C-V2X Beam Prediction in the Era of Big Data Using Sequence to Sequence Model

Abstract

The introduction of wireless connectivity to a world of automobiles brought many challenges which was transformed to many innovations. Researchers worldwide continuously strive hard to develop new technologies to improve the connectivity problems and enhance the user’s comfort and enhance the safety of the users. Vehicle connectivity feature such as cellular connectivity for Wi-Fi connection provides better user convenience and Cellular-Vehicle to Everything connectivity provides safety features. Cellular Vehicle to Everything (C-V2X) has been an interesting technology which includes various connectivity methods such as Vehicle to Vehicle (V2V) connectivity, Vehicle to Network (V2N), Vehicle to Pedestrians (V2P), Vehicle to Infrastructure (V2I) and many more. The main goal of C-V2X system is to improve the safety of the vehicle and its surroundings. 3rd Generation Partnership Project (3GPP) has been working on standardizing the C-V2X which is referred as PC5. Department of Transportation (DOT) and National Highway Traffic Safety Administration (NHTSA) governs the C-V2X system which issued Notice of proposed Rulemaking (NPRM) for the V2V communication which is based on the Dedicated Short-Range Communication (DSRC) defined in SAE J2735. A 360 degree “awareness” is expected from the V2V communication which provides the complete coverage for the vehicle with a range of 300 meters which leads to the adoption of omnidirectional antenna. Omnidirectional radiational antenna provides the 360 degree “awareness” and provides us the 300-meter coverage, but it also increases the congestion factor, which has been regulated in SAE J2945/1. In a highly congested location where there are multiple vehicles present the congestion factor is high i.e., there are high loads of data present everywhere and to reduce the congestion factor, the radiation power will be reduced. But, if the radiation power is reduced, it reduces the coverage requirement. To communicate for longer range without the requirement of reducing the power and increasing the congestion would be through beam i.e., beamforming. The method to steer an array of antenna in an intended direction is called Beamforming. The radiated energy is concentrated into a narrow beam by adding the radio frequency (RF) signal either constructively or destructively based on the phase of the input RF signal. In various standards such as Wi-Fi and 5G, all the beams of the antenna perform scanning for each beacon interval (BI) and based on the various received signal, the optimum beam is chosen and adopted during the whole BI. If we implement the same methodology for our beam, we will get a medium or significant non-optimum selection. To avoid the non-optimal selection of beam, in this dissertation, a novel beam selection, “Intelligent Beam Selection” (IBS), was proposed, based on sequence-to-sequence machine learning prediction which enhances the selection of beam in real time with better accuracy compared to the traditional machine learning model. In this research, IBS predicts the optimal beam to choose from various beams integrated to the vehicle as part of the C-V2X system. Deep learning (DL) models are developed by mapping the signal strength of the various antennas which are collected over the simulation and real drive scenario. The trained functional were utilized to predict the future beam of the vehicle, reflecting better signal reception without increasing the congestion factor. The IBS model is developed for the beam selection, but the model shall also be used for other time series feature prediction in real world scenarios.