Abstract:
This paper considers a cost function level analysis of
the Sum-squared Autocorrelation Minimization (SAM) channel
shortening algorithm. We point out that the actual cost the blind
adaptive stochastic gradient descent algorithm is minimizing is
only indirectly related to the sum squared autocorrelation. We
study the asymptotic regimes under which the actual cost yields
a reliable surrogate for the sum squared autocorrelation. We
investigate the relationship between the minima of the actual
cost and sum squared autocorrelation. We also study the upper
bound of the approximate cost as a function of the window size
used in the approximate autocorrelation calculation
