Research Direction 02
Integrated Communication, Sensing, and Computing: Seeing Risk Before the Mountains Move
We read changes in mountains and their environment from wireless signals, using deep learning and AI models to fuse communication, sensing, and multisource monitoring data. For natural hazards such as landslides and debris flows, we explore low-power, low-cost, deployable technologies for pre-disaster monitoring and risk prediction.
Direction Lead
Direction Lead
Chengxuan YangUndergraduate Student
ycxuan0517@gmail.com
Research Partners
China Comservice
FutureComm Lab, SEU
Research Direction
Integrated Communication, Sensing, and Computing
At around 10:30 a.m. on August 26, 2026, a debris-flow disaster on the Nepal side caused major casualties and missing persons at Gyirong Port in Gyirong County, Shigatse, Tibet. What if a base station could have “seen” the mountain changing before the disaster? Landslides and debris flows do not occur entirely without warning. Tiny surface displacements, changes in soil moisture, prolonged rainfall, and changes in mountain structure can all leave slow, subtle, long-term signals before a disaster. Yet mountainous areas are often where power is hardest to supply, communications are weakest, and monitoring equipment is most difficult to maintain over time.
Core mission
Working with China Comservice and FutureComm Lab at Southeast University, we study integrated communication, sensing, and computing for natural-disaster scenarios based on the practical needs of mountainous regions. We aim to make wireless networks do more than carry data: communication signals themselves become a new kind of sensor for observing mountain changes, while AI identifies anomalies, assesses trends, and detects risks in long-term observations.
Industry background
Natural-hazard monitoring does not lack sensors. The real challenge is enabling them to operate reliably, affordably, and over long periods in remote mountainous areas.
In mountainous regions such as Tibet, many areas have inadequate power supplies and communication links may be interrupted frequently. When a link fails, sensing nodes deployed on a slope can lose connectivity and monitoring data cannot be returned in time. The more devices are deployed, the greater the burden on power, communications, construction, and ongoing maintenance.
Traditional solutions usually build communication networks and sensing systems separately. Our research takes a further step: since wireless signals such as 4G and 5G already propagate continuously through the environment, can the same signal communicate and sense at the same time?
Wireless signals undergo reflection, scattering, diffraction, and multipath propagation. Changes in mountains, ground surfaces, vegetation, rainfall, and the surrounding environment alter signal paths and are reflected in wireless features such as CSI, RSSI, delay, phase, and multipath structure. Deep learning models can then discover patterns of change in long-term signals that are difficult for people to observe directly.
The communication network therefore becomes more than a transport channel for data; it can also become part of how we sense the natural environment.
Research directions
Communication, sensing, and computing are not three isolated modules, but three mutually reinforcing capabilities. For pre-disaster monitoring, we focus on the following three intersecting directions so that communication, sensing, and computing ultimately form a complete technical pipeline.
Communication + Sensing
We study how wireless networks can sense the environment while maintaining reliable connectivity in mountainous areas. For scenarios with limited power and intermittent signals, we explore low-power communication and coverage solutions using 4G/5G, LoRa, satellites, and UAVs. We then use communication-signal features such as CSI, RSSI, phase, and multipath to detect subtle changes in mountains, ground surfaces, and the surrounding environment—one network for both communication and sensing.
Sensing + Computing
We study how to transform multisource observations—including wireless signals, displacement, inclination, rainfall, and soil moisture—into meaningful disaster-risk information. Deep learning, time-series forecasting, multimodal fusion, and large models extract anomalous features from complex long-term data, identify changes in mountain conditions and evolving risk, and move the system from sensing change to understanding change and predicting risk.
Communication + Computing
We study cooperation between intelligent computing and networks in mountainous areas with weak coverage and tight power budgets. Through cloud-edge-device collaboration and coordination between large and small models, lightweight models run on devices or edge nodes to detect local anomalies even when communications are constrained or briefly interrupted, while the cloud handles long-term analysis and complex reasoning. Computing results also guide communication-resource scheduling so critical monitoring information receives priority under limited bandwidth and energy.
Future directions
In the mountains of the future, communication base stations may do more than connect phones and sensors. Wireless signals continuously cross valleys and slopes, recording subtle environmental changes in their channels. A small number of low-power sensing nodes supplement key geological and meteorological information, edge models continuously assess local anomalies, and cloud models search weeks, months, or longer periods of observation for patterns of mountain change.
As a mountain slowly shifts, signals change first. Prolonged rainfall alters soil conditions and multisource data begins to show anomalies. Before risk develops into disaster, the system has already identified areas requiring closer attention. Monitoring gradually shifts from deploying more devices to giving existing infrastructure more capabilities.
Communication, sensing, and computing likewise cease to be separate systems and instead work together within one network. When a base station can not only connect mountainous regions but also understand what is happening there, natural-hazard monitoring can move beyond traditional device-driven approaches toward low cost, broad coverage, and continuous intelligent sensing.
And all of this can begin with one communication signal, one long-term dataset, and one model.