Novel direct digital frequency synthesis architecture based on artificial neural networks

This paper presents a new architecture for direct digital frequency synthesis (DDS), which employs artificial neural networks for phase to sine amplitude conversion. Also, as a contribution to electronic implementation of neural networks, a quasi-sigmoid non-linearity function is proposed which is easily implemented by using standard digital hardware. Designed in a 0.5-μm standard CMOS process, the system consists of 156205 transistors, and consumes about 24 mm² silicon area.