direction-of-arrival estimation

To address the challenge of detecting maneuvering small targets under low signal-to-noise ratio (SNR) conditions in mono-, bi-, and multistatic radar systems, this paper proposes a joint trajectory estimation and cross-coherent processing interval (CPI) long-term coherent integration method. By unifying the modeling of target motion, complex reflectivity, and clock offsets, the approach integrates extended Kalman filtering (or nonlinear optimization) with Neyman–Pearson detection to achieve time–frequency–phase coherence compensation under asynchronous multistatic configurations. Crucially, trajectory estimation is embedded directly into the long-term coherent integration framework for the first time, thereby overcoming conventional CPI duration constraints and enabling arbitrary-length cross-CPI integration. Simulation results demonstrate that the method successfully detects targets undetectable by conventional approaches under extremely low SNR, achieving significant improvements in detection probability and range estimation accuracy while maintaining a constant false alarm rate.

Conventional matrix pencil (MP) methods for direction-of-arrival (DoA) estimation in hybrid analog-digital (HAD) receivers suffer severe performance degradation under extremely limited snapshots (e.g., <10), primarily due to signal entanglement induced by analog combiners and SNR loss from low-dimensional projections. Method: To decouple entangled signals, we propose two MP extensions: (i) leveraging periodicity of source signals—applicable to both fully and partially connected HAD architectures; and (ii) modeling the analog combiner as a block-diagonal structure—removing the periodicity assumption and tailored for partially connected architectures. Both integrate analog combiner cyclic switching and signal subspace recovery. Contribution/Results: The proposed methods achieve high-accuracy DoA estimation with ultra-low snapshots, approaching the Cramér–Rao bound. Simulations demonstrate substantial gains over state-of-the-art techniques, effectively mitigating signal attenuation caused by low-dimensional projections.

This paper addresses the degradation of direction-of-arrival (DoA) estimation performance in uncalibrated arrays caused by hardware impairments—specifically, antenna position errors and channel complex gain mismatches. To tackle this, we propose a physically parameterized, fully differentiable MUSIC algorithm that embeds the array’s physical model into a differentiable spectral estimation framework. The method supports both supervised and unsupervised learning, enabling end-to-end joint estimation of DoAs and hardware errors. Its key innovation lies in the first-ever full-chain differentiability of MUSIC, integrating principles from physics-informed neural networks (PINNs) to perform gradient propagation and subspace optimization directly in the complex domain. Simulation results demonstrate that the proposed approach accurately recovers hardware imperfections and achieves significantly higher DoA estimation accuracy than classical MUSIC—particularly under low signal-to-noise ratios and severe hardware mismatches—while exhibiting superior robustness.

To address the dual bottlenecks of prohibitive interconnect overhead and soaring computational complexity in Extremely Large-Scale Antenna Array (ELAA) systems, this paper proposes a practically deployable distributed signal processing framework. We first establish a unified taxonomy for ELAA distributed processing, categorizing it into three paradigms: single-base-station ELAA, cooperative distributed antenna systems, and ELAA integrated with emerging technologies. The framework systematically incorporates distributed optimization, graph-model-driven antenna grouping, clustering-based processing, edge-cloud coordination, approximate message passing, and sparse reconstruction algorithms. We rigorously characterize the performance limits and operational conditions of each approach. Finally, we identify three key future research directions: scalability enhancement, low-latency coordination, and hardware-aware optimization. Our work provides a comprehensive, scalable, low-overhead, and robust signal processing design guideline for 6G air interfaces.

Speech recognition and speaker change detection in smart glasses degrade significantly in noisy environments. Method: This paper proposes a multi-microphone directional speech enhancement method tailored for wearable devices, jointly modeling neural beamforming and multichannel source separation within an end-to-end optimized framework integrating separation and automatic speech recognition (ASR). The approach employs a lightweight Conv-TasNet variant and a differentiable beamformer to enable efficient directional source separation on resource-constrained edge devices. Contribution/Results: Directional separation alone reduces the word error rate (WER) of wearer’s speech by 32%. Joint training further enhances robustness, achieving state-of-the-art ASR performance for smart glasses under real-world noise conditions. Crucially, this work presents the first empirical validation of end-to-end joint optimization of source separation and ASR for wearable audio applications.

This work addresses the challenges of source separation in distributed microphone arrays, where performance is hindered by cross-array permutation inconsistency and strong inter-array dependencies. To overcome these limitations, the authors propose a geometry-constrained decentralized independent vector analysis (Dec-IVA) method that leverages direction-of-arrival (DOA) information to align sources across arrays and introduces a weakly dependent source model to reduce inter-array coupling. By integrating geometric constraints to resolve permutation ambiguity and incorporating power-based statistical exchange to enhance robustness against noise, the proposed approach achieves improved separation accuracy and cross-array consistency. Experimental results demonstrate that the method significantly outperforms existing techniques in both separation performance and alignment reliability across distributed arrays.

This work addresses the limitations in degrees of freedom (DOF) and virtual aperture inherent in direction-of-arrival (DOA) estimation for noncircular signals. To overcome these constraints, a novel nested array configuration based on an extended coprime structure is proposed. By incorporating a sliding translation strategy to optimize sensor placement, the design achieves, for the first time, a redundancy-free fusion of sum and difference coarrays while preserving the continuity of the difference coarray. This integration effectively enlarges the virtual aperture and substantially enhances the achievable DOF. Simulation results demonstrate that the proposed array significantly outperforms conventional nested arrays and existing structures such as ESNA, yielding improved DOA estimation accuracy and an increased number of resolvable sources.

This study addresses the problem of computing the best rank-1 Hankel or Toeplitz approximation of an arbitrary matrix under both L₂ and L₁ norms, and applies this framework to small-sample direction-of-arrival (DoA) estimation. To this end, the authors propose an efficient and exact structured matrix factorization algorithm and derive maximum-likelihood-optimal DoA estimators under Gaussian and Laplacian noise assumptions, respectively. By integrating structured matrix approximation, norm-based optimization, and signal processing techniques, the proposed method achieves superior estimation accuracy and robustness in limited-data regimes. Extensive simulations and real-world experiments demonstrate its significant performance gains over existing approaches under low-snapshot conditions.

This study addresses the high computational complexity of conventional SRP-PHAT methods in three-dimensional direction-of-arrival (3D DOA) estimation, which stems from exhaustive searches over a dense grid of candidate directions and hinders real-time applicability. To overcome this limitation, the authors propose a two-stage strip-based search strategy: first, a coarse-to-fine search within azimuth strips identifies and retains multiple prominent peaks; then, elevation angles are refined along the corresponding great circles. By leveraging the higher reliability of azimuth estimates and integrating geometric search techniques—including spherical cap multi-peak retention, azimuth strip partitioning, and great-circle-based elevation refinement—the method achieves substantial computational savings while preserving estimation accuracy. Simulations and real-world experiments demonstrate that the proposed approach attains comparable precision to state-of-the-art 3D DOA estimators but with significantly improved computational efficiency.