Prediction and Modeling of Spectrum Occupancy for Dynamic Spectrum Access Systems

Abstract

In a dynamic spectrum allocation (DSA) system, reliable prediction of spectrum occupancy based on a spectrum consumption model (SCM) is critical for system design, performance analysis, and evaluation. In this article, we focus on a low-level abstracted measured dataset from a massive campaign and investigate the occupancy of representative frequency bands. First, we apply an autoregressive-moving-average (ARMA) model combined with a low-pass filter, given the stationarity of the channel measurement dataset and thanks to the computational simplicity of the model. The average received power and off-state probability are extracted from the measured data. According to the results, the measured and predicted data are in good agreement. Comparing the proposed model-based ARMA with the popular long short-term memory learning algorithm, they have similar error accuracy with pre-processed data, while ARMA has a much lower training complexity. In the second step, we develop an SCM describing the spectrum usage for designing and examining the DSA system. We extract the periodic, aperiodic low-frequency, and burst components of the time series. Also, a binary sequence is extracted from a sparse occupancy channel, and modelled by a non-homogeneous Markov chain. Results show that the model-generated data can maintain the same statistics as the measured data.