resource Selection (RS) Scheme As mentioned above, D-Pairs select their resources in a distributed manner with the help of enb. Fig. 3 shows the flow chart of the proposed RS scheme, which works as follows: 1) SRS reception: AsC-UE 1,, C-UE N transmit SRS, DRx 1,, DRx M learn the existence of neighboring C- UEs. 2) Beacon transmission: As DTx 1,, DTx M transmit Beacons, DRx 1,, DRx M learn the existence of neighboring DTxs. 3) NL transmission: Based on the acquired information from Steps 1 and 2, each DRx k generates and transmits NL k. By receiving NLs from DRxs, DTxs learn not only the existence of neighboring DRxs but also the existence of neighboring DTxs and C-UEs of its target DRx. 4) Cellular scheduling information reception: When enb informs cellular scheduling information to C-UEs, D- Pairs can also receive the information, which can be DRx4 NL DTx1 NL (D-field) 1110 Beacon DRx1 (a) DTx2 DTx3 Beacon resource D-Pair 1 D-Pair 2 Higher D-Pair 3 priority D-Pair 4... (b) NDPL1 D-Pair 1 D-Pair 2 D-Pair 3 D-Pair 4 Fig. 4: RS examples: (a) NDPL and (b) Priority of D-Pairs. used to reuse the resources. Note that the scheduling information of a C-UE s uplink transmission is scrambled with the C-UE s identifier in LTE, and hence, we need to change the C-UE s identifier to a common key so that both C-UE and D-UEs can receive cellular scheduling information. Still, security is provisioned between the C- UE and LTE network since C-UE s data is encrypted with security key established during authentication. 5) D-Pair scheduling: Now,DTx 1,, DTx M have accumulated enough information to select resources. First, DTx k determines whether it can transmit data at a given time or not. To this end, DTx k makes a Neighboring D-Pair List (NDPL) by using the information about the existence of neighboring DRxs of DTx k and DTxs of DRx k.d-pair l will be the element of NDPL k if DRx l is nearby DTx k or DTx l is nearby DRx k.ifthereisnod- Pair l which has higher priority than D-Pair k in NDPL k, DTx k determines to transmit data at the given time. If not, DTx k determines not to transmit. DTx k finds out which C-UEs transmit with specific resources at a given time by receiving the cellular scheduling information from enb. By comparing the existence of C-UEs in NL k from DRx k and cellular scheduling information, DTx k determines which resources to reuse. In here, DTx k need not report the self-determined scheduling decision to DRx k.drx k is also aware of the existence of C-UEs in the NL and cellular scheduling information so that DRx k can predict the resources to reuse if DTx k transmits. Fig. 4 illustrates an example of the above-described resource selection scheme. As shown in Fig. 4a, DTx 1 finds out that D-Pair 1 (including itself), D-Pair 2, and D-Pair 3 are nearby by decoding NL 1 (which is transmitted by DRx 1 ) and that D-Pair 4 is nearby by receiving NL 4 (which is transmitted by DRx 4 ). Then, NDPL 1 is determined accordingly. If the priority is set by enb as shown in Fig. 4b, then DTx 1 cannot use resources this time because DTx 1 has lower priority

4 C-UE C-UE DRx FC (a) (b) DRx Fig. 5: ICI examples: (a) ICI from C-UE to enb and (b) ICI from C-UE to DRx. than neighboring D-Pairs. However, if the priority was set by enb the other way around (i.e., D-Pair 1 has the highest priority), then DTx 1 would use the resources. Note that the physical location of Beacon and NL resources is much more complicated than Fig. 4 to resolve a half duplex problem among D-UEs. However, we here show only logical location to help understanding. B. Interference Avoidance (IA) in Multicell Environments Using the above-described resource selection scheme, intra-cell interference can be minimized but ICI is not considered yet. Basically, in our simulation, enb has 3 directional antennas (i.e., each enb has 3 sectors). In the case a of 3- sector based multicell model, ICI from C-UEs to enb can be efficiently managed thanks to enb antenna pattern as shown in Fig. 5a. The antenna gain G(θ) is calculated as a function of the direction θ by [14] [ ( ) 2 θ G(θ) = min 12 65, 20] (db). In this case, as θ increases (up to 60 degrees), the ICI to other neighboring enbs decreases. However, when it comes to ICI from C-UEs to DRx (which is located in different cells), DRx is more vulnerable since DRx uses an omnidirectional antenna. Moreover, the location of a DRx can be closer than enb (whose location is fixed) to C-UEs as shown in Fig. 5b. To deal with ICI, Fractional Frequency Reuse (FFR) is used as shown in Fig. 6. In the conventional FFR scheme, center C-UE (which receives strong signal from enb, e.g., those in the white region in Fig. 6) uses center resources denoted by F C and edge C-UE (which receives weak signal from enb, e.g., those in one of the colored regions in Fig. 6) uses one group of edge resources, depending on the location (or color) of the sector where the edge C-UE belongs, denoted by F E1, F E2, and F E3, respectively. In the system, total resources F T is F C + F E1 + F E2 + F E3. For C-UEs, a FFR scheme is the same as the conventional one. In the case of D-Pairs, however, a D-Pair reuses either only edge resources or both edge and center resources depending on the DRx s condition. In order to determine DRx s condition, enb shares the information of SRS and whether each C-UE is a center C-UE or an edge C-UE with neighboring enbs. By doing so, a DRx can overhear not only SRS of inner cell C-UEs Fig. 6: An FFR scheme. FE1 FE2 FE3 but also SRS of neighbor cell C-UEs. Then, a DRx determines to reuse either edge resources or both edge resources and center resources according to whether DRx receives strong interference from center C-UEs in neighboring cells. Please note that edge C-UEs in neighboring cells are not of our concern because they use only edge resources. C. D2D Power Control Using the above-described RS scheme, D-Pairs may achieve good throughput performance, but the degradation of the cellular performance is not considered. enb has to guarantee the performance of both C-UEs and D-Pairs since enb is responsible for managing the whole network performance. One way to prevent the degradation of cellular performance is DTx power control. Interference from DTxs can be easily predicted and prevented since enb is the only victim (from the perspective of cellular network) of D2D communication using uplink resources. enb controls DTx power as follows. P DTx = min(p max,p 0 + α PL+ 10 log 10 M,P thres ) where P max denotes the maximum power that a device can generate, P 0 denotes the target power, α denotes the path loss compensation factor, M is the number of Resource Blocks (RBs), P thres is the maximum power such that the interference from DTx to enb is less than (i.e., P thres = +PL). M is not yet determined when DTxs negotiate their transmission power with enb, and hence, M is set to be the maximum number of RBs that DTxs can utilize in a cell. In short, DTx controls its power so that the received power at the DRx side can be P 0. However, if the interference to enb exceeds, D-Tx should transmit with a lower power, P thres, instead. IV. PERFORMANCE EVALUATION In this section, the performance of the proposed resource selection scheme for D2D communication is evaluated. We use system level simulator developed with MATLAB and C++. A. Simulation Environment The simulation model and parameters are summarized in Table I. 7 enbs with 3 antennas and wrap around model are

5 TABLE I: Simulation Parameters. Cell deployment 7 enbs with 3 sectors, 21 cells (wrap around model) Inter-site distance 750 m System bandwidth 5 MHz (24 RBs for PUSCH) Carrier frequency 2.0 GHz Channel model Path loss + multipath fading + shadowing Path loss (PL) PL = log 10 d (km) UE speed 3km/s Shadowing Log normal distribution with 8 db std. Correlation between sites/cells 0.5/1.0 Max power of enb 40 dbm Max power of UE (P max) 23 dbm Noise power per RB dbm HARQ Synchronous adaptive with chase combining Power control of C-UE min(p max,p 0 + α PL+10log 10 M) α 0.8 Number of C-UEs per cell 10 Number of D-Pairs per cell 10 Scheduling method for C-UEs Round robin Simulation time 20,000 subframes (20 s) used to reflect realistic environments. Parameters and traffic models are adopted from [14] [16]. Each value is obtained by averaging results from 50 iterations. B. Impact of RS and IA Schemes Here, we examine the performances of RS and IA schemes. Fig. 7 shows the average sum throughput of D-Pairs per cell according to the portion of edge C-UEs. The ratio between one group of edge resources and total resources is 0.2. That is, each F Ei (i =1, 2, 3) occupies 20% of total resources and F C occupies 40% of total resources. The traffic model of C-UE is a full buffered model. First, we compare our proposed schemes with the scheme in [12], which considers a resource reusing method in multicell environment. To deal with multicell environment, the authors have adopted FFR scheme which is also used in our proposed scheme. Since the environment and the system model in [12] are similar to those in our work, we compare this scheme with our proposed scheme. Since the comparison scheme considers only whether a D-Pair is in the inner region or outer region based on the received signal strength from enb, and resources to reuse are determined according to that information, D-Pairs may receive strong interference from C-UEs in neighboring cells. That is because the received signal strength from enb does not exactly reflect the distance between UE and enb due to antenna gain in sectorized-cell environment. Therefore, it always performs worse than our schemes, where D-Pairs basically use the received signal strength of SRS from C-UEs to reuse resources. Note that our proposed scheme outperforms the comparison scheme by as much as 64% (when the portion of edge C-UEsis0.1). Next, we examine the impact of IA scheme while RS scheme is being applied. As the portion of edge C-UEs decreases, RS+IA scheme more outperforms RS scheme. The main difference between RS+IA scheme and RS scheme is whether to consider the C-UEs in neighboring cells or not. The small portion of edge C-UEs means that there are more center C-UEs which are relatively close to cell edge. So, when the portion of edge C-UEs is small, the probability that DRx receives strong interference from neighboring cells is relatively high. However, in the case of the large portion of edge C- UEs, RS+IA scheme and RS scheme perform about the same Avg. sum throughput of D Pairs per cell (Mbps) RS+IA RS [12] Portion of edge C UEs Fig. 7: The performance of D2D communication with RS and RS+IA. because the probability becomes small. Still, RS+IA scheme is more efficient since IA scheme makes D-Pairs utilize fewer resources as explained in Section III-B while the performance is not degraded. C. Impact of Power Control Scheme Fig. 8 shows the average sum throughput of D-Pairs and C-UEs per cell according to Resource Utilization (RU), which is defined during a given observation time as follows: # RBs used by C-UEs per cell RU = Total # available RBs per cell. RU represents the ratio of the resources utilized by cellular traffic. For example, RU = 1 means the traffic model of C-UE is basically a full buffered model and RU < 1 means the traffic model of C-UE is a non-full buffered model. We evaluate the performance by varying RU while maintaining D2D traffic to be generated in a full-buffered manner. Here, both RS and IA schemes are applied. As RU increases, the sum throughput of D-Pairs decreases and the sum throughput of C-UEs increases since the possibility that a specific resource is used by a C-UE becomes higher. Accordingly, D-Pairs have fewer opportunities to reuse resources and the interference from C-UEs is also stronger. In contrast, C-UEs have more opportunities to use resources. Power control parameter,, is another important factor. indicates the maximum interference that enb may receive from each D-Pair in the worst case. As increases, the amount of interference at enb increases while D-Pairs can transmit with higher power and eventually it results in degrading cellular communication performance while improving D2D communication performance. However, the performance gain of D2D communication when is changed from 120 dbm to infinity (i.e., no power control case) is not much while the performance degradation of cellular communication is severe. That is because interference among D-Pairs is very strong, and hence, not many D-Pairs can coexist (i.e., transmit simultaneously). Therefore, it is important to set a proper power control parameter so that both D2D and cellular communication can achieve good performance.

6 Avg. sum throughput of D Pairs per cell (Mbps) Avg. sum throughput of C UEs per cell (Mbps) = infinity = 120 dbm = 130 dbm = 140 dbm RU w/o D Pairs = 140 dbm = 130 dbm = 120 dbm = infinity (a) RU (b) Fig. 8: ICI examples: (a) the performance of D2D communication, and (b) the performance of cellular communication. V. CONCLUSION AND FUTURE WORK In this paper, we propose a D2D communication procedure for a semi-distributed resource selection scheme and evaluate its performance under a realistic system model with 3-sector antenna and ICI. We assume that the same resources can be utilized by many devices at the same time, and hence, it is important to minimize interference. For that purpose, IA and D2D power control are proposed to maximize the system performance and their performance is evaluated via realistic system level simulation. While the degradation of cellular performance caused by D2D communication is inevitable, we demonstrate that the degradation can be minimized by using D2D power control scheme while D-Pairs achieve reasonable performance. In addition, our proposed scheme outperforms the existing scheme (which does not consider 3-sector based multicell environment) by as much as 64% in terms of sum throughput of D2D users under realistic system model assumptions. As future work, we plan to optimize the power control parameters which are introduced in this paper. The optimal parameters are affected by many factors such as user locations, traffic pattern, and scheduling method. We will take those factors into account so that the system can operate in various situations with high performance. ACKNOWLEDGMENT This work was supported by ICT R&D program of MSIP/IITP. {B , Multiple Access Technique with Ultra-Low Latency and High Efficiency for Tactile Internet Services in IoT Environments}. REFERENCES [1] C. H. Yu, K. Doppler, C. B. Ribeiro, and O. Tirkkonen, Resource Sharing Optimization for Device-to-Device Communication Underlaying Cellular Networks, IEEE Transactions on Wireless Communications, vol. 10, no. 8, pp , Aug [2] X. Zhu, S. Wen, G. Cao, X. Zhang, and D. Yang, QoS-based Resource Allocation Scheme for Device-to-Device (D2D) Radio Underlaying Cellular Networks, in Proc. ICT, 2012, pp [3] Y. Tao, J. Sun, and S. Shao, Radio Resource Allocation Based on Greedy Algorithm and Successive Interference Cancellation in Deviceto-Device (D2D) Communication, in Proc. IETICT. IET, 2013, pp [4] H. Min, J. Lee, S. Park, and D. Hong, Capacity Enhancement Using an Interference Limited Area for Device-to-Device Uplink Underlaying Cellular Networks, IEEE Transactions on Wireless Communications, vol. 10, no. 12, pp , Aug [5] H. Xing and S. Hakola, The Investigation of Power Control Schemes for a Device-to-Device Communication Integrated into OFDMA Cellular System, in Proc. IEEE PIMRC, 2010, pp [6] M. Belleschi, G. Fodor, and A. Abrardo, Performance Analysis of a Distributed Resource Allocation Scheme for D2D Communications, in Proc. IEEE GLOBECOM Workshops, 2011, pp [7] C. Xu, L. Song, Z. Han, D. Li, and B. Jiao, Resource Allocation Using a Reverse Iterative Combinatorial Auction for Device-to-Device Underlay Cellular Networks, in Proc. IEEE GLOBECOM, 2012, pp [8] C. Xu, L. Song, Z. Han, Q. Zhao, X. Wang, and B. Jiao, Interference- Aware Resource Allocation for Device-to-Device Communications as an Underlay Using Sequential Second Price Auction, in Proc. IEEE ICC, 2012, pp [9] L. Su, Y. Ji, P. Wang, and F. Liu, Resource Allocation Using Particle Swarm Optimization for D2D Communication Underlay of Cellular Networks, in Proc. IEEE WCNC, 2013, pp [10] D. Zhu, J. Wang, A. L. Swindlehurst, and C. Zhao, Downlink Resource Reuse for Device-to-Device Communications Underlaying Cellular Networks, IEEE Signal Processing Letters, vol. 21, no. 5, pp , May [11] B. Wang, L. Chen, X. Chen, X. Zhang, and D. Yang, Resource Allocation Optimization for Device-to-Device Communication Underlaying Cellular Networks, in Proc. IEEE VTC Spring, 2011, pp [12] H. S. Chae, J. Gu, B. Choi, and M. Y. Chung, Radio Resource Allocation Scheme for Device-to-Device Communication in Cellular Networks Using Fractional Frequency Reuse, in Proc. IEEE APCC, 2011, pp [13] 3GPP TR , LTE Device to Device (D2D) Proximity Services (ProSe); User Equipment (UE) radio transmission and reception, ver , Mar [14] 3GPP TR , Evolved Universal Terrestrial Radio Access (E- UTRA): Further Advancements for E-UTRA Physical Layer Aspects, Mar [15] 3GPP TS v12.2.0, Evolved Universal Terrestrial Radio Access (E-UTRA); Physical Channels and Modulation, Mar [16] 3GPP R , LTE Physical Layer Framework for Performance Verification, Sep