Abstract:

At the center of the explosive growth of multimedia occurring today is the popular MPEG family of compression algorithms. Although currently available MPEG decoders are suitable for almost any application, the greater complexity of the encoding algorithm means that compromises need to be made. Our goal is to develop an encoder that will combine a distributed implementation with a finer degree of parallelism than other similar products provide. This will allow for an improved compression ratio while maintaining the cost advantages of a distributed approach.

Problem Statement:

There are several approaches to compressing digital video. One approach would be to encode each frame as if it were a still picture. However, it is often possible to predict much of the content of a frame by looking at the frame that precedes it and/or the one that follows it. Taking advantage of that fact is the basis of MPEG compression.

The disadvantage of MPEG compression is that it takes considerable effort to search for the data that can be safely omitted. Although decoding can be done in real time using modest hardware, rapid encoding requires equipment beyond the reach of most people. Because the information needed to encode a frame is contained in the frame itself and its closest neighbors, software has been developed that runs on a number of computers in parallel, providing reasonable speed with much less costly hardware.

The goal of that software is to provide the greatest possible throughput. A better compression ratio can be achieved by looking at smaller units of data, but this more thorough method can now be found only in software that runs on a single machine. Our aim is to develop software that provides a more complete search for dispensable data in a parallel implementation.

Design Objectives:

The primary objective is that our software should retain the cost benefits of the current parallel implementation. This implies that it should run on the same inexpensive network of computers. Although the new approach will add more network overhead that will necessarily slow down the process, the throughput should not be reduced by more than half. Furthermore, the software should be based on freely available toolkits so that the resulting software is free. The software should have a command line interface. Also, reasonable defaults should be offered for any parameter possible so that minimal learning is required for initial use.

To some extent, our software will be based on the Berkeley MPEG Encoder. This free, parallel software is distributed with source code and can be compiled without modification on a computer running Linux. It may be possible to modify this software to fulfill the objectives outlined above. If more drastic alterations are needed, PVM is a freely available software package that provides an approachable environment for developing parallel applications for Unix. In either case, the Berkeley implementation of the encoding algorithm will form the core of our application. Based on the reference software produced by the MPEG Software Simulation Group, it is the most efficient code that is publicly distributed.

Although Widows software is nearly ubiquitous, it is not suited for this application because of the limits of its networking capabilities. Linux was chosen because it is free and available to anyone, and because its flexibility makes it an ideal development platform.

Our project takes advantage of various software and hardware that is freely available. Cost implications should be minimal and restricted to miscellaneous expenses only. Below is a list of the resources we will be using.

4 – Inexpensive network capable CPU’s (Pentium 120-166Mhz expected but dependent upon availability)

LINUX OS (Freeware)

PVM (Parallel Virtual Machine distributed computing management software - Freeware)

MPEG Encoding Software (Freeware from Berkley Media Research Center)

Miscellany – could include but not limited to networking accessories (approx. $50)

Human Effort Budget:

Name: Bryan A. Burkhardt