Title:Machine Learning Based Cooperative Relay Selection in Virtual MIMO

Abstract:In cellular systems, virtual multiple-input multiple-output (V-MIMO) technology promises to achieve performance gains comparable to conventional MIMO. In this paper, we propose cooperative relay selection algorithm based on machine learning techniques. Willingness of user to cooperate in V-MIMO depends on his current battery power, time and day along with incentives offered by service provider. Every user has different criterion to participate in V-MIMO, but allows a specific behavior pattern. Therefore, it is required to predict willing users in the neighborhood of source user (SU), before selecting users as cooperative nodes. Only inactive users belonging to Virtual Antenna Array (VAA) cell of SU are assumed to cooperate. This reduces control overheads in cooperative node discovery. In this paper, we employ prediction algorithm using two machine learning techniques i.e. ANN and SVM to find out inactive willing users within VAA cell. The parameters such as MSE, accuracy, precision and recall are calculated to evaluate performance of ANN and SVM model. Prediction using ANN has MSE of 3% with average accuracy of 97% (variance 0.37), whereas SVM has MSE of 2.58% with average accuracy of 97.56% (variance 0.17). We also observe that proposed prediction method reduces the node discovery time by approximately 29%.