采用向量内积的并行相关算法

针对软件接收机相关器计算的实时性问题,通过分析扩频信号的接收过程,建立一种基于向量内积的并行相关信号接收模型。利用图形处理单元中大量的浮点运算单元进行矩阵与向量运算,并行计算各通道相关值,提高了信号相关运算的实时性。仿真验证结果表明,利用基于GPU的向量内积软件并行相关算法计算25 MHz采样率时长1 ms的信号相关值,25个通道共150个相关运算耗时967μs,与CPU上基于数学核心函数库的实现相比速度约提高了61.4倍,能够实现宽带扩频信号软件实时相关接收。

Parallel correlation algorithm based on vector dot product

The correlator is the core module of serial search acquisition and channel tracking in a navigation receiver. Computing resource requirements are high for real-time correlation of several tens of channels in software receiver, but most of the software receiver algorithms are sequential algorithms based on a single core processor. For solving the real-time problem of the software correlator, a parallel receiving model of spread spectrum signal based on vector dot product is proposed. By parallel computing correlation value of each channel with a large number of ALU in GPU, the calculation speed of the signal correlation is improved. The simulation results show that, with the Vector Dot Product parallel correlation algorithms based on GPU, 25 channels with 150 correlators consume 967us in all for 1ms signal of 25MHz sampling rate. Compared with the implementation on CPU based on MKL, the method speeds up to 61.4 times, then real-time correlation of the broadband spread spectrum signals in software can be realized.