Super-Resolution Blind Channel Modeling

We consider the problem of extracting a wideband channel model when only measurements in parts of this band are available, specifically in disjoint frequency subbands. Conventional channel modeling techniques cannot model at all those parts of the band where no sounding signals are available; or, if they use conventional interpolation, suffer from poor performance. To circumvent this obstacle, we develop in this paper a three-step super-resolution blind algorithm. First, the path delays are estimated by exploiting super-resolution algorithms such as MUSIC or ESPRIT based on the transfer function of each subband, separately. Exploiting such a set of delay estimates, the proposed algorithm performs blind (i.e., without training signal) channel estimation over the unmeasured subbands, and subsequently derives the frequency response over the whole wideband channel. Finally, estimates derived from different subbands are combined via a soft combining technique. Computer simulations show that the proposed super-resolution blind algorithm can achieve a significant performance gain over conventional methods.