Wavelength-tunable or frequency-swept lasers have proven instrumental across a wide scope of applications, including spectroscopic sensing,1 biological imaging,2 and light detection and ranging (LIDAR).3 In spectroscopy, combining a tunable laser with a photodetector creates a simple yet highly versatile instrument, which obviates the need for dispersive optics and sensor arrays in spectral analysis, delivering unmatched spectral resolution compared with traditional dispersive spectrometers. Notably, frequency-swept spectroscopy surpasses Fourier-transform spectroscopy in both signal-to-noise ratio (SNR) and measurement accuracy. Fourier-transform techniques suffer from penalties associated with multiplexed detection4 and instrumental line shape distortions arising from the mechanical constraints of a moving arm.5 The advent of optical frequency combs and dual-comb spectroscopy5^–10 mitigates some of these problems by enabling motionless multiheterodyne detection using two frequency combs but remains constrained by limitations in SNR and sensitivity.4 In the realm of LIDAR, integrating a tunable laser with dispersive optics for spectral–spatial encoding enables inertia-free parallel distance metrology3 and facilitates multidimensional imaging.11 These advancements underscore the pivotal role of tunable lasers in driving progress across diverse scientific and technological domains.12

For many of these applications, accessing the mid-infrared (mid-IR) regime (typically 2 to $20\mu \mathrm{m}$) is crucial. In spectroscopy, molecular fundamental ro-vibrational transitions are prominent in the mid-IR, enabling strong light absorption and enhanced sensitivity for chemical sensing.13 In LIDAR, the mid-IR spectrum encompasses two critical atmospheric transmission windows (3 to $5\mu \mathrm{m}$ and 8 to $12\mu \mathrm{m}$),14^,15 which support long-range applications with reduced attenuation. For optical imaging, mid-IR light can penetrate materials opaque to visible or near-IR wavelengths, such as plastics, making it invaluable for imaging through barriers or within subsurface structures.16 Despite these advantages, mid-IR tunable sources remain less developed compared with their visible/near-IR counterparts. Challenges such as limited tuning speed and constrained spectral coverage hinder their broader adoption in these applications.

As the workhorses of this domain, quantum cascade lasers (QCLs) are available across an exceptionally broad spectral range, spanning the mid-IR to terahertz regimes.17^,18 However, individual QCLs cover only a narrow portion of the spectrum and are limited in tuning speed or scan rates (typically 10 s of kHz).19^,20 Moreover, although most QCLs operate in continuous-wave mode—achieving tremendous success in frequency-swept spectroscopy—they face challenges in time-resolved or nonlinear measurements.21 Recent advancements in QCL-based frequency comb technology show great promise in addressing some of these limitations.22^,23 Alternatively, nonlinear optical techniques such as optical parametric oscillation (OPO)24^,25 and difference-frequency generation (DFG)26^–28 can produce mid-IR light with a broad tunable range in both continuous-wave and pulsed modes. However, neither mechanical adjustments (e.g., changing the OPO cavity length or the incident angle) nor stationary measures (e.g., tuning pump wavelengths via laser diode currents or using acousto-optic tunable filters16) are sufficiently time-efficient. As a result, ultra-rapid, broadband spectral tuning remains challenging, particularly in the critical mid-IR region.

In this paper, we introduce a spectral-focusing DFG scheme to address this challenge. Spectral focusing, commonly employed to enhance the spectral resolution in coherent Raman spectroscopy,29^,30 is adapted here—coupled with controllable asynchronous pulses—for mid-IR light generation. Innovatively, this approach allows fast spectral tuning on a pulse-to-pulse basis. In particular, our approach generates a time-domain sequence of mid-IR pulses, each with a specific center wavelength (or “color”) and a narrow bandwidth ($\sim 8{\mathrm{cm}}^{-1}$), spanning a spectral range from 2600 to $3780{\mathrm{cm}}^{-1}$. Experimentally, our setup comprises a broadband mode-locked fiber laser and a frequency-modulated electro-optical (EO) laser comb. These lasers operate with slightly different repetition frequencies, akin to a dual-comb system. A critical feature of our scheme is the modulation of the EO comb’s repetition frequency, enabling effective control of the asynchronous time delay. This modulation facilitates rapid spectral tuning at a scan rate of up to $2\text{Mscans}/\mathrm{s}$. From an application standpoint, the mid-IR source supports broadband spectral measurements ($>1000{\mathrm{cm}}^{-1}$) within a few microseconds, or even faster. We demonstrate its capabilities through the analysis of chemical liquids, showcasing its potential for high-speed, high-resolution spectroscopic applications.

In our approach [Fig. 1(a)], mid-IR pulses are generated via DFG between two trains of chirp-matched pulses—referred to as the pump and signal pulses—in a chirped-poling lithium niobate (CPLN) crystal, which supports broadband frequency conversion. Notably, our method integrates two key techniques.

First, a spectral focusing technique is employed to narrow the spectral bandwidth of the mid-IR pulses,31^,32 thereby defining the resolution limit for spectroscopic applications. The principle of spectral focusing is illustrated in Fig. 1(b). For a linearly chirped pulse, the instantaneous central frequency $f(t)$ can be expressed as $$f(t)=f_c+\beta ·t,$$where the chirp parameter $\beta$ describes the linear slope of the central frequency ($f_c$) and $t$ is the time. In DFG, a mid-infrared pulse is generated at the instantaneous frequency difference, $f_D(\tau )$, which linearly changes with the relative time delay ($\tau$) between the pump and the signal pulses as $$f_D(\tau )={\mathrm{\Omega }}_0+\beta ·\tau ,$$where ${\mathrm{\Omega }}_0$ is the pump and signal center frequency difference. By selecting signal and pump lasers spectrally centered at $\sim 1.03\mu \mathrm{m}$ (280 THz) and $1.55\mu \mathrm{m}$ (192 THz), respectively, we generate idler pulses in the $3\text{-}\mu \mathrm{m}$ region (${\mathrm{\Omega }}_0\sim 90\mathrm{THz}$). The bandwidth of each pulse is determined by the degree of chirp matching between the pump and signal pulses.

Second, an asynchronous time-delay scheme is utilized to tune the idler frequency ($f_D$) in an inertia-free manner.33 As shown in Fig. 1(c), the pump and signal lasers have slightly different repetition frequencies ($f_r$ and $f_r+\mathrm{\Delta }{f}_r$, respectively), causing their relative delay ($\tau$) to change linearly with a step size of $\mathrm{\Delta }{f}_r/f_r^2$. However, this scheme results in a poor duty cycle, meaning the effective time window, during which the two pulses that temporally overlap are much smaller than those in the scan period ($T=1/\mathrm{\Delta }{f}_r$), as shown in Fig. 1(c). The primary reason is that the pulse durations of the two lasers are much shorter than their repetition periods. This leads to significant dwell time and limits the refresh rate of spectral measurements.34^,35

To address this issue, we modulate the signal’s repetition frequency between $f_r-\mathrm{\Delta }{f}_r$ and $f_r+\mathrm{\Delta }{f}_r$ at a modulation frequency of $f_m$, as shown in Fig. 1(d). Consequently, we can scan $\tau$ within the pulses’ overlapping region, achieving a scan time equal to the effective time, or a 100% duty cycle. This approach, which has proven successful in dual-comb spectroscopy,29^,30 has not yet been explored for boosting the scan rate of a tunable source. An additional advantage of our scheme lies in the use of a cavity-free EO comb, with a repetition frequency that can be rapidly modulated by a radio-frequency synthesizer at speeds far exceeding those of mode-locked lasers.29

As shown in Fig. 2(a), our mid-IR source utilizes a broadband fiber laser as the pump light and an EO laser as the signal light. The pump light originates from a mode-locked Yb-doped fiber oscillator at $1.03\mu \mathrm{m}$. With the help of an intracavity piezoelectric actuator and a servo system, the repetition rate is stabilized at 60.5 MHz with a standard deviation of $<3\mathrm{mHz}$. This laser contains a ytterbium-doped fiber amplifier that boosts the average power to $\sim 1.6\mathrm{W}$. The output pulses are compressed to $\sim 120\mathrm{fs}$ using a couple of transmission gratings ($1250\text{lines}/\mathrm{mm}$). These pulses are then focused into a 10-cm-long photonic crystal fiber (PCF; SC-3.7-975, NKT Photonics, Birkerød, Denmark) using an aspheric lens (C330TMD-B, Thorlabs, Newton, New Jersey, United States, $f=3.1\mathrm{mm}$). This setup generates a spectral range exceeding $1300{\mathrm{cm}}^{-1}$ (from 1000 to 1150 nm) with an average power of $>200\mathrm{mW}$, as shown in Fig. 2(b). A pair of prisms is used for precise chirp control.

In the EO laser, a narrow-linewidth continuous-wave (CW) laser (linewidth $<10\mathrm{kHz}$, 100 mW; ID Photonics, Ottobrunn, Germany) is sent into an intensity modulator (IM) with a 40-GHz bandwidth (KY-MU-15-DQ-A, Keyang Photonics, Beijing, China). The IM is driven by a pulse generator (LaseGen) that produces 30-ps electrical pulses, triggered by a radio-frequency signal generator (DSG3060, Rigol, Beijing, China). As a result, the IM outputs 30-ps optical pulses. After amplification by a custom-built erbium-doped fiber amplifier (EDFA), the optical power is increased to $\sim 350\mathrm{mW}$. The output undergoes spectral broadening in a segment of highly nonlinear fiber (HNLF) with a linear dispersion slope of $0.03\mathrm{ps}/({\mathrm{nm}}^2·\mathrm{km})$ and a nonlinear coefficient of $\sim 10{\mathrm{W}}^{-1}·{\mathrm{km}}^{-1}$. This is followed by dispersion compensation using a long polarization-maintaining fiber. Finally, the EO laser is further amplified to $>0.5\mathrm{W}$ using a double-clad EDFA.

The net group velocity dispersion for each laser is carefully managed to ensure that the pump and signal pulses are linearly dispersed with matched chirps before being used for DFG. For a highly chirped Gaussian pulse, the linear chirp parameter $\beta$ can be estimated as $\beta \approx \mathrm{\Delta }f/\mathrm{\Delta }\tau$, where $\mathrm{\Delta }f$ is the spectral full width at half-maximum (FWHM) and $\mathrm{\Delta }\tau$ is the pulse duration. As shown in Fig. 2(b), the pump spectral FWHM is $\sim 11\mathrm{THz}$, and the signal FWHM is $\sim 0.09\mathrm{THz}$. The uneven pump spectrum arises from nonlinear effects such as self-phase modulation and four-wave mixing during PCF broadening. However, its impact on our spectral measurements is negligible, as background normalization cancels this variation. The measured autocorrelation traces of the pump and signal pulses, shown in Fig. 2(c), indicate a pump pulse duration of $\sim 90\mathrm{ps}$ and a signal pulse duration of $\sim 750\mathrm{fs}$. Consequently, $\beta$ is estimated to be $\sim 0.12\mathrm{THz}/\mathrm{ps}$ for both lasers.

To produce mid-IR pulses, the chirp-matched pump and signal beams are spatially overlapped using a dichroic mirror (DMLP1180, Thorlabs) and then focused into a 2-cm-long chirped-poling lithium niobate (CPLN) crystal (Castech, Fuzhou, China) for DFG. The CPLN crystal features a linearly increasing poling period (from 23 to $31\mu \mathrm{m}$), facilitating adiabatic quasi-phase matching across a broad spectral range. The idler beam is collimated and filtered through a long-pass filter (cutoff wavelength at $2.6\mu \mathrm{m}$), which blocks the fundamental beams. Note that the pump fiber laser and signal EO comb are referenced to the same 10-MHz clock generated by a hydrogen maser.

We confirm the mid-IR spectral tuning range by operating the pump and signal lasers in synchronized mode. By adjusting their relative pulse delay, we measure the mid-IR spectra using a standard Fourier-transform infrared spectrometer (FTIR, resolution of $0.1{\mathrm{cm}}^{-1}$, PerkinElmer, Waltham, Massachusetts, United States). Figure 2(d) exemplifies several recorded spectra, spanning from 2600 to 3700 nm (or 2703 to $3704{\mathrm{cm}}^{-1}$). The FWHM of each tuning spectrum is $\sim 8{\mathrm{cm}}^{-1}$, indicating the spectral width of a mid-IR pulse achieved in asynchronous mode. This also sets a resolution limit for molecular spectroscopy using our source. In addition, the maximum output power of 1 mW is achieved at $3100{\mathrm{cm}}^{-1}$ [blue dots in Fig. 2(d)], corresponding to a maximum energy per spectral element of 16.5 pJ.

We then demonstrate the capability of our source for broadband molecular fingerprinting. To this end, the mid-IR light is directed into a thin cuvette filled with chemical liquids and subsequently detected by a single mid-infrared detector (bandwidth of 250 MHz, Vigo, Warsaw, Poland). The detector output is recorded by a digital oscilloscope (bandwidth of 1 GHz, LeCroy, Chestnut Ridge, New York, United States) with a sample rate of $500\mathrm{MS}/\mathrm{s}$. Figure 3(a) shows two mid-IR pulse trains recorded with and without a liquid chemical sample (a mixture of benzene and ethanol in a 2:1 concentration ratio). For these measurements, the repetition frequency of the EO laser is set to $302.5\mathrm{MHz}+14\mathrm{kHz}$, close to the fifth harmonic of the fiber laser’s repetition frequency. The EO laser’s overall performance is optimal at this frequency. Consequently, each recorded time-domain trace lasts $\sim 6.3\mu \mathrm{s}$, corresponding to a full spectral coverage of $1180{\mathrm{cm}}^{-1}$ (or 35.4 THz), determined by the pump spectral width. The available bandwidth is defined as the range where the pulse peak SNR exceeds 10. This results in a spectral tuning speed of $5.6\mathrm{THz}/\mu \mathrm{s}$, far exceeding that of other advanced frequency-agile sources (e.g., $267\mathrm{GHz}/\mu \mathrm{s}$ reported in the near-IR12). As detailed in Fig. 3(b), the chemical fingerprints are encoded on the pulses. Each pulse corresponds to a spectral element or point of the mid-IR spectrum. A time-encoded pulse sequence contains over 380 pulses or spectral points, resulting in a spectral point spacing of $\sim 3.1{\mathrm{cm}}^{-1}$. This is consistent with the result ($3.08{\mathrm{cm}}^{-1}$) derived from the relationship between $f_D$ and $\tau$. Specifically, the asynchronous temporal interval ($\mathrm{\Delta }{f}_r/f_r^2$) is 0.77 ps, and $\beta =0.12\mathrm{THz}/\mathrm{ps}$ or $4{\mathrm{cm}}^{-1}/\mathrm{ps}$. Note that when using our tunable source for spectral measurements, the spectral resolution is set by the upper bound of the point spacing and the pulse spectral width (e.g., $\sim 8{\mathrm{cm}}^{-1}$).

To validate our spectral results, we convert the temporal data into the spectral domain (in wavenumbers) using the calculated spectral point spacing ($3.08{\mathrm{cm}}^{-1}$). In addition, we calibrate the absolute wavenumber axis in advance using known absorption line centers of neat benzene. By normalizing the peaks of the mid-IR pulses against the background, we reconstruct the absorption spectrum of the chemical mixture. Figure 3(c) shows a portion of the normalized spectral data (red dots) on the calibrated wavenumber axis. Our results, which reveal the absorption bands of benzene (e.g., around $3072{\mathrm{cm}}^{-1}$) and ethanol (around $2900{\mathrm{cm}}^{-1}$), align well with the absorption spectrum [black curve in Fig. 3(c)] measured by the FTIR spectrometer (spectral resolution of $8{\mathrm{cm}}^{-1}$). Notably, our data are obtained in a single-shot measurement of $\sim 6.3\mu \mathrm{s}$, whereas the FTIR measurement takes $\sim 200\mathrm{ms}$.

Moreover, data averaging improves the SNR of a spectral element, defined as the ratio between the pulse peak and the standard deviation (SD) of the noise floor. The SNR averaged over the spectral FWHM in a single-shot measurement is $\sim 100$, which improves to 3200 with 1000-fold averaging [Fig. 3(d)]. The fitting line in Fig. 3(d) indicates that the SNR evolves as the square root of the number of averages. Our setup exhibits good stability and reproducibility, with spectral variations remaining below $2.8{\mathrm{cm}}^{-1}$ and relative intensity fluctuations within $\pm 1.5\%$. However, for extended measurements lasting several hours, re-calibration of the wavenumber axis becomes necessary to compensate for gradual drifts.

The broad spectral coverage allows our approach to interrogate molecular fingerprints of various organic and biological samples. For instance, the reconstructed absorption spectra of a dimethyl sulfoxide (DMSO) solution [Fig. 4(a)] and an ethanol solution [Fig. 4(b)], with volume fractions of $\sim 0.5\%$ and 1%, respectively, diluted in carbon tetrachloride, are presented in the range of 2800 to $3200{\mathrm{cm}}^{-1}$. The spectral results generally agree with the data obtained from the FTIR spectrometer. In addition, we measure absorption spectra of biological samples, such as flaxseed oil [Fig. 4(c)], in the spectral range from 2600 to $3780{\mathrm{cm}}^{-1}$. The margin of error in Fig. 4(c) represents the fluctuations from 200 consecutive measurements. The FTIR data (black curve) are displayed for comparison.

A key merit of our source lies in the frequency-modulated EO comb, which enables high scan rates for tuning mid-infrared wavelengths. For instance, spectral tuning across $\sim 1000{\mathrm{cm}}^{-1}$ at a scan rate of $2f_m=600\mathrm{kHz}$, i.e., $600\text{kScans}/\mathrm{s}$, is shown in Fig. 5(a). The scan rate can be further increased to $2\text{Mscans}/\mathrm{s}$ with $2f_m=2\mathrm{MHz}$, as shown in Fig. 5(b), at the cost of reduced pulse number and, consequently, an enlarged spectral point spacing. For example, there are $\sim 100$ pulses, spaced by $10{\mathrm{cm}}^{-1}$, for one scan in Fig. 5(a), and $\sim 30$ pulses, spaced by $33{\mathrm{cm}}^{-1}$, in Fig. 5(b). Nonetheless, the maximum tuning speed achievable with our setup reaches $\sim 2000{\mathrm{cm}}^{-1}/\mu \mathrm{s}$ or $\sim 60\mathrm{THz}/\mu \mathrm{s}$, which is higher than that of a time-stretch mid-IR source (e.g., $19{\mathrm{cm}}^{-1}$ at $50\text{MScans}/\mathrm{s}$, or $\sim 30\mathrm{THz}/\mu \mathrm{s}$26). These capabilities make our mid-IR source highly promising for multiplexed spectral interrogation of chemical dynamics on the $\mu \mathrm{s}$ scale.36 In addition, our source has immediate applications in multiplexed spectral imaging,16 optical coherence tomography,37 and parallel range measurements.3

In particular, for spectral measurements, the frequency-modulated method we used offers significant advantages over conventional asynchronous schemes. In the conventional case [Fig. 5(c)], the scan rate (set by $\mathrm{\Delta }{f}_r$) is directly linked with the spectral point spacing. When the spacing exceeds the pulse spectral width [e.g., indicated by point A in Fig. 5(c)], the continuous increase of $\mathrm{\Delta }{f}_r$ degrades spectral resolution (black line), while simultaneously reducing the effective number ($N$) of spectral elements (blue line). Here, $N$ is defined as a ratio between a full spectral width and spectral resolution. When $N=1$, only one spectral element is measured.

By contrast, in our case, the scan rate is determined by $2f_m$, which is independent of spectral resolution, as indicated by the black line in Fig. 5(d). This means that a high scan rate can be achieved without sacrificing spectral resolution, provided other parameters (e.g., $\mathrm{\Delta }{f}_r$ and $\beta$) remain unchanged. However, when the scan rate reaches a point [point B on the blue line in Fig. 5(d)] where the scan duration equals the effective time window [see Fig. 1(d)], any further increase in the scan rate narrows the spectral tuning range, causing a reduction in $N$. Nonetheless, for spectroscopic applications, the frequency-modulated scheme can provide a scan rate that is several orders of magnitude higher than that of the unmodulated case while preserving the spectral resolution and the number of spectral elements.

Currently, the finest spectral resolution offered by our mid-IR source is $\sim 8{\mathrm{cm}}^{-1}$, limited by the spectral focusing process. This process can be improved through refined dispersion management of the pump and signal pulses, including compensation of high-order dispersion.38 Nevertheless, this resolution is sufficient for characterizing condensed samples, which typically exhibit broad absorption bands rather than sharp absorption lines, as seen in gas-phase samples.

Our tunable mid-IR source and its application in spectroscopy share a similar architecture to dual-comb spectroscopy, utilizing two asynchronous lasers and a photodetector. Dual-comb spectroscopy is a transformative technique that offers unparalleled spectral resolution, precision, and accuracy for rapid broadband molecular spectroscopy.4^–10^,39 However, it requires two mutually coherent combs, which presents technical challenges6 and adds complexity to the system. Furthermore, achieving broadband ($>1000{\mathrm{cm}}^{-1}$) mid-IR spectra within a few microseconds remains a significant challenge for dual-comb spectrometers, primarily due to Nyquist limitations.5 By contrast, our approach circumvents the need for coherent comb generation and phase stabilization, enabling wavelength-tunable spectral detection directly in the time domain without relying on Fourier transformation.

Another technique with similarities to our approach is photonic time-stretching. In this method, a laser pulse is temporally dispersed, establishing an intra-pulse time-wavelength correspondence for ultrafast spectral analysis and signal processing.40 Recent demonstrations of mid-IR time-stretch lasers for spectroscopy have enabled single-shot spectral measurements on the nanosecond timescale, with scan rates at the pulse repetition frequency (typically tens of MHz26^,36^,41^,42). However, the pulse energy is dispersed, limiting the SNR per spectral element (e.g., a peak SNR of 14 reported in Ref. 26), and the spectral width in these systems is typically restricted to tens of ${\mathrm{cm}}^{-1}$. Furthermore, these systems face significant challenges in high-speed mid-IR detection (e.g., bandwidth of $>10\mathrm{GHz}$36) as well as data storage and processing. By contrast, our source enables broadband mid-IR spectral tuning ($1180{\mathrm{cm}}^{-1}$) on a pulse-to-pulse basis, offering a high SNR ($\sim 100$) while alleviating the burdens on detection and digitization, as evidenced by our use of a 250 MHz detection bandwidth and a moderate sampling rate ($500\mathrm{MS}/\mathrm{s}$).

Finally, with recent advancements in nanophotonics, particularly the development of on-chip electro-optical combs,43 our approach holds the potential to lead to compact, wavelength-swept laser sources with broad applications across chemistry, biology, and materials science.

In summary, we present a method for ultra-rapid, broadband wavelength sweeping in the mid-IR region, utilizing spectral focusing DFG between a broadband fiber laser and an asynchronous, frequency-modulated EO comb. This approach bypasses mechanical scanning limitations, enabling automatic spectral tuning at $5.6\mathrm{THz}/\mu \mathrm{s}$ and pulse-to-pulse spectral encoding. Applied to spectroscopic sensing, it offers broad spectral detection ($1180{\mathrm{cm}}^{-1}$ or 35.4 THz) with moderate resolution ($8{\mathrm{cm}}^{-1}$) and rapid measurement time ($\sim 6.3\mu \mathrm{s}$). The cavity-free EO comb facilitates fast repetition frequency modulation, achieving high scan rates (up to 2 MHz, corresponding to a tuning speed of $\sim 60\mathrm{THz}/\mu \mathrm{s}$), albeit with reduced resolution ($33{\mathrm{cm}}^{-1}$). Despite this, our method shows promise for broadband spectral sensing of condensed-phase samples and ultra-rapid mid-IR wavelength-space encoding, with potential applications in chemical monitoring, mid-IR flow cytometry, biomedical imaging, and scanner-free LIDAR.

Biographies of the authors are not available.