Sparsity-Cognizant Multiple-Access Schemes for Large Wireless Networks with Node Buffers

Abstract

This paper proposes efficient multiple-access schemes for large wireless networks based on the transmitters' buffer state information and their transceivers' duplex transmission capability. First, we investigate the case of half-duplex nodes where a node can either transmit or receive in a given time instant. In this case, for a given frame, the transmitters send their buffer states to the destination which assigns the available time duration in the frame for data transmission among the transmitters based on their buffer state information. The network is said to be naturally sparse if the number of nonempty-queue transmitters in a given frame is much smaller than the number of users, which is the case when the arrival rates to the queues are very small and the number of users is large. If the network is not naturally sparse, we design the user requests to be sparse such that only few requests are sent to the destination. We refer to the detected nonempty-queue transmitters in a given frame as frame owners. Our design goal is to minimize the nodes' total transmit power in a given frame. In the case of unslotted-time data transmission, the optimization problem is shown to be a convex optimization program. We propose an approximate formulation to simplify the problem and obtain a closed-form expression for the assigned time durations to the nodes. The solution of the approximate optimization problem demonstrates that the time duration assigned to a node in the set of frame owners is the ratio of the square-root of the buffer occupancy of that node to the sum of the square-roots of each occupancy of all the frame owners. We then investigate the slotted-time data transmission scenario, where the time durations assigned for data transmission are slotted. In addition, we show that the full-duplex capability of a node increases the data transmission portion of the frame and enables a distributed implementation of the proposed schemes. Our numerical results demonstrate that the proposed schemes achieve higher average bits per unit power than the fixed-assignment scheme where each node is assigned a predetermined fraction of the frame duration.