Chapter 4 THE FEEDBACK SCHEME OF WIMEDIA MAC FOR H.264/AVC SVC
4.4 Proposed Feedback Scheme
4.4.2 Adaptation
<Z<0 is represented of available percentage 0% and probability density isP0. 0 <Z<1 is represented of available percentage 1% and probability density isP1.99 <Z<
is represented of available percentage100% and probability density isP100. The formula for expectation value (E_MAS) is listed below:
consider the mechanism of frame transaction and the fragmentation, we can find out how much time can use to transmit data. After knowing the available transmission time, by multiplying the modulation rate shown in Table 3-3 which is affected by the SNR, we can get the bandwidth for transmission in each super-frame in the next GOP period.BW = Available_Time*Modulation_Rate (5)
4.4.2 Adaptation
The performance of the proposed estimation method is better than the compared methods as shown in Figure 5-2, but it is thought can be further improved.
Besides proper estimation, dynamical adjustment of the estimation which takes the video latency issue into consideration can efficiently handle the bandwidth fluctuations which may cause the video latency.
4.4.2.1 Relationship between MACbuffer and Decoderbuffer
In our performance analysis, we observed the variation of the fullness of decoderbuffer by inspecting the fullness of MACbuffer successively. As shown in Figure 4-1, we store the multimedia bitstream after the extraction process from SVC extractor in the MACbuffer and wait to be transmitted. Similarly, the decoder needs a decoderbuffert to store the received multimedia data for replay at the receiver.
We can adapt our estimation according to the fullness of MACbuffer because the fullness of MACbuffer can indicate the level of miss-estimation which also affects the fluctuation of fullness of decorderbuffer. If there are over-estimations for a period
of time as shown in Figure 4-4, the fullness of MACbuffer increase and the fullness of decoderbuffer decrease.
Figure 4-2 Relation between MACbuffer and decoderbuffer
4.4.2.2 Content-Based Adaptation
In order to help to reduce the chance of video latency, we take advantages of the scalabilities of SVC to propose an adaptation mechanism.
The adaptation mechanism modifies the estimation bitrate from step1 according on the fullness of MACbuffer dynamically. The modified size is shown in (6).
RateModify : the modified size of estimation bitrate (kb/s)
α : indicates the adjust ratio of the modified size and 0≦α ≦1 DataMACbuffer : the remaining data in the MACbuffer (kb)
Nsuperframe : the number of super-frames in a GOP period Tsuperframe : the time interval of a super-frame (s)
After obtaining the adjustment, the extraction decision engine will adjust the estimation bitrate according to (7) and send the modified transmission bitrate back to the SVC extractor for extraction.
Rate
Report= Rate
Estimation– Rate
Modify (7)RateEstimation : the estimation bitrate from step1 (kb/s)
RateModify : the modified size (kb/s)
RateReport : the modified transmission bitrate which is reported back to the extractor (kb/s)
Adjust the extraction multimedia bitrate by reference to the fullness of MACbuffer can efficiently reduce the chance of video latency when there comes up successively miss-estimations. If the fullness of MACbuffer gets much, it may note that the estimations are too high to fit the available bandwidth, so the adaptation will decrease the estimations more. On the contrary, if the fullness of MACbuffer is few, the adaptation will decrease little which may stay the quality of pictures. Figure 4-3 shows the flow of content-based adaptation
Figure 4-3 Flow of content-based adaptation
4.4.2.3 Improved Observation
Figure 4-4 and Figure 4-5 illustrate an example of the improvement of applying the content-based adaptation. After a period of miss-estimation, the conventional estimation method in Figure 4-4 leaded to the happening of video latency in the 8th super-frame. However with the content-based adaptation in Figure 4-5, the extraction decision engine will modify the estimation bitrate and report the modified bitrate, Report rate, back to the extractor. As shown in Figure 4-5, the content-based adaptation reduced the happening of underflow in the decoderbuffer in the 8th super-frame.
Figure 4-4 Example of conventional method
Figure 4-5 Example of content-based adaptation
CHAPTER 5
SIMULATION RESULT
5.1 Simulation Setup 5.1.1 SNR Generation Model
The working environment of WPAN may be indoor residential and commercial buildings and in [16] had shown the time-invariant characteristic of this kind indoor UWB channel. As introduced in [17], it measured the time snapshot of the channel at each spatial point over 20 commercial building and 20 residences to build database. In each building, it performed 25 measurements on a grid around 30 locations with transmitter-receiver separating from 0.8m to 10.5m for both line-of-sight (LOS) and non-line-of-sight (NLS) as illustrated in Figure 5-1.
Figure 5-1 Illustration of the spatial measurement in a typical building [17]
It invested the path loss model described below by data regression and proved the result of model is similar to the measured data through simulations. Because the model is regressed from the experimented database, it has included the effect of multipath and the shadowing effect but without mobility. The path loss model fixed PL0 at the mean taken over all buildings while , , and indicate the mean and standard deviation of independent Gaussian distribution and
respectively. The parameters of path loss model are different among building environments and listed in Table 5-1 Parameters of path loss model.
The deviation from the average path loss is affected by the combination of the zero-mean, unit variance Gaussian random variables x1, x2 and y. The part of
yx2y is usually referred as the shadowing [17].
Table 5-1 Parameters of path loss model
Environment
5.1.2 Available MASs Generation Model
We perform the simulations with two kinds of distribution of the available MASs in reservation to confirm if the proposed scheme works in other distribution.
5.1.2.1 Normal Distribution
Generate a Gaussian random variable with a set of pre-determined mean and variance repeatedly while keeping the mean and the variance fixed during the
PL0 r r
simulation. The Gaussian random variables Gi represent the percentage of available MASs and N indicates the total available MASs for the ith super-frame.
(9)
5.1.2.2 Poisson Distribution
Select a number as the mean in advance to build the Poisson distribution and choose a random number Pi as the percentage of available MASs of the ith super-frame successively.
(10)
5.1.3 Fragment Model
After WiMedia MAC was informed the information of SNR and negotiated reservation, the extraction decision engine is required to calculate how many bandwidth is available in each super-frame. Due to the overheads in WiMedia are heavy, the reserved MASs are not totally for transmission. We first check the DRP_available_IE to see how is the MASs‟ separation in the super-frame. Within each successively MASs, by adopting the I-mm-ACK policy as shown in Figure 3-5, take the overheads into consideration to calculate how many frames and the frame-length can be transmitted in the super-frame with the transmission rate provided by the modulation.
5.2 Simulation Results
In the performance analysis, we first check the performance of the proposed estimation. Then combined the content-based adaptation on the proposed estimation as our proposed scheme and compared with two other methods, Infinite Impulse Response (I.I.R) and least square method in three aspects, the probability of the decoderbuffer falling into null stage, the required size of MACbuffer, and the average percentage of the unused bandwidth, to analyze the performance of proposed scheme.
We performed simulations and got the average of results which are 10 times over 10000 super-frames each, on per aspect.
5.2.1 Compared Methods
5.2.1.1 Infinite Impulse Response (I.I.R)
Infinite Impulse Response (I.I.R) refers to the impulse responses with in infinite range. But it is impossible to implement the references of infinite steps, so we adopt I.I.R by regression. Because of the procedure of regression, the output of I.I.R is associated with not only the current data but also the output of last time as shown in the formula below.
In the following simulation, we apply the weightiness of current data P equals to 0.8 as reference to [13]. and the summation of the square of vertical distances which is from each data to the curve is shown as
To minimize E, the parameters m and bmust satisfy the following two equations.
Therefore, we can achieve the linear equation which best fits the data through
)
5.2.2 Performance of Proposed Estimation Method
5.2.2.1 Probability of Underflow
The meaning of underflow is defined as the happening that the decoderbuffer falls into the null stage where there is no data in the buffer while decoder is accessing, and it will cause video latency which majorly discomfort users and has the highest priority to be prevented. We speculated on the fullness of decoderbuffer by monitoring the variation of MACbuffer as discussed above and calculate the probability of happening of underflow over each 10000 super-frames.
In Figure 5-2, we show the average results of proposed estimation method in the simulations of normal distribution with different variance.
Figure 5-2Probability of underflow of the proposed estimation
5.2.2.2 Required Size for MACbuffer
The required size for MACbuffer is defined as the maximum value of monitored loading of MACbuffer ever seen in the simulation process in order to keep away from the overflow which may lose data from transmission happened in the transmitter.
The maximum value of MACbuffer can also indicate the stability of MACbuffer and the level of overestimation. The more overestimation will result in more data stored in the MACbuffer.
Figure 5-3Required size of MACbuffer of the proposed estimation
5.2.2.3 Conclusion of the Proposed Estimation
The proposed estimation method decides the bitrate for the SVC extractor.
The extractor uses the estimation bitrate to extract the multimedia bitstream for each GOP. Since the proposed estimation can‟t fit the actual bandwidth perfectly, it also causes the chance of video latency. From the simulation results in Figure 5-2 and Figure 5-3, we can see that the performance of the proposed estimation in reducing the video latency and the data stored in the MACbuffer are better than the compared
methods, but still can be improved. Therefore we may need the proposed content-based adaptation to adjust the estimation.
5.2.3 Performance of Proposed Scheme
We combined the proposed estimation and the content-based adaptation together as our proposed scheme to compare with the other two methods to see how is the performance of the proposed scheme in the three aspects.
5.2.3.1 Probability of Underflow
Through Figure 5-4, we can observe that the probability of underflow by the proposed scheme which combines the proposed estimation and adaptation, remains zero in cases with variances equal to 1%, 5% and 10%, and are near to 0.0005 with variances equal to 20% and 25%, while other methods‟ get higher and higher, up to 0.0142.
The probabilities of underflow result from the proposed scheme are much less than from the compared methods. The simulation results indicate that the proposed scheme has better performance in decreasing the probability of happening of the video latency, compare with the other two methods in the cases of normal distribution.
Figure 5-4 Probability of underflow (Normal distribution)
Figure 5-5 presents the simulation results in Poisson distribution. Obviously, the probabilities of underflow of the proposed scheme which remains zero in these
cases are much less than the other two‟s which can be up to 0.0176 in Poisson distribution. The observations are similar to the cases of normal distribution.
Figure 5-5 Probability of underflow (Poisson distribution)
5.2.3.2 Required Size for MACbuffer
The MACbuffer may need two separated parts to meet the requirement in the proposed content-based adaption while conventional methods may need only one. But from the simulation results presented in Figure 5-6, the required MACbuffer size is reduced by the proposed scheme in every cases of normal distribution. And in the cases of Poisson distribution, the proposed method can also obtain improved performance as shown in Figure 5-7.
Figure 5-6 Required size for MACbuffer (Normal distribution)
Figure 5-7 Required size for MACbuffer (Poisson distribution)
5.2.3.3 Percentage of Wasted Bandwidth
The wasted bandwidth is the bandwidth that available by reservation but unused due to the underestimations and may decrease the utilization of bandwidth resource as discussed. We averaged the wasted bandwidth and the total actual
available bandwidth within a super-frame. Then, calculate their ratio to be the percentage of wasted bandwidth. And we log the results of these percentages to study their current.
The simulation results shown in Figure 5-8 and Figure 5-9 demonstrate that the percentage of wasted bandwidth of the proposed scheme is higher than the two compared methods. The poor percentage of wasted bandwidth of proposed scheme also results from the idea that we treat the avoidance of video latency as higher priority.
It is really a pity to see that the proposed scheme results in more wasted bandwidth than the compared methods in the simulation. But, fortunately, due to the transmission bandwidth of WiMedia is so huge that the modulus of percentage result from each method is actually pretty small while the utilization of bandwidth will not decrease too much. And according to the reservation specifications in WiMedia MAC, there exist releasing scheme which allows of releasing the reserved but unused bandwidth to other devices, in soft-DRP or hard DRP as introduced in section 3.1.5.
Because the WiMedia MAC provides the release mechanisms help to solve the problem of utilization, so we treat the percentage of wasted bandwidth as a slight weak point of the proposed scheme but not deadly.
Figure 5-8 Log of Percentage of wasted bandwidth (Normal distribution)
Figure 5-9 Log of Percentage of wasted bandwidth (Poisson distribution)
CHAPTER 6 CONCLUSION
6.1 Conclusion
In this thesis, we clarify the flow of transmission within WiMedia and SVC in advance and propose the system architecture with a feedback scheme which takes advantages of the scalabilities from SVC and the characteristics of the WiMedia. The proposed scheme consists of the appropriate estimation and content-based adaptation in the transmitter in order to provide a better environment. The estimation is used to catch the current of the bandwidth variation and content-based adaptation can reasonably and efficiently reduce the probability of happening of video latency.
From simulation results of comparing with other two methods, I.I.R and least square method, the proposed scheme can offer not only a striking reduced probability of happening underflow in the decoderbuffer but also decrease the required size for MACbuffer. Especially when the variation of bandwidth is great, the effects of proposed scheme are evident. We also applied the simulations on different distributions of available transmission time and obtain similar conclusion which confirms the proposed scheme can work on other general distributions.
There exists a weak point of the proposed scheme through the simulation results. Because the major mission of the scheme not only precisely predict the bandwidth but also concern the latency issue, it obtained the higher decrease of bandwidth. Fortunately, in WiMedia system, the transmitted bandwidth is large and WiMedia supports to release the unused bandwidth in the DRP mechanism. These make the disadvantage of the proposed scheme not deadly but solvable.
6.2 Future Work
Besides the discussion had done above, there are some future works remain can be done for further analysis.
Implement the system with consideration of mobility together. Since the WiMedia is used for WPAN such as at home now, the location of devices will not change sometimes. So, in the simulations, we fixed the distances to get the SNR. Maybe WiMedia can be capable on other mobile applications to transmit
large scale multimedia such as HD, one day. The estimation and prediction methods can be further improved.
To consider and confirm the more precise situations on available time reservation. Because the causes for reserving the available MASs are too much and complicated, there doesn‟t exist a general form or distribution to module it.
Soft adaptation to improve the utilization of bandwidth of the proposed scheme.
Soft adaptation may consider more issues and details to discover a way.
References
[1] Hsiang-Chun Huang, Wen-Hsiao Peng, Tihao Chiang and Hsueh-Ming Hang,
"Advances in the scalable amendment of H.264/AVC," Communications Magazine, IEEE, vol. 45, pp. 68-76, 2007.
[2] Hsiang-Chun Huang, Chung-Neng Wang and Tihao Chiang, "A robust fine
granularity scalability using trellis-based predictive leak," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 12, pp. 372-385, 2002.[3] Yu-Jen Chen, Chen-Han Tsai, and Liang-Gee Chen, “Analysis and
Architecture Design for Multi-Symbol Arithmetic Encoder in H.264/AVC,”in VLSI Design/CAD Symposium, 2005
[4] J. del Prado Pavon, Sai Shankar N, V. Gaddam, K. Challapali and Chun-Ting
Chou, "The MBOA-WiMedia specification for ultra wideband distributed networks," Communications Magazine, IEEE, vol. 44, pp. 128-134, 2006.[5] http://www.multibandofdm.org [6] http://www.wimedia.org [7] http://www.ecma.org
[8] http://ieee802.org/15/pub/TG3a.html [9] http://www.edca.org
[10] Sun, Shenjin (December 2006). Performance Analysis Of The Beacon Period
Contraction Mechanism In Wimedia Mac For Uwb. University of Texas at Arlington. Available electronically from http://hdl.handle.net/10106/181 .[11] Chien-Yu Chen, “Comparison of MAC Layer Mechanisms for WiMedia
UWB and UWB Forum,”
[12] Jian-Liang Lin, Wen-Liang Hwang and Soo-Chang Pei, "SNR scalability
based on bitplane coding of matching pursuit atoms at low bit rates:fine-grained and two-layer," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 15, pp. 3-14, 2005.
[13] Peng Chen, Jeongyeon Lim, Bumshik Lee and Munchurl Kim, “A
network-adaptive SVC Streaming Architecture,” the 9th international conference on advanced communication technology, pp: 955-960, Feb.2007.[14] Haechul Choi, Jung Won Kang and Jae-Gon Kim, "Dynamic and
Interoperable Adaptation of SVC for QoS-Enabled Streaming," Consumer Electronics, IEEE Transactions on, vol. 53, pp. 384-389, 2007.[15] Q. Zhang, W. Zhu and Y. Zhang, "End-to-End QoS for Video Delivery Over
Wireless Internet," Proceedings of the IEEE, vol. 93, pp. 123-134, 2005.[16] D. Cassioli, M. Z. Win and A. F. Molisch, "A statistical model for the UWB
indoor channel," Vehicular Technology Conference, 2001. VTC 2001 Spring.IEEE VTS 53rd, vol. 2, pp. 1159-1163 vol.2, 2001.
[17] S. S. Ghassemzadeh, L. J. Greenstein, A. Kavcic, T. Sveinsson and V. Tarokh,
"UWB indoor path loss model for residential and commercial buildings,"
Vehicular Technology Conference, 2003. VTC 2003-Fall. 2003 IEEE 58th, vol. 5, pp. 3115-3119 Vol.5, 2003.