Enhancing WiFi CSI Fingerprinting: A Deep Auxiliary Learning Approach

We propose SignFi to recognize sign language gestures using WiFi. SignFi uses Channel State Information (CSI) measured by WiFi packets as the input and a Convolutional Neural Network (CNN) as the classification algorithm. Existing WiFi-based sign gesture recognition technologies are tested on no more than 25 gestures that only involve hand and/or finger gestures. SignFi is able to recognize 276 sign gestures, which involve the head, arm, hand, and finger gestures, with high accuracy. SignFi collects CSI measurements to capture wireless signal characteristics of sign gestures. Raw CSI measurements are pre-processed to remove noises and recover CSI changes over sub-carriers and sampling time. Pre-processed CSI measurements are fed to a 9-layer CNN for sign gesture classification. We collect CSI traces and evaluate SignFi in the lab and home environments. There are 8,280 gesture instances, 5,520 from the lab and 2,760 from the home, for 276 sign gestures in total. For 5-fold cross validation using CSI traces of one user, the average recognition accuracy of SignFi is 98.01%, 98.91%, and 94.81% for the lab, home, and lab+home environment, respectively. We also run tests using CSI traces from 5 different users in the lab environment. The average recognition accuracy of SignFi is 86.66% for 7,500 instances of 150 sign gestures performed by 5 different users.

With the development of the internet of things, localization and activity recognition based on WIFI signals have received much attentions. In order to improve the performance, in this paper, a convolutional neural network (CNN) based localization and activity recognition algorithm using channel state information (CSI) measurements and decision fusion is proposed. In the off-line phase, the original CSI measurements obtained from multiple receivers are preprocessed to remove the outliers and noise by Hampel filter and Gaussian filter. Then the normalized CSI measurements are rendered into RGB images. At last, at each receiver, the CNN and distributed training is used for classification learning of localization and activity recognition, respectively. In the on-line phase, the decision-level fusion is used to fuse the intermediate estimation and obtain the final localization and activity recognition results. Experiment results show the better performance of the proposed algorithm.

Radio frequency (RF) fingerprint representing the inherent hardware characteristics of mobile devices has been employed to classify and identify wireless devices for the security of Internet of Things (IoT). Existing works on RF fingerprinting are usually based on the amplitude or phase of RF signal envelope, which leads to relatively coarse features. Moreover, the classification performance over small sample dataset is poor. To solve the problem, a novel device identification method based on RF fingerprinting with on deep learning is proposed. In particular, the RF signal are transformed into two dimensional representations by image preprocessing. Then the gray images representing the RF fingerprints are classified by employing classical CNN. To verify the performance of the proposed approach, a testbed is constructed by using MATLAB build framework of gray image preprocessing. Extensive experiment results show that the identification accuracy can reach at least 90%. Even with the sample rate of 20Gsps. Particularly, the accuracy of iPhone can reach 100%. It is verified that the proposed method can effectively classify mobile devices even with small sample RF fingerprints represented two dimensional gray images,

A transmitted packet that fails to reach its intended destination may be correctly received by neighbor nodes due to the broadcast nature of the wireless medium. In a cooperative retransmission scheme, these neighbor nodes, known as relays, can retransmit the failed packet on behalf of the original source node. The challenge is that multiple concurrent transmissions may lead to collision at the destination, and thus the problem is to decide which relay should help in retransmitting the failed packet so that the destination can successfully receive it. This paper proposes a decentralized partially observable Markov decision process (DEC-POMDP) model for selecting the relays to perform the cooperative retransmission. The proposed DEC-POMDP model does not require global channel state information (CSI). In addition, it is robust to noise in CSI measurements. Furthermore, the proposed DEC-POMDP scheme utilizes the gradient descent learning method to eliminate the need for a wireless channel model. We show that the proposed learning method based on the DEC-POMDP model can perform near optimally in the absence of a channel model and despite its implementation simplicity.

Location-Based Service (LBS) has been widely deployed. One of the key components of LBS is the positioning algorithm. For outdoor environments, the Global Positioning System (GPS) has been used as the default positioning scheme. However, GPS requires the line of sight to the satellites. When the line of sight is blocked, GPS simply stops working. To tackle the problem with GPS, varied WiFi-based positioning schemes have been proposed. However, the positioning precision of the existing methods is not satisfactory. In this paper, we present a high-precision positioning scheme named Deep Learning based Positioning (DLP). Technically, DLP utilizes both Received Signal Strength Indicator (RSSI) and Channel State Information (CSI) to improve the positioning precision. In detail, a deep neural network is used to model the received RSSI and CSI measurements, which leads to satisfactory positioning accuracy. Our experimental results acquired from a large-scale testbed indicate that DLP outperforms the existing positioning schemes in terms of positioning precision.

Recently, network security has become a major challenge in communication networks. Most wireless networks are exposed to some penetrative attacks such as signal interception, spoofing, and stray. Radio frequency (RF) fingerprinting is considered to be a promising solution for network security problems and has been applied with various improvements. In this paper, extensive data from Bluetooth (BT) devices are utilized in RF fingerprinting implementation. Hilbert-Huang transform (HHT) has been used, for the first time, for RF fingerprinting of Bluetooth (BT) device identification. In this way, time-frequency-energy distributions (TFED) are utilized. By means of the signals' energy envelopes, the transient signals are detected with some improvements. Thirteen features are extracted from the signals' transients along with their TFEDs. The extracted features are pre-processed to evaluate their usability. The implementation of three different classifiers to the extracted features is provided for the first time in this paper. A comparative analysis based on the receiver operating characteristics (ROC) curves, the associated areas under curves (AUC), and confusion matrix are obtained to visualize the performance of the applied classifiers. In doing this, different levels of signal to noise ratio (SNR) levels are used to evaluate the robustness of the extracted features and the classifier performances. The classification performance demonstrates the feasibility of the method. The results of this paper may help readers assess the usability of RF fingerprinting for BT signals at the physical layer security of wireless networks.

Wireless infrastructure protection using low-cost radio frequency fingerprinting receivers

In recent years, radio frequency (RF) fingerprinting has attracted more and more attention. Many different types of RF fingerprints have been proposed, such as carrier frequency offset (CFO), sampling frequency offset and error vector magnitude. Among them, the CFO fingerprint is recognized as a promising RF fingerprint. However, for commonly used smartphones, we find that its CFO fingerprint is unstable, because the temperature of crystal oscillator varies greatly and large fluctuations of temperature significantly affect its CFO fingerprint. Therefore, the solutions of CFO-based fingerprinting will no longer be effective for smartphones if the temperature of crystal oscillator is not involved. To this end, we propose a more reliable and applicable CFO-based fingerprinting approach called temperature-aware radio frequency fingerprinting (TeRFF). First, we construct a dataset by extracting crystal oscillator's temperature and the corresponding CFO value on multiple smartphones over a period. In the dataset, the extracted temperature values constitute a set of temperature values, and each registered temperature value corresponds to a group of CFO samples. On this basis, we train multiple Naive Bayes models, each tagged with a registered temperature value. Moreover, since there are many temperature values which are not in the temperature set, we design a CFO estimation method to estimate the CFO fingerprint at the unregistered temperature. Finally, the experimental results demonstrate that our proposed solution TeRFF makes the CFO fingerprinting still effective for smartphone identification, and its performance is better than other existing RF fingerprinting schemes.

In mobile communications networks, if we can estimate the location of each user equipment (UE) with high accuracy, efficient cell planning and network optimization become possible. However, it is difficult for operators to estimate the location of most commercial UEs because they cannot feedback their location information directly to their serving base stations with the exception of UEs with Minimization of Drive Test (MDT) functions specified in LTE Release 10. Radio Frequency (RF) fingerprint method is known as an effective localization method, with which we can estimate the location of UEs with only RF signature and location information beforehand. In LTE, RF fingerprint database can be collected by conducting drive tests or using measurement data from UEs compatible with MDT. Although the estimation accuracy can be better with more RF signatures, the increase in the amount of RF signatures causes the increase in the feedback, which consumes the uplink capacity and UE battery. Furthermore, keeping all RF signatures in database is inefficient because some RF signatures do not have effect on improving estimation accuracy. Hence, it is important to clarify the effect of each RF signature and use only effective ones in the localization. In this paper, we conducted a field experiment to create an RF fingerprint database in dense urban area, and evaluated the effectiveness of each RF signature by making several localization models based on machine learning. Eventually, we clarified minimum RF signatures required to operate RF fingerprint localization in LTE network.

Aiming at the actual application scenario of radio frequency (RF) fingerprint identification, a low-cost, small-volume embedded system for specific identification of RF devices is designed and implemented, and the improved VGG-16 model is used to extract the RF fingerprint features of RF signals. The hardware of the system consists of two parts: the RF Signal Acquisition Module and the RF Fingerprint Identification Module. The RF Signal Acquisition Module uses Zedboard as the main control board, and the AD9361 chip is the RF signal receiving circuit; the RF Fingerprint Identification Module uses FPGA to accelerate the deep learning algorithm, and displays the identification results on the host computer. Using 6 wireless routers of the same model for testing, the system has an identification accuracy rate of 96% for wireless routers in the actual electromagnetic environment, and the real-time identification feedback time is about 6 seconds, which has a certain practical application value.

Fingerprint-based indoor positioning systems (IPSs) are being explored to aid in location-based services due to their robustness in nonline-of-sight (NLOS) conditions. Current systems use high-dimensional radio frequency (HDRF) fingerprints, such as Wi-Fi channel state information (CSI), to achieve higher positioning precision. Since data acquisition is labor-intensive, researchers proposed to enrich the dataset with generative models. It, however, faced challenges arising from capturing the intricate HDRF distribution using simplistic models and the lack of a framework that simultaneously addresses the generative model training, sample evaluation, and selection. In order to synthesize high-quality HDRF fingerprints, this article proposes an HDRF fingerprint generation framework using a conditional diffusion model (CDM) that learns the packet-level feature distribution by decomposing HDRF fingerprints using grid points, anchors, and frequency channel information while preserving the feature spatial correlation within a fingerprint. A sample selection process using the Mahalanobis distance and the principal component analysis (PCA) Q-statistic is used to ensure the sample fidelity. An adaptive learning strategy is further developed to integrate the generated synthetic HDRF fingerprints into downstream positioning tasks. Experimental results on two HDRF datasets quantitatively and qualitatively showcase the diversity and fidelity of the synthetic samples. Compared to solely using the original dataset, integrating the synthetic HDRF fingerprints from the developed framework to train downstream positioning models can, furthermore, decrease the positioning error by up to 16%.

This paper studies the downlink of a cloud radio access network (C-RAN) that incorporates a baseband central processor (CP), multiple remote radio units (RUs), and a network of wireless fronthaul links that connect the RUs to the CP. The fronthaul network utilizes dedicated point-to-point free-space optical (FSO) links along with a broadcast radio frequency (RF) channel. The spectrum of the RF channel is also utilized for downlink transmission from the RUs to the mobile users. That is, the available RF spectrum is time-shared (in a half-duplex manner) among the fronthaul and downlink. The data symbols intended for different users are linearly precoded at the CP in the form of quantized in-phase and quadrature (IQ) samples. These samples are compressed then delivered via the fronthaul network to the corresponding RUs. The RUs, in turn, perform decompression and IQ modulation, before broadcasting their RF signals to the users. We focus on the joint design of the linear precoders, quantizers, and capacity of the RF fronthaul links, along with the time allocation of the RF spectrum, in order to maximize the weighted sum-rate of the users, subject to power constraints and capacity limitations of the hybrid fronthaul network. The resulting problem is nonconvex and difficult to handle. Therefore, we propose a computationally-tractable algorithm that utilizes line search and alternating convex optimization in order to obtain a high-quality suboptimal solution. We provide numerical examples to demonstrate the performance of the proposed algorithm under different weather conditions. We also show the performance gain of hybrid RF/FSO fronthaul, as compared to FSO-only fronthaul, during unfavorable weather conditions.

Radio transmitters can be identified by their unique radio frequency (RF) fingerprints, which are imparted by their distinct hardware differences and manufacturing variations. However, traditional classification approaches require all classes to be known during training - a constraint that may not hold in many real-world scenarios, with new RF devices appearing during data collection. In this paper, we present an approach for novel device detection in RF data that concurrently performs RF fingerprinting and novel device detection. We introduce a combined clustering loss that explicitly encourages a clustered representation on a lower dimensional manifold, and demonstrate when and where it is most helpful. Finally, we show how to improve performance by including additional devices during the training procedure in a semi-supervised setting.

Media-based modulation (MBM) is an attractive modulation scheme that offers MIMO benefits by placing digitally controlled parasitic elements as radio frequency (RF) mirrors near the transmit antenna. Different realizations of the ON/OFF status of these mirrors result in different channel fades which constitute the channel modulation alphabet. In this paper, we investigate full-duplex (FD) communication using MBM. First, we study full-duplex MBM (FD-MBM) scheme with full channel state information (CSI). Our numerical results show that, for the same spectral efficiency, FD-MBM outperforms FD with conventional modulation (FD-CM). Next, in order to relax the full CSI assumption, we investigate FD with differential MBM (FD-DMBM) with no CSI/partial CSI, where partial CSI refers to the availability of either self-interference CSI or desired-signal CSI. We propose detectors for the FD-DMBM with no CSI/partial CSI and illustrate their bit error performance.

Person re-identification (Re-ID) is a challenging task that tries to recognize a person across different cameras, and that can prove useful in video surveillance as well as in forensics and security applications. However, traditional Re-ID systems analyzing image or video sequences suffer from well-known issues such as illumination changes, occlusions, background clutter, and long-term re-identification. To simultaneously address all these difficult problems, we explore a Re-ID solution based on an alternative medium that is inherently not affected by them, i.e., the Wi-Fi technology. The latter, due to the widespread use of wireless communications, has grown rapidly and is already enabling the development of Wi-Fi sensing applications, such as human localization or counting. These sensing procedures generally exploit Wi-Fi signals variations that are a direct consequence, among other things, of human presence, and which can be observed through the channel state information (CSI) of Wi-Fi access points. Following this rationale, in this paper, for the first time in literature, we show how the pervasive Wi-Fi technology can also be directly exploited for person Re-ID. More accurately, Wi-Fi signals amplitude and phase are extracted from CSI measurements and analyzed through a two-branch deep neural network working in a siamese-like fashion. The designed pipeline can extract meaningful features from signals, i.e., radio biometric signatures, that ultimately allow the person Re-ID. The effectiveness of the proposed system is evaluated on a specifically collected dataset, where remarkable performances are obtained; suggesting that Wi-Fi signal variations differ between different people and can consequently be used for their re-identification.