1. Introduction

THz technology has attracted a great deal of interest in recent years in applications ranging from security screening to medical imaging [1–8]. This is the consequence of the unique properties of THz waves, namely (i) their ability to penetrate different types of materials, from biological tissue, to plastic and cloth; (ii) their non-invasive nature; and (iii) their relatively short wavelength, allowing for a spatial resolution in the 100 μm range [3,4].

Pulsed photoconductive THz generation/detection is a powerful technique which is now currently employed in commercial imaging systems used for research and to carry out online quality control [4–8]. Applications such as security screening are instead more demanding in terms of system sensitivity [1]. Indeed, compared to current values, the latter will need to be enhanced by several orders of magnitude to allow for the penetration of thick layers of material, an essential condition to successfully carry out passenger and luggage inspection. The exploitation of pulsed photoconductive generation will also prove challenging in applications requiring the ability to perform remote sensing on moving targets. In fact, in this case, the distance between the target and the detector must be tracked and measured with very high accuracy in real time, to allow for the time-resolved detection of the reflected THz pulses[1]. In this context the use of sources emitting in continuous wave (CW) is certainly advantageous.

A compact CW imaging system for inspection applications has been recently demonstrated, using a 0.2 THz Gunn oscillator and a Schottky diode detector [9]. At such low frequencies the penetration depth through most materials is large, for instance allowing to image objects inside a luggage [10]. In principle THz radiation can also be used to identify concealed substances that do not have a distinctive geometrical shape, by exploiting their spectroscopic properties For example many common plastic explosives have been shown to possess distinctive THz absorption features in the region between 0.5 and 3 THz [1,2]. The availability of sources emitting in the lower end of this spectral window, i.e. in the range 0.5 – 2 THz, would be ideal for hidden explosives identification, since there are a number of spectral features, yet penetration through barriers is higher than at the upper end of the frequency range [10].

Millimeter-wave Gunn oscillators are commercially available with an output power of ~ 10 mW at ~ 100 GHz. To achieve operation at higher frequencies these devices are combined with multiplier chains, yielding power levels of a few hundreds of μW up to ~ 0.5 THz. As the emission frequency goes deeper into the submillimeter/THz range the power drops rapidly and prototype devices have been demonstrated emitting a few μW at ~ 2 THz [11]. This is in spite of the fact that power levels of the order of the mW are needed to pump room temperature heterodyne mixers in the 0.3 –2 THz band.

The advent of THz Quantum Cascade Lasers (QCLs) has opened new interesting perspectives for THz technology [12]. In these semiconductor laser sources the advantage of compactness is merged with the ability to generate CW powers of several tens of mW [13,14]. So far QCLs have been demonstrated with emission frequencies between 2.0 and 4.4 THz and maximum CW operating temperatures of 117K [15–16]. Extending their operation at least down to 1.5 THz appears to be a realistic objective [17].

In this work we demonstrate raster scan imaging at 2.9 THz using a QCL and a point contact Schottky diode mixer (SDM) operating at room temperature. So far imaging with THz QCLs has been performed using liquid-helium cooled bolometers or pyroelectric detectors [18,19]. Both detection techniques are inherently slow, with response times in the hundreds of ms range, rendering them impractical for many applications. Here we demonstrate a system with a 60 dB signal-to-noise ratio and a potential response time of 10 μs.

2. Experimental technique

Heterodyne mixing by pumping a GaAs membrane-diode with two far-infrared gas lasers and a THz QCL emitting at 4.7 THz was demonstrated by Barkan et al [20]. Subsequently we have reported heterodyne mixing using two independent QCLs emitting at 3.3 THz and a GaAs-based point contact SDM [21]. In that case, owing to thermal instabilities in the devices, we found that the Intermediate Frequency (IF) signal experienced a fluctuation of the order of several tens of MHz. In addition, the IF amplitude and frequency stability suffered from unwanted optical feedback on the QCLs. These facts hinder the achievement of ultimate levels of sensitivity, by essentially limiting the possibility of IF filtering in a narrow band. Frequency stabilisation of THz QCLs by phase locking to a reference oscillator is certainly the most effective solution to this problem [22]. In this work we adopted a simpler approach, based on the detection of the beating between adjacent Fabry-Perot modes from the same QCL, leading to an IF signal oscillating at the cavity round-trip frequency c/2NL, where c is the speed of light, N the group effective refractive index, and L is the cavity length. In comparison, the use of longitudinal modes from two independent QCLs, emitting at ν₁ = nc/2N₁L₁ and ν₂ = mc/2N₂L₂, considerably deteriorates the stability of the downconverted signal since (i) effective indices N₁ and N₂ are modified by independent thermal fluctuations, and (ii) integers n an m amplify the effect of changes of N₁ and N₂ on ν₁, ν₂ (n ~ 200, for a 3 mm long cavity at 2.9 THz).

 

Fig. 1. Layout of the experimental apparatus. Laser emission and IF signal spectra are also represented schematically. The separation of ~13 GHz between neighbouring longitudinal Fabry-Perot modes corresponds to a QCL cavity length of 3mm.

Download Full Size | PDF

The experimental set up is shown in Fig. 1. The sample is mounted on motorised translation stages. Radiation from the QCL is collected, collimated, focused, and re-collected, after reflection on the sample surface, with the help of three 3inch-diameter, f/0.9 parabolic mirrors. The beam is then focused on a point contact GaAs-SDM (Farran CM(X)-5) biased with a regulated constant-voltage source. The device is mounted inside a 90° corner-cube reflector, with an extended “whisker” used to both point-contact the diode and act as a travelling-wave antenna. This leads to a main lobe of approximately 24° width (at the 10dB points), at an angle of 28° to the vertical. The IF signal is extracted with a bias-tee, fed into a two-stage broad-band preamplifier with a gain of 50 dB (MITEQ, JSD3 and JSD2 series), and recorded with a spectrum analyser (Agilent E4407B). As radiation source we used a back-facet gold-coated, 3mm-long THz QCL emitting at 2.9 THz (for details on laser performance see Ref. [13]). This was mounted on the cold head of a continuous flow cryostat, and driven in CW at 10K with a dc power supply.

3. Results and discussion

In Fig. 2 we display the IF power spectrum obtained by placing a gold-coated mirror on the sample holder, and by driving the QCL at maximum power, i.e., approximately 20 mW. With resolution bandwidths of 1 kHz and 5 MHz, signal-to-noise ratios are of 80 and 50 dBm respectively. The line is centered at ~ 13 GHz, corresponding to the laser round-trip frequency. As expected this value is in agreement with the longitudinal mode separation obtained from the laser emission spectrum displayed in the inset (here the linewidth is limited by the maximum resolution, of 0.25 cm^-1, i.e., 7.5 GHz, of the Fourier transform infrared spectrometer). Compared to the use of two independent QCLs we found a much improved line stability, with a maximum drift of the order of 1 MHz over a time scale of several hours. Here we note how this fact, together with the extremely narrow instantaneous linewidth of 10 kHz, obtained from the 1 kHz resolution bandwidth spectrum, are compatible with stable self-mode locked oscillations of the laser [23]. Additional support to this hypothesis was found by analysing the voltage across the QCL with the help of a high frequency bias-tee. In this case (not shown) we measured an ac modulation superimposed to the constant dc bias, with a power spectrum essentially identical to the one reported in Fig. 2 (i.e., same center frequency and linewidth). Further investigations to clarify this issue are under way.

 

Fig 2. IF power spectrum detected at the output of the SDM with two different resolution bandwidths of the spectrum analyser. The QCL was driven in CW with a current of 1.4 A at T = 10K. The IF signal was amplified with a two-stage, 50 dB gain, 12–14 GHz low noise preamplifier (MITEQ, JSD3 and JSD2 series). Inset: QCL emission spectrum recorded with a Fourier transform infrared spectrometer (0.25 cm^-1 resolution).

Download Full Size | PDF

In Fig. 3 we report some examples of images obtained with the present set-up. The scanner step was set to 100 and 200 μm for the left and right panel respectively. Images were acquired by integrating, pixel by pixel, the IF spectrum in a 5 MHz window. This was accomplished by continuously sweeping the internal oscillator of the spectrum analyser with the resolution bandwidth set to 100 kHz, yielding a total sweep time of 5 ms. Despite this fact the scan time had to be limited to 100 ms/pixel. This value was essentially dictated by the time required for data acquisition from the spectrum analyser. To assess the system spatial resolution we used a gold pattern evaporated on top of a glass window. The corresponding THz image is displayed in the left panel, showing that 350 μm wide metal strips can be clearly resolved (λ = 103 μm). Next we imaged a metal razor blade placed behind a sheet of standard, 100 μm thick, A4 paper. Owing to attenuation as well as reflection at the paper/air interface, we found that the additional sheet of paper reduced the signal to approximately - 40 dBm from the original ~ -15 dBm obtained from the surface of the bare blade. Despite the sizeable magnitude of this residual dynamic range (~ 35 dBm) we could not image the metal blade when the latter was placed just on top of the paper sheet [24]. This resulted from the mixer detecting simultaneously reflections from both the metal/paper and the paper/air interfaces. In particular the signal from the latter interface varied significantly over the scanned area, owing to the paper sheet not being perfectly flat (see below). We partially solved these problems by placing a 1mm thick high density polythene spacer between the blade and the paper sheet, resulting in the image displayed in the right panel of Fig. 3. The non-uniform intensity detected from the blade surface is the consequence of the signal being extremely sensitive to the position of the focal plane (this leads to the top-left and bottom-right corners of the blade in Fig. 3 to disappear). This is in contrast with what we found by imaging the bare blade using a 5mm squared area thermopile detector (not shown). We believe that this effect is the result of (i) the extremely reduced dimensions of the whisker antenna, approximately 500 μm long and 1 μm thick, and of (ii) an unwanted feedback on the laser source produced by impedance mismatches at the QCL and the antenna embedded in the corner cube [21]. Measurements to eliminate the latter effect are under way.

 

Fig 3. Examples of THz images. Left: Image of a gold pattern evaporated on top of a TPX window (100 μm² pixels). Right: A razor blade imaged through a standard A4 paper sheet (200 μm² pixels). Regions of different colours correspond to different intensities detected by the mixer as a consequence of the blade surface not being perfectly flat (see text). The light concentric rings visible on the image are produced by rings patterned on the surface of the circular 1mm thick polythene spacer placed between the blade and the paper. Additional non-uniformity of the horizontal edges of the blade (see for instance the bottom edge) is due to the translation stage not regaining exactly the same position in subsequent vertical scans.

Download Full Size | PDF

In the experimental setup used in this work (see Fig. 1), both the local oscillator (LO) and signal “beams”, given by two neighbouring Fabry Perot lines, experience attenuation from the sample. This is unlike standard heterodyne detection, where radiation from the LO directly impinges on the mixer element. This way the mixer response scales with the amplitude and not with the power (as in the present work) of the detected signal, which is the reason for the extremely high levels of sensitivity achievable with this detection scheme. For instance, when pumped with an LO power in the 10 mW range, the present SDM has a nominal NEP of 10^-19 W/Hz^1/2 at room temperature, i.e., exceeding by several orders of magnitude the NEP of standard bolometers. In principle one could modify the present set up by generating two spatially separated beams from the QCL with the help of a beam splitter, and subsequently filter out one single FP line from each of them, however the required filtering selectivity goes beyond the capabilities of present THz technology. Another option is to recombine the two beams on the mixer and perform homodyne detection.

3. Conclusions

In summary we have reported THz imaging using a 2.9 THz QCL and a SDM operating at room temperature. Image acquisition was performed with a 100 kHz resolution bandwidth, yielding a signal-to-noise ratio of ~ 60 dB and an acquisition time of 100 ms/pixel, solely limited by the acquisition software. With a dedicated circuit for the down conversion of the IF signal, the acquisition time could be reduce to ~ 10 μs/pixel. The present imager already outperforms by several orders of magnitude systems based on pyroelectric or thermopile detection, both in terms of signal to noise and acquisition speed, showing the potential of THz QCLs combined with heterodyne detection [21]. Our next goal is to improve the system’s performance for inspection applications, by (i) implementing “true” heterodyne mixing with two independent frequency stabilised QCL sources and by (ii) extending the operation of QCLs to the frequency range between 1 and 2 THz. Indeed, compared to 2.9 THz, in this spectral region the attenuation through materials such as plastic, paper and cloth is reduced by 2 orders of magnitude or more. In addition, at lower THz frequencies the use of planar SDMs offers the potential to realise integrated mixer cameras with video rate imaging capability [25].