Using Hierarchical Filters to Detect Sparseness in Unknown Channels

A novel algorithm particularly suited for determining and identifying the sparseness and the associated features within an unknown channel model is proposed.This is achieved by virtue of hierarchical filters which exhibit relatively steady state characteristics, when more than one of their constitutive sub-filters have to identify non-zero parts of the unknown channel. This way, sparseness is detected by comparing, in an on-line fashion, the performance of a hierarchical filter with that of the standard finite impulse response (FIR) filter trained by the least mean square (LMS) algorithm. In addition, the character and type of sparseness can be detected based on the architecture of the underlying hierarchical filter. The analysis is supported by simulations evaluated in a rigourous statistical framework.