基于混合遗传算法及最低误码率准则的几何特征均衡器

Geometric feature equalizers based on minimum bit error rate criterion and hybrid genetic algorithm

针对高效调制通信系统中带内干扰抑制问题,提出一种基于最低误码率准则的非线性几何特征均衡器,并用径向基函数神经网络来实现。为优化非线性均衡器的参数训练,本文构造了一种新的遗传随机梯度混合算法。仿真表明:对于扩展的二元相移键控信号,在相对强的窄带干扰下,匹配滤波器及线性均衡器已失效,而基于最低误码率准则的几何特征均衡器仍能表现出良好的性能,也大大优于基于最小均方误差准则的非线性均衡器。

A nonlinear geometric feature equalizer based on minimum bit error rate is proposed for removing interference whose frequency band overlaps with the desired signal in high efficient modulation communications, and the equalizer is realized by radial basis function neural network. For optimizing the parameters of nonlinear equalizers, a novel hybrid genetic algorithm-stochastic gradient algorithm is also proposed in the paper. Simulation results show that when extended binary phase shifting keying signals are contaminated by the relatively strong narrow band interference, the performances of matched filters and linear equalizers are degenerate rapidly, but geometric feature equalizers based on minimum bit error rate provide very low bit error rate, and their performance is also much better than that of nonlinear equalizers based on minimum mean square error.