DOA Estimation on One-Bit Quantization Observations through Noise-Boosted Multiple Signal Classification

Yan Pan, Li Zhang, Liyan Xu, Fabing Duan2024 · DOI

TL;DR

This paper introduces a novel "Noise-Boosted Quantizer Unit" (NBQU) for Direction-of-Arrival (DOA) estimation using low-cost, one-bit quantized sensor data. By strategically injecting noise before quantization, the proposed method significantly improves estimation accuracy and resolution, approaching the performance of systems using full-precision analog-to-digital converters.

What Problem Does This Paper Address?

Direction-of-Arrival (DOA) estimation is crucial for applications like wireless communications, radar, and sonar. Traditional methods often rely on high-resolution analog-to-digital converters (ADCs), which are expensive and power-hungry. One-bit ADCs offer a low-cost, low-power alternative, but their coarse quantization severely degrades estimation accuracy and resolution. This paper tackles the challenge of achieving high-performance DOA estimation with these limited one-bit observations, aiming to bridge the performance gap between one-bit systems and their analog counterparts.

Methods

The core innovation is the Noise-Boosted Quantizer Unit (NBQU). Instead of directly quantizing the received signal, the NBQU first adds a controlled amount of random noise to the signal. This noise injection has a surprising beneficial effect: it smooths the input-output characteristic of the one-bit quantizer. The authors then apply the Multiple Signal Classification (MUSIC) algorithm to the outputs of an array of these NBQUs, resulting in the NBQU-based MUSIC method.

Key aspects of the method include:

  • Noise Injection: A controlled amount of Gaussian noise is added to the incoming signal before it's quantized.
  • Averaging: Multiple noise-added samples are generated and averaged to form the final one-bit measurement. This averaging process, combined with the noise, effectively smooths the non-linear quantizer's response.
  • NBQU Structure: The paper details the mathematical formulation of the NBQU, showing how the noise injection transforms the discrete sign function into a more continuous and derivable function, especially as the number of noise samples (L) increases.
  • MUSIC Algorithm: The standard MUSIC algorithm is then applied to the covariance matrix derived from the NBQU outputs. The key idea is that by optimizing the injected noise's RMS value, the performance of the MUSIC algorithm can be significantly enhanced.

Datasets

Not discussed in the paper. The evaluation relies on simulated data generated according to the described one-bit observation model.

Key Insights and Ideas

  • Noise as a Feature, Not a Bug: The paper demonstrates that carefully controlled noise injection can actually improve the performance of systems with non-linearities like one-bit quantizers, a concept rooted in stochastic resonance principles.
  • Smoothing the Quantizer: The primary mechanism for improvement is the smoothing of the quantizer's transfer function, which makes the subsequent covariance matrix estimation more robust and the signal/noise subspaces better separated.
  • Approaching Analog Performance: The NBQU-based MUSIC method shows a remarkable ability to approach the performance of traditional MUSIC algorithms that use full-precision analog data, especially with a sufficient number of noise samples.
  • Trade-off Between Noise and Computation: While more noise samples (larger L) improve accuracy, they also increase computational complexity. The paper suggests this is a worthwhile trade-off for improved performance.

Key Results

  • Improved Accuracy and Resolution: The NBQU-based MUSIC method consistently outperforms existing one-bit MUSIC methods in terms of Root Mean Square Error (RMSE) for DOA estimation across various Signal-to-Noise Ratios (SNRs) and snapshot numbers.
  • Resolution of Closely Spaced Sources: Figure 3 shows that the proposed method can successfully resolve two closely spaced sources (10° and 5.5° apart) where other one-bit methods fail.
  • Near-Analog Performance: With an optimal noise RMS and a sufficiently large number of noise groups (L=500), the NBQU-based method's RMSE is very close to that of the unquantized MUSIC algorithm.
  • Low SNR Advantage: The proposed method shows a significantly superior performance compared to other one-bit methods under low SNR conditions.
  • Resolution Probability: Figure 6 demonstrates that the NBQU-based MUSIC method achieves a resolution probability comparable to unquantized MUSIC, especially in low SNR and low snapshot scenarios.

Limitations and Future Work

  • Optimal Noise Determination: While grid search is used to find the optimal noise RMS, the paper suggests that more efficient optimization algorithms (e.g., genetic algorithms, particle swarm optimization) could be explored.
  • Noise Type: The study focuses on Gaussian noise. Investigating other noise types (e.g., generalized Gaussian, Lévy noise) for potential further improvements is a promising avenue.
  • Computational Cost: Achieving near-analog performance requires a large number of injected noise sequences (L). The paper notes that this might not be practical and suggests exploring optimal weight vectors to reduce L.
  • Real-World Experiments: The authors emphasize the need for real-world field experiments to validate and further refine the proposed algorithm, particularly for applications like microphone arrays.
  • Generalizability: The paper suggests that the noise-boosting concept could be applied to other DOA estimation algorithms (ESPRIT, root-MUSIC, etc.) and other array configurations (sparse arrays), which warrants further investigation.

Summary updated

BibTeX
@misc{pan2024doa,
  author = {Yan Pan and Li Zhang and Liyan Xu and Fabing Duan},
  title = {{DOA Estimation on One-Bit Quantization Observations through Noise-Boosted Multiple Signal Classification}},
  year = {2024},
  doi = {10.3390/s24144719},
}
Figure