Research on signal processing technology using wavelet-based hidden Markov models

The problem of signal processing in north-finders has been studied. The wavelet-based hidden Markov models (WHMM) are used to denoise the Gyro's output signals in continuous rotary north-finders. The WHMM employs Gaussian mixture model and transition probabilities between hidden states to model the individual wavelet coefficient and relationships between wavelet coefficients in different layers, respectively. Furthermore, an expectation maximum (EM) method is used to train the WHMM coefficients. Finally, the wavelet coefficients are reestimated through the trained WHMM, and used in inverse wavelet transform to realize signal denoising processing. The practical examples show that the WHMM can effectively depress the noise in Gyro's output signals, improve the precision of north-finders.