Mid-infrared optical coherence tomography with a stabilized OP-GaP optical parametric oscillator

We demonstrate mid-infrared time-domain optical coherence tomography (OCT) with an orientation-patterned GaP optical parametric oscillator. Instantaneous broadband mid-infrared spectra provide reduced scattering for OCT applications including cultural heritage, quality assurance, and security. B-scan calibrations performed across the wave-length tuning range show depth resolutions of 67 µm at 5.1 µm and 88 µm at 10.5 µm. Volumetric imaging inside a plastic bank card is demonstrated at 5.1 µm, with a 1 Hz A-scan rate that indicates the potential of stable broadband OPO sources to contribute to mid-infrared OCT.


Mid-infrared optical coherence tomography with a stabilized OP-GaP optical parametric oscillator
Imaging at mid-infrared (MIR) wavelengths can provide unique insights and contrast mechanisms because of the low scattering of MIR light and the chemical specificity of MIR absorption.Compared to a common optical coherence tomography (OCT) wavelength λ = 1.3 µm, the scattering loss (∝1/λ 4 ) is orders of magnitude lower at λ = 5 µm (220×) and λ = 10 µm (3500×), assuming the particle is small in comparison to the wavelength [1].Consequently, new light sources in the MIR offer previously unavailable capabilities for depth-resolved imaging [2,3].MIR OCT applications include an examination of the cultural heritage [4] and marine coatings [5], quality assurance in ceramics additive manufacturing [6], and possible use in identification document validation for security, previously applied in the near IR [7].
Light sources exploited in preceding MIR OCT works have included quantum cascade lasers (QCLs), supercontinuum (SC) sources, and spontaneous parametric downconversion (SPDC) for analysis of a range of samples including biological samples, ceramic plates, paint samples, and bank cards.Colley et al. [8] first demonstrated MIR OCT in 2007 with a time-domain (TD-) OCT device that used a QCL to cover 6-8 µm; however, the structured emission spectrum significantly reduced the axial resolution.Since this seminal work, MIR OCT demonstrations have focused on sources from SC [9,10] or SPDC [11,12] processes with spectral ranges up to 5 µm, immediately beyond which water absorption (5-7.5 µm) restricts imaging in most working environments.
Complementing the above-mentioned sources, ultrafast optical parametric oscillators (OPOs) can provide broadband idler light tunable from 5 to 12 µm at tens-of-mW average powers [13,14].The quasi-phasematched nonlinear crystal of choice for MIR OPOs above 5 µm is orientation-patterned GaP (OP-GaP), notable for its compatibility with mature 1.0 and 1.5 µm pump lasers, as well as its wide 0.55-12 µm transparency range and its high 70.6 pm/V nonlinear-optical coefficient [15].Here, we demonstrate the MIR time-domain (TD-) OCT with an OP-GaP OPO tunable from 5 to 11 µm providing instantaneously broad bandwidths, and as an example we present volumetric imaging of a bank card at 5.1 µm.
The synchronous OPO was a ring cavity built around an OP-GaP crystal to be singly resonant for the signal and to output the idler beam.It was pumped by 120-fs pulses from a 99.86-MHz 2.8-W Yb-doped fiber laser (Chromacity 1040) with a central wavelength of 1036 nm.The beam propagated through a half-wave plate for polarization tuning before an N-BK7 lens (L, f = 50 mm) focused the beam through a ZnSe curved mirror (CM1, ROC = 100 mm) to form a 15-µm beam waist in the center of the OP-GaP crystal, calculated by using a Gaussian beam propagation model.The OPO crystal was an OP-GaP on OP-GaAs crystal (BAE Systems) with dimensions of (1.9 × 12.8 × 1.1) mm 3 .The crystal was grown to have 13 distinct quasi-phasematching gratings along its 12.8 mm length from Λ = 21-34 µm in discrete steps of 1 µm.The operational region of the OP-GaP was known from previous works to be approximately within 150 µm of its border with OP-GaAs [16].
As illustrated in Fig. 1(a), a plane intracavity silver mirror (M2) was positioned on a micrometer-driven linear stage (LS1) to allow precise matching of the OPO and pump laser cavity lengths needed to achieve oscillation.A concave intracavity silver mirror (CM2, ROC = 100 mm) collimated the signal and idler beams toward a ZnSe plane mirror (M1).This and the other ZnSe cavity optics had custom coatings (TwinStar) which reflected the signal wavelength and transmitted the pump and idler wavelengths.The ZnSe plane mirror served as the output coupler for the idler.The idler beam was isolated by a Ge window (W) that transmitted the idler and reflected the accompanying visible and pump light into a beam dump.A total of 12 of the 13 distinct grating periods were operatable, with Λ = 34 µm inaccessible due to obscuration by the crystal mount.Idler spectra centered from 5.1 to 10.7 µm were measured, with average powers ranging from 85 mW at 5.1 µm to 11 mW at 10.7 µm.The decrease in average power at longer wavelengths can be attributed to the decrease in idler photon energies and the lower nonlinear gain experienced further from degeneracy [17].A further explanation is the poorer spatial mode overlap of the pump and idler beams within the OP-GaP at longer idler wavelengths because of a greater mismatch in their respective confocal parameters [16].
Active stabilization of the cavity length was required to enable measurements over many hours, as environmentally induced changes to the cavity length on the order of tens of nm led to power fluctuations and spectral changes in the idler output, and eventually to loss of oscillation.Visible light was emitted from the OP-GaP crystal due to the nonlinear mixing between the idler, signal, and pump photons, with an intense red beam observed from the second-harmonic of the signal.Any change in the center wavelength of the signal and idler pulses was evident in the spectrum of the second-harmonic light, and this was leveraged to achieve long-term stabilization of the idler spectrum.The second-harmonic beam was dispersed across two photodiodes (PDs) using a reflective blazed diffraction grating (DG).The photodiodes were mounted side-by-side to act as a basic position-sensitive detector, and their signals sent respectively to the inverting and non-inverting inputs of a high-speed servo controller (New Focus LB1005) to generate an error signal.The modification of the distribution of spectral intensities in the second-harmonic of the signal spectrum, from fractional changes to the cavity length, manifested as a non-zero error signal from the servo controller.The output signal from the servo controller was sent, via a high voltage 20× amplifier (Falco Systems WMA-005), to a piezoelectric transducer (PZT, Thorlabs PK25LA2P2) that was placed between the micrometer and LS1 that translated M2.  powers and idler spectrum remained the same throughout 2 h measurements of the 5.1 µm idler beam.This was in comparison to the free running case, in which a stable output was observed for ∼20 min and oscillation was lost after 1 h.
A schematic of the TD-OCT device is shown in Fig.The 780.2 nm reference wavelength was provided by a narrowline Rb-stabilized cw laser beam (Vescent Photonics D2), and the detector was an amplified Si detector (NIR PD, Thorlabs PDA36A-EC).Data from both interferometer detectors were acquired on an oscilloscope (TiePie Handyscope HS5) with an arbitrary waveform generator (WFG) that was used to displace the linear stage ±1.5 mm with a 2 Hz sine waveform.The sample was laterally imaged by displacing the sample stage in a raster scan pattern.Sample stage positions were probed to generate the two P R × P C × P A interferometric data cubes from the OPO interferometer and cw interferometer, where P R is the number of lateral row pixels, P C is the number of lateral column pixels, and P A is the number of sampled points per scan.The acquisition times were large for volumetric OCT scans, as each A-scan took 1 s to acquire.The reconstruction to obtain an OCT image was performed independently for each lateral pixel  Estimations of the axial and lateral resolutions were performed across all available grating periods of the OP-GaP crystal by an analysis of B-scans from a 1-mm-thick uncoated CaF 2 window, chosen for its broadband IR transparency.B-scans at each idler wavelength were acquired by recording an interferogram every 40 µm through a range of 4 mm across the uncoated window surface.For each idler wavelength, the optical spectrum and a B-scan were reconstructed from the interferometric data cube, from which the axial resolution was estimated.For the spectral reconstruction, the interferometric data cube was cropped such that only a single reflection from the uncoated upper window surface was present in the interferogram trace.Spectra from ten neighboring interferograms were averaged to estimate the OP-GaP OPO idler spectrum.Axial resolution estimates in air for each idler spectrum were obtained by measurement of the fullwidth-at-half-maximum (FWHM) of the A-scan feature from the reflection of the uncoated upper window surface.FWHMs from ten neighboring A-scans were averaged to give the axial resolution estimate.The diffraction-limited lateral resolution was estimated from the FWHM of the lateral point spread function (PSF), taken as the squared modulus of the normalized field at the focal plane [18].The lateral resolution calculations were based on experimental values for the idler central wavelength and beam radius at the OAP along with the OAP focal length and diameter.Spectral reconstruction and lateral and axial resolution estimates were obtained for each idler wavelength and shown in Fig. 3.The axial resolution estimates ranged between 60 and 134 µm in air, with lower axial resolution around ∼6.2 µm due to the water absorption limiting the bandwidth.The axial resolutions available may not be suitable for some applications and are lower than those of SC-based systems demonstrated at up to 5 µm utilizing InF 3 [9] or ZBLAN [10] sources that provided ∼5-10 µm axial resolution, or the SPDC-based demonstration by Vanselow et al. [12] that achieved 10 µm axial resolution with a 3.3-4.3µm source generated from 660-nm pumping of a PPKTP crystal.The lateral resolution estimates derived from experimental-based calculations ranged from 47 to 70 µm.
A bank card was imaged as a proof-of-principle test of the MIR TD-OCT system.The polyvinyl chloride (PVC) bank card, with dimensions (85.60 × 53.98 × 0.76) mm 3 , was chosen as PVC absorbs in the visible and NIR but transmits in parts of the MIR spectral range, as first shown in work by Israelsen et al. [19].The bank card (Lloyds bank) was clamped to the sample stage such that the upper surface displayed text, the magnetic strip, and signature strip and the lower surface contained the embedded Europay, Mastercard, and Visa (EMV) chip and bank logo.A volumetric scan of the bank card was carried out at 5.1 µm.Volumetric scans at idler wavelengths greater than 5.6 µm were prohibited by strong absorption of PVC at these wavelengths [20].The experimental axial resolution at 5.1 µm was 67 µm in comparison to the diffraction-limited calculated value of 64 µm.The signal-to-noise ratio was ∼10 4 , estimated using measured interferogram intensity as I 2 max /σ 2 [21].The raster scan acquisition created a 100 µm-pitch grid for the lateral positions, (Fig. 4) where the total x (y) displacement was 55.1 mm (64.9 mm).The total number of A-scans acquired was 650 × 552 = 358800, which took 100 h at a 1 Hz A-scan rate.OPO stability prohibited a single continuous measurement although with stabilization single measurements could readily last over 10 h.The data were recorded in 12 separate measurements in a two-week period, where it was possible to spatially co-register the separate volumes to create the composite volumetric image.Co-registration was achieved by fitting lines to image features and cross-correlating the fits at neighboring volume-volume boundaries.
The results are displayed in Fig. 4, where the internal structure of the card can be observed that includes a near-field communication (NFC) antenna and the EMV chip.Figures 4(a In summary, a MIR TD OCT has been demonstrated using a 100 MHz OP-GaP OPO with a 5-11 µm tunable bandwidth.To our knowledge, this is the first demonstration of a high repetition-rate broadband MIR OPO used for OCT [22].Cavity stabilization has proven effective for cavity length control over tens of hours.Axial resolution estimates from instantaneous bandwidths were in the range 60-134 µm in air from idler spectra centered from 5.1 to 10.7 µm.This spectral range is within the fingerprint region and as such could be leveraged for molecular identification by the addition of a hyperspectral modality, perhaps based on dual-comb spectroscopy [23].Reduced scattering was demonstrated by volumetric imaging of a bank card at 5.1 µm.Further automation of the OPO could facilitate rapid tuning to sweep through idler wavelengths to utilize the full tunable bandwidth to provide ∼5 µm axial resolution, employing a technique such as a conditional generative adversarial network [24].An SD-OCT embodiment using a MIR linear array could increase the acquisition rate by a factor of 10-100.Exploitation of NIR spectrometer refresh rates would increase the acquisition rate to kHz levels, which could be made possible by an upconversion technique such as that demonstrated by Israelsen et al. [10].Further studies of the MIR OCT with samples that are transparent in the full bandwidth, including plastics and ceramic plates, are a clear next step.
Figure 1(b) exemplifies typical stabilization performance compared to the free running case.The average
2. The idler output beam from the OP-GaP OPO entered a Michelson interferometer where it was incident on a ZnSe 50:50 beam splitter (BS1, Thorlabs BSW710).The reference arm consisted of a gold retroreflector (RR1) fixed to a linear stage (LS2, Thorlabs VCFL35) and a flat gold mirror (M3).The sample arm consisted of a 25.4 mm diameter gold off-axis parabolic mirror (OAP) with f = 76.2 mm (Edmund Optics 35-529), and the sample was placed on a two-axis motorized translation stage (2AS, ASI Imaging MS-2000).The MIR detector was a HgCdTe thermoelectrically cooled photodiode (MIR PD, VIGO PC-4TE-10.6).A secondary Michelson interferometer used a cw source to calibrate the linear stage displacement.The cw interferometer consisted of a UV fused silica plate 50:50 beam splitter (BS2, Thorlabs BSW26), a reference arm with a gold retroreflector (RR2) fixed to the opposite side of LS2, and a silver flat mirror (M5), and a sample arm with a silver flat mirror (M4).

Fig. 3 .
Fig. 3. (a) Instantaneous spectra (shaded, normalized intensity) and corresponding lateral (black, left axis) and axial (red, right axis) resolutions from a range of OP-GaP grating periods.(b) Example of PSFs for resolution characterization at 5.1 µm.
) and 4(b) show B-scans at different slices of the volumetric image.The upper and lower surfaces of the bank card with their matching curvature, due to clamping, are visible.Below the lower surface of the bank card, the flat steel base, to which the card is clamped, is observable.This highlights that penetration depth was not limited to the bank card thickness.Horizontal-line artifacts present in the B-scans result from an etalon caused by the output coupler of the OPO.En face views, shown in Figs.4(d)-4(f), were generated by maximum intensity projections over three axial ranges that related to the upper surface, the internal structure, and the lower surface of the bank card.

Fig. 4 .
Fig. 4. Volumetric imaging of a bank card at 5.1 µm.B-scans (a) and (b) taken from lines as illustrated in the top-down image (c).En face views (d)-(f), with color bar representing a logarithmic signal scale.