Forecasting of Throughput across Heterogeneous Boundaries in Wireless Communications - Algorithm and Performance

In this paper we demonstrate how an estimated functional kernel-regression polynomial from a particular RF technology can be created by the mobiles being served by that technology. A 3rd order polynomial description of the regression can be used to predict future throughput by observing the "pilot" quality prior to handover. The possibly heterogeneous technologies can transmit the polynomial to the UE. The UE may use it to predict the throughput it will get in the new technology 50 to 200 m-sec prior to handover. This is accomplished by observing less than 10 samples of the new technology's pilot quality. The prediction can inform the Transport Control Protocol (TCP) layer or the application layer of the upcoming handover and the throughput expected after handover so that the user application receives the best quality of service. In this paper we: 1. propose that there is a way to predict the unobservable quality metrics in the new cell prior to commencement of the handover. This is achieved by 1) a prediction mechanism and 2) a signaling mechanism. In this paper we focus on the prediction mechanism. 2. propose that the observable metric ("pilot" quality) is predicted with prediction error of 2 to 9% with prediction step sizes of 200 m-sec. 3. show that the unobservable quantity (we choose bits/physical-resource-block = $beta$ as the link quality metric) can be predicted with error of 2 to 8% with prediction step size of 200 m-sec.