Peter Eck

A new compression scheme for syntactically structured messages (programs) and its application to Java and the Internet

Summary form only given. The immense demand of required network bandwidth to load computer programs like Java applets or applications within an acceptable time makes efficient compression schemes highly desirable. Syntax-oriented coding (SOC) is a new compression scheme that is able to eliminate redundancy caused by syntactical restrictions. Current compression schemes with a lexical view of the source cannot efficiently exploit knowledge of these restrictions. The paper illustrates the current state in transmitting Java applets (compiled bytecode) over the net, as well as the integration of SOC into this framework. The smart and seamless integration of SOC in the existing Java architecture is obvious. In the compression performance of SOC, the paper depicts the file sizes of various representations of a typical Java applet: the size (containing no comments) of the (ASCII-) source code, the size of the class file (bytecode), the size of the compressed (Ziv-Lempel coding) class file and the size of the SOC-code. Using SOC a compression ratio of more than 6 is obtained compared to the Java class file, and even if the class file is Ziv-Lempel coded a ratio of more than 4 is still reached, reducing required bandwidth by 75 percent.

On MPEG-2 decoding of noisy input data

Due to the high compression ratio of the MPEG-2 video compression scheme, the effect of corrupted bits during transmission over a noisy channel is tremendously increased compared to uncompressed transmission. Moreover, due to the hierarchical structure of the MPEG-2 bitstream, the damage caused by channel noise depends on the position of the corrupted bit in the bitstream. Analysis shows that the most important (or most sensitive) bits are those of the MPEG-2 start codes. Loss of a start code usually causes the unrecoverable damage of an entire slice or even a whole sequence of pictures. An optimum detector for start codes is presented. In contrast to classical MPEG-2 decoders, this detector is based on the real-valued channel output (soft values), instead of already detected binary values (hard decisions). Performance analysis shows an SNR gain of about 6 dB compared to the classical hard decision detector.

Syntax-oriented coding (SoC): a new algorithm for the compression of messages constrained by syntax rules

A data compression algorithm suited for syntactically structured sources is presented. This algorithm is based on a syntactic source model, which formulates the syntactic information of a message as a sequence of SoC-symbols. Analysis using some exemplary mini-languages shows that this model is quite compact in the sense of capacity. A typical application is implemented for Java, the compression ratio is satisfactory.

A low complexity soft-output decoder for a broad class of trellis coded modulation schemes

This work introduces a general concept to extend the principles of soft output Viterbi decoding to M-ary transmission. The proposed algorithm matches ideally the family of punctured pragmatic TCM (P/sup 2/-TCM) schemes. Due to the puncturing technique employed, such a decoder can be realized by a basically traditional SOVA with a slightly modified metric and time-variant labeling, operating at a multiple clock speed.