Network Animator

Figure 4.3 is the network animator (NAM) result of our AFEC in RTP session of NS-2.

We can simulate the RTP and RTCP packet transmission with Ethernet and packet loss event.

Because of RTCP report packets may be loss in our simulation environment, it’s more close to real world network environment. We can also control the time of node join RTP session by Tcl (Tool Command Language) in NS-2, it means that we can discuss about multicast AFEC in our future work.

Figure 4.3 AFEC simulation

Random loss is the most familiar event of network transmission when network congestion occurred. We assume that packet random loss is Poisson distribution. Poisson counting process has three assumption 1)At most one event can occur at any time instant 2)Event count are mutually independent random variables for any interval 3)Event count may depend on the interval length, but it is independent of the time instants for any interval.

4.4 Summary

We propose a H-Q equation to design three AFEC schemes– static AFEC (SAFEC), dynamic AFEC (DAFEC), and advanced AFEC (AAFEC). SAFEC is simply use H-Q equation as adaptive controller in sender, but its original data block size is static. According to our simulation result that original data block size is also has effective to receiving quality in our H-Q equation, so that we modify SAFEC to be DAFEC. DAFEC can dynamically change original data block size by computing network packet loss rate. One more important thing is that it is needed to reduce the transmission overhead when network environment too worse to match our expected receiving quality, because it may cause more serious network congestion to effect other network traffic. Therefore we propose advanced AFEC (AAFEC) having self-close AFEC scheme can release network bandwidth if necessarily.

Chapter 5

Analysis and Comparison

5.1 Analysis

We propose a mathematical equation, which named H-Q equation, to adaptively control the number of redundant packets in AFEC. We use this equation to propose three AFEC schemes – SAFEC, DAFEC, and DAFEC, to adaptive deal with random loss network environment effectively.

5.1.1 SAFEC

The H-Q equation is used to design the adaptive controller of AFEC, which called static AFEC (SAFEC). It is compared SAFEC with non-AFEC in Figure 5.1, SAFEC has better receiving quality and performance to tolerate more packet loss rate than non-AFEC, because SAFEC quality curve is always higher than non-AFEC. We set receiving quality lower bound is 90%, so that the adaptive controller do not start when packet loss rate is lower than 10%.

Thus the SAFEC curve is similar to non-AFEC when loss rate is lower than 10%. In this case, SAFEC can only tolerate network loss rate below 20% if we set receiving quality lower bound is 90%. When loss rate is larger than 70%, RTCP report may lost in this network environment, so that the static AFEC can not adjust redundant packets immediately.

SAFEC Performance

Figure 5.1 SAFEC vs. non-AFEC

○

○

Generally, as the transmission overhead (H) increase the receiving quality (Q) increasing too. But we can not increase H boundless, because H has great effect upon the packet redundancy. By the way, in our adaptive control design program, we set redundant packets transmission can be interrupt when new RTCP report packets are received. Therefore redundant packets of each transmission block should be limited. Figure 5.2 shows our analysis of Q vs. H up-bound. We can find that maximum H=2 has better receiving quality in different packet loss rate. Therefore we set maximum H=2 to analysis following 3 AFEC schemes simulation.

From above analysis we know that transmission block size also have great effect of receiving quality in our H-Q equation design. H up-bound is set to constant 2, and then the transmission block size is related to original data block size. Figure 5.3 analyzes the effect of data block size versus the receiving quality in different packet loss rate when we fix number of redundant packets. We can find that small data block size has better receiving quality but higher packet redundancy. In packet redundancy illustrate, Figure 5.4, we expect that packet redundancy can be 0% when packet loss rate is lower than 10%. When packet loss rate is too large, it is expected that adaptive controller can stop redundant packet scheme in order to release network bandwidth.

Performance

Figure 5.3 SAFEC performance analysis

SAFEC Performance

Figure 5.4 SAFEC packet redundancy analysis

SAFEC has better performance than non-AFEC in packet loss rate tolerably. The performance of adaptive controller is related to the receiving quality tolerate value, redundant overhead up bound, and original data block size. General multimedia transmission can tolerate packet loss 8%~10%, so that we set our threshold receiving quality is 90%. According to our simulation result we set up-bound of H to 2, because it has better receiving quality in

○

○

our design program although it has higher packet redundancy. Finally, we find that original data block size has effect upon the receiving quality. We can let original data block size larger when packet loss rate is lower than 10% to reduce packet redundancy. On the other hand, small original data block size has more redundant packets per unit time, according to simulation result that it can tolerate more packet loss rate but it also increase packet redundancy.

5.1.2 DAFEC

It is added dynamic original data block size scheme in adaptive controller of AFEC. We combine dynamic H-Q model into AFEC to be dynamic AFEC (DAFEC). Figure 5.5 shows that DAFEC can tolerate 30% packet loss rate to maintain receiving quality higher than 90%.

It shows that DAFEC has better performance than SAFEC. The DAFEC curve is similar to non-AFEC when L≦10%, because we expect DAFEC do not inject too much redundant packets in our design when receiving quality Q≧90%. In Figure 5.6 we can find that packet redundancy still below 5% at L=10%. The receiving quality maintain 91% at L=32%, but packet redundancy is 62% this moment. The maximum tolerate packet redundancy we concerned, as the case may be.

DAFEC Performance

Figure 5.5 DAFEC performance

○

○

DAFEC Performance

Figure 5.6 DAFEC packet redundancy

DAFEC has better performance than SAFEC and non-AFEC. The packet redundancy is lower than 65% in our design program. We still can adjust the redundancy up-bound by setting H of DAFEC. As the packet loss rate increasing, the receiving quality can not fit our threshold quality (90%). DAFEC still can not release the network bandwidth. The H-Q model is still working even though receiving quality too bad to reach QW. It may effect others network traffic when we inject too much useless redundant packets.

5.1.3 AAFEC

Although DAFEC has good performance, it can not release network bandwidth in worse network environment. Therefore we add self-close functionality into DAFEC to be AAFEC.

In Figure 5.7, we set the criteria of self-close scheme is Q≦50%. The adaptive controller will start self-close scheme when receiving quality lower than 50%. Because the packet loss rate is oscillatory, the AAFEC curve is not immediately release network bandwidth when 0.5≦L≦0.7. The adaptive controller will check the gape between Q and QW when the receiving quality higher than 50%. If the gape is too large, we will re-estimate packet loss rate.

Otherwise, we will base on the RTCP report to compute new H. If the Q is better than QW, means that network environment is better than before, so that we try to reduce the R and raise original data block size. On the other hand, we will reduce Nori and increase redundant packets to converge the Nred transmission interval time. Finally, we take new H and new original data block size to compute number of R.

AAFEC Performance

Figure 5.7 AAFEC performance

AAFEC Performance

Figure 5.8 AAFEC packet redundancy

The Figure 5.8 shows that packet redundancy is 0% at L≧0.7, because the self-close scheme is stop the advanced H-Q model to inject redundant packets. At this time, AAFEC is similar to non-AFEC. AAFEC don’t start adaptive controller, until receiving quality Q≧50%.

The different between AAFEC and DAFEC is that AAFEC has self-close scheme to prevent critical network congestion. It not only inherits all the advantage from DDAFEC, but also concern about others network traffic during network congestion. The most advantage of AAFEC is that it can release network bandwidth when network congestion.

○

○

○

○

5.2 Performance Analysis

We propose a H-Q equation and develop three AFEC schemes which have efficiently enhancement of receiving quality performance within network packet random loss. Figure 5.9 shows three versions AFEC and non-AFEC performance comparison. The receiving quality of theses three AFEC schemes are higher than non-AFEC. It shows that we propose adaptive controller can efficiently enhance receiving quality. DAFEC performance is the best of all, but it is also shows that adaptive controller is still working even though network congestion occurred. AAFEC inherit DAFEC advantage, its curve is fit DAFEC when L≦0.5. The most important thing is that AAFEC has self-close scheme to release network bandwidth to avoid serious network congestion occurred.

SAFEC Performance

Figure 5.9 Three AFEC various performance

SAFEC Performance

Figure 5.10 Three AFEC schemes packet redundancy analysis

Figure 5.10 shows that SAFEC, DAFEC, and AAFEC occupy 60% network bandwidth when 0.2≦L≦0.6. This transmission lower bound is related to redundant overhead we set.

This packet redundancy up-bound can be adjusted, if we change H maximum value.

We analyze the performance of SAFEC, DAFEC, AAFEC, and non-AFEC when network environment sudden change. Figure 5.11 is the network environment. Figure 5.11 shows that we propose three AFEC various and non-AFEC receiving quality oscillation when packet loss rate changes from 0.1, 0.3, 0.5, 0.1, 0.7, and 0. These three AFEC various receiving quality are higher than non-AFEC. The SAFEC performance is not good enough as DAFEC and AAFEC when packet loss rate L≧0.3. At the same time, DAFEC and AAFEC still can maintain receiving quality around 90% and do not have terrible quality dip. In particular, DAFEC and AAFEC quality curves are similar beside at packet loss rate L = 0.7, because AAFEC has self-close scheme when network congestion occurred. At this time, AAFEC stop redundant RTP packets so that receiving quality curve similar to non-AFEC.

Packet Loss Rate

0 200 400 600 800 1000 1200 1400 1600

Time

Rloss

Figure 5.11 Network environment

Performance

0 200 400 600 800 1000 1200 1400 1600

Time(Sec.)

Figure 5.12 SAFEC, DAFEC, and AAFEC performance analysis

5.3 Comparison

FEC and AFEC are using redundant packets to recover lost packets to enhance perception quality. But FEC estimates network environment only once before it start. If network environment is static and accurately estimated, the FEC has the best performance. The problem is network environment always dynamically change, FEC can not dynamically adjust number of redundant packets to fit this situation but AFEC can do it. AFEC can dynamically

estimate network packet loss rate to adjust number of redundant packets. The feedback controller of AFEC using objective function to enhance AFEC performance but this function is hard to drive a closed-form. In order to simplify the objective function, it assumes that if the number of received packets is greater or equal than the original packets, then it is assumed no packet is lost during the transmission. Its network environment assumes that the loss probability of each packet is static and independent loss. EAFEC under the same assumption of feedback controller, it provide a algorithm to improve video delivery in wireless Access Point (AP). EAFEC tunes FEC packet numbers according to network traffic load and wireless channel state. However, there are four threshold values of adaptive controller can not dynamically adjust. According to Table 5.1, we know that FEC can not dynamically adjust the redundant packets quantity, although feedback controller and EAFEC can solve it but otherwise they are only executed under some unrealistic assumption. Therefore feedback controller and EAFEC can not fit every network environment. We propose AFEC schemes in this thesis can efficiently adjust number of redundant packets by dynamically estimate network environment.

Table 5.1 Comparison of FEC and AFEC schemes AFEC FEC

Feedback Controller EAFEC Our schemes

Description ¾ Network

environment never greater or equal than the original packets, then it is assumed no packet is lost during the transmission

¾ The loss probability of each packet is static

¾ Independent loss

¾ If the number of received packets is greater or equal than the original packets, then it is assumed no packet is lost during the transmission

¾ Static block size to 8 video packets

¾ Receiving quality only related to packet loss and delay

¾ The loss probability of each packet is static

¾ Independent loss

¾ Upper bound of the overhead (H) is adjustable

Disadvantage ¾ Can’t adaptive

adjust redundancy to suit network environment

¾ Underestimate the optimal

We propose H-Q equation can find the optimal value between perception quality and number of redundant packets. Furthermore, the redundant packets will occupy partial network bandwidth and then reduce the transmission data quantity per unit time. Thus packet

redundancy is also an important issue, but many papers do not discussion about this but in this thesis we have detailed analysis. Because the packet redundancy is a very important problem, we set upper bound of the overhead (H) to limited number of redundant packets. In order to avoid too large percentage of network bandwidth are wasted on sending redundant packets.

Besides, the network packet lost event can also cause the report packets lost. It makes the adaptive controller very unreliable. The report packet lost may cause the adaptive controller out of control and starve for receiver information. But many papers do not mention to this problem and do not consider this event influence into their network simulation. Otherwise FEC, EAFEC, and feedback controller can not release the network bandwidth which occupied in sending redundant packets when network packet rate below the tolerable value. In our simulation result shows that data block size is also influence the receiving quality. Our schemes consider much probable situation in our simulation environment (Table 5.2).

Table 5.2 Comparison of adaptive controller and simulation result

AFEC

FEC Feedback Controller EAFEC Our schemes

Adaptability No Yes Yes Yes

Mathematical Equation No Yes* Yes Yes

Loss of the Report Packets N/A Inaccurate estimation

Inaccurate estimation

Accurate estimation Disable the Redundancy for

high loss rate No No No AAFEC

Adjustable Size of Data Block No No No Yes

*The objective function is unimodal and monotone on each side of the peak

In Table 5.3, we compare the performance of FEC and AFEC on loss rate 50%. The 4_FEC means that producing 4 redundant packets per 8 original packets in FEC. Although recovery rate in FEC is better than EAFEC under random loss model, the FEC can not dynamic adjust its number of redundant packets. We can find that recovery rate of our schemes are better than EAFEC at Hmax =1, but it is trade-off redundancy rate. The EAFEC recovery rate and redundant rate are worse than SAFEC, but similar to DAFEC at Hmax =2.

But in this table, we can not realize the advantage of AAFEC. Because we set self-close scheme of AAFEC threshold value is loss rate 50%, therefore AAFEC starts to close it self at this time.

Table 5.3 Performance of FEC and AFEC on loss rate 50%

(1-BL/BL_nonAFEC) 98% N/A 60%

AAFEC 27% 57%

In Table 5.4, we change the loss rate on 70%. We can find that the recovery rate of FEC is decreasing to 65% when redundancy rate is keeping in 50%. The recovery rate of our schemes (SAFEC and DAFEC) is higher than other schemes; moreover redundancy rate is lower than other schemes. The performance of our schemes is better than EAFEC and feedback controllers. We can find that our schemes can tolerate more packet loss rate than other schemes. At loss rate 70%, AAFEC has already started self-close scheme so that recovery rate and redundancy rate are both zero.

Table 5.4 Performance of FEC and AFEC on loss rate 70%

AFEC

Service provider can choose one of our schemes according to network environment to provide better transmission performance. According to our analysis tables, all schemes are similar when packet loss rate is lower. As the packet loss rate is increasing, our schemes are getting better than other schemes. Therefore our schemes can accurate at estimation of network environment.

5.4 Summary

We propose a H-Q equation to design three AFEC schemes– static AFEC (SAFEC), dynamic AFEC (DAFEC), and advanced AFEC (AAFEC). SAFEC is simply use H-Q equation as adaptive controller in sender, but its original data block size is static. According to our simulation result that original data block size is also has effective to receiving quality in our H-Q equation, so that we modify SAFEC to be DAFEC. DAFEC can dynamically change original data block size by computing network packet loss rate. One more important thing is that it is needed to reduce the transmission overhead when network environment too worse to match our expected receiving quality, because it may cause more serious network congestion to effect other network traffic. Therefore we propose advanced AFEC (AAFEC) with self-close scheme to release network bandwidth if necessarily.

Chapter 6 Conclusions

In this paper, we propose a mathematical equation, which named H-Q equation, to adaptively control number of redundant packets in AFEC. We use this equation to propose three AFEC schemes – SAFEC, DAFEC, and DAFEC, to adaptive deal with random loss network environment effectively. SAFEC is simply using H-Q equation without consider about parameter sensitivity in forward error correction. We combine dynamic adaptive controller into AFEC to be DAFEC. It is adding dynamic original data block size scheme in DAFEC. The simulation results show that DAFEC can enhance more receiving quality than SAFEC, but it also increases more packet redundancy than SAFEC. SAFEC and DAFEC have too many useless redundant packets when network loss rate is too large, because the quality enhancement is limited by network packet lost rate. On the other hand, too much redundant packets may cause serious congestion problem. Therefore we propose advanced adaptive controller into AFEC to be AAFEC which can solve this useless redundant packets problem. AAFEC has self-close scheme to stop sending useless redundant packets to release network bandwidth, when network loss rate is too large. We not only analyze the performance of these three AFEC schemes but also packet redundancy. The simulation results show that receiving quality enhancement always trade-off packet redundancy whatever AFEC type we choose.

Chapter 7 Future Work

In our paper, we propose an equation, H-Q, for designing the adaptive controller in AFEC to facilitate end-to-end transport of real-time traffic. The receiver can use redundant packets to recover lost packets. But packets lost events, packet searching, and receiver packet queue length maybe cause congestion problem of receiver side. This is another real-time constraint problem. So that a good packets sorting scheme can improve packets recover speed and receiver queue length. If we change redundant packets to redundant frames (data) can also improve this problem. One packet can carry many frames inside including redundant data.

Therefore it can save the space of redundant packet header.

Packet lost event of our simulation environment is using random model. In real network environment, the burst event is also an important issue, especially in wireless network.

Packet lost event of our simulation environment is using random model. In real network environment, the burst event is also an important issue, especially in wireless network.