Unconstrained Hartley Domain Least Mean Square Adaptive Filter

Peak detection is a basic data analysis problem which isessential in a large number of applications. In applications such asimage processing, the large computational effort to locate peaks mayprohibit operation in real-time. A Hopfield neural network is proposedfor the peak detector to solve the real-time problem. Analytical expres-sions are derivedfor input separation, neuron gain, and restrictions oninitial conditions. Hardware limitations are discussed and a modifiedcircuit model is suggested for the Hopfield neuron. Solution time underthirty microseconds is obtainable with general purpose operationalamplifiers independent of the number of inputs. Results obtained from atwenty-five neuron hardware implementation of the network lend cre-dence to the theoretical results.