FPGA-based data compressor based on prediction by partial matching

Abstract

We design and develop a data compression engine on a single FPGA chip that is used as part of a text-classification application. The implementation of the prediction by partial matching algorithm and arithmetic coding data compression is totally in hardware without any software code. Our design implements a dynamic data structure to store the symbol frequency counts up to maximal order of 2. The computation of the tag-interval that encodes the data sequence in arithmetic coding is done in a parallel architecture that achieves a high speedup factor. Even with a relatively slow 50 Mhz clock our hardware engine performs more than 70 times faster than a software-based implementation in C on a CPU running on a 3 Ghz clock.