Low noise, self-phase-modulation-enabled femtosecond fiber sources tunable in 740-1236 nm for wide two-photon fluorescence microscopy applications

: We have demonstrated widely tunable Yb:fiber-based laser sources, aiming to replace Ti:sapphire lasers for the nJ-level ultrafast applications, especially for the uses of nonlinear light microscopy. We investigated the influence of different input parameters to obtain an expansive spectral broadening, enabled by self-phase modulation and further reshaped by self-steepening, in the normal dispersion regime before the fiber damage. We also discussed the compressibility and intensity fluctuations of the demonstrated pulses, to reach the transform-limited duration with a very low intensity noise. Most importantly, we have demonstrated clear two-photon fluorescence images from UV-absorbing fluorophores to deep red dye stains.


Introduction
Two-photon fluorescence microscopy (TPFM), along with the use of femtosecond lasers operated within the spectral range from 700 nm to 1200 nm [1,2], has been considered as a revolutionary imaging tool for biomedical studies in many ways [3][4][5]. First, the compatibility of using two-photon fluorescence (TPF) for molecular imaging via maturely developed fluorescent tools has contributed to a better understanding of advanced biomedical research. For example, visualizing green fluorescent proteins (GFPs) excited by 920-nm femtosecond pulses have led to the observation of complete tissue morphology [6], tracking the transfer of nervous signals [7], and revealing cell-matter interactions by localizing the sub-cellular units [8]. The use of longer excitation wavelengths, which suffer less light attenuation in the specimen, links to the realization of deeper-tissue imaging even in turbid samples: TPFM excited by 1200-nm pulses has thus revealed details of blood vessels in tumor with a penetration depth up to 924 µm in the mouse brain [9]. Second, the excitation spectral window ranging from 750 nm to 820 nm is also of great importance to image ultraviolet and blue light-absorbing fluorophores with reduced photodamage. It thus allows TPFM applications such as virtual optical biopsy [10,11] and physiological monitoring [12]: The TPFM-based virtual biopsy of breast tissues, using 780-nm femtosecond pulses, has offered more accurate evaluations for surgical margins, with both high specificity of 93.3% and sensitivity of 95.4% [11]; TPFM-enabled intravital imaging also paves the way for in situ observation of cell metabolism [13], endocrinological studies [14], and microenvironment monitoring [15].
The possibility of finding an efficient generating process with the ease of dispersion management holds the key to the development of tunable femtosecond sources. Nonlinear optical imaging by amplitude and phase shaping a fiber supercontinuum (SC) from 900 to 1160 nm has thus been demonstrated [24]; however, the need for programable pulse shapers, for active dispersion pre-compensation of the whole spectrum, increased the system complexity with high losses. Besides, soliton self-frequency shift (SSFS) offers a nearly chirp-free mechanism for three-photon imaging based on an energetic Er:fiber laser [25], and its frequency doubling has been applied for different nonlinear imaging modalities [26][27][28][29]. A high repetition rate (i.e., 80 MHz) femtosecond laser using a master-oscillator-power-amplifier approach has also been demonstrated recently tunable from 810-1000 nm [30]. Nevertheless, a high pump power of up to 50 W has been employed due to the less efficient pumping of Er-doped amplifiers and the need for sequential nonlinear conversion via SSFS and SHG [30].
Recently, an efficient way to obtain tunable femtosecond sources has been demonstrated via self-phase modulation enabled spectral selection (SESS) [31][32][33][34][35][36]: Spectral broadening induced by self-phase modulation (SPM) features isolated spectral lobes, and the leftmost and the rightmost spectral lobes of the broadened spectrum can be continuously tuned, by controlling the input pulse energy, and then selected (e.g., using edgepass filters) as tunable sources. Spectral broadening via SPM plays an essential role in ultrafast techniques, such as Kerr-lens mode-locking [38], nonlinear pulse compression [39], applications of Mamyshev regenerators [40,41], and seed generation for optical parametric amplifiers [42]. The energy scalability of SESS has been investigated for third/second harmonic generation imaging [31,32]. However, the feasibility of using SESS for wide TPFM applications, to our best knowledge, has not been discussed. First, although the previous work has demonstrated nJ-level femtosecond (70-120 fs) pulses continuously tunable from 825 nm to 1210 nm [35], it cannot cover the main excitation window for intravital imaging [11][12][13][14]. The possibility of an expansive tuning range, particularly in the covering from the center wavelength of Ti:S lasers around 780 nm, has not been investigated for ultraviolet and blue light-absorbing fluorophores. Second, the pulse compressibility of the generated tunable femtosecond sources from SESS has not been addressed, and the full dispersion compensation of the exciting femtosecond pulses contributes to a more efficient excitation of TPF [43] before photobleaching and photodamage. Although the leftmost/rightmost spectral lobes of the SPM-dominated spectrum are produced at the end of the nonlinear propagation, their spectral chirp, which mainly originates from the accumulation of the nonlinear phase, may still significantly distort the pulses due to higher-order dispersion, especially the cases with substantial spectral detuning from the pumping wavelength via the highly nonlinear process. Therefore, it still requires a detailed optimization of the SESS method to replace Ti:S lasers for broad applications.
To even explore the possibilities for other ultrafast applications, the intensity noise of a light source is also crucial for uses demanding high dynamic range detections, such as optical metrology [44], stimulated Raman scattering (SRS) microscopy [45], pump-probe experiments/microscopy [46], and optical coherence tomography (OCT) [47]. SC generation features a broad spectral coverage, but it is typically regarded as a noisy process resulting from modulation instabilities [48,49] and the nonlinear amplification of the pumping noises [50]. The use of shorter pumping sources (e.g., femtosecond pulses) and shorter fiber lengths for spectral broadening in the normal dispersion regime has contributed to a significant reduction of the intensity noise of the generated spectra [51], and it is thus interesting to investigate the noise behavior of our tunable source based on SESS.
This study investigated the influences of different input parameters on the SPM-dominated spectral broadening and the fiber damage threshold. We were thus able to optimize the spectral coverage of SESS by closely examining the role of self-steepening before wave breaking. As a result, we have produced widely tunable femtosecond sources, ranging from 740 nm to 1236 nm, pumped by a femtosecond Yb:fiber laser. The demonstrated tunable sources featured sub-100-fs pulse durations direct after the fiber output with >1-nJ pulse energies. Moreover, with the use of commercially available double chirped mirrors (DCMs), the tunable femtosecond sources were easily compressible to its transform-limited (TL) duration (e.g., down to 14 fs) with negligible losses (i.e., less than 0.5% per bounce on a single mirror). Besides, we also measured the relative intensity noise (RIN) of the femtosecond sources. With an optimized spectral broadening in the normal dispersion regime, our result shows that the characterized RIN over the whole spectral tuning range is similar to the one from the driving source. We also applied the light sources for TPFM imaging, and we thus obtained clear TPF images from commonly used blue to red fluorophores, including DAPI, GFP, tdTomato, and SYTOX Deep Red. These results indicated the potentials and feasibilities of our demonstrated sources for wide ultrafast applications, especially for TPFM imaging.

Optimization of SPM-dominated spectral broadening
The essence of SESS is to optimize the SPM-dominated spectral broadening, which manifests isolated spectral lobes [31,35]. The generation of a broader spectrum via SPM demands a systematic optimization of different input conditions and fiber parameters, including the damage assessment of optical fibers using femtosecond pulses. Photonic crystal fibers (PCFs) are great candidates as nonlinear media for spectral broadening due to the modifiable mode-field diameters (MFDs) and dispersion profiles [52]. To minimize the temporal pulse broadening from the fiber dispersion followed by the previous work [46], we employed a commercial PCF (NL-1050-ZERO-2, NKT Photonics) with an MFD of 2.2 µm, which features slightly positive dispersion over most of the desired spectral coverage, as shown in Fig. 1(a). The generation of new frequencies based on SPM comes from the time-dependent nonlinear phase shift, ϕ NL (t). As a rule of thumb, the nonlinear phase is proportional to the fiber length L and the pulse intensity, I(t). The frequencies of light, ω(t), can be described as Eq. (1) [53], where ω 0 is the center frequency of the input light, n 2 is the nonlinear refractive index, and k 0 is the propagation constant in a vacuum.
According to Eq. (1), a substantial spectral broadening requires a long fiber length L, a high peak power P (i.e., a high pulse energy E and a short pulse duration τ), and a small mode-field area A. To be more precise, we employed a numerical model to simulate the nonlinear pulse propagation in the PCF, considering the influences of the fiber dispersion, SPM, self-steepening, and the Raman response. The simulation solved a generalized nonlinear Schrödinger (GNLS) equation based on the split-step method [54]. We illustrated a simulated spectral evolution against the fiber length in Fig. 1(b). The leftmost and rightmost spectral lobes are significantly detuned from the pumping wavelength during the propagation, and most of the energy is transferred into the leftmost and rightmost spectral lobes. With an increase in fiber length above 4 mm, the occurrence of wave breaking exhibits a spectral protrusion around 635 nm, and it limits the further blueshift of the leftmost spectral lobes. We thus defined an optimal fiber length by where the peak spectral intensity of small protrusions reached 20% of the leftmost lobe one, as indicated by the red dashed line. (c1-c3) Comparison of spectral broadening at the optimal fiber length, using 190-fs (the dotted line) and using 85-fs (the solid line) input pulses, with the specified coupled pulse energies. The input pulses were simulated with a Gaussian profile, and the pulse duration is specified as its FWHM. (d) Test of potential damage mechanisms using the PCF under different input pulse widths. The damage threshold on the fiber surface is around 1360 mW of the input power, corresponding to the coupled pulse energy of 16.6 nJ with a coupling efficiency of 58.5%, and the damage fluence on the effective mode area of 3.8 µm 2 is 0.73 J/cm 2 . It does not show a significant dependence on the input pulse widths. MFD: Mode-field diameter; FWHM: Full width at half maximum.
A better understanding of self-steepening and wave breaking leads to further optimizations of the spectral broadening using femtosecond pulses. The spectral broadening based on SPM is initially symmetric with the assumption of applying symmetric input pulse shapes. The occurrence of self-steepening will break the symmetry of the SPM-dominated spectral broadening: The nonlinear propagation of femtosecond optical bursts suffers from intensity-dependent group velocity, leading to an increasing slope of the trailing edge of the ultrafast pulse, and the break of temporal shape symmetry results in more blueshift than the redshift but with less power [53]. Followed by more nonlinear phase shifts, wave breaking occurs after the shock formation on the trailing edge, resulting in a spectral protrusion at the spectral margin [55], as shown in the spectral evolution around 635 nm in Fig. 1(b). With the increase of the fiber length, the role of wave breaking becomes more significant, and it hampers the further blueshift of the leftmost spectral lobe. Although the spectral protrusion arising from wave breaking exhibits the largest blueshift, it features a hardly tunable spectral peak with a relatively narrow bandwidth. Therefore, we tried to avoid the spectral protrusion, and we defined an optimal fiber length by where the peak spectral intensity of the small protrusions reached 20% of the leftmost lobe one, as labeled by the red dashed line in Fig. 1(b).
To discuss the impact of different parameters on the SPM-dominated broadening, we compared the spectral broadening with different input pulse energies, as shown in Fig. 1(c1-c3) at different input pulse durations (the solid and dotted lines in Fig. 1(c1-c3)). The optimal fiber lengths were modified accordingly, and the higher input peak intensity links to a shorter optimal fiber length. As a result, applying pulses with a higher peak intensity (e.g., employing a short duration of 85 fs with a high coupled pulse energy of 16 nJ) leads to a wider spectral broadening: The possibility to acquire more nonlinear phase shifts with a shorter optimal fiber length results in a broader spectral broadening due to the less temporal spreading from the fiber dispersion.
Although a wider broadening is theoretically achievable by increasing the input peak intensity, the fiber damage sets the upper bound of input power. Therefore, to discuss the damage mechanism, we measured the damage threshold at different input pulse durations, as shown in Fig. 1(d), and we experimentally evaluated the fiber damage by the inability of light coupling. In our case, focused 1.36-W pulses under a coupling efficiency of 58.5% at a 48-MHz repetition rate would tend to damage the front surface of the fiber. Our result shows no significant dependency between the damage threshold and the input pulse durations, confirming the average power rather than the peak power induces the fiber damage. Furthermore, the coupling inability due to the surface damage can be fully recovered by slightly polishing the fiber head. We want to note that the power-handling capability may be further increased by splicing a proper fiber end cap or realizing a higher coupling efficiency. The damage condition in our case corresponds to the fluence of 0.73 J/cm 2 on the effective mode area of 3.8 µm 2 , and the peak intensity on the fiber surface ranges from 1.5-7.1 TW/cm 2 . Our result reaches 57% of the fused silica damage fluence, around 1.3 J/cm 2 , with sub-ps pulses [56].
After discussing the damage threshold, we evaluated the SPM-dominated broadening and the optimal fiber lengths at different input pulse durations, as shown in Fig. 2(a), under the maximum applicable pulse energy of 16 nJ. Using the definition followed by the red dashed line in Fig. 1(b), the defined optimal fiber length exhibits a linear trend with a slope of 1.8 against the input pulse durations with a logarithmic scale, which indicates a longer input pulse duration leads to the late happening of wave breaking after an exponentially longer propagation length. To investigate the spectral tuning range, we defined the broadening ratio, B, as the frequency difference, ∆ω, divided by the center frequency of the input laser, ω 0 , where ∆ω is the difference of the intensity peak of the leftmost and rightmost spectral lobes (i.e., ω b,peak and ω r,peak , respectively). That is, When the input pulse duration is longer than 50 fs, the results in Fig. 2(a) agree well with the simulation results in Fig. 1(c1-c3): Minimizing the role of fiber dispersion by using shorter pulses leads to a larger broadening ratio. We also marked the spectral peaks of the leftmost and rightmost lobes against different input pulse durations, as shown in Fig. 2(b). When using shorter input pulses, self-steepening plays a more critical role, resulting in a more significant blueshift of the leftmost lobe than the red-shifted one. Nevertheless, the broadening ratio, B, starts to decrease when the input pulse duration is below 40 fs, and the inflection of the detuning trend is due to an earlier occurrence of wave breaking. From our simulation using 50-fs input pulses, we achieved an optimal spectral broadening with the leftmost and rightmost lobes centered at around 738 nm and 1261 nm, respectively. We want to note that the temporal reshaping from self-steepening during the nonlinear propagation in the fiber does not increase the blue-shifted amount but significantly reducing the red-shifted one. In comparison, we also show the simulations without self-steepening as the dotted lines in Fig. 2(b): In the cases without self-steepening, the optimal fiber length, as shown in the green dotted line in Fig. 2(a), is longer than the one with self-steepening, and the late onset of wave breaking results in more redshift, as shown in the red dashed line in Fig. 2(b).

SPM-dominated spectral broadening and spectrum selection
For experimental demonstration, we showed a schematic of our imaging system in Fig. 3, including the driving source, the pulse characterization units, and the microscope setup. We employed a femtosecond Yb-fiber laser (Kasmoro-1030-HE, mRadian) as the pumping source, which produced 48-MHz ultrashort pulses with a 3-W average power. The autocorrelation trace of the laser, as shown in Fig. 3(a), is well fitted with a Gaussian curve of 120-fs full width at half maximum (FWHM), which features a pulse width of 85 fs, given by a deconvolution factor of 0.707, as shown in Fig. 3(a). The center wavelength of the driving source is 1025 nm, and the laser spectrum is shown as the blue line in Fig. 3(b). We coupled the laser into the PCF with a typical output spectrum as the yellow line in Fig. 3(b), and we employed the leftmost and rightmost spectral lobes of the broadened spectra for TPFM. The leftmost and rightmost lobes, as shown in the specified areas in Fig. 3(b), were selected by the shortpass and longpass filters, respectively. To continuously tune the center wavelengths of the selected spectral lobes, we varied the input laser power before the PCF by using a polarization beam splitter (PBS) and a half-wave plate (HWP). The experimental spectrum shared very similarly broadening features with the simulation result but with slightly broader spectral coverage, which may derive from the variance of the input pulse shape and extrapolation inaccuracy of the fiber dispersion in the spectral margin. We employed a pair of DCMs (DCM7, Laser Quantum) and placed glass wedges under Brewster's angle for dispersion pre-compensation. These mirrors exhibit over 99.5% reflectivities, enabling a pulse compression scheme with negligible losses. Before guiding the pulses to the microscope for TPFM, we characterized the pulses with an RF spectrum analyzer (DSA815-TG, Rigol), a home-built second harmonic generation (SHG) frequency-resolved optical gating (FROG) system, and an optical spectral analyzer (MS9710C, Anritsu). To demonstrate the continuous tuning range of the spectral lobes, we measured the spectral evolution versus the coupled pulse energies using a 7-mm-long PCF, as shown in Fig. 4(a). The fiber length was optimized to obtain an extensive blueshift coverage. The filtered spectral lobes, centered from 740 nm to 1236 nm, are shown in Fig. 4(b). We performed FROG measurements using a home-built autocorrelator via a 10-µm-thick Beta Barium Borate (BBO) under type-1 phase matching. Also, we employed broadband dispersion-matched beam splitters (UFBS5050, Thorlabs) to ensure consistent measurements of all the lobes under the same configuration. We used the RANA approach as a robust retrieval algorithm [57], and the Matlab code is available on the Trebino Group website. The output details are listed in Table 1. The filtered spectral lobes contain 15% to 28% of the total output power, and the output power of the leftmost lobes is slightly less than the rightmost lobes, which is the consequence of the asymmetric broadening induced by self-steepening. The corresponding pulse widths of all the lobes from the direct output are less than 100 fs.  Although the generation of the leftmost and rightmost spectral lobes happens close to the end of the nonlinear propagation in the fiber, the chirp of the filtered spectral lobes still significantly broadens the pulse in the time domain. The possibility of compressing the pulse down to its TL duration is vital for TPFM applications to obtain an optimal excitation efficiency. Therefore, we compressed the spectral lobes around 740 nm, as an example shown in Fig. 4(c), which is the one with the most significant spectral chirp and detuning from the pumping wavelength at the direct fiber output. With anomalous GDD of around -1500 fs 2 provided mainly by multiple bounces of a DCM pair, we obtained the compressed pulses with the duration of 14.1 fs (the blue line in Fig. 4(c)), close to its TL width of 13.5 fs (the green line in Fig. 4(c)). The peak intensity of the compressed pulse is thus increased by more than four times than the direct output, followed by the pulse compression from 62 fs to 14 fs with negligible losses from DCMs.
We measured the RIN power spectral density (PSD) of the selected spectral lobes, as shown in Fig. 5. To obtain a uniform frequency response, we used a high-speed Si-based photodiode (S5971, Hamamatsu) with a 70-MHz bandwidth under the reverse voltage of 5 V, and we employed an RF spectrum analyzer (DSA815-TG, Rigol) supporting the measurement from 9 kHz. To obtain consistent measurements using a single photodetector over the whole tuning range, we fixed the detected photocurrent in all the measurements before the signal saturation, and we characterized the spectral lobes above 1100 nm using their SHG. All the measurements were well above the instrumental noise floor, as shown in the gray line in Fig. 5(a). The black line indicates the RIN of the Yb:fiber laser: The PSD follows the 1/f slope (i.e., -20 dB/decade) from 9 kHz to 60 kHz; a plateau around the range of 60-400 kHz and the white noise above 400 kHz are closely linked to the cavity finesse and dispersion of the femtosecond laser [49,58]. The noise PSD of the resulting sources over the spectral tuning range only increases by less than 2.8 dB (i.e., 1.4 times) compared with the driving laser's PSD at 1 MHz, as shown in Fig. 5(b). For reference, we also measured the noise PSD of an SC source (SuperK EXTREME, NKT Photonics) as shown in the red dashed line in Fig. 5(a), and the RIN PSD of the commercial SC source at 1 MHz is 45-dB greater than our demonstrated results. With the use of photomultiplier tubes (PMTs), the dynamic range of the TPF detection is typically around 10 3 . Therefore, the excessive integrated RIN, which reaches the level of 5 × 10 −4 , can be detectable as the intensity fluctuation between the adjacent pixels in TPF imaging. The intensity noise of a light source is even crucial for high dynamic range detections, such as optical metrology [44], SRS microscopy [45], pump-probe experiments/microscopy [46], and OCT [47]. Moreover, we characterized the long-term stability, and we have obtained a robust light coupling with a power variation of less than 3% for at least 3 hours without any feedback control. Furthermore, we applied an active stabilization to the fiber alignment and achieved low power fluctuations of less than 1% for more than 8 hours, potentially allowing for long-period and power-consistent measurements.

Two-photon microscopy images
To demonstrate the applicability of our tunable sources for TPFM uses, we used a home-built scanning microscope with a customized processing and acquisition unit (Jade Bio, SouthPort Co.) and a 40X 1.15NA objective lens (CFI Apo LWD Lambda S 40XC WI, Nikon). We separated the epi-detected signals from the excitation by a longpass dichroic beam splitter (DMLP650R, Thorlabs). The fluorescence signals were isolated using the corresponding bandpass filters and detected by PMTs (C9305-03, Hamamatsu). We acquired all the images under a 400µm×400µm field of view (FOV) with 1024×1024 pixels, and the pixel dwell time is four µs without frame averaging. We also collected the emission spectra by an uncooled spectrometer (SE2030, OTO) with an exposure time of 5 seconds.
We employed the resulting tunable sources to obtain multi-color fluorescence contrasts from the commonly used labeling tools in bioscience, as shown in Fig. 6. We fine-tuned the spectral peaks to match the excitation cross-sections, and we controlled the illumination power by a neutral density filter before the microscope. Figure 6(a) shows the DAPI-labeled nucleus in mouse brain pons, which indicates the location and amount of DNA in the brainstem. DAPI is a blue fluorophore, and it is mostly used to label DNA for nucleus indication because of its high photostability and affinity to the targets. The dots in Fig. 6(a) indicate the nucleus, enabling biologists to monitor cell differentiation and distribution. The signals were excited by the blue spectrum in Fig. 6(f) with a 19-mW excitation power after the objective lens, and the emission spectrum is correspondingly shown in Fig. 6(g), isolated by a 60-nm bandpass filter (FF02-447/60, Semrock). Figure 6(b) is the image of brain neurons from a Thy1-GFP transgenic mouse. GFP is a widely used fluorescent protein for various tissues and cell labeling through genetic expression. The GFP signals indicate the distribution of neurons and axons and the connection of neural networks, providing information about the nervous system if the mice were given specific stimulations. We used the 921-nm spectral lobe with 21-mW power after the objective lens, shown as the green spectrum in Fig. 6(f). We also showed the signal spectrum in Fig. 6(g) in green color using a designed bandpass filter for GFP uses (MF 525-39, Thorlabs). Figure 6(c) shows neurons and axons in the dorsal root ganglion (DRG) of a mouse labeled with tdTomato, a commonly used orange-red fluorescence designed for deep-tissue in vivo imaging as an alternative for GFP. The excitation and emission spectra were respectively shown as the orange lines in Fig. 6(f) and Fig. 6(g), isolated by a bandpass filter (FF01-607-60-25, Semrock), and here we used the direct output of the Yb:fiber laser as the excitation source with a 28-mW power. Figure 6(d) is the image of MCF-7 cell, a kind of breast cancer cell line, and the sample was labeled with SYTOX Deep Red to present target nucleic acids. We employed the 1236-nm spectral lobe, shown as the red spectrum in Fig. 6(f), with an excitation power of 17 mW. The emission spectrum is also shown as the red line in Fig. 6(g), isolated with both a 600-nm longpass filter (FEL0600, Thorlabs) and a 700-nm shortpass filter (FESH700, Thorlabs). The tumor cells grew in a high density with rough surface morphology, as shown in Fig. 6(d). To observe endogenous contrasts, we also demonstrated two-color autofluorescence imaging from banyan leaves, as shown in Fig. 6(e), and the blue and red signals were simultaneously excited using the 740-nm lobe (under 19 mW) and the 1236-nm lobe (under 12 mW), respectively. The autofluorescence may come from different chemical compounds, but we can still deduce the primary emission sources. The red fluorescence is predominantly derived from Chlorophyll, which is the key for photosynthesis, and the signals were isolated using the same filters used for SYTOX Deep Red. The blue fluorescence would emanate from NADH, an important associated product of photosynthesis and metabolism, and the blue signals were collected after the same filters used for DAPI.
According to the results of Fig. 6(a-e), the required power level for TPFM is well below the fiber output, even with the consideration of the microscope system loss. Our demonstrations have provided enough imaging details in biological tissues, such as axons and roughness of tissue surfaces. The efficient contrast generation benefits from both the fine spectral tunability and optimized dispersion compensation of our resulting sources.

Conclusion
In conclusion, we have optimized and produced the tunable femtosecond sources ranging from 740 nm to 1236 nm via the SPM-dominated spectral broadening. The filtered spectra lobes featured sub-100-fs pulse durations directly at the fiber output with the conversion efficiencies from 15% to 28%. Besides, this work provided a detailed discussion on the optimization of the spectral broadening via fiber nonlinearity. By investigating the roles of different parameters (i.e., the input pulse duration, the damage threshold, and the fiber length) during the nonlinear pulse evolution, we optimized the spectral tunability of SESS, aiming to replace Ti:S lasers for wide TPFM applications. We employed a Yb:fiber laser and a millimeter-long PCF with a slightly positive dispersion profile over the spectral tuning range, and we showed that the power-handling capacity of the fiber is related to the average power instead of the peak power. We thus optimized the spectral tuning range of SESS by using shorter input pulses. On the other hand, when the input pulses are excessively short (i.e., less than 50 fs in our case), the early onset of wave breaking due to the asymmetric pulse reshaping from self-steepening also limits the tuning range in the normal dispersion regime.
Moreover, we demonstrated the pulse compression down to its TL duration of 14 fs, more than four times shorter than the uncompressed duration, at the leftmost spectral lobe around 740 nm with negligible losses. The spectral tunability and compressibility lead to optimizing the TPF efficiency and reducing the potential photodamage and photobleaching. Besides, employing a short pumping pulse of 85 fs and a millimeter-long fiber for spectral broadening in the normal dispersion regime has significantly reduced the excessive intensity noise during the generating process. As a result, we have shown that the RIN of our resulting sources was close to the noise property of the driving source, which is beneficial for applications demanding high dynamic range detections, such as metrology applications [44], SRS microscopy [45], pump-probe experiments [46], and OCT [47]. Furthermore, the long-term stability of a light source is of great importance for practical uses, and the demonstrated femtosecond sources featured power variations of less than 1% for more than 8 hours under a proper active stabilization of the fiber alignment. To demonstrate the feasibility of our tunable fiber sources for TPFM, we have obtained clear images labeled with DAPI, GFP, tdTomato, and SYTOX Deep Red, covering the excitation range from UV-absorbing fluorophores to deep red dye stains. In addition, we have also demonstrated simultaneous two-color TPFM from endogenous contrasts, and we would like to note that using the tunable leftmost and rightmost lobes and the original laser pulses enables simultaneous three-color TPF imaging for respectively blue, red, and yellow fluorescences, such as for TPF brainbow imaging [1]. We believe that this tunable femtosecond laser system can be a robust, quiet, and cost-effective alternative to femtosecond Ti:S lasers in nJ-level applications, especially for biomedical studies using TPFM.