Entropy-based stochastic gradient algorithm with adaptive neuron slope for all-pole whitening

The stochastic gradient algorithm presented employs the time-varying neuron's slope to optimise performing of the all-pole filter-whitener maximising the joint Shannon's entropy. By using the time adaptive slope, which matches the unknown probability density function of an input process, a neuron slope selection issue of the original algorithm is facilitated and its tracking of non-stationary statistics is improved. The performing of algorithm is verified using the whitener as a front-end amplitude equaliser of the blind equaliser.