Cross-Layer Protocol Fingerprint for Large-Scale Fine-Grain Devices Identification

Dan Yu, Haoguang Xin, Yongle Chen, Yao Ma, Junjie ChenIEEE Access 2020 · DOI

Abstract

Internet-connected Internet of Things (IoT) devices are exploding, which pose a significant threat for their management and security protection. IoT device identification is a prerequisite for discovering, monitoring, and protecting these devices. Although the existing proactive identification methods based on protocol fingerprint can discover and identify large-scale IoT devices, the fingerprint granularity is difficult to meet the requirements of security risk assessment for large-scale IoT devices. Since IoT devices usually support multiple network protocols for specific collection and control tasks, we propose a cross-layer protocol fingerprint to achieve large-scale fine-grained devices identification instead of traditional single protocol fingerprint. We first design a probing scheme for gathering HTTP and TCP cross-layer packets. Then we select the specific field of the HTTP and TCP protocols based on the diversity and consistence of field value. Finally, we utilize convolutional neural network (CNN) and long-term memory network (LSTM) to extract and construct feature fingerprint of these specific fields, and achieve a fine-grain IoT devices identification with high accuracy. The experimental results show that our identification accuracy of devices model reaches 96.6%, the recall rate reaches 97.4%.

Introduction

The rapid expansion of the Internet of Things (IoT) has outpaced traditional security management capabilities. Effective network protection begins with precise asset identification. However, identifying IoT devices at scale presents a significant challenge. Existing methods typically rely on application-layer banners (e.g., Shodan), which are fast but lack granularity, or active fingerprinting (e.g., Nmap), which is accurate but resource-intensive and potentially intrusive.

This research introduces a "cross-layer" fingerprinting approach that balances granularity and overhead. By analyzing both HTTP (application layer) and TCP (transport layer) traffic, the proposed system identifies specific device models with high precision while minimizing network impact.

Methodology

The proposed identification framework consists of three primary stages: lightweight probing, cross-layer feature selection, and deep learning-based classification.

Overview of the proposed IoT device identification framework, illustrating the stages from data collection to final model probability output.
  1. Lightweight Probing: Instead of exhaustive scanning, the system sends a single standard HTTP request. This triggers a natural sequence of packets—including the TCP handshake, HTTP response, and connection teardown—providing sufficient data for analysis without excessive overhead.
  2. Cross-Layer Feature Selection: The method extracts features from both the HTTP and TCP layers. HTTP fields like Server, Content-Type, and Content-Length are combined with TCP metrics such as Average TCP Segment Size and Average Window Size. The researchers utilized mathematical metrics (Coefficient of Variation and Euclidean distance) to ensure selected fields are consistent for a specific model but diverse across different models.
  3. CNN+LSTM Classifier: A hybrid neural network processes the extracted features. The Convolutional Neural Network (CNN) extracts spatial patterns from protocol field content, while the Long Short-Term Memory (LSTM) network captures the temporal relationships between sequential packets. A final SoftMax layer outputs the probability of the device belonging to a specific brand or model.

Experimental Evaluation

The study utilized a dataset of 15,000 real-world devices sourced via the Censys API, balanced across three categories: network cameras, routers, and printers. This dataset included 238 brands and hundreds of unique models.

The CNN+LSTM model demonstrated superior performance compared to traditional machine learning architectures:

  • Identification Accuracy: Reached 96.6% for specific device models.
  • Recall Rate: Achieved 97.4%.
  • Brand Identification: Accuracy exceeded 98%.

The system maintained high performance even as identification granularity increased from brand-level to model-level, whereas simpler models like CNN+SVM experienced significant accuracy degradation.

Conclusion

By leveraging cross-layer protocol features and hybrid deep learning, this research provides a robust, non-intrusive solution for fine-grained IoT device identification. While factors like network congestion and the shift toward encrypted protocols (HTTPS) present ongoing challenges, this framework establishes a high-precision baseline for large-scale IoT asset management and security monitoring.

Summary updated

Abstract

Internet-connected Internet of Things (IoT) devices are exploding, which pose a significant threat for their management and security protection. IoT device identification is a prerequisite for discovering, monitoring, and protecting these devices. Although the existing proactive identification methods based on protocol fingerprint can discover and identify large-scale IoT devices, the fingerprint granularity is difficult to meet the requirements of security risk assessment for large-scale IoT devices. Since IoT devices usually support multiple network protocols for specific collection and control tasks, we propose a cross-layer protocol fingerprint to achieve large-scale fine-grained devices identification instead of traditional single protocol fingerprint. We first design a probing scheme for gathering HTTP and TCP cross-layer packets. Then we select the specific field of the HTTP and TCP protocols based on the diversity and consistence of field value. Finally, we utilize convolutional neural network (CNN) and long-term memory network (LSTM) to extract and construct feature fingerprint of these specific fields, and achieve a fine-grain IoT devices identification with high accuracy. The experimental results show that our identification accuracy of devices model reaches 96.6%, the recall rate reaches 97.4%.
BibTeX
@inproceedings{yu2020cross,
  author = {Dan Yu and Haoguang Xin and Yongle Chen and Yao Ma and Junjie Chen},
  title = {{Cross-Layer Protocol Fingerprint for Large-Scale Fine-Grain Devices Identification}},
  booktitle = {IEEE Access},
  year = {2020},
  doi = {10.1109/access.2020.3026818},
}
Figure