Searching for short-span Convolutional Doubly Self-Orthogonal Codes: A parallel implicitly-exhaustive -search algorithm

In this paper we present an efficient algorithm to search for optimal/short-span convolutional self-doubly orthogonal codes (CSO2C) using a novel parallel exhaustive-search algorithm. The error correcting codes are used in the new CSO2C encoding/iterative threshold decoding technique that was introduced as an alternative to turbo codes. Since the iterative threshold decoding algorithm does not use interleaving, the generator connections in these codes must satisfy specific orthogonality properties to ensure the independence of observables over the first two iterations. The code constraint length (or ldquospanrdquo of the code) corresponds to the latency of each decoding iteration, and minimizing it for a given error-correcting capability is important when building high-performance, low-latency codecs. The proposed algorithm drastically speeds up the implicitly-exhaustive search for shorter span codes and scales well on todaypsilas multi-core computer systems. On a 16-core server, the computational time for obtaining optimal J = 6 and J = 7 CSO2C was reduced by 93%.