24 October 2019 in IEEE Transactions on Cognitive Communications and Networking
Due to the unprecedented scale of the Internet of Things, designing scalable, accurate, energy-efficient and tamper-proof authentication mechanisms has now become more important than ever. To this end, in this paper we present ORACLE, a novel system based on convolutional neural networks (CNNs) to “fingerprint” (i.e., identify) a unique radio from a large pool of devices by deep-learning the fine-grained hardware impairments imposed by radio circuitry on physical-layer I/Q samples. First, we show how hardware-specific imperfections are learned by the CNN framework. Then, we extensively evaluate the performance of ORACLE on several first-of-its-kind large-scale datasets of WiFi-transmissions collected “in the wild”, as well as a dataset of nominally-identical (i.e., equal baseband signals) WiFi devices, reaching 80-90% accuracy is many cases with the error gap arising due to channel-induced effects. Finally, we show through an experimental testbed, how this accuracy can reach over 99% by intentionally inserting and learning the effect of controlled impairments at the transmitter side, to completely remove the impact of the wireless channel. Furthermore, to scale this approach for classifying potential thousands of radios, we propose an impairment hopping spread spectrum (IHOP) technique that is resilient to spoofing attacks.
11 May 2018 in IEEE Transactions on Network Science and Engineering
Next generation network access technologies and Internet applications have increased the challenge of providing satisfactory quality of experience for users with traditional congestion control protocols. Efforts on optimizing the performance of TCP by modifying the core congestion control method depending on specific network architectures or apps do not generalize well under a wide range of network scenarios. This limitation arises from the rule-based design principle, where the performance is linked to a pre-decided mapping between the observed state of the network to the corresponding actions. Therefore, these protocols are unable to adapt their behavior in new environments or learn from experience for better performance. We address this problem by integrating a reinforcement-based Q-learning framework with TCP design in our approach called QTCP. QTCP enables senders to gradually learn the optimal congestion control policy in an on-line manner. QTCP does not need hard-coded rules, and can therefore generalize to a variety of different networking scenarios. Moreover, we develop a generalized Kanerva coding function approximation algorithm, which reduces the computation complexity of value functions and the searchable size of the state space. We show that QTCP outperforms the traditional rule-based TCP by providing 59.5% higher throughput while maintaining low transmission latency.
13 September 2018 in IEEE Transactions on Mobile Computing
As the dominant protocol on the Internet, TCP has attracted significant attention and has been implemented in various ways, each of which optimizes for a single objective such as high throughput or low delay. However, in today's mobile networks that carry traffic from diverse types of flows, this approach may lead to misconfiguration of TCP congestion control algorithms and further degrade performance for many applications. In this paper, we propose Janus, a new transport-layer framework that automatically selects among existing congestion control variants to optimize traffic in accordance with application demands. Janus is easy to deploy because it reuses existing, well-tested congestion control implementations, and does not require any in-network or client-side changes. To explore the potential for this approach, we implement Janus in the Linux kernel and extensively evaluate its performance with both emulated and real Internet traffic. We show Janus outperforms alternative protocols by offering fast convergence times in response to changing network conditions, achieving 5-10X lower delay with comparable or higher throughput. Our approach also significantly improves user-perceived performance according to QoE metrics, with up to 5X fewer interruptions for video streaming applications and 2X faster page loading for web-browsing applications.