Noise Cancellation in Knee Joint Vibration Signals Using a Time-Delay Neural Filter and Signal Power Error Minimization Method

Removal of random noise is an essential signal preprocessing procedure before the diagnostic analysis of knee joint vibration signals. This article presents a novel noise cancellation method for the knee joint vibration signal processing using a time-delay neural filter. The filter contains a single neuron that convolves the reference input with synaptic weights. The synaptic weights of the filter can be adaptively optimized by a novel instantaneous signal power error minimization method. The results of knee joint vibration signal filtering experiments showed that the time-delay neural filter outperformed the wavelet matching pursuit decomposition method, with less execution time consumption and a higher signal-to-noise ratio.