23
FEC and Multi-layer Video Coding For ATM Networks
MJ. Riley and I.E.G. Richardson
The Robert Gordon University, Aberdeen AB9 IFR, Scotland
Abstract
Significant advances have been made in recent years on standards [1][2][3] for video
compression. Compressed video data resulting from these coding standards is however
susceptible to quality degradation when cell losses occur during transmission over an ATM
network.
In [4][5] we have shown that multi-layer video coding using the Progressive Spectral
Selection (PSS) mode of the JPEG algorithm produces parallel traffic streams (or scans) which
show different tolerances to cell loss. Therefore, we apply varying quantities of FEC to each
of the video traffic scans. The scans which are least sensitive to cell loss are transmitted with
little or no error correction whereas the highly sensitive DC scan is heavily forward error
corrected. A real video sequence is transmitted through a simulated ATM network before re-
animation and subjective quality evaluation. We compare the video quality with that resulting
from FEC on a baseline JPEG CODEC.
We conclude that the quality degradations introduced by cell loss have different effects on
the two coding schemes. The base-line JPEG scheme with FEC provides excellent picture
quality until the FEC fails where-upon a severely degraded frame results. The PSS JPEG
scheme provides much higher FEC protection to the critical scan 0 data but less FEC to the
other scans. As the cell error rate increases this will result in a more gradual overall
deterioration in picture quality. In some applications the characteristics of the multi-layer
scheme may be more desirable.
Keyword Codes: 14.2, C2.1, C2.2
Keywords: Compression (Coding), Network Architectures and Design, Network Protocols
1. INTRODUCTION
The International Telecommunications Union (ITU) have proposed that future public
Broadband Integrated Service Digital Networks (B-ISDN) use the Asynchronous Transfer
Mode (ATM) of operation. ATM networks transfer data in a continuous sequence of small
fixed size cells. Cells from separate links are statistically multiplexed together over common
links within the network. When traffic peaks coincide the capacity of the common link is
exceeded resulting in congestion which if maintained for a sufficient length of time causes
buffers to overflow and cells to be dropped. The majority of cell losses within ATM networks
are attributable to buffer overflows and therefore cell loss occurs in bursts during times of
congestion. The probability of cell loss therefore depends on the dimensioning of the network
and the network management strategies adopted. An inverse relationship exists between
D. D. Kouvatsos (ed.), Performance Modelling and Evaluation of ATM Networks
© Springer Science+Business Media Dordrecht 1995