Optimal quantization for soft-decision turbo decoder

We present an optimization of the quantization of the channel input to turbo decoders by maximizing the cutoff rate. Three types of quantizers are analyzed: uniform, non-uniform, and logarithmic quantizers. We propose a logarithmic quantizer because it is not very sensitive to a quantizer mismatch relative to the input level and keeps the BER constant over a broad range of input variances. In addition, we quantify the quantization effect of the extrinsic information via simulation. We show that 6-bit channel input and 8-bit extrinsic information quantization incur little degradation of the decoding error rate performance with proper choice of the clipping level