K Foysal Haque

Wi-Fi sensing has gained significant attention forits non-intrusive nature and the widespread avail-ability of Wi-Fi-enabled devices that can be lever-aged for sensing. Given the complexity of wirelesspropagation, sensing applications are increasing-ly relying on machine learning tools, which alloweffectively extracting meaningful features from theradio measurements. However, developing robust,domain-adaptive algorithms that perform wellacross diverse environments remains challenging,requiring extensive datasets with rich diversity interms of environments, subjects, and configurations.In this view, here we present and release two com-prehensive datasets for human activity recognition(HAR) through Wi-Fi consisting of the channel stateinformation (CSI) and beamforming feedback infor-mation (BFI) collected through commercial IEEE802.11ac devices in both line of sight (LoS) andnon-line of sight (NLoS) scenarios in different days.The first dataset entails traces related to 20 activitieswith six subjects across six different setups. The sec-ond dataset focuses on simultaneous multi-subjectactivity classification, with three individuals performing distinct activities concurrently. Our datasets arethe first that also provide corresponding BFI mea-surements — a newly investigated and appealingprimitive for Wi-Fi sensing given its ease of collec-tion procedure. Overall, the datasets contain about240 GB of CSI data and 230 GB for BFI.

In this letter, we introduce M3-channel frequency response (CFR), a bistatic millimeter-wave (mmWave) multiple-input multiple-output (MIMO) CFR dataset for fine-grained microgesture recognition under domain shifts. As wireless systems move toward integrated sensing and communication (ISAC), sensing will increasingly be realized within communication infrastructures rather than via standalone sensing hardware. Yet most existing mmWave sensing datasets are radar-centric and sensing-specific, which misaligns with communication-driven ISAC deployment. In contrast, M3-CFR provides communication-native channel frequency response (CFR) from an 8 × 8 fully digital mmWave MIMO ISAC testbed at 58 GHz with 1 GHz bandwidth. It contains ten microgestures from three subjects across three indoor environments (3000 synchronized CFR instances). The dataset consists of raw and domain-adaptively processed CFR. The latter includes path-loss compensation, frequency normalization, and an entropy-guided subchannel filtering pipeline, enabling cross-environment and cross-subject sensing benchmarking.

High-throughput wireless communications and ac- curate sensing rely on high-resolution channel estimates. Es- timation algorithms should be able to accurately capture the underlying wireless multipath propagation to compensate for the channel in data transmission while leveraging it for acquiring sensing information. However, due to the frequency-dependent superposition of delayed propagation paths, certain components may undergo destructive interference at specific frequencies, obscuring their contribution. This leads to partial reconstruction of the channel, thereby degrading the overall estimation fidelity. To address this challenge, we propose BANDWEAVE, a new multi-band channel frequency response (CFR) fusion framework that significantly enhances channel estimation by leveraging complementary multipath information across multiple frequency bands. BANDWEAVE includes a three-phase progressive learning algorithm enabling adaptive, robust, and efficient fusion in dynamic multi-band wireless environments. We experimentally evaluate BANDWEAVE with an IEEE 802.11ac multi-user mul- tiple input multiple output (MU-MIMO) testbed and a 60 GHz mmWave single-user multiple input multiple output (SU-MIMO) system in three different propagation environments with different fading conditions. Experimental results show that BANDWEAVE channel estimation outperforms single-band estimation achieving more than 16% of throughput improvement and bit error rate (BER) reduction. Also, BANDWEAVE achieves more than 4.9× gain in communication performance compared to state of the art (SOTA) band-merging approaches. Latency measurements on Jetson Nano, Jetson Orin Nano and Raspberry Pi 5 show that BANDWEAVE reduces inference time and energy consumption by up to 17× and 18×, respectively, with respect to SOTA methods.

Emerging applications require mobile devices to continuously execute complex deep neural networks (DNNs). While mobile edge computing (MEC) may reduce the computation burden of mobile devices, it exhibits excessive latency as it relies on encapsulating and decapsulating frames through the network protocol stack. To address this issue, we propose PhyDNNs, an approach where DNNs are modified to operate directly at the physical layer (PHY), thus significantly decreasing latency, energy consumption, and network overhead. Conversely from recent work in Joint Source and Channel Coding (JSCC), PhyDNNs adapt already trained DNNs to work at the PHY. To this end, we developed a novel information-theoretical framework to fine-tune PhyDNNs based on the trade-off between commu-nication efficiency and task performance. We have prototyped PhyDNNs with an experimental testbed using a Jetson Orin Nano as the mobile device and two USRP software-defined radios (SDRs) for wireless communication. We evaluated PhyDNNs performance considering various channel conditions, DNN models, and datasets. We also tested PhyDNNs on the Colosseum net-work emulator considering two different propagation scenarios. Experimental results show that PhyDNNs can reduce the end-to-end inference latency, amount of transmitted data, and power consumption by up to 48×, 1385×, and 13 × while keeping the accuracy within 7 % of the state-of-the-art approaches. Moreover, we show that PhyDNNs experience 4.3 times less latency than the most recent JSCC method while incurring in only 1.79% performance loss. For replicability, we shared the source code for the PhyDNNs implementation.

State-of-the-art covert routing in heterogeneous network (HetNet) focuses on balancing covertness and throughput, but often overlooks explicit energy optimization. While covert communication inherently limits transmit power, meeting throughput demands without coordinated design can still lead to high energy consumption. In contrast, our approach jointly considers covertness, throughput, and energy efficiency, addressing all three objectives simultaneously. To this end, we propose DECOR, a Decentralized Energy-efficient COvert Routing framework that jointly optimizes covertness, throughput, and energy efficiency. Unlike traditional methods that use a single wireless technology, DECOR leverages the diversity of available wireless communication technologies in HetNet to enable simultaneous multi-modal routing. The core idea behind DECOR is that optimal simultaneous utilization of multiple modalities improves throughput and overall energy efficiency. It minimizes the end-to-end energy consumption while satisfying stringent constraints on throughput and covertness through two core steps: (1) link-level optimization using sequential least squares programming (SLSQP), and (2) network-level optimization through a custom cluster-based routing strategy. DECOR introduces a novel clustering-based strategy that aggregates intra-cluster link information and delegates routing decisions to cluster heads, significantly reducing control overhead and enabling scalable, energy-efficient covert communication. Extensive numerical analysis demonstrates that DECOR significantly outperforms existing approaches in terms of energy-efficiency and data overhead.

In this paper, we propose BeamSense, a completely novel approach to implement standard-compliant Wi-Fi sensing applications. Existing work leverages the manual extraction of the uncompressed channel state information (CSI) from Wi-Fi chips, which is not supported by the 802.11 standards and hence requires the usage of specialized equipment. On the contrary, BeamSense leverages the standard-compliant compressed beamforming feedback information (BFI) (beamforming feedback angles (BFAs)) to characterize the propagation environment. Conversely from the uncompressed CSI, the compressed BFAs (i) can be recorded without any firmware modification, and (ii) simultaneously captures the channels between the access point and all the stations, thus providing much better sensitivity. BeamSense features a novel cross-domain few-shot learning (FSL) algorithm for human activity recognition to handle unseen environments and subjects with a few additional data samples. We evaluate BeamSense through an extensive data collection campaign with three subjects performing twenty different activities in three different environments. We show that our BFAs-based approach achieves about 10% more accuracy when compared to CSI-based prior work, while our FSL strategy improves accuracy by up to 30% when compared with state-of-the-art cross-domain algorithms. Additionally, to demonstrate its versatility, we apply BeamSense to another smart home application – gesture recognition – achieving over 98% accuracy across various orientations and subjects. We share the collected datasets and BeamSense implementation code for reproducibility – https://github.com/kfoysalhaque/BeamSense.