Accelerating convolution Coding & Viterbi decodingon GPUs using OpenCL

This paper is an attempt to accelerate the convolution Encoding and Viterbi most-likelihood decoding algorithms on Graphics Processing Units (GPUs) using Open Computing Language (OpenCL). Several parallel decomposition schemes are explored, to harness the massively parallel architecture of the GPU. We observed encoding throughput up to 467 Gbps and decoding throughput up to 391 Mbps for a Constraint Length of 7 and a Code Rate of ½ using AMD Radeon HD 7970 GPU. We have also tuned our OpenCL implementation by varying a number of configurable attributes of our implementation and have identified the best performing configurations on AMD GPU, NVIDIA GPU and Intel CPU.