Competitive learning algorithms for channel optimized vector quantizers

This paper proposes some modifications of known competitive learning rules for designing vector quantizers optimized for noisy channels. The modified learning rules take into account the knowledge of the channel to further reduce overall distortion. It is shown that the noisy competitive learning rule outperforms both noisy and noiseless generalized Lloyd algorithm in quantizing speech signals. Furthermore, it appears very robust in case of over estimation of the bit error rate when only partial knowledge of the channel is available