Predictive analytics for non-stationary V2I channel

Vehicle to Infrastructure (V2I) channels are particularly difficult to analyze because of high mobility and localized scattering from nearby vehicles and road-side features. The spatio-temporal variation of the scattering environment makes the channel a non-stationary stochastic process, which renders conventional, receiver-side channel conditioning techniques ineffective for this emerging application. Our work takes a radically different approach to introduce predictive analytics at the Road-Side Unit (RSU) to proactively compensate for channel variations over time and frequency while precisely fitting into contemporary protocols like Dedicated Short Range Communication (DSRC) and Wireless Access in Vehicular Environment (WAVE). By assimilating the channel state feedback built into these protocols, we employ an iterative learning algorithm to gather localized knowledge of the channel profile. This acquired knowledge is used to pre-condition the downlink waveform to lower the Bit Error Rate (BER) by ≈ 100 times, when compared to the current vehicular communication standards even at a relatively high Signal to Noise (SNR) of 17 dB. Further, our algorithm is able to predict the non-stationary V2I channel with an average absolute error of 10
^-2
in dense scattering environment.