Neural networks for sinusoidal frequency estimation

We present a new approach to the problem of sinusoidal frequency estimation using neural networks. The developed neural networks can simultaneously estimate frequencies, amplitudes and phases of a sinusoidal signal from noisy measurements. Furthermore, by integrating the conjugate gradient technique into the neural networks, the convergent rate of the solution is significantly improved. The developed networks are also able to track any frequency variation in signal sources. Due to the neural networksʹ massive parallelism and high processing speed, this new method is superior to the existing techniques in that the estimation can be carried out in real time. The results are illustrated by stimulation examples.