Turbo decoding with nonuniform quantization

The performance of fixed point turbo decoders is usually evaluated assuming a uniform quantizer and an additive white Gaussian noise (AWGN) channel. However, if a uniform quantizer is employed in a fading channel, significant quantization noise can occur, resulting in a degradation in bit error rate (BER) performance. In this paper, we propose turbo decoding with a nonuniform quantizer. In a Rayleigh flat fading channel, it is shown that with a 6-bit nonuniform quantizer, fixed point turbo decoding performance is approximately equal to floating point performance. It is also shown that a 3-bit nonuniform quantizer is as effective as a 4-bit uniform quantizer. In addition, we determine that a low complexity turbo decoding algorithm with a nonuniform quantizer performs the same as a higher complexity algorithm with a uniform quantizer.