IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS.VOL.14.NO.10.OCTOBER 2015 leads to overhead.When there are a larger number of clients more trialsare needed tocollect historical information.Thus.as in overh and 13.Ho in the ith different frame engths (ie in Fis 13 there exists additional overhead.For example,when Client A the frame length of 24 Bytes es much earlier than the AP 0-802.11DC ing suc ufor the scheme without SIR MAP the throughput in 10 20 Fig.13 drops more quickly.scheme Fig.12.The th not have any gain as compared to pe 15 3)Packer Loss and Delay:We evaluate packet loss and delay under the influence of client movement.In A-Duplex R MAP IE I-link setup deper y upda of th 13 ts SIR MAP is not undated in time 12缚 fail.To evaluate the impact of mobility to SIR MAP.we only consider unsaturat on mode. 10 be brought back to the AP in a timely fashion The random w aypoint mobility model [2]is used to sim 7 omy choose O SIR MAP 64 Number of cients t moving speed.In our simulations moving ped we rs per se Fig.13.The throughput under different frame lengths ed MAC tin LAN under a practical environment.The network topology is slightly with se the shown in Fig.10 where there are 10 clients.The total number overhea cting the his the t d to be in a is 37%higher than this scheme The frame length of a dat 的 nts. The packet le ngth i 13T of A-D to 7.The e d o pack is clearly illustrated.For example,with 25 clients,A-Duplex and 122%as compared to smitted packet in a dual-link is not counte that of th with RTS/CT with on.Howet n the d is duplexw scheme in the second case is lower than that in the first case as a retransmission.Thus in full duplex with SI MAPC) The reason is that in the second case the frame length is shorter the number of retransmissions for some packets from the AP than that in the first case.which makes the verhead in the times.As a he second DCE information of dual links via blind trial.More specifically to higbest collisions The packet los s of full duplex with SIR the AP MAP(2)is nearly equal to IEE duple t SIR MAP5882 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 14, NO. 10, OCTOBER 2015 Fig. 12. The throughput under the same frame length. Fig. 13. The throughput under different frame lengths. slightly with an increasing number of clients, because the overhead of collecting the historical information of dual links increases. Considering 25 clients, the throughput of A-Duplex is 37% higher than this scheme. In the second case, we consider different frame lengths, i.e., 240, 576, 1000, and 1500 Bytes. The frame length of a data packet is randomly selected from these values. The results of this case are shown in Fig. 13. The advantage of A-Duplex is clearly illustrated. For example, with 25 clients, A-Duplex improves throughput by about 34% and 122% as compared to that of the 802.11 DCF with RTS/CTS and without RTS/CTS, respectively. Furthermore, the saturation throughput of each scheme in the second case is lower than that in the first case. The reason is that in the second case the frame length is shorter than that in the first case, which makes the overhead in the second case larger and results in a lower throughput. As for the scheme without SIR MAP [6], the AP collects the historical information of dual links via blind trial. More specifically, the AP searches the historical information of dual links to select clients to set up asymmetric dual links. Such historical information is accumulated via trial-and-error. Failure in trails leads to overhead. When there are a larger number of clients, more trials are needed to collect historical information. Thus, as the number of clients increases, the overhead increases and the throughput drops, as show in both Figs. 12 and 13. However, in the scenario with different frame lengths (i.e., in Fig. 13), there exists additional overhead. For example, when Client A sends a packet to the AP with the frame length of 240 Bytes, the AP tries to select Client B to form dual links and then send a packet with a frame length of 1500 Bytes. Whether this trial succeeds or not, Client A experiences additional overhead, because its transmission finishes much earlier than the AP’s transmission. Due to blind trial, the chance of having such additional overhead grows if the number of clients increases. As a result, for the scheme without SIR MAP [6], the throughput in Fig. 13 drops more quickly. Considering 25 clients, the scheme without SIR MAP does not have any gain as compared to 802.11 DCF with RTS/CTS. However, throughput performance of A-Duplex remains steady as the number of clients grows. 3) Packet Loss and Delay: We evaluate packet loss and delay under the influence of client movement. In A-Duplex the success of dual-link setup depends on timely update of the SIR MAP. If a client moves away from the previous place but its SIR MAP is not updated in time, the dual-link setup may fail. To evaluate the impact of mobility to SIR MAP, we only consider unsaturation mode. The reason is simple: when nodes constantly send packets (i.e., in saturation mode), the SIR MAP can always be updated in time, because the SIR information can be brought back to the AP in a timely fashion. The random waypoint mobility model [26] is used to sim￾ulate client mobility, i.e., each client can randomly choose a direction and a distance for each step. Each client moves a step every 0.25 s, but the step distance varies in different experi￾ments to simulate different moving speed. In our simulations, we set the maximum moving speed to 4 meters per second to evaluate the performance of our proposed MAC protocol since we consider the situation of the people’s movement in a wireless LAN under a practical environment. The network topology is shown in Fig. 10 where there are 10 clients. The total number of packets generated in the network is controlled to be in an unsaturation mode. More specifically, every 0.25 s we allocate 25 packets to the 10 clients in a random way and assign 25 packets to the AP with destinations selected randomly from the 10 clients. The packet length is set to 1500 Bytes. In our simulation, the number of packet retransmission is set to 7. The packet loss rate and the packet delay are shown in Figs. 14 and 15, respectively. In the case of full duplex with SIR MAP(1), a retransmitted packet in a dual-link is not counted as a retransmission. However, in the case of full duplex with SIR MAP(2), whenever a packet is retransmitted, it is regarded as a retransmission. Thus, in full duplex with SIR MAP(1), the number of retransmissions for some packets from the AP to a client may exceed 7 times. As a result, the full duplex with SIR MAP(1) achieves the lowest packet loss. IEEE 802.11 DCF without RTS/CTS results in the highest packet loss due to highest collisions. The packet loss of full duplex with SIR MAP(2) is nearly equal to IEEE 802.11 DCF with RTS/CTS, but both achieve lower packet loss than that of full duplex without SIR MAP