System identification with denoising

When the signal-to-noise ratio (SNR) is low, classical system identification methods can not produce accurate results. The results can be improved by using denoising methods with time-frequency decompositions. The chirp signal is used as a training sequence to make the time-frequency domain denoising possible. Chirplet decomposition is proposed for separation of signal and noise components. The results are compared with the Gabor transform denoising. The chirplet denoising method proposed here is less sensitive to SNR changes than the Gabor denoising proposed before. Also, the accuracy of the estimates in chirplet case is superior to the Gabor transform method