880 LEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS,VOL 14.NO.10.OCTOBER 2015 Mode Modulation Coding rate (dB)7 (dB) BPSK 12 5-6 1-4 BPSK 4 67 45 QPSK 12 7-9 57 9-13 7-12 16-QAM 12 13-17 12-16 616-QAM 3/417-2016-21 64-QAM 320-2221-23 864-QAM 34≥222223 Number of The path loss in the channel model is given by Fig.9.The throughput results under the protocol model PL(d=PL(do)+l0×n×log(d/do)[dBl, where we set do I m and n =3.and let PL(1)=48 dB to re Parameter Value Parameter flect the indoor environment.Moreover.the minimum required SNK to nt modu PHY header (PH) tion and coding schemes fo 20 PHY P amble RTS(802.1 effect((B))in Table I.where is adopted from PHY ion [23]andy is from experimental results in [12]. Pavload idefault 14Byte +PH me adopted in our simulatio Link rate 18 Mbps 16 three times Capture rate 12 Mbps Wo 16 next lower level.When a client succeeds in 5 consecutive Slot time 9 us m 6 SIFS 16us mo transmissions or the rate is not changed over 10 times.the DIFS 345 capture threshold 5dB rate rises to AP does not establish dual links when the SIR MAP is empty in the beginning can get n Throughput an Fa app in Ea (3).Since it is e for in s valid.as simu ractical s stem.we consider discrete values of=n.(n ts are con y23w res up .2.3. .and then determine a proper value of n.When we ans the new SIR e and 246 e is the latest u and 54%over 802.11 DCF without RTS/CTS for 5 clients and dual links is about 55%.When we set =3.the succe 40clients.respectively. nt can be achieved.As a result.we set n 3 in B.Performance Results Based on the Physical Model In the fr of 1)Plysical Model Setup:The physical model simulation is considered,and 40%clients(sl.s2.s3.s4)have hidden nodes conducted in Matlab.In addition to thr oughput,other perfo The frame length is set to1500 Bytes.In this.8 case s such .packet re sh ind compare own in la The fir parameters are given in Table I except that link rate.capture rat 802.11 DCF without using RTS/CTS.The second case is the (i.e.the link rate under capture effect).and capture threshold same as the first case except that RTS/CTS is enabled.The third und the AP.W and eh fad study the ing.Thus,the channel is a time varying channel.However.we A-Duplex when the parameter B is set to 1.2,and 3.respec tively.The last case is that the APal ways chooses the client with (in dua nin Fig.I5880 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 14, NO. 10, OCTOBER 2015 Fig. 9. The throughput results under the protocol model. TABLE I PARAMETERS USED IN SIMULATIONS to be 12 Mbps. To support such a rate, the capture threshold is 5 dB [12]. From Eq. (23) we can get Pca = 0.4371. Both simulation (marked by symbols) and analytical (represented by line) results of throughput are shown in Fig. 9. From these results, we know that: 1) The analytical model is valid, as simu￾lation results are consistent with analytical results; 2) A-Duplex improves throughput by 23% and 24% over 802.11 DCF with RTS/CTS for 5 clients and 40 clients, respectively, and 24% and 54% over 802.11 DCF without RTS/CTS for 5 clients and 40 clients, respectively. B. Performance Results Based on the Physical Model 1) Physical Model Setup: The physical model simulation is conducted in Matlab. In addition to throughput, other perfor￾mance metrics such as fairness, packet loss, and packet delay are also considered in performance evaluation. The system parameters are given in Table I except that link rate, capture rate (i.e., the link rate under capture effect), and capture threshold are determined dynamically. The clients are set in a wide area uniformly distributed around the AP. We consider a propagation model with deterministic power attenuation and Rayleigh fad￾ing. Thus, the channel is a time varying channel. However, we assume that the channel only changes between packets, i.e., the channel parameters remain the same during the transmission of each packet. TABLE II THE MINIMUM REQUIRED SNR FOR A CLEAN CHANNEL (γ1 (dB)) AND THE MINIMUM SIR FOR CAPTURE EFFECT (γ2 (dB)) The path loss in the channel model is given by PL(d) = PL(d0) + 10 × n × log(d/d0) [dB], where we set d0 = 1 m and n = 3, and let PL(1) = 48 dB to re- flect the indoor environment. Moreover, the minimum required SNR to decode different modulation and coding schemes for a clean channel (γ1 (dB)) and the minimum SIR for capture effect (γ2 (dB)) are given in Table II, where γ1 is adopted from [23] and γ2 is from experimental results in [12]. The rate adaptation scheme adopted in our simulation is ARF [24]. More specifically, when a client misses an ACK three times consecutively, its transmission rate moves to the next lower level. When a client succeeds in 5 consecutive transmissions or the rate is not changed over 10 times, the transmission rate rises to the next higher level. The capture rate is determined according to the SIR MAP table. Note that the AP does not establish dual links when the SIR MAP is empty in the beginning. 2) Saturation Throughput and Fairness: To conduct exper￾iments, we first need to determine an appropriate value for θ in Eq. (3). Since it is hard to find an optimal value for θ in a practical system, we consider discrete values of θ = 1/n, (n = 1, 2, 3 ···) and then determine a proper value of n. When we set n = 1, which means the new SIR value is the latest updated value SIRnew = SIRupdate, the successful transmission ratio of dual links is about 55%. When we set n = 3, the successful ratio increases to 80%. However, when we increase n further, little improvement can be achieved. As a result, we set n = 3 in the following simulations. In the first set of experiments, 10 clients shown in Fig. 10 are considered, and 40% clients (s1, s2, s3, s4) have hidden nodes. The frame length is set to 1500 Bytes. In this scenario, 8 cases are simulated and compared as shown in Table III. The first case is that all clients and the AP transmit packets based on the 802.11 DCF without using RTS/CTS. The second case is the same as the first case except that RTS/CTS is enabled. The third case is our MAC but setting no limit to the additional time Tadd. The fourth case is that the AP only establishes dual links with hidden nodes. The next three cases study the performance of A-Duplex when the parameter β is set to 1, 2, and 3, respec￾tively. The last case is that the AP always chooses the client with the best capture rate to establish dual links. As shown in Fig. 11, if we set no limit to Tadd (i.e., case three), the throughput is