On algorithms for determination of turbo code weight distribution

Let C = C(C/sub a/, C/sub b/, N, /spl pi/) be a parallel concatenated turbo code with constituent convolutional encoders C/sub a/ and C/sub b/, information length = interleaver length N, and interleaver /spl pi/. Further, let /spl Pi/(C, /spl tau/) be the problem of finding all codewords of C of Hamming weight /spl les/ /spl tau/. Solutions of /spl Pi/(C, /spl tau/) characterize the code's performance at moderate to high SNR's and under a maximum-likelihood decoding assumption. We present some improvements to the algorithm presented by Garello et al. (see IEEE Journal of Selected Areas in Communications, p.800-12, May 2001) for solving /spl Pi/(C, /spl tau/). As an example we apply the improved algorithm to find the minimum distance of all UMTS turbo codes. We consider parallel concatenated codes, but the techniques are also applicable to other turbo-like codes.