Defending Against Packet-Size Side-Channel Attacks in Iot Networks

Motivated by privacy issues in the Internet of Things (IoT), we generalize a previously proposed privacy-preserving packet obfuscation scheme to guarantee differential privacy. We propose a locally differentially private packet obfuscation mechanism as a defense against packet-size side-channel attacks in IoT networks. We formulate the problem as an optimization over a conditional probability distribution (channel) between the original and obfuscated packet sizes and show that the optimal set of obfuscated packet sizes is a strict subset of the set of original packet sizes. We study the optimal mechanisms for minimizing the (average or min-max) bandwidth overhead subject to a privacy constraint by solving the corresponding (linear or convex) program. We demonstrate our methods on synthetic and real data to illustrate privacy-bandwidth tradeoffs in different settings. Systems with many bandwidth-intensive devices can easily mask low-bandwidth devices. For data collected from actual smart home $\mathrm{IoT}$ devices, we show how the packet size distributions become increasingly indistinguishable as the level of privacy protection increases. The proposed mechanism highlights the possibility for bandwidth-constrained users to optimally tune their privacy preferences and trade off privacy with bandwidth.