Searcharxiv⌕ Search

arXiv subjects

Shangyuan Li

Publications and source records attributed to Shangyuan Li.

10 recordsLinked to original sources
arXiv2026

Dispersion-managed octave soliton microcombs in heterostructured microresonators

Controlling group velocity dispersion is of fundamental importance in ultrafast optics, particularly for supercontinuum generation and optical frequency comb synthesis. However, the simultaneous, independent tailoring of multiple dispersion coefficients over an ultra-broad bandwidth remains a formidable challenge in conventional nanophotonic platforms. Here, we demonstrate a robust strategy for broadband dispersion management using heterostructured microresonators comprised of adiabatically concatenated waveguides with distinct geometries. By meticulously engineering the local dispersion profiles, we can flexibly synthesize the global effective dispersion coefficients of different orders, effectively expanding design degrees of freedom beyond conventional limits. As a benchmarking demonstration, we fabricate heterostructured silicon nitride microresonators using a commercial foundry process and successfully generate octave-spanning soliton microcombs with a repetition rate as low as 118 GHz. These microcombs feature deterministically tunable dispersive waves operating across the 290-310 THz range. Such octave-spanning microcombs with detectable repetition rates and carrier-envelope offset frequencies are readily applicable to f-2f self-referencing in optical clocks and frequency synthesizers. The proposed heterostructured architecture establishes a versatile paradigm for generating ultrawideband soliton microcombs with tailorable spectral profiles.

physics.optics↗
arXiv2026

Vibrational resonance in coupled self-learning Duffing oscillators and its application in noisy radio frequency signal processing

This work presents a new coupled array of frequency-adaptive Duffing oscillators. Based on learning rules, the natural frequency of each oscillator changes with the external excitation to achieve the frequency-adaptive capability in the response. The frequency range of vibrational resonance in the response is greatly extended through the frequency-adaptive learning rule. Moreover, the theoretical condition for vibrational resonance is derived and its validity is verified numerically. The coupled self-learning Duffing oscillators can also perform signal denoising in strong noise environment, and its performance in signal denoising has been verified through processing the simulated signal and the wireless radio frequency signal under two scenarios. The superiority of vibrational resonance to the conventional denosing methods such as wavelet transform and Kalman filter has also been illustrated by experimental radio frequency signal processing. The combination of broadband frequency adaptability and strong noise-reduction capability suggests that these oscillators hold considerable potential for engineering applications.

nlin.AO↗
arXiv2025

Hybrid Attribution Priors for Explainable and Robust Model Training

Small language models (SLMs) are widely used in tasks that require low latency and lightweight deployment, particularly classification. As interpretability and robustness gain increasing importance, explanation-guided learning has emerged as an effective framework by introducing attribution-based supervision during training; however, deriving general and reliable attribution priors remains a significant challenge. Through an analysis of representative attribution methods in classification settings, we find that although these methods can reliably highlight class-relevant tokens, they often focus on common keywords shared by semantically similar classes. Because such classes are already difficult to distinguish under standard training, these attributions provide insufficient discriminative cues, limiting their ability to improve model differentiation. To overcome this limitation, we propose Class-Aware Attribution Prior (CAP), a novel attribution prior extraction framework that guides language models toward capturing fine-grained class distinctions and producing more salient, discriminative attribution priors. Building on this idea, we further introduce CAP Hybrid, which combines priors from CAP with those from existing attribution techniques to form a more comprehensive and balanced supervisory signal. By aligning a model's self-attribution with these enriched priors, our approach encourages the learning of diverse, decision-relevant features. Extensive experiments in full-data, few-shot, and adversarial scenarios demonstrate that our method consistently enhances both interpretability and robustness.

cs.LG↗
arXiv2025

Instance Relation Learning Network with Label Knowledge Propagation for Few-shot Multi-label Intent Detection

Few-shot Multi-label Intent Detection (MID) is crucial for dialogue systems, aiming to detect multiple intents of utterances in low-resource dialogue domains. Previous studies focus on a two-stage pipeline. They first learn representations of utterances with multiple labels and then use a threshold-based strategy to identify multi-label results. However, these methods rely on representation classification and ignore instance relations, leading to error propagation. To solve the above issues, we propose a multi-label joint learning method for few-shot MID in an end-to-end manner, which constructs an instance relation learning network with label knowledge propagation to eliminate error propagation. Concretely, we learn the interaction relations between instances with class information to propagate label knowledge between a few labeled (support set) and unlabeled (query set) instances. With label knowledge propagation, the relation strength between instances directly indicates whether two utterances belong to the same intent for multi-label prediction. Besides, a dual relation-enhanced loss is developed to optimize support- and query-level relation strength to improve performance. Experiments show that we outperform strong baselines by an average of 9.54% AUC and 11.19% Macro-F1 in 1-shot scenarios.

cs.CL↗
arXiv2025

Frequency-comb-steered ultrawideband quasi-true-time-delay beamformer for integrated sensing and communication

Phased array antennas (PAAs) possessing broadband beamforming capabilities are crucial for advanced radar and wireless communication systems. Nevertheless, traditional phase-shifter-based PAA beamformers frequently encounter the beam-squint issue, which substantially restricts their instantaneous bandwidth. Photonic true-time-delay (TTD) beamformers have the potential to overcome this challenge, offering ultrawide bandwidth and immunity to electromagnetic interference. However, their practical application is impeded by the high complexity, which typically involves a vast array of optical switches and delay lines. Here, we introduce a novel frequency-comb-steered photonic quasi-TTD beamformer that eliminates the need for delay lines by leveraging the concepts of frequency-diverse arrays and photonic microwave mixing arrays. This beamformer enables squint-free beamforming of ultrawideband linear frequency modulation waveforms, which is essential for high-resolution radar applications. It ensures seamless and continuous beam steering, effectively delivering infinite spatial resolution. We present a prototype with an 8-element PAA, demonstrating an instantaneous bandwidth of 6 GHz across the entire Ku-band. Additionally, we explore the system's capabilities in integrated inverse synthetic aperture radar imaging and high-speed communication, achieving a high imaging resolution of 2.6 cm * 3.0 cm and a transmission rate of 3 Gbps. Compared to conventional delay-line-based beamformers, our new concept markedly reduces hardware complexity and enhances scalability, positioning it as a potent enabler for future integrated sensing and communication applications.

physics.optics↗
arXiv2023

Ultrasensitive vibrational resonance induced by small disturbances

We have found two kinds of ultra-sensitive vibrational resonance in coupled nonlinear systems. It is particularly worth pointing out that this ultra-sensitive vibrational resonance is a transient behavior caused by transient chaos. Considering long-term response, the system will transform from transient chaos to periodic response. The pattern of vibrational resonance will also transform from ultra-sensitive vibrational resonance to conventional vibrational resonance. This article focuses on the transient ultra-sensitive vibrational resonance phenomenon. It is induced by a small disturbance of the high-frequency excitation and the initial simulation conditions, respectively. The damping coefficient and the coupling strength are the key factors to induce the ultra-sensitive vibrational resonance. By increasing these two parameters, the vibrational resonance pattern can be transformed from an ultra-sensitive vibrational resonance to a conventional vibrational resonance. The reason for different vibrational resonance patterns to occur lies in the state of the system response. The response usually presents transient chaotic behavior when the ultra-sensitive vibrational resonance appears and the plot of the response amplitude versus the controlled parameters shows a highly fractalized pattern. When the response is periodic or doubly-periodic, it usually corresponds to the conventional vibrational resonance. The ultra-sensitive vibrational resonance not only occurs at the excitation frequency, but it also occurs at some more nonlinear frequency components. The ultra-sensitive vibrational resonance as a transient behavior and the transformation of vibrational resonance patterns are new phenomena in coupled nonlinear systems.

nlin.PS↗
arXiv2021

Dispersion-less Kerr solitons in spectrally confined optical cavities

Solitons are self-reinforcing localized wave packets arising from a balance of linear and nonlinear effects. This definition encompasses the interplay of nonlinear gain and loss, leading to the concept of dissipative solitons that has been instrumental in understanding the wide variety of mode locking phenomena in ultrafast optics. To date, most studies have involved the group velocity dispersion as a key ingredient for soliton generation. Here, we report on a novel kind of soliton, both theoretically and experimentally, which builds up in spectrally confined cavities when dispersion is practically absent. Precisely, the interplay between the Kerr nonlinearity and spectral filtering results in an infinite hierarchy of eigenfunctions which, combined with optical gain, allow for the generation of stable dispersion-less dissipative solitons in a previously uncharted regime. When the filter order tends to be infinite, we find an unexpected link between dissipative and conservative solitons, in the form of Nyquist-pulse-like solitons endowed with an ultra-flat spectrum. In contrast to the dispersion-enabled solitons, these dispersion-less Nyquist solitons build on a fully confined spectrum and their energy scaling is not constrained by the pulse duration. This study broadens the fundamental scope of dissipative soliton physics and opens new avenues for engineering optical solitons endowed with superior temporal and spectral features.

physics.optics↗
arXiv2019

Closed Loop Load Model Identification Using Small Disturbance Data

Load model identification using small disturbance data is studied. It is proved that the individual load to be identified and the rest of the system forms a closed-loop system. Then, the impacts of disturbances entering the feedforward channel (internal disturbance) and feedback channel (external disturbance) on relationship between load inputs and outputs are examined analytically. It is found out that relationship between load inputs and outputs is not determined by load itself (feedforward transfer function) only, but also related with equivalent network matrix (feedback transfer function). Thus, load identification is closed loop identification essentially and the impact of closed loop identification cannot be neglected when using small disturbance data to identify load parameters. Closed loop load model identification can be solved by prediction error method (PEM). Implementation of PEM based on a Kalman filtering formulation is detailed. Identification results using simulated data demonstrates the correctness and significance of theoretical analysis.

eess.SY↗
arXiv2018

Microcomb-based true-time-delay network for microwave beamforming with arbitrary beam pattern control

Microwave phased array antennas (PAAs) are very attractive to defense applications and high-speed wireless communications for their abilities of fast beam scanning and complex beam pattern control. However, traditional PAAs based on phase shifters suffer from the beam-squint problem and have limited bandwidths. True-time-delay (TTD) beamforming based on low-loss photonic delay lines can solve this problem. But it is still quite challenging to build large-scale photonic TTD beamformers due to their high hardware complexity. In this paper, we demonstrate a photonic TTD beamforming network based on a miniature microresonator frequency comb (microcomb) source and dispersive time delay. A method incorporating optical phase modulation and programmable spectral shaping is proposed for positive and negative apodization weighting to achieve arbitrary microwave beam pattern control. The experimentally demonstrated TTD beamforming network can support a PAA with 21 elements. The microwave frequency range is $\mathbf{8\sim20\ {GHz}}$, and the beam scanning range is $\mathbf{\pm 60.2^\circ}$. Detailed measurements of the microwave amplitudes and phases are performed. The beamforming performances of Gaussian, rectangular beams and beam notch steering are evaluated through simulations by assuming a uniform radiating antenna array. The scheme can potentially support larger PAAs with hundreds of elements by increasing the number of comb lines with broadband microcomb generation.

physics.app-ph↗