Learning-Based Multiplexing of Grant-Based and Grant-Free Heterogeneous Services with Short Packets

In this paper, we investigate the multiplexing of grant-based (GB) and grant-free (GF) device transmissions in an uplink heterogeneous network (HetNet), namely GB-GF HetNet, where the devices transmit their information using low-rate short data packets. Specifically, GB devices are granted unique time-slots for their transmissions. In contrast, GF devices can randomly select time-slots to transmit their messages utilizing the GF non-orthogonal multiple access (NOMA), which has emerged as a promising enabler for massive access and reducing access latency. However, random access (RA) in the GF NOMA can cause collisions and severe interference, leading to system performance degradation. To overcome this issue, we propose a multiple access (MA) protocol based on reinforcement learning for effective RA slots allocation. The proposed learning method aims to guarantee that the GF devices do not cause any collisions to the GB devices and the number of GF devices choosing the same time-slot does not exceed a predetermined threshold to reduce the interference. In addition, based on the results of the RA slots allocation using the proposed method, we derive the approximate closed-form expressions of the average decoding error probability (ADEP) for all devices to characterize the system performance. Our results presented in terms of access efficiency (AE), collision probability (CP), and overall ADEP (OADEP), show that our proposed method can ensure a smooth operation of the GB and GF devices within the same network while significantly minimizing the collision and interference among the device transmissions in the GB-GF HetNet.