Joint Sensor and Relay Power Control in Tracking Gaussian Mixture Targets by Wireless Sensor Networks

Johann A. Bengua, Hoang Duong Tuan, Trung Q. Duong, H. Vincent Poor

Abstract

This paper considers a wireless sensor network for locating a static target or tracking a dynamic target, which is characterized a priori by a Gaussian mixture distribution. An amplify-and-forward relay node acts as a wireless bridge in relaying the sensor's independent observations of the target to a fusion center (FC). Joint power allocation is considered for the sensors and relay to optimize a Bayesian filter, which is deployed at the FC for a global estimate of the target. The mean squared error of the Bayesian filter is already computationally intractable for fixed sensor and relay transmitter power, so power allocation to minimize its mean squared error is a very challenging problem. In this paper, the problem is addressed by an iterative procedure of very low computational complexity. Simulations are provided to support the efficiency of our proposed power allocation.

Introduction

In Wireless Sensor Networks (WSNs), sensors are often battery-powered and geographically dispersed. To save energy and overcome signal fading, a "relay" node can act as a bridge, amplifying sensor signals before sending them to a Fusion Center (FC) for final processing.

System model of the wireless sensor network showing sensors (S), a relay, and the fusion center (FC).

The core challenge lies in Gaussian Mixture (GM) targets. While standard targets are often modeled as a single "blob" of probability (a Gaussian), real-world targets are often multimodal—they might be in one of several locations or following complex paths. GMMs represent these targets much better. However, calculating the Bayesian Mean Squared Error (MSE) for GMMs is "computationally intractable"—there is no simple formula for it. This makes it incredibly difficult to determine exactly how much power each sensor and the relay should use to minimize tracking errors without draining their batteries instantly.

Methods: Majorant Minimization

The authors propose a "Majorant Minimization" approach to bypass the mathematical deadlock of intractable MSE.

  1. Surrogate Function: Instead of trying to minimize the true Bayesian MSE directly, they identify a "majorant" function—a tractable upper bound based on the Linear Minimum Mean Squared Error (LMMSE).
  2. Iterative Optimization: They developed an iterative procedure (Algorithm 1) that solves this surrogate problem. At each step, the algorithm finds the optimal power levels for both the sensors and the relay.
  3. Closed-Form Solutions: A key contribution is deriving closed-form solutions for these power variables using compressed cubic equations. This allows the FC to calculate optimal settings almost instantly without massive computing power.
  4. Handling Nonlinearity: For scenarios where sensors have nonlinear outputs (like measuring angles or ranges) or targets move unpredictably, the authors incorporate the Unscented Transformation (UT) to approximate the joint statistics of the target and the observations.

Key Results

The proposed method was evaluated using rigorous Monte Carlo simulations across static localization, linear dynamic tracking, and nonlinear maneuvering tracking scenarios.

Normalized MSE (dB) vs. Power Budget (PT) for linear tracking, demonstrating that the proposed 2H-Algorithm 1 outperforms one-hop and equal power schemes.
  • Accuracy Gains: In both linear and nonlinear simulations, the proposed "2H-Algorithm 1" consistently provided the lowest Normalized Mean Squared Error (NMSE) compared to one-hop (1H) systems and equal-power schemes.
  • Efficiency at Low Power: The performance gap was most noticeable at low power budgets, where the algorithm's ability to intelligently "spend" its limited decibels gave it a massive advantage (often several dB of improvement).
  • Fast Convergence: The iterative algorithm is highly efficient, typically converging to a solution in fewer than 10 iterations, making it suitable for real-time applications.
  • Qualitative Success: In maneuvering target tests, the proposed method was the only one that accurately reconstructed the "true track" of the target, while other methods diverged significantly during turns.

Conclusion

This research demonstrates that controlling sensor and relay power together is far superior to optimizing them separately. By using a tractable proxy for the MSE, the proposed algorithm achieves near-optimal tracking performance for complex GMM targets with very low computational overhead. Future work may extend this to multi-target tracking and decentralized network architectures.

Summary updated

Abstract

This paper considers a wireless sensor network for locating a static target or tracking a dynamic target, which is characterized a priori by a Gaussian mixture distribution. An amplify-and-forward relay node acts as a wireless bridge in relaying the sensor's independent observations of the target to a fusion center (FC). Joint power allocation is considered for the sensors and relay to optimize a Bayesian filter, which is deployed at the FC for a global estimate of the target. The mean squared error of the Bayesian filter is already computationally intractable for fixed sensor and relay transmitter power, so power allocation to minimize its mean squared error is a very challenging problem. In this paper, the problem is addressed by an iterative procedure of very low computational complexity. Simulations are provided to support the efficiency of our proposed power allocation.
BibTeX
@article{bengua2017joint,
  author = {Johann A. Bengua and Hoang Duong Tuan and Trung Q. Duong and H. Vincent Poor},
  title = {{Joint Sensor and Relay Power Control in Tracking Gaussian Mixture Targets by Wireless Sensor Networks}},
  journal = {IEEE Transactions on Signal Processing},
  year = {2017},
  doi = {10.1109/tsp.2017.2749211},
}
Figure