SNR Estimation Based on Finite Sample Data

This paper first demonstrats that signal subspace has finite dimensions from signal model. A conclusion is obtained by simulation test that both eigenvalues of signal subspace and noise subspace can't be evaluated precisely due to small size data. Then signal subspace dimensions are estimated accurately with Modified-AIC criterion using likelihood function depending only on noise space eigenvalues. A new robust SNR estimation method is proposed with the application of modified-AIC criterion. Regarding a variety of SNR and data size, simulation results show that the new method can be applied in lower SNR and shorter data samples environment.