一种LMS算法的实现及应用

Implementation and Application of LMS Algorithm

定步长LMS算法虽然结构简单,易于实现,但也存在后期信号变弱,步长过大,导致收敛变慢的问题。针对该问题,首先在硬件设计上采用了射频信号正交分解的思想,降低滤波器维度;然后软件上设计了一种实时采样决定步长的方法。该方法使得不同频点、不同强度的信号有不同的步长。通过降低滤波器维度和变步长的方式加快LMS算法的收敛速度。利用C语言编程在DSP6416芯片上实现了这种变步长LMS算法。实物测试表明:该算法具有收敛速度快,对单载波、FM、AM信号的噪声都能达到40dBm的对消效果。

The fixed-step LMS algorithm, although being simple in structure and easy for implementation, has the problem that the signal becomes weak and the step size is too large in the later stage, which leads to the slow convergence. To solve this problem, firstly, the orthogonal decomposition idea of RF signals is adopted in hardware design, thus to reduce the filter dimension, then the real-time sampling method to determine the step size is designed on the software. This method makes different frequency points and signals of different strengths have different step sizes. The convergence speed of the LMS algorithm is accelerated by reducing the filter dimension and the variable step size. This variable step size LMS algorithm is implemented on the DSP6416 chip by using C language programming. The physical test indicates that this algorithm has fast convergence speed and can achieve a cancellation effect of 40 dBm for single-carrier, FM, and AM signals.