A parallel LTE Turbo decoder on GPU

Turbo codes were developed by Claude Berrou in 1993, which were based on convolutional codes and Viterbi algorithm and could closely approach the channel capacity. Software Defined Radio (SDR) promises to revolutionize the communication industry by delivering low-cost, flexible software solutions. Graphics Processing Units (GPUs) offer unprecedented application performance since they have many cores. GPUs have been widely used to accelerate a wide range of applications. In this paper, a 3GPP LTE Turbo decoder on GPU is designed, which can offer high throughput. Sub-frame-level parallelism and trellis state-level parallelism have been implemented by some researchers. In this paper, we propose a new method to accelerate our decoder: sub-decoder-level parallelism. In addition, shared memory is utilized to keep memory access fast. Finally, the bit error rate (BER) performance and throughput of the decoder are presented.