Traffic Prediction for Reconfigurable Access Scheme in Correlated Traffic MTC Networks

This paper presents a learning-based multiple-access scheme for machine-type communication networks where the base station (BS) is equipped with a large-scale antenna array and devices have spatially correlated and event-based traffic. Each access time frame is divided into two segments: grant-based and grant-free, which are dynamically configured according to device traffic statistics to achieve high throughput. As device traffic statistical characteristics might not be available at the BS in reality, we propose a learning-based algorithm that exploits device traffic correlations to predict the probability of having data packets of each device in the next time frame based on its traffic history. The BS uses this predicted probability for its scheduling and access probability calculation algorithms to maximize the system throughput. Performance of the proposed learning-based multiple-scheme is evaluated and illustrative results show its superior achievable throughput as compared to both random scheduling and optimized random access schemes.