在电力变压器有源降噪系统实际运行时,声传感器采集的参考信号和误差信号可能存在冲击干扰,当采用传统的滤波最小均方(filtered-x least mean square,Fx LMS)算法对控制器的参数自适应更新时,控制器易受冲击性干扰的影响而出现振荡甚至是发散。为了抑制冲击性干扰对控制器的影响,提出了归一化滤波语音压缩?律变换的最小均方(normalized filtered-x?law mean square,NFx?LMS)算法。该算法采用经语音压缩?律函数变换后的误差信号的均方作为代价函数,并对参考信号做归一化处理以提高算法鲁棒性。算法可以根据不同声场环境的初级噪声特点选取不同的压缩系数以提高算法的实用性。仿真显示,在冲击性干扰的环境下,论文提出的算法表现出很好的鲁棒性和收敛效果。
【英文摘要】
In actual application of power transformer active noise control system,signals produced by error and reference microphone contain signals relating to both transformer noise and interference.The interference,especially impulsive interference,deteriorates ANC system by deviating weights of ANC controller from its converged values.In the worst case,it may cause adaptation divergence.In order to reduce effect of the interference,normalized filtered-x μ law LMS algorithm is proposed with minimized squared μ-law ...