A New Vision-Aided Beam Prediction Scheme for mmWave Wireless Communications

Beamforming is one of the most critical technologies for millimeter wave (mmWave) wireless communications. However, the classical beam selection methods generally require the large system overhead, such as frequent channel estimation. To this end, this paper proposes a new vision-aided beam prediction scheme based on object detection for mmWave wireless communications. In this scheme, the optimal beams are directly predicted based on the RGB images captured by cameras. Specifically, we first adopt the object detection model to locate the positions of all targeted users in the RGB images. Then, based on multilayer perceptron, an angle prediction model is used to estimate the angles between the users and the cameras. In addition, we select the optimal beam indexes from a pre-defined beam codebook according to the information of angle and codebook. Finally, the experimental results reveal that our proposed scheme has the lower overhead, higher accuracy and better scalability.