A complex analogue neural network with digitally controlled parameters
Owen E. Holland · 1991
Summary form only given, as follows. An ongoing project to construct a modular, digitally controlled, fully interconnected analogue neural network of up to 48 cells is discussed. The neural unit models synapse and membrane events at the level of ionic currents and conductances, and includes a representation of hypothetical presynaptic processes leading to the phenomena of fatigue and sensitization. The satisfactory implementation of analog synapses using field-effect transistors as voltage-controlled conductances is shown to require a calibration stage not found in previous models; a circuit modification for producing a substantially linear control of synapse strength, thus enabling improved accuracy of calibration, has been obtained. The system architecture is of interest in the context of applications in the field of autonomous mobile robots.>