Vector classification by a winner-take-all neural network with digital frequency-locked loop

Abstract

This paper discusses performance of a winner-take-all neural network (WTANN) that is based on digital frequency-locked loops (DFLLs), especially its speed as well as its vector classification capability are studied. In the proposed WTANN input and weight vectors are conveyed by frequency-modulated signals, and neuron computation is carried out by the DFLL. Each DFLL uses a direct digital frequency synthesizer (DDS) as its local oscillator. Frequency resolution of signal generated by the DDS is decided by the size of internal register. Winner search operation is implemented by using frequency comparators distributed among all neurons, which makes it easier to increase the number of neurons. The proposed WTANN architecture was described by very high speed integrated circuit (VHSIC) hardware description language (VHDL) and its feasibility was tested and verified through simulations. Simulation results show that the proposed DFLL-based WTANN can find winner neuron faster than digital phase-locked Loop (DPLL) based WTANN. Another simulation was carried out by using IRIS and WINE data set to verify the classification performance of the WTANN.