On finite-precision implementation of a decoder for turbo codes

In this paper, we have investigated the finite-precision effects on a decoder for turbo codes based on the Log-BCJR algorithm. As for the numerical accuracy, we have empirically found that three bits are adequate for representing the decimal fractions of the variables in this algorithm. Furthermore, these variables have different numerical ranges for different signal-to-noise ratios, thus requiring different numbers of bits for representing them. By observing their minimum and maximum values through computer simulations, we have decided the proper number of bits for representing these variables. Simulation results have indicated that the scheme we have developed for determining the finite-precision parameter set of a turbo decoder based on the Log-BCJR algorithm is effective in approximating the performance of the same decoder with infinite-precision implementation