System identification at an extremely low SNR using energy density in DCT domain

This paper presents a new approach for system identification at an extremely low signal-to-noise ratio (SNR) like -10 dB. At such a low SNR, many of the system peaks in the signal spectrum are indistinguishable from the noise peaks that lead to erroneous identification of system poles. In the proposed method, the system properties like energy density in discrete cosine transform (DCT) domain, which is less susceptible to noise even at such a low SNR, has been utilized to estimate autocorrelation function. Successive autocorrelations of this estimated function are taken for sequential identification of system parameters.