Low-complexity direction-of-arrival (DOA) estimation for large-scale antenna arrays requires dimensionality reduction without losing full field-of-view (FOV) identifiability and large-aperture advantage. A novel subarray beamspace DOA estimation architecture is proposed in this paper. For equally-sized uniform subarrays, a two-stage unambiguous DOA estimation algorithm, termed U-ES-DOA, is developed. Specifically, the intra-subarray beamspace provides full-FOV angular information, while the inter-subarray convolutional beamspace exploits the aperture across subarrays for high-resolution estimation. To improve robustness against coherent sources, a beam smoothing technique is incorporated, resulting in the S-U-ES-DOA algorithm. For non-equally-sized uniform subarrays, theoretical conditions on subarray sizes are derived to enable direct unambiguous DOA estimation in the inter-subarray convolutional beamspace, leading to the U-NES-DOA algorithm. The number of intra-subarray beams can be flexibly selected to balance estimation performance and computational complexity, and a single wide beam per subarray is sufficient for unambiguous wide-FOV multi-target DOA estimation only by the processing of inter-subarray convolutional beamspace. Simulations verify the advantages of the proposed architecture in ambiguity removal, coherent-source resolvability, resolution, estimation accuracy, and computational efficiency compared with conventional methods.
Accurate emitter localization is critical for civil and industrial applications, yet conventional two-step methods suffer from suboptimality under challenging signal conditions. While direct position determination (DPD) offers superior performance by jointly processing raw sensor signals, existing approaches assume reliable fusion center (FC) measurement with perfect sensor-to-FC transmission, an impractical assumption in resource-constrained or contested environments where the measurements collected in FC are subject to bit errors. Focusing on LOS-dominated narrowband localization scenarios, where strong multipath components are absent or not resolvable, this paper pioneers robust DPD methodologies specifically designed for contaminated signal scenarios.We first establish a novel statistical framework modeling composite errors (sensor noise + transmission artifacts) through multivariate Student’s t-distributions, capturing heavy-tailed characteristics overlooked by Gaussian assumptions. Building on this foundation, we develop two computationally efficient algorithms: For known waveforms, a three-stage estimator integrates adaptive measurement truncation with closed-form parameter solutions; for unknown waveforms, an iterative subspace projection technique maintains robustness against signal ambiguity. Both algorithms dynamically suppress outliers without prior error characterization. Extensive numerical validations demonstrate the accuracy improvements over classical and modern benchmarks across diverse SNR regimes, with particular efficacy in low-SNR scenarios. Furthermore, the proposed solutions maintain consistent performance within 5-20% truncation ranges, accommodating unknown bit error rates in practical deployments.
Uplink sensing in perceptive mobile networks (PMNs), where the user equipment transmits and the base station receives, represents an attractive and important type of integrated sensing and communication (ISAC). However, its bi-static configuration introduces clock asynchrony that leads to time-varying timing offset (TO) and composite phase offset (CPO), which disrupts phase coherence, thereby invalidating conventional sensing algorithms that rely on synchronization assumptions. Moreover, even with phase coherence restored, the delay ambiguity induced by the global TO cannot be fully resolved without prior relative location information, thereby hindering accurate sensing. To address these fundamental challenges, this paper proposes a novel three-step prior-free sensing framework. First, we develop two closed-form subspace-based high-resolution relative TO (RTO) estimators with respect to a reference frame, which enable TO alignment in both line-of-sight (LoS)-dominant and LoS-absent scenarios. Then, we propose a subspace-based relative CPO (RCPO) estimation method, which establishes an analytical relationship between RCPO and the signal subspace and solves the resulting constant-modulus constrained optimization problem efficiently using Riemannian gradient descent (RGD). Finally, leveraging the estimated delay, AoA, and AoD, we develop a geometric-consistency-based association method that resolves the delay ambiguity and establishes parameter associations without requiring prior information. Our simulations demonstrate that the proposed methods robustly deliver superior accuracy even under low signal-to-noise ratio (SNR) conditions.
Reconfigurable intelligent surface (RIS)-aided integrated sensing and communications (ISAC) systems are promising for next-generation wireless networks, enabling high spectral efficiency and seamless coverage. However, large-scale deployment is constrained by the complexity and cost of high-precision hardware components. Although low-resolution architectures provide a cost-effective alternative, existing studies lack a comprehensive system-level design tailored to low-resolution RIS-aided ISAC. This paper proposes a full-chain low-resolution framework that jointly optimizes the transmit waveform, receive beamforming, and RIS reflection coefficients under low-resolution DAC/ADC constraints and discrete phase-shifter limitations. A closed-form expression for the signal-to-quantized-clutter-plus-noise ratio (SQCNR) is derived, and the symbol error rate (SER) of symbol-level precoding (SLP) with PSK/QAM modulation is analyzed. Both sensing and communication metrics are then unified into a joint optimization framework that explicitly balances sensing and data transmission performance. An Alternating Projection (AP) strategy is employed to decouple the transmit waveform and RIS phase-shift design, and the resulting subproblems are solved via Majorization–Minimization using convex surrogate relaxations. Simulation results validate that the proposed framework achieves superior joint sensing–communication performance compared with existing benchmark schemes.
This paper studies the problem of blindly recovering the global centrality vectors using partially observed graph signals taken from multiple graphs. Our setting is motivated by practical scenarios where simultaneous access to all nodes can be infeasible, while we aim to infer the global centrality vector. We show that the multi-graph observation model enables blind global centrality recovery. Assuming that the partial graph signal observations are generated by sparse and nonnegative external sources filtered through low-pass graph filters, we propose to estimate the eigen-centrality vector of each graph by integrating sparse non-negative matrix factorization (NMF) and robust principal component analysis (RPCA), and an alternating optimization strategy is shown to converge to a stationary solution. Lastly, we show that a necessary condition for identifiability depends on whether the number of external sources, i.e., the rank of observed graph signal data, exceeds the number of hidden nodes. Numerical experiments support our findings.
