A smart ultrasonic beam controller

This paper presents the system design of a smart controller which simulation studies show has the potential of achieving ultrasonic-linked, voice selective control through neurofuzzy network and FFT software embedded in a dual microprocessor system. The neurofuzzy network serves as a speech recognition processor which is speaker independent. The second microprocessor performs an FFT function on the input signal. The output of the neurofuzzy network is a digital pattern in which the 1´s switch on members of an array of ultrasonic transducers each tuned to a different frequency, Integration techniques which hold promise are discussed. Performance of the neurofuzzy network software trained by five different training algorithms was measured and compared in terms of metrics: recognition accuracy, learning rate, speaker sensitivity error, recall/forget ratio, and false command reject/accept ratio. Algorithms which provide the best results are determined. In addition the effects of varying the number of input nodes and the number of fuzzy logic classes are determined