Hybrid stochastic gradient for convergence of adaptive filter

Claim 2

Stochastic control circuitry for adaptively tracking the output of a communication channel, the stochastic control circuitry comprising: (a) an adaptive filter; (b) means for driving both the adaptive filter and the communications channel utilizing a common input signal to produce an adaptive filter output signal and a communications channel output signal, respectively; (c) first comparator means for comparing the adaptive filter output signal and the communications channel output signal in accordance with a first criteria to generate a first comparator output signal; (d) second comparator means for comparing the adaptive filter output signal and the communications channel output signal in accordance with a second criteria to produce a second comparator output signal; and (e) third comparator means, which receives the first and second comparater output signals as inputs, for comparing the first comparator output signal and a preselected value and wherein the third comparator means provides the first comparator output signal as a third comparator output signal if the first comparator output signal is greater than or equal to the preselected value and provides the second comparator output signal as the third comparator output signal if the first comparator output signal is less than the preselected value.

Claim 3

Stochastic control circuitry as in claim 1 wherein the adaptive filter comprises convergence means for converging the input signal in accordance with adaptive filter impulse response coefficients and means for updating the adaptive filter impulse response coefficients utilizing the third comparator output signal.

Claim 4

Stochastic control circuitry responsive to an input signal S(n), where s'((n)=[S(n), S(n-1),... S(n-m)], where n and m are time variables and denotes transposition, the stochastic control circuitry comprising: (a) a communications channel including means for implementing a communications channel impulse response function H, where H'=(h 0 , h 1 , h 2 ,...,h m ), to generate a communications channel output signal y(n), where ##EQU5## (b) an adaptive filter including means for implementing an adaptive filter impulse response function A(n), where A'(n)=(a 0 (n), a 1 (n),... a m (n)), where a 0 (n) through a m (n) are adaptive filter coefficients, and having a next state A(n+1) where A(n+1)=A(n)+g(n)*f 1 (e(n))*f 2 (S(n)) and e(n)=y(n)-x(n), where e(n) is a prediction error, g(n) measures a gain, and f 1 and f 2 are generalized functions, to generate an adaptive filter output signal x(n), where x(n)=S'(n)*A(n), the adaptive filter connected in parallel to the communications channel; (c) first comparator means for comparing the communications channel output signal y(n) and the adaptive filter output signal x(n) in accordance with a first convergence algorithm to generate a first error prediction signal; (d) second comparator means for comparing the communications channel output signal y(n) and the adaptive filter output signal x(n) in accordance with a second convergence algorithm to generate a second error prediction signal; and (e) third comparator means, responsive to the first and second error prediction signals, for comparing the first error prediction signal and a preselected value and for providing the first error prediction signal as a selected error prediction signal if the first error prediction signal is greater than or equal to the preselected value and for providing the second error prediction signal as the selected error prediction signal if the first error prediction signal is less than the preselected value.

Claim 5

Stochastic control circuitry responsive to an input signal S(n), where s'((n)=[S(n), S(n-1),... S(n-m)], where n and m are time variables and denotes transposition, the stochastic control circuitry comprising: (a) a communications channel including means for implementing a communications channel impulse response function H, where H'=(h 0 , h 1 , h 2 ,...,h m ), to generate a communications channel output signal y(n), where ##EQU5## (b) an adaptive filter including means for implementing an adaptive filter impulse response function A(n), where A'(n)=(a 0 (n), a 1 (n),... a m (n)), where a 0 (n) through a m (n) are adaptive filter coefficients, and having a next state A(n+1) where A(n+1)=A(n)+g(n)*f 1 (e(n))*f 2 (S(n)) and e(n)=y(n)-x(n), where e(n) is a prediction error, g(n) measures a gain, and f 1 and f 2 are generalized functions, to generate an adaptive filter output signal x(n), where x(n)=S'(n)*A(n), the adaptive filter connected in parallel to the communications channel; (c) first comparator means for comparing the communications channel output signal y(n) and the adaptive filter output signal x(n) in accordance with a first convergence algorithm to generate a first error prediction signal; (d) second comparator means for comparing the communications channel output signal y(n) and the adaptive filter output signal x(n) in accordance with a second convergence algorithm to generate a second error prediction signal; and (e) third comparator means, responsive to the first and second error prediction signals, for comparing the first error prediction signal and a preselected value and for providing the first error prediction signal as a selected error prediction signal if the first error prediction signal is greater than or equal to the preselected value and for providing the second error prediction signal as the selected error prediction signal if the first error prediction signal is less than the preselected value.

Claim 6

Stochastic control circuitry as in claim 3 wherein the adaptive filter further comprises means responsive to the selected error prediction signal for updating the adaptive filter coefficients.

Claim 7

Stochastic control circuitry as in claim 4 wherein the first convergence algorithm is the Least mean Squares Algorithm and the second convergence algorithm is the Sign Algorithm.