DeepRadioID

Abstract

Network slicing of multi-access edge computing (MEC) resources is expected to be a pivotal technology to the success of 5G networks and beyond. The key challenge that sets MEC slicing apart from traditional resource allocation problems is that edge nodes depend on tightly-intertwined and strictly-constrained networking, computation and storage resources. Therefore, instantiating MEC slices without incurring in resource over-provisioning is hardly addressable with existing slicing algorithms. The main innovation of this paper is Sl-EDGE, a unified MEC slicing framework that allows network operators to instantiate heterogeneous slice services (e.g., video streaming, caching, 5G network access) on edge devices. We first describe the architecture and operations of Sl-EDGE, and then show that the problem of optimally instantiating joint network-MEC slices is NP-hard. Thus, we propose near-optimal algorithms that leverage key similarities among edge nodes and resource virtualization to instantiate heterogeneous slices 7.5x faster and within 0.25 of the optimum. We first assess the performance of our algorithms through extensive numerical analysis, and show that Sl-EDGE instantiates slices 6x more efficiently then state-ofthe-art MEC slicing algorithms. Furthermore, experimental results on a 24-radio testbed with 9 smartphones demonstrate that Sl-EDGE provides at once highly-efficient slicing of joint LTE connectivity, video streaming over WiFi, and ffmpeg video transcoding.

SignCRF: Scalable Channel-agnostic Data-driven Radio Authentication System

“…An alternative approach that has recently gained traction in the research and industry communities is the concept of Radio Frequency Fingerprinting through Deep Learning (RFFDL). The RFFDL is as an emerging data-driven authentication approach that has shown promising results in authenticating IoT radios in a more energy-efficient and scalable fashion [2], [3], [4], [5], [6]. It relies on the hardwarelevel imperfections that are uniquely associated with the Radio Frequency (RF) analog circuitry of every single device, referred to as RF "fingerprint" or "signature".…”

“…In spite of the promising progress in this domain, recent studies have revealed a number of challenges in developing robust and reliable RFFDL models for practical applications. Key among them are the non-stationary characteristic of the wireless channel and the dynamic nature of the environmental actions in a mobile scenario, which significantly limit fingerprinting accuracy by obscuring the hardware impairments associated with the transmitted waveform [2], [3], [7]. In particular, without intervention of sophisticated digital signal processing (DSP) algorithms, the existing RFFDL models suffer from significant performance degradation when training and testing of their deep learning classifier are done on separate segments of the data that are collected in different days/environments.…”

“…In particular, without intervention of sophisticated digital signal processing (DSP) algorithms, the existing RFFDL models suffer from significant performance degradation when training and testing of their deep learning classifier are done on separate segments of the data that are collected in different days/environments. This problem is introduced as the worst-case scenario in [3], [2], [7], and [8], but to the best of our knowledge has not been resolved yet.…”

“…In an effort to address these unique challenges, the authors of [3] proposed the DeepRadioID system that leverages a carefully optimized digital Finite Impulse Response (FIR) filter at the transmitter's side, which slightly modifies its base-band signal to compensate for current channel conditions; however, the system depends on an overhead DSP at the transmitter with feedback required from the receiver. Sign.…”

“…Most importantly, the dynamic behavior of the non-stationary wireless channel along with the environmental changes alter the transmitted signals in ways that makes it challenging even for the finest and most sophisticated classifiers to extract the unique impairments associated to each device. This significantly impacts the fingerprinting accuracy especially when the RFFDL Neural Network (NN) is trained on a dataset that is collected in a day or environment that is different from the day or environment that is used for classification inference -the main reason that none of the prior work were successful in achieving good performance in this worst-case scenario [2], [3], [7], [8], [9].…”