On Medium and Long Term Channel Conditions Prediction for Mobile Devices

With wireless spectrum being a precious commodity, the knowledge of the future wireless channel conditions is very valuable. Channel condition maps enable wireless service providers to identify and fix coverage gaps as they build their networks and deploy a variety of self-optimization technologies. Knowing future channel conditions for an individual mobile device is one of the key ingredients for predicting wireless link throughput. The latter is much sought after since it enables bandwidth demanding mobile applications (e.g. video) to proactively and gracefully adapt to varying wireless network throughput, resulting in significant improvement in end user quality of experience and efficiency of utilization of the existing wireless resources. In this paper, we analyze the crucial impact of the neighboring cell load on LTE channel conditions of the UE. As a metric characterizing channel conditions we use an average number of useful (without retransmissions) bits per LTE physical resource block. In particular, we come to a conclusion that popular crowdsourcing methods of assembling wireless signal strength maps based upon mobile device reports are inadequate for accurate throughput prediction due to inherent lack of the neighbor cell load information. We show how to draw a dynamic channel conditions map with the corresponding confidence intervals for both stationary and moving users, parameterized by the neighboring cell load. We then define and characterize the procedure for predicting UE channel conditions given the UE geographic pixel location and serving cell load characteristics. We find that this prediction for both stationary and moving users can have 90% or higher accuracy.