Papers

This paper proposes an encrypted state observer that is capable of detecting sensor attacks without decryption. We first design a state observer that operates over a finite field of integers with the modular arithmetic. The observer generates a residue signal that indicates the presence of attacks under sparse attack and sensing redundancy conditions. Then, we develop a homomorphic encryption scheme that enables the observer to operate over encrypted data while automatically disclosing the residue signal. Unlike our previous work restricted to single-input single-output systems, the proposed scheme is applicable to general multi-input multi-output systems. Given that the disclosed residue signal remains below a prescribed threshold, the full state can be recovered as an encrypted message.

Lightweight, real-time text-to-speech systems are crucial for accessibility. However, the most efficient TTS models often rely on lightweight phonemizers that struggle with context-dependent challenges. In contrast, more advanced phonemizers with a deeper linguistic understanding typically incur high computational costs, which prevents real-time performance. This paper examines the trade-off between phonemization quality and inference speed in G2P-aided TTS systems, introducing a practical framework to bridge this gap. We propose lightweight strategies for context-aware phonemization and a service-oriented TTS architecture that executes these modules as independent services. This design decouples heavy context-aware components from the core TTS engine, effectively breaking the latency barrier and enabling real-time use of high-quality phonemization models. Experimental results confirm that the proposed system improves pronunciation soundness and linguistic accuracy while maintaining real-time responsiveness, making it well-suited for offline and end-device TTS applications.

The Advanced Light Source (ALS) at Lawrence Berkeley National Laboratory, a third-generation synchrotron light source operational since 1992, is undergoing a comprehensive upgrade of its storage ring RF control system. The legacy Horner PLC controllers and remote I/O modules, now at end-of-life, are being replaced with an Allen-Bradley PLC platform to improve maintainability, reliability, and long-term support. This paper presents the planning, design, and current status of the upgrade project.

Due the complexity of modern power systems, modeling based on first-order principles becomes increasingly difficult. As an alternative, dynamical models for simulation and control design can be obtained by black-box identification techniques. One such technique for the identification of continuous-time systems are neural ordinary differential equations. For training and inference, they require initial values of system states, such as phase angles and frequencies. While frequencies can typically be measured, phase angle measurements are usually not available. To tackle this problem, we propose a novel structure based on augmented neural ordinary differential equations, learning latent phase angle representations on historic observations with temporal convolutional networks. Our approach combines state-of-the art deep learning techniques, avoiding the necessity of phase angle information for the power system identification. Results show, that our approach clearly outperforms simpler augmentation techniques.

We present an efficient algorithm to compute the explicit open-loop solution to both finite and infinite-horizon dynamic games subject to state and input constraints. Our approach relies on a multiparametric affine variational inequality characterization of the open-loop Nash equilibria and extends the classical explicit constrained LQR and MPC frameworks to multi-agent non-cooperative settings. A key practical implication is that linear-quadratic game-theoretic MPC becomes viable even at very high sampling rates for multi-agent systems of moderate size. Extensive numerical experiments demonstrate order-of-magnitude improvements in online computation time and solution accuracy compared with state-of-the-art game-theoretic solvers.

Piezoelectric micromachined ultrasonic transducers (PMUTs) are widely utilized in applications that demand mechanical resilience, thermal stability, and compact form factors. Recent efforts have sought to demonstrate that single-crystal lithium niobate (LN) is a promising PMUT material platform, offering high electromechanical coupling (k2) and bidirectional performance. In addition, advances in LN film transfer technology have enabled high quality periodically poled piezoelectric films (P3F), facilitating a bimorph piezoelectric stack without intermediate electrodes. In this work, we showcase a bimorph PMUT incorporating a mechanically robust, 20 $\mu$m thick P3F LN active layer. We establish the motivation for LN PMUTs through a material comparison, followed by extensive membrane geometry optimization and subsequent enhancement of the PMUT's k2. We demonstrate a 775 kHz flexural mode device with a quality factor (Q) of 200 and an extracted k2 of 6.4%, yielding a high transmit efficiency of 65 nm/V with a mechanically robust active layer. We leverage the high performance to demonstrate extreme-temperature resilience, showcasing stable device operation up to 600 $^\circ$C and survival up to 900 $^\circ$C, highlighting LN's potential as a resilient PMUT platform.

Given the importance of monitoring the operational status of high-voltage cables in coal mines, this study investigates the application of intelligent sensing technology to the online monitoring of such cables. Taking an actual coal mine as a case study, a three-layer architecture high-voltage cable monitoring system was designed. The system employs high-frequency current sensors and distributed optical fiber temperature sensors to achieve real-time acquisition of partial discharge signals and temperature distribution data. Data analysis and fault diagnosis are performed through a combined approach of edge computing and cloud computing. The research results demonstrate that the system can accurately identify cable insulation defects and potential overheating hazards, with a diagnostic accuracy exceeding 95%, thereby significantly enhancing the reliability of power supply in mines.

The standard linear quadratic Gaussian (LQG) framework assumes a Brownian noise process and relies on classical stochastic calculus tools, such as those based on Itô calculus. In this paper, we solve a generalized linear quadratic optimal control problem where the process and measurement noises can be non-Markovian and non-semimartingale stochastic processes with sample paths that have low Hölder regularity. Since these noise models do not, in general, permit the use of the standard Itô calculus, we employ rough path theory to formulate and solve the problem. By leveraging signature representations and controlled rough paths, we derive the optimal state estimation and control strategies.

A segmented waveguide-enabled pinching-antenna system (SWAN)-assisted integrated sensing and communications (ISAC) framework is proposed. Unlike conventional pinching antenna systems (PASS), which use a single long waveguide, SWAN divides the waveguide into multiple short segments, each with a dedicated feed point. Thanks to the segmented structure, SWAN enhances sensing performance by significantly simplifying the reception model and reducing the in-waveguide propagation loss. To balance performance and complexity, three segment controlling protocols are proposed for the transceivers, namely i) segment selection to select a single segment for signal transceiving, ii) segment aggregation to aggregate signals from all segments using a single RF chain, and iii) segment multiplexing to jointly process the signals from all segments using individual RF chains. The theoretical sensing performance limit is first analyzed for different protocols, unveiling how the sensing performance gain of SWAN scales with the number of segments. Based on this performance limit, the Pareto fronts of sensing and communication performance are characterized for the simple one-user one-target case, which is then extended to the general multi-user single-target case based on time-division multiple access (TDMA). Numerical results are presented to verify the correctness of the derivations and the effectiveness of the proposed algorithms, which jointly confirm the advantages of SWAN-assisted ISAC.

The conventional Artificial Potential Field (APF) is fundamentally limited by the local minima issue and its inability to account for the kinematics of moving obstacles. This paper addresses the critical challenge of autonomous collision avoidance for Unmanned Aerial Vehicles (UAVs) operating in dynamic and cluttered airspace by proposing a novel Direction and Relative Velocity Weighted Artificial Potential Field (APF). In this approach, a bounded weighting function, $\omega(\theta,v_{e})$, is introduced to dynamically scale the repulsive potential based on the direction and velocity of the obstacle relative to the UAV. This robust APF formulation is integrated within a Model Predictive Control (MPC) framework to generate collision-free trajectories while adhering to kinematic constraints. Simulation results demonstrate that the proposed method effectively resolves local minima and significantly enhances safety by enabling smooth, predictive avoidance maneuvers. The system ensures superior path integrity and reliable performance, confirming its viability for autonomous navigation in complex environments.

Optimal pulse patterns (OPPs) are a modulation method in which the switching angles and levels of a switching signal are computed via an offline optimization procedure to minimize a performance metric, typically the harmonic distortions of the load current. Additional constraints can be incorporated into the optimization problem to achieve secondary objectives, such as the limitation of specific harmonics or the reduction of power converter losses. The resulting optimization problem, however, is highly nonconvex, featuring a trigonometric objective function and constraints as well as both real- and integer-valued optimization variables. This work casts the task of OPP synthesis for a multilevel converter as an optimal control problem of a hybrid system. This problem is in turn lifted into a convex but infinite-dimensional conic program of occupation measures using established methods in convex relaxations of optimal control. Lower bounds on the minimum achievable harmonic distortion are acquired by solving a sequence of semidefinite programs via the moment-sum-of-squares hierarchy, where each semidefinite program scales in a jointly linear manner with the numbers of permitted switching transitions and converter voltage levels.

Accurate segmentation of spinal structures in X-ray images is a prerequisite for quantitative scoliosis assessment, including Cobb angle measurement, vertebral translation estimation and curvature classification. In routine practice, clinicians acquire coronal, left-bending and right-bending radiographs to jointly evaluate deformity severity and spinal flexibility. However, the segmentation step remains heavily manual, time-consuming and non-reproducible, particularly in low-contrast images and in the presence of rib shadows or overlapping tissues. To address these limitations, this paper proposes R2MF-Net, a recurrent residual multi-path encoder–decoder network tailored for automatic segmentation of multi-directional spine X-ray images. The overall design consists of a coarse segmentation network and a fine segmentation network connected in cascade. Both stages adopt an improved Inception-style multi-branch feature extractor, while a recurrent residual jump connection (R2-Jump) module is inserted into skip paths to gradually align encoder and decoder semantics. A multi-scale cross-stage skip (MC-Skip) mechanism allows the fine network to reuse hierarchical representations from multiple decoder levels of the coarse network, thereby strengthening the stability of segmentation across imaging directions and contrast conditions. Furthermore, a lightweight spatial-channel squeeze-and-excitation block (SCSE-Lite) is employed at the bottleneck to emphasize spine-related activations and suppress irrelevant structures and background noise. We evaluate R2MF-Net on a clinical multi-view radiograph dataset comprising 228 sets of coronal, left-bending and right-bending spine X-ray images with expert annotations.

Accurate three-dimensional delineation of liver tumors on contrast-enhanced CT is a prerequisite for treatment planning, navigation and response assessment, yet manual contouring is slow, observer-dependent and difficult to standardise across centres. Automatic segmentation is complicated by low lesion-parenchyma contrast, blurred or incomplete boundaries, heterogeneous enhancement patterns, and confounding structures such as vessels and adjacent organs. We propose a hybrid framework that couples an attention-enhanced cascaded U-Net with handcrafted radiomics and voxel-wise 3D CNN refinement for joint liver and liver-tumor segmentation. First, a 2.5D two-stage network with a densely connected encoder, sub-pixel convolution decoders and multi-scale attention gates produces initial liver and tumor probability maps from short stacks of axial slices. Inter-slice temporal consistency is then enforced by a simple three-slice refinement rule along the cranio-caudal direction, which restores thin and tiny lesions while suppressing isolated noise. Next, 728 radiomic descriptors spanning intensity, texture, shape, boundary and wavelet feature groups are extracted from candidate lesions and reduced to 20 stable, highly informative features via multi-strategy feature selection; a random forest classifier uses these features to reject false-positive regions. Finally, a compact 3D patch-based CNN derived from AlexNet operates in a narrow band around the tumor boundary to perform voxel-level relabelling and contour smoothing.

The present document is Part 1 of a 2-part introduction to ambisonics and aims at readers who would like to work practically with ambisonics. We leave out deep technical details in this part and focus on helping the reader to develop an intuitive understanding of the underlying concept. We explain what ambisonic signals are, how they can be obtained, what manipulations can be applied to them, and how they can be reproduced to a listener. We provide a variety of audio examples that illustrate the matter. Part 2 of this introduction into ambisonics is provided in a separate document and aims at readers who would like to understand the mathematical details.

Cross-modal learning has become a fundamental paradigm for integrating heterogeneous information sources such as images, text, and structured attributes. However, multimodal representations often suffer from modality dominance, redundant information coupling, and spurious cross-modal correlations, leading to suboptimal generalization and limited interpretability. In particular, high-variance modalities tend to overshadow weaker but semantically important signals, while naïve fusion strategies entangle modality-shared and modality-specific factors in an uncontrolled manner. This makes it difficult to understand which modality actually drives a prediction and to maintain robustness when some modalities are noisy or missing. To address these challenges, we propose a Dual-Stream Residual Semantic Decorrelation Network (DSRSD-Net), a simple yet effective framework that disentangles modality-specific and modality-shared information through residual decomposition and explicit semantic decorrelation constraints. DSRSD-Net introduces: (1) a dual-stream representation learning module that separates intra-modal (private) and inter-modal (shared) latent factors via residual projection; (2) a residual semantic alignment head that maps shared factors from different modalities into a common space using a combination of contrastive and regression-style objectives; and (3) a decorrelation and orthogonality loss that regularizes the covariance structure of the shared space while enforcing orthogonality between shared and private streams, thereby suppressing cross-modal redundancy and preventing feature collapse. Experimental results on two large-scale educational benchmarks demonstrate that DSRSD-Net consistently improves next-step prediction and final outcome prediction over strong single-modality, early-fusion, late-fusion, and co-attention baselines.

Estimation of the conditional independence graph (CIG) of high-dimensional multivariate Gaussian time series from multi-attribute data is considered. Existing methods for graph estimation for such data are based on single-attribute models where one associates a scalar time series with each node. In multi-attribute graphical models, each node represents a random vector or vector time series. In this paper we provide a unified theoretical analysis of multi-attribute graph learning for dependent time series using a penalized log-likelihood objective function formulated in the frequency domain using the discrete Fourier transform of the time-domain data. We consider both convex (sparse-group lasso) and non-convex (log-sum and SCAD group penalties) penalty/regularization functions. We establish sufficient conditions in a high-dimensional setting for consistency (convergence of the inverse power spectral density to true value in the Frobenius norm), local convexity when using non-convex penalties, and graph recovery. We do not impose any incoherence or irrepresentability condition for our convergence results. We also empirically investigate selection of the tuning parameters based on the Bayesian information criterion, and illustrate our approach using numerical examples utilizing both synthetic and real data.

In this paper, we propose a data-driven robust safety verification framework for stochastic dynamical systems modeled as Markov decision processes with time-varying and uncertain transition probabilities. Rather than assuming access to the exact nominal transition kernel, we consider the realistic setting where only samples from multiple system executions are available. These samples may correspond to different transition models inside an ambiguity set around the nominal transition kernel. Using these observations, we construct a unified ambiguity set that captures both inherent run-to-run variability in the transition dynamics and finite-sample statistical uncertainty. This ambiguity set is formalized through a Wasserstein-distance ball around a nominal empirical distribution and naturally induces an interval Markov decision process representation of the underlying system. Within this representation, we introduce a robust safety function that characterizes reach-avoid type probabilistic safety under all transition kernels consistent with the interval Markov decision process. We further derive high-confidence safety guarantees for the true, unknown time-varying system. A numerical example illustrates the applicability and effectiveness of the proposed approach.

Global Navigation Satellite System (GNSS) signals are fundamental in applications across navigation, transportation, and industrial networks. However, their extremely low received power makes them highly vulnerable to radio-frequency interference (RFI) and intentional jamming. Modern data-driven methods offer powerful representational power for such applications, however real-time and reliable jamming detection on resource-limited embedded receivers remains a key challenge due to the high computational and memory demands of the conventional learning paradigm. To address these challenges, this work presents a dictionary-based contrastive learning (DBCL) framework for GNSS jamming detection that integrates transfer learning, contrastive representation learning, and model compression techniques. The framework combines tuned contrastive and dictionary-based loss functions to enhance feature separability under low-data conditions and applies structured pruning and knowledge distillation to reduce model complexity while maintaining high accuracy. Extensive evaluation across varying data regimes demonstrate that the proposed algorithm consistently outperforms modern CNN, MobileViT, and ResNet-18 architectures. The framework achieves a substantial reduction in memory footprint and inference latency, confirming its suitability for real-time, low-power GNSS interference detection on embedded platforms.

Electrical brain stimulation relies on externally applied currents to modulate neural activity, but safety constraints require each stimulation cycle to be charge-balanced, enforcing a zero net injected charge. However, how such charge-balanced stimulation works remains poorly understood. This paper investigates the ability of charge-balanced inputs to steer state trajectories in discrete-time linear systems. Motivated by both open-loop and adaptive neurostimulation protocols, we study two practically relevant input structures: periodic (repetitive) charge-balanced inputs and non-repetitive charge-balanced inputs. For each case, we derive novel reachability and controllability conditions. The theoretical results are further validated through numerical demonstrations of minimum-energy control input design.

This work proposes a novel distributed framework for verifying the incremental stability of large-scale systems with unknown dynamics and known interconnection structures using graph neural networks. Our proposed approach relies on the construction of local incremental Lyapunov functions for subsystems, which are then composed together to obtain a suitable Lyapunov function for the interconnected system. Graph neural networks are used to synthesize these functions in a data-driven fashion. The formal correctness guarantee is then obtained by leveraging Lipschitz bounds of the trained neural networks. Finally, the effectiveness of our approach is validated through two nonlinear case studies.

This paper investigates the impact of Reconfigurable Intelligent Surface (RIS) orientation on the throughput performance of Unmanned Aerial Vehicle (UAV)-assisted wireless communication systems. Specifically, we study how physical rotation of the RIS, through controlled azimuth and elevation adjustments, influences the effective channel and data rate. A UAV-mounted RIS enables directional alignment to serve ground users in scenarios where the direct Base Station (BS)-to-user path is blocked. Using the SimRIS channel simulator, we analyze the system under various rotation angles and present performance heatmaps that highlight optimal RIS orientations. The study shows that RIS alignment has a substantial effect on achievable rates, thereby motivating orientation-aware optimization in practical deployments.

We study the orbit recovery problem under the rigid-motion group SE(n), where the objective is to reconstruct an unknown signal from multiple noisy observations subjected to unknown rotations and translations. This problem is fundamental in signal processing, computer vision, and structural biology. Our main theoretical contribution is bounding the sample complexity of this problem. We show that if the d-th order moment under the rotation group SO(n) uniquely determines the signal orbit, then orbit recovery under SE(n) is achievable with $N\gtrsim \sigma^{2d+4}$ samples as the noise variance $\sigma^2 \to \infty$. The key technical insight is that the d-th order SO(n) moments can be explicitly recovered from (d+2)-order SE(n) autocorrelations, enabling us to transfer known results from the rotation-only setting to the rigid-motion case. We further harness this result to derive a matching bound to the sample complexity of the multi-target detection model that serves as an abstract framework for electron-microscopy-based technologies in structural biology, such as single-particle cryo-electron microscopy (cryo-EM) and cryo-electron tomography (cryo-ET). Beyond theory, we present a provable computational pipeline for rigid-motion orbit recovery in three dimensions. Starting from rigid-motion autocorrelations, we extract the SO(3) moments and demonstrate successful reconstruction of a 3-D macromolecular structure. Importantly, this algorithmic approach is valid at any noise level, suggesting that even very small macromolecules, long believed to be inaccessible using structural biology electron-microscopy-based technologies, may, in principle, be reconstructed given sufficient data.

We consider the problem of recovering an unknown low-dimensional vector from noisy, underdetermined observations. We focus on the Generalized Projected Gradient Descent (GPGD) framework, which unifies traditional sparse recovery methods and modern approaches using learned deep projective priors. We extend previous convergence results to robustness to model and projection errors. We use these theoretical results to explore ways to better control stability and robustness constants. To reduce recovery errors due to measurement noise, we consider generalized back-projection strategies to adapt GPGD to structured noise, such as sparse outliers. To improve the stability of GPGD, we propose a normalized idempotent regularization for the learning of deep projective priors. We provide numerical experiments in the context of sparse recovery and image inverse problems, highlighting the trade-offs between identifiability and stability that can be achieved with such methods.

We study safety-critical control on fully actuated mechanical systems by means of Zeroing Control Barrier Functions (ZCBFs) defined on Lie Groups. Specifically, we introduce and theoretically validate two classes of ZCBFs. The first enforces kinematic constraints, suitable for implementing obstacle avoidance algorithms. The second enforces kinetic energy limits along prescribed inertial-frame translational and rotational directions, relevant for ensuring safe physical interaction. Numerical simulations involving slit traversal and safe landing scenarios are presented to validate the effectiveness and versatility of the proposed methodology.

Analog computing at the edge is an emerging strategy to limit data storage and transmission requirements, as well as energy consumption, and its practical implementation is in its initial stages of development. Translating properties of biological neurons into hardware offers a pathway towards low-power, real-time edge processing. Specifically, resonator neurons offer selectivity to specific frequencies as a potential solution for temporal signal processing. Here, we show a fabricated Complementary Metal-Oxide-Semiconductor (CMOS) mixed-signal Resonate-and-Fire (R&F) neuron circuit implementation that emulates the behavior of these neural cells responsible for controlling oscillations within the central nervous system. We integrate the design with asynchronous handshake capabilities, perform comprehensive variability analyses, and characterize its frequency detection functionality. Our results demonstrate the feasibility of large-scale integration within neuromorphic systems, thereby advancing the exploitation of bio-inspired circuits for efficient edge temporal signal processing.

Mountain huts are buildings located at high altitude, offering a place for hikers and providing shelter. Energy supply on mountain huts is still an open issue. Using renewable energies could be an appropriate solution. Jiujiu Cabins, a famous mountain hut in Shei-Pa National Park, Taiwan, has operated an off-grid solar energy storage system (ESS) with lead-acid batteries. In 2021, a serious system failure took place, leading to no electricity. After a detailed on-site survey, a reorganization and repair project implemented, the energy system came back to operate normally. Meanwhile, an eco-friendly lithium iron phosphate battery (LFP battery) ESS replaces part of the lead-acid battery ESS, forming a hybrid ESS, making a better and green off-grid solar ESS. In this case report, the energy architecture, detailed descriptions, and historical status of the system are provided. An on-site survey of the failed energy system, a system improvement project, and future plan are listed.

Extremely large-scale multi-input multi-output (XL-MIMO) is a promising technology for the sixth generation (6G) wireless networks, thanks to its superior spatial resolution and beamforming gains. In order to realize XL-MIMO costeffectively, an innovative ray antenna array (RAA) architecture with directly-connected uniform linear array (ULA) was recently proposed, which achieves flexible beamforming without relying on traditional analog phase shifters or digital beamforming. However, RAA suffers from the signal blockage issue since its ray-configured ULAs are placed in the same plane. To address this issue, this paper proposes a novel antenna array architecture termed cylinder directly-connected antenna array (DCAA), which is achieved via multiple simple uniform circular array (sUCA) with carefully designed orientations in a layered three-dimensional structure. The so-called sUCA partitions the uniform circular array (UCA) into two sub-arrays where each sub-array has all antenna elements directly connected to achieve a desired beam direction corresponding to the sub-array's physical orientation, thus achieving full spatial coverage. Compared with the conventional ULA architecture with hybrid analog/digital beamforming (HBF), the proposed cylinder DCAA can achieve uniform spatial resolution, enhanced communication rate and lower hardware costs. Simulation results are provided to validate the promised gains of cylinder DCAA, demonstrating its great potential for high-frequency systems such as millimeter wave (mmWave) and Terahertz (THz) systems.

Uncrewed Surface Vehicles (USVs) are a popular and efficient type of marine craft that find application in a large number of water-based tasks. When multiple USVs operate in the same area, they may be required to dock to each other to perform a shared task. Existing approaches for the docking between autonomous USVs generally consider one USV as a stationary target, while the second one is tasked to reach the required docking pose. In this work, we propose a cooperative approach for USV-USV docking, where two USVs work together to dock at an agreed location. We use a centralized Model Predictive Control (MPC) approach to solve the control problem, obtaining feasible trajectories that also guarantee constraint satisfaction. Owing to its model-based nature, this approach allows the rejection of disturbances, inclusive of exogenous inputs, by anticipating their effect on the USVs through the MPC prediction model. This is particularly effective in case of almost-stationary disturbances such as water currents. In simulations, we demonstrate how the proposed approach allows for a faster and more efficient docking with respect to existing approaches.

We present a neural network-based framework for solving the quantitative group testing (QGT) problem that achieves both high decoding accuracy and structural verifiability. In QGT, the objective is to identify a small subset of defective items among $N$ candidates using only $M \ll N$ pooled tests, each reporting the number of defectives in the tested subset. We train a multi-layer perceptron to map noisy measurement vectors to binary defect indicators, achieving accurate and robust recovery even under sparse, bounded perturbations. Beyond accuracy, we show that the trained network implicitly learns the underlying pooling structure that links items to tests, allowing this structure to be recovered directly from the network's Jacobian. This indicates that the model does not merely memorize training patterns but internalizes the true combinatorial relationships governing QGT. Our findings reveal that standard feedforward architectures can learn verifiable inverse mappings in structured combinatorial recovery problems.

Automatic speech recognition for low-resource languages remains fundamentally constrained by the scarcity of labeled data and computational resources required by state-of-the-art models. We present a systematic investigation into cross-lingual continuous pretraining for low-resource languages, using Perso-Arabic languages (Persian, Arabic, and Urdu) as our primary case study. Our approach demonstrates that strategic utilization of unlabeled speech data can effectively bridge the resource gap without sacrificing recognition accuracy. We construct a 3,000-hour multilingual corpus through a scalable unlabeled data collection pipeline and employ targeted continual pretraining combined with morphologically-aware tokenization to develop a 300M parameter model that achieves performance comparable to systems 5 times larger. Our model outperforms Whisper Large v3 (1.5B parameters) on Persian and achieves competitive results on Arabic and Urdu despite using significantly fewer parameters and substantially less labeled data. These findings challenge the prevailing assumption that ASR quality scales primarily with model size, revealing instead that data relevance and strategic pretraining are more critical factors for low-resource scenarios. This work provides a practical pathway toward inclusive speech technology, enabling effective ASR for underrepresented languages without dependence on massive computational infrastructure or proprietary datasets.