Unequal Energy Allocation with Turbo Codes for Nonuniform Sources

Abstract:
Non-systematic channel codes can have a significant advantage over systematic codes when utilizing redundancy left in the data for channel decoding. However, results previously attained show that even with such codes there is a performance gap to the theoretical limits. We study techniques that allow reducing this gap by using unequal energy allocation for both systematic and non-systematic turbo codes. For both types of codes, there are regions of source non-uniformity, for which they outperform codes with equal energy allocation. Systematic codes with unequal allocation perform well at extreme non-uniformities. The non-systematic codes perform well at moderate non-uniformities and still outperform their equal energy allocation counterparts at extreme non-uniformities. A theoretical study of the mutual information attainable by the different classes of codes for different source non-uniformities is presented, and supports the advantage of non-systematic codes with unequal allocation over the other classes of codes. While it also points out that the non-systematic codes should be better for extreme source non-uniformities, simulation results show that the current designs of the systematic codes with unequal energy allocation are still better than those of the non-systematic codes in this region.