The error control experiment comprised of five measurements, with the measurement of no error control being the control case of the experiment.  The metric for measuring the quality of the video is the comparison of the signal-to-noise (SNR) ratio of the resulting video to the original video.  The measurements were no error control (the control case), HEC, and 4-, 8-, and 16-bit symbol RS FEC.  Figures 2 through 4 show the results of the experiments for each error control method with error rate as a function of the SNR.  Higher SNR ratio means better image quality.  A SNR ratio of 47 is a perfect duplicate of the original video, and an SNR of 0 indicates a premature termination of MPEG player due to decoder error.

figure3.gif (4934 bytes)
Figure 3
Error Rate and SNR Using RS FEC.

Generally the quality of the video decreases as error rate and/or burst error size increases.  It is observed from the experiment that an error rate less than or equal to 10-6, with small burst size (less than 10 bits), yields nearly perfect image quality.  Visually, errors are not perceived. However, this is subjective.  With error rates greater than 10-6, video quality starts to degrade severely without error control. Especially with a burst error size larger than or equal to 5-bit, the MPEG decoder encounters errors and thereby terminates, causing the MPEG player to quit prematurely.  Furthermore, the data does not show that with an error rate greater than 10-6 and a small burst error size, the video will not be played with the correct frame rate without error control.   It pauses every now and then and usually skips frames. Thus error control must be provided to guaranteed quality of service.

The RS FEC error control method provided superb error control.  It corrected all errors under all experimented conditions, except the 4-bit symbol RS FEC when the burst error size was larger than 10 bits. The

 

video quality degraded slightly in this case. It was found that a larger code symbol size of RS code is favored to handle larger burst size error; a smaller code symbol size is favored to handle scattered bit errors. Furthermore, by increasing the number of redundancies appended, the number of burst errors it can handle also increases. However, this is a tradeoff, which causes a larger overhead processing time in the encoding and decoding process.   Despite the superb error correction ability of RS FEC, it imposed a large overhead redundancy and overhead processing time. The redundancies imposed by the 4-, 8-, and 16-bit symbol RS FEC were 25, 14.4, and 32 percent, respectively.  The imposed redundancy by the 16-bit symbol RS FEC was unusually large because its block size is relatively large compared to the packet size; therefore extra padding was required. Large overhead redundancy and overhead processing time resulted in lower throughput and higher latency delay.  In either case, the number of concurrent clients a single server can handle decreased.  The performance of the system was decreased to maintain relatively perfect image quality.

figure4.gif (5008 bytes)
Figure 4
Error Rate and SNR using HEC.

The HEC error control method provided a relatively perfect image quality with error rate less than or equal to 10-6, an acceptable image quality with error rate of 10-5, and ensured the continuing playing of the video with error rate of 10-4.  The overhead redundancy was very small compared to the RS FEC error control method, 1.34 percent.  However, the overhead processing time is about the same as with RS FEC because of the parsing and the rearrangement of sequence headers. The software implementations for both proposed error control methods are not practical because of the large overhead processing time imposed.   Thus hardware implementation is recommended for such tasks.

back.gif (221 bytes) next

 

Page 5

 

Chin-Yu Chen - Error Control For Video On Demand ... [1] [2] [3] [4] [5] [6]