TANG AND WANG:MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL AND HALF-DUPLEX COMMUNICATION 5881 20 e Station AP 16 0 10 -15 -20 252520-15-1050 101520 x coordinates (m) Fig.10.A typical topology with hidden nodes ie.11.The throughput in a wireless LAN with 40 hidden node TABLE IV 802.11 DCF without using RTS/CTS casel case2 case3 case4 case5 case6 case7 cases g K am1110.60.910991038 oud robin alg possible that the default round robin order is changed but no proper client can be selected for dual-link setup.which leads to roughput and f establishes dual links with hidden nodes (ie..case four).the for case four.However.as explained in the end of Section V.we throughput is improved by 16%and 6%as compared to case respectively.However, ith captu talimit to the negative deficit to improve the four.In cases fve to s onut of case six (B=2)is a little higher than that of case five or case achieved in both throughput and faimess. In the second set of ix,so the average trans threspect to th er o client average transmission rate of dual links is higher than case six age throughput ogies.is set to2.which but the capture probability is lower,which leads to a little lower is shown to be a proper value in the first set of experiments As the first Fig.12 a ram the best c collisions occur more frequently.The throughput of 802.11 with RTS/CTS is stable even if the numb of clients wn omp A-Duple muc MAC protocol improvesth with RTS/CTS and without three,the roughput is the owe out pe rfect fair /CTS,respectively. ulate the sc e propose on The eve AP for ua-link nd th cal inf used t setup.so the virtual deficit round robin algorithm works like This scheme is indicated by"full duplex without SIR MAP eight for captu MAC case four,neither throughput nor fairness is satisfactory.In random access.As shown in Fig.12.its throughput decreases TANG AND WANG: MAC FOR EFFICIENT COEXISTENCE BETWEEN FULL- AND HALF-DUPLEX COMMUNICATIONS 5881 Fig. 10. A typical topology with 40% hidden nodes. TABLE III THE EXPLANATION OF THE 8 CASES IN THE FIRST SET OF EXPERIMENTS even lower than that of case one and case two, because some dual links lead to lower transmission rate. When the AP only establishes dual links with hidden nodes (i.e., case four), the throughput is improved by 16% and 6% as compared to case one and case two, respectively. However, with capture effect, the throughput of case six can be improved by up to 39% and 26% as compared to case one and case two. The throughput of case six (β = 2) is a little higher than that of case five or case seven. The reason is as follows. β of case five is lower than that of case six, so the average transmission rate of dual links in case five is lower than case six. However, for case seven, the average transmission rate of dual links is higher than case six but the capture probability is lower, which leads to a little lower throughput than case six. The throughput of the last case is the highest, since the AP establishes dual links with a client with the best capture rate. We consider fairness of downlink transmissions among all clients with respect to transmission time. Jain index [25] is used to measure the fairness. The results are shown in Table IV. In case one and case two, the deficit round robin algorithm is applied, so perfect fairness is achieved. From case three to case eight, the virtual deficit round robin algorithm is applied. In case three, the throughput is the lowest, but perfect fairness can be achieved. The reason is that, without setting a limit on Tadd, most clients can be selected by the AP for dual-link setup, so the virtual deficit round robin algorithm works like the deficit round robin algorithm. In case eight, the highest throughput is achieved, but the fairness is lowest, because no fair scheduling is considered in downlink transmissions. In case four, neither throughput nor fairness is satisfactory. In Fig. 11. The throughput in a wireless LAN with 40% hidden nodes. TABLE IV THE SHARE TIME OF DOWNLINKS FOR EACH CLIENT FOR THE CASES IN FIG. 11 case four, only hidden nodes are selected for dual-link setup. When the virtual deficit round robin algorithm is applied, it is possible that the default round robin order is changed but no proper client can be selected for dual-link setup, which leads to poor performance in both throughput and fairness. This result indicate that the virtual deficit round robin algorithm is not fit for case four. However, as explained in the end of Section V, we can set a limit to the negative deficit to improve the performance of case four. In cases five to seven, the virtual deficit round robin algorithm works effective, as both hidden nodes and capture effect are considered in dual-link setup, so high performance is achieved in both throughput and fairness. In the second set of experiments, we evaluate throughput with respect to the number of clients. For each number of clients, we create 50 random topologies and calculate the aver￾age throughput among all these topologies. β is set to 2, which is shown to be a proper value in the first set of experiments. In the first case, we set the data frame length to 1500 Bytes. As shown in Fig. 12, the throughput of 802.11 DCF without RTS/CTS decreases as the number of clients grows, since collisions occur more frequently. The throughput of 802.11 DCF with RTS/CTS is stable even if the number of clients increases. In comparison, A-Duplex achieves stable and much higher throughput. For example, considering 25 clients, our MAC protocol improves throughput by about 48% and 188% as compared to the 802.11 DCF with RTS/CTS and without RTS/CTS, respectively. We also simulate the scheme proposed in [6] where only the historical information is used to search asymmetric dual links and the capture effect is not considered. This scheme is indicated by “full duplex without SIR MAP”. Although the MAC protocol in [7] considers non-hidden nodes for capture effect, it is not compared with our MAC protocol, because it is a reservation based MAC protocol instead of random access. As shown in Fig. 12, its throughput decreases