Einsteinian neural network for spectrum estimation

Abstract A model-based neural network is developed for spectrum estimation. Its architecture and learning mechanism are founded on the Einsteinian interpretation of the spectrum as a probability distribution of photons. By considering a spectrum as an ensemble of photons, we derive the neural learning mechanism from the basic physical principle of entropy maximization of a canonical ensemble. This neural network is applied to characterizing a recently observed phenomenon known as equatorial ionospheric clutter that significantly affects operations of over-the-horizon (OTH) radars and communication links using high frequency radiowaves propagating through the ionosphere. We utilize a specific parameterization of the internal spectral model, which is derived from the physical principles of the propagation of electromagnetic waves through a turbulent ionosphere. A set of parameters characterizing equatorial ionospheric clutter is estimated. The developed technique may have a broad applicability in scientific data analysis.