Blood pressure (BP) measurement is an indispensable tool in diagnosing and treating many diseases such as cardiovascular failure and stroke. Traditional direct measurement can be invasive, and wearable-based methods may have limitations of discomfort and inconvenience. Contact-free BP measurement has been recently advocated as a promising alternative. In particular, Millimeter-wave (mmWave) sensing has demonstrated its promising potential, however it is confronted with several challenges including noise and vulnerability to human's tiny motions which may occur intentionally and inevitably. In this paper, we propose mmBP+, a contact-freemmWave-basedBPmeasurement system with high accuracy and motion robustness. Due to the high frequency and short wavelength, mmWave signals received in the time domain are dramatically susceptible to ambient noise, and deteriorating signal quality. To reduce noise,we propose a novel approach to exploit mmWave signal's characteristics and features in the delay-Doppler-fractional Fourier domain to significantly improve signal quality for pulse waveform construction. We also propose a periodic signal feature based functional link adaptive filter leveraging on the periodic and correlation characteristics of pulse waveform signals to alleviate the impact of human's tiny motions. Extensive experiment results achieved by the leave-one-out cross-validation (LOOCV) method demonstrate that mmBP+ achieves the mean errors of 0.65mmHg and 1.31mmHg for systolic blood pressure (SBP) and diastolic blood pressure (DBP), respectively; and the standard deviation errors of 3.92mmHg and 3.99mmHg for SBP and DBP, respectively.

The accuracy of traditional localization methods significantly degrades when the direct path between the wireless transmitter and the target is blocked or non-penetrable. This paper proposesN LoS, a novel approach for precise non-line-of-sight (NLoS) localization using a single mmWave radar and a backscatter tag.N LoSleverages multipath reflections from both the tag and surrounding reflectors to accurately estimate the target's position.N LoSintroduces several key innovations. First, we designHFD(Hybrid Frequency-Hopping and Direct Sequence Spread Spectrum) to detect and differentiate reflectors from the target. Second, we enhance signal-to-noise ratio (SNR) by exploiting the correlation properties of the designed signals, improving detection robustness in complex environments. Third, we proposeFS-MUSIC(Frequency-Spatial Multiple Signal Classification), a super-resolution algorithm that extends the traditional MUSIC method by constructing a higher-rank signal matrix, enabling the resolution of additional multipath components. We evaluateN LoSusing a 24 GHz mmWave radar with 250 MHz bandwidth in three diverse environments: a laboratory, an office, and an around-the-corner corridor. Experimental results demonstrate thatN LoSachieves median localization errors of10.69 cm (X)and11.98 cm (Y)at a 5 m range in the laboratory setting, showcasing its effectiveness for real-world NLoS localization.

Deep learning (DL) has become pivotal in advancing integrated sensing and communication (ISAC) systems. However, conventional DL models often require frequent updating or retraining to adapt to dynamic ISAC environments. To address these limitations, this work creatively proposes a large language model (LLM)-based ISAC system, called LLM-ISAC, to enable concurrent sensing-communication processing in a unified framework, with enhanced generalization and environmental robustness. To realize LLM-ISAC, we design a novel signal encoder to transform ISAC signals into LLM-compatible representations through a delay-Doppler-spatial transformer, enabling discriminative cross-domain signal feature extraction for downstream tasks. Moreover, we develop an innovative ISAC-specific context prompt to construct structured machine-readable prompts, dynamically guiding the LLM’s reasoning without retraining and ensuring robust generalization to unseen scenarios. To the best of the authors’ knowledge, this is the first work leveraging the property of LLM in ISAC systems. Extensive simulations demonstrate that LLMI-SAC achieves significant superiority in sensing accuracy, communication reliability, and environmental robustness, compared to state-of-the-art DL-based ISAC methods.

Deep learning (DL) has emerged as a key enabler for unlocking the potential of integrated sensing and communication (ISAC). Despite recent progress, current DL methods primarily handle sensing and communication as independent tasks, overlooking potential performance enhancement through a joint approach. Moreover, existing methods rely on fully annotated data for training, which is often challenging to obtain, especially in multi-task scenarios where labeled data may be scarce or only exist for a subset of tasks. Motivated by these shortcomings, this paper proposes a novel scheme, MTL-DFM, to enable simultaneous sensing and communication with partially labeled training data, which leverages multi-task learning (MTL) and a diffusion model (DFM). In particular, we introduce an initial feature extraction module (IFEM) to jointly capture shared information across tasks and explore inherent cross-task connections for enhanced feature extraction. Next, we design a signal denoising with incomplete labeling (SDIL) module to effectively remove noise from extracted information and construct comprehensive feature representations for all tasks with partially labeled datasets, which is difficult for conventional DL methods. Simulation results verify the superior performance offered by MTL-DFM over prior state-of-the-art methods.

The accuracy of the current localization methods degrades significantly when the direct path between the wireless transmitter and the target is blocked. This paper considers the problem of using a single mmWave radar and a tag to facilitate localization in the non-penetrable non-line-of-sight (NLoS) scenario. We present mN2LoS (short for mmWave based Non-penetrable NLoS LOCalization), which accurately localizes the tag by using the multipath reflections. mN2LoS has a few novel features. First, we design HTRD for detecting reflectors and surroundings while distinguishing them from the tag, using Hybrid utilization of Tag localization code and Reflector localization code based on Direct sequence spread spectrum techniques. Second, we enhance the signal-to-noise ratio by exploiting the correlation features of the designed signal. Evaluation results demonstrate that the developed mN2LoS can achieve median errors (at 5m range) of 12.9cm and 3.8° for distance and AOA estimations for the office configuration, respectively.

Non-line-of-sight (NLoS) sensing represents a significant advancement in sensor technology. Unlike traditional sensing methods that rely on direct line-of-sight, NLoS sensing allows for the detection and localization of objects obscured from the sensor's view. In this paper, we introduce mmMirror, a novel Van Atta Array based millimetre-wave (mmWave) reconfigurable intelligent reflecting surface (IRS) that provides: (i) NLoS localization at a range of approximately 3 meters, (ii) seamless communication between radar and IRS using existing frequency-modulated continuous-wave (FMCW) signals, and (iii) support for multiple targets. The mmMirror system is implemented on commodity 24 GHz radars, and the IRS is prototyped on printed circuit boards (PCBs).

The recently emerged multiple-input multiple-output over orthogonal delay-Doppler division multiplexing (MIMO-ODDM) is gaining paramount interest as a promising solution to provide reliable communication performance for high-mobility systems. To achieve its full potential, signal detection becomes a critical issue, while the performance of existing methods is yet to be satisfactory. In this paper, we develop a novel signal detection approach for MIMO-ODDM systems by leveraging the spatial-based generative adversarial network, namely SG-ODDM, for accurate, interference-resilient and environment-robust performance. We creatively design a spatial-based generative adversarial network (spatial-based GAN) for comprehensive feature extraction and interference mitigation. We propose a transfer learning-based adaptive updating (TAU) to enhance environmental robustness by reducing the frequency and effort in updating the detection model. Extensive simulation results verify that the proposed SG-ODDM is considerably superior to state-of-the-art related works, in terms of detection accuracy, interference resilience and updating effort reduction.

The multiple-input multiple-output over orthogonal delay-Doppler division multiplexing (MIMO-ODDM) has recently attracted great interest as a promising solution for high-mobility systems. To achieve its full potential, signal detection becomes a critical issue, while the performance of the existing methods is yet to be satisfactory. To address this issue, we propose a novel signal detection approach called SG-ODDM, which utilizes a spatial-based generative adversarial network (spatial-based GAN) for accurate and interference-resistant performance. We creatively design a spatial-based GAN for comprehensive feature extraction and interference mitigation. In the spatial-based GAN, we develop an attention-based generator with multi-domain feature (AGMF) to effectively reconstruct signals for detection by extracting and utilising signal characteristics across multiple domains, e.g., delay, Doppler, and spatial domains. Moreover, we develop a self-attention-based discriminator with multi-domain feature (SDMF) to guide AGMF to mitigate the impact of interference in MIMO systems, thereby improving the quality of the generated/reconstructed data from AGMF. Additionally, we design a novel hybrid loss function to fully exploit signal features in the multiple domains for detection. Through extensive simulations, we demonstrate that SG-ODDM outperforms state-of-the-art related works regarding detection accuracy and interference resilience.

Orthogonal delay-Doppler division multiplexing (ODDM) modulation was recently proposed as a promising solution for high-mobility communication systems. To achieve the potential of ODDM, reliable signal detection is essential, hence, in this work, we propose a contrastive learning-based signal detection approach for ODDM systems, named CL-ODDM. Unlike the conventional deep learning-based methods which focus on positive samples alone, we creatively leverage contrastive learning to exploit both positive and negative samples in the training dataset. By doing so, more distinguishable information of signals can be captured and extracted, contributing to reliable detection results. Moreover, we employ a convolutional neural network and recurrent encoder-decoder (CREN) to represent the comprehensive properties and features of ODDM signals. In addition, an adaptive correction method (ACM) is proposed to increase the convergence rate and improve the stability of the detection model. Extensive simulation results validate that the proposed CL-ODDM is significantly superior state-of-the-art related work, regarding the detection accuracy and convergence rate.

Facial expression recognition plays a vital role to enable emotional awareness in multimedia Internet of Things applications. Traditional camera or wearable sensor based approaches may compromise user privacy or cause discomfort. Recent device-free approaches open a promising direction by exploring Wi-Fi or ultrasound signals reflected from facial muscle movements, but limitations exist such as poor performance in presence of body motions and not being able to detect multiple targets. To bridge the gap, we propose mmFER, a novel millimeter wave (mmWave) radar based system that extracts facial muscle movements associated with mmWave signals to recognize facial expressions. We propose a novel dual-locating approach based on MIMO that explores spatial information from raw mmWave signals for face localization in space, eliminating ambient noise. In addition, collecting mmWave training data can be very costly in practice, and insufficient training dataset may lead to low accuracy. To overcome, we design a cross-domain transfer pipeline to enable effective and safe model knowledge transformation from image to mmWave. Extensive evaluations demonstrate that mmFER achieves an accuracy of 80.57% on average within a detection range between 0.3m and 2.5m, and it is robust to various real-world settings.

Deep learning (DL) demonstrates tremendous potential in high-mobility communication systems, especially from the perspective of signal detection. However, most existing DL-based detection methods are data/environment specific and the re-training process is resource intensive. To address these shortcomings, we propose a contrastive learning-based environmentally robust signal detection approach in orthogonal delay-Doppler division multiplexing (CL-ODDM) to achieve fast convergence, high accuracy and strong robustness to variations in wireless environments. Specifically, unlike conventional methods which explore only positive samples in the dataset for detection, in this work, we propose to leverage contrastive learning to fully exploit both positive and negative samples in the training dataset. This enables us to extract more comprehensive features of signals, which can accelerate convergence, improve detection accuracy, and enhance the generalized ability of our CL-ODDM. Moreover, we creatively employ a convolutional neural network and recurrent encoder-decoder (CREN) to represent the underlying properties of ODDM signals and extract high-quality features. To further improve environmental robustness, we propose novel training strategies, i.e., data augmentation (DA) and adaptive updating scheme (AUS). The proposed DA method is expected to increase the diversity of the dataset and represent more effective signal features. The designed AUS leverages transfer learning to adapt partial layers of CREN to make our CL-ODDM suitable for various wireless environments. Numerous simulation results validate that the proposed CL-ODDM significantly outperforms state-of-the-art related works, in terms of detection accuracy, environmental robustness and convergence rate.

Deep learning (DL)-based signal detection techniques have demonstrated significantly superior performance than the conventional methods in orthogonal time frequency space (OTFS) systems. Despite the effectiveness, existing methods using DL techniques are environment-specific. For instance, a detection model trained in one environment may become ineffective if environmental changes occur, e.g., user scheduling, inter-user interference and network scheduling. A re-training process is required to refine the model using numerous samples from the new/unseen environment, which is not always accessible in practice. To address the above concern, in this work, we propose an environment-robust approach to detect OTFS signals, by leveraging the property of the matching network (MatNet), referred as to OTFS-MatNet. Specifically, we propose to employ two functional blocks of MatNet to automatically capture generalized features shared among seen environments and the potential new environment. We also develop a novel loss function and a two-step training strategy to improve the generalized ability and detection accuracy. Therefore, the proposed OTFS-MatNet can realize accurate detection with a limited number of training samples, i.e., one sample from the new environment and the dataset from one seen environment. Numerous simulation results demonstrate that the developed OTFS-MatNet is significantly superior to state-of-the-art OTFS detection methods, in terms of improving detection accuracy and reducing the required number of training samples.

Blood pressure (BP) measurement is an indispensable tool in diagnosing and treating many diseases such as cardiovascular failure and stroke. Traditional direct measurement can be invasive, and wearable-based methods may have limitations of discomfort and inconvenience. Contact-free BP measurement has been recently advocated as a promising alternative. In particular, Millimetre-wave (mmWave) sensing has demonstrated its promising potential, however it is confronted with several challenges including noise and vulnerability to human's tiny motions which may occur intentionally and inevitably. In this paper, we propose mmBP, a contact-free mmWave-based BP measurement system with high accuracy and motion robustness. Due to the high frequency and short wavelength, mmWave signals received in the time domain are dramatically susceptible to ambient noise, and deteriorating signal quality. To reduce noise, we propose a novel delay-Doppler domain feature transformation method to exploit mmWave signal's characteristics and features in the delay-Doppler domain to significantly improve signal quality for pulse waveform construction. We also propose a temporal referential functional link adaptive filter leveraging on the periodic and correlation characteristics of pulse waveform signals to alleviate the impact of human's tiny motions. Extensive experiment results achieved by the leave-one-out cross-validation (LOOCV) method demonstrate that mmBP achieves the mean errors of 0.87mmHg and 1.55mmHg for systolic blood pressure (SBP) and diastolic blood pressure (DBP), respectively; and the standard deviation errors of 5.01mmHg and 5.27mmHg for SBP and DBP, respectively.

Deep learning has demonstrated its great potential in channel state information (CSI)-based human activity recognition (HAR), and hence has attracted increasing attention in both the industry and academic communities. While promising, most existing high-accuracy methodologies require to retrain their models when applying the previous-trained ones to a new/unseen environment. This issue has limited their practical usabilities. In order to overcome this challenge, this article proposes an innovative scheme, which combines an activity-related feature extraction and enhancement (AFEE) method and matching network (AFEE-MatNet). The proposed scheme is “one-fits-all,” meaning that the trained model can be directly applied in new/unseen environments without any retraining. We introduce the AFEE method to enhance CSI quality by eliminating noise. Specifically, the approach mitigates environmental noises unrelated to activity while better compressing and preserving the behavior-related information. Moreover, the size of feature signals generated by AFEE are reduced, which in turn significantly shortens the training time. For effective feature extraction, we propose to use the MatNet architecture to learn transferable features shared among source environments. To further improve the recognition performance, we introduce a prediction checking and correction scheme to rectify some classification errors that do not abide by the state transition of human behaviors. Extensive experimental results demonstrate that our proposed AFEE-MatNet significantly outperforms existing state-of-the-art HAR methods, in terms of both recognition accuracy and training time.

Deep Learning plays an increasingly important role in device-free WiFi Sensing for human activity recognition (HAR). Despite its strong potential, significant challenges exist and are associated with the fact that one may require a large amount of samples for training, and the trained network cannot be easily adapted to a new environment. To address these challenges, we develop a novel scheme using matching network with enhanced channel state information (MatNet-eCSI) to facilitate one-shot learning HAR. We propose a CSI correlation feature extraction (CCFE) method to improve and condense the activity-related information in input signals. It can also significantly reduce the computational complexity by decreasing the dimensions of input signals. We also propose novel training strategy which effectively utilizes the data set from the previously seen environments (PSE). In the least, the strategy can effectively realize human activity recognition using only one sample for each activity from the testing environment and the data set from one PSE. Numerous experiments are conducted and the results demonstrate that our proposed scheme significantly outperforms state-of-the-art HAR methods, achieving higher recognition accuracy and less training time.

Deep learning (DL) has been envisioned as a plausible solution to spectrum sensing, demonstrating an influential role in dynamic spectrum access. Despite their effectiveness, existing DL based sensing methods are heavily environment-sensitive. In other words, the sensing model trained in one environment usually cannot be applied to another, and a large number of labeled samples from the new environment are required to re-train DL architectures. To address the above challenge, we propose a novel approach leveraging the matching network (MN) for environment-robust spectrum sensing (MN-ERSS). Specifically, to improve the quality of input signals of MN, we propose to use a cross-correlation feature of the cyclic prefix (CP) of orthogonal frequency division multiplexing (OFDM) signals as the input data. Then, we propose to employ an advanced technique of one-shot learning, i.e., MN, to automatically extract inherent features from input signals. Moreover, we propose a tailored training strategy to better utilize the data set from the previous environment. The proposed training strategy can accomplish a successful spectrum sensing with the data set from only one previous environment and one sample from the new/testing environment. To the best of our knowledge, this is the first to investigate the environment-robust spectrum sensing by exploring one-shot learning. Extensive simulation results demonstrate that the proposed MN-ERSS significantly outperforms state-of-the-art sensing approaches, i.e., achieving a higher sensing accuracy with only one sample from the testing environment and the data set from one previous environment.

Deep learning has shown a strong potential in device-free human activity recognition (HAR). However, a fundamental challenge is ensuring accuracy, without re-training, when exposing a previously trained architecture to a new or unseen environment. To overcome the aforementioned challenge, this paper proposes an environment-robust channel state information (CSI) based HAR by leveraging the properties of a matching network (MatNet) and enhanced features (HAR-MN-EF). To improve the CSI quality, we propose a CSI cleaning and enhancement method (CSI-CE) that includes two key stages: activity-related information extraction (ARIE) and correlation feature extraction based on principal component analysis (CFE-PCA). The ARIE stage is able to effectively enhance the activity-dependent features whilst mitigating behavior-unrelated information. The CFE-PCA stage further improves the extracted features by filtering out the residual activity-unrelated data and the residual noise contained in signals from the former stage. The extracted features are then sequenced into the MatNet to create an environment-robust HAR. Experimental results confirm that an architecture trained by the proposed HAR-MN-EF can be directly adapted to a new environment, achieving reliable sensing accuracies without requiring additional effort.

Full duplex (FD) wireless technology, which enables simultaneous transmission and reception on the same frequency, has shown its great potential for doubling the spectral efficiency as well as spectrum sensing while transmitting in cognitive radio networks (CRNs). However, the self interference (SI) suppression, the underlying technique of FD, is often imperfect, resulting in non-negligible residual SI that severely affects the test statistics of sensing methods. The residual SI thereby significantly deteriorates the spectrum sensing accuracy. In this work, we aim to address this issue by proposing a novel sensing approach in FD systems leveraging the Pilot-Tone (PT) structure of Orthogonal Frequency Division Modulation (OFDM) signals. In comparison with the conventional sensing methods in FD systems, the developed sensing approach holds the advantage in the robustness not only to residual SI but also the carrier frequency offset (CFO). Besides, the proposed sensing method is able to accomplish sensing tasks in low SNR conditions with much lower computational complexity. Numerical simulations results demonstrate that the probability of detection of our proposed approach can be improved up to 34.9%, compared with state- of-the-art sensing methods in FD systems, suffering from residual SI and CFO.

Device free WiFi Sensing using channel state information (CSI) has been shown great potentials for human activity recognition (HAR). However, extracting reliable and concise feature signals remains as a challenging problem, especially in a dynamic and complex environment. In this paper, we propose a novel scheme for CSI-based HAR using deep learning network (CH-DLN), with an innovative CSI correlation feature extraction (CCFE) method. The CCFE method pre-processes the signals input to the DLN in two steps. Firstly, it uses a recursive algorithm to reduce non-activity-related information from the signal and hence enhance the activity-dependent signals. Secondly, it computes the correlation over both the time and frequency domain to disclose better signal structure and compress the signal. From such enhanced and compressed signals, we utilize the recurrent neural networking (RNN) to automatically extract deeper features, and then apply the softmax regression algorithm for classifying activities. Through extensive experimental results, our proposed scheme is shown to outperform state-of-the-art methods in recognition accuracy, with much less training time.

Spectrum sensing plays a critical role in dynamic spectrum sharing, a promising technology to address the radio spectrum shortage. In particular, sensing of orthogonal frequency division multiplexing (OFDM) signals, a widely accepted multi-carrier transmission paradigm, has received paramount interest. Despite various efforts, noise uncertainty, timing delay and carrier frequency offset (CFO) still remain as challenging problems, significantly degrading the sensing performance. In this work, we develop two novel OFDM sensing frameworks utilizing the properties of deep learning networks. Specifically, we first propose a stacked autoencoder based spectrum sensing method (SAE-SS), in which a stacked autoencoder network is designed to extract the hidden features of OFDM signals for classifying the user’s activities. Compared to the conventional OFDM sensing methods, SAE-SS is significantly superior in the robustness to noise uncertainty, timing delay, and CFO. Moreover, SAE-SS requires neither any prior information of signals (e.g., signal structure, pilot tones, cyclic prefix) nor explicit feature extraction algorithms which however are essential for the conventional OFDM sensing methods. To further improve the sensing accuracy of SAE-SS, especially under low SNR conditions, we propose a stacked autoencoder based spectrum sensing method using time-frequency domain signals (SAE-TF). SAE-TF achieves higher sensing accuracy than SAE-SS using the features extracted from both time and frequency domains, at the cost of higher computational complexity. Through extensive simulation results, both SAE-SS and SAE-TF are shown to achieve notably higher sensing accuracy than that of state of the art approaches.

Full duplex (FD) capability enables a "listen and talk" protocol for spectrum sensing that has been used as a new paradigm to increase the spectrum utilization in cognitive radio networks (CRNs). However, the spectrum sensing performance suffers from the imperfect self-interference suppression (SIS). This could significantly degrade the performance of FD systems in CRNs. In this paper, we investigate the issue of spectrum sensing with imperfect SIS in FD systems. By drawing support from a cyclic prefix (CP) of Orthogonal Frequency Division Modulation (OFDM) signals, we propose a novel spectrum sensing mechanism that is robust to self- interference. Comparing with other conventional sensing approaches in FD systems, the proposed method is independent of timing delay. That significantly improves the sensing performance, even without requiring a complex process for timing delay estimation. As a result, it also reduces the overhead of spectrum sensing. Extensive simulation results indicate that even with serious self-interference and timing delay, the presented approach is still able to achieve much higher performance than the conventional energy detection and waveform-based detection approaches.