Performance Comparison

3 4 5 6

BER

System Throughput (Mbps)

N=5, sim N=5, ana N=10, sim N=10, ana N=20, sim N=20, ana

Figure 2.7: Performance validation for ARCR protocol: system throughput versus BER.

It can be intuitively observed that the system throughput increases as the total number of WSs in the network is augmented. Moreover, Fig. 2.7 illustrates the throughput versus BER under N = 5, 10, and 20. The throughput performance decreases as the BER values are increased. It can be seen from both figures that the proposed analytical model can match with the simulation results under different numbers of WSs and BER values.

2.5.2 Performance Comparison

As shown in Figs. 2.8 to 2.11, the proposed ARCR protocol is compared to the DCF, GDCF [20, 21], PCF, and EBA-2 [17] schemes through a series of simulations in terms of both the number of WSs and the BER values. The system parameters in Table 1 are uti-lized in performance comparison with saturation traffic considered for each WS. It is also assumed that the successful counter c of GDCF is set equal to 2. Fig.2.8shows the perfor-mance comparison of system throughput w.r.t. different numbers of WSs under BER = 0 and 10⁻⁵. It can be observed that the proposed ARCR scheme possesses higher through-put performance than the other three protocols under different numbers of WSs. The only exception is at the case of N = 1 that the EBA-2 scheme provides better performance than the ARCR protocol mainly due to the reason that the parameter W is selected to be 8 and 32for EBA-2 and ARCR schemes, respectively. It is intuitively that smaller W will

en-0 10 20 30 40 50

Figure 2.8: Performance Comparison for ARCR protocol: system throughput versus num-ber of WSs.

Figure 2.9: Performance Comparison for ARCR protocol: system throughput versus BER.

0 2 4 6 8 10

Figure 2.10: Fairness comparison for ARCR protocol: fairness index versus time.

hance the throughput performance if there exists only one WS in the network. The EBA-2 protocol has better performance than the GDCF and DCF schemes since most of the WSs are not allowed to choose the same slot to transmit their RTS packets. Collision overhead and backoff delay can therefore be reduced. Moreover, the GDCF method is slightly supe-rior to the DCF scheme with larger number of WSs in the network. The reason is that the GDCF scheme has higher probability of staying at the stages with larger backoff window sizes compared to the DCF protocol. Less packet collisions will be incurred by adopting the GDCF scheme especially under larger number of WSs, which results in higher system throughput.

Fig. 2.9 illustrates the comparison of throughput performance versus different BER values under N = 5 and 50. The proposed ARCR protocol still outperforms the other three schemes under various BER values. It can also be observed that the system throughput of four schemes decrease and converge with the augmentation of BER values. At higher BER values, the proposed ARCR protocol behaves similar to the DCF scheme since almost all the WSs in the network will be removed from the reservation table due to occurrence of packet error . On the other hand, with higher BER values, the GDCF method is also comparable to the DCF scheme owing to the reason that its backoff stage will eventually remain at the maximum value. Furthermore, the fairness index F [36] is introduced to

estimate the fairness of each protocol, which is formulated as

F = (∑N

i=1S_i )2

N ·∑_N

i=1S_i² (2.16)

where Sirepresents the throughput of WS i. The maximum value of F is equal to 1 which indicates the fairest situation; while minimal value F = 1/N denotes the most unbalanced case. Intuitively, the value of F will be approximated to 1 for the steady state response of each protocol since every WS has equal opportunity to contend the channel. Therefore, the rates of achieving steady state for all the schemes are compared in order to determine the fairness levels. Fig.2.10shows the fairness comparison F versus time under different numbers of WSs. It can be observed that the proposed ARCR protocol can quickly achieve the steady state value of the fair index, i.e., F = 1, compared to the other schemes. On the other hand, both the GDCF and EBA-2 schemes result in worse performance compared to that of the DCF protocol due to their inherent designs. The WS with RTS packet collision will suffer from larger backoff delay for the GDCF scheme than that for the DCF proto-col since the WS will slowly return back to the minimum CW size in the GDCF scheme, which makes the GDCF scheme become more unfair between the WSs. Furthermore, the WSs that receive the EBA field from the other WS should reselect another backoff counter, which is advantageous for the WS to easily win the channel contention. Therefore, the EBA-2 scheme possesses inferior fairness compared to the DCF protocol. Note that it takes more time for the fairness index to achieve steady state with the N = 30 scenario compared to the N = 10 case.

Moreover, the proposed ARCR protocol is compared with the distributed DCF scheme and the centralized PCF protocol given the condition of non-saturated queue in each WS with M/G/1 queuing system. In order to illustrate the pure reservation-based system, the PCF scheme is implemented only with the CFP while the CP is not considered in performance comparisons. It is assumed that there are two types of WSs in the network, including the WSs with high packet-arrival rate (λ1 bits/sec (bps)) and with low packet-arrival rate (λ2bps). Let n_λ1 and n_λ2 be respectively defined as the numbers of WSs with λ1

and λ2as the packet-arrival rates, the corresponding average throughput for each WS with λ₁and λ2is respectively denoted as µ_λ1 and µ_λ2 with the unit of bps. It is considered that there are total of 10 WSs in the network for performance comparison, i.e., n_λ1 + n_λ2 = 10.

The performance comparisons of average throughput for each WS (i.e., either µ_λ1 or µ_λ2) versus the number of WSs with the packets arrive rate equal to λ1 are shown in Fig.

1 2 3 4 5 6 7 8 9 Average Throughput of Each WS λ 1 or λ 2 (Mbps)

ARCR, λ

Figure 2.11: Performance Comparison for ARCR protocol: average throughput of each WS versus number of WSs with λ1 (n_λ1) (n_λ2 = 10− n_λ1, λ1 = 2Mbps, λ2 = 0.2Mbps).

150 200 250 300 350 400 450 500 550 600 650

0 Average Throughput of Each WS λ 1 or λ 2 (Mbps)

ARCR, λ₁

Figure 2.12: Performance Comparison for ARCR protocol: average throughput of each WS versus packet arrival rate λ2 (n_λ1 = 2, n_λ2 = 8, λ1 = 2Mbps).

2.11. Noted that the packet arrival rates λ1 = 2Mbps and λ2 = 200Kbps, and the num-ber of WSs with packets arrive rate λ2 becomes n_λ2 = 10− n_λ1. It can be observed from Fig. 2.11that the average throughput for the WSs with λ2 is approximately the same for all these three protocols, i.e., µ_λ2 ≃ 0.2 Mbps. The results indicate that all three schemes can provide satisfactory services for the WSs with packet arrival rate λ2 = 0.2Mbps since µ_λ2 is around the same as the theoretically maximal throughput for each WS with λ2. On the other hand, the effectiveness of proposed ARCR scheme can be revealed by observing the average throughput µ_λ1 for the WSs with λ1. As n_λ1 is small, the polling-based PCF scheme becomes inefficient comparing with the ARCR and the DCF protocols since the network bandwidth is wasted as the AP is scheduled to periodically poll the larger num-bers of WS with λ2. The proposed ARCR protocol and the DCF scheme can provide higher throughput for WSs with λ1 since there is more opportunity for these WSs to frequently transmit their data packets. Furthermore, with larger values of n_λ1, the contention-based DCF scheme will spend significant amount of time to resolve for packet collisions, which results in reduced system throughput of µ_λ1. The proposed ARCR protocol and the PCF scheme can provide higher throughput performance since there is greater chance for the larger amount of WSs with λ1 to be scheduled for packet transmission. As a result, the ARCR protocol can provide better throughput performance under different arrival rates of the WSs in the network.

Fig. 2.12illustrates the average throughput of each WSs (µ_λ1 or µ_λ2) versus the packet arrival rate λ2 on the conditions that n_λ1 = 2, n_λ2 = 8, and λ1 = 2 Mbps. As can be ex-pected, with the augmentation of λ2, the throughput µ_λ2 will be increased, however, the throughput performance µ_λ1 for the WSs with λ1 is reduced for all three schemes. A sat-uration point will be reached by the WSs with either λ1or λ2for all three protocols due to the availability of total network bandwidth. Owing to the severe packet collision, the DCF scheme will result in the lowest throughput among the three protocols with the earliest saturation point at λ2 ≃ 400 Kbps. Furthermore, the ARCR scheme will provide higher throughput performance with around µ_λ1 ≃ µ_λ2 ≃ 0.58 Mbps. On the other hand, the pro-posed ARCR scheme and DCF can outperform the PCF protocol under smaller values of packet arrival rate λ2owing to the polling overheads resulting from the centralized-based PCF scheme.

Moreover, in each channel competition, based on our proposed ARCR protocol, the first WS in the reservation table possesses the same range of CW size compared to the newly joining WSs. In other words, unsaturated WSs that may not have constant packet

to be delivered will have the same level of opportunity for channel contention with the saturated WSs. As shown from Fig. 2.12 that the average throughput of WSs with λ1 is around 1.32 Mbps and that with λ2 is around 0.2 Mbps at packet arrival rate λ2 = 200 Kbps. It can be seen that the throughput of the unsaturated WS with λ2 will always be identical to its packet arrival rate before the saturation occurs at around 580 Kbps. That is to say, the WSs with λ2will not encounter large backoff timer due to packet collisions. As the packet arrival rate λ2is increased to 580 Kbps, it can be observed that all the WSs will achieve around the same throughput performance. Therefore, the merits of adopting the proposed ARCR scheme can be perceived.