The free distance [1] of a convolution code determines to a large extend the error rate in the case of maximum likelihood (ML) decoding. This report describes a method to derive upperbounds on the bit error probability for a Vertebra decoder [2] with finite path register length. These bounds are compared with measurements for both the Vertebra- and for the Stack-algorithm [3]. An exhaustive search is carried out for several classes of R = k/n codes, including the classes mentioned by Schalkwijk et al [4,5]. These latter codes allow for exponentional hardware savings in the (syndrome) decoder. Tradeoffs between complexity and performance are considered.
展开▼
