Abstract
Naturally occurring radio noise above approximately 100 MHz is well modeled for most applications as a Gaussian random process. Below 100 MHz, however, radio noise is impulsive in nature and is not well modeled as Gaussian. Individual atmospheric events (mainly "sferics", the electromagnetic emissions from lightning) produce large noise impulses, so power is not spread evenly in time but instead depends on the occurrence of discrete, clustered events. Due to large variations in the occurrence of sferics on a seasonal and diurnal basis (and with the passing of individual storms), atmospheric noise is non-stationary.

The objective of this work is the statistical characterization and modeling of atmospheric radio noise in the range 10 Hz -- 60 kHz (denoted "low-frequency" noise), with the specific goal of improving communication systems operating in this range. The statistics analyzed include seasonal and diurnal variations, amplitude probability distributions (APD's), impulse interarrival time distributions, impulse counting statistics, background noise statistics, and power spectral densities. The analyses are based on many thousands of hours of noise measurements made by the Stanford Radio Noise Survey System.

Based on the statistical characteristics of the noise data, a new clustering Poisson atmospheric noise model is developed. This model is based on several previously known statistical-physical models, but in addition it takes into account the clustering aspects of sferics. It is verified that the clustering model accurately predicts the statistical features found in the data.

Low-frequency noise is shown to have an alpha-stable noise envelope distribution for locations removed from heavy sferic activity. (The alpha-stable distribution as applied to low-frequency noise is similar to a Gaussian distribution but with heavier tails.) Since many low-frequency communication systems operate at such locations (e.g., the polar regions), it is of value to compare the performance of receivers specifically designed for alpha-stable noise with that of the receivers currently in use. Simulations are performed using real time-series data, and alpha-stable receivers are shown to give improved performance for certain signal formats.