Dimensional reduction of analog signals with a neural processor

We describe a neurally inspired processor that transforms complex analog signals into linearly independent representations of these signals. The processor uses on-chip learning to adapt weights to provide detection of principle features in complex waveforms. The chips consist of a linear sum of products section, a principle components weight adaptation section, and a lateral inhibition section. We use several elementary principles from biology to construct our neural processor. Experimental data demonstrates the chip detecting and encoding principle features found in complex temporal signals.<>