Hendrik Vogt

Hendrik Vogt, Stefan Buehler, Mark Schutera

In this paper, we propose a method for addressing the issue of unnoticed catastrophic deployment and domain shift in neural networks for semantic segmentation in autonomous driving. Our approach is based on the idea that deep learning-based perception for autonomous driving is uncertain and best represented as a probability distribution. As autonomous vehicles' safety is paramount, it is crucial for perception systems to recognize when the vehicle is leaving its operational design domain, anticipate hazardous uncertainty, and reduce the performance of the perception system. To address this, we propose to encapsulate the neural network under deployment within an uncertainty estimation envelope that is based on the epistemic uncertainty estimation through the Monte Carlo Dropout approach. This approach does not require modification of the deployed neural network and guarantees expected model performance. Our defensive perception envelope has the capability to estimate a neural network's performance, enabling monitoring and notification of entering domains of reduced neural network performance under deployment. Furthermore, our envelope is extended by novel methods to improve the application in deployment settings, including reducing compute expenses and confining estimation noise. Finally, we demonstrate the applicability of our method for multiple different potential deployment shifts relevant to autonomous driving, such as transitions into the night, rainy, or snowy domain. Overall, our approach shows great potential for application in deployment settings and enables operational design domain recognition via uncertainty, which allows for defensive perception, safe state triggers, warning notifications, and feedback for testing or development and adaptation of the perception stack.

Christan Zenger, Hendrik Vogt, Jan Zimmer et al.

Channel-reciprocity based key generation (CRKG) has gained significant importance as it has recently been proposed as a potential lightweight security solution for IoT devices. However, the impact of the attacker's position in close range has only rarely been evaluated in practice, posing an open research problem about the security of real-world realizations. Furthermore, this would further bridge the gap between theoretical channel models and their practice-oriented realizations. For security metrics, we utilize cross-correlation, mutual information, and a lower bound on secret-key capacity. We design a practical setup of three parties such that the channel statistics, although based on joint randomness, are always reproducible. We run experiments to obtain channel states and evaluate the aforementioned metrics for the impact of an attacker depending on his position. It turns out the attacker himself affects the outcome, which has not been adequately regarded yet in standard channel models.

Hendrik Vogt, Zohaib Hassan Awan, Aydin Sezgin

Full-duplex (FD) communication is regarded as a key technology in future 5G and Internet of Things (IoT) systems. In addition to high data rate constraints, the success of these systems depends on the ability to allow for confidentiality and security. Secret-key agreement from reciprocal wireless channels can be regarded as a valuable supplement for security at the physical layer. In this work, we study the role of FD communication in conjunction with secret-key agreement. We first introduce two complementary key generation models for FD and half-duplex (HD) settings and compare the performance by introducing the key-reconciliation function. Furthermore, we study the impact of the so called probing-reconciliation trade-off, the role of a strong eavesdropper and analyze the system in the high SNR regime. We show that under certain conditions, the FD mode enforces a deteriorating impact on the capabilities of the eavesdropper and offers several advantages in terms of secret-key rate over the conventional HD setups. Our analysis reveals as an interesting insight that perfect self-interference cancellation is not necessary in order to obtain performance gains over the HD mode.

Hendrik Vogt, Aydin Sezgin

Secret-key generation in a wireless environment exploiting the randomness and reciprocity of the channel gains is considered. A new channel model is proposed which takes into account the effect of reflections (or re-radiations) from receive antenna elements, thus capturing an physical property of practical antennas. It turns out that the reflections have a deteriorating effect on the achievable secret-key rate between the legitimate nodes at high signal-to-noise-power-ratio (SNR). The insights provide guidelines in the design and operation of communication systems using the properties of the wireless channel to prevent eavesdropping.