Application of unimodular matrices to signal compression

In the past decade, the application of orthonormal or paraunitary (PU) matrices to subband coding has drawn considerable attention. In this paper, we will study signal compression using unimodular matrices. Like PU matrices, unimodular matrices have FIR inverse. However unimodular matrices do not have the energy preservation property. Applying unimodular matrices to subband coding is not always a good choice as quantization noise might be amplified and coding gain can be less than one. We introduce a new structure for unimodular coders. The new structure consists of a closed-loop vector DPCM structure followed by a transform coder. Using such a structure, it is shown that the coding gain of unimodular coders is never less than 1. Simulation example shows that despite having the smallest system delay, unimodular coders have a better performance than subband coding with the same complexity for AR(1) input.