基于GPU的Viterbi并行解码算法的设计与实现

Design and impelementation of parallel viterbi decoding on GPU

针对GPU并行计算特征,对Viterbi解码自身做了并行处理探索,并提出使用Zero-Termination卷积码来实现基于GPU的Viterbi解码分块并行处理。设计的实现结果表明:Zero-Termination卷积码的简单而适用于GPU分块并行;误码率降低,特别是在信噪比低的情况下,Zero-Termination卷积码误码率比不损失码率的卷积码要低。同时,还实现了基于GPU的7,9,15三种不同约束长度的Viterbi解码,获得了良好的误码性能曲线及高吞吐率表现。

In this paper, GPU-oriented parallel Viterbi decoding algorithm for convolutional code is explored. Zero-Termination convolutional code is also proposed to achieve parallel block Viterbi decoding on GPU is proposed. The results show that Zero-Ter- mination convolutional code is simple and suitable for GPU-based block Viterbi decoding. It also showsthat. It also shows that BER decreases especially when the ebn0 is low. Meanwhile, 7,9,15 these three kinds of constrain length convolutional codes＇ parallel block Viterbi decoding are implemented on GPU with low BER and high I/O performance.