| Popis: |
Internet-of-Things (IoT) has widely permeated our daily lives and is used for various applications ranging from agriculture to industrial IoT. Recent years have seen the emergence of Low-Power Wide-Area Networks (LP-WANs) as a promising technology to connect the Internet of Things. LP-WAN technologies allow devices to send data at low data rate (few kbps) to base stations several miles away while lasting 5-10 years powered by batteries. These LP-WAN deployments are inherently asymmetric where more resourceful base stations and cloud infrastructure lie on the other side of the clients’ bandwidth starved link. This precludes the possibility of data-driven applications on LP-WAN deployments for large scale sensing such as smart agriculture, automated traffic signals and micro-climate monitoring. While most smart-home sensors can communicate large amounts of data to enable low-latency complex inference applications, the big data revolution has eluded such outdoor sensors from realizing their true potential. This thesis overcomes fundamental limitations of such low-power wireless technologies and builds systems that enable such data driven applications at scale. Firstly, we develop solutions that can push the wireless and compute operations from these low-power IoT clients to the much more powerful base stations. We further take a cross-layer approach to enable data-driven applications on these IoT sensors at scale by co-optimizing their wireless physical layer to improve client range and throughput while reducing latency for higher level network and inference objectives. My solutions apply these principles for improving LoRa, a popular LP-WAN technology, where this asymmetry is particularly stark. First, we enable base station collaboration which allows us to push components of the LoRa physical layer from the clients to the base stations and the shared cloud infrastructure. This collaboration allows us to reduce power expenditure at clients while improving communication range as well as throughput. Further, this collaborative base station framework can also enable decameter-level location tracking. Next, we develop a distributed LP-WAN physical layer that can perform complex low-latency statistical, spatial and inference queries on thousands of low-power clients. Finally, we show how we can empower LoRa-enabled CubeSats deployed in low-earth orbit by co-designing higher level data-driven applications within communication constraints for multi-fold improvement in inference and data retrieval quality. |
