Turbo码Log-MAP译码算法简化实现的研究

 Log-MAP算法实现Turbo译码时,要用查表等方法进行修正项的近似计算。用误差传播理论推导出Log-MAP算法大规模集成电路实现时状态度量和对数似然值等的精度要求,明确了量化方案,并由此得出计算修正项所需要的精度由信道接收值量化精度决定。理论上解释了信道接收值量化间隔为0.25时,8级查表就可以满足译码精度要求;也解释了高信噪比条件下,由于量化误差的传播,修正项可以忽略不计,此时Max-Log-MAP算法与Log-MAP算法一样,也是最优译码算法。根据推导出的对修正项精度的要求,结合修正项函数值分布特点,提出用6级非均匀查表计算修正项的方法。该方法能达到16级查表精度,在实现上比8级均匀查表简单。计算机仿真证明了上述结论。

Study on the Simplification of Log-MAP Algorithm for Turbo Decoding

When turbo decoder is implemented by Log-MAP algorithm, it is necessary to calculate the correction term by look-up table (or other ways). The required precision of the state metrics, log-likelihood ratio, etc. , are calculated and a quantization scheme is presented. The required precision of the correction term is determined by the quantile interval of channel value for the quantizing error propagation. The quantizing error propagation shows the reason why the excellent results could be obtained with 8 stored values in the table when the quantile interval of channel value is 0. 25. Also it shows that the Max-Log-MAP algorithm just like Log-MAP algorithm is optimal one because of the ignorable correction term for the quantizing error propagation under high signal noise ratio. Then a non-uniform look-up, table with 6 stored values is developed to calculate the correction term. It reaches the precision of the look-up table with 16 stored values, which is more accurate than 8 stored values table. At the same time it is easier to be implemented than the 8 stored values table. The simulation verifies these conclusions.