A modified bethe free energy approximation for codeword quantization

A novel codeword quantization algorithm based on message-passing using a low density generator matrix formulation is proposed and analyzed. The scheme is a seemingly subtle variant on a recently proposed “truthiness” propagation algorithm, but one which affords a more explicit connection to a modified Bethe free energy function. Applications to distributed coding in sensor networks are also included in the simulation examples, where the algorithm is observed to outperform conventional LDPC belief propagation decoding using side information, in a practical setting when the reliability of the side information diminishes.