Few-cycle pulse generation with short pulse duration and broad spectrum has been studied and developed over the past few decades.1^–3 It is becoming a significant tool for attosecond science,4 extreme ultraviolet and X-ray generation,5 and high harmonic generation.6 In particular, few-cycle pulses at ∼2μm have attracted much attention for their unique property of directly generating mid-infrared optical frequency combs, which contain a significant “molecular fingerprint” regime.7^,8 Therefore, these combs have promising applications in mid-infrared spectroscopy, biological sensing, and medical techniques.

For the generation of 2-μm laser pulses, a common way to obtain high-power ultrashort pulses is to utilize chirped pulse amplification. By stretching the pulses before amplification, the peak power of the pulses is significantly reduced, thereby suppressing nonlinear effects in the fiber. This approach effectively maintains the pulse quality during compression.9 However, the narrow spectral bandwidth fundamentally limits pulse compressibility, preventing the attainment of few-cycle durations despite optimal dispersion management.10 Contrary to the strategy of avoiding the nonlinear effects, the nonlinear self-compression method benefits from the spectral broadening induced by the nonlinearity and combines the processes of broadening and compression in the same fiber. In 2018, this approach obtained 55-fs pulses with a record average power of 20 W utilizing a thulium (Tm)-doped large-pitch fiber.11 In the following year, a smaller core photonic crystal fiber generated remarkably short pulses with a duration of 13 fs.12 Though these results are attractive, the complexity of the experimental structures and the careful control of fiber parameters prevent them from being used in more applications. Multistage nonlinear compression is an effective method to simplify the experimental configuration. Although an impressive all-fiber system has obtained single-cycle pulses (6.8 fs) through a two-stage nonlinear compressor seeded by 2-μm Raman soliton self-frequency-shifted pulses,13 the current performance ceiling for direct mode-locked Tm-doped fiber lasers remains constrained to 29-fs pulses.14 This significant performance gap highlights the great challenges in generating few-cycle pulses directly from Tm-doped fiber laser systems.

In this paper, we present, to our knowledge, the shortest few-cycle pulses generated from nonlinear pulse compression based on a dispersion-managed mode-locked Tm-doped fiber laser operating at a repetition rate of ∼199.74MHz. Sub-three-cycle pulses (19.8 fs, 3.4 nJ) are obtained with a pre-chirp-managed nonlinear amplifier and double-stage nonlinear compressor. Notably, the whole system is compact and robust, benefiting from the all-fiber configuration.

Figure 1 illustrates the experimental configuration, consisting of four main parts: an oscillator, a pre-chirp-managed nonlinear fiber amplifier, and a double-stage nonlinear compressor. The mode-locked Tm-doped fiber laser is based on nonlinear polarization evolution. A commercial 1560 nm laser source serves as the pump through an integrated wavelength division multiplexer (IWDM, 1565/2000 nm). The IWDM contains an isolator (ISO) to ensure unidirectional transmission and minimize the cavity length. Two quarter-wave plates (QWPs) and a half-wave plate (HWP) in combination with a polarization beam splitter (PBS) implement nonlinear polarization evolution-based mode-locking, whereas the reflection port of the PBS is used as the output. A 15-cm highly doped Tm fiber (SM-TSF-5/125, Nufern, East Granby, Connecticut, United States) serves as the gain medium, with a group velocity dispersion (GVD) value of −0.051ps2/m at 2μm and an absorption coefficient of (340±50)dB/m at 1565 nm. The pigtails of all devices are the single-mode fiber (SMF-28, Corning, New York, United States) with a length of 22 cm and a GVD value of −0.071ps2/m at 2μm. As all the fibers are of negative dispersion at 2μm, a 53-cm piece of ultrahigh numerical aperture fiber (UHNA4, Nufern) with a normal GVD value of +0.091ps2/m is used to compensate for the anomalous dispersion. Along with a 15-cm free-space path, the laser operates in the dispersion-managed soliton regime with a net dispersion value of 0.025ps2. The total optical length of the cavity is 1.5 m, corresponding to a repetition rate of 199.74 MHz. By carefully setting the pump to 1.44 W, a stable and sustained single-pulse operation is obtained with an average output power of 130 mW corresponding to a pulse energy of 0.65 nJ. Coupled from the reflection port of the PBS, the seed enters an ISO and then an output coupler (OC), where 10% of the power is monitored and 90% propagates into the pre-chirp-managed nonlinear amplifier.

The pre-chirp-managed nonlinear amplifier comprises a stretcher and a double-clad Tm-doped fiber (DC-TDF, Nufern), backward pumped by a commercial 793 nm pump laser.15 A 2.5-m length of UHNA4 is employed as the stretcher to achieve better amplification and compression performance. The stretched pulses are then compressed by a 32-cm piece of SMF while stripping the residual pump. Linear compression within the SMF occurs solely through dispersion; therefore, the pulse duration is indeed limited by the transform limit of the spectrum.

To obtain the shortest possible pulse duration, nonlinear effects, typically self-phase modulation (SPM), should be considered to obtain a broadband spectrum. The nonlinear compressor used in this study comprises two stages. In the first stage, a 38-cm piece of UHNA4 is applied to slightly broaden the spectrum, followed by a 32-cm-long SMF that compensates for the normal dispersion to compress the pulse duration. Given that the power dramatically determines the SPM, we carefully adjust the efficiency of fiber splicing to 1.7-dB losses, providing high-power and short-duration pulses to seed the second-stage compression in the short piece of standard highly nonlinear optical fiber (HNLF, OFS, Norcross, Georgia, United States). Nonlinear self-compression accompanied by significant spectral broadening enables the few-cycle pulse generation. Finally, an off-axis parabolic mirror is used to collimate the output pulse.

Figure 2(a) shows the spectrum of the oscillator, represented by the blue line and measured with an optical spectrum analyzer (AQ6375B, Yokogawa, Musashino, Japan), covering 1200 to 2400 nm. The smooth spectrum confirms the operating state of the dispersion management soliton. The asymmetrical position of side lopes is induced by modulation instability, which is a typical characteristic of dispersion-managed lasers with positive net cavity dispersion.16 Besides, the birefringence and gain profile of the laser cavity could be attributed to the asymmetric intensity distribution of the sidebands.17 The corresponding Fourier-transform limit of this spectrum is 57 fs. Further characterizations of the operating status are performed with a spectrum analyzer (N9340B, Agilent Technologies, Santa Clara, California, United States) and an oscilloscope (DSOV084A, Keysight, Santa Rosa, California, United States) with a 12 GHz InGaAs photo-detector (ET-5000F, EOT, Pendleton, Oregon, United States). Figures 2(c) and 2(d) illustrate the single-pulse operation and a repetition rate of 199.74 MHz, which aligns with the cavity length.

Figure 2(a) (red line) and Fig. 2(b) depict the spectral and time domain features of the amplifier, respectively. A notable red shift is observed due to the absorption feature of the Tm-doped gain fiber, and the decline in the spectrum width is attributed to the gain narrowing effect.10 If the seed is directly amplified without pre-chirp management, the high peak intensity may induce soliton splitting and cause damage to the whole amplifier system. Pre-chirp management dramatically influences the duration and quality of the amplifier, and this process drives the temporal pulse evolution from a Gaussian profile toward a hyperbolic secant (sech2) profile, a characteristic shape consistent with observations reported in Ref. 18. An insufficient UHNA4 fails to provide adequate chirp, resulting in soliton fission during amplification.19 Conversely, an excessive UHNA4 introduces significant positive chirp to the pulses, reducing nonlinearity during amplification and leading to the gain narrowing effect. By trimming the UHNA4 in 10-cm increments, we introduced appropriate pre-chirp, and 80.8-fs pulses were generated after compression in the SMF. The classic sech2 shape indicates sufficient compression, and the subtle pedestal is due to the higher-order dispersion to be compensated.

To further compress the pulses to a few cycles, nonlinear compression should be employed. Points A and B, marked in Fig. 1, represent the monitoring points of the two nonlinear compression stages. A comparison of the two spectra in Fig. 3(a) shows the slightly broadened spectrum at point A, whereas the spectrum at point B is significantly expanded to a 10-dB bandwidth of 437 nm. Two factors contribute to the different broadening behavior. One is that the HNLF has a higher nonlinear coefficient, and the other is attributed to the different widths of the pulses entering these two stages. Shorter input pulses lead to a higher peak power, further strengthening the nonlinear effects. To characterize the output performance more thoroughly, we utilized the well-known phase and intensity from correlation and spectrum only (PICASO) method to retrieve the pulse and the phase distribution,20 as shown in Fig. 3(b). From the red line in Fig. 3(b), we observe that the pulses from A are relatively clean, with only a tiny amount of energy shifting to the sideband, and the green line indicates some chirps resulting from the dispersion at the output. As calculated, it closely matches the transformed limit of 44 fs, indicating that the pulses are well-compressed and suitable to be the pump of the subsequent compression stage. The total loss in the second stage includes splicing loss and Fresnel reflection loss. By carefully discharging during fusion splicing and polishing the pigtail to an angle of 8 deg, the loss is controlled to be less than 1 dB, and the back reflection is effectively eliminated.

By optimizing the length of HNLF in a step of 2 mm, we achieved three-cycle pulse generation in a 2-cm HNLF, maintaining the balance between nonlinearity and anomalous dispersion in the fiber. Figure 3(c) shows that the interference autocorrelation (AC) trace exhibits sparse stripes, following an intensity ratio of 1 to 8. This indicates successful suppression of pulse width and well-organized dispersion management. The width of the PICASO retrieved pulse [blue line in Fig. 3(d)] is calculated to be 19.8 fs, equivalent to 2.9 optical cycles. The gray region, referring to the transformed-limited pulse, aligns well with the retrieved pulse, showing that our compression result is relatively close to the limit. The fragmented parts far from the main pulse may be derived from soliton splitting, which results in extremely high intensity in the HNLF fiber. To investigate the long-term stability of the few-cycle pulses, we monitored the output power and spectrum over a 2-h period. As shown in Fig. 4(a), the average power is ∼674mW, showing excellent stability with a root mean square (RMS) fluctuation of 0.781%. Figure 4(b) displays the evolution of the spectrum, spanning continuously from 1200 to 2400 nm. The spectrum demonstrates high stability, with RMS fluctuations of 0.2% in center wavelength and 2.87% in 10-dB bandwidth. Higher stability of the few-cycle pulses can be achieved by active thermal control and isolation from environmental disturbance.

To better understand the evolution within the double nonlinear compression stages, we performed numerical simulations based on the generalized nonlinear Schrodinger equation.21 The entire nonlinear compression stages are included in the simulation by considering the terms of high-order dispersion along with nonlinear effects consisting of self-steepening, SPM, and stimulated Raman scattering.22 The linear propagation loss is ignored because of the short fiber length. The equation is constructed as $$\begin{gathered} ∂A∂z=−α2−(∑k≥2βnin−1n!∂n∂Tn)A+iγ(1+1ω0∂∂T)× \\ [(1−fR)A|A2|+fRA∫0∞hR(τ)|A(z,T−τ)|2d]. \end{gathered}$$(1)The parameters used in the simulation are consistent with the experiment. The simulation is initiated with an anomalously chirped sech2-shaped pulse sharing the same characteristics as the output of the amplifier. The fourth-order Runge–Kutta method models the evolution within these fibers.23 The HNLF has a GVD value of −17.4ps2/km and a nonlinear parameter of 3.1W−1/km.

Figure 5 describes the simulation results in the first compression stage. In the segment of UHNA4, a monotonic broadening of the spectrum is observed, whereas the evolution in the time domain is significantly more complex. At the beginning of the UHNA4, dispersion and nonlinearity jointly shape the pulse profile, whereas nonlinearity primarily drives the broadening process in the frequency domain, as depicted in Fig. 5(a). The pulse is constantly narrowing in the middle of the UHNA4 as the anomalous dispersion from the amplifier is gradually compensated and reaches its shortest when the dispersion is completely compensated. However, as dispersion accumulates in the rest of the UHNA4, the pulse finally expands to a pulse width of 1.5 ps. Subsequently, the pulse is compressed to 55 fs to seed the second compression stage, and the spectrum remains unchanged due to the low nonlinearity of the SMF.

In Fig. 6(a), we note that a symmetrical broadening occurs at the beginning of the HNLF, which indicates that SPM dominates the evolution of the spectrum. This broadening is beneficial to the compression process.24 However, the spectra become asymmetric in the last 1 cm caused by self-steepening.25 When this effect is taken into account, the pulses undergo a time-domain shift, where the peak of the pulses moves more slowly than the wings. As a result, the peak of the pulses tends to shift toward the trailing edge, leading to the asymmetric temporal profile.26 The pulse will be narrowed when the anomalous dispersion in the HNLF completely compensates for the nonlinearity induced by the SPM effect. Figure 6(b) displays the corresponding pulse compression in the time domain, and the output of the HNLF is estimated to be 19.5 fs. Figures 6(c) and 6(d) describe the simulation outputs in the spectral and time domains, respectively. The dip at the center of the spectrum and the two peaks around the dip prove the strong SPM effect, and the asymmetrical side lobes in the time domain are caused by the nonlinear phase shift induced by self-steepening and higher order dispersion. The simulation results demonstrate quantitative agreement with the experimental data and intuitively reflect the evolution trend of the pulses. There are some deviations between the simulation and experimental results, primarily due to the unavailability of some key simulation parameters. For example, the loss of quartz fibers above 2μm and higher-order dispersion parameters of the high-nonlinearity fiber, for which we could not obtain data either from the literature or the supplier. Further increase of the HNLF length in the simulation leads to continuous spectral broadening but ceases pulse compression due to the accumulated dispersion. This confirms that the shortest pulse duration is achieved with this compression strategy.

We developed a compact and efficient all-fiber double-stage nonlinear compression system in the 2-μm wavelength range. By carefully balancing the dispersion and nonlinearity in the fiber and employing proper pre-chirp management, we generated 19.8-fs (2.9 cycles) pulses with a high repetition of 199.74 MHz and a maximum output power of 674 mW (corresponding to a pulse energy of ∼3.37nJ). To the best of our knowledge, our result demonstrates the shortest pulses generated from a dispersion-managed mode-locked Tm-doped fiber laser. We utilized a simulation model to illustrate the evolution in the double-stage nonlinear compression, and this model can be inversely applied to guide our experimental investigations, such as selecting pre-parameters. Besides, the all-fiber configuration of the amplifier and compressors makes the method accessible and straightforward. With its simple configuration and compatible pulse quality, we believe this could be a general method for generating few-cycle pulses. The spectrum of our few-cycle pulses lies at the edge of the mid-infrared region, which inherently makes it particularly suitable for applications in mid-infrared frequency metrology and molecular spectroscopy. Given an effective nonlinear frequency conversion process, i.e., intra-pulse different frequency generation and optical parametric amplification, we can generate mid-infrared pulses covering longer wavelength regions,13 thereby accessing major absorption bands of numerous gas molecules.27 In future work, we will commit to achieving an octave-spanning spectrum with a narrower pulse to realize high-quality mid-infrared optical frequency combs.28^–31

Acknowledgment. This work was supported by the National Natural Science Foundation of China (Grant Nos. 62227821 and 42461050) and the Key Project of the National Institute of Metrology, China (Grant Nos. AKYZD2411 and AKYZD2511-2).

Yan Wu received his BS degree in electronic science and technology (optoelectronics) from Tianjin University, Tianjin, China, in 2022. He is currently working towards his PhD in optical engineering specialty with the School of Precision Instruments and Opto-electronics Engineering, Tianjin University, Tianjin, China. His research interests include fiber lasers, nonlinear optics, and optical frequency combs.

Biographies of the other authors are not available.