DOA estimation exploiting coprime frequencies

Si Qin, Yimin D. Zhang, Moeness G. AminProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2014 · DOI

TL;DR

This paper introduces a novel approach to direction-of-arrival (DOA) estimation by leveraging a sparse uniform linear array (ULA) with two coprime frequencies. This method effectively creates a "coprime array" structure, significantly increasing the number of detectable signals compared to traditional ULAs, and the authors analyze its performance and limitations.

What Problem Does This Paper Address?

Estimating the direction of arrival (DOA) of multiple signals is a fundamental problem in array signal processing. Traditional uniform linear arrays (ULAs) have a limited number of "degrees of freedom" (DOFs), meaning they can only resolve a number of sources equal to one less than the number of sensors. To overcome this, researchers have explored sparse array configurations that create a larger "difference co-array" with virtual sensor positions. However, designing optimal sparse arrays like Minimum Redundancy Arrays (MRAs) can be challenging. This paper addresses the need for a systematic way to achieve high DOFs for DOA estimation, proposing an alternative to physical sparse arrays by using a single ULA with carefully chosen coprime frequencies.

Methods

The core idea is to use a single ULA with two distinct frequencies, $f_1$ and $f_2$, that are coprime (their ratio is a ratio of coprime integers). This setup effectively creates two virtual subarrays with different spatial sampling intervals, mimicking the structure of a physical coprime array. The paper derives the analytical expression for the number of unique "lags" in the resulting difference co-array, which directly relates to the achievable DOFs. DOA estimation is then framed as a sparse signal reconstruction problem, specifically using compressive sensing (CS) techniques. The authors explore two scenarios: one using only cross-correlations between the virtual subarrays (yielding maximum DOFs) and another that also incorporates self-correlations (improving estimation performance). They employ group Lasso for DOA estimation when both self- and cross-correlations are used, recognizing the shared sparsity support across different signal components.

Datasets

Not discussed in the paper. The evaluation is based on simulated data.

Key Insights and Ideas

  • Frequency as a Spatial Tool: The paper demonstrates that by selecting coprime frequencies, a single ULA can be transformed into a sparse array with enhanced DOA estimation capabilities, offering a flexible alternative to physically sparse arrays.
  • Coprime Frequencies Unlock DOFs: The number of unique lags in the difference co-array, and thus the number of resolvable sources, is directly tied to the coprimality of the frequencies and the number of sensors.
  • Co-array Structure is Key: The analysis hinges on understanding the structure of the "difference co-array" formed by the lag differences between virtual sensors. The number of unique lags in this co-array dictates the maximum number of DOFs.
  • Self- and Cross-Correlations Synergy: While cross-correlations alone maximize DOFs, combining them with self-correlations (from each virtual subarray) can improve estimation accuracy, albeit requiring group sparsity techniques.

Key Results

  • The paper provides analytical expressions for the number of unique lags (DOFs) in the co-array, showing it depends on the number of sensors ($L$) and the coprime integers ($M_1, M_2$) defining the frequency ratio.
  • For a ULA with $L=6$ sensors, using coprime frequencies ($M_1=5, M_2=6$) allowed resolving 19 sources, whereas non-coprime frequencies ($M_1=4, M_2=6$) could only resolve fewer than 19 sources.
  • Simulations show that using both self- and cross-lags leads to improved DOA estimation performance compared to using only cross-lags.

Limitations and Future Work

  • Alias-Free Condition: The paper notes that a sufficient condition for alias-free DOA estimation is achieved by choosing coprime frequencies. However, the analysis of achievable DOFs is presented for cases where $M_2 > L$, and the case where $M_2 < L$ (which can lead to redundant lags) is not extensively considered.
  • Sidelobe Levels: The paper mentions that a very large $M_2$ might lead to high sidelobe levels in the co-array's point spread function, which could impact estimation performance.
  • Future Work: The paper implicitly suggests further exploration into optimizing the choice of coprime frequencies and sensor configurations to balance DOF maximization with practical considerations like sidelobe levels and frequency separation. The use of group Lasso for joint self- and cross-lag processing is a key algorithmic contribution that could be further investigated.

Summary updated

Abstract

Coprime array, which utilizes a coprime pair of uniform linear subarrays, is an attractive structure to achieve sparse array configurations. Alternatively, effective coprime array configurations can be implemented using a uniform linear array with two coprime sensing frequencies. This enables the integration of the coprime array and filter concepts to achieve high capabilities in meeting system performance and complexity constraints. This paper examines its performance for direction-of-arrival estimations. In particular, we analyze the number of detectable signals and the estimation accuracy as related to the array configurations and sensing frequencies.
BibTeX
@inproceedings{qin2014doa,
  author = {Si Qin and Yimin D. Zhang and Moeness G. Amin},
  title = {{DOA estimation exploiting coprime frequencies}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2014},
  doi = {10.1117/12.2051000},
}
Figure