All-inorganic CsPbX₃ quantum dots (QDs) are promising gain media for next-generation lasers due to their excellent optoelectronic properties and superior stability. However, achieving ultralow-threshold amplified spontaneous emission (ASE) remains a challenge. While hybrid perovskites integrated with photonic crystals (PCs) suffer from environmental instability, the integration of stable CsPbX₃ QDs with PC microcavities is still in its infancy. Here, we engineer a PC microcavity that synergistically couples with CsPbBr₃ QDs to enhance the spontaneous emission rate via the Purcell effect, thereby dramatically reducing the ASE threshold under single-photon pumping and elucidating the underlying gain dynamics.

A strategy that integrates CsPbBr₃ QDs with SiO₂-based PCs to construct a composite microcavity (CsPbBr₃-QD-PCs) was proposed. The PC microcavity was first optimized using the finite-difference time-domain (FDTD) method. The physical model consisted of close-packed SiO₂ microspheres (refractive index n=1.46) on a quartz substrate (n=1.54), coated with CsPbBr₃ QDs (n=2.3–2.5). The resonance wavelength was governed by a modified Bragg diffraction equation, and the Purcell factor F_p was computed to quantify spontaneous emission enhancement. The simulation parameter space included SiO₂ microsphere diameters d=50–1000 nm, diffraction orders m=1 and 2, and QD overlayer thickness increments ΔH. Experimentally, CsPbBr₃ QDs were synthesized via the ligand-assisted reprecipitation (LARP) method at room temperature, using CsBr and PbBr₂ as precursors with oleic acid and oleylamine as ligands dissolved in DMF/DMSO, followed by a two-step centrifugation purification. The PC template was fabricated via an SDS-assisted lifting method using SiO₂ microspheres of approximately 253 nm diameter, forming a hexagonal close-packed periodic structure. CsPbBr₃ QD solution was infiltrated into the PC template under low vacuum (0.3–0.6 Pa). A dense CsPbBr₃ QD film on quartz served as the control. Characterizations included XRD, SEM, steady-state absorption and PL spectroscopy, time-resolved photoluminescence (TRPL), and femtosecond transient absorption (TA) spectroscopy. The pump source was a Ti:sapphire femtosecond laser amplifier (Spectra-Physics Solstice Ace) frequency-doubled via a BBO crystal to produce 400 nm pulses (~35 fs, 1 kHz).

FDTD simulations revealed three key features of F_p within the CsPbBr₃ QD emission window (λ=500–550 nm): 1) high F_p at diffraction orders m=1 and m=2; 2) the F_p peak red-shifts with increasing SiO₂ diameter d owing to variations in the lattice constant and filling fraction; 3) the maximum F_p exhibits a near-periodic distribution with m. Fine spectral scans identified d=170 nm (m=1) and d=250 nm (m=2) as optimal diameters, both with F_p peaking near 540 nm. The d=250 nm design was selected for its superior robustness: the ΔH tolerance range maintaining F_p > 2.5 was approximately 300 nm, substantially broader than the ~200 nm range for d=170 nm, attributable to stronger electric field localization.

XRD confirmed the orthorhombic phase of CsPbBr₃ QDs. SEM revealed the periodic hexagonal close-packed structure and uniform QD coverage, with a continuous QD layer (D₁≈436 nm) fully encapsulating the SiO₂ microspheres (D₂≈253 nm), yielding ΔH≈183 nm consistent with FDTD optimization. Despite lower absorption due to the SiO₂ volume, the PL intensity of CsPbBr₃-QD-PCs was enhanced 3.8-fold compared with bare QD films, accompanied by a PL redshift of approximately 15 nm. This redshift is primarily attributed to the spectrally selective Purcell enhancement (F_p peaking at λ≈540 nm), with additional contributions from the modified local dielectric environment and enhanced reabsorption of shorter-wavelength photons. TRPL measurements showed that the PL lifetime was shortened from 1.96 ns (bare QDs) to 1.24 ns (QD-PCs), directly evidencing the Purcell-enhanced spontaneous emission rate.

Under 400 nm femtosecond laser pumping, CsPbBr₃-QD-PCs exhibited clear ASE: a narrow emission peak emerged at ~540 nm with superlinear intensity growth. The ASE threshold was determined to be 8.7 μJ/cm², with the FWHM narrowing to approximately 6.7 nm. Bare CsPbBr₃ QD films showed no ASE under identical conditions, confirming the essential role of the PC microcavity.

TA spectroscopy further elucidated the gain dynamics. Both samples shared four spectral features: excited-state absorption (470–500 nm), ground-state bleaching (GSB, 500–530 nm), transient photoinduced absorption (530–550 nm), and its subsequent inversion into stimulated emission. The PC microcavity shifted the GSB center wavelength consistently with the PL redshift and accelerated the GSB rise time from 1501 ps to 960 ps, indicating Purcell-enhanced carrier relaxation to the band edge. Critically, the optical gain lifetime at ~540 nm was drastically shortened from 1924 ps (bare QDs) to 618 ps (QD-PCs), a ~3.1-fold acceleration. This demonstrates that the PC structure activates a highly efficient stimulated emission channel, enabling the excited state to be rapidly depleted via stimulated emission and outcompeting nonradiative recombination. A systematic comparison with reported CsPbBr₃-based gain media—including thermally evaporated films (~35 μJ/cm²), spin-coated films (26–100 μJ/cm²), and glass-embedded nanocrystals (~54.5 μJ/cm²)—confirms the competitiveness of the 8.7 μJ/cm² threshold achieved here. The solution-based self-assembly approach offers the advantage of large-area, low-cost fabrication without complex lithography.

In summary, we have demonstrated ultralow-threshold ASE by integrating CsPbBr₃ QDs with a PC microcavity. FDTD optimization yielded a CsPbBr₃-QD-PCs composite with F_p > 2.5 and strong electric field confinement, fabricated via vacuum-assisted self-assembly. The PC microcavity enhanced the PL intensity 3.8-fold, shortened the PL lifetime from 1.96 ns to 1.24 ns, and accelerated the optical gain lifetime from 1924 ps to 618 ps, enabling an ASE threshold of 8.7 μJ/cm² under single-photon pumping. This work elucidates the Purcell effect-driven mechanism by which PC microcavities enhance perovskite optical gain, and provides a platform for developing low-power, high-performance all-inorganic perovskite laser devices.

Femtosecond laser direct writing has emerged as a powerful tool for fabricating three-dimensional photonic circuits inside transparent materials. However, conventional transverse writing geometry suffers from aberration-induced cross-sectional distortion, resulting in pronounced polarization sensitivity of the inscribed waveguides. Moreover, the single-focus point-by-point scanning strategy severely limits fabrication throughput when dealing with large-scale waveguide arrays or complex integrated structures. This work aims to develop a high-efficiency, polarization-insensitive waveguide fabrication strategy that simultaneously addresses the symmetry issue of transverse writing and the throughput bottleneck of serial scanning.

A longitudinal femtosecond-laser multi-focus writing (LFLMW) technique based on a reflective phase-only spatial light modulator (SLM) was proposed. The SLM was loaded with computer-generated holograms designed by the GSW algorithm to convert a single incident beam into programmable multi-focus arrays at the focal plane of the objective lens. The sample was translated parallel to the laser propagation direction, ensuring that the axial direction of the multi-focus array coincided with the sample scanning direction. An optical system incorporating a 1030 nm, 300 fs, 76 kHz femtosecond laser amplifier, a 4f imaging system, and a 20× objective (NA = 0.4) was employed. Fused silica substrates with dimensions of 10 mm × 10 mm × 2 mm were used. Through systematic optimization of pulse energy, scanning speed, and inter-focus spacing, Type I waveguide arrays with different channel counts and cladding waveguides with varied track numbers were fabricated in a single longitudinal scan. Free-space end-face coupling at 633 nm was used to characterize the insertion loss, near-field mode profiles, and polarization dependence of the fabricated waveguides. Based on this technique, a 1×9 beam splitter was fabricated by dynamically switching between a ring-shaped hologram for the input cladding waveguide and a rectangular hologram for the output Type I waveguide array.

Multi-focus arrays including 2×2, 3×3, 4×4 rectangular matrices and ring-shaped patterns with 8 to 16 foci were generated with high fidelity, as verified by CCD imaging at the Fourier plane. For Type I waveguide arrays, a 2×2 array was fabricated with a total pulse energy of 8.4 μJ and a scanning speed of 2 mm/s. The inter-focus spacing of 9 μm effectively suppressed thermal cross-talk between adjacent channels. All waveguides exhibited approximately circular cross-sections with an average mode-field diameter of ~5 μm. Average propagation losses of 1.1 dB/cm (TE) and 1.0 dB/cm (TM) were achieved. For 3×3 and 4×4 arrays, the total pulse energies were increased to 17.9 μJ and 21.7 μJ, respectively, with a reduced scanning speed of 0.4 mm/s to guarantee sufficient energy deposition at each focal spot. All channels maintained regular matrix arrangement without observable inter-channel distortion. Single-mode guidance was confirmed for both polarizations. The 3×3 array exhibited average losses of 0.4 dB/cm (TE) and 0.5 dB/cm (TM), while the 4×4 array achieved losses as low as 0.1 dB/cm (TE) and 0.2 dB/cm (TM).

Cladding waveguides with 8, 10, 12, 14, and 16 damage tracks were fabricated using a total pulse energy of 21.7 μJ and a scanning speed of 0.4 mm/s. The tighter focal spacing in the ring-shaped arrays induced significantly enhanced cumulative energy deposition through superimposed stress and thermal effects, pushing the local modification beyond the positive refractive index change regime into the negative change region dominated by micro-damage and material rarefaction. All cladding waveguides had a diameter of approximately 32 μm. Single-mode transmission was observed under both TE and TM polarizations for all track numbers. The measured propagation losses ranged from 3.62 to 4.83 dB/cm for TE polarization and from 2.98 to 4.17 dB/cm for TM polarization. Increasing the track number densified the cladding boundary, theoretically enhancing transverse confinement; however, it also introduced additional scattering interfaces, resulting in no monotonic trend in propagation loss as a function of track count.

A 1×9 beam splitter was demonstrated by cascading a 16-track cladding waveguide (core diameter 35 μm) as the input stage with a 3×3 Type I waveguide array as the output splitting stage. Hologram switching was executed without sample repositioning, ensuring precise alignment between the two sections. At 633 nm, total insertion losses of 2.20 dB (TE) and 1.37 dB (TM) were measured.

A SLM-based longitudinal femtosecond-laser multi-focus writing technique was successfully developed for parallel and efficient fabrication of optical waveguides and integrated devices in glass. By active wavefront modulation, programmable multi-focus arrays were generated at the objective focal plane, enabling simultaneous inscription of multiple waveguide tracks in a single longitudinal scan. Type I waveguide arrays scaled from 2×2 to 4×4 and cladding waveguides with tunable track numbers from 8 to 16 were fabricated. Type I waveguide exhibited high cross-sectional circularity, polarization-insensitive guidance, and propagation losses comparable to those fabricated by conventional single-focus serial writing. Cladding waveguides shown a slightly larger TE/TM loss difference, while all cladding waveguides maintain stable single-mode guidance under arbitrary polarization. The cascaded integration of a cladding waveguide input and a Type I waveguide array output was achieved through dynamic hologram switching, realizing a 1×9 polarization-independent beam splitter with low insertion loss and high splitting uniformity at 633 nm. Compared with conventional serial writing, the proposed method theoretically reduces the processing time of an N-track structure to 1/N. Compared with transverse writing approaches, the longitudinal geometry eliminates aberration-induced cross-sectional asymmetry. This technique provides a promising route toward rapid prototyping and scalable manufacturing of high-performance photonic integrated devices for applications in quantum photonics, optical interconnects, and waveguide lasers.

Single-layer PB phase metasurfaces, while compact and broadband, face inherent limitations in complex wavefront shaping and arbitrary polarization control. Specifically, they cannot achieve beam translation, which requires spatially separated elements, nor can they independently manipulate the diagonal terms of the Jones matrix, limiting their ability to modulate linear polarization. Multilayer cascaded devices address these issues but often introduce fabrication complexities such as interlayer Fresnel reflections, reduced refractive index contrast, or demanding alignment processes. This study aims to overcome these limitations by leveraging the true three-dimensional direct writing capability of femtosecond laser processing to fabricate a compact, cascaded PB phase beam splitter within a single fused silica substrate. The objective is to realize a device that simultaneously achieves beam splitting, wavefront collimation, and linear-to-linear polarization conversion, all while maintaining high efficiency and robustness for high-power applications.

The proposed device is a three-stage cascaded structure integrated monolithically in fused silica. It consists of two identical PB phase polarization gratings and a quarter-wave plate (QWP). The first PG acts as a beam splitter, diffracting the orthogonal circular polarization components of an incident linear polarization into ±1 orders. The second PG, placed at a distance, acts as a collimator, redirecting these divergent beams back to their original propagation direction. The final QWP converts the circularly polarized outputs back to linear polarization.

A rigorous vector diffraction analysis was first conducted to theoretically evaluate the impact of fabrication errors—specifically, non-uniform retardance and nonlinear azimuthal angle variations—on the diffraction efficiency and polarization state of the ±1, 0, and higher orders. This analysis established strict requirements for the linearity of the azimuthal angle and the uniformity of the retardance across the grating to achieve ideal 50:50 beam splitting with high polarization purity.

To meet these stringent requirements, a dynamic pulse line density compensation strategy was developed to address the polarization-dependent fabrication characteristics inherent in femtosecond laser writing. This method compensates for variations in birefringence strength induced by different laser polarization states during scanning, which arise from spatio-temporal asymmetries like pulse front tilt and angular dispersion introduced by the focusing objective. Instead of modifying the laser’s spatio-temporal profile, this approach dynamically adjusts the pulse line density for each scanning line based on its polarization orientation. This allows for uniform birefringence retardance across the entire device, regardless of the slow-axis orientation, while maintaining low propagation loss. The three-layer structure was fabricated in Corning 7980 fused silica using a 1030 nm femtosecond laser and a low numerical aperture objective. The final device was characterized at a 510 nm operating wavelength.

The theoretical analysis confirmed that deviations from ideal conditions—a uniform π retardance and a strictly linear azimuthal angle variation across the grating period—lead to severe performance degradation. Such errors cause energy coupling between the ±1 and 0 orders, introduce higher-order diffractions, and degrade the polarization ellipticity of the output beams. The analysis quantitatively showed that azimuthal angle nonlinearity is a primary source of efficiency loss, while non-uniform retardance causes unbalanced energy distribution between the two output beams.

The experimental investigation of the fabrication process revealed a significant polarization-dependent writing effect. Birefringent structures written with laser polarization at 90° (slow axis horizontal) exhibited a lower modification threshold, resulting in wider, longer, and higher-retardance structures compared to those written with 0° polarization. This "aggressive" growth of structures in the preferred polarization direction distorted the designed linear azimuthal angle variation, effectively pulling the orientation of neighboring structures towards the preferred angle. This phenomenon was found to be objective-dependent, with different objectives (5×, 20×, 40×) showing reversed trends, confirming the role of the optical system’s spatio-temporal aberrations.

The dynamic pulse line density compensation was highly effective in mitigating this polarization-dependent non-uniformity. Compared to a non-compensated grating, the compensated grating showed a reduction in normalized retardance deviation from 30%–40% to just 12%. This significantly improved the linearity of the azimuthal angle profile. The fabricated polarization grating, when tested alone, demonstrated remarkable input polarization robustness. For any linear input polarization, the power splitting ratio between the ±1 orders remained above 90.69%, with the device operating at its design state (90° linear input) achieving a 97.01% splitting ratio and a total diffraction efficiency of 94.8%, far exceeding the ~81% of conventional binary optics.

The complete three-stage collimated beam splitter exhibited excellent performance. The output beams showed a measured separation angle of 0.16 mrad, closely matching the theoretical design, indicating successful collimation. The final output beams were linear with a high splitting ratio of 94.01% and a polarization angle difference of only 3.3°. The device also demonstrated outstanding power handling capability, maintaining a stable ~83% efficiency under 3.6 W of continuous 510 nm laser irradiation over 2400 seconds, with a calculated power tolerance exceeding 52.67 W/cm².

This work successfully demonstrates a new paradigm for fabricating multifunctional PB phase devices by leveraging the true three-dimensional capability of femtosecond laser direct writing. A cascaded PB phase collimated beam splitter was integrated into a single fused silica substrate, overcoming the wavefront shaping and polarization modulation limitations of single-layer devices. The critical challenge of polarization-dependent writing non-uniformity was effectively addressed by a novel dynamic pulse line density compensation method, which is more practical than complex optical path compensation. This approach enabled precise control over the retardance and azimuthal angle distributions, yielding a high-performance device with a splitting ratio of 94.01%, an inter-beam angle of 0.16 mrad, a polarization angle difference of 3.3°, and high-power tolerance. This work paves a robust and scalable way for creating compact, high-performance, and damage-resistant integrated photonic elements for applications such as high-power laser parallel processing, atomic electric field sensing, and advanced beam manipulation systems.

Nitrogen-vacancy (NV) centers in diamond are a robust platform for vector magnetometry with high sensitivity and nanoscale spatial resolution using optically detected magnetic resonance (ODMR). The Zeeman splitting of each NV center is proportional to the absolute magnetic field projection along its quantization axis. Because diamond hosts four possible NV orientations, full spectral separation of all eight resonances is required for unambiguous three-dimensional field reconstruction. In practice, an inappropriately chosen bias magnetic field can cause the projected field magnitudes on different NV axes to become equal, merging the corresponding resonance lines and creating ambiguity in field inversion. This work systematically identifies all conditions that lead to such projection equivalence, formulates rigorous mathematical criteria, and experimentally verifies their influence on ODMR spectra. The objective is to establish a quantitative framework for predicting and preventing spectral overlap, thereby improving the accuracy and reliability of NV-based vector magnetometry.

Taking into account the tetrahedral symmetry of diamond, a coordinate transformation was established between the laboratory frame and an internal frame aligned with the cubic crystal axes. The four NV quantization axes correspond to the body-diagonal directions of the cube. Using direction cosines, the projections of an arbitrary external magnetic field onto each NV axis were expressed as linear combinations of the laboratory field components. From these projection expressions, a complete set of eight independent equivalence conditions was derived algebraically; each condition describes the equality of absolute projected values on a specific pair of NV axes. Under fixed bias components, these conditions produced characteristic curves in the parameter space spanned by the variable field component and the azimuthal angle. An ODMR experimental setup was built with a 532 nm laser for spin initialization and readout, a microwave source for spin manipulation, and a three-axis Helmholtz coil system to generate precisely controlled bias magnetic fields. The coil currents were proportional to the field components. Two field-related currents were fixed, while the third current and the azimuthal orientation were systematically adjusted. Seven representative points were selected for measurement: one intersection of three equivalence curves, several points lying on a single equivalence curve, and points where two independent pairs of NV axes simultaneously satisfied the equivalence condition. At each point, ODMR spectra were recorded, and the number of distinct resonance peaks was identified as a direct indicator of spectral overlap.

The eight equivalence conditions, labeled B_X1 through B_X8, were classified by the NV-axis pairs involved. For example, B_X1 denoted equal projections on NV₁ and NV₄, B_X6 equal projections on NV₁ and NV₃, and B_X8 equal projections on NV₃ and NV₄. Under the fixed bias currents, these conditions produced curves in the parameter space defined by B_X and the azimuthal angle. When only one pairwise equality held, two resonance lines coincided, reducing the number of distinct peaks from eight to six. When two independent equalities occurred simultaneously— for instance, at the intersection of B_X1, B_X6, and B_X8, where NV₁, NV₃, and NV₄ shared identical projection magnitudes—only four peaks remained. The experimental intersection of these three curves yielded four peaks. Points on a single curve, such as B_X8 or B_X6, displayed six peaks. Points on B_X3 (where projections on NV₁ equaled NV₃ and on NV₂ equaled NV₄) and on B_X4 (where projections on NV₁ equaled NV₂ and on NV₃ equaled NV₄) both resulted in four peaks, confirming that two separate pairs could overlap simultaneously. All seven measured characteristic points matched the predictions without exception. These findings demonstrated that the equivalence of projected magnetic fields is the fundamental mechanism underlying ODMR spectral overlap. The degeneracies stem from the cubic symmetry of diamond combined with the linear Zeeman dependence on field projection. The derived criteria enable reliable identification of spectral degeneracies in experimental data. Moreover, they offer direct guidance for selecting bias field directions and magnitudes that avoid equivalence, ensuring complete peak separation essential for unambiguous vector magnetometry.

A systematic investigation of projection equivalence and its role in ODMR spectral overlap was conducted for NV-center-based vector magnetometry. An explicit coordinate transformation between the laboratory frame and the four NV quantization axes was formulated, yielding analytical magnetic field projection formulas. From these, all possible degeneracy conditions where absolute projections become equal were identified. These conditions were experimentally verified using a controlled ODMR setup with three-dimensional Helmholtz coils, and perfect agreement between theory and measurement was observed. The findings clarify the physical origin of spectral overlap and deliver a practical framework for optimizing experimental configurations. The criteria act as a quantitative tool for selecting bias fields that avoid resonance merging, thus enhancing the reliability of three-dimensional magnetic field measurements. This work advances both the fundamental understanding of NV quantum sensing and its application in high-precision magnetometry.

High-speed coherent optical communication, driven by 5G/6G, data-center interconnects, and ultra-high-definition video transmission, requires advanced modulation formats such as QPSK and high-order QAM, which impose stringent demands on integrated 90° optical hybrids. These devices perform coherent mixing of signal and local oscillator (LO) light, converting the in-phase (I) and quadrature (Q) components into optical intensity signals. Their insertion loss, channel crosstalk, phase error, and splitting uniformity directly determine receiver sensitivity and error vector magnitude (EVM). Currently, most integrated 90° optical hybrids are fabricated on planar platforms such as silica-based PLC and SOI. Restricted by two-dimensional geometry, the I and Q arms must be arranged within the same plane, leading to severe crosstalk from waveguide crossings and large device footprints—the overall chip area far exceeds that of the core functional units. In contrast, femtosecond laser direct writing offers true three-dimensional (3D) fabrication capability, enabling arbitrary waveguide geometries inside transparent materials and overcoming planar layout limitations. This work exploits this 3D capability to propose a compact, high-performance 90° optical hybrid with ultra-low crosstalk and low polarization-dependent phase error.

A fully symmetric 3D orthogonal architecture is proposed based exclusively on 2×2 directional couplers for both splitting and phase shifting. The theoretical analysis is grounded in coupled-mode theory. For a lossless 2×2 directional coupler, the through and cross port outputs satisfy E_through = E_incos(κL) and E_cross = E_injsin(κL), where κ is the coupling coefficient, L is the coupling length, and j is the imaginary unit corresponding to the intrinsic 90° phase shift. This phase difference is determined solely by coupled-mode theory and has no direct dependence on κ or L, making it highly robust against process fluctuations. Even when input polarization varies and causes κ to fluctuate, the 90° phase difference remains stable. Moreover, the I and Q interferometric arms adopt completely symmetric spatial layouts and identical fabrication parameters, ensuring that for any polarization state, the propagation phase difference remains zero. Based on this theory, the I and Q arms are inscribed at depths of 250 µm and 300 µm, respectively, with a 50 µm vertical separation to virtually eliminate evanescent-wave crosstalk. Identical writing parameters and symmetric paths maintain amplitude balance and phase orthogonality, while a waveguide trimming structure (ΔL) provides precise optical-path-length matching. Fabrication uses a Yb:KGW femtosecond laser system (1030 nm, 100 kHz, 197 fs) with second-harmonic generation at 515 nm through a BBO crystal. The 515 nm pulses are focused by a 20× objective (NA = 0.75), and the process is monitored in real time by a CCD camera.

Systematic optimization of fabrication parameters is performed. Under optimized conditions (56 mW, 0.1 mm/s), waveguides achieve stable single-mode transmission at 1550 nm with a minimum propagation loss of 0.16 dB/cm. Performance at different depths (250 µm and 300 µm) shows excellent consistency, with a loss difference below 0.02 dB/cm and a mode-field overlap integral of 0.986, demonstrating outstanding process stability across depths. The directional coupler splitting characteristics are systematically characterized. With a coupling gap of 12 µm, a coupling length of 1.712 mm achieves 1∶1 uniform splitting at 1550 nm. The coupling coefficient κ deviation between the two depths is less than 0.5%, confirming that core functional units do not require separate design adjustments for different depths, significantly simplifying fabrication. The crosstalk between the I and Q arms, measured by the end-face injection method, is below −72 dB—far superior to the −20 to −40 dB levels of most reported devices. This fully verifies that the 3D orthogonal layout effectively suppresses evanescent-wave crosstalk. In practice, due to measurement system limitations, the actual crosstalk does not exceed −72 dB. The fabricated device exhibits a compact footprint of 25 mm × 90 µm and an insertion loss of 2.3 dB. The splitting amplitude imbalance is maintained within 0.95–1.05, achieving near-ideal 1:1 uniformity. Phase demodulation performance is evaluated using Lissajous ellipse fitting. Under vertical polarization (V-light), the phase error is −0.35° with an amplitude imbalance of 1.04; under 30° polarization, −0.83° and 1.10; under 60° polarization, −3.53° and 1.26; and under horizontal polarization (H-light), −3.37° and 1.26. Across the full polarization range, the phase error remains within ±5°, demonstrating significant suppression of polarization-dependent phase errors by the symmetric interferometric arm structure. A digital calibration matrix is derived from the measured transmission matrix. After applying a single fixed calibration matrix, the residual phase error across the full polarization range is reduced to within 2°, and the amplitude imbalance is controlled within 1 ± 10%. The calibrated Lissajous ellipses converge to nearly perfect circles that almost completely overlap with the ideal unit circle, with no observable rotation or axis deviation. This intuitively verifies that the digital calibration method effectively compensates for amplitude imbalance and phase errors, requiring only a single calibration to achieve near-ideal quadrature demodulation across the full polarization range.

This work demonstrates a novel paradigm for high-performance 90° optical hybrids using femtosecond-laser 3D photonic integration. By synergistically combining a 3D orthogonal waveguide layout with a fully symmetric directional-coupler-based architecture, the device simultaneously achieves ultra-low crosstalk (<−72 dB), minimal polarization-dependent phase error (<±5° across the full polarization range, and <2° after calibration), and an extremely compact form factor (25 mm × 90 μm). The device offers a high-performance, rapidly prototypal 3D integration solution for high-speed coherent optical communication systems, and also provides new design concepts for the application of femtosecond laser direct writing technology in 3D integrated photonic signal processing and quantum optics.

Microwave photonic links (MPLs) serve as a critical physical-layer infrastructure that bridges microwave engineering and integrated photonics, enabling low-loss, high-isolation, and wideband transmission and processing of high-frequency microwave signals. They represent a key enabling technology for next-generation 6G communications, high-resolution radar, and integrated sensing-communication (ISAC) systems. The performance of MPLs is fundamentally governed by two core metrics: working bandwidth and spurious-free dynamic range (SFDR). Working bandwidth determines the link capacity to handle high-frequency and multi-octave signals, while SFDR—comprehensively reflecting system gain, third-order intermodulation distortion (IMD3), and noise floor—quantifies the link's capability to faithfully transmit desired signals in the simultaneous presence of strong interference and weak targets. An inherent design tension exists between these two metrics: broadening the bandwidth typically demands flatter frequency responses from modulators and photodetectors, often at the cost of linearity, while achieving high SFDR requires precise bias control and nonlinear suppression that conflict with low-loss wideband design. Simultaneously realizing ultra-wide bandwidth and large dynamic range therefore remains a long-standing core challenge in this field.

This review presents a systematic survey of recent advances in high-performance MPLs, organized around on-chip integrated micro-nano photonic devices as the central theme.

At the device level, substantial breakthroughs have been achieved across all key enabling components. In on-chip laser sources and optical frequency combs, heterogeneously integrated lasers with fundamental linewidths below 1 kHz and RIN below −160 dB/Hz have been demonstrated; femtosecond pulse generators on thin-film lithium niobate (TFLN) have produced flat-top optical frequency combs with conversion efficiencies an order of magnitude higher than prior integrated sources; and a triple-resonance electro-optic comb on thin-film lithium tantalate (LiTaO₃) has generated over 2000 comb lines spanning 450 nm (>60 THz), while Kerr microcomb sources with conversion efficiencies exceeding 90% provide low-noise multi-wavelength foundations for wideband MPL architectures. In electro-optic modulators, the TFLN platform has advanced the 3-dB bandwidth from 45 GHz in early demonstrations to beyond 250 GHz, supporting 512 Gbps fiber-wireless transmission and 400 Gbps terahertz-band wireless records; a heterogeneously integrated LiTaO₃-on-Si₃N₄ modulator achieved near-100 GHz bandwidth with 333 Gbit/s PAM4 and 581 Gbit/s 16-QAM transmission; and plasmonic modulators extended the modulation bandwidth beyond 1 THz, covering MHz to THz in a single device. In photodetectors, germanium-on-silicon devices achieved 3-dB bandwidths of 265 GHz, while ultrahigh-power Ge-Si photodetectors delivered record saturation photocurrents of 471.4 mA with OIP3 of 32.3 dBm and SFDR of 114.7 dB·Hz^2/3; modified uni-traveling-carrier photodetectors (MUTC-PDs) maintained frequency responses beyond 140 GHz while preserving high linearity.

At the system level, advances are reviewed along three principal directions. For large dynamic range links and nonlinear distortion suppression, three complementary layers of techniques were developed: at the modulator bias and architecture level, schemes including SD-DPMZM, dual-parallel polarization modulator, and DP-DPMZM achieved IMD3 suppression of 20–40 dB and SFDR improvements up to 20 dB; in the optical sideband manipulation domain, a gain-penalty-free IMD3 suppression scheme based on optical single-sideband modulation achieved SFDR exceeding 128 dB·Hz^4/5—the first linearization technique to suppress nonlinear distortion without any link gain sacrifice—while multi-order optical sideband vector manipulation (MOSVM) simultaneously improved SFDR by over 20 dB and system gain by over 12 dB; in the algorithm compensation domain, genetic algorithm-assisted coherent detection enabled stable 100.8 km transmission with SFDR above 122 dB·Hz^2/3 across −40 to 70°C, and deep neural network architectures including complex-biGRU and focal-loss-based networks provided robust nonlinear compensation in ultra-high-speed and long-distance scenarios. For ultra-wideband microwave photonic communication and spectrum sensing, chromatic dispersion compensation strategies spanning optical double-sideband pre-compensation, optical phase conjugation, TFLN differential phase-diversity electro-optic modulators, and silicon photonic neural network equalization were systematically investigated; a dispersion-immune microwave photonic phase shifter achieved continuous 360° phase shifting from 5 to 25 GHz; a TFLN-integrated real-time spectrum sensing system demonstrated 57.5 GHz instantaneous bandwidth with sub-110 ns latency for 6G ISAC applications; photonics-assisted W-band dual-hop radio-over-fiber transmission achieved 80 Gbit/s 16-QAM at 90 GHz over a 21-km fiber link; and a self-injection-locked microcomb-driven silicon photonic transceiver achieved a WDM coherent transmission record of 61.2 Tbit/s with a net spectral efficiency of 8.01 bit/s/Hz. For microwave photonic intelligent signal processing and computing, a programmable TFLN microwave photonic notch filter simultaneously achieved high dynamic range, high link gain, low noise figure, and ultra-high rejection ratio on a single chip; a silicon photonic true-time-delay processor integrated arbitrary waveform generation, tunable filtering, and ultra-broadband beamforming; a silicon photonic analog parallel processor demonstrated simultaneous 6 GHz radar signal processing and 8 Gbps wireless communication on a single photonic integrated circuit; and microcomb-driven optical convolutional accelerators achieved computing speeds exceeding 3 TOPS for tensor convolution, complex-valued convolution, and transposed convolution tasks, while a photonic edge intelligence chip (PEIC) realized multi-modal analog signal sensing and on-chip inference with a response latency of only 1.33 ns and an energy efficiency of 29 fJ/OP.

On-chip integrated micro-nano photonic devices have substantially advanced the performance boundaries of MPLs in both bandwidth and dynamic range, driving the evolution of MPLs systems from signal transmission conduits toward multi-functional intelligent processing platforms. Looking forward, chip-to-fiber coupling loss, wafer-scale fabrication yield, and heterogeneous integration compatibility remain the primary bottlenecks for TFLN-based systems; topological photonic protection mechanisms offer a promising route to improve fabrication robustness and reduce waveguide losses. The fundamental tension among gain, noise figure, and nonlinearity has yet to be resolved at the system level; deep co-design of device innovations—continuously reducing $V_{\text{π}} $ and improving linearity and saturation photocurrent—with adaptive wideband digital linearization and online learning algorithms represents the fundamental path to breaking this bottleneck. As wafer-scale multi-layer heterogeneous integration matures, fully integrated MPLs transceiver systems incorporating laser sources, modulators, photodetectors, and signal processing units on a single chip are expected to become the core physical-layer infrastructure for next-generation 6G intelligent wireless networks and integrated sensing-communication-computation platforms.

Accurate railway track detection is a critical component of intelligent railway safety systems, including obstacle intrusion warning, autonomous train navigation, and infrastructure monitoring. Reliable detection improves operational safety and enables real-time decision-making in complex railway environments. However, railway track detection remains challenging due to various environmental and structural factors. Railway tracks are often affected by strong background interference caused by ballast, vegetation, shadows, and surrounding infrastructure. These background elements share similar visual characteristics with track structures, which reduces feature discriminability. In addition, railway tracks exhibit significant scale variation across image regions, ranging from large, clearly visible structures in near-view areas to thin and indistinct structures in distant regions. This variation requires the model to maintain strong multi-scale feature representation capability. Furthermore, railway scenes frequently contain complex geometric configurations, such as curved tracks, intersections, and switches, which require the model to capture long-range spatial dependencies and preserve structural continuity. Existing deep learning-based methods, particularly those derived from lane detection frameworks, improve detection performance by leveraging convolutional neural networks and structured prediction strategies. However, these methods often suffer from insufficient receptive fields, limited multi-scale context modeling capability, and weak suppression of background interference. These limitations reduce detection accuracy and robustness in challenging railway scenarios. Therefore, developing an efficient and robust railway track detection method capable of enhancing multi-scale feature representation and suppressing background interference is essential for improving intelligent railway safety systems.

A railway track detection network named dilated and masked attention network (DMA-Net) was developed to address these challenges. The proposed method was built upon a row-based anchor framework, which efficiently localizes horizontal track positions and preserves structural consistency. Several key modules were introduced to improve feature representation and discrimination capability.A dilated multi-scale fusion module (DMSFM) was designed to enhance multi-scale feature extraction. This module employed parallel dilated convolutions with different dilation rates to capture contextual information at multiple spatial scales. Dilated convolutions expanded the receptive field without reducing spatial resolution or increasing computational complexity. The parallel structure enabled effective aggregation of both local structural features and global contextual information, improving representation of curved, distant, and intersecting tracks.An adaptive multi-head masked attention (AMMA) mechanism was introduced to suppress background interference and enhance track-related feature representation. This module applied a learnable mask to the attention map, which selectively suppressed irrelevant background regions and emphasized important track areas. The multi-head attention structure enabled the model to learn diverse feature relationships and improve discrimination capability in complex environments.In addition, the SimAM attention module and DropBlock regularization were incorporated to further improve feature quality and model robustness. SimAM enhanced feature discriminability by evaluating neuron importance without introducing additional parameters. DropBlock improved generalization capability by randomly suppressing contiguous feature regions during training. These modules improved robustness while maintaining a lightweight network structure.

Experiments were conducted on the public Rail-DB benchmark dataset to evaluate the effectiveness of DMA-Net. The proposed method achieved a detection accuracy of 94.53%, which exceeded the original baseline rail detection method by 1.73%. Compared with other mainstream railway track detection methods, DMA-Net improved accuracy by an average of 8.07%. These results demonstrate the effectiveness of the proposed multi-scale fusion and masked attention mechanisms.The DMSFM module significantly improved the model’s ability to capture track features at different spatial scales. The enlarged receptive field enabled better representation of long-range contextual information, which improved detection performance in distant and curved track regions. The AMMA mechanism effectively suppressed background interference and enhanced attention to track-related features, which improved discrimination capability in complex environments containing ballast, vegetation, and structural interference.Visual analysis further demonstrated that DMA-Net produced smoother and more continuous detection results compared with existing methods. Performance improvements were particularly evident in challenging scenarios such as nighttime conditions, occlusion, and complex track intersections. These results confirm that the proposed modules effectively improve feature representation and detection robustness. DMA-Net maintained a lightweight design while achieving high detection accuracy. The proposed network achieved a processing speed of 268 f/s, which satisfies real-time processing requirements. The efficient architecture and lightweight attention mechanisms enable practical deployment in real-world railway safety systems.

DMA-Net provides an effective and robust solution for railway track detection in complex environments. The DMSFM module enhances multi-scale feature representation and enlarges the receptive field, improving the model’s ability to capture track geometry across different spatial scales. The AMMA mechanism improves discrimination capability by suppressing background interference and emphasizing track-related features. The integration of SimAM attention and DropBlock regularization further improves robustness while maintaining computational efficiency. Experimental results demonstrate that DMA-Net achieves superior detection accuracy, robustness, and real-time performance compared with existing methods. The proposed network provides a reliable technical solution for intelligent railway monitoring, autonomous train navigation, and railway safety applications.

Fluorescence-guided surgery (FGS) has emerged as a critical technique in modern medical procedures, utilizing fluorophores such as indocyanine green (ICG) to visualize anatomical structures and pathological tissues that are invisible to the naked eye. While effective, the clinical utility of fluorescence imaging relies heavily on its integration with white-light (WL) imaging to provide necessary anatomical context. Conventional dual-camera systems, which employ beam splitters and separate sensors, offer high performance but suffer from bulky form factors, high costs, and physical registration errors. Conversely, single-camera solutions that utilize time-division multiplexing often face significant technical bottlenecks. These include data congestion when processing high-bandwidth 4K video streams via software, and, more critically, the inability to balance exposure between high-dynamic-range white-light scenes and weak fluorescence signals using a single sensor. The primary objective of this study was to overcome these limitations by developing a "virtual dual-camera" architecture. This system aimed to achieve the performance and independent image processing capabilities of a physical dual-camera setup within a single-sensor hardware framework, specifically targeting real-time performance, hardware-level synchronization, and optimized automatic exposure control (AEC) for composite endoscopic imaging.

The proposed system integrated a custom hardware platform with a novel control architecture. The optical front-end utilized a high-resolution CMOS image sensor (Sony IMX334) paired with a dual-bandpass filter capable of blocking excitation light (780 nm) while transmitting visible light and fluorescence signals (above 820 nm). Illumination was provided by a synchronized dual-LED source delivering high-power white light and 780 nm near-infrared excitation.

The core innovation lay in the field-programmable gate array (FPGA)-based control logic. The FPGA functioned as the central timing master, generating precise frame synchronization signals. It coordinated the CMOS exposure window with the alternating strobe of white and near-infrared LEDs, achieving frame-level synchronization accuracy. To resolve the bottleneck of software-based frame separation, the system implemented a hardware-level video stream demultiplexing architecture. The FPGA intercepted the raw MIPI CSI-2 video stream from the sensor. Based on the current lighting state (odd frames for white light, even frames for fluorescence), the FPGA routed the image data to two independent direct memory access (DMA) channels and image signal processor (ISP) pipelines. This hardware separation allowed for the decoupling of image processing parameters; the white-light channel utilized standard color correction and gamma settings, while the fluorescence channel employed aggressive black level correction, higher gamma values for contrast enhancement, and temporal noise reduction, which relied on the frame continuity preserved by the hardware splitting.

Furthermore, a dual-modal joint automatic exposure control (AEC) algorithm was developed to address the drastic intensity disparity between the two imaging modes. The algorithm operated in two distinct phases separated by a single frame delay. In Phase 1, the system calculated the luminance of the white-light frame and adjusted the white LED intensity and the sensor’s global analog gain. Crucially, a constraint strategy prioritized the prevention of white-light overexposure. In Phase 2, utilizing the analog gain fixed during Phase 1, the system optimized the fluorescence image brightness by independently adjusting the near-infrared excitation power and applying digital gain. This decoupled approach ensured that the shared analog gain did not compromise the white-light image quality while still allowing sufficient amplification for the weaker fluorescence signal.

Experimental validation was conducted using a porcine liver model injected with ICG, imaged at varying working distances (10 mm, 25 mm, and 40 mm) to simulate clinical conditions. The hardware synchronization mechanism achieved a control precision of 16.67 ms, strictly aligning the exposure windows with the respective light sources and eliminating channel crosstalk.

The proposed AEC algorithm demonstrated superior stability and speed compared to conventional ISP-based exposure algorithms. Quantitative analysis of convergence time revealed that the system recovered to a stable brightness state within 0.8 seconds (approximately 23 frames for white light and an additional 15–20 frames for fluorescence) even under extreme lighting changes. In contrast, standard AEC algorithms exhibited significant oscillation and "flashing" artifacts due to the coupled feedback loop between the alternating bright and dark frames. The proposed method maintained the white-light luminance within a tight tolerance (50 luma) without the overshoot observed in conventional methods.

Image quality was assessed using structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) metrics. At a standard working distance of 10 mm, the system achieved an SSIM of 0.9994 and a PSNR of 46.23 dB for white-light images, indicating negligible information loss compared to reference images. Fluorescence images at the same distance yielded an SSIM of 0.9893 and a PSNR of 42.25 dB. As the working distance increased to 40 mm, the fluorescence image quality naturally degraded due to the reliance on higher digital gain to compensate for signal attenuation, resulting in an SSIM of 0.9224. However, the system successfully maintained diagnostic visibility without introducing significant noise artifacts or false positives, a benefit attributed to the independent ISP pipeline tuning enabled by the virtual dual-camera architecture. Visual inspection confirmed that the fused images retained clear anatomical textures from the white-light channel and distinct, high-contrast tumor margins from the fluorescence channel.

The study successfully implemented a virtual dual-camera imaging system that resolves the fundamental conflicts inherent in single-sensor fluorescence endoscopy. By shifting the frame synchronization and video demultiplexing tasks to the FPGA hardware layer, the design eliminated the latency and bandwidth limitations associated with software processing, enabling smooth 4K real-time imaging. The novel two-stage automatic exposure control algorithm effectively decoupled the brightness regulation of white-light and fluorescence modes, allowing for optimal dynamic range utilization for both anatomy and pathology without physical filter switching or dual-sensor alignment.

The experimental results confirmed that this architecture delivers image quality comparable to dual-camera systems, with high structural similarity and signal-to-noise ratios, while retaining the cost and size advantages of a single-chip solution. The fast convergence time of the exposure algorithm ensures seamless transitions during intraoperative maneuvering. Consequently, this technology provides a robust, low-cost, and high-performance solution for fluorescence-guided surgery, offering surgeons reliable real-time navigation aids with precise anatomical and functional information fusion. The proposed architecture establishes a viable technical pathway for the next generation of miniaturized, multi-modal endoscopic platforms.

Corn, as one of the world’s most vital staple crops, plays an indispensable role in agricultural production, industrial applications, and economic development. In China, corn planting area and yield currently account for over 40% of total grain output, a proportion that continues to grow steadily, underscoring its strategic significance for national food security and economic stability. However, throughout the growth cycle, corn is persistently threatened by various pests and diseases, including corn borer, leaf spot, rust, stalk rot, and armyworm, which severely restrict both yield and quality. Traditional pest and disease detection methods rely heavily on manual field inspection—a labor-intensive, time-consuming, and inefficient approach that struggles to meet the demands of large-scale cultivation. Moreover, manual detection is inherently subjective and discontinuous, often resulting in missed diagnoses and misjudgments. Consequently, developing an automated, efficient, and accurate detection method for corn pests and diseases has become an urgent necessity.

In recent years, deep learning-based object detection technologies have achieved remarkable success across various domains, with the YOLO (You Only Look Once) series of algorithms gaining particular prominence due to their exceptional balance of detection speed and accuracy. YOLOv11, the latest iteration in this series, incorporates further optimizations in training strategies, network architecture, and detection precision, providing a solid technical foundation for agricultural pest identification tasks. Nevertheless, when applied to corn pest detection in real-world field scenarios, mainstream models including YOLOv11 continue to exhibit critical limitations: 1) Complex environmental interference—the intricate backgrounds, variable lighting conditions, diverse pest morphologies, and frequent occlusions characteristic of field environments substantially impair the model’s ability to accurately localize and recognize targets; 2) Small-target detection difficulty—early-stage or physically small pests and disease lesions occupy extremely limited pixel regions in images and display indistinct features, leading to elevated rates of missed detections and false negatives.

To address these challenges, this paper proposes an improved YOLOv11 model designated as MSAF-YOLO (Multi-Scale Adaptive Feedback YOLO). The core innovation lies in reconstructing the backbone network using a novel MSAF-Net (Multi-Scale Adaptive Feedback Network) architecture, which synergistically integrates two specially designed modules: the Hierarchical Feature Feedback Module (HFFM) and the Adaptive Dynamic Receptive Field Module (ADRF).

The Hierarchical Feature Feedback Module addresses the fundamental problem of semantic information deficiency in shallow feature layers and the progressive loss of small-target features during repeated downsampling operations. Conventional feedforward convolutional networks propagate information unidirectionally from shallow to deep layers; while deep layers acquire rich semantic representations, shallow layers remain dominated by fine-grained spatial details but lack high-level semantic context. HFFM introduces residual-style feedback pathways coupled with channel attention mechanisms to establish cross-layer feature calibration and enhancement. Specifically, deep feature maps containing abstracted semantic information are selectively fed back and fused with corresponding shallow feature maps through learnable gating mechanisms. The channel attention component adaptively recalibrates feature responses, emphasizing informative channels while suppressing less useful ones. This feedback-driven feature refinement significantly enhances the model’s capacity to extract and preserve discriminative characteristics of small pests and incipient disease symptoms, effectively mitigating the information attenuation typically observed across deep convolutional architectures.

The Adaptive Dynamic Receptive Field Module confronts the inherent inflexibility of standard convolutional neural networks, which employ fixed geometric receptive fields that remain constant regardless of input content. Such rigidity proves particularly disadvantageous when processing agricultural images containing objects at vastly different scales—from millimeter-scale rust pustules to centimeter-scale expanded leaf spot lesions—under heterogeneous background conditions. ADRF achieves pixel-level receptive field adaptation through an elegantly designed integration of spatially adaptive weights and multi-branch dilated convolutions. For each spatial location in the feature map, the module dynamically computes context-dependent weights that govern the relative contributions of parallel convolutional branches employing different dilation rates. This mechanism enables the network to expand its receptive field when processing large contextual regions while maintaining fine resolution for small-target localization, all modulated by the local structural characteristics of the input features. The resulting adaptive receptive field substantially improves model robustness and generalization across diverse pest and disease categories, growth stages, and environmental conditions.

The proposed MSAF-YOLO model was rigorously evaluated on two datasets: a self-constructed corn pest and disease dataset (CORN) encompassing representative categories including corn borer, leaf spot, and rust under various field conditions, and the publicly available general plant disease dataset (diseases) to assess cross-domain generalization capability. Experimental results demonstrate substantial and consistent improvements across multiple evaluation metrics. On the CORN dataset, MSAF-YOLO achieved mean Average Precision at 50% IoU (mAP50) of 85.7% and mAP50:95 of 68.9%, representing significant gains of 3.2 and 2.7 percentage points respectively over the baseline YOLOv11 model. On the diseases dataset, mAP50 and mAP50:95 reached 80.9% and 62.8%, improving by 3.7% and 3.1% respectively. Notably, these accuracy enhancements were accomplished alongside parameter optimization, demonstrating that MSAF-YOLO achieves a favorable trade-off between detection precision and computational efficiency.

In conclusion, this research makes three principal contributions: 1) it systematically identifies and characterizes the specific challenges confronting corn pest and disease detection in complex field environments, particularly the coupled difficulties of small-target detection and background interference; 2) it proposes two novel modules—HFFM and ADRF—that respectively address semantic feature deficiency and fixed receptive field limitations through innovative feedback mechanisms and dynamic convolution strategies; 3) it delivers a practically deployable model that achieves state-of-the-art detection performance on corn pest and disease tasks while maintaining architectural efficiency suitable for real-time agricultural applications. The proposed MSAF-YOLO framework not only advances the methodological frontier of object detection in agricultural contexts but also offers a viable technical pathway toward intelligent, automated crop health monitoring systems.