Abstract
The research has witnessed three stages. It started with theoretical channel modelling for physical layer secrecy under active and passive eavesdropping and designing algorithms for agreeing on shared secrets using wireless channel fading. The protocols were implemented on software-defined radios and evaluated for efficacy under single and multiple antenna systems. The second phase of the research was about designing RF antenna-polarization assisted secret agreement on the fly between unauthenticated COTS devices such as smartphones- Wave2Share. Wave2Share supports scenarios where two parties can securely exchange information by just waving the phone for a few seconds. It exploits the physical layer of WiFi Direct protocols and varying polarization in the line of sight RF channel for agreeing on a shared secret with sufficient entropy. The third phase of our research investigates security and usability tradeoff in two-factor authentication (TFA) used by industry. Current solutions limit usage to less than 20%. First, we demonstrate attacks on most usable forms of two-factor authentication, i.e., push-based TFA, which affects almost all push TFA in practice today. Then, we develop algorithms and designs for four protocols and methods to address both the security and usability issues. The core contribution behind the proposes architecture is removing friction at the user end through an algorithmic confluence of wireless sensing and peripheral sensors on a typical smartphone. We propose PhyAuth (radiofrequency assisted TFA), AudiAuth (acoustics based TFA to support only proximity logins and proof of possession of the token device), TeethAuth (TFA based on the exploitation of earphones as receivers and sensing of teeth related gestures in place of traditional OTPs and TOTP codes) and PushAuth (a set of design interventions for making TFA more engaging and resilient to attacks). We propose multi-modal sensing and verification techniques which negates possibilities of both remote and collocated attacks. Since a large part of the research is based on extracting randomness from ambient sources, we also proposed methods to induce randomness across sensing modalities through oscillator anonymization and varying polarization. Further, we demonstrated how randomness could be used to enhance jamming resilience of devices which intermittently communicate without any prior authentication and trust.