Sub-Sampling Framework Comparison for Low-Power Data Gathering: A Comparative Analysis

Bojan Milosevic, Carlo Caione, Elisabetta Farella, Davide Brunelli, Luca Benini

TL;DR

This paper tackles the critical challenge of balancing data quality and energy consumption in wireless sensor networks (WSNs) for long-term monitoring. They compare two sub-sampling techniques, Compressive Sensing (CS) and a Latent Variable (LV) model, demonstrating that CS generally offers better reconstruction accuracy and energy efficiency, especially when aggressive sampling reduction is employed.

What Problem Does This Paper Address?

Deploying wireless sensor networks (WSNs) for long-term monitoring, such as environmental monitoring of historical buildings, faces a fundamental trade-off: collecting high-resolution data requires significant energy, which limits the network's lifespan. To extend network lifetime, researchers aim to reduce the amount of data sampled and transmitted. This paper addresses how to achieve this reduction by intelligently sampling data at sub-Nyquist rates and then reconstructing the original signal, all while minimizing energy consumption. The core problem is to find the optimal balance between data accuracy and energy efficiency when using these advanced data gathering techniques.

Methods

The paper compares two primary approaches for sub-sampling and data reconstruction:

  • Compressive Sensing (CS): This technique leverages the sparsity of signals in a particular domain (e.g., frequency or time) to reconstruct the original signal from a significantly smaller number of measurements than traditionally required by the Nyquist theorem. The paper specifically explores a form of CS that involves skipping samples during acquisition, effectively reducing the sampling rate. They also investigate a "group sparse" version of CS (GS-CS) that exploits correlations among multiple sensors.
  • Latent Variable (LV) based Statistical Model: This approach models the observed multivariate data using a smaller set of unobserved "latent variables." The paper utilizes a tensor factorization technique to represent the spatio-temporal data from the heterogeneous sensor network. They further enhance this model by incorporating correlations between sensors and time (LV-MAP) to improve reconstruction.

Both methods are evaluated on a real-world WSN deployment, with a focus on detailed power consumption modeling and analysis.

Datasets

The paper utilizes data collected from a real-world WSN deployment for environmental monitoring of a historical building (Palazzina della Viola in Bologna, Italy). This dataset includes readings from temperature, humidity, ambient light, and metal-oxide gas sensors. The authors do not introduce a new benchmark dataset but rather use their experimental setup as a testbed.

Key Insights and Ideas

  • Sub-sampling is key for energy efficiency: By intelligently reducing the sampling rate, significant energy savings can be achieved, directly impacting network lifetime.
  • Correlations are crucial: Exploiting spatio-temporal and inter-sensor correlations significantly improves reconstruction accuracy for both CS and LV methods, especially for more challenging signals like light.
  • CS generally outperforms LV: For most sub-sampling scenarios, Compressive Sensing (CS) provides better reconstruction accuracy and overall energy efficiency compared to the Latent Variable (LV) model.
  • Aggressive sub-sampling favors CS: The advantage of CS becomes more pronounced as the sub-sampling factor increases.
  • LV excels with explicit correlation models: While standard LV struggles, the LV-MAP approach, which explicitly incorporates correlations, shows significant improvement, particularly for humidity and temperature signals.
  • Signal nature matters: The difficulty of reconstructing certain signals (e.g., light, which is highly irregular and noisy) impacts the performance of both methods.
  • Block size (N) trade-off: For LV-MAP, there's an optimal block size (around 256-512 samples) that balances reconstruction accuracy and data collection delay. Larger blocks don't necessarily improve accuracy and can increase delay.

Key Results

  • Reconstruction Accuracy:
    • In general, CS achieved better Signal-to-Noise Ratio (SNR) than the standard LV approach across various sub-sampling factors.
  • The LV-MAP method, incorporating correlations, showed significant improvement over standard LV, sometimes outperforming CS for specific signals (humidity, temperature) at lower sub-sampling factors.
  • For temperature signals, SNR values ranged from 40 to 70 dB with increasing sub-sampling factors.
  • Energy Efficiency:
    • The CS approach generally led to better energy efficiency, especially with aggressive sub-sampling.
  • For similar energy consumptions, CS often achieved higher reconstruction quality than LV-MAP.
  • Maximum energy savings of up to 58% were reported.
  • Parameter Impact:
    • The sub-sampling factor (p) had a significant impact on CS performance, while LV was more stable.
  • The length of the data block (N) had a less pronounced effect on GS-CS but was more critical for LV-MAP, with an optimal range identified.

Limitations and Future Work

  • Limitations:
    • The paper notes that for extremely aggressive sub-sampling policies, the LV-MAP approach can sometimes outperform CS, particularly when its explicit correlation models are highly beneficial.
  • The light signal, due to its inherent noisiness and irregularity, proved challenging for both reconstruction methods.
  • The study focuses on a specific type of historical building environment, and results might vary in different settings.
  • Future Work:
    • The authors suggest that their findings can serve as guidelines for WSN developers and inform future design specifications for sub-sampling in heterogeneous WSNs.
  • Further exploration into optimizing the trade-off between reconstruction quality, energy consumption, and data collection delay for different WSN applications is implied.

Summary updated

Abstract

A key design challenge for successful wireless sensor network (WSN) deployment is a good balance between the collected data resolution and the overall energy consumption. In this paper, we present a WSN solution developed to efficiently satisfy the requirements for long-term monitoring of a historical building. The hardware of the sensor nodes and the network deployment are described and used to collect the data. To improve the network’s energy efficiency, we developed and compared two approaches, sharing similar sub-sampling strategies and data reconstruction assumptions: one is based on compressive sensing (CS) and the second is a custom data-driven latent variable-based statistical model (LV). Both approaches take advantage of the multivariate nature of the data collected by a heterogeneous sensor network and reduce the sampling frequency at sub-Nyquist levels. Our comparative analysis highlights the advantages and limitations: signal reconstruction performance is assessed jointly with network-level energy reduction. The performed experiments include detailed performance and energy measurements on the deployed network and explore how the different parameters can affect the overall data accuracy and the energy consumption. The results show how the CS approach achieves better reconstruction accuracy and overall efficiency, with the exception of cases with really aggressive sub-sampling policies.
BibTeX
@inproceedings{milosevic2015sub,
  author = {Bojan Milosevic and Carlo Caione and Elisabetta Farella and Davide Brunelli and Luca Benini},
  title = {{Sub-Sampling Framework Comparison for Low-Power Data Gathering: A Comparative Analysis}},
  booktitle = {Sensors},
  year = {2015},
  url = {https://openreview.net/forum?id=Re4dsp8KI2},
}
Figure