1. Introduction

As technology continues advancing towards the midinfrared wavelength range, there are new opportunities for applications that make use of the direct interaction between midinfrared light and molecular vibrations such as trace gas sensing [1] and chemically-sensitive microscopy [2]. Widespread use of new applications will require simple, robust, and integrated sources. Fiber lasers are a compact laser type that can provide high average power and femtosecond pulses. Erbium and thulium doped fibers amplify light around 1.6 and 1.9 µm in wavelength, the energy difference of which is conveniently around 9 µm. We use a new type of quasi-phase-matched nonlinear semiconductor, orientation-patterned gallium phosphide (OP-GaP) [3], to enable efficient broadband difference frequency generation (DFG) between erbium and thulium outputs, resulting in over 60 mW of midinfrared power.

Difference frequency generation and fiber lasers are well suited for frequency comb generation. A frequency comb has a stable pulse repetition rate and carrier envelope offset (CEO) frequency. The repetition rate of a fiber laser can be controlled with an end mirror on a piezoelectric stack. The CEO of the difference frequency is automatically zero when the pump and signal seed have the same CEO, which we ensure by using a single oscillator. For a case where a non-zero CEO might be needed, such as an optical enhancement cavity [1], we add an acousto-optic (AO) frequency shifter to the signal seed, which also shifts the midinfrared output for a midinfrared frequency comb with a tunable CEO, without requiring CEO measurement or feedback. We interfere each pulse with its preceding pulse to observe stability and control of the interferogram, verifying that the midinfrared pulses are a frequency comb with tunable carrier envelope offset.

2. Difference frequency system

Fiber lasers are available in three main wavelength regions around 1, 1.6, and 2 µm, corresponding to ytterbium (Yb), erbium (Er), and thulium (Tm) or holmium (Ho) doping. Ytterbium fiber amplifiers offer very high powers, but the high photon energy can make multiphoton absorption in the nonlinear crystal a problem [4]. We chose to pump at erbium wavelengths, and to seed the signal wave at thulium wavelengths. The system is illustrated in Fig. 1. We start with an Er nonlinear amplifying loop mirror oscillator similar to [5] with a 93.4 MHz repetition rate. This oscillator includes an end mirror on a piezoelectric transducer for fast repetition rate control, and a piezoelectric stack for slow repetition rate control by fiber stretching. We stabilize the repetition rate to within about ±1 Hz by optical referencing to a diode laser with a specified short-term linewidth of 3 kHz. The diode laser slow drift is corrected by temperature adjustments based on the laser repetition rate as referenced against a GPS-disciplined Rb frequency standard. The oscillator CEO is not stabilized.

 

Fig. 1 Illustration of midinfrared difference frequency generation comb system. One erbium oscillator seeds both pump and signal seed arms, using supercontinuum (SC) generation to reach thulium wavelengths in the signal seed arm. Orientation-patterned gallium phosphide provides high efficiency conversion to the 6 to 11 µm range. The time delay is stabilized by monitoring nonlinear parasites. A large delay interferometer verifies the frequency comb stability.

Download Full Size | PDF

The oscillator pulses are split into a pump and a signal seed arm. The pump arm amplifies the beam to 1.6 W average power with 110 fs pulse duration by nonlinear amplification in large core Er fiber, which is scalable to high pulse energies [6]. The amplified spectrum has three peaks at around 1.54, 1.55, and 1.57 µm.

The signal seed is first amplified in erbium-doped fiber and recompressed for supercontinuum generation in highly nonlinear fiber. This broadens the spectrum past 2 µm for seeding a thulium fiber [7,8], with output of about 0.6 W average power with 60 fs pulse duration, and a roughly flat spectrum from about 1.8 to 1.96 µm. The passively cooled thulium fiber was core pumped by about 3 W from a continuous wave Er fiber laser at 1566 nm. It is important that the spectral broadening be a coherent process in order for the final output to be a frequency comb. For example, we previously used Raman frequency shifting [9], which provides tunability, but the process was likely incoherent, so the DFG CEO would not have been stable.

The pump and signal seed exit to free space, and the beams are spatiotemporally overlapped on a dichroic mirror. They are focused by a 4 cm focal length lens into the 1 mm thick, anti-reflection coated OP-GaP crystal. The crystal has adjacent grating periods of 52.25, 54, 55.5, 57.25, 60, 62.25, and 64 µm that can be selected by translating the crystal in order to choose the output spectrum. Difference frequency between the pump and signal seed generate midinfrared idler output from about 6 to 11 µm which is collimated with a zinc selenide lens. A 3.4 µm longpass filter passes the midinfrared output, and reflects the input beams, as well as the many parasitic nonlinear combinations generated along with the idler.

Midinfrared spectra and the corresponding average powers are shown in Fig. 2 for different quasi-phase matching periods. The DFG after the longpass filter is about 60 mW. Over 70 mW has been observed with temporarily stronger pumping. The output was verified to be midinfrared by transmission through different materials with known transmission spectra such as plastic, glass, and various midinfrared filters. A 60 mW output with 1.6 W of pumping corresponds to a photon conversion efficiency of about (60 mW / 1.6 W) × (8 µm / 1.55 µm) = 19% without accounting for losses from the lenses and filter. Some previous systems [10] had efficiencies ranging from 1% from GaSe with an erbium-based system without a thulium amplifier for 4 mW output [11], 18% in CdSiP₂ with a Tm:Ho amplifier for 15 mW output [12], and 43% from a 3 mm long periodically-poled lithium niobate crystal at about 3 µm wavelength for 500 mW output [13].

 

Fig. 2 Midinfrared spectra and their average powers for different quasi-phase matching periods of OP-GaP, with 69 mW output at 7.4 µm.

Download Full Size | PDF

3. Pulse timing

The pump and signal seed arms are made of several meters of fiber, making the pulse timing sensitive to temperature. The relative pulse timing is adjusted by a delay stage. Left to drift, the system output power could vary from full to no output. We use a piezoelectric-based 120 µm range translation stage for active feedback control of the delay at about 10 Hz speeds.

The choice of diagnostic for the pulse delay offset is an important parameter for the stability of the lock. We use one of the many coincidentally generated parasitic nonlinear combinations such as frequency doubling and sum frequency mixing. These are separated from the midinfrared output at the long pass filter, and can be isolated with dielectric filters.

The simplest way to diagnose the pulse timing would be to take the photodiode signal of one of the parasites, and directly use this as the error signal for a side-of-fringe lock for a pulse delay with nearly peak midinfrared generation. This method works, but is sensitive to amplitude fluctuations. Compare this to the balanced cross-correlator, where there is a signal that peaks for a positive and one for a negative pulse delay. The difference of the two signals provides a linear region in the middle which serves as a good error signal and is ideally insensitive to intensity. We found two parasites with similar behavior, one being the frequency-doubled signal seed near 1 µm wavelength, and the other being the sum of a signal and two pump photons near 550 nm.

The dependence of these signals with pulse delay is shown in Fig. 3, where the delay slowly oscillates around peak DFG. The top curve is the relative midinfrared output power. The two parasite photodiode signals are in the middle, showing their opposing behaviour. The detector gain is adjusted to roughly match their signal levels. The error signal is the difference of the parasite signals shown by the bottom curve, and can be offset so that the DFG is maximized when the error signal is locked to 0 V. This combination of parasites is not completely intensity independent, but is more robust than using a single detector as the source of the error signal.

 

Fig. 3 Oscilloscope traces while sweeping the pump and signal seed pulse delay. From top to bottom, relative midinfrared output power, photodiode signals of two nonlinear parasites, and the error signal from their difference, having a broad linear region near maximum DFG.

Download Full Size | PDF

A time trace of the stabilized midinfrared output power as measured by a thermal power meter is shown in Fig. 4. The regular peaks coincide with the cycling of the air conditioner. However, even with these peaks, the standard deviation is only 0.4% over the 100 minutes shown in the figure. The system can operate stably for hours.

 

Fig. 4 Temporal stability of the midinfrared output power, with a standard deviation of 0.4% over 100 minutes. The regular variations are from room temperature changes.

Download Full Size | PDF

4. Frequency comb

To show that our midinfrared output is a frequency comb, we use a large-delay interferometer to interfere each pulse with its previous pulse. The interference is monitored by a mercury-cadmium-telluride detector, and the optical path delay is continuously tracked by two-quadrature measurement of a frequency-stabilized HeNe laser. This makes the interferometer a Fourier transform spectrometer, but with an additional 3 m delay in one arm. If the midinfrared beam is a frequency comb, it will have a stable intensity fringe pattern as a function of path delay. If the repetition rate changes, the interferogram will shift in path delay. If the CEO frequency is changed, the phase of the interference fringes will change.

Figure 5 shows a few fringes of the interference burst when the two pulses are overlapped. The different curves are for different frequency shifts applied to the signal seed. Four shift values were used, 164, 186, 207, and 229 MHz. For the first shift, about four interferograms are measured and coherently averaged, resulting in a purple curve in Fig. 5. The frequency shift is then stepped to the next value, and four more interferograms are averaged to make a blue curve. After finishing the 229 MHz shift (red curve), the measurement sequence is repeated in reverse order, back down to 164 MHz, plotted with colours matching the frequency shift. There are two main points to Fig. 5: each colour has its own distinct phase, indicating a different CEO frequency; and two curves of the same colour mostly overlap, indicating that the repetition rate and CEO frequency are stable for at least the few minutes duration of the full measurement.

 

Fig. 5 Intensity interference fringes as each pulse is interfered with its previous pulse. Four different optical frequency shifts are applied to the signal seed pulses, shown by curve colour, stepping up then back down in frequency. The grouped colours and shifted phases show that the midinfrared output is a frequency comb with tunable carrier envelope offset frequency.

Download Full Size | PDF

To be more quantitative, we can calculate the phase of the interferogram as a function of delay (Fourier transform, zero the negative frequencies, then inverse transform back to get the complex interferogram). The result is plotted on the left of Fig. 6, with the phases converted to CEO by taking one of the phase curves to be a baseline, assuming it has the CEO value expected to result from the applied frequency shift (in this case, one of the red curves), and scaling to the repetition rate. A frequency shift applied to the seed signal is expected to become the negative of the CEO frequency, as seen by writing the difference of the pump and seed frequencies: f_DFG = f_pump − f_seed = (n f_rep + f₀) − (m f_rep + f₀ + f_shift) = p f_rep − f_shift, where n, m, and p are integers, f₀ is the CEO frequency of the oscillator, f_rep is the repetition rate of the oscillator, and f_shift is the applied frequency. Following convention, the CEO frequency is taken to be between 0 and f_rep by adding or subtracting multiples of f_rep as needed.

 

Fig. 6 Interferogram phases converted to CEO frequency relative to one interferogram in the temporal and frequency domains. The phase shifts match the frequency shifts applied to the signal seed before difference frequency generation, as marked by the circles.

Download Full Size | PDF

The curves in Fig. 6 all show the expected relative phases corresponding to having a CEO from the applied frequency shift, as marked by the circles. The phase in the spectral domain (after one Fourier transform of the interferogram) is shown on the right side. The phase of the interferogram is noisier as it includes amplitude and spectral fluctuations, while the cleaner spectral phase is like an average over the full interferogram. The spectral phase curves are within about ±1 MHz of the expected CEO, indicating that the midinfrared output is a frequency comb with tunable CEO.

This measurement does not say much about the phase noise of the midinfrared comb as various noise sources, particularly amplitude fluctuations, will couple into the phase. The detector used here was also relatively slow, on the kilohertz level, so any fast modulations would not be detected. However, the minutes long stability of the fringe pattern shows that the average frequency values of the comb are likely quite good, as would be expected given the tight repetition rate stabilization and the direct control of the CEO frequency.

5. Summary

We have demonstrated a stable and broadband 60 mW midinfrared frequency comb in the 6 to 11 µm wavelength region with tunable carrier envelope offset frequency. This is made possible by the recent development of orientation-patterned gallium phosphide crystals, which provide quasi-phase matched difference frequency generation between erbium and thulium fiber lasers. These crystals are remarkably robust, showing no noticeable degradation after months of use, and hours of continuous exposure to over 2 W of tightly focused light.

This midinfrared frequency comb source may be particularly useful for comb applications where the value of the CEO frequency is important. For example, in cavity-enhanced spectroscopy, there will be a particular CEO value that has the best coupling of the cavity and the frequency comb. For a source like a doubly-resonant optical parametric oscillator, adjusting the CEO frequency involves large changes to the output spectrum and intensity [14]. Since it is based on fiber lasers, this system should be scalable well beyond 100 mW, and while this was a laboratory experiment, the system could be engineered to be almost fully fiber-based, making it a compact and stable broadband midinfrared frequency comb for important future applications such as environmental monitoring and chemical microscopy.