Assessing Interference with Regression Analysis Techniques

Reliably predicting aggregate interference is critical for deploying next-generation Dynamic Spectrum Access(DSA) systems. Next-generation DSA systems must operate in real-time. In current large-scale networks, aggregate interference prediction takes hours to complete. This work proposes regression analysis as a method for predicting interference. The model proposed in this work performs comparably to other machine learning approaches. Regression techniques can be fit quickly, unlike many machine learning methods which require long training. This rapid model training enables applications in dynamic spectrum environments. Our model achieved a mean absolute percentage error(MAPE) of 5.1% interference prediction with an 80/20 training validation split based on simulated data with added thermal noise density to approximate the noise of the transmission channel.