Turbo coding behavior in Rayleigh fading channels without perfect interleaving

Turbo coding is investigated for uninterleaved and partially interleaved Rayleigh fading channels. It is compared to ordinary convolutional coding with the same rate and memory 2, 4 and 8 best d_free encoders. When turbo frames are very long, turbo coding with a channel interleaver gets better bit error rates (BER) than ordinary convolutional coding with a channel interleaver. When the frames are shorter, lower complexity convolutional coding is as good as turbo coding. It is shown by experiment that only after a certain critical frame size does turbo coding get better BER than convolutional coding and the length of this threshold is linearly dependent on the inverse of the fading bandwidth, 1/BT. The effect of the constituent encoders on the error performance is also tested. Memory 2, 4 and 6 constituent encoders are compared. For short to moderate turbo frame size, the memory 2 constituent encoder is as good as the memory 4 and better than the memory 6 encoder. For very long block-lengths, the memory 4 encoder is the best. The memory 6 encoder is always the worst.