There are two major categories of embodied navigation: Vision-Language Navigation (VLN), where agents navigate by following natural language instructions; and Object-Goal Navigation (OGN), where agents navigate to a specified target object. However, existing work primarily evaluates model performance under nominal conditions, overlooking the potential corruptions that arise in real-world settings. To address this gap, we present NavTrust, a unified benchmark that systematically corrupts input modalities, including RGB, depth, and instructions, in realistic scenarios and evaluates their impact on navigation performance. To our best knowledge, NavTrust is the first benchmark that exposes embodied navigation agents to diverse RGB-Depth corruptions and instruction variations in a unified framework. Our extensive evaluation of seven state-of-the-art approaches reveals substantial performance degradation under realistic corruptions, which highlights critical robustness gaps and provides a roadmap toward more trustworthy embodied navigation systems. Furthermore, we systematically evaluate four distinct mitigation strategies to enhance robustness against RGB-Depth and instructions corruptions. Our base models include Uni-NaVid and ETPNav. We deployed them on a real mobile robot and observed improved robustness to corruptions. The project website is: https://navtrust.github.io.

Uncrewed aerial vehicles (UAVs) are expected to enhance connectivity, extend network coverage, and support advanced communication services in sixth-generation (6G) cellular networks, particularly in public and civil domains. Although multi-UAV systems enhance connectivity for IoT networks more than single-UAV systems, energy-efficient communication systems and the integration of energy harvesting (EH) are crucial for their widespread adoption and effectiveness. In this regard, this paper proposes a hierarchical ad hoc UAV network with non-linear EH and non-orthogonal multiple access (NOMA) to enhance both energy and cost efficiency. The proposed system consists of two UAV layers: a cluster head UAV (CHU), which acts as the source, and cluster member UAVs (CMUs), which serve as relays and are capable of harvesting energy from a terrestrial power beacon. For the considered IoT network architecture, the outage probability expressions of ground Internet of things (IoT) devices, each CMU, and the overall outage probability of the proposed system are derived over Nakagami-m fading channels with practical constraints such as hardware impairments and non-linear EH. We compare the proposed system against a non EH system, and our findings indicate that the proposed system outperforms the benchmark in terms of outage probability.

Large language models (LLMs) have been widely used as knowledge backbones of Large Audio Language Models (LALMs), yet how much auditory knowledge they encode through text-only pre-training and how this affects downstream performance remains unclear. We study this gap by comparing different LLMs under two text-only and one audio-grounded setting: (1) direct probing on AKB-2000, a curated benchmark testing the breadth and depth of auditory knowledge; (2) cascade evaluation, where LLMs reason over text descriptions from an audio captioner; and (3) audio-grounded evaluation, where each LLM is fine-tuned into a Large Audio Language Model (LALM) with an audio encoder. Our findings reveal that auditory knowledge varies substantially across families, and text-only results are strongly correlated with audio performance. Our work provides empirical grounding for a comprehensive understanding of LLMs in audio research.

Autonomous driving (AD) requires safe and reliable decision-making among interacting agents, e.g., vehicles, bicycles, and pedestrians. Multi-agent reinforcement learning (MARL) modeled by Markov games (MGs) provides a suitable framework to characterize such agents' interactions during decision-making. Nash equilibria (NEs) are often the desired solution in an MG. However, it is typically challenging to compute an NE in general-sum games, unless the game is a Markov potential game (MPG), which ensures the NE attainability under a few learning algorithms such as gradient play. However, it has been an open question how to construct an MPG and whether these construction rules are suitable for AD applications. In this paper, we provide sufficient conditions under which an MG is an MPG and show that these conditions can accommodate general driving objectives for autonomous vehicles (AVs) using highway forced merge scenarios as illustrative examples. A parameter-sharing neural network (NN) structure is designed to enable decentralized policy execution. The trained driving policy from MPGs is evaluated in both simulated and naturalistic traffic datasets. Comparative studies with single-agent RL and with human drivers whose behaviors are recorded in the traffic datasets are reported, respectively.

Camera pipelines receive raw Bayer-format frames that need to be denoised, demosaiced, and often super-resolved. Multiple frames are captured to utilize natural hand tremors and enhance resolution. Multi-frame super-resolution is therefore a fundamental problem in camera pipelines. Existing adversarial methods are constrained by the quality of ground truth. We propose GenMFSR, the first Generative Multi-Frame Raw-to-RGB Super Resolution pipeline, that incorporates image priors from foundation models to obtain sub-pixel information for camera ISP applications. GenMFSR can align multiple raw frames, unlike existing single-frame super-resolution methods, and we propose a loss term that restricts generation to high-frequency regions in the raw domain, thus preventing low-frequency artifacts.

Generating audio that is acoustically consistent with a scene is essential for immersive virtual environments. Recent neural acoustic field methods enable spatially continuous sound rendering but remain scene-specific, requiring dense audio measurements and costly training for each environment. Few-shot approaches improve scalability across rooms but still rely on multiple recordings and, being deterministic, fail to capture the inherent uncertainty of scene acoustics under sparse context. We introduce flow-matching acoustic generation (FLAC), a probabilistic method for few-shot acoustic synthesis that models the distribution of plausible room impulse responses (RIRs) given minimal scene context. FLAC leverages a diffusion transformer trained with a flow-matching objective to generate RIRs at arbitrary positions in novel scenes, conditioned on spatial, geometric, and acoustic cues. FLAC outperforms state-of-the-art eight-shot baselines with one-shot on both the AcousticRooms and Hearing Anything Anywhere datasets. To complement standard perceptual metrics, we further introduce AGREE, a joint acoustic-geometry embedding, enabling geometry-consistent evaluation of generated RIRs through retrieval and distributional metrics. This work is the first to apply generative flow matching to explicit RIR synthesis, establishing a new direction for robust and data-efficient acoustic synthesis.

Dynamic metasurface antennas (DMAs) are an emerging hybrid-MIMO technology distinguished by an ultrathin form factor, low cost, and low power consumption. In real-world DMA prototypes, mutual coupling (MC) between meta-elements is generally non-negligible; some architectures even deliberately exploit strong MC to enhance wave-domain flexibility. In this paper, we address channel estimation (CE) for DMAs with known MC by formulating it as a tensor-decomposition problem. We develop a generalized block Tucker alternating least squares (BTALS) algorithm, together with specialized variants for cases with known direct and/or feed channel. We also develop a reciprocity-aware bilinear factorization method for the case with known direct channel. We experimentally validate our algorithms using an 18 GHz DMA prototype whose 7 feeds and 96 meta-elements are strongly coupled via a chaotic cavity. Our general BTALS algorithm reaches an accuracy of 43.1 dB, only 0.3 dB below the upper bound imposed by experimental noise. All proposed algorithms generally outperform the prior-art reference scheme thanks to the superior noise rejection enabled by the tensor-based framework. We further study the minimum number of required measurements as a function of the number of feeds and demonstrate the importance of MC awareness by comparison with an MC-unaware benchmark. Finally, we apply BTALS to a second setup enabling the prediction of the DMA's full dual-polarization 3D radiation diagram. We also measure the latter for DMA configurations optimized for channel-gain enhancement based on the estimated channels. Altogether, our work establishes the practical relevance of MC-aware tensor methods; beyond DMAs, it applies to all wireless systems with wave-domain programmability enabled by tunable lumped elements.

The growing use of inverter-based resources in modern power systems has made grid-following inverters a central topic in power-system modeling, control, and simulation. Despite their widespread deployment, introductory material that explains grid-following inverter operation from first principles and connects control design to time-domain simulation remains limited. To address this need, this tutorial presents a circuit-theoretic introduction to the modeling and simulation of a grid- following inverter connected to an electrical power grid. We describe the inverter synchronization with the grid (PLL), power control, and current control structure and show how these elements can be represented within an electromagnetic transient (EMT) simulation framework using companion model-based formulations similar to those used in circuit simulators such as SPICE and Cadence. In this tutorial, we use the grid-following inverter as the primary example to illustrate how its governing equations, control loops, and network interface can be formulated and simulated from first principles. By the end of the document, readers should gain a clear introductory understanding of how to model and simulate a grid-following inverter in an EMT platform.

Analog multiplexing for sigma delta modulated Digital to Analog Converters has been recently proposed as a means of achieving robustness. This preprint analyses said scheme via simulations. The main limitation introduced by the proposed architecture comes from mismatch in the DACs gain, which can drastically impact performances. A new technique of dynamic elements matching is proposed here to overcome this problem.

Federated learning (FL) enables collaborative model training across distributed devices without sharing raw data, but applying FL to multi-modal settings introduces significant challenges. Clients typically possess heterogeneous modalities and model architectures, making it difficult to align feature spaces efficiently while preserving privacy and minimizing communication costs. To address this, we introduce CoMFed, a Communication-Efficient Multi-Modal Federated Learning framework that uses learnable projection matrices to generate compressed latent representations. A latent-space regularizer aligns these representations across clients, improving cross-modal consistency and robustness to outliers. Experiments on human activity recognition benchmarks show that CoMFed achieves competitive accuracy with minimal overhead.

Radar target detection in the presence of a mixture of non-Gaussian clutter and white thermal noise is a challenging problem. This paper proposes a Rectified Flow Matching-based method for radar detection, termed D-RFM. Unlike existing detectors, D-RFM learns a mapping from a standard Gaussian distribution to radar observations by capturing the underlying velocity field. Detection is then performed by inverse mapping test samples into the latent Gaussian space using the learned velocity field, with targets identified as deviations from the learned distribution. Experimental results demonstrate the efficacy of the proposed method under both Gaussian and non-Gaussian clutter plus additive white Gaussian noise, highlighting its accuracy, robustness, and computational efficiency.

This paper addresses the fundamental question of determining the minimum number of distinct control laws required for global controllability of nonlinear systems that exhibit singularities in their feedback linearising controllers. We introduce and rigorously prove the (k+1)-Controller Lemma, which establishes that for an nth order single-input single-output nonlinear system with a singularity manifold parameterised by k algebraically independent conditions, exactly k+1 distinct control laws are necessary and sufficient for complete state-space coverage. The sufficiency proof is constructive, employing the approximate linearisation methodology together with transversality arguments from differential topology. The necessity proof proceeds by contradiction, using the Implicit Function Theorem, a dimension-counting argument and structural constraints inherent to the approximate linearisation framework. The result is validated through exhaustive analysis of the ball-and-beam system, a fourth-order mechanical system that exhibits a two-parameter singularity at the third output derivative.

This work introduces two lightweight model predictive control (MPC) approaches for attitude tracking with reaction wheels during spacecraft rendezvous synchronization. Both approaches are based on a novel attitude deviation formulation, which enables the use of inherently linear constraints on angular velocity. We develop a single-loop and a dual-loop MPC; the latter embeds a stabilizing feedback controller within the inner loop, yielding a linear time-invariant system. Both controllers are implemented with CasADi - including automatic code generation - evaluated across various solvers, and validated within the Basilisk astrodynamics simulation framework. The experimental results demonstrate improved tracking accuracy alongside reductions in computational effort and memory consumption. Finally, embedded delivery to an ARM Cortex-M7 - representative of commercial off-the-shelf devices used in New Space platforms - confirms the real-time feasibility of these approaches and highlights their suitability for onboard attitude control in resource-constrained spacecraft rendezvous missions.

This paper presents a safety-guaranteed, runtime-efficient imitation learning framework for spacecraft close proximity control. We leverage Control Barrier Functions (CBFs) for safety certificates and Control Lyapunov Functions (CLFs) for stability as unified design principles across data generation, training, and deployment. First, a nonlinear Model Predictive Control (NMPC) expert enforces CBF constraints to provide safe reference trajectories. Second, we train a neural policy with a novel CBF-CLF-informed loss and DAgger-like rollouts with curriculum weighting, promoting data-efficiency and reducing future safety filter interventions. Third, at deployment a lightweight one-step CBF-CLF quadratic program minimally adjusts the learned control input to satisfy hard safety constraints while encouraging stability. We validate the approach for ESA-compliant close proximity operations, including fly-around with a spherical keep-out zone and final approach inside a conical approach corridor, using the Basilisk high-fidelity simulator with nonlinear dynamics and perturbations. Numerical experiments indicate stable convergence to decision points and strict adherence to safety under the filter, with task performance comparable to the NMPC expert while significantly reducing online computation. A runtime analysis demonstrates real-time feasibility on a commercial off-the-shelf processor, supporting onboard deployment for safety-critical on-orbit servicing.

Air-dispersed sensor networks deployed from aerial robotic systems (e.g., UAVs) provide a low-cost approach to wide-area environmental monitoring. However, existing methods often rely on active actuators for mid-air shape or trajectory control, increasing both power consumption and system cost. Here, we introduce a passive elastic-folding hinge mechanism that transforms sensors from a flat, stackable form into a three-dimensional structure upon release. Hinges are fabricated by laminating commercial sheet materials with rigid printed circuit boards (PCBs) and programming fold angles through a single oven-heating step, enabling scalable production without specialized equipment. Our geometric model links laminate geometry, hinge mechanics, and resulting fold angle, providing a predictive design methodology for target configurations. Laboratory tests confirmed fold angles between 10 degrees and 100 degrees, with a standard deviation of 4 degrees and high repeatability. Field trials further demonstrated reliable data collection and LoRa transmission during dispersion, while the Horizontal Wind Model (HWM)-based trajectory simulations indicated strong potential for wide-area sensing exceeding 10 km.

Integrating large language models (LLMs) into wireless communication optimization is a promising yet challenging direction. Existing approaches either use LLMs as black-box solvers or code generators, tightly coupling them with numerical computation. However, LLMs lack the precision required for physical-layer optimization, and the scarcity of wireless training data makes domain-specific fine-tuning impractical. We propose BeamAgent, an LLM-aided MIMO beamforming framework that explicitly decouples semantic intent parsing from numerical optimization. The LLM serves solely as a semantic translator that converts natural language descriptions into structured spatial constraints. A dedicated gradient-based optimizer then jointly solves the discrete base station site selection and continuous precoding design through an alternating optimization algorithm. A scene-aware prompt enables grounded spatial reasoning without fine-tuning, and a multi-round interaction mechanism with dual-layer intent classification ensures robust constraint verification. A penalty-based loss function enforces dark-zone power constraints while releasing optimization degrees of freedom for bright-zone gain maximization. Experiments on a ray-tracing-based urban MIMO scenario show that BeamAgent achieves a bright-zone power of 84.0\,dB, outperforming exhaustive zero-forcing by 7.1 dB under the same dark-zone constraint. The end-to-end system reaches within 3.3 dB of the expert upper bound, with the full optimization completing in under 2 s on a laptop.

Simultaneous wireless information and power transfer (SWIPT) critically depends on waveform design, which governs both reliable data delivery and efficient energy harvesting. Among waveform characteristics, the peak-to-average power ratio (PAPR) plays a pivotal role: low-PAPR signals improve power amplifier (PA) efficiency, while high-PAPR signals exploit rectifier nonlinearities to boost harvested energy. This duality makes PAPR a fundamental design challenge in SWIPT systems. To tackle this issue, we establish a unified analytical framework that characterizes the PAPR-dependent behaviors of both the PA and the rectifier, thereby revealing how waveform statistics determine end-to-end energy transfer efficiency. Building on this insight, we propose a frequency-domain resource allocation strategy for power-splitting SWIPT, where spectral segments are adaptively assigned to balance communication throughput with energy harvesting performance. Here, a key contribution is to extend SWIPT to MIMO-OFDM architectures. Despite concerns over excessive PAPR in large-scale antenna-subcarrier configurations, we demonstrate that appropriate waveform adaptation and resource optimization can transform MIMO-OFDM into an energy-efficient platform for joint data and power transfer. Finally, simulation results confirm significant improvements in PA efficiency, rectifier output, and overall energy transfer, thereby validating the practical benefits of the proposed approach.

Energy consumption is a significant concern for mobile network operators, and to enable further network energy improvements it is also an important target when developing the emerging 6G standard. In this paper we show that, despite the existence of many energy-saving features in 5G new radio (NR) networks, activating them in isolation yields only suboptimal savings and often compromises other network key performance indicators (KPIs) such as coverage or latency. We first introduce a compact taxonomy that distinguishes hardware capabilities from higher-layer features. Features fall into two classes: (i) signaling and scheduling mechanisms that create idle windows, and (ii) features that utilize those windows to save energy. We then present a feature orchestrator as a logical node to coordinate between features to maximize the gain. Using a 3GPP-aligned simulator with product-realistic parameters, we show that coordinating lean NR, scheduling, and advanced sleep modes significantly reduces gNodeB (gNB) energy consumption with negligible throughput loss, compared to the uncoordinated scenario. We conclude by outlining open issues in observability, system dynamics, coordination, and intelligent automation for energy performance management.

Understanding the impact of vegetation and small-scale antenna movements on signal propagation is important for the design and optimization of high-frequency wireless communication systems. This paper presents an experimental study analyzing signal propagation at 60 GHz and 80 GHz in the presence of vegetation, with a focus on forward scattering and microdiversity effects. A controlled measurement campaign was conducted in an indoor environment, where the influence of a potted plant placed in the line-of-sight (LOS) path between the transmitter and receiver was investigated. The study examines the effects of antenna micro-shifts on the channel impulse response (CIR), highlighting variations in received power due to small positional changes of the antennas. The results indicate that the 80 GHz band exhibits higher sensitivity to micro-movements compared to the 60 GHz band, leading to greater fluctuations in received power.

A stochastic modeling methodology for 3-D foliage is presented, aimed at enhancing ray-tracing simulations. The model supports adjustable stochastic geometry, density, and shape to capture variability in foliage structures. The model is validated through experimental measurements of representative vegetation. The influence of foliage density and size on path loss and root mean square delay spread is analyzed to demonstrate the applicability of the model in the 80 GHz frequency band.

In this paper, we show a stochastic approach to generate a 3D model of a foliage, which is then used for deterministic ray-tracing channel modeling. This approach is verified by a measurement campaign at 60 and 80 GHz with 2 GHz bandwidth. The wireless channel is characterized by path-loss and RMS delay spread and we show the angular dependency of those parameters when the receiver is placed on a half-circle around the tree. Besides electromagnetic material properties, the 3D model is characterized by several interpretable parameters, including tree volume, leaf size, leaf density, and the tree crown shape parameter.

We propose a physics-grounded mechanism design for dynamic spectrum sharing that bridges the gap between radiometric retrieval constraints and economic incentives. We formulate the active and passive users coexistence problem as a Vickrey-Clarke-Groves (VCG) auctions mechanism, where the radiometer dynamically procures ``quiet'' time-frequency tiles from active users based on the marginal reduction in retrieval error variance. This approach ensures allocative efficiency and dominant-strategy incentive compatibility (DSIC). To overcome the computational intractability of exact VCG on large grids, we derive an approximation algorithm by using the monotone submodularity induced by the radiometer equation. AMSR-2-based simulations show that the approach avoids high-cost tiles by aggregating low-cost spectrum across time and frequency. In an interference-trap case study, the proposed framework reduces procurement costs by about 60% over a fixed-band baseline while satisfying accuracy targets.

A major bottleneck in Computer-Assisted Preoperative Planning (CAPP) for fracture reduction is the limited availability of annotated data. While annotated datasets are now available for evaluating bone fracture segmentation algorithms, there is a notable lack of annotated data for the evaluation of automatic fracture reduction methods. Obtaining precise annotations, which are essential for training and evaluating automatic CAPP algorithm, of the reduced bone therefore remains a critical and underexplored challenge. Existing approaches to assess reduction methods rely either on synthetic fracture simulation which often lacks realism, or on manual virtual reductions, which are complex, time-consuming, operator-dependant and error-prone. To address these limitations, we propose a hybrid physical-digital framework for generating annotated fracture reduction data. Based on fracture CTs, fragments are first 3D printed, physically reduced, fixed and CT scanned to accurately recover transformation matrix applied to each fragment. To quantitatively assess reduction quality, we introduce a reproducible formulation of clinically relevant 3D fracture metrics, including 3D gap, 3D step-off, and total gap area. The framework was evaluated on 11 clinical acetabular fracture cases reduced by two independent operators. Compared to preoperative measurements, the proposed approach achieved mean improvements of 168.85 mm 2 in total gap area, 1.82 mm in 3D gap, and 0.81 mm in 3D step-off. This hybrid physical--digital framework enables the efficient generation of realistic, clinically relevant annotated fracture reduction data that can be used for the development and evaluation of automatic fracture reduction algorithms.

Sleep disturbances are tightly linked to cardiovascular risk, yet polysomnography (PSG)-the clinical reference standard-remains resource-intensive and poorly suited for multi-night, home-based, and large-scale screening. Single-lead electrocardiography (ECG), already ubiquitous in Holter and patch-based devices, enables comfortable long-term acquisition and encodes sleep-relevant physiology through autonomic modulation and cardiorespiratory coupling. Here, we present a proof-of-concept Holter-to-Sleep framework that, using single-lead ECG as the sole input, jointly supports overnight sleep phenotyping and Holter-grade cardiac phenotyping within the same recording, and further provides an explicit analytic pathway for scalable cardio-sleep association studies. The framework is developed and validated on a pooled multi-center PSG sample of 10,439 studies spanning four public cohorts, with independent external evaluation to assess cross-cohort generalizability, and additional real-world feasibility assessment using overnight patch-ECG recordings via objective-subjective consistency analysis. This integrated design enables robust extraction of clinically meaningful overnight sleep phenotypes under heterogeneous populations and acquisition conditions, and facilitates systematic linkage between ECG-derived sleep metrics and arrhythmia-related Holter phenotypes. Collectively, the Holter-to-Sleep paradigm offers a practical foundation for low-burden, home-deployable, and scalable cardio-sleep monitoring and research beyond traditional PSG-centric workflows.

We study a continuous-time dynamical system of nodes diffusively coupled over a hierarchical network to examine the efficiency and performance tradeoffs that organizations, teams, and command and control units face while achieving coordination and sharing information across layers. Specifically, after defining a network structure that captures real-world features of hierarchical organizations, we use linear systems theory and perturbation theory to characterize the rate of convergence to a consensus state, and how effectively information can propagate through the network, depending on the breadth of the organization and the strength of inter-layer communication. Interestingly, our analytical insights highlight a fundamental performance tradeoff. Namely, networks that favor fast coordination will have decreased ability to share information that is generated in the lower layers of the organization and is to be passed up the hierarchy. Numerical results validate and extend our theoretical results.

Many applications involving multi-agent systems require fulfilling safety constraints. Control barrier functions offer a systematic framework to enforce forward invariance of safety sets. Recent work extended this paradigm to mean-field scenarios, where the number of agents is large enough to make density-space descriptions a reasonable workaround for the curse of dimensionality. However, an open gap in the recent literature concerns the development of mean-field control barrier functions for Fokker-Planck (advection-diffusion) equations. In this work, we address this gap, enabling safe mean-field control of agents with stochastic microscopic dynamics. We provide bounded stability guarantees under safety corrections and corroborate our results through numerical simulations in two representative scenarios, coverage and shepherding control of multi-agent systems.

This paper investigates a cell-free massive multiple-input-multiple-output (CF-mMIMO) integrated sensing and communication (ISAC) system that addresses the critical challenge of information leakage to potential eavesdroppers located within sensing zones. A novel access point (AP) selection strategy is proposed, which partitions the distributed APs into two functional groups: communication APs (C-APs), dedicated exclusively to data transmission, and sensing APs (S-APs), responsible for target detection and eavesdropper suppression. Closed-form expressions for the achievable communication rate, eavesdropping rate, and mainlobe-to-average-sidelobe ratio (MASR) are derived to evaluate system performance. Two complementary optimization problems are formulated using the successive convex approximation (SCA): (i) maximizing user rates under security constraints and (ii) minimizing eavesdropping rates while satisfying quality of service (QoS) requirements. The proposed joint optimization framework determines the optimal AP operational modes and power allocation across communication and sensing links. Extensive numerical results validate the theoretical analysis and demonstrate significant performance gains, revealing inherent trade-offs among communication efficiency, sensing accuracy, and security. These insights offer practical guidelines for designing secure CF-mMIMO ISAC systems.

Future sixth-generation of wireless system is expected to provide data-rates in the order of 1 Tbps and latencies below 1 ms. Among others, one of the most promising strategies to meet these requirements is to operate at higher frequencies than millimeter wave bands: the THz bands. Nevertheless, because of the higher losses and the detriment of classical propagation mechanisms, deploying systems operating at these frequencies becomes a real challenge. Consequently, short-range scenarios are of special interest since these effects of THz bands can be managed. This work conducts an extensive campaign within corridors at frequencies within the H-band in the range from 250 GHz to 330 GHz. For the first time in literature, an ultra wideband of 80 GHz is studied simultaneously. Large scale effects are assessed by estimating and modeling path gain. The path gain exponent varies between -2.1 and -1.6, which is explained by a guiding effect also observed at millimeter wave bands. Small scale effects are also evaluated in terms of parameters such as rice $K$-factor, root mean squared delay spread and coherence bandwidth. Additionally, an analytical approximation based on the classical N-rays model is proposed obtaining an accurate representation of the wireless channel which is coherent with the empirical analysis. The full analysis reveals the suitability of these THz bands for deploying point-to-point links due to the predominance of the line-of-sight contribution respect to the reflected components.

We introduce DiscoPhon, a multilingual benchmark for evaluating unsupervised phoneme discovery from discrete speech units. DiscoPhon covers 6 dev and 6 test languages, chosen to span a wide range of phonemic contrasts. Given only 10 hours of speech in a previously unseen language, systems must produce discrete units that are mapped to a predefined phoneme inventory, through either a many-to-one or a one-to-one assignment. The resulting sequences are evaluated for unit quality, recognition and segmentation. We provide four pretrained multilingual HuBERT and SpidR baselines, and show that phonemic information is available enough in current models for derived units to correlate well with phonemes, though with variations across languages.

We propose to model parallel streams of data, such as overlapped speech, using shuffles. Specifically, this paper shows how the shuffle product and partial order finite-state automata (FSAs) can be used for alignment and speaker-attributed transcription of overlapped speech. We train using the total score on these FSAs as a loss function, marginalizing over all possible serializations of overlapping sequences at subword, word, and phrase levels. To reduce graph size, we impose temporal constraints by constructing partial order FSAs. We address speaker attribution by modeling (token, speaker) tuples directly. Viterbi alignment through the shuffle product FSA directly enables one-pass alignment. We evaluate performance on synthetic LibriSpeech overlaps. To our knowledge, this is the first algorithm that enables single-pass alignment of multi-talker recordings. All algorithms are implemented using k2 / Icefall.