Secure and Cost-Effective Distributed Aggregation for Mobile Sensor Networks

Kehua Guo, Ping Zhang, Jianhua Ma2016 · DOI

TL;DR

This paper introduces MFSDA, a novel secure data aggregation scheme for mobile sensor networks that supports multiple statistical queries (both summation-based and comparison-based) in a single query. It offers two versions: MFSDA-I for accurate results and MFSDA-II for reduced communication costs at the expense of slight accuracy loss, making it suitable for dynamic and resource-constrained environments.

What Problem Does This Paper Address?

Mobile sensor networks (MSNs) are increasingly used in various applications, but their limited energy capacity and the need for secure data processing pose significant challenges. Traditional data aggregation schemes often struggle with:

  • Limited Functionality: Most schemes support only one type of statistic (e.g., summation or comparison), requiring multiple queries for comprehensive analysis.
  • Inefficiency in Dynamic Networks: Handling networks where nodes frequently join or leave (due to sleep modes or failures) is computationally expensive and can create bottlenecks.
  • Security Vulnerabilities: Protecting data integrity and confidentiality during aggregation is crucial, especially in hostile environments.
  • High Communication Costs: Transmitting raw data or intermediate results can consume significant energy, limiting network lifespan.

This paper aims to address these issues by proposing a multi-functional, secure, and cost-effective distributed aggregation scheme that can handle various statistics simultaneously and adapt to dynamic network conditions.

Methods

The core of the proposed MFSDA scheme lies in its innovative approach to data processing before aggregation:

  • Mapping and Encoding: Sensor data ($x_k$) is first mapped to a new domain ($y_k$) and then encoded into a binary vector ($\vec{v}_k$). This step is crucial for enabling both value-preserving (for summation-based statistics) and order-preserving (for comparison-based statistics) properties simultaneously.
  • Mapping: $y_k = f_m(x_k) = \frac{x_k - X_{LB}}{a}$, where $X_{LB}$ is the lower bound and $a$ is the accuracy requirement.
  • Encoding: $\vec{v}_k$ is a binary vector of length $L$, where the $y_k$-th element is 1 and others are 0. This effectively represents the mapped data in a way that facilitates subsequent aggregation.
  • Homomorphic Encryption: For inter-cluster data transmission (between cluster heads or to the server), homomorphic encryption is employed. This allows aggregation operations to be performed directly on encrypted data without decryption, significantly reducing computation and key management overhead. The scheme uses an EC-ElGamal-based homomorphic encryption algorithm.
  • Identity-Based Signature (IBS): To ensure data integrity and authenticity, an IBS scheme is used for all packets. This simplifies key management compared to traditional public-key infrastructure.
  • MFSDA-I (Accurate Version): This version focuses on achieving accurate results. It includes an additional compression step for MFSDA-I to further reduce packet size.
  • MFSDA-II (Approximate Version): This version introduces a compression step before encoding. This compression reduces the vector length ($L$ to $L'$) and significantly cuts down communication costs, but at the expense of some accuracy loss.

Datasets

  • Simulated Datasets:
    • Uniform distribution dataset generated using Matlab's unifrnd function.
  • Poisson distribution dataset generated using Matlab's poissrnd function (Lambda = 200).
  • Real-world Dataset:
    • Wind direction data from the TAO (Tropical Atmosphere Ocean) project by NOAA. This dataset contains 2000 samples from 1992-1993, with a measurement range of [0, 359] and an accuracy of 5-7.8.

Key Insights and Ideas

  • Unified Statistics Handling: The mapping and encoding steps are the key innovation that allows a single aggregation process to support both summation-based (e.g., SUM, AVG, VAR) and comparison-based (e.g., MAX, MIN) statistics. This is a significant improvement over prior work that typically required separate queries.
  • Trade-off Between Accuracy and Cost: MFSDA-II demonstrates that by introducing lossy compression, communication costs can be drastically reduced, offering a practical solution for energy-constrained MSNs where perfect accuracy might not always be necessary.
  • Dynamic Network Adaptability: The scheme is designed to be efficient in dynamic networks by implicitly handling varying numbers of active nodes without requiring explicit reporting mechanisms that incur extra communication costs.
  • Security through Encryption and Signatures: The combination of homomorphic encryption for secure in-network aggregation and IBS for data integrity provides a robust security framework.

Key Results

  • Functionality: MFSDA can obtain multiple statistics (summation-based and comparison-based) in a single query, unlike many existing schemes.
  • Communication Cost:
    • MFSDA-I shows comparable communication costs to RCDA-HETE in the first stage and is more efficient than RCDA-HOMO and EERCDA when the network size ($N$) is greater than $L$.
  • MFSDA-II significantly reduces communication costs. For example, with a compression factor $c=2$, the data length was reduced by 50% compared to MFSDA-I, and with $c=4$, it was reduced to 25%.
  • Accuracy (MFSDA-II):
    • For the TAO wind direction dataset, with $c=2$, the maximum error rate was 2% and the minimum was 0.2%.
  • With $c=4$, the maximum error rate remained around 2%, while the minimum was 0.5%.
  • The error rates are generally within acceptable reference boundaries, especially for datasets similar to uniform distributions.
  • Security: MFSDA is shown to be resilient against eavesdropping, replay attacks, data tampering, and DoS attacks. It also handles compromised nodes by detecting falsified values outside the bound.

Limitations and Future Work

  • Accuracy Loss in MFSDA-II: The primary limitation of MFSDA-II is the introduction of errors due to lossy compression. While the errors are often offset, further analysis is needed to precisely quantify the impact on different statistics and datasets.
  • Reference Boundary Prediction: While the reference boundary formula provides a good prediction for uniform distributions, it may not be directly applicable to highly non-uniform datasets, requiring careful selection of the compression factor.
  • Future Work: The paper suggests that further analysis is needed to understand the positive and negative errors that can occur simultaneously in MFSDA-II and how they offset each other. Additionally, detailed calculations for specific statistical error rates are referred to Appendix D, implying that a deeper dive into these calculations could be a future direction.

Summary updated

BibTeX
@misc{guo2016secure,
  author = {Kehua Guo and Ping Zhang and Jianhua Ma},
  title = {{Secure and Cost-Effective Distributed Aggregation for Mobile Sensor Networks}},
  year = {2016},
  doi = {10.3390/s16040583},
}
Figure