My research asks how a low-power wireless network can deliver data on time, reliably, and within a tight energy budget when thousands of devices share one channel across several kilometers. Most of this work targets low-power wide-area networks (LPWANs), and LoRa in particular, which connect battery-powered and intermittently powered sensors to a gateway in industrial settings such as oil fields, smart grids, and remote process plants. Industrial wireless standards were built for short-range networks, and they do not carry over to this scale. I design protocols and scheduling methods that give LPWANs real-time behavior, and I validate them on real hardware.
LoRa networks were designed for sporadic, delay-tolerant traffic, so they offer no notion of a deadline. RTPL is a real-time communication protocol for LoRa that exploits the ability of a gateway to receive on several channels in parallel, which improves the fraction of packets that meet their deadlines (IoTDI ’22; ACM TECS ’24). An event in an industrial plant often triggers many devices to report at once, and the resulting burst of traffic collides and misses deadlines exactly when timely data matters most. Burst-MAC coordinates this bursty traffic to reduce collisions during time-critical events (RTSS ’24). Both protocols are evaluated in NS-3 and on physical testbeds.
When an LPWAN carries the messages of a control loop, the network is part of that loop rather than a fixed delay. I co-design real-time scheduling and control for industrial CPS over LPWANs, combining schedulability analysis with adaptive sensor sampling so that control performance is optimized while every message still meets its deadline (ICCPS ’24; ACM TCPS ’26).
Many IoT devices run on harvested energy stored in a supercapacitor, so they can transmit only when they hold enough charge. I developed methods that schedule charging and transmission together, using integer linear programming and online heuristics, so that such devices still communicate in a timely way (DCOSS-IoT ’26). For deployments where a gateway cannot reach every device, I also worked on latency-aware data harvesting with a mobile sink, with a 2-approximation guarantee on collection latency (RTAS ’26).
Long Range Frequency Hopping Spread Spectrum (LR-FHSS) extends LoRa’s reach to tens of kilometers and to satellites in low Earth orbit, but it does not interoperate with the commodity LoRa receivers already deployed. We built cross-technology communication from LR-FHSS to commercial LoRa receivers (SenSys ’26). Building on this, I am developing per-message delivery certificates and signature-bound frame recovery for direct-to-satellite LR-FHSS, implemented with GNU Radio and a USRP.
I have also contributed to hierarchical multi-agent reinforcement learning for energy- and thermal-aware scheduling of parallel OpenMP workloads on multicore platforms (RTSS ’24 Brief Presentations).
Going forward, I aim to extend timeliness, reliability, and energy guarantees to direct-to-satellite IoT and to 5G/6G IoT, where devices beyond terrestrial coverage and massive numbers of connected machines raise the same questions at a larger scale.
I build and measure real systems. My experiments use software-defined radio (SDR) prototypes on USRP B200 with GNU Radio, commercial off-the-shelf (COTS) LoRa nodes and gateways, and multi-node testbeds. At scale I use NS-3 simulation, PyTorch for the learning components, and Gurobi for optimization.
The papers behind each part are listed on the publications page.