The role of filter banks in sinusoidal frequency estimation

The problem of estimating the frequencies of sinusoids buried in noise has been one of great interest to the signal processing community for many years, especially to those involved in the field of array processing. While many methods have been proposed to solve this problem, most involve processing in the fullband. In this paper, we investigate the effects of carrying out estimation in the subbands of an analysis bank of a multirate filter bank and show that there are some benefits to be reaped. In particular, we observe that with properly chosen analysis filters, the local signal-to-noise ratio (SNR) and line resolution in the subbands will exceed that in the fullband. Also, through the use of the spectral flatness measure, we show that if the input noise is colored, then the noise processes seen in the subbands will be more flat on average. This can be useful if the exact statistics of the input noise process are not known. Various examples are shown giving evidence to the fact that estimation in the subbands is superior to that in the fullband.