Maximizing Uplink and Downlink Transmissions in Wirelessly Powered IoT Networks

This paper considers the problem of scheduling uplinks and downlinks transmissions in an Internet of Things (IoT) network that uses a mode-based time structure and Rate Splitting Multiple Access (RSMA). Further, devices employ power splitting to harvest energy and receive data simultaneously from a Hybrid Access Point (HAP). To this end, this paper outlines a Mixed Integer Linear Program (MILP) that can be employed by a HAP to optimize the following quantities over a given time horizon: (i) mode (downlink or uplink) of time slots, (ii) transmit power of each packet, (iii) power splitting ratio of devices, and (iv) decoding order in uplink slots. The MILP yields the optimal number of packet transmissions over a given planning horizon given non-causal channel state information. We also present a learning based approach to determine the mode of each time slot using causal channel state information. The results show that the learning based approach achieves 90% of the optimal number of packet transmissions, and the HAP receives 25% more packets as compared to competing approaches.

💡 Research Summary

**
This paper addresses the joint scheduling of uplink and downlink transmissions in a wireless‑powered Internet‑of‑Things (IoT) network that employs Rate‑Splitting Multiple Access (RSMA) and simultaneous wireless information and power transfer (SWIPT) with power‑splitting receivers. Unlike conventional time‑division duplex (TDD) systems where downlink and uplink phases are fixed, the authors introduce a “mode‑based” time structure: each time slot can be assigned either to downlink or uplink operation depending on instantaneous channel conditions. In downlink slots the hybrid access point (HAP) transmits a common RSMA message and private messages to the devices; each device splits the received RF power into an energy‑harvesting branch and an information‑decoding branch. In uplink slots the devices, powered by the harvested energy, simultaneously transmit two RSMA sub‑messages to the HAP, which decodes them using successive interference cancellation (SIC) in a chosen order.

The core contribution is a mixed‑integer linear program (MILP) that jointly decides (i) the mode of every slot over a planning horizon, (ii) the transmit power allocated to each common and private sub‑message, (iii) the power‑splitting ratio of each device in downlink slots, and (iv) the SIC decoding order in uplink slots. The objective maximizes the total number of successfully decoded packets (both at the devices and at the HAP) while respecting per‑slot power limits, SINR thresholds for common and private messages, and the energy‑causality constraints that couple device battery levels across slots. This MILP assumes non‑causal (future) channel state information (CSI) and therefore yields an upper bound on achievable performance.

Because future CSI is unavailable in practice, the authors propose a reinforcement‑learning (RL) based scheduler that operates with only causal CSI. The RL agent observes the current channel gains and the residual energy states, then selects the slot mode (downlink or uplink). The reward is the number of packets successfully transmitted in that slot, penalized heavily if any energy‑causality constraint is violated. Once the mode is fixed, a separate linear program (LP) is solved for that slot to obtain the optimal power allocation, power‑splitting ratios, and, for uplink slots, the optimal SIC order. This two‑stage approach dramatically reduces computational complexity while still adapting to real‑time channel variations.

Simulation results consider a network with ten devices, twenty time slots, a path‑loss exponent of three, and realistic non‑linear energy‑harvesting models. The proposed RL‑based scheduler achieves about 90 % of the packet throughput obtained by the optimal MILP solution. Compared with simple baselines—round‑robin mode selection and random mode selection—the RL scheduler delivers 15 % and 25 % more packets, respectively. The gains stem from the ability to defer uplink transmissions to slots with favorable device‑to‑HAP channels and to allocate downlink slots to periods with strong HAP‑to‑device channels, thereby improving both harvested energy and data‑rate efficiency.

The paper’s contributions can be summarized as follows: (1) introduction of the first RSMA‑enabled SWIPT IoT model with mode‑based time slots; (2) formulation of a comprehensive MILP that jointly optimizes mode selection, power allocation, power‑splitting, and SIC order; (3) design of a reinforcement‑learning framework for online mode selection using only causal CSI; and (4) extensive performance evaluation demonstrating near‑optimal throughput and substantial improvements over conventional TDD schemes.

Nevertheless, the work has limitations. The MILP’s computational burden grows rapidly with the number of devices and slots, making it impractical for large‑scale deployments without decomposition or approximation techniques. The RL approach relies on sufficient training data and a well‑designed reward structure; its performance may degrade under highly non‑stationary channels or bursty traffic. The current system model assumes a single‑antenna HAP and a single frequency band; extending the framework to multi‑antenna beamforming, multi‑carrier OFDM, or multi‑cell scenarios would introduce additional variables and constraints. Future research directions include (i) incorporating multi‑antenna RSMA precoding and multi‑carrier resource allocation, (ii) developing distributed or federated learning methods for scalable online adaptation, (iii) multi‑objective optimization that balances throughput, fairness, latency, and energy efficiency, and (iv) experimental validation on hardware testbeds to assess real‑world feasibility.