Mitigating Packet-level Signature Attacks in PINGPONG via Differential Privacy

In this digitally connected world, Internet of Things (IoT) devices are ubiquitous. From smart homes to healthcare, there has been a proliferation of these devices. However, smart devices contain personal and sensitive information, which poses a significant risk to a user’s privacy. Although IoT networks are encrypted end-to-end, inference attacks concerning the IP address, length, direction, and timestamp of packets are possible. One of the convenient techniques to launch such attacks is to analyze the packet-level signatures of IoT devices through a tool called PINGPONG. In this work, we propose using the notion of differential privacy to preserve the contextual privacy of these devices. The proposed approach involves adding noise to the packet length attribute of the devices’ PCAP files, thereby hindering their detection. Thus, our model prevents the identification of individual device activity while allowing for their aggregated data analysis. For the utility part, we have computed the Root Mean Square Error (RMSE) corresponding to various values of the privacy budget. Since we do not introduce any dummy packets or synthetic traffic in the network, our model incurs minimum overhead (which is an essential requirement in the IoT ecosystem).