A Recursive Estimator of Spectral Noise Floor in the Presence of Signals

In this paper we propose a recursive noise floor estimator, which operates on spectral data and in the presence of intermittent signals. We use a Bayesian approach where we treat data samples consisting of signals as outliers in a Gaussian mixture distribution. The signal is modelled with a Gaussian distribution having larger power with respect to that of pure noise samples. In addition to noise floor estimation, this technique also leads to a novel detection statistic for the detection of signals in noise. We demonstrate our approach in two scenarios. Firstly, we consider a noise only case and calculate the root mean square error for the estimated noise variance over the number of spectral samples used. In the second scenario, we simulate representative radio frequency (RF) signals in the presence of noise and evaluate the performance of our approach by estimating the noise and the detection of signals within the noise. We also compare our method to another classical technique, in the form of a median filter and show that our proposed approach performs better for certain signal scenarios of interest.