A computer search for convolutional codes with good distance profiles

A nested step-by-step computer search was undertaken to find new convolutional codes with good distance profiles and information rates R=k/n, where the information symbols k and the n codeword symbols are: 4⩽n⩽8, 2⩽k⩽n-2. These codes are intended for sequential decoding. The search was set up in such a way that information about the weight distributions of all code sequences, or a truncated version of this weight distribution, was available after each step. The algorithm used processes all retained information sequences at each step, but bounds the computational work by not storing any information about an information sequence once its encoded weight exceeds a target minimum weight which is specified at the beginning of the search. If an information sequence is thus deleted at a given step of the search, it is not processed at the next or any further steps to evaluate the next connection vector or any further sets of connection vectors. Although the search will eventually run on a Cray supercomputer, it was conducted using an IBM PC. The new codes are summarized in tabular form