Optimum design of microwave oscillator using Hopfield Neural Network

A new method is presented to optimally design analog RF circuits such as oscillators using Hopfield Neural Networks (HNN). For this purpose, an HNN is trained with some sample realization of a specific structure of an analog circuit. Then, considering random initial points, the related HNN converges to spurious local minima of energy function which may be associated with an optimum design of the analog circuit. To demonstrate this idea in practice, a 2.4 GHz fixed-frequency oscillator is considered. The associated design parameters of the oscillator are trained in the weights of an HNN. Starting from random initialization, the HNN provides some design realizations at the convergence state (associated spurious local minima). Some of these design realizations show a better performance in terms of phase noise, power consumption and harmonic distortions.