Multi-Bit Distributed Detection of Sparse Stochastic Signals Over Error-Prone Reporting Channels

Linlin Mao, Shefeng Yan, Zeping Sui, Hongbin Li

TL;DR

This paper develops a robust framework for detecting sparse stochastic signals in wireless sensor networks using multi-bit quantization. By optimizing how sensors compress data before sending it over noisy channels, the authors show that Likelihood Ratio Quantization (LQ) significantly outperforms traditional raw data quantization, nearly matching the performance of uncompressed systems even with very few bits.

What Problem Does This Paper Address?

In Wireless Sensor Networks (WSNs), sensors often need to detect "sparse" signals—events that are geographically or temporally rare (like a gas leak or a battlefield intrusion). However, these sensors are constrained by limited battery life and narrow communication bandwidth.

To save energy, sensors "quantize" their data (turning high-resolution readings into just a few bits) before sending it to a central Fusion Center (FC). The challenge is two-fold:

  1. Information Loss: Reducing a signal to 1 or 2 bits naturally throws away useful information.
  2. Channel Noise: The "reporting channels" between sensors and the FC are often "error-prone," meaning bits can flip during transmission, leading to false alarms or missed detections.

While previous research focused on one-bit quantization or perfect channels, this paper tackles the more realistic scenario of multi-bit quantization over noisy reporting channels.

Methods

The authors propose two primary strategies for local sensors to process data before transmission:

  • Raw Quantization (RQ): Sensors quantize the actual observed signal values.
  • Likelihood Ratio Quantization (LQ): Sensors first calculate the likelihood ratio (a statistical measure of how likely the signal is present vs. absent) and then quantize that value.

At the Fusion Center, they employ the Locally Most Powerful Test (LMPT). This is a statistical framework designed to be optimal for detecting very weak signals.

The Secret Sauce: Threshold Optimization The performance of a quantizer depends heavily on its "thresholds" (the boundaries that decide if a reading becomes a '00', '01', etc.). The authors use Particle Swarm Optimization (PSOA) to find the optimal thresholds that maximize the Fisher Information (FI). Essentially, they tune the sensors to be as sensitive as possible to the specific signal they are looking for, while accounting for the fact that some bits will inevitably be corrupted by channel noise.

Datasets

The paper relies on synthetic datasets generated via Monte Carlo simulations (5000 trials per scenario).

  • Signal Model: Sparse signals are modeled using a Bernoulli-Gaussian distribution, which is standard for representing signals that are mostly zero but occasionally have Gaussian-distributed peaks.
  • Channel Model: Reporting channels are modeled as Binary Symmetric Channels (BSCs) with varying crossover probabilities ($P_e$), representing different levels of environmental interference.

Key Insights and Ideas

  • LQ > RQ: Quantizing the Likelihood Ratio (LQ) is almost always superior to quantizing raw data (RQ). This is because LQ inherently "knows" what information is most relevant for the detection task.
  • The Hamming Distance Advantage: A fascinating insight is that LQ naturally maps high-probability signal regions to binary codewords that are far apart in "Hamming distance." This makes the system more resilient to bit-flip errors; a single bit error is less likely to turn a "signal present" message into a "signal absent" message.
  • Threshold Adaptation: The optimal quantization threshold actually decreases as channel noise increases. To provide the Fusion Center with enough information to overcome noise, sensors must become "braver" and report potential signals at lower local thresholds.
  • Diminishing Returns of Bits: Moving from 1-bit to 3-bit quantization provides a massive performance boost, bringing the system very close to the "Clairvoyant" detector (the theoretical limit of a system with no quantization at all).

Key Results

  • Efficiency: Using Asymptotic Relative Efficiency (ARE) analysis, the authors show that RQ often requires 2 to 4 times as many sensors as LQ to achieve the same detection performance in noisy environments.
  • Robustness: In a scenario with a Signal-to-Noise Ratio (SNR) of -9.2 dB and a 10% bit-error rate on the channel, the proposed 3-bit LQ detector maintained high detection probability while 1-bit methods failed significantly.
  • Near-Optimal Performance: Simulation results confirmed that with just 3 bits of quantization, the LMPT detectors achieve near-optimal performance, validating the theoretical closed-form expressions derived in the paper.

Limitations and Future Work

  • Homogeneity Assumption: Much of the analytical derivation assumes a "homogeneous environment" where all sensors have similar observation quality. Real-world deployments might have "near" and "far" sensors with vastly different SNRs.
  • Known Channel State: The optimization assumes the crossover probability ($P_e$) of the channel is known. While the authors tested the impact of inaccurate $P_e$ estimates (showing significant degradation), a more robust system might need to estimate channel conditions in real-time.
  • Future Directions: The authors suggest extending this work to non-Gaussian noise environments and exploring heterogeneous networks where different sensors might use different bit-depths based on their local energy reserves.

Summary updated

Abstract

We consider a distributed detection problem within a wireless sensor network (WSN), where a substantial number of sensors cooperate to detect the existence of sparse stochastic signals. To achieve a trade-off between detection performance and system constraints, multi-bit quantizers are employed at local sensors. Then, two quantization strategies, namely raw quantization (RQ) and likelihood ratio quantization (LQ), are examined. The multi-bit quantized signals undergo encoding into binary codewords and are subsequently transmitted to the fusion center via error-prone reporting channels. Upon exploiting the locally most powerful test (LMPT) strategy, we devise two multi-bit LMPT detectors in which quantized raw observations and local likelihood ratios are fused respectively. Moreover, the asymptotic detection performance of the proposed quantized detectors is analyzed, and closed-form expressions for the detection and false alarm probabilities are derived. Furthermore, the multi-bit quantizer design criterion, considering both RQ and LQ, is then proposed to achieve near-optimal asymptotic performance for our proposed detectors. The normalized Fisher information and asymptotic relative efficiency are derived, serving as tools to analyze and compensate for the loss of information introduced by the quantization. Simulation results validate the effectiveness of the proposed detectors, especially in scenarios with low signal-to-noise ratios and poor channel conditions.
BibTeX
@article{mao2024multi,
  author = {Linlin Mao and Shefeng Yan and Zeping Sui and Hongbin Li},
  title = {{Multi-Bit Distributed Detection of Sparse Stochastic Signals Over Error-Prone Reporting Channels}},
  journal = {IEEE Trans. Signal Inf. Process. Over Networks},
  year = {2024},
  doi = {10.1109/tsipn.2024.3496253},
}

Loading related content...

Figure