UCLA

Wireless technologies have been evolving significantly over the past few decades, forming the backbone of many modern applications. Two key paradigms have emerged for different purposes: communication and sensing. Wireless communication systems, such as WiFi and 5G, are optimized to encode as much information as possible into the limited time-frequency resources for high-throughput, low-latency data transmission. Wireless sensing systems, such as radar, focus on extracting information about the environment, such as the location and movement of objects, by analyzing the wireless signals reflected off them. These two paradigms have been evolving separately, developing different hardware and software stacks. Next-generation wireless systems seek to integrate sensing and communication into a unified system to enhance performance and enable new applications. Many challenges exist in achieving this integration. One key challenge is designing unified radio software and hardware that can support both sensing and communication functionalities. Another challenge is to design synergistic mechanisms that create mutual benefits between sensing and communication. The advancement of machine learning and artificial intelligence has opened up new opportunities for addressing these challenges. This dissertation introduces three systems towards this objective. First, we present MilBack, a unified design of a millimeter-wave sensing and communication system based on the backscatter principle. MilBack enables high-accuracy target localization and high-throughput, two-way communication with ultra-low power consumption. Next, we present two systems, NeWRF and mmDiff, that materialize the notion of sensing the wireless environment for improving communication. NeWRF is a deep learning framework that reconstructs the wireless radiation field from sparse channel measurements and predicts the channel state information at unseen locations. It enables operators to identify coverage dead zones and optimize access point placement. mmDiff is a differentiable ray-tracing simulation framework that learns the radio material properties of the environment from sparse measurements. It enables robust mmWave beam prediction in complex 3D environments, which is critical for high-throughput, low-latency communication.