1. Introduction

Three-dimensional scanning can be a waste of time, if one wants to image a planar surface or interface only. However, sometimes accumulation of non-relevant data cannot be avoided, either because the position of the target is not known, or the scanning scheme is not adjusted to the surface to be inspected. We present a method for reducing the number of scans needed to cover a planar surface in pulsed terahertz (THz) imaging, thus making this technology more viable for non-destructive material testing.

THz technology deals with the generation and detection of electromagnetic radiation with wavelengths roughly between 3 mm and 30 μm (0.1–10 THz). The science behind it has come a long way from its early beginnings [1–4] to modern day imaging techniques (see, e.g., [5,6]). Applications range from defect detection in polymers and foams [7–11], food inspection [12], biomedical and pharmaceutical applications [13, 14] to the detecion of hazardous [15, 16] or illicit [17, 18] substances. For a comprehensive review, refer to, e.g., [19].

A multitude of detection schemes have been established so far. On the one hand, these are systems using intensity dependent THz detectors such as bolometers [20], pyroelectric detectors [21] or superconducting tunnel junctions [22]. However, they have their drawbacks by not recording the phase of the THz waves and by being either slow, very sensitive to their surroundings or requiring expensive cooling. On the other hand, phase sensitive detection schemes, which are also referred to as pump and probe techiques, have the advantage of providing direct information on the THz refractive index and absorption coefficient, as well as on the three-dimensional (3d) structure of the sample. Well established pump and probe techniques either use photo-conductive antennas (PCAs) [1] made from semiconductor materials [23], or non-linear optical crystals (e.g., LiNbO₃ [24], CdTe [25], ZnTe [26]) through optical rectification and electro-optic sampling (EOS). The full potential of phase sensitive measurement is exploited only in reflection geometry, when the echoes from interfaces at different depths are recorded separately [27]. Hence, this measurement method is inherently sensitive to the depth position of the sample or interface. In contrast, measurement in transmission geometry only gives an integral information of the sample, i.e., for 3d imaging of irregular objects or inhomogeneities, the sample has to be rotated and computed tomography has to be performed [28].

In order to obtain lateral information on the specimen, one can move the focused THz beam relative to the sample by translating the sample with a gantry [13, 29], drag the THz emitter/detector across it [10], or scan the THz beam over a fixed object [30, 31]. Alternatively, single line scanners [32, 33], line scanners with a temporal axis [34] and full field [35–38] imaging systems have been realized. They provide much faster acquistion times than a single pixel imaging system by removing one or even two scanning axes. However, the THz source has to illuminate a line or the whole object at once, and the scan area is restricted mainly by the size of the non-linear optical crystal. Hence, single pixel scanning systems are still the better choice if high signal-to-noise ratios (SNR), dynamic ranges (DR) or large scan regions are needed.

In this work, we have realized a phase sensitive THz scanner in reflection geometry, which allows for scanning of the THz beam over a fixed object. The setup will be explained in detail in section 2. For scanning of the THz beam, a two mirror galvanoscanner in combination with a single scanning lens is used. One of the scanning mirrors is necessarily positioned out of the focal plane of the lens, which thus leads to a distortion of the image of the scanned object, i.e., a planar surface or interface of an object appears to be curved along one direction. The effect of bending of the scanning plane is demonstrated in Fig. 1. The imaged sample consists of six sectors cut from a 2 mm thick aluminum sheet, and a second sheet used as back panel. This pattern, also known as a Böhler star [39], is the 3d equivalent of the Siemens star for determining the resolution of optical imaging systems. A 3d scanner cannot reproduce the correct height of the two planes once the spacing between the sectors becomes smaller than the beam diameter. Hence, the resolution limit can be determined from the diameter of the central plateau and the number of sectors. In Fig. 1(a), a photo of the sample and in Fig. 1(b), the corresponding surface reconstruction obtained from a THz 3d scan are depicted. In one direction (in our case the vertical), the flat surface is not correctly reproduced, but appears curved. In order to correct the image, usually a large number of scans have to be performed, covering a scanning volume that contains the curvature of the surface, followed by data post-processing. This is a severe downside of such scanning systems, as it unnecessarily increases the measurement time.

 

Fig. 1 (a) Photo of a Böhler star, which is used as a resolution target. (b) Reconstruction of the sample surface, obtained from the time delay of the reflected THz pulse. The flat surface appears curved along one axis, which is an artefact due to scanning.

Download Full Size | PDF

In this work, we present an improved measurement routine for our scanning THz system, which takes care of the problem of the deformed scanning plane already during measurement. By aligning the scanning plane with the surface of the sample, the necessary scanning time is decreased by one order of magnitude and therefore, efficient THz imaging of planar surfaces and interfaces becomes possible.

2. Materials & methods

2.1. Experimental setup

In Fig. 2, the schematic measurement setup of the scanning THz system is shown, which allows for phase sensitive measurements in reflection geometry. The source for the femtosecond gating pulses is a Mai Tai laser (Newport Spectra-Physics), tuned to 790 nm and delivering 1.15 W average power with <100 fs short pulses. We use PCAs as THz emitter and detector, which are made from low temperature grown gallium arsenide (LT-GaAs). For the emitter, a strip line antenna is used. The separation of the electrodes is 10 μm, and the applied voltage is modulated at 5 kHz. In order to measure the THz beam reflected by the sample, we use a THz beamsplitter made from a 2 mm thick wafer of high resistive float zone silicon (HRFZ-Si) (Tydex). The part of the emitted THz pulse that is reflected by the beam splitter is guided towards the target, where it is reflected, passes through the beam splitter and is finally recorded by the detector. The part of the emitted THz pulse that directly passes the beam splitter (dashed lines in Fig. 2) is received by a second detector. This second signal can be measured simultaneously with the beam from the sample and provides a reference for the emitter. We later refer to these two parts as the reflected and the transmitted THz beam, respectively. The two detectors include a 20 μm long and 5 μm wide dipole structure [40]. Their signal is amplified with two current amplifiers (DLPCA-200 from Femto) before being fed into a lock-in amplifier (eLockin 203 from Anfatec). Data acquistion is realized by a DAQ card (PCI-6281 from National Instruments) and a personal computer (Dell), which also controls the translation stage (M-403.8PD from Physik Instrumente) and the two axes galvanoscanner (model 6900 from Cambridge Technologies). The THz beam is collimated and focused by hyperhemispherical lenses made from HRZF-Si (attached to the PCAs) and off-axis parabolic mirrors.

 

Fig. 2 Schematic drawing of the measurement setup. BS = non polarizing infrared beam splitter, OAPM = off-axis parabolic mirror, THzBS = THz beam splitter

Download Full Size | PDF

Special attention is given to the aspheric, rotationally symmetric f-θ scanning lens, which is custom made from a 20 mm thick slab of polytetrafluoroethylene (PTFE, Teflon). The front and back surfaces are not symmetric and are optimized for a beam diameter of 50 mm, corresponding to the clear aperture of the galvanoscanner. The focal length of around 145 mm results from geometric requirements (maximum scan angle of ±10° and lens diameter of 101.6 mm or 4”). The shape of the lens is the result of a raytracing simulation of the optic system consisting of the scanning mirror, the lens and the imaging plane. The merit function, which is minimized for optimization, includes 1.) the spot size on a flat focal plane, 2.) the linear dependence of the image height on the deflection angle, and 3.) the incidence angle of the chief ray of the THz beam on the focal plane. The realization of the scanning optics with a single lens not only simplifies the adjustment of the system, but also maintains a high DR by reducing Fresnel losses and the optical path length through absorbing material.

For our THz scanning system, the principal mode of operation is transversal or so-called en-face scanning. There, the two mirrors of the galvanoscanner deflect the THz pulses along the horizontal and the vertical direction, respectively, while the delay time is held fixed. These scans provide information from a distinct scan depth or about the phase of the THz pulse. As a second measurement mode, cross-sectional scanning along one lateral direction and the depth axis can be performed. Our system permits a maximum lateral scanning area of about 100 mm x 100 mm. The scan speed is limited to a few Hertz by the speed of the galvanoscanner, and is typically set to 1 line/s. Depending on the number of lines per image, one en-face scan therefore requires 1–2 minutes. Much faster THz imaging systems and scanners with up to 10 lines/s have been have been demonstrated [41]. With a lighter mirror this scan rate would be possible with our galvanoscanner as well, but at the same time the integration time of the lock-in amplifier needs to be decreased, which leads to a lower dynamic range and noisier images.

As the mirrors of the galvanoscanner act as virtual apertures, one mirror can be positioned in the back focal point of the scanning lens. This has the advantage that one obtains a telecentric system for this direction, i.e., the THz beam always stays parallel to the optical axis. However, the second mirror is necessarily positioned out of the focal plane, and therefore the scanning plane becomes curved along the second scanning axis, since the optical path length of the THz beam is now dependent on the deflection angle. An estimation of the additional measurement time due to the bending of the scanning plane can be made, based on the following rough numbers: the jitter (i.e., the standard deviation of the position of the maximum of the THz pulse), was found to be around 0.01 ps, equivalent to an uncertainty in the depth position of about 3 μm. The curvature of the scanning plane amounts to about 2 mm, as was determined from images of the Böhler star. In order to accurately measure the pulse amplitude, the enface scans are performed every 20 μm, corresponding to a temporal resolution of 133 fs or a Nyquist frequency of 3.75 THz. Therefore, about 100 scans are needed to cover a single surface. Without the bending of the scanning plane only 5 scans would suffice to cover a scan depth of 100 μm, a surely adequate number for the flat surface of this specimen. The bending will therefore increase the imaging time of a single interface by a factor of 20. In section 2.3., we will introduce a scanning scheme based on vertical cross-sectional scans, that compensates these bending effects already during en-face scanning.

2.2. Dynamic range and quality of the reflection arm

As the mechanical stage delays the laser pulse for the emitting PCA, it is possible to record both, the transmitted and the reflected THz beam simultaneously by adjusting the optical path lengths of the gating pulses for the two receiver antennas (see Fig. 2). A mirror with a diameter of 50.8 mm (2”) is used as object for the reflected signal. In Fig. 3(a), the average of 16 time-domain traces for the reflection and the transmission arm, respectively, is shown. Every trace is resampled with equidistant supporting points and a straight line is subtracted by linear regression to eliminate an offset or linear drift of the signal. For calculating the THz spectra, a fast Fourier transform (FFT) is applied to the single traces prior to any averaging.

 

Fig. 3 (a) Time-domain traces for transmitted and reflected THz beam, averaged over 16 measurements. The transmitted signal is scaled by a factor 0.5. (b) Average of 16 specta, obtained by Fourier transform for the transmitted and reflected THz beams and scaled by their respective noise levels.

Download Full Size | PDF

In Fig. 3(b), the spectra of both, the reflected and the transmitted THz signals are shown, normalized by their respective noise levels, which corresponds to the DR of the measurements [42]. It can be seen that the much longer optical pathlength in ambient air, Fresnel losses from the THz beam splitter and lens, as well as the double transmission through the PTFE lens entail a smaller DR and useable bandwidth for the reflected beam.

For characterization of the scanning THz system, the maximum DR is determined along a horizontal and a vertical scan line through the center of the mirror. The DR is calculated following the procedure described in [43]. The noise level is determined from a measurement with blocked THz beam as the standard deviation (root-mean-square, RMS) of the THz trace in time-domain (TD). In frequency-domain (FD), the noise level can be calculated as the RMS value of the magnitude of the amplitude spectrum of the blocked beam, however, both values (TD and FD) of the noise level correspond to each other. The maximum DR is obtained from the ratio $DR=\frac{\text{mean}\hspace{0.17em}\text{peak}\hspace{0.17em}\text{value}}{\text{noise}\hspace{0.17em}\text{level}}$ in FD at the maximum of the spectrum.

The result of these measurements is shown in Fig. 4. The DR stays nearly constant (around 1000 or 60 dB) across the horizontal scan line, up to the edge of the mirror with 2” diameter. The slight sagging in the center can be attributed to higher absorption by the scanning lens, which has its maximum thickness in the center as well. For the vertical scan, the signal drops to around 40% at the edge of the mirror, which indicates that the THz beam does not stay parallel to the optical axis and therefore is not perpendicularly incident on the target mirror. Thus, the reflected THz pulse does not completely pass through the scanning lens anymore, and some signal is lost. This is a consequence of the fact that only one mirror can be at the back focal point of the scanning lens, i.e., fulfills the requirements for a telecentric system.

 

Fig. 4 Maximum dynamic range as a function of vertical and horizontal beam deflection calculated in frequency-domain.

Download Full Size | PDF

2.3. Phase correction

In our THz imaging setup, the horizontal scanning mirror is located at the back focal point of the scanning lens. The out-of-focus position of the vertical scanning mirror generates not only a loss of amplitude, but also a dependence of the optical path length on the vertical deflection. Because of the phase sensitivity of the THz detection method this prevents an easy en-face scan of a planar surface, i.e., if an xy-scan at fixed delay line position is performed, one would cut through different phases of the THz pulse (much like a sagittal cut of an onion).

In order to overcome this limitation, a cross-sectional scan perpendicular to the sample surface is made prior to en-face imaging. Figure 5 shows the THz signal amplitude for (a) vertical and (b) horizontal deflection (abscissa) and position of the delay line (ordinate). The midpoints between the maximum and minimum of the THz pulse lie on a straight line for a horizontal trace (dashed line in Fig. 5(b)). However, a phase shift of the echoes can be observed for the vertical scan. This additional optical path length has a quadratic dependence on the deflection and can be nicely fitted by a second order polynomial (dashed line in Fig. 5(a)).

 

Fig. 5 (a) Vertical and (b) horizontal cross-sectional depth scans of a planar surface. The vertical phase shift can be fitted by a second order polynomial (dashed line).

Download Full Size | PDF

Hence, it is possible to compensate the non-planarity of the isophase surface already during scannning if the following scheme is used:

3. Results

The benefit of this procedure is first demonstrated using the sample (Böhler star) from Fig. 1. From this object 512 en-face scans, each with a resolution of 128x128 pixels, are taken, covering a total scan depth of 8 mm. For each pixel, the position of the absolute maximum of the amplitude is determined. The result is shown in Fig. 6(a), which can be interpreted as the topology of the sample. Figure 6(b) shows the height profile of the two diagonal cutlines in Fig. 6(a). Note that without additional post-processing the planar surfaces are accurately reproduced, and the curvature of the image is neglible.

 

Fig. 6 (a) Position of the absolute maximum of the time signal from the sample in Fig. 1. The dashed circle indicates the limiting gray ring for determining the resolution. (b) Height profile from the solid and dashed cutlines in (a).

Download Full Size | PDF

The diameter of the outer circle of the Böhler star, which amounts to 40 mm, is used to determine the conversion factor between the voltage applied to the two mirrors of the galvanoscanner and the corresponding deflection after the scanning lens by correctly scaling the image (horizontally 5.1 mm/V, vertically 5.4 mm/V). The resolution l of the scanning system is given by $l=\frac{\pi d}{2N}$, where d is the diameter of the inner ’gray ring’ (dashed circle in Fig. 6(a)), and N = 6 is the number of rays of the star. d is determined by calculating the Fourier coefficient of the 6^th harmonic by inscribing circles of variable radius. The limiting radius is taken at 50% of the maximum value and is found to be d = 10.5 mm. Thus, the resolution of the imaging system is roughly l = 2.7 mm. With a beam radius of w₀ = l/2 ≈ 1.5 mm, the Rayleigh length of the THz pulse amounts to about $z_R=\frac{\pi w_0^2}{\lambda }=23.5\text{mm}$ (at 1 THz). Thus, the depth of focus is about b = 2z_R = 47 mm (in air) and much larger than typical scan depths. Therefore, the spot size for a depth scan can be assumed constant for all practical purposes and a dynamic focus is not needed.

3.1. Application

The potential of the scanning THz system in combination with the improved scanning routine is demonstrated for a sample consisting of a metal cylinder attached to a 3 mm thick and 59 mm wide plastic plate by an around 0.6 mm thick adhesive layer (see Fig. 7). This is a component as used in the automobile industry for attaching a plastic part to the metal car body. Since THz waves easily penetrate the plastic, it is possible to examine the adhesive layer by scanning through the opaque plastic plate, which would not be possible with visible light.

 

Fig. 7 Photo of the disassembled composite material sample. The white plastic plate is shown on the left hand side and is usually attached on top of the black ring of glue. The holes drilled into the adhesive layer are indicated by white arrows on the right hand side.

Download Full Size | PDF

For the subsequent measurements, the parameters for phase correction are obtained from an initial cross-sectional scan of the front surface of the plastic plate. In addition, the pulse amplitude originating from the front surface is used to normalize the images along the vertical direction, according to the solid curve displayed in Fig. 4.

In the first 3d measurement, a scan volume including the front surface is chosen, giving an overall impression of the sample. Thus, a total of 512 frames with a scan depth of 10 mm are acquired. The lateral resolution is 128x128 pixels with a scan area of about 65 mm x 80 mm. For the further data analysis, the part of the scanning volume containing the front surface is discarded. In Fig. 8(a), the maximum of the pulse amplitude originating from the intact adhesive layer is shown. The little circle on the right hand side is a nose on the front surface, produced from injection molding of the plastic plate. In Fig. 8(b), a single en-face scan is shown, which is taken at an optical depth corresponding to the plastic/glue interface.

 

Fig. 8 (a) and (c): THz amplitude images of a ring of adhesive layer between a plastic plate and a metal cylinder. Total acquisition time for (a) was about 18 hours, and $2\frac{1}{4}$ hours for (c). (b) and (d): Single en-face scans, each taking about 2 minutes. (a) and (b) are taken with intact glue, (c) and (d) ( Media 1) after two holes were drilled into the glue and the components reattached. All images are amplitude corrected along the vertical direction. An animation through the sequence of enface scans is available online.

Download Full Size | PDF

The well pronounced boundaries of the adhesive layer are interference effects between the THz pulses reflected from the plastic/glue and the plastic/air interfaces. In Fig. 8(a), the high contrast is due to the fact that the plastic/air interface represents a large change of the refractive index compared to the plastic/glue interface, and hence the amplitude of the reflected THz beam is larger. In Fig. 8(b), the two regions are even more pronounced, but here the enhanced contrast can be attributed to the difference between internal and external reflection, i.e., the reflection at a boundary from higher to lower or from lower to higher refractive index, respectively. At the plastic/air interface, the refractive index changes from high to low, and hence the reflected light does not undergo a phase change. On the other hand, at the plastic/adhesive layer the refractive index increases, and therefore the reflected pulse amplitude changes sign. This results in an increased contrast between the two regions in the en-face image. As it is visible from Fig. 8(b), the polynomial coefficients obtained from scanning the front surface cannot completely remove the cutting through different phases of the THz beam, suggesting that using the plastic/air interface at the back for phase compensation would be more favorable.

After this initial scan, the plastic plate is removed and two holes with diameters of 2 mm and 4 mm are drilled into the adhesive (indicated by the arrows in Fig. 7). The two parts are reattached and the measurement is repeated. The lateral resolution is set again to 128x128 pixels with a scan area of about 65 mm x 70 mm, but this time only 64 en-face scans with a total depth of 1 mm containing the adhesive layer are collected. Although the glue is not as strong as it was initially, it is still strong enough to hold the metal cylinder. Figure 8(c) shows the corresponding amplitude image. The two voids from drilling are visible, but also a region of delamination of the adhesive can be observed left to the nose. Figure 8(d) represents a single en-face scan. Again, the contrast is higher due to the phase shift when external reflection occurs, and the edges are enhanced such that the drillings are visible even more than in the amplitude image.

It should be pointed out that for the last measurement, the covered depth is ten times smaller compared to the previous measurement, and the adhesive layer is only tightly contained in the scanning volume. Nevertheless, the whole information is obtained with the same quality, but with a reduced measurement time by a factor of about ten. In addition, it is shown that even single en-face scans can be used for the investigation of planar structures or interfaces. These results clearly underline the effectiveness of our scanning routine for phase correction.

4. Conclusion

We have presented a scanning THz system for measurements in reflection geometry. Our design has the advantage that neither the THz emitter and source, nor the object has to be moved. Instead, an xy-galvanoscanner and a single scanning lens made from PTFE allow performing 3d scans of partially transparent objects in the THz wave range.

For planar structures, we have developed an advanced scanning routine to compensate the bending of the scanning plane. This has the benefit of not only making geometric compensation during post-processing superfluous, but also of significantly reducing the scanning time because non-relevant scanning volume is spared. Therefore, efficient THz imaging becomes feasible. We want to emphasize that this method is not restricted to our system, but could be implemented in other phase sensitive line scanning THz imaging setups.

Furthermore, we have demonstrated that areas of interest can be identified in single en-face scans with our phase compensation enabled. This means that preparation and execution of measurements becomes much more simple and effective. Therefore, as we have shown for a plastic-metal composite sample, this advanced THz scanning system is highly promising for the investigation of layered parts and components, hopefully bringing THz technology a step further on its route to application.