We present the RIS-VSign system, an active reconfigurable intelligent surface (RIS)-assisted multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) framework for vital signs extraction under an integrated sensing and communication (ISAC) model. The system consists of two stages: the phase selector of RIS and the extraction of respiration rate. To mitigate synchronization-induced common phase drifts, the difference of Möbius transformation (DMT) is integrated into the deep learning framework, named DMTNet, to jointly configure multiple active RIS elements. Notably, the training data are generated in simulation without collecting real-world measurements, and the resulting phase selector is validated experimentally. For sensing, multi-antenna measurements are fused by the DC-offset calibration and the DeepMining-MMV processing with CA-CFAR detection and Newton's refinements. Prototype experiments indicate that active RIS deployment improves respiration detectability while simultaneously enabling higher-order modulation; without RIS, respiration detection is unreliable and only lower-order modulation is supported.
Driven by the vision of sixth-generation (6G) wireless networks, ISAC is widely regarded as a cornerstone technology for enabling native environmental perception and context awareness on top of high-throughput connectivity. By unifying sensing and communications, ISAC allows the two functions to share spectrum, waveforms, and infrastructure, thereby meeting requirements for spectral and hardware efficiency, low latency, high reliability, and deployability in realistic environments with blockage and multipath. However, in practical MIMO-OFDM systems, non-contact vital-sign sensing is often fragile due to weak micro-motion returns and synchronization impairments [1,2], e.g., sampling frequency offset (SFO), carrier frequency offset (CFO), and packet detection delay (PDD), which induce common phase drifts.
To be effective in real-world ISAC systems, an RIS with the improvement ability of signal quality is suitable to support fastly because of the low control overhead. Despite D.-M. Chian, C.-K. Wen, and F.-J. Chen are with the Institute of Communications Engineering, National Sun Yat-sen University, Kaohsiung 804, Taiwan (e-mail: icefreeman123@gmail.com, chaokai.wen@mail.nsysu.edu.tw, king19635@gmail.com).
Y.-J. Sun and F.-K. Wang are with the Department of Electrical Engineering, National Sun Yat-sen University, Kaohsiung 804, Taiwan (e-mail: spring1968bear@gmail.com, fkw@mail.ee.nsysu.edu.tw).
The source code for DMTNet will be released soon. rapid progress, practical ISAC deployments face a persistent control bottleneck, particularly when a RIS is introduced as a programmable propagation interface [3]. Optimizationbased phase configuration and pilot-intensive strategies can be computationally expensive and sensitive to phase drift, while purely data-driven solutions are often limited by the scarcity of labeled real-world measurements and potential domain mismatch between simulation and deployment. These challenges motivate a model-informed learning approach that (i) reduces reliance on extensive calibration and data collection, and (ii) maintains sensing robustness without sacrificing communication performance.
The proposed RIS-VSign system consists of two steps: the phase selector of RIS and vital signs extraction. The key contributions of this study are as follows:
• The proposed phase selector of RIS is based on the deep learning, and verified by the experiments. • Without using any real-world data, the training dataset is constructed based on a modified channel model [4] in the simulation. This implies that the proposed ISAC model is suitable for practical deployment. • We evaluate that the deployment of an active RIS in a MIMO system simultaneously enhances the experimental performance of both sensing and communication.
In Fig. 1, we consider a 5G NR-compatible OFDM downlink system, where a Tx equipped with T transmit antennas arXiv:2602.16637v1 [eess.SP] 18 Feb 2026 and K active RIS elements serve an Rx with R receive antennas. We assume that the echo signal from a person is only influenced by the RIS AM path, implying that the static channels do not contain vital signs. For respiration rate estimation, P consecutive packets are required [2].
The static components of the channel include the RIS SM path and the direct Tx-Rx path, while the dynamic components include the RIS AM path and the Doppler-affected reflections. The channel frequency response (CFR) of the static channel, observed at the r-th receive antenna from the t-th transmit antenna, is given by h r,t s . Moreover, the CFR of the dynamic channel from the k-th active RIS element is given by
where a k = g k e -jθ k is the response of the k-th active RIS element with a constant gain g k and a phase shift θ k , h r,k o and h r,k w represent the complex attenuation components without and with vital signs, respectively, l r,k w denotes the chest wall displacement, and λ is the wavelength.
However, when applying the channel models in practical scenarios, an additional challenge arises: the lack of perfect time synchronization between Tx and Rx. This results in timevarying phase offsets [2], including SFO, CFO, and PDD. Since all Rx antennas share the same RF oscillator, they experience the same phase offsets when receiving from a given Tx antenna, making the phase offset independent of the Rx antennas. Then, the CFRs of the static and dynamic channels considering time-varying phase offsets are given by c r,t s = h r,t s e -jψ t and c r,t d,θ k = h r,t d,θ k e -jψ t , respectively, where ψ t represents the time-varying phase offset. Assuming an unit transmission power per antenna, the signal power of the r-th receivr antenna with the t-th transmit antenna and the k-th active RIS element is defined as:
To simplify the optimization problem of E, we consider a single RIS element. For the k-th active RIS element with phase shift θ k , the maximization of (2) in a MIMO system is expressed as:
where p r,t θ k = h r,t s (h r,t d,θ k ) * is the
This content is AI-processed based on open access ArXiv data.