Learning methods for long-term wireless channel prediction

Spearker: Federico Chiariotti, Date: Friday the 29th of April 2016, Time: 14:30, Room: 201 (DEI/A)

Abstract. The ever stricter Quality of Service (QoS) conditions demanded by high-bandiwdth and real-time applications require a foresighted use of the wireless channel; a prediction with a timescale of several seconds might be needed in several contexts. Accurate short-term prediction methods were developed to calculate MIMO pre-coding matrices, but long-term channel prediction is still a largely unexplored field. In this work, we propose three different learning methods to predict the future channel gain of a mobile terminal using only the average Received Signal Strength Indicator (RSSI). This presentation will focus on the Bayesian graphical method, as we don’t yet have complete results for the other two techniques. As the data-gathering is quite time-consuming, the work is still in progress and all our results and conclusions are still tentative.