Blind multichannel identification via frequency selection

In a recent study on blind multichannel identification, we presented a fast Fourier transform based algorithm, called the BI-FFT algorithm. The BI-FFT algorithm is capable of dealing with channel outputs of a very short length, which is of practical importance in communication applications where wireless channels involve high mobility. However, when the output sample size is relatively large, the BI-FFT algorithm becomes cumbersome and computationally costly. To address this issue, we develop a frequency selection approach in this paper. The proposed method strategically chooses those frequency bins that significantly contribute to the error dynamics in the channel estimation process. As a result, the row dimension of a key matrix in the identification equation is considerably reduced, effectively decreasing the computational cost. With a comparable performance, the new method is much more efficient in computation than the original BI-FFT algorithm in case of long data sequences.