Moment-Matching Array Processing Technique for diffuse source estimation

Direction of Arrival (DOA) estimation is a fundamental problem in signal processing. Diffuse sources, whose power density cannot be represented with a single angular coordinate, are usually characterized based on prior assumptions, which associate the source angular density with a specific set of functions. However, these assumptions can lead to significant estimation biases when they are incorrect. This paper introduces the Moment-Matching Estimation Technique (MoMET), a low-complexity method for estimating the mean DOA, spread, and power of a narrow diffuse source without requiring prior knowledge on the source distribution. The unknown source density is characterized by its mean DOA and its first central moments, which are estimated through covariance matching techniques which fit the empirical covariance of the measurements to that modeled from the moments. The MoMET parameterization is robust to incorrect model assumptions, and numerically efficient. The asymptotic bias and covariance of the new estimator are derived and its performance is demonstrated through simulations.

💡 Research Summary

This paper addresses the challenging problem of Direction‑of‑Arrival (DOA) estimation for diffuse (spatially extended) sources, where the source power density cannot be represented by a single angle. Conventional high‑resolution methods such as MUSIC or ESPRIT assume a small number of point‑like sources and therefore require an a‑priori model for the angular distribution of a diffuse source (e.g., Gaussian, uniform). When the assumed model is wrong, large estimation biases occur.

The authors propose the Moment‑Matching Estimation Technique (MoMET), a low‑complexity algorithm that estimates three key characteristics of a narrow diffuse source—its mean DOA (ω₀), its angular spread (σ_ω), and its total power (P)—without any prior knowledge of the shape of the source power density f(ω). The core idea is to parameterize f(ω) through its central moments rather than by a specific functional form. By normalizing f(ω) as f(ω)=P·σ_ω·p((ω−ω₀)/σ_ω), where p(x) is a standardized distribution with unit mass, zero mean and unit variance, the authors express the array covariance matrix as

 R = P·a(ω₀)a(ω₀)ᴴ ⊙ B + σ²_ε I,

where B_{k,l}=˜p((u_k−u_l)σ_ω) and ˜p(·) is the characteristic function of p(x). The characteristic function is expanded in a Taylor series around zero, retaining the first D+1 terms:

 ˜p(ξ)=∑_{d=0}^D j^d μ_d ξ^d / d!,

with μ₀=1, μ₁=0, μ₂=1 fixed and higher‑order moments μ₃…μ_D treated as unknowns. By defining ν_d = P·μ_d·σ_ω^d, all unknowns except ω₀ appear linearly in the model. This leads to a compact representation

 pR(θ) = a(ω₀)a(ω₀)ᴴ ⊙ pB(α) + σ²_ε I,

where α collects the linear parameters (including ν_d and σ²_ε) and pB(α)=∑ α_k A_k is a linear combination of known matrices A_k that depend only on the array geometry.

Estimation is performed within the COMET (Covariance Matching Estimation Technique) framework, i.e., a generalized least‑squares problem that minimizes

 J(θ)=½‖W^{½}( \bar R_N – pR(θ) )W^{½}‖_F²,

with \bar R_N the sample covariance and W a weighting matrix (W=I for ordinary LS, W=\bar R_N^{-1} yields asymptotic efficiency). For a fixed ω₀ the cost is quadratic in α, giving a closed‑form solution (23). The remaining scalar ω₀ is obtained by a one‑dimensional search over the criterion (24). The algorithm therefore consists of: (1) compute \bar R_N, (2) solve the 1‑D optimization for ω₀, (3) compute α via the closed‑form expression, and (4) recover P and σ_ω from α.

The authors derive the asymptotic bias and covariance of the MoMET estimator, showing that as the Taylor order D increases the model bias diminishes and the estimator approaches the Cramér‑Rao bound, while the variance grows modestly due to the estimation of additional moments. This trade‑off is quantified analytically and validated numerically.

Simulation results cover uniform linear arrays (M=10) and non‑uniform arrays, and test three representative power densities: Gaussian, uniform, and exponential. MoMET is benchmarked against weighted pseudo‑subspace fitting (WPSF‑MUSIC) and other moment‑based subspace methods. In all cases, especially when the assumed source model is misspecified, MoMET yields lower root‑mean‑square error (RMSE) for the mean DOA (2–3 dB improvement) and markedly smaller bias for the spread estimate. Computationally, MoMET avoids eigen‑decomposition; its dominant cost is solving a linear system and performing a scalar search, leading to O(M²) complexity versus O(M³) for conventional subspace techniques.

In conclusion, MoMET offers (i) a distribution‑agnostic model for diffuse sources, (ii) a moment‑based linear‑nonlinear parameterization that enables efficient covariance matching, and (iii) explicit control over the bias‑variance trade‑off via the Taylor order D. Limitations include the need for sufficient snapshots to reliably estimate higher‑order moments and reduced accuracy for extremely wide spreads where low‑order expansions become inadequate. Future work is suggested on adaptive order selection, extension to multiple simultaneous diffuse sources, and robustness to colored or non‑Gaussian noise.