Noise Estimation and Automatic Filter Design Based on Silent Slot Prior

The optimal parameter of a filter is usually dependent on the degree and the style of the noise. Therefore, noise estimation plays a critical role in filter design. In this study, we propose a new method of noise estimation in the time-plane. We apply the fact that natural signals without noise must have a sparse time-frequency distribution. That is, their energy must be concentrated in some regions of the time-frequency plane. By contrast, the time-frequency distribution of the background noise is usually much more uniform. For a noise-interfered signal, if there are regions in the time-frequency plane that do not have a concentrated distribution and the average energy in the region is relatively smaller, the region is probable to be the noise part. We call it the silent slot. After performing the weighted Fourier transform on the silent slot, the degree of the noise is well estimated. With the estimation results, the optimal filter is automatically designed, and it is unnecessary to adjust the parameter for each signal manually. The proposed algorithm is helpful for adaptive and real-time filter design.