Adaptive log domain filters for system identification using floating gate transistors

We describe an adaptive log domain filter with integrated learning rules for model reference estimation. The system is a first-order low pass filter implemented using multiple input floating gate transistors operating in subthreshold to realize on-line learning of gain and cut-off frequency. We use adaptive dynamical system theory to derive robust control laws for gain and cut-off frequency adaptation in a system identification task. Simulation results show that convergence is slower using simplified control laws but still occurs within milliseconds. Experimental results confirm that the estimated gain and cut-off frequency track the parameters of the reference filter. The adaptive log domain filter has measured power consumption of 33 μW. During operation, deterministic errors are introduced by mismatch within the analog circuit implementation. An analysis is presented which attributes the errors to current mirror mismatch.