Use of Multilevel PWM in the VLSI Implementation of Neural Networks
This paper presents a novel method for implementation of Neural Networks using pulse stream technique. We utilize multilevel pulse width modulation (PWM) as the pulse stream signals. Pulse streams based on multilevel PW1VI are waveforms which convey analog information both in amplitude and time axis. The multilevel PWM signals act as the input and output signals in the artificial neural network (ANN). One of the most frequently used operation in ANN is multiplication which is achieved by controlling the integration time of the current through integrator. Using PWM signals to control the flow duration of current, multiplication can be realized by simple mixed mode circuits. Multilevel implementation helps in reducing the frequency as well as the chip area of the designed neural network. Neural network to solve for any linearly separable function is designed and simulated and simulation results verify the functioning of the ANN. The designed circuit has good linearity and large dynamic range after compensating for non ideal effects of switches. The measured results on the learning examples of AND function, OK function and the real world application in finger print based gender recognition have successfully verified the functions correctness and performance of the designed 1 neural network.
Neural Networks ulse Stream Multilevel PWM.
Keshavan Namboothiri K K Sreejith C Sreekesh Lakshminarayanan Vishnu Suresh
Department of Electronics Engineering College Of Engineering Trivandrum. Kerala, India
国际会议
海口
英文
250-254
2011-02-22（万方平台首次上网日期，不代表论文的发表时间）