A reduced size lattice-ladder neural network

An idea to incorporate filters in the common structure of an artificial neural network (ANN) is motivated by the need to process time-varying signals using ANN. IIR filters are more advantageous than FIR filters - they require less coefficients, can handle long duration signals, however their stability is not guaranteed and requires special attention. We propose new ANN structure, which is based on the lattice-ladder realization of IIR filters incorporated as multilayer perceptron synapses. It is of small size, has simple monitoring of filter stability, trains rapidly, and is easy expandable in filter orders.