Owing to their excellent performance in terms of short pulse duration and high peak power, ultrashort pulse lasers have gained much attention in the fields of fiber-optics communication, biophotonics, and industrial processing.1^–8 Recently, because of the advantages of compactness, robustness, and freedom from maintenance over conventional solid-state lasers,9^–13 fiberized ultrashort pulses operating in the wavelength range from 900 to 950 nm produced by passive mode-locking techniques have become highly desirable for use in two-photon bioimaging (TPM) systems for the efficient excitation of green fluorescent proteins and their variants.14^–18 Combined with nonlinear frequency conversion, $1.0\text{-}\mu \mathrm{m}$${\mathrm{Yb}}^{3+}$-doped,12$1.5\text{-}\mu \mathrm{m}$${\mathrm{Er}}^{3+}$-doped,19 and $1.9\text{-}\mu \mathrm{m}$${\mathrm{Tm}}^{3+}$-doped10^,13 ultrashort fiber lasers have been utilized to achieve 900- $\sim 950\text{-}\mathrm{nm}$ ultrashort pulses, which have propelled the advances of TPM systems, with the only drawback being the reduced all-fiber configuration as a result of using these additional nonlinear media. As a comparison, the direct generation of ultrashort pulses at 920 nm using ${\mathrm{Nd}}^{3+}$-doped fibers is a more straightforward method9^,20^–24 and has attracted considerable research interest because the emission of its three-level (${{}^4\mathrm{F}}_{3/2}\to {{}^4\mathrm{I}}_{9/2}$) transition falls into the 900 to 950 nm wavelength band.

Taking ${\mathrm{Nd}}^{3+}$-doped multicomponent fibers and silica fibers as gain media, ultrafast fiber lasers ranging from 905 to 932 nm have been previously studied using different passive mode-locking techniques, including nonlinear amplification loop mirrors,21^–23 nonlinear polarization rotation,24^,25 and semiconductor saturable absorber mirrors (SESAMs).26^,27 In particular, due to high rare-earth (RE) ions’ solubility, these ${\mathrm{Nd}}^{3+}$-doped multicomponent fibers enable gigahertz repetition rate 920 nm ultrashort pulses.26^–28 Nevertheless, owing to the inherent difference in glass softening temperature and thermal expansion coefficient, the low-loss and high-strength fusion splicing between multicomponent fibers and silica fibers has been an obstacle to their commercialization.29 Furthermore, the low damage threshold of multicomponents is another concern for their long-term running. In terms of practical applications, silica fibers are still an irreplaceable host for RE ion doping in the near-infrared (NIR) region.

However, in terms of ${\mathrm{Nd}}^{3+}$-doped silica fibers, two issues have been limiting the performance improvement of 920 nm ultrashort pulses; taking the pulse repetition rate as an example, it is still well below 100 MHz21^–25—much lower than that of these multicomponent fibers,26^–28 which hardly meets the requirement of TPM imaging speed increase, as higher repetition rates are required to shorten the pixel dwell time via dense two-photon excitation.27 The first issue, generally regarded as the most challenging one, is the competitive $1.06\text{-}\mu \mathrm{m}$ amplified spontaneous emission (ASE) of the four-level transition (${{}^4\mathrm{F}}_{3/2}\to {{}^4\mathrm{I}}_{11/2}$), which originates from the same upper laser level (${{}^4\mathrm{F}}_{3/2}$) as the 920 nm but with a stronger intensity.30^,31 This usually develops into noise or even parasitic oscillation that affects the self-starting of the mode-locking process. To weaken the $1.06\text{-}\mu \mathrm{m}$ ASE, W-type ${\mathrm{Nd}}^{3+}$-doped silica fibers are ordinarily adopted because this waveguide structure can induce an effective cutoff wavelength near 1060 nm.25^,32 Yet, the pulse repetition rate by these W-type fibers, as mentioned above, is limited to relatively low values of $<100\mathrm{MHz}$.22^,25 The reason for this repetition rate limitation comes from the second issue—the low RE ion solubility of silica glass, which leads to low pump absorption and optical gain. Co-doping with Al or P is a feasible method to improve the doping level of RE ions in silica glass. However, this method is literally not beneficial for the $0.9\text{-}\mu \mathrm{m}$ laser oscillation when considering the spectral properties because the photoluminescence of ${\mathrm{Nd}}^{3+}$ tends to enhance the $1.06\text{-}\mu \mathrm{m}$ transition probability once these cations are introduced. In addition, the emission peak also exhibits an obvious blueshift, such as shifting from 935 to 900 nm or even shorter after P incorporation,33^,34 causing deviation of the fiber gain spectrum from the desired wavelength window. How to elevate the 920 nm optical gain and suppress the $1.06\text{-}\mu \mathrm{m}$ ASE in ${\mathrm{Nd}}^{3+}$-doped silica fiber still remains a challenge in improving the 920-nm ultrashort pulse performance.

In this work, we demonstrate a coordination engineering strategy to enhance the gain performance at 920 nm, while suppressing the $1.06\text{-}\mu \mathrm{m}$ ASE intensity of ${\mathrm{Nd}}^{3+}$-doped silica fibers. Sulfur is introduced into the fiber core glass using the modified sol-gel (M-Sol-Gel) method with high-temperature sintering technology, creating a coordinate environment of ${\mathrm{Nd}}^{3+}$ that is more favorable for the three-level transition instead of the four-level transition. As a result, the 920-nm passively mode-locked laser built by this novel ${\mathrm{Nd}}^{3+}$-doped silica fiber yields a repetition rate over 200 MHz. To the best of our knowledge, the 207 MHz fundamental repetition rate of 920-nm ultrashort pulses reported here is the highest value realized in ${\mathrm{Nd}}^{3+}$-doped silica fibers, which can be a promising light source in high-speed TPM systems.

Manipulating the coordinate environment of RE ions is an effective strategy for achieving desirable optical properties.35 As analyzed in our previous research,36 we detailed in theory the suitable environment favorable for the ${{}^4\mathrm{F}}_{3/2}\to {{}^4\mathrm{I}}_{9/2}$ transition of ${\mathrm{Nd}}^{3+}$ by the Judd–Ofelt theory. The theoretical spontaneous emission probabilities $A(J\to J^{\prime })$ are determined by the following equation: $$A(J\to J^{\prime })=\frac{64{\pi }^4{e}^2}{3h{\lambda }^3(2J+1)}\times \frac{n{(n^2+2)}^2}{9}\times \sum _{t=\mathrm{2,4},6}{\mathrm{\Omega }}_t{|⟨{\phi }^a|U^{(t)}|{\phi }^b⟩|}^2.$$(1)

The fluorescence branching ratio (${\beta }_x$, $x={{}^4\mathrm{I}}_{9/2}$, ${{}^4\mathrm{I}}_{11/2}$ or ${{}^4\mathrm{I}}_{13/2}$) for these three transitions of ${\mathrm{Nd}}^{3+}$ is calculated based on spontaneous emission probability as $${\beta }_x=\frac{A(J\to J^{\prime })}{\sum _{J^{\prime }}A(J\to J^{\prime })},$$(2)where $J$ is the total angular momentum and equals to 3/2 for the transitions from the ${{}^4\mathrm{F}}_{3/2}$ level, $J^{\prime }$ is the total angular momentum of the lower state and equals to 9/2, 11/2, and 13/2, respectively, $\lambda$ is the emission peak wavelength, and $n$ is the refractive index. $⟨|{\phi }^a|U^{(t)}|{\phi }^b⟩{|}^2$ ($t=2$, 4, 6) are the doubly reduced unit tensor operators and independent of the matrix;37 the values are listed in Table 1.

The ${|⟨{\phi }^a|U^{(2)}|{\phi }^b⟩|}^2$ for these three transitions is null, i.e., ($J\to J^{\prime }$) are dominated by ${\mathrm{\Omega }}_4$ and ${\mathrm{\Omega }}_6$ as evaluated by Eq. (1). Combined with Eq. (2), it can be derived that the ${\beta }_x$ is governed only by the ratio of ${\mathrm{\Omega }}_4/{\mathrm{\Omega }}_6$; $\beta ({{}^4\mathrm{I}}_{9/2})$ will increase only by increasing the ${\mathrm{\Omega }}_4/{\mathrm{\Omega }}_6$ while both $\beta ({{}^4\mathrm{I}}_{11/2})$ and $\beta ({{}^4\mathrm{I}}_{13/2})$ decrease. Consequently, a larger ${\mathrm{\Omega }}_4$ or a smaller ${\mathrm{\Omega }}_6$ is desired for larger $\beta ({{}^4\mathrm{I}}_{9/2})$. The ${\mathrm{\Omega }}_6$ is closely linked with the overlap integrals of the 4f and 5d orbitals and decreases as the outermost electron cloud density of ${\mathrm{Nd}}^{3+}$ increases, which can be achieved by increasing the covalency between ${\mathrm{Nd}}^{3+}$ and the bond anion.38 Therefore, the key to enhancing $\beta ({{}^4\mathrm{I}}_{9/2})$ is to create a coordinate environment of ${\mathrm{Nd}}^{3+}$ featuring stronger covalency with the bond anion.36

Different from multicomponent glasses, the coordination engineering of RE ions in silica-based glasses is challenging because their rigid network limits the components that can be incorporated.29 In this case, we recognize these non-oxygen anions, such as S, can be used as they play the same role as oxygen in silicate glass networks, bridging with cations.39 More importantly, the electronegativity of S (2.5) is much smaller than that of O (3.5); we can evaluate accurately the electronegativity differences of the Nd-S bond (1.3), which is much smaller than that of Nd-O (2.3), given the electronegativity of Nd (1.2). In other words, it can therefore be reasonably deduced that ${\mathrm{Nd}}^{3+}$-doped silica glass would benefit to three-level transition instead of the four-level transition when using sulfur to tailor the coordinate environment of ${\mathrm{Nd}}^{3+}$. Here, we use ${\mathrm{Nd}}_2{\mathrm{S}}_3$ as a dopant to create such an environment for ${\mathrm{Nd}}^{3+}$, and the glass (labeled as N-S-G) was fabricated by our M-Sol-Gel method combined with the high temperature sintering technique,40^,41 which also permits high doping level and uniform dispersity. Furthermore, ${\mathrm{Nd}}_2{\mathrm{O}}_3$-doped silica glass, denoted as N-O-G, was also fabricated for comparison.

As shown in Fig. 1(a), the photoluminescence spectra of the emission bands of ${\mathrm{Nd}}^{3+}$ reveal several intriguing features after the introduction of sulfur. First, under the excitation of an 808 nm laser diode (LD), a notable enhancement in the overall NIR emission intensity for N-S-G is observed, and the peak positions keep identical with N-O-G. The second feature is that the increased intensity at 900 nm is much higher than that at 1064 nm. As shown in Fig. 1(b), compared to N-O-G, the $\beta ({{}^4\mathrm{I}}_{9/2})$ of the three-level transition ($0.9\mu \mathrm{m}$) of N-S-G increased from 0.37 to 0.43, while $\beta ({{}^4\mathrm{I}}_{11/2})$ of the four-level transition decreased from 0.52 to 0.44; that is, the $\beta ({{}^4\mathrm{I}}_{9/2})$ of the three-level transition in N-S-G is indeed increased, as expected. The $\beta ({{}^4\mathrm{I}}_{9/2})$ in N-S-G is much higher than that of ${\mathrm{Nd}}^{3+}$-doped silica glasses34 and multicomponent glasses.26 The lifetime of ${{}^4\mathrm{F}}_{3/2}$ level in sample N-S-G also shows a slight increase, from 364 to $373\mu \mathrm{s}$ (Fig. S1 in the Supplementary Material), showing that no concentration quenching occurs.

To verify that the spectral variation originated from the environment change surrounding ${\mathrm{Nd}}^{3+}$, we first performed the transmission electron microscopy (TEM) analysis combined with energy-dispersive X-ray spectrometry (EDS) to confirm the sulfur was indeed introduced into the silica glass. Figure 2(a) presents the photograph of N-S-G, from which we can observe that the glass sample retains high transmittance and homogeneity. Figures 2(c)–2(g) are the measurement results of TEM. With the addition of Si, Al, O, and Nd, an obvious EDS signal of S is detected in N-S-G; it presents the same distribution characteristics as the other elements, unambiguously showing the presence of sulfur. Furthermore, the electron probe microanalysis (EPMA) mapping analysis of Nd (Fig. S2 in the Supplementary Material) also presents its uniform distribution at a larger scale. These characteristics of element distribution also confirm the glass homogeneity of the N-S-G sample.

The variation in the environment surrounding ${\mathrm{Nd}}^{3+}$ was further validated by the spectroscopic intensity parameters (${\mathrm{\Omega }}_t$, $t=2$, 4, 6) analysis by the above-mentioned Judd–Ofelt theory. Based on the absorption spectra of N-O-G and N-S-G (Fig. S3 in the Supplementary Material), the ${\mathrm{\Omega }}_t$ were derived using the least squares fitting method.36 The low root-mean square deviation, ${\delta }_{\mathrm{rms}}$ (${10}^{-7}$), shows the reliability of these parameters. As listed in Table 2, compared to N-O-G, N-S-G has a smaller ${\mathrm{\Omega }}_6$ and larger ${\mathrm{\Omega }}_4$, i.e., a larger value of ${\mathrm{\Omega }}_4/{\mathrm{\Omega }}_6$, indicating a higher covalence degree between ${\mathrm{Nd}}^{3+}$ and the bonding anion. Furthermore, the ${\mathrm{\Omega }}_2$ for N-S-G is also higher than that of the N-O-G, which also reflects the asymmetry degree around ${\mathrm{Nd}}^{3+}$, and covalence between ${\mathrm{RE}}^{3+}$ and the bonding anions are elevated.34 Combining the aforementioned analyses, we infer that sulfur breaks the Si–O rigid network by replacing some of the oxygen and enters into the direct neighbor coordination with ${\mathrm{Nd}}^{3+}$, creating an environment that is more beneficial to the $0.9\text{-}\mu \mathrm{m}$ emission.

The key performance that needs to be explored is whether the tailored coordinate environment can be transferred to fiber when subjected to a secondary high-temperature process during fiber drawing. Using the well-polished Nd-S-G and Nd-O-G rods as cores, we directly fabricated two standard ${\mathrm{Nd}}^{3+}$-doped silica single-mode fibers, denoted as N-S-F and N-O-F, respectively. The detailed fiber parameters are listed in Table 3. From the fiber refractive index profiles (Fig. S4 in the Supplementary Material), the numerical apertures (NAs) for N-S-F and N-O-F were calculated to be 0.14 and 0.15, respectively. Moreover, as shown in Figs. 3(a) and 3(b), the fiber absorption coefficient of N-S-F improved; the 808 nm absorption coefficient reached $4.1\mathrm{dB}/\mathrm{cm}$, which is higher than the $3.7\mathrm{dB}/\mathrm{cm}$ of N-O-F. The differences in the fiber NA and absorption coefficient are additional supports for the different of coordinate environment of ${\mathrm{Nd}}^{3+}$. Furthermore, the preservation of sulfur and its intriguing effect were confirmed by the fiber-forward ASE spectral investigation. As shown in Fig. 3(c), a prominent decrease in the $1.06\text{-}\mu \mathrm{m}$ ASE intensity for the N-S-F was observed and, impressively, almost no ASE intensity change or shift around 920 nm occurred. The measurement setup is provided in Fig. S5 in the Supplementary Material. The ASE spectra with linear coordinates [Fig. 3(d)] provide a clearer indication of the weakened $1.06\text{-}\mu \mathrm{m}$ ASE, where the peak intensity of N-S-F at 1062 nm was only 0.63 times that of N-O-F. These data confirm that the created coordinate environment of Nd is preserved in fiber that weakens the $1.06\text{-}\mu \mathrm{m}$ ASE without affecting the required 920 nm. Furthermore, the main reason for 920 nm ASE without exhibiting an obvious improvement, we think, is the fiber absorption coefficient at 920 nm also shows a slight increase in N-S-F, as can be seen from Figs. 2(a) and 2(b).

Successful tuning of the fiber ASE properties prompted us to exploit their lasing performance. As presented in Fig. 4(a), we first evaluate the continuous wave (CW) laser efficiency, as it is a direct reflection of the optical gain. The laser cavity was formed by splicing a pair of fiber Bragg gratings (FBGs) centered at 920 nm with a section of ${\mathrm{Nd}}^{3+}$-doped silica fiber. Benefiting from the well-matched fiber NA and core diameter with the passive fiber, the cavity loss was less than 0.2 dB. The reflectivity and 3-dB bandwidth of high-reflection FBG and low-reflection FBG were 99% and 0.5 nm, and 70% and 0.04 nm, respectively. Additionally, two commercial ${\mathrm{Nd}}^{3+}$-doped silica single-mode fibers (Nufern, NDF-PM-5/125 and Coractive, ND103) were also selected for comparison. The detailed fiber properties of the fibers are listed in Table 3. To provided adequate pump absorption, a total of around 40-dB pump absorption was fixed for these four types of fibers. Before injection into an optical spectrum analyzer (OSA), the 920-nm laser powers were attenuated to the same level for better comparison of the laser signal-noise-ratio (SNR) and the $1.06\text{-}\mu \mathrm{m}$ ASE intensity. Figure 4(b) shows the measured laser spectra; clearly, the 920-nm laser were obtained using these four ${\mathrm{Nd}}^{3+}$-doped silica fibers. Detailed laser spectra are provided in Fig. S6 in the Supplementary Material. Among the laser spectra of these four ${\mathrm{Nd}}^{3+}$-doped silica fibers, the $1.06\text{-}\mu \mathrm{m}$ ASE from N-S-F exhibits the lowest intensity, more than 10 dB lower than that of ND103, and thus the 920-nm CW laser by N-S-F presents the highest laser SNR of more than 56 dB. This result further validates the role of the designed environment of ${\mathrm{Nd}}^{3+}$ in suppressing the $1.06\text{-}\mu \mathrm{m}$ ASE. Moreover, as shown in Fig. 4(c), the laser conversion efficiency by N-S-F also presents remarkable elevation, reaches to 25.4%, compared to the 15.0% for N-O-F, 9.2% for NDF-PM-5/125, and 13.0% for ND103. A maximum slope efficiency of 27.4% was obtained with optimizing N-S-F length (Fig. S7 in the Supplementary Material). We attributed the significant increase of laser efficiency to the advantage of boosting the radiative transition from ${{}^4\mathrm{F}}_{3/2}$ level to the ground level (${{}^4\mathrm{I}}_{9/2}$) when the transition at $1.06\mu \mathrm{m}$ is suppressed after co-doping with sulfur. Moreover, our featured M-Sol-Gel method also permits high ${\mathrm{Nd}}^{3+}$ ion doping with excellent dispersion. It is worth noting that the CW laser operations at $890-910\mathrm{nm}$ were also achieved by N-S-F (Fig. S8 in the Supplementary Material), showing a gain bandwidth of at least 30 nm and the potential of tunable fiber laser.

To demonstrate the ability of fiber in generating ultrashort pulses at 920 nm, we selected a linear-cavity passive mode-locking scheme using a commercial SESAM as the mode locker. The experimental setup is illustrated in Fig. 5(a). The SESAM with a 13% modulation depth, 3-ps recovery time, and $15\text{-}\mu \mathrm{J}/{\mathrm{cm}}^2$ saturable fluence was directly butt-coupled with the fiber end facet. A uniform FBG centered at 920-nm with a 3-dB bandwidth of 1.2 nm was selected to define the laser wavelength and render more oscillating longitudinal modes. The laser cavity is compact, and the shortest length was optimized to $\sim 50\mathrm{cm}$ with a 207-MHz repetition rate picosecond pulse output. We had literally realized a higher repetition rate near 300 MHz by further reducing the cavity length, but its initiating pump threshold was over 300 mW due to the decrease of optical gain, which we think is not beneficial for application, as the maximum power of current 808 nm single-mode LD is around 250 mW. Detailed analyses of this 207-MHz pulse characteristics are shown in Figs. 5(b)–5(f). No obvious intensity fluctuation can be observed either in the 50-ns time span or the wider time span in the typical temporal pulse trains recorded by the oscilloscope [Fig. 5(b)], indicating its temporal stability. The 4.8-ns temporal pulse period, i.e., a 207 MHz repetition rate, conforms to the cavity length and shows the pulse is fundamentally mode-locked. Benefiting from the suppressed $1.06\text{-}\mu \mathrm{m}$ ASE and boosted fiber gain performance, the repetition rate of N-S-F is much higher than that of the commercial silica fibers.24^,25 The measured laser spectrum confirms that the laser peak wavelength is at $\sim 920\mathrm{nm}$ with a 3-dB bandwidth of 0.8 nm [Fig. 5(c)], corresponding to a transform-limited ${\mathrm{sech}}^2$ pulse width of 1.1 ps. The spectral bandwidth of the laser is narrower than that of the FBG, which may be because not all modes located in the FBG bandwidth are phase-locked, owing to the mode competition.42 Considering the gain spectrum of this newly developed fiber is over 30 nm, it is believed broader laser spectrum—supporting narrower transform-limited pulse width—can be enabled if adopting wideband output mirrors, such as fiber-type dielectric film. The inset of Fig. 5(b) shows the spectral evolution in the 2-h time span, during which no wavelength drifting occurred, indicating the long-term stability of the wavelength. The accurate pulse repetition rate is 206.93 MHz rate as recorded using radiofrequency (RF) spectrum [Fig. 5(d)]. The 67.9-dB SNR as well as the clean and flat 5-GHz RF spectrum in the inset of Fig. 5(f) confirms the stability of the mode-locking (ML) state. Assuming a ${\mathrm{sech}}^2$-pulse shape, the pulse duration was determined to be 4.3 ps [Fig. 5(e)], which is larger than the transform-limited pulse width, indicating that the output pulse was chirped. Figure 5(f) shows the average output power as a function of the launched pump power. Self-starting CWML is achieved when the pump power exceeds 175 mW. The output power is low, only a few milliwatts, which may be caused by the high connection loss between SESAM and N-S-F as well as the fiber reabsorption. The output power would be further amplified by combining our $30/125\mu \mathrm{m}$${\mathrm{Nd}}^{3+}$-doped silica fiber.36 These results demonstrate this modified ${\mathrm{Nd}}^{3+}$-doped silica fiber is a suitable medium for producing ultrashort pulses at 920 nm and hold potential as light source in biophotonics.

We have presented a strategy of surrounding environment modification using anion to tune the preferable transition of ${\mathrm{Nd}}^{3+}$-doped silica glass. We have shown that sulfur can be stabilized in the fiber and suppresses the competitive $1.06\text{-}\mu \mathrm{m}$ ASE. Based on this novel ${\mathrm{Nd}}^{3+}$-doped silica fiber, we demonstrate that the CW laser performance at 920 nm can be enhanced, including efficiency and SNR. Especially, the laser efficiency reached up to 27.4%, twice that of commercial ${\mathrm{Nd}}^{3+}$-doped silica fiber. Moreover, this fiber allows for the stable picosecond pulse generation at 920 nm with a high 207 MHz fundamental repetition rate under a proper initiating pump threshold of 175 mW, which is expected to obtain promising applications in biophotonics. The present method also offers an opportunity to tune the properties of RE ion-doped silica glasses and fibers toward ultrafast lasers.

${\mathrm{Nd}}^{3+}$-doped silica glasses were prepared using a self-developed M-Sol-Gel method. For the N-O-G sample, high-purity tetraethoxysilane (TEOS, Kermel), ${\mathrm{AlCl}}_3{·6\mathrm{H}}_2\mathrm{O}$ (99.9995%, Alfa), and ${\mathrm{NdCl}}_3{·6\mathrm{H}}_2\mathrm{O}$ (99.99%, Alfa) were selected as the precursors, and deionized water was added to sustain the hydrolysis reaction. All reagents were mixed and thoroughly stirred at 30°C to form a homogeneous sol, which was then put into an 80°C water bath for 48 h to obtain millimeter gel particles. In oxygen atmosphere, the gel particles were heated at 1100°C for 5 h to completely remove HCl, moisture, and organic matter. The obtained powder was ball-milled to micrometer level and further melted under vacuum at 1750°C for 2.5 h to form a glass. The fabrication process was the same for the N-S-G sample as for the N-O-G, and sulfur was doped via ${\mathrm{Nd}}_2{\mathrm{S}}_3$. Argon was also used as the shielding gas atmosphere during the heat treatment process to avoid the oxidation of the sulfur. Both the N-O-G and N-S-G glass samples were doped with Al to improve the dispersity of ${\mathrm{Nd}}^{3+}$ and its doping concentration. The doping levels of Nd and Al in N-O-G and N-S-G were identical, which were 0.3 and 3.0 mol%, respectively. The obtained glasses were cut and well polished to 2-mm thick slices for spectroscopic measurements. The residual glass samples were mechanically shaped into a core-glass rod with a well-polished surface. Then, it was inserted into a silica tube (Heraeus, Suprasil F300) into a preform, which was then drawn into $125\text{-}\mu \mathrm{m}$ fiber by a drawing tower. The fibers were coated with ultraviolet curing adhesive during the drawing process.

The emission spectra and decay curves of the glass samples were measured on a high-resolution spectrofluorometer (FLS 920, Edinburgh Instruments) with an 808-nm LD as the excitation source. The content of the various elements was measured using an EPMA (JXA8230), with a test error of less than 10%. The refractive indices of the fibers were measured using a fiber index profiler (fiber analysis, IFA-100). The fiber absorption coefficient and propagation loss were measured using a stabilized broadband light source (SLS201L, Thorlabs) and the cutback method. The laser power was measured using a power meter (S145C, Thorlabs) combined with a console (PM100D, Thorlabs). The laser spectra were recorded by an OSA (AQ6374, Yokogawa). The pulse train was measured using a real-time oscilloscope with a bandwidth of 8 GHz (DSOV084A, Keysight) and a photodetector with a span of 5 GHz (DET08CFC, Thorlabs). The RF spectrum was obtained using an electrical spectrum analyzer (N9020B, Keysight). The pulse duration was measured using an autocorrelator (Pulsecheck150, APE).

Biographies of the authors are not available.