Tracking of sinusoidal frequencies by neural network learning algorithms

An adaptive signal subspace estimation algorithm with a natural interpretation in terms of neural network concepts is considered. This algorithm contains only relatively simple operations and has self-orthornormalizing properties. It is demonstrated that the algorithm can learn and track the frequency information in an unsupervised manner from the data samples. After convergence, the connection weights of the network directly define a frequency estimator. Practical issues and some related algorithms are discussed