Uncovering the Complexity of WSPR Daemon Noise Through Kernel Density and Gaussian Mixture Models

This study describes and analyzes the multimodal nature of the noise measurements in Weak Signal Propagation Reporter (WSPR) daemon data. We statistically modeled the noise data from several stations using a combination of non-parametric and parametric methods to explore the underlying probability structures. Non-parametric kernel density estimation (KDE) was used to define a smooth, continuous estimate of the empirical probability density function, revealing that the observed noise values do not conform to a simple Gaussian distribution and instead exhibit strong multimodal characteristics. To explore these features further, Gaussian Mixture Models (GMM) were applied to estimate the overall distribution as a weighted sum of multiple Gaussian components. The number of components presented was chosen using both visual analyses, the Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) considering both goodness of fit and model parsimony. The fitted density curves and component contributions varied widely between stations. Most showed that the WSPR daemon noise profile can be decomposed into several overlapping sub distributions, suggesting multiple contributing noise sources related to propagation conditions, receiver and antenna configuration, and possible anthropogenic interference.