1. Introduction

The Internet continues to grow at an exponential rate. As shown in Figure 1, Figure 2 and Table 1 there were approximately 16.1 million host computers on the Internet in January of 1997. A year earlier there were only an estimated 9.5 million hosts [1]. The ongoing explosion of new Internet users brings with it a higher demand for multimedia applications on the World Wide Web (WWW). The majority of these users, however, do not have high speed communication paths. Instead they are using dial-up connections with 28.8 kbps or 14.4 kbps modems [2]. This poses a challenge to audio and video application developers because both require tremendous amounts of bandwidth to operate in real-time, typically 1.41 Mbps for CD quality audio and 221.18 Mbps for broadcast quality video [3].

	.....
Figure 1: Internet Hosts 1989-1996 [4]	.....	Figure 2: Internet Hosts - Overall Trend [4]

The solution to this problem lies in the realm of lossy data compression and perceptual encoding. Limitations in the body's auditory and visual senses can be exploited so that data that would otherwise be ignored can be thrown out. At low bitrates deciding which data to discard becomes a complex task. Several algorithms have been developed to handle this task for both audio and video, many of which have resulted in worldwide standards. Few, however, address very low bitrate encoding at speeds which would allow real-time transfer of data through modems.

One encoding method that offers very low bitrate, high quality audio is MPEG-1 Layer III. MPEG-1 Layer III decoders are just beginning to appear and it is expected that soon the encoders will become widely available. The Fraunhofer Institut für Integrierte Schaltungen (IIS) in Germany offers both an encoder and a stream decoder. MPEG-1 Layer II encoders and decoders are more prevalent but are only designed to reach bitrates down to 32 kbps.

The goal of this project was to use the existing MPEG-1 Layer II architecture to encode audio for real time delivery across the Internet. This meant testing new lossy MPEG-1 Layer II algorithms to determine if any were suitable for use at bitrates lower than 32 kbps. This posed a unique challenge because the extent the encoder could be changed was limited, because the resulting bitstream must adhere to the standard to be decipherable by existing MPEG decoders. This project began with the assumption that the Fraunhofer MPEGiis encoder was as efficient and effective as possible. From there it explored additional methods by which the bitrate could be reduced further. It also explored pre and post signal processing methods that improved the encoder's results.

The results of this project show that MPEG-1 Layer II can be used to encode very low bitrate files (VLBR) for delivery across the Internet. The quality of the new algorithms is equivalent to I-Wave and Real Audio 2.0 encoded at 28.8 kbps and better than Real Audio 2.0 files encoded at 14.4 kbps. The quality is not, however, as good as MPEG-1 Layer III or Real Audio 3.0.

Over the lifetime of this project the 28.8 kbps modem barrier has been broken twice to yield 33.6 and 56 kbps modems. In the future these modems will become more prevalent and they will eventually give way to integrated services digital network (ISDN), Asymmetrical Digital Subscriber Line (ADSL), and cable modem technologies. These technologies offer more bandwidth, reducing the need for compression. The methods and tools employed in this project, however, may be useful in cases where multiple MPEG-1 Layer II streams are desired or where MPEG-1 Layer III or MPEG-2 encoding methods do not achieve a low enough bandwidth to meet specific requirements. The tools developed in this project give the user control over the encoding process that is not available in any other encoders. This allows the encoding process to be refined, ensuring the quality is as good as possible, rather than leaving it up to the encoder.

The first section of chapter two in this paper discusses data communication networks to give the reader an overview of the medium by which multimedia data is transmitted. The second section of chapter two discusses the basics of digital audio and the problems faced when transmitting it over communication networks. The first section of chapter three offers a look at the current audio compression schemes which attempt to solve these problems. The second section of chapter three gives an overview of the scheme used in this project, MPEG-1. The third section covers various aspects of MPEG-1 in detail, including all three layers and both psychoacoustical models.

The fourth chapter of this paper gives and overview of this project, including its objective and one previously attempted method. The first section of the fifth chapter discusses the resources that were used to implement this project. The second section covers the modifications that were made to the MPEG encoder and decoder to create three new encoding controls. The third and fourth sections discuss pre- and post-processing methods that can be used to improve the quality of an encoded file.

The first section of chapter six discusses the results of the qualitative tests on various new MPEG algorithms that were created using the three new encoder controls. The second section takes a look at the compatibility of the new low bitrate files with the various MPEG decoders that are available. Chapter seven draws some conclusions from the results of this project. Chapter eight suggests some directions for further study and what might be done to improve the quality of the low bitrate files even more.