1. Introduction

Pulsed, wavelength tunable sources are desirable for numerous applications in areas such as communications [1–3], optical coherence tomography [4] and analogue-to-digital conversion [5]. The soliton self-frequency shift [6] (SSFS) has been exploited in the realization of such sources. It occurs as a result of the red-shift induced by intra-pulse Raman scattering. A wavelength tunable source employing the SSFS is attractive because (i) the red-shift can be continuously tuned by varying the power of the input pulse, (ii) it can provide a very large tuning range, and (iii) sources with short output pulses (pulse widths ∼ 10’s femtoseconds) can be obtained.

In standard fibre-based supercontinuum (SC) generation the SSFS is responsible for extending the long wavelength edge of the spectrum [7]. In that context it is desirable to generate a large number of solitons which overlap spectrally, thus leading to a continuum. This may be achieved through the fission of a pump pulse injected at a wavelength where the fiber exhibits low anomalous dispersion. In contrast, in the realization of a tunable source based upon the SSFS, the relevant output is the single soliton which undergoes the largest red-shift. Thus, in order to maximize the efficiency of the source, it is required that as few solitons as possible are generated, for which high dispersion is a favourable pumping condition.

There have been several demonstrations of the SSFS in different types of silica fiber [8–14]. The maximum shift achievable in these experiments is limited on the long wavelength side by the intrinsic absorption of silica at wavelengths above λ ≈ 2 μm [9, 14]. Since the SSFS requires anomalous dispersion, the separation of the first and second zero-dispersion wavelengths (ZDWs), λ _ZD1 and λ _ZD2, also limits the achievable red-shift. The λ _ZD1 can be lowered by using photonic crystal fibers (PCFs) with small cores, but this dramatically increases the OH loss peak around a wavelength λ ≈ 1380 nm [15]. Attenuation due to OH loss has been observed to limit the long wavelength edge of supercontinua generated in PCFs [16–20]. Although the fiber and pulse parameters employed here are somewhat different to those optimized for SC generation, we find that OH absorption is likewise a limiting factor in the SSFS when pumping below 1380 nm in a small core (< 2 μm diameter) silica PCF. It must therefore be mitigated in order to realize a large shift beyond this wavelength.

Demonstrations of the SSFS over an octave are implied in reports of some very broad supercontinua [21]. However, the conversion of energy into the most red-shifted soliton is necessarily very inefficient. Indeed, there is a distinction in the literature between an SSFS experiment, wherein a single, spectrally resolved soliton is considered, and the generation of a supercontinuum consisting of many overlapping spectral features [8–14]. Here we present the first experimental demonstration of efficient octave spanning SSFS [22], in which light emitted by a mode-locked Ti:Sapphire laser at a wavelength λ = 801 nm is shifted to λ = 1708 nm in a PCF. We measure the photon conversion efficiency from the pump to the output soliton to be 22%, which corresponds to an average power of 9.5 mW at output. For a fundamental soliton with the observed spectral width, the inferred peak power is 1.1 kW. This remarkably large and efficient frequency shift is made possible by two key characteristics of our PCF, which was selected especially for the purpose of this experiment. First, it limits the OH loss to a maximum of 0.06 dB/m at 1380 nm. Second, the zero-dispersion wavelengths, calculated to lie at λ _ZD1 ≈ 700 nm and λ _ZD2 ≈ 1870 nm, are (i) widely spaced, while maintaining anomalous dispersion (β ₂ = −24 ps²km⁻¹) at the pump wavelength of 801 nm, and (ii) are at sufficiently short wavelengths so as not to be affected by the intrinsic infrared absorption over the length scales considered. Furthermore, the presence of λ _ZD2 implies a reduction in the magnitude of the GVD beyond the OH loss region. This leads to pulse compression at longer wavelengths which helps compensate for a decreasing nonlinear coefficient, as well as energy loss due to both the water peak and the inelastic scattering from which the red-shift originates.

2. Theoretical background

The concept of the SSFS spanning an octave is shown in Fig. 1: the widely spaced ZDWs ensure a large wavelength interval with anomalous dispersion, which, in principle, is available for the SSFS. The presence of λ _ZD2 also constrains the magnitude of the GVD within this interval. Beyond a wavelength of λ ≈ 1500 nm the absolute GVD falls with increasing wavelength and the SSFS is enhanced by the increase in the soliton peak power following from the associated adiabatic pulse compression. Additional red-shift beyond the anomalous dispersion region is limited by the spectral recoil associated with the resonant emission by the soliton of dispersive waves (DWs), also termed Cerenkov emission [23], in the normal dispersion region when it impinges upon λ _ZD2 [24].

 

Fig. 1 Schematic of the soliton self-frequency shift showing an input soliton shifting from near λ _ZD1 to λ _ZD2 while crossing the region of OH absorption. The generation by the soliton of DWs above 2 μm occurs when it reaches λ _ZD2 and the SSFS is halted due to spectral recoil. Inset shows an SEM of the fiber core region.

Download Full Size | PDF

For a pulse with peak power P ₀ and temporal width T ₀, the characteristic dispersion and nonlinear lengths are defined as $L_{\text{D}}\hspace{0.17em}=\hspace{0.17em}{T}_0^2/\left|{\beta }_2\right|$ and L _NL = (P ₀ γ)⁻¹, respectively. Pumping a fiber such that L _D ≥ L _NL at a wavelength where the fiber has anomalous dispersion, leads to the generation of one or more fundamental solitons. When multiple solitons are generated, one such soliton has a higher energy and peak power than the others, and it is with this soliton that we are concerned here. Increasing the peak power of the initial pulse increases that of the largest emergent soliton, upon which the SSFS rate is sensitively dependent [25, 26]. It is this property which facilitates the tuning of the final wavelength of the soliton by varying the input pulse power. In order to optimize the efficiency of our SSFS source it is desirable to have a low soliton number $N\hspace{0.17em}=\hspace{0.17em}\sqrt{L_{\text{D}}/L_{\text{NL}}}$ associated with the input pulse. For a given pump source, this may be achieved by increasing the level of anomalous dispersion, and thus reducing L _D, at the pump wavelength through reduction of the PCF core dimensions. As a consequence, the nonlinear coefficient of the fiber is also increased due to the reduced mode area, which enhances the SSFS.

3. Experiment

The experimental setup is shown schematically in Fig. 2. Pulses from a mode-locked Ti:Sapphire laser in the short pulse regime with a repetition rate of 83 MHz were launched into the PCF using a 40× microscope objective. An optical isolator was used at the output of the laser to avoid disruptive feedback. The pulse width was measured using an autocorrelator with the assumption of a Gaussian pulse shape. Simultaneous measurements of the pulse spectrum using an Ocean Optics HR2000 optical spectrometer showed that the pulses were close to transform limited. Due to the birefringence of our PCF, a half-wave plate was used to optimize the input polarization such that the red-shift was maximized at the highest input power. Subsequently, the input power was tuned using a variable attenuator and the output spectrum recorded using two optical spectrum analyzers (OSAs), an Ando AQ6317B covering the range from 600 nm to 1750 nm and a Yokogawa AQ6375 covering the range from 1400 nm to 2200 nm. A removable free-space, low-pass filter (transmitting >1500 nm) was placed between the output of the sample and the long wavelength OSA when measuring at wavelengths above 1750 nm to avoid spurious spectral features arising due to higher order diffraction effects.

 

Fig. 2 Schematic of the experimental set-up: pulses from a mode-locked Ti:Sappire laser are coupled into the PCF, via an attenuator and a polarization controller. The input pulse is characterized using an autocorrelator and an optical spectrometer while the output from the PCF was measured on an optical spectrum analyzer (OSA).

Download Full Size | PDF

The fiber employed here has a core diameter of 1.5 μm; an SEM of the cross section of the fiber core area is shown in the inset of Fig. 1. The OH-associated attenuation for the various lengths of fiber employed in the experiment was determined using transmission measurements with a commercially available Fianium supercontinuum source. Using the SEM of the fiber cross-section and a commercially available finite-element software package, the anomalous dispersion region was calculated to range from λ _ZD1 = 700 nm to λ _ZD2 = 1870 nm, as shown in Fig. 1. This window, wherein the nonlinear coefficient is calculated to vary monotonically from γ = 0.11 (Wm)⁻¹ at λ _ZD1 to γ = 0.03 (Wm)⁻¹ at λ _ZD2, permits an SSFS over an octave for an input wavelength of 801 nm.

The fiber was fabricated using the standard stack-and-draw process but with some additional steps to reduce the spectral attenuation. Previously published data on attenuation in such small-core PCF’s shows a dramatic increase in the spectral attenuation for core diameters below about 2 μm. This effect is due to a combination of extrinsic OH contamination during stacking, and structural damage done to the silica matrix during the fiber-drawing process. These together cause strongly increased attenuation both at the OH overtones and also at other wavelengths within the transparency window of silica. Previous efforts to reduce these effects using halogen-based dehydration have been only partially successful [27]. In our fibers we have greatly reduced these effects by annealing the preform in a dry environment immediately prior to fiber drawing. This enables us to fabricate small-core fibers with low attenuations reproducibly [15].

4. Results and discussion

Fibers corresponding to the structure shown in the inset in Fig. 1 with lengths 21 m and 40 m were used in the experiment. The role of OH absorption was established by exposing the 21 m length of the PCF to a humid environment for 3 months, leading to a peak OH-associated attenuation of 2.41 dB/m. This exposed fiber was then pumped with pulses having a wavelength of 785.5 nm and a temporal FWHM of 100 fs. A series of output spectra are shown in Fig. 3. Once the soliton reaches the OH absorption peak it is observed to undergo significant attenuation. As the input power is increased the largest soliton manifests with sufficient energy to traverse the OH loss region. However, further increasing the pump peak power from 13.3 kW to 15.1 kW does not result in any appreciable advancement of the final wavelength of 1464 nm. In conducting the same experiment in a 21 m length of unexposed fiber with a lower peak OH loss of 0.08 dB/m under similar pump conditions, we were able to achieve a final wavelength of 1614 nm [22]. The fiber ends were sealed between measurements using a fiber fusion splicer in order to prevent the diffusion of air into the dry PCF.

 

Fig. 3 Experimentally measured output spectra showing the SSFS for various pump powers in a 21 m length of PCF exposed to a humid environment with a peak OH absorption of 2.41 dB/m. Vertical dashed lines indicate the approximate extent of the OH absorption peak. The shifted signal is clearly attenuated by the high OH absorption.

Download Full Size | PDF

Clearly, the presence of OH loss can limit the achievable red-shift and, once the available power is fully exploited for a given fiber length, the water peak must be reduced in order to increase the final wavelength of the soliton. We also note that the total loss experienced by the soliton on account of OH absorption is not simply a product of the attenuation per unit distance and the fiber length. The soliton undergoes attenuation only for that portion of the fiber length wherein its spectrum overlaps appreciably with the water peak. Using basic theory [25] we estimate this length to be ∼ 2 m for the largest red-shifts reported below where the attenuation was low, but this is expected to be much larger for fibers with high attenuation, since the subsequent reduction in the SSFS rate leads to the soliton spending more time in the OH loss region.

In order to achieve a larger red-shift, a 40 m fiber with a peak OH associated attenuation of 0.06 dB/m was employed. Fig. 4 shows the output spectra for different input powers. The wavelength of the strongest soliton increases continuously as the pump peak power is varied from 0.8 kW to 12.5 kW. At a power of 9.8 kW, where the pump pulse with a temporal FWHM of 105 fs has an associated soliton number of N ≈ 8, the dominant soliton in the output spectrum has a wavelength of λ = 1708 nm, implying a red-shift of more than an octave relative to the pump wavelength of 801 nm. To the best of our knowledge, this is the largest frequency shift obtained in an SSFS experiment, and the first time an octave shift has been achieved. Although we do not present them here, we note that spectral measurements show the energy in the wavelength region below the pump to be negligible. Our experiments thus combine a very large SSFS with a high efficiency. As mentioned, while the most red-shifted soliton contains 22% of the photons incident on the fibre, it represents about 10% and 52% of the incident and output powers, respectively.

 

Fig. 4 Experimentally measured output spectra for a range of pump peak powers in a 40 m length of dry PCF with a peak OH absorption of 0.06 dB/m. Vertical dashed lines indicate the approximate extent of the OH absorption peak. The inset figure shows the dispersive-wave feature enclosed by the solid box in the upper-most spectrum.

Download Full Size | PDF

For pump peak powers of 12.0 kW and above the dominant soliton is disrupted by the proximity of λ _ZD2, and at 12.5 kW and above the red-shift saturates at 1883 nm and a spectral feature arises in the normal dispersion region at 2040 nm, corresponding to DWs shed by the soliton. Thus, a larger SSFS cannot be achieved by increasing the fiber length. Our results are summarized in Fig. 5, which shows the wavelength of the most red-shifted soliton versus pump peak power, as well as the location the DW feature above 2 μm when present.

 

Fig. 5 The final wavelength of the most red-shifted soliton versus input power (filled circles) as well as the wavelength of the dispersive-wave feature (hollow circles) for the 40 m fiber.

Download Full Size | PDF

5. Conclusion

We have demonstrated a record SSFS of a spectrally isolated soliton over more than an octave, from 801 nm to 1708 nm using a specially selected PCF with reduced OH loss. For this shift, the fraction of the incident photons in the most red-shifted soliton is 22%, which represents a high efficiency, pulsed source with wavelength tunable over more than an octave. For higher input powers the soliton attains a wavelength of λ = 1883 nm but is disrupted due to the presence of λ _ZD2, which leads to the emission of DWs at λ = 2040 nm. This demonstrates the generation of radiation with a wavelength λ > 2 μm directly from a Ti:Sapphire laser.