TL;DR
This paper introduces a novel, non-iterative blind calibration method for 2-level nested sensor arrays. It demonstrates that a Least-Squares (LS) approach, contrary to some prior claims, is effective for these sparse arrays, yielding optimally-weighted estimates of sensor gains and phases. This method significantly improves calibration accuracy and computational efficiency compared to existing state-of-the-art techniques.
What Problem Does This Paper Address?
Accurate sensor calibration is crucial for many array processing tasks, such as determining the direction of arrival (DOA) of signals. "Blind calibration" is particularly desirable because it doesn't require known calibration signals, making it more practical in real-world scenarios. However, blind calibration is challenging, especially for modern "sparse" array structures like nested arrays, which offer advantages in terms of the number of detectable sources versus the number of physical sensors. While methods exist for simpler array types, blind calibration for nested arrays has been less explored, with some literature even suggesting that standard LS approaches are inapplicable. This paper aims to fill this gap by providing an effective and efficient blind calibration method for 2-level nested arrays.
Methods
The core of the paper's contribution lies in developing an Optimally-Weighted Least-Squares (OWLS) approach for blind calibration. The authors leverage the specific "partially-Toeplitz" structure of the covariance matrix of signals received by a nested array. By applying a log transformation to the covariance matrix elements, they derive a system of linear equations relating the unknown sensor gains and phases to observable quantities.
Key methodological points include:
- Revisiting the LS Approach: The paper directly challenges claims that LS is unsuitable for nested arrays, showing its applicability when the array geometry is chosen appropriately (specifically, when the inter-element spacing of the second ULA is equal to that of the first, i.e., $L=N_1$).
- Linear System Formulation: They transform the problem into solving a linear system of equations by extracting real and imaginary parts of the log-transformed covariance elements. This system relates the unknown parameters (gains and phases) to the observed data.
- Optimally-Weighted LS (OWLS): Instead of using a simple Ordinary Least Squares (OLS) estimate, which can be suboptimal due to estimation errors in the covariance matrix, the authors derive an OWLS estimate. This involves accounting for the statistical properties of these estimation errors to achieve more accurate results. They use a "Maximum Likelihood (ML)-based OWLS" estimate by approximating the covariance of the noise terms in their linear system.
- Handling Degeneracies: The method addresses inherent ambiguities in blind calibration by setting reference sensor gains and phases to zero, a standard practice.
Datasets
Not discussed in the paper. The evaluation is performed using simulated data.
Key Insights and Ideas
- LS is Applicable to Nested Arrays: A significant insight is the demonstration that the Least-Squares approach can be used for blind calibration of 2-level nested arrays, contrary to some prior literature. This opens up simpler and more efficient calibration pathways.
- Geometry Matters for LS: The paper highlights that a specific geometric configuration of the nested array (where $L=N_1$) is crucial for enabling a unique LS-based blind calibration solution.
- OWLS for Improved Accuracy: The core innovation is the OWLS approach, which goes beyond basic LS by incorporating statistical information about estimation errors. This leads to a more robust and accurate estimation of sensor parameters.
- Coupling of Gain and Phase Errors: Contrary to some assumptions, the authors show that the noise terms in the gain and phase estimation equations are statistically coupled, necessitating a joint estimation approach for optimal results.
- Computational Efficiency: The non-iterative, closed-form nature of the OWLS solution results in dramatic speedups compared to iterative or optimization-based methods.
Key Results
- Superior Calibration Accuracy: Simulations show that the proposed ML-OWLS method significantly outperforms the state-of-the-art Semidefinite Programming (SDP)-based method ([14]) in terms of the Mean Squared Error (MSE) of estimated sensor gains and phases. For example, Figure 2 shows substantial MSE reductions.
- Enhanced DOA Estimation: The improved calibration accuracy directly translates to more accurate Direction-of-Arrival (DOA) estimation using standard algorithms like Spatial Smoothing MUSIC (SS-MUSIC), as demonstrated in Figure 3.
- Orders of Magnitude Speedup: Table 1 shows that the proposed method is dramatically faster than the SDP approach, with average runtimes of 0.0032 seconds compared to 1.616 seconds for an 8-sensor array, representing a reduction by orders of magnitude.
Limitations and Future Work
- 2-Level Nested Arrays: The current work focuses specifically on 2-level nested arrays.
- Gaussian Signals: While the method is derived under the assumption of Gaussian signals, the authors note that it can still provide significant performance improvements for non-Gaussian signals with analytical guarantees.
- Future Work: The authors explicitly state that their proposed calibration scheme can be extended to K-level nested arrays ($K > 2$), and these extensions will be presented in future work. They also mention that the concepts can be applied to non-Gaussian signals.
Summary updated