SBIR Phase I: Scalable Mesh Routing to Augment Low-Power Wide Area Internet of Things (IoT) Networks

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project seeks to enhance the understanding of wireless mesh networks and deep reinforcement learning algorithms in order to significantly expand the coverage and robustness, accelerating Internet of Things (IoT) adoption across the globe. These improvements to sensor systems seek to benefit a range of applications including public safety, smart agriculture, supply chain logistics, smart cities, wildlife monitoring, healthcare and other markets. The coverage and cost benefits are especially impactful for the U.S population in rural or economically-disadvantaged areas that lack cost effective connectivity and are unable to take advantage of the IoT benefits. Further, the project will enhance industry-academia partnership, enable the technology transition of innovations, and expand the participation of women in science, technology, enducation and mathematics (STEM).This Small Business Innovation Research (SBIR) Phase I project seeks to enable scalable, longer-lasting, and higher-throughput Low Power Wide Area (LPWA) Internet of Things (IoT) networks by using LPWA IoT Mesh Augmentation (LIMA) devices to augment the connectivity between end-nodes and gateways in a cost-effective and easy-to-deploy manner. The mesh network of LIMA devices will adaptively multi-hop relay messages to maximize effective capacity and range. The teams seeks to augment the connectivity of LoRaWAN (a standard for Long Range Wide Area Networks) with a goal of increasing coverage range several-fold, reducing the battery drain of end-nodes, and enabling higher uplink bitrates. The core challenge for such LIMA is the development of a scalable multi-hop mesh routing protocol that, unlike existing protocols, accommodates the ultra-low bit rates that characterize LPWANs, and is generalizable across the diverse applications of LPWAN technology. The LIMA solution builds upon two synergistic innovations: (a) embedded-control routing, that uses a very small amount of bits in the data packet header in lieu of control packets, and thereby achieves scalability and energy-efficiency in ultra-low-capacity, energy-constrained networks and (b) relational deep reinforcement learning based routing that combines Reinforcement Learning and Deep Neural Networks (DeepRL) with the use of relational features to learn routing policies that adapt to a variety of link dynamics and traffic conditions. The LIMA solution does not require changes to end-nodes. The anticipated technical deliverables include mesh networking software, a simulation model, a 4-node prototype using a LoRa integrated circuit, and experimental analysis.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.