Fixed-point performance of low-complexity turbo decoding algorithms

Efficient fixed-point turbo decoders require accurate low-complexity algorithms using small word sizes. A procedure is provided to determine the word sizes for fixed-point turbo decoders. The procedure is used to design small word sizes for the low-complexity constant-log-MAP and linear-log-MAP turbo decoding algorithms, as well as the lookup table based log-MAP. At a 1% FER, the performance of low-complexity fixed-point turbo decoders designed accordingly is shown to reside within 0.05 dB of the performance obtained using the floating point log-MAP algorithm. New results on the optimization of the linear-log-MAP are also provided