Abstract
Millimeter wave communications employ the use of high frequency signals for higher bandwidth and data rate. However, the signal attenuation over short distances is a main issue since these higher frequency signals are easily blocked by physical objects such as trees and buildings. The signals can be usually boosted for longer ranges with the cost of additional power consumption. To overcome the problem of the power consumption, smart antenna systems are used which can predict the direction in which high gain and highly directive beam needs to be generated to maximize signal efficiency and minimize communication loss. In this report, the system employs cameras installed in base station to identify the location of users and blockages to determine the direction of the output beam. Therefore, the direction of output beam can be predicted, by using machine learning approach. ML models can then be fed into the system to predict direction of beams for future beams. In this project, three neural network algorithms are employed to predict future beams by using Long short-term memory network, Bidirectional LSTM network and Bidirectional GRU network. Also, the dataset used for the project is provided by ViWi (Vision Wireless) for Vision-Aided Millimeter Wave Beam Tracking Competition which has respective beams mapped to these images. Using long short-term memory neural networks and Bidirectional recurrent neural networks results are improved to 87% with a gain of 3 percent when compared to baseline score.