We present a two-step method for computing the High-Resolution Spectrogram (HRS) of nonstationary signals. In the first step, a Weighted Short-Term Fourier Transform decomposes the signal using a bank of symmetric, tapered convolution operators, and the Hilbert transform extends each resulting component to its analytic form. In the second step, a single partial fraction decomposition is applied, corresponding to a first-order prediction error operator whose single pole is the first-order Burg reflection coefficient. A maximum-entropy autoregressive estimator then gives the instantaneous frequency of each locally monochromatic component within a sliding time window. The energy of each component is reassigned along the frequency axis to the pole location, producing sharply focused energy tracks and substantially improved time-frequency resolution. Summing all reassigned components reconstructs the original signal, which confirms invertibility and enables signal separation, denoising, and feature extraction. A systematic sensitivity analysis characterizes the method's valid operating range and parameter trade-offs, and quantitative evaluations show its robustness to white, colored, and impulsive noise across multiple SNR levels. Benchmarks against established high-resolution tools show competitive performance. The method is validated on synthetic signals, including a crossing chirp and a multi-component signal with stationary, frequency-modulated, swept, and impulsive constituents. It is then applied to a marine seismic trace and to the GW150914 gravitational wave signal recorded by the LIGO Hanford detector, demonstrating its effectiveness across a broad range of signal types and application domains.
In this paper, we develop two approaches for estimating the directions of arrival (DoAs) in hybrid analog/digital (HAD) receivers when the number of snapshots is too small for statistical averaging to be reliable. To the best of our knowledge, this is the only work to date that extracts DoAs in HAD receivers under stringent constraints on the number of available snapshots. To avoid averaging, the proposed approaches invoke the structure of the received signals in a matrix pencil method (MPM) framework. Unfortunately, the analog combiners at the front end of HAD receivers tangle the output signals, thereby hindering the application of the MPM framework. In addition to entanglement, these analog combiners effectively project the received signal onto a low-dimensional subspace, causing signals arriving from particular DoAs to be nullified or, at least, heavily attenuated. To eliminate this possibility, the proposed approaches cycle over an exhaustive set of analog combiners, which collectively span the entire space. The first approach can be applied to both fully-connected (FC) and partially-connected (PC) HADs and relies on the availability of periodic, potentially unknown, signals to disentangle the output of the HAD receiver. The second approach applies only to PC-HADs and eliminates the contingency on periodic signals by exploiting the block-diagonal structure underlying the PC-HAD architecture. Our numerical simulations and comparisons with the Cramér-Rao lower bound show that both approaches can perform reliably with as little as one snapshot per analog combiner, and that they can achieve a $97\%$ reduction in the root-mean-square error compared to their existing counterparts.
For testing underwater communication performance, model-driven channel simulators have high computational complexity, and the required dynamic environmental parameters are difficult to provide, making it challenging to generate realistic time-varying channels. In contrast, data-driven simulators do not suffer from these problems. Traditional underwater acoustic channel replay methods employ time-varying impulse responses; however, herein, a time-varying transfer function (TVTF) is employed to overcome the complexities associated with the large number of scattered acoustic paths in the time-delay domain. Singular value decomposition (SVD) is employed to implement stochastic acoustic channel replay by decomposing TVTF into principal and random components. A new random component can be generated randomly via the inverse process of the variable-parameter autoregressive (VPAR) model. To solve VPAR model, basis functions were used to translate frequency-dependent coefficients into constant coefficients, which can be estimated via a least square algorithm. SVD-VPAR replay (SVD-VPARR) method was proposed to generate realistic channel samples by inputting different complex noises into the inverse process of VPAR. The proposed simulator is validated using two measured channels from the WATERMARK benchmark and is compared with direct replay and an empirical mode decomposition-based stochastic replay method. Local time-frequency statistics and binary phase-shift keying / orthogonal frequency-division multiplexing communication results show that SVD-VPARR preserves the principal frequency-dependent channel characteristics and the major bit-error-rate variations with signal-to-noise ratio, time, and frequency.
Inverse rendering decomposes a scene into geometry, materials, and lighting to enable physically consistent view synthesis and relighting. Inverse rendering approaches built upon Gaussian Splatting benefit from its efficient reconstruction pipeline, yet they struggle in the presence of strong shadows. Shadows are often absorbed into albedo, leading to view synthesis in which the original shadows remain, while newly computed shadows are simply overlaid during relighting. To address this issue, we present a shadow-aware inverse rendering method for 2D Gaussian Splatting that explicitly disentangles illumination effects from surface reflectance. Our approach estimates pixelwise directional visibility in image space from observed geometry, enabling accurate modeling of cast shadows. We further introduce material segmentation, which groups shadowed and non-shadowed pixels of the same surface into a single material, enforcing consistent reflectance and preventing shadows from being baked into the albedo. Our method produces shadow-free albedo and models shadows as directional attenuation, enabling stable and high-fidelity relighting under novel illumination.
This paper proposes the Block Double Proportionate Sign Adaptive Filter (BDP-SAF) for sparse system identification in impulsive noise environments. The algorithm brings three mechanisms that are individually established in the adaptive filtering literature, namely block-level proportionate adaptation, zero-attraction and the sign-error update, and combines them with a single block-level method. The algorithm is derived from a regularized cost function combining $\ell _{2}$ and block-level $\ell _{1}$ penalty terms, which yields stationary-point expressions for a block proportionate gain matrix and a block zero-attractor; these are then reduced using approximations to a pair of low-cost implementable gains. The resulting update rule simultaneously accelerates adaptation in active blocks and suppresses inactive ones through a unified block energy estimate, while the sign-error update improves robustness to heavy-tailed disturbances under only a symmetry assumption on the noise distribution, without requiring knowledge of its higher-order moments or parametric form. To handle practical scenarios where the block partition is unavailable or misaligned with the true impulse response support, an adaptive block boundary detection scheme is developed that progressively realigns the partition online with negligible computational overhead. A convergence analysis establishes sufficient conditions for mean stability, and a semi-analytical steady-state mean-square deviation (MSD) expression is derived, together with an explicit statement of the assumptions on which both results rely. Simulation results are consistent with the theoretical findings and demonstrate that BDP-SAF achieves lower steady-state MSD with or without abrupt channel changes compared to the considered competing algorithms, including the case of several simultaneously active blocks.
