Universal data compression with LDPC codes
Giuseppe Caire · 2003
We present a new universal noiseless compressor of sources with memory based on the concatenation of the Burrows–Wheeler block sorting transform (BWT) with the syndrome former of an LDPC code. The proposed scheme makes use of a library of LDPC paritycheck matrices of different rates and of a simple method to estimate and encode the tree source model from the BWT output. Unlike existing works that use error-correcting codes for data compression, our scheme can deal with sources with memory and achieves lossless compression. Our method offers competitive performance over existing methods such as Lempel-Ziv (gzip) and standard BWT– based schemes (bzip), while being amenable to joint source-channel decoding.