基于GPU的混沌弱信号检测临界阈值确定

Determination of threshold value of chaos-based weak signal detecting based on GPU

混沌检测系统临界阈值的确定是建立混沌检测系统的核心问题,临界阈值的精度决定了可检测信号的精度。目前相轨迹图观察法已经无法满足快速确定精确的系统临界阈值的需求,利用Lyapunov指数等量化指标检测临界阈值的方法计算量大、算法复杂、时间消耗大,且消耗大量计算资源,无法在GPU上实现并行程序设计;系统相轨迹过零周期数相变判别算法在检测相同精度阈值情况下较Lyapunov指数算法有相同的检测精度,同时更易于利用现代高性能计算工具GPU实现并行程序设计。因此,在系统相轨迹过零周期数阈值判别算法的基础上提出了基于GPU的混沌弱信号检测临界阈值并行检测算法,实验显示,GPU并行系统临界阈值检测算法在幅值递增步长10-6下较CPU串行算法可以实现近90倍的加速比,使高精度临界阈值能够在短时间内准确确定。

Determination of the threshold value is the key to chaos-based weak signal detection. The precision of the threshold value invariably defined the precision of the detection. Currently, determining the threshold value by observing the phase space trajectory could hardly satisfy the demand for an accurate threshold value in a short time, while the Lyapunov index method was complicated and it took a lot of time and computing resources,which was not convenient to program on the GPU platform. Com- parably, the detecting method based on zero-crossing number could achieve an accurate threshold value with lower computing resources and complexity,which could well applied to the modem high performance programmable GPU. By using the zero- crossing number criterion, this paper proposed a new parallel algorithm of detecting the accurate threshold value based on GPU platform. Experiments show that the parallel algorithm based on GPU can accelerate the determination of the threshold value up to 90 times compared with the single-thread algorithm at the amplitude step length of 10-6. This parallel algorithm brings light to the quick detection of the threshold value of different chaos detecting system.