Wireless Surfaces Monitor Breathing Without Contact

Researchers are developing innovative methods for non-contact vital signs monitoring, and a new study by De-Ming Chian, Chao-Kai Wen, and Feng-Ji Chen from the Institute of Communications Engineering, National Sun Yat-sen University, working with colleagues Yi-Jie Sun and Fu-Kang Wang from the Department of Electrical Engineering, National Sun Yat-sen University, details a promising active reconfigurable intelligent surface (RIS)-assisted multiple-input multiple-output system for integrated sensing and communication. Their work introduces the RIS-VSign system, which extracts vital signs like respiration rate using a novel deep learning framework called DMTNet to configure active RIS elements and mitigate synchronization issues. Significantly, the team validated their simulated training data with prototype measurements, demonstrating that active RIS deployment not only enhances respiration detection accuracy but also supports higher-order modulation schemes, representing a substantial advancement over existing methods that struggle with reliable detection and limit communication capabilities.

For years, combining wireless signals with health monitoring has been a complex challenge, often limiting both communication speed and accurate readings. Now, a new system utilises actively controlled surfaces to simultaneously enhance signal transmission and detect breathing rate. Prototype measurements demonstrate improved respiration detection alongside support for faster data transfer.

Scientists are developing a system integrating sensing and communication technologies to extract vital signs, specifically respiration rate, using reconfigurable intelligent surfaces (RIS). This approach addresses a key challenge in wireless systems, reliably detecting subtle body movements for health monitoring while simultaneously maintaining high data transmission rates.

Current methods often struggle with weak signals and synchronization errors, limiting both sensing accuracy and communication bandwidth. The RIS-VSign system employs an active RIS, a programmable surface that intelligently controls radio waves, to enhance signal quality and enable more dependable vital sign detection. Unlike previous attempts relying on extensive real-world data for training, this system generates its training data through simulation, offering a practical advantage for deployment in diverse environments.

Achieving this integrated functionality requires overcoming issues related to signal synchronization. Common phase drifts, caused by factors like frequency offsets and timing delays, can corrupt the weak signals reflecting off the body. To counter this, researchers incorporated a difference of Möbius transformation (DMT) into a deep learning framework, termed DMTNet, to precisely configure the active RIS elements.

This configuration aims to improve signal clarity and ensure accurate respiration rate estimation. Prototype experiments demonstrate that deploying an active RIS not only improves the reliability of respiration detection but also allows for the use of higher-order modulation schemes, increasing data throughput. The core of the system lies in a two-stage process: first, selecting the optimal phase configuration for the RIS, and second, extracting the vital signs from the received signals.

For sensing, multi-antenna measurements are processed using DC-offset calibration and a technique called DeepMining-MMV, coupled with constant false alarm rate (CA-CFAR) detection and Newton’s refinements. By fusing these techniques, the system aims to isolate the subtle movements caused by breathing from background noise. At the heart of this work is the ambition to create a system that doesn’t just communicate data, but actively perceives the surrounding environment and the well-being of individuals within it.

Researchers found that the simulated training data effectively guided the phase selector, validating the system’s potential for real-world application. For instance, the channel response between the transmitter, RIS, and receiver is modelled to account for both static and active components, including the subtle chest wall displacement caused by respiration.

The model incorporates the effects of time-varying phase offsets, which are inherent in practical wireless systems. Instead of requiring extensive calibration, the DMTNet learns to compensate for these offsets, ensuring strong sensing performance. The system’s ability to simultaneously improve sensing and communication is a key advancement. By maximising a reward function that considers both signal strength and phase alignment, the RIS configuration optimizes performance for both functions.

Unlike traditional approaches that treat sensing and communication as separate tasks, this integrated approach allows for efficient use of spectrum and hardware resources. Since the system is designed to operate within the existing 5G NR framework, it offers a pathway towards smooth integration with future 6G wireless networks. Under these conditions, the system promises to unlock new possibilities for remote health monitoring, personalized healthcare, and proactive wellness management.

The researchers emphasize the importance of a model-informed learning approach, combining simulation with experimental validation to reduce the need for large amounts of labelled real-world data. This approach also helps to mitigate the potential for domain mismatch between simulation and deployment, ensuring that the system performs reliably in real-world conditions.

By carefully modelling the channel characteristics and incorporating techniques to address synchronization errors, the researchers have created a system that is both accurate and practical. Beyond respiration rate, the framework could be extended to monitor other vital signs, such as heart rate and body temperature, opening up new avenues for non-contact health monitoring.

Active RIS control enhances respiration sensing and signal power

Initial experiments with the RIS-VSign system demonstrate that active RIS deployment markedly improves respiration detectability alongside the enablement of higher-order modulation schemes. Without active RIS elements, reliable respiration detection proved unattainable, restricting operation to lower-order modulation only. Prototype results reveal that the system achieves performance gains through the precise control of active RIS elements, configuring them to optimise signal reception.

Specifically, the research focuses on maximising the reward function Q(θk) for each RIS element, where θk represents the phase shift applied. Calculations show that maximising this function is equivalent to enhancing the real component of the complex inner product, pr,t θk, which captures both communication and sensing channels. The work establishes a theoretical link between the phase shift and the received signal power, Er,t θk, demonstrating how RIS elements can simultaneously improve both communication and sensing performance.

To address time-varying phase offsets, specifically, carrier frequency offset, symbol timing offset, and phase drift, the research introduces the Difference of Möbius Transformation (DMT) parameter, δr,t θk. This parameter is defined as mr,t θk −mr,t s, where mr,t θk represents the Möbius transformation of the total channel with phase shift θk and mr,t s is the transformation of the static channel.

Theorem 1, central to the work, establishes a relationship allowing prediction of the complex inner product pr,t θ⋆ k for any phase shift θ⋆ k, given knowledge of δr,t θk and mr,t s. Also, the developed DMTNet architecture, a deep learning model, processes inputs of dimensions 4x4x32 for phase and 4x4x4 for magnitude, in the end outputting a phase selection for each of the two active RIS elements.

The network incorporates upsampling, downsampling, convolutional layers, and fully connected layers to learn the optimal phase configuration. The use of this network allows for simultaneous selection of phases for all RIS elements, overcoming the challenges of interference between them.

Simulated MIMO channel modelling and active RIS element configuration via DMTNet

A 4×4 active RIS array underpins the research, designed to simultaneously support sensing and communication within an integrated framework. Initially, a detailed model of the RIS was modified to represent a 4R4T MIMO system, incorporating both static and active components. Two feeding ports within the antenna pair function as transmit and receive points, connected to other antenna pairs to improve isolation across the RIS array.

The antenna pair operates effectively within a 3.4-3.8GHz frequency range, exhibiting reflection and transmission coefficients below -10 dB and -20 dB, respectively, as measured in the operating band. To verify the predictor, the 4R4T MIMO system transmitted 5G OFDM signals in the 3.7-3.72GHz band. All RIS elements were initially set to the same phase shift state, with the receiver positioned directly in front of the RIS.

By first disabling all relays to capture the Möbius transformation of the static channel, and then activating them to obtain the transformation for all phase shift states, the predictor’s accuracy was assessed. Averaging continuous packets for each state allowed for the achievement of stable and reliable channel information, as demonstrated through analysis of the channel’s transformation across different states.

Reconfigurable surfaces enable simultaneous respiration detection and data transmission

Once considered a distant prospect, the convergence of wireless communication and physiological sensing is rapidly becoming a practical reality. This work details a system, RIS-VSign, that uses reconfigurable intelligent surfaces to not only transmit data but also to detect human respiration, achieving both goals simultaneously. For years, extracting subtle physiological signals from radio waves has been hampered by the need for strong, consistent signals and the difficulty of separating these signals from noise.

Previous attempts often relied on dedicated sensing equipment or struggled to maintain reliable communication alongside sensing functions. The ability to generate training data entirely through simulation, rather than requiring extensive real-world data collection, represents a considerable step forward. By bypassing the need for labelled datasets gathered from individuals, the researchers have lowered a significant barrier to deployment and adaptation.

The demonstrated increase in data throughput, reaching 388 Mbps with active RIS deployment, is not merely an incremental improvement. It suggests a pathway towards higher bandwidth applications that can coexist with continuous health monitoring. Beyond respiration, this framework could, in principle, be extended to detect other vital signs, such as heart rate or even subtle movements, opening possibilities for remote patient monitoring and preventative healthcare.

While the system successfully detects respiration, its performance in cluttered or active environments remains unproven. Further research should focus on refining the algorithms to better filter out interference and adapt to changing conditions. Exploring the potential of integrating this technology with existing wireless infrastructure, such as 5G networks, could unlock even greater opportunities for widespread adoption and real-world impact.