People of ACM - Mo Li

January 21, 2020

Why is this an exciting time to be working in the wireless and sensing fields?

Wireless communications have become embedded in our daily life. The latest 5G initiative has promised extremely high-speed low-latency wireless technology, and the growing Internet-of-Things (IoT) technology offers pervasive connectivity for vast computing and sensing devices.

In our most recent studies, we exploited the wireless signal propagation itself as a novel way to see, hear, and feel the physical world. The wireless signal, when traveling in the air, interacts with intermediate objects/obstacles before it reaches the receiver. Such interactions (reflection, diffraction, scattering, etc.) lead to attenuation of signal amplitudes and rotation of signal phases, which conveys essential information about the environment and, if measured, can be used to construct key knowledge such as the location or motion of objects/obstacles in the environment. We developed the Atheros CSI tool for accessing the channel state information from commodity WiFi interface. The tool has been open-sourced and used by thousands of researchers and engineers worldwide.

The research in this direction is still at its early stage and is pioneered by a few leading research groups in the world. I believe the progress in this direction will once again greatly benefit our life and work. I am very excited to contribute to such a technological revolution.

In the recent paper "Known and Unknown Facts of LoRa: Experiences from a Large-scale Measurement Study,” you and your co-authors stated that low-power wide-area networks (LPWANs) will be instrumental in the future scalability of the Internet of Things. Will you briefly describe what LPWANs are and why they hold such promise?

The LPWAN is a type of wireless wide-area network designed to allow long-range wireless communications at a low data rate. LPWAN operates at lower cost and at a much higher energy efficiency than other networks, and is suitable for supporting a greater number of end devices over a wide area. The term “LPWAN” also usually refers to a group of similar-purposed technologies such as long-range (LoRa), narrowband Internet of Things (NB-IoT), and Sigfox, a French global network, among others.

LPWAN’s specific attributes—such as long range, low power, and high multiple access—distinguish it from conventional wireless technologies, and provide a dimension that has been lacking in IoT. The high bandwidth and high throughput abilities which had been the main focus of the wireless technology research in past decades are now exchanged for power efficiency and scalability. LPWAN fits with general use cases of IoT applications where a scaled deployment that requires low maintenance is more desirable than the frequency and accuracy of the data.

As a fairly new technology, LPWAN has been constantly growing and is still far from mature. We see need for further research in areas including wireless spectrum limitation; coexistence with other wireless technologies; detailed power modeling and understanding; wireless medium access control; physical improvement; and so on. We have done a preliminary study and reported our measurement experience with regard to some of those aspects in the paper you mention.

Recently, you and your co-authors also researched how deep learning may improve vehicular traffic prediction in "Urban Traffic Prediction from Mobility Data Using Deep Learning.” How might a deep learning approach improve the way in which we are currently modeling traffic?

Traditional traffic analysis in transportation builds mathematical models to describe rational behaviors of humans, and based on these models, applies their impact to vehicle traffic. Recent studies in computer science and transportation take a different, data-driven approach, aimed at understanding real human behaviors and their impact.

Deep learning is very efficient at discovering intricate structures in datasets that are large and complex. Such an ability greatly benefits traffic modeling, since traffic, by its nature, is a result of complex interplay among factors including travel demand, road connectivity, transport regulations, weather conditions, various events, and many others. Recent advances in applying neural networks on graph-structured data allow us to model traffic propagation in road networks as graphs—its most natural form, from which I see great potential in building more accurate and robust traffic models.

What is another exciting research area in your field that will play an important role in how we live and/or work in the coming years?

There has been recent interest in the combination of artificial intelligence (AI) technology with the IoT, also referred to as AIOT. While IoT provides us with unprecedented ability to penetrate our living environment, the use of sophisticated AI models would help us to more deeply understand sensing data and make the most sense of it. A key challenge lies in the fact that the AIOT system is highly heterogeneous; the available computation and communication resources at different levels of the system are diverse and therefore require novel AI designs that can be appropriately distributed and incorporated into the IoT. Early adoption of AIOT may be seen in Industry 4.0 applications, where AI may play an important role in proactive decision-making processes.