Amalinda Gamage

I am actively looking for faculty positions.

Publications | Talks | Teaching | Service

Hello, I'm Amalinda from Sri Lanka 🇱🇰. ආයුබෝවන්! (āyubōwan!)—a warm wish for a long and prosperous life in Sinhalese.

Central to my research is spectral optimization, specifically in the implementation of physical-layer techniques that enhance the coexistence and fairness of LPWANs. I also have a keen interest in all things RF—particularly unconventional low-power transceivers, which I believe will be essential to power the next billion IoT devices.

My research has had real-world impact. For example, LMAC was adopted into the LoRaWAN standard by the LoRa Alliance. Our efforts marked the very first industry-academia collaboration of the LoRa Alliance to have resulted in a technical standard: a collaboration between Semtech (France), the patent holder of LoRa, and NTU, SG. For these contributions, I was named one of the 2023 Honorees of the LoRa Alliance Contribution Award at LoRaWAN Live’24, an honor shared among contributors who advanced the LoRaWAN standard globally. Since 2023, LMAC has been integrated into the official LoRaWAN library. More recently, LMAC has also found its way into the firmware of some of Semtech’s newest LoRa radios. Consequently, LMAC now lives inside millions of LoRaWAN devices worldwide. Two of my other works on LPWAN reliability have also been licensed to industry through NTU.

I earned a Ph.D. degree in Computer Science from NTU, SG, and my Bachelor's degree in Electronic Engineering with first-class honors from SHU, UK. I learned the ABCs of research and much more from Prof. Mo Li, my PhD mentor. Prior to joining NTU, I was a lecturer at both SLIIT and UNIVOTEC in Sri Lanka. I am also an elected member of the Institution of Engineers Singapore (IES), a signatory of the Washington Accord.

[SUMMARY] LMAC, my Ph.D. work that brings CSMA to LoRa, is now an industry standard. In 2022, we proposed LMAC to the LoRa Alliance, and the Alliance formed the CSMA Task Force (CSMA-TF) to take it forward. Within the CSMA-TF, we worked closely to find the best way to bring CSMA to LoRaWAN. We also modified LMAC together to suit global regulations. The CSMA-TF then forwarded TR-013 to the LoRa Alliance Technical Committee (TC), where the members voted to accept it in June 2023, and it became part of the LoRaWAN standard. This was the very first industry-academia collaboration in the LoRa Alliance to result in a technical standard: a collaboration between Semtech and NTU. For this work, my Ph.D. mentor (Prof. Mo Li) and I were named among the 2023 Honorees of the LoRa Alliance Contribution Award at LoRaWAN Live’24 in Germany (PR).

Since then, TR-013 has been part of the official LoRaWAN codebase. More recently, it has also been built into the radio firmware of newer generations of Semtech's LoRa radios. Consequently, LMAC is now embedded in millions of LoRaWAN devices. Thank you, Geoffroy, Miguel and Olivier at Semtech, and the LoRa Alliance, for this wonderful collaboration. Merci beaucoup !

[SUMMARY] On-demand downlinks can enable spontaneous actuation of end-devices. But this requires them to listen continuously, which costs about 43.5 mW on COTS LoRa radios. LoRaWAN Class-C enables always-on listening at the protocol level, but that energy cost drains a coin cell in under a day. In practice, only mains-powered end-devices run Class-C long term. Therefore, most end-devices operate under Class-A, where the gateway has to wait for an uplink to deliver a downlink. Hark aims to bridge the gap between what the protocol allows and what practical energy budgets can sustain. Hark is an ultra-low-power LoRa receiver that listens continuously for frames from any unmodified COTS LoRa transmitter. It achieves this through a hardware-software co-design. First, in hardware, it replaces the conventional PLL and mixer with an autodyning tunnel diode oscillator. The oscillator generates its own local carrier and mixes it with the incoming frame. One unmodulated emitter can injection-lock the oscillators of many Hark receivers, which keeps their local carriers stable. Second, in software, Hark proposes a novel multi-chirp stacking algorithm that detects LoRa frames at sub-Nyquist rates, even when frames fall below the noise floor. Sub-Nyquist sampling reduces both ADC and MCU power. Hark listens at just 1.90 mW, over 22× below the COTS receiver.

Project Website. 90-second teaser video.

[SUMMARY] SateRIoT enables reliable IoT connectivity in rural areas by complementing satellite communication with LoRa. It experimentally demonstrates the temporal and spatial challenges of satellite links and effectively mitigates their impact on network performance through a bursty link model and a multi-hop flooding protocol. SateRIoT significantly improves overall network energy efficiency and latency, and it is implemented using cost-effective COTS hardware.

[SUMMARY] This work implements collision avoidance across large-scale LoRa networks. CSMA is a technique used by many wireless protocols to avoid collisions. However, LoRa modulation is unique. A single LoRa channel can accommodate multiple frames simultaneously. LoRa frames also have the ability to traverse underneath the noise floor. Therefore, we need a CSMA protocol that doesn't debilitate LoRa's key advantages. LMAC is a simple protocol that does just that reliably with an extremely small energy overhead.

Project Website

[SUMMARY] This work extends LMAC, originally published in MobiCom'20. The extension examines several additional aspects of LoRa CSMA, such as the quantified limitations of RSS-LBT, how LMAC enhances weak LoRa transmissions, the impact of gradual adoption and coexistence of CSMA-enabled LoRa with ALOHA, global regulatory implications, and finally, how to comply with them.

Project Website

[SUMMARY] Prior works on real-time LoRa assume gateway capabilities that don’t scale well in practice, such as full-duplex operation or unlimited concurrent demodulation. This makes real-life deployment impractical. We designed LoRaHART from the ground up to work within the hardware limits of LoRa gateways. It accounts for two key constraints: the half-duplex nature of COTS gateways and their ability to demodulate at most eight frames concurrently. At the core, LoRaHART uses a TDMA scheduler with CSMA-based retransmissions when necessary, within a single superframe. On a 40-node COTS testbed, it achieves 98% PRR and 45% higher spectral efficiency than the best baseline.

[SUMMARY] We experimentally evaluate LoRa’s real-world performance using over 50 nodes across a 3×3 km testbed. LoRa can exceed 10 km under line-of-sight, but its range drops drastically in urban and indoor environments—by as much as 90%. We show that achieving long battery life requires tight parameter tuning; naïve configurations can cut node lifetime down to a year. We measure power profiles and derive lifetime models using real hardware and power measurement traces. We also profile the capacity of a real SX1301-based gateway and estimate the limits of its concurrency. Surprisingly, we find that using higher spreading factors (like SF12) supports more nodes in terms of PRR, due to lower collision rates, despite slower throughput. We then propose a simple algorithm to jointly optimize node lifetime and coverage. Finally, we highlight that LoRaWAN could add energy overhead and that it doesn’t always improve goodput, especially when downlinks are enabled.

[SUMMARY] We designed a kinetic energy harvester to power GPS tags on elephants using their natural movement. The device uses magnetic levitation and ferrofluid bearings to reduce friction and boost energy output. We collected 10 days of real acceleration data from an Asian elephant and tuned the system accordingly. The harvester generates ~89 J per day—enough to transmit location 24 times daily over 114 km line-of-sight, with a ~35 J surplus. This surplus can support higher update rates or fallback options like GPRS. The results show that long-term, maintenance-free tracking of wild elephants is feasible, which is critical as human-elephant conflict zones expand.