CHAPTER 4 RESULTS AND DISCUSSION
4.2 Evaluating the Algorithm on Adaptation Instance
The graph in Figure 4.1 shows the fact that in IEEE 802.11 WLANs, congestion does not happen gradually rather it happens instantly. It means sometimes network can sustain the extra traffic but sometimes by adding of a tiny extra traffic network tends to switch from an un-congested condition with good perceived speech quality to a congested condition with bad perceived speech quality .
In order to demonstrate that every transmission rate reduction does not need adaptation, some scenarios are conducted with different transmission rate and in all of them the calls use G.711 codec with one frame per packet and the transmission rate in one of the calls (a pair of source and destination) changes in descending form (from 11 to 5.5 then to 2 and end up with 1 Mbps).
As Figure 4.1 shows the first three rate reductions in one of the calls do not need adaptation mechanism, since the average MOS is maintained high (almost 4.3). But if this call drops to transmission rate of 1 Mbps, the MOS decreases to 1 which is unacceptable quality, so the adaptation approach is needed in this instance (not for all previous transmission rate reductions). Consequently, the system can withstand some transmission rate reductions until the network gets congested.
Figure 4.1: MOS results when the transmission rate of one call drops gradually.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
11 5.5 2 1
Transmission rate reduction in one call (in Mbps)
Figure 4.1 also shows only the transmission rate changing cannot be an adequate indication for triggering the adaptation mechanism. Unlike the previous works that they changed the coding rate with all variations in transmission rate (MAC alarm), the proposed method in this study only changes the coding rate when is needed.
The adaptation instance in the proposed algorithm is determined by monitoring the quality factors such as jitter and delay besides monitoring the transmission rate variations (as discussed in the methodology chapter). That is, when jitter and delay are going to increase sharply and beside that, MOS is going to decrease to the unacceptable range, adaptation should be commenced. Consequently, unlike some previous studies, the proposed algorithm by this study determines the adaptation instance not only by MAC alarm rather, beside MAC, quality factors will be checked to perform the adaptation timely and accurately.
in Figure 4.2, Codec-Adaptive graph demonstrates the codec adaptation method by Anna‟s work , for the scenario in that one of the call‟s transmission rate is changed to the lower transmission rate gradually. Besides, the non-adaptive graph also presented to compare these two methods in term of perceived quality.
Obviously, in the sufficient bandwidth, G.711 codec provides the best quality among the codecs (MOS 4.3) and other codecs (like G.729) provide lower quality (MOS 4) . In regard to the first three points of both graphs in Figure 4.2, when the bandwidth is still enough and network is not congested yet, the system can still continue with higher bitrate codecs and adaptation is not needed. In this situation, switching the codec from G.711 to G.729 causes quality degradation from 4.3 to 4 (first three points of Codec Adaptive graph). As previous codec adaptation approaches like Anna‟s work  change the codec for all transmission rate, they scarify the good quality of G.711, even in non-congested situation. But the proposed algorithm in this study trys to keep the G.711 till the real need of adaptation time (which is determined by quality factors beside transmission rate changes).
Figure 4.2: Comparison between MOS results of Anna‟s work  & non-adaptive method when the transmission rate in one of the calls is reduced gradually.
Evidently, as Figure 4.2 shows, for the last transmission rate reduction to 1Mbps, codec adaptation acts better than no adaptation. However, for the first three transmissions rate changes from 11 to 5.5 and then to 2 Mbps “No Adaptation” has the top quality, even better than codec-adaptive method, because it uses G.711 with higher output rate.
Accordingly, with the mechanism used in this study, the system does not enter to the adaptation phase till it would be required. The main advantage of this approach is cost reduction in term of changing the call's coding parameters frequently during the call.
So far, the right instance to commence adaptation is evaluated and hereafter the adaptation method will be evaluated. In fact, adaptation phase could be accomplished using codec rate adaptation and/or using frame size adaptation. The results in chapter 3 showed that in most of the cases frame size adaptation would be enough for the system to recover from congestion. It is due to grouping of more frames in one packet which cause less overhead to carry in the network so lower congestion would be expected. Also frame adaptation is more flexible and has less cost in terms of modification impact and it is free of paying license fees. Moreover, when the
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5
11 5.5 2 1
Rate reduction in one call (in Mbps)
transmission rate is changed slightly, the frame size adaptation can control the rate of traffic carried by network but codec adaptation changes the traffic volume hugely. So in the small congestion, frame size adaptation could be used instead of codec adaptation. That is why the proposed mechanism increases the number of frames up to three steps and checks the quality factors after each of them. If the speech quality turns into the acceptable range, the system does not undergo the codec adaptation but if frame size adaptation does not rectify the congestion then codec adaptation would be commenced. The rest of this chapter is to evaluate the proposed algorithm.