Wireless Sensing of radios

Sensing of Radios for autonomous systems, indoors and outdoors:

In the previous theme, we developed algorithms to sense the spectrum and spectral activity to enable AI-driven networking and/or sensing-driven communication. Next, my research work has endeavored to build enhanced sensing of radios and the devices they are attached to for both indoors and outdoors; enabling indoor location and sensing services for navigation, analytics and smart control for IoT devices, and upcoming applications in smart warehouse and Industrial IoT 4.0 is well known.

Outdoor location services have thrived over the last couple of decades because of well-established platforms for both these components (e.g., Google Maps for mapping and GPS for positioning). Due to the lack of GPS signals indoors, extensive work has been done on using wireless signals, even ambient light, for localization indoors. However, this work lacked the ability to use existing off-the-shelf infrastructure like COTS WiFi access points, which limited their deployment. Besides leveraging existing infrastructure, the multi-path in the indoor environments limited the accuracy of localization. To this end, our work developed the first indoor sub-meter accurate localization system on COTS WiFi access point by leveraging the channel information and developing a two-dimensional (time-of-flight and angle-of-arrival) based MUSIC algorithm. The algorithm enabled eliminating the multi-path and achieved accurate localization. We developed a first sub-meter accurate localization for the COTS BLE devices developing protocol which provides the channel estimate, following a time-difference of arrival based technique to provide sub-meter accurate localization. The protocol was concurrently adopted into IEEE standards (hold patent which would be required for anyone to commercialize BLE localization using this protocol [IV.A.12]).

Our work developed sub-meter accurate location of wireless devices using standard WiFi and Bluetooth, but lacks mapping. To deploy any indoor navigation or location services, one needs a map of the environment with mapping of the WiFi/BLE infrastructure. The challenge is that the community has studied localization without the map’s context; the knowledge of maps is lacking in the localization provided. The follow-up challenge is that even if we had some maps, these maps do not have enough detail of WiFi/BLE infrastructure anchor points. These problems are the bottleneck of the lack of any localization system deployment. In my work, we developed a context-driven localization using WiFi for the first time, solving one of the core challenges with indoor localization. We developed and designed autonomous indoor robots, which helped with the mapping and leveraged it to enable contextual WiFi localization [Fig. 3, IV.A.19]. Furthermore, for the first time, this work developed a mechanism to project the RF signals or RF sensing information in the form of images and developed a deep learning approach to leverage the maps in the localization and provide accurate localization. Our work created open datasets for the entire community, creating image-net (the largest dataset of images for the computer vision community) equivalent for indoor navigation and localization. With Google and many other companies looking for ways to build indoor maps and localization – this work would form the first work in this direction and overcome the lack of maps in localization. With the power of AI beating the multi-path, it has shown worse case performance to 1-meter accurate localization.

Another fundamental problem in enabling deployable indoor localization dimension was to overcome the lack of knowledge of access points within the map, which has started a research area of its own. It was assumed that knowledge of the location of the access points within the map was sufficient, but it turns out you need a very accurate orientation of the access points to be known as well – a shocking realization was made by my recent work [IV.A.17]. The need for precise orientation led to achieving mm-accurate localization for the access points in the map i.e., with an accurate orientation of the antenna and location of the anchor points, one can achieve sub-meter precise localization for a user to provide indoor services. The solution presented in this work used a robot that built maps indoors, was augmented to combine the rate of change of bearing information to deduce the orientation of the anchor points (access points). It was an impressive technique, creative and groundbreaking contribution. It would have many more applications in AR/VR to provide mm-accurate localization. Furthermore, this would enable robots to use WiFi as a sensor in the SLAM (simultaneous localization and mapping) like a camera. Leverage existing WiFi access points as accurate landmarks (anchor points) to develop a SLAM algorithm that requires no loop-closure for precise mapping (submission to RAL–Robotics and Automation Letters).

In recent times, smart warehouses and smart manufacturing are emerging as applications enabled IoT 4.0, which requires localization. However, the localization has new requirements for safety applications, in specific, they require low latency and low power per location. My group has developed an area of low-power, low-latency localization systems for these applications [Fig. 4, IV.A.30]; we have developed atop of UWB a first localization system that provides a low-power and low-latency localization. Specifically, our work shows a 10x energy-efficient protocol and 10x lower latency than the latest protocol [IV.A.30]. Our localization accuracy is close to enabling a motion tracking system for humans. Looking towards the future, we would venture into applications beyond indoor localization, AR/VR applications, smart manufacturing, and end-to-end application considering compute and communication and sensing.