A class of unbiased identification for inverse system with input noises
一类有输入噪声扰动的逆系统无偏参数辨识算法研究

In identifying the inverse system, the input is the output from the original system. This signal is corrupted by noises with unknown variance. When the ordinary least-squares method is applied to estimate the parameters of the inverse system, the estimates turn out to be biased. A new identification algorithm for bias compensation is proposed. Therein, the noise variance of the inverse system input is first estimated using the wavelet transform, and then, a recursive least-squares method with bias-elimination is used to estimate the parameters of the inverse system. Thus, the proposed algorithm does not require the input signal to be the white noise with a zero mean. Since the computation is recursive, it can be implemented online for estimating parameters of the inverse system. Experimental results show that the approach is effective.