Abstract: Secure signal authentication is arguably one of the most challenging problems in the Internet of Things (IoT) environment, due to the large-scale nature of the system and its susceptibility to man-in-the-middle and eavesdropping attacks. In this paper, a novel deep learning method is proposed for dynamic authentication of IoT signals to detect cyber attacks. The proposed learning framework, based on a long short-term memory (LSTM) structure, enables the IoT devices (IoTDs) to extract a set of stochastic features from their generated signal and dynamically watermark these features into the signal. This method enables the cloud, which collects signals from the IoT devices, to effectively authenticate the reliability of the signals. Moreover, in massive IoT scenarios, since the cloud cannot authenticate all the IoTDs simultaneously due to computational limitations, a game-theoretic framework is proposed to improve the cloud's decision making process by predicting vulnerable IoTDs. The mixed-strategy Nash equilibrium (MSNE) for this game is derived and the uniqueness of the expected utility at the equilibrium is proven. In the massive IoT system, due to a large set of available actions for the cloud, it is shown that analytically deriving the MSNE is challenging and, thus, a learning algorithm proposed that converges to the MSNE. Moreover, in order to cope with the incomplete information case in which the cloud cannot access the state of the unauthenticated IoTDs, a deep reinforcement learning algorithm is proposed to dynamically predict the state of unauthenticated IoTDs and allow the cloud to decide on which IoTDs to authenticate. Simulation results show that, with an attack detection delay of under 1 second the messages can be transmitted from IoT devices with an almost 100% reliability.
1. Introduction
The Internet of Things (IoT) is expanding at an unprecedented scale, yet the "perceptual layer"—comprising sensors like accelerometers and RFID tags—remains notoriously difficult to secure. These devices are often too low-power for heavy cryptographic encryption, making them prime targets for man-in-the-middle and data injection attacks. This paper addresses two primary hurdles: the vulnerability of traditional static watermarking to reverse-engineering and the computational bottleneck at gateways in massive IoT scenarios, where authenticating every incoming packet in real-time is often impossible.
2. Dynamic LSTM-Based Watermarking
To counter static vulnerabilities, the authors propose a dynamic watermarking scheme. Instead of embedding a fixed bitstream, an IoT device (IoTD) employs a Long Short-Term Memory (LSTM) network to extract stochastic features—such as spectral flatness, skewness, and kurtosis—directly from its generated sensor data. These time-varying features are then watermarked into the signal. Because the underlying sensor data is dynamic, the resulting watermark is inherently difficult for an attacker to spoof or replay, effectively tying the security key to the physical reality the sensor observes.
3. Strategic Authentication and Resource Allocation
In massive IoT environments, the gateway must strategically decide which devices to authenticate due to limited computational resources. This is modeled as a non-cooperative game between the defender (gateway) and an attacker.
- Fictitious Play (FP): Used in full-information scenarios to reach a Mixed-Strategy Nash Equilibrium (MSNE), determining the optimal probability distribution for device verification.
- Deep Reinforcement Learning (DRL): In realistic "incomplete information" scenarios, a DRL agent with an LSTM-based Q-learning architecture predicts device vulnerability based on historical patterns, allowing for proactive defense.
4. Key Results and Insights
The proposed framework treats security as a resource management problem. A key theoretical finding suggests that gateways should always utilize maximum available computational resources for authentication to minimize risk. Experimental results using real-world accelerometer data demonstrate:
- High Reliability: Near 100% message reliability with an attack detection delay of less than 1 second.
- Superior Accuracy: The LSTM-based feature extraction achieved bit error rates 1 to 2 orders of magnitude lower than traditional static methods across various signal-to-noise ratios.
- Efficiency at Scale: In simulations with 1,000 devices, the DRL strategy reduced the number of successfully compromised devices by up to 30% compared to uniform random checking.
5. Conclusion
This research demonstrates that tying security to the physical "fingerprint" of sensor signals via deep learning provides a robust defense against sophisticated IoT attacks. While training costs and hardware constraints for the lowest-end devices remain areas for future optimization, the combination of dynamic watermarking and strategic resource allocation offers a scalable path forward for securing massive IoT ecosystems.
Summary updated