基于MISEP算法的瞬时混合模型非线性电路的信噪盲分离

摘要

This paper proposes a mutual information based blind signal noise separation algorithm of non-linear circuit, to compensate the measured circuit signals. For the instant mixing non-linear circuits, this method constructs the multilayer perceptron network by feedback cascading the signal separation block and the parameter adjustment block; aiming at the minimum mutual information, it trains the network using the measured circuit signals with random noises, until the cost function value converges to pre-set error range;then the trained network is applied to the blind signal noise separation for the nonlinear circuits. The ex-perimental results on the fore-nonlinear circuit, the post-nonlinear circuit and the single-stage amplifier cir-cuit show that the signals and noises separated from this method approximately follows the circuit inputs on the time domain waveforms and power spectrum characteristics.%针对非线性电路信噪分离问题, 提出基于互信息的信噪盲分离算法, 用以对非线性电路的测量信号进行补偿. 首先面向瞬时混合模型下的非线性电路, 将用于计算分离信号的模块与用于调整参数的模块进行反馈级联, 构造多层感知机网络; 然后采用含随机噪声的电路测量数据, 以最小互信息为目标对该网络进行训练, 直至代价函数值收敛于预设的误差范围; 最后利用训练好的网络进行非线性电路盲源分离问题的解算. 对前非线性电路、后非线性电路和单级放大器的实验结果表明, 该算法分离出的信号和噪声在时域波形和功率谱特征方面均与输入相符.