Abdallah Abdallah

The vulnerability of wireless devices to a well-known set of probable cyberattacks has made safeguarding the networks to which these devices connect a tremendous security issue, threatening the safety and security of thousands, if not millions, of private and public networks. Due to the rapid growth of embedded and wearable wireless devices on the market, wireless Internet of Things (IoT) devices are now one of the most vulnerable entry points because they don’t have advanced authentication procedures. This article provides a summary of our most recent findings in the development of a novel authentication and identification method for IoT ZigBee and Long Range (LoRa) devices based on the physical signals they emit. Our method relies on the extraction of a collection of unique features from the received modulated signal in order to construct a features vector for each device and then train a machine learning model using the acquired features. Following training, the trained model is evaluated by testing its ability to identify and recognize the authorized devices (i.e., those previously included in the training set) from the testing set, which contains an evenly distributed random mix of new and authorized devices. Our method employs differential constellation trace Figure (DCTF)-based features for the features vector and computationally-efficient machine learning methods, such as Quadratic Discriminant Analysis (QDA) and Gaussian Naive Bayes classifiers, which resulted in a recognition accuracy greater than 90 percent.

Media synthesis and manipulation has reached unprecedented levels of realism owing to the proliferation of deep learning. Deepfake has been the de-facto tool for media manipulation. Although this technology has potential in the entertainment industry, its threats include political manipulation and bypassing biometric security systems. As a result, deepfake detection has garnered widespread attention among research communities. The intuition is to use deep learning to fix the problems created by deep learning. Although convolutional neural networks have shown their dominance in the filed of pattern recognition, the receptive field-model size dilemma still persists along with the lack of interpretation for such models. While the traditional Gabor function was proposed to fix these problems, it can only generate limited linear Gabor filters which makes it optimal for limited data and applications. The contribution of this paper is quadruple: (i) proposing a unified Gabor function capable of generating linear, elliptical, and circular Gabor filters. (ii) leveraging the back-propagation learning framework to incorporate the proposed function in convolutional neural networks and generate adaptive Gabor filters. (iii) presenting a dual scale large receptive field network for deepfake image recognition. (iv) demonstrating where the proposed model stands in terms of performance and architecture size compared to state-of-the-art models. The proposed model is evaluated on four benchmark datasets: Celeb-DF (v2), DeepFake Detection Challenge Preview, FaceForensics++ and Wilddeepfake. Experimental results show that the proposed adaptive Gabor filters reduce the model size by 64.9% compared to adaptive weighted filters without performance reduction.

Radio Access Networks (RAN), Edge Computing (EC), and Network Slicing are some of the critical components in 5G networks used to provide different transport services over the same infrastructure and to enable new features, e.g., low latency and context awareness. It is important to have tools for evaluating present and future applications over these components; however, there is a gap between theoretical work on resource allocation and its deployment in simulated, emulated, or real-world experiments. This paper addresses these issues through a new emulation software named eXP-RAN, which enables experimenting with network slicing in virtualized RAN nodes and EC scenarios. In this version, we have focused on the computing part of RANs. Experiments can be created manually or imported from an optimization model. We describe the software architecture and its advantages over traditional tools such as ns-3 and Mininet, as well as more recent tool targets to EC simulations such as EdgeCloudSim. We also illustrate how two use-cases, a virtual RAN (vRAN) / Multi-Access Edge Computing (MEC) orchestration and a network slicing for video service providers can be emulated and explored using eXP-RAN.

The accurate detection of people in indoor environments requires high-cost devices, while low-cost devices, in addition to low accuracy, offer little information about the monitored events. The perturbations that result from indoor movements affect the signals received by 802.11 interfaces. Hence, an 802.11 device becomes a widely available, low-cost, and reasonably accurate solution for several applications. This paper presents WiDMove, a proposed technique to detect the entry and exit of persons, within an indoor environment, using the channel state information (CSI) measurements, which is provided by the IEEE 802.11n compliant devices. Based on the gathered CSI measurements, we utilized frequency-time analysis methodology to build an efficient features vector based on Short-Time Fourier Transform (STFT) and Principal Component Analysis (PCA). We used the extracted features to train and develop a Support Vector Machine (SVM) classifier, which provided very promising initial results. Our initial results have an accuracy near 80 %.

Long Term Evolution in unlicensed bands (LTE-U) has emerged as a promising solution to address the unprecedented growth of mobile data traffic. LTE-U extends the advantages of LTE protocols to the 5GHz unlicensed bands, which are primarily used in USA by IEEE 802.11a/n/ac WLANs and radar systems. However, LTE-U deployment is a challenging task for cellular operators due to several regulations and factors (e.g., uncertainty of available capacity). In this paper, we present a stochastic programming model for the allocation of LTE-U resources that maximizes the service capacity for LTE users, while controls the risk of collision with Wi-Fi traffic by operating under a predefined minimum collision probability.

Network functions virtualization (NFV) is an evolving field with great potentials, similarly is the software defined networks (SDNs) field, which is considered as a complement of the former. As the cloud services grow to dominate most of the web services, it becomes obvious to consider cloud-based centralized radio access networks (CRANs) as an excellent approach toward NFV. The three technologies have significant potential to improve the quality of service (QoS) while decreasing networks operating cost. Several new research proposals discussed how to deploy these technologies under new infrastructure, but only few investigated incremental transition from the current networks to the new deployments allowing a hybrid mix. The transition process faces three major issues: 1) which cell sites to be updated; 2) how much updates to be applied (i.e., change to fully virtualized or not); and 3) where the virtualized cells should be positioned. A common factor among these issues is that all are influenced by how much centralization is employed in the RAN. Therefore, this paper presents a new integer linear programming (ILP) model to allow decision makers (e.g., business managers and network architects) to investigate the impact of centralization level on decision criteria (e.g., required investment) by controlling the centralization level and its associated weight (reward). Our model shows how the investment should be positioned based on the level of centralization and the relationship between the costs of different resources. Additionally, we propose a time efficient heuristic algorithm to provide a reasonable suboptimal solution for the model under scenarios with a large number of network nodes.

Dynamic spectrum access (DSA) has emerged as an enabling technology to allow more intensive sharing of the radio spectrum. A requirement for most proposed DSA techniques is prior knowledge of the primary user's access pattern or the ability to predict primary user activities. Therefore, spectrum surveys are taking place on an even wider scale to provide data on spectrum usage and occupancy for developing new prediction models and for spectrum planning by regulators. This paper investigates the potential of mining spectrum data for correlation between human activities in a neighborhood and the resulting spectrum occupancy across different bands. We propose a systematic approach based on two clustering techniques: Gaussian mixture models (GMMs) and self-organizing map neural networks (SOMNNs). We mine spectrum measurements gathered by our network of spectrum observatories in Virginia and Illinois. The results confirm the existence of observable correlation and show that our proposed techniques detect correlation across various land mobile radio (LMR) and cellular bands under a wide range of scenarios with a high detection ratio. These results inspire us to develop more efficient prediction models for applications in opportunistic spectrum access (OSA) or self-organized networks.

The volume of video streaming traffic over the Internet has increased dramatically over the last 10 years. By the end of 2014, it is expected that 77% of the US internet traffic will be video streaming [1]. In parallel, the last ten years have seen an incredible deployment of IEEE 802.11 technologies, also known as WiFi, at homes and public places. Although, the industry and research communities have given a lot attention to address the challenges of video streaming over wireless access networks (e.g., Cellular, WiMAX, WiFi), there are still interesting problems to address to enhance the user quality of experience (QoE). In this paper, we highlight the challenges that face adaptive video streaming over wireless access networks. Then, we propose a a heuristic cross-layer adaptation algorithm named E2E-MAC to improve the video client application response to changing conditions in WiFi environments. The algorithm, which runs on the client side, combines end to end (E2E) throughput measurements with MAC-based measurements from the WiFi link to choose the appropriate streaming quality-to request from the server. Simulation results have shown that our algorithm reduces the occurrence of buffer under-run, increases the average achieved streaming quality, and lowers the rate of lost video frames. This is achieved with a minor overhead on the client application's side.

This paper describes the Flexible Internetwork Stack (FINS) Framework, an open-source tool to enable implementation-based experimental research in computer networking. The FINS Framework uses a module-based architecture that allows cross-layer behavior and runtime reconfiguration of the protocol stack. The FINS Framework is general enough to enable experimental setups under various network architectures (e.g., MANET, infrastructure, mesh) and to accelerate prototyping solutions for evolving areas (e.g., cognitive networks, cross-layer design, context-aware applications). Version 1.0 of the framework makes use of existing physical and data link layer functionality, while enabling modifications to the stack at the network layer and above, or even the implementation of a clean-slate, non-layered protocol architecture. Protocols, stubs for communicating with intact layers, and management and supervisory functions are implemented as FINS Framework modules, interconnected by a central switch. This paper describes the FINS Framework architecture, presents an initial assessment along with experiments enabled by the tool, and documents an intuitive mechanism for transparently intercepting socket calls that maintains efficiency and flexibility. Performance testing shows that the FINS Framework is capable of supporting experiments requiring IEEE 802.11g hardware speeds and operating in varying networking architecture and experimental scenarios on both Ubuntu laptops and Android devices.