Sensor Placement and Resource Allocation for Energy Harvesting IoT\n Networks

Osama M. Bushnaq, Anas Chaaban, Sundeep Prabhakar Chepuri, Geert Leus, Tareq Y. Al-Naffouri

TL;DR

This paper presents a framework for jointly optimizing sensor placement, transmission power, and bandwidth in energy-harvesting IoT networks to minimize source estimation errors. By comparing analog and digital transmission schemes, the authors demonstrate that digital systems offer superior performance when bandwidth is ample and introduce a novel randomized rounding algorithm to solve the complex underlying non-convex optimization problem.

What Problem Does This Paper Address?

In the burgeoning world of the Internet of Things (IoT), deploying "self-powered" sensor networks is a major goal. However, designing these networks is a difficult balancing act. Sensors must harvest enough energy from their environment (such as solar or RF signals) to stay operational while accurately measuring and transmitting data to a central Fusion Center (FC).

The challenge lies in the "where" and the "how": Where should sensors be placed? How much power should they use? And how much bandwidth should they be allocated? Most existing research tackles these questions in isolation. This paper addresses the gap by jointly optimizing these three dimensions—location, power, and bandwidth—specifically for networks where sensors have varying energy-harvesting (EH) efficiencies, battery capacities, and costs.

Methods

The authors model two source estimation scenarios:

  1. Static Source Estimation: Estimating a fixed vector of correlated sources.
  2. Dynamic State Estimation: Tracking a scalar source over time using a steady-state Kalman filter.

They compare two communication strategies:

  • Analog: Sensors amplify and forward their noisy observations directly.
  • Digital: Sensors quantize and encode their data. The authors use an information-theoretic approach (Rate-Distortion theory and Shannon capacity) to determine the optimal number of quantization levels based on the allocated power and bandwidth.

The resulting design problem is a Boolean non-convex optimization problem—essentially a hard combinatorial puzzle where the system must decide "yes" or "no" for every possible combination of sensor type, location, and bandwidth. To solve this, the authors relax the Boolean constraints to continuous values (between 0 and 1), solve the resulting convex problem, and then apply a newly proposed generalized randomized rounding algorithm. This algorithm is specifically designed to handle joint constraints on both the total financial budget and the available spectral bandwidth.

Datasets

The paper relies on numerical simulations rather than real-world datasets. They simulate a field of 400m × 400m with 36 candidate sensor locations. The environment includes:

  • Sources: Either 5 static sources or 1 dynamic source, modeled with Gaussian noise and specific diffusion parameters.
  • Energy Sources: Two cellular base stations providing RF energy for harvesting, plus a baseline solar energy component.
  • Sensor Pool: Three types of sensors with varying costs, battery capacities, and harvesting efficiencies.

Key Insights and Ideas

  • Digital vs. Analog Trade-offs: Digital transmission generally outperforms analog because it is immune to channel noise once decoded. However, this advantage is only realized when there is sufficient system bandwidth to support high-rate quantization.
  • Strategic Resource Allocation: The optimization naturally places "expensive" sensors (with better EH and batteries) at strategic locations far from the Fusion Center where communication is difficult, while "cheaper" sensors are utilized for easier, nearby tasks.
  • The Power of Flexibility: Allowing the system to choose from a variety of sensor types and bandwidths significantly lowers the estimation error compared to a "one-size-fits-all" deployment.
  • Rounding Robustness: Standard rounding techniques often fail to produce a feasible solution when multiple constraints (like cost and bandwidth) are involved. The proposed randomized rounding algorithm effectively navigates these boundaries by taking the joint selection into account.

Key Results

  • Near-Optimal Performance: The proposed rounding algorithm achieved estimation errors very close to the theoretical lower bound (the relaxed solution), proving its efficiency in finding high-quality feasible solutions.
  • Bandwidth Sensitivity: In digital systems, increasing bandwidth directly reduces quantization distortion. In contrast, for analog systems, increasing bandwidth beyond a certain point does not improve performance because the signal-to-noise ratio (SNR) per channel remains constant if total energy is fixed.
  • Cost vs. Accuracy: The authors demonstrated a clear "Pareto front" where increasing the system budget leads to diminishing returns in error reduction, helping engineers identify the "sweet spot" for cost-effective deployment.

Limitations and Future Work

A primary limitation is that the dynamic source model is restricted to a scalar (a single value). Extending this to a vector of dynamic sources is mathematically complex because it involves solving the Discrete Algebraic Riccati Equation (DARE) within the optimization loop. Additionally, the model assumes orthogonal channels (no interference between sensors), which may not hold in extremely dense IoT deployments. Future work could explore more complex interference scenarios, multi-antenna (MIMO) configurations at the Fusion Center, and vector-based dynamic source estimation.

Summary updated

Abstract

The paper studies optimal sensor selection for source estimation in energy\nharvesting Internet of Things (IoT) networks. Specifically, the focus is on the\nselection of the sensor locations which minimizes the estimation error at a\nfusion center, and to optimally allocate power and bandwidth for each selected\nsensor subject to a prescribed spectral and energy budget. To do so,\nmeasurement accuracy, communication link quality, and the amount of energy\nharvested are all taken into account. The sensor selection is studied under\nboth analog and digital transmission schemes from the selected sensors to the\nfusion center. In the digital transmission case, an information theoretic\napproach is used to model the transmission rate, observation quantization, and\nencoding. We numerically prove that with a sufficient system bandwidth, the\ndigital system outperforms the analog system with a possibly different sensor\nselection. Two source models are studied in this paper: static source\nestimation for a vector of correlated sources and dynamic state estimation for\na scalar source. The design problem of interest is a Boolean non convex\noptimization problem, which is solved by relaxing the Boolean constraints. We\npropose a randomized rounding algorithm which generalizes the existing\nalgorithm. The proposed randomized rounding algorithm takes the joint sensor\nlocation, power and bandwidth selection into account to efficiently round the\nobtained relaxed solution.\n
BibTeX
@article{bushnaq2020sensor,
  author = {Osama M. Bushnaq and Anas Chaaban and Sundeep Prabhakar Chepuri and Geert Leus and Tareq Y. Al-Naffouri},
  title = {{Sensor Placement and Resource Allocation for Energy Harvesting IoT\textbackslash\{\}n Networks}},
  journal = {Digit. Signal Process.},
  year = {2020},
  doi = {10.1016/j.dsp.2020.102659},
}

Loading related content...

Figure