December 1, 2008 / Vol. 33, No. 23 / OPTICS LETTERS 2725

Optical detection of terahertz using nonlinearparametric upconversion

M. Jalal Khan,* Jerry C. Chen, and Sumanth KaushikMIT Lincoln Laboratory, 244 Wood Street, Lexington, Massachusetts 02420, USA

*Corresponding author: mjkhan@ll.mit.edu

Received June 10, 2008; accepted October 7, 2008;posted October 21, 2008 (Doc. ID 97312); published November 18, 2008

We extend our work to perform sensitive, room-temperature optical detection of terahertz (THz) by usingnonlinear parametric upconversion. THz radiation at 700 GHz is mixed with pump light at 1550 nm in abulk GaAs crystal to generate an idler wave at 1555.6 nm. The idler is separated, coupled into optical fiber,and detected using a gated Geiger-mode avalanche photodiode. The resulting THz detector has a power sen-sitivity of 4.5 pW/�Hz and a timing resolution of 1 ns. © 2008 Optical Society of America

OCIS codes: 190.0190, 190.2620, 190.4410, 190.7220, 040.0040, 230.4320.

Terahertz radiation is increasingly being applied toremote sensing that spans a variety of applications,such as explosive detection [1,2], penetrative imaging[3], nondestructive evaluation [4], and vibrometry[5]. The capability of these systems depends directlyon the availability of high-power terahertz (THz)sources and ultrasensitive, fast THz detectors. Boththese technology goals are areas of active research[6,7]. However, compared with the near IR, THz tech-nology is relatively immature. Commercially avail-able, room-temperature direct detectors, such as Go-lay cells and pyroelectric, have poor sensitivities.Other commercial THz detectors, such as bolometers,are much more sensitive but require liquid heliumcooling and like their room-temperature counterpartshave small electrical bandwidths. By contrast, opticaldetectors are a mature and commercial technologythat offers excellent sensitivity with large band-widths and room-temperature operation. Photonmultiplier tubes and Geiger-mode avalanche photo-diodes (GM-APDs) enable detection down to thesingle-photon limit at room temperature, with veryfast performance [8,9].

We have previously reported our scheme to lever-age mature optical technology to perform sensitivedetection of THz waves [10]. The THz signal wasparametrically upconverted [11] in GaAs, a ��2� non-linear crystal, to an optical telecom signal by mixingit with a pump beam at 1550 nm. The generated op-tical sideband or idler was then separated from thepump and detected with a p-i-n diode [10]. Other re-search groups have also recently reported upconvert-ing THz to the optical frequency regime using othercrystals such as GaP [12] and MgO:LiNbO3 [13] us-ing near 1 �m sources. Here we extend our work bycoupling the idler into an optical fiber and detectingit using a GM-APD. The technique results in a THzdetector with a noise equivalent power (NEP) of4.5 pW/Hz1/2 and 1 ns temporal resolution.

We used (110)-cut GaAs as the nonlinear materialto perform upconversion owing to its high ��2� nonlin-ear coefficient [14] and low absorption losses at THzand optical frequencies [15]. Transmission measure-ments revealed losses of 0.2 and 0.065 cm−1 at the

THz �0.5–0.7 THz� and optical frequencies (near

0146-9592/08/232725-3/$15.00 ©

193.4 THz), respectively. Figure 1 shows the upcon-version experiment layout. This setup is very similarto that reported in [10], with some key differences. Asbefore, the THz source is a backward wave oscillator(BWO) operating at 700 GHz. The choice was drivenby the relatively high power, 2.5 mW, that was avail-able at this frequency. The multimoded BWO beam iscollected using a pair of off-axis parabolic gold mir-rors and focused on the 4 mm GaAs bulk crystal. The4 mm crystal length corresponds to the coherentbuildup length [11] and allows maximal collinearparametric conversion without having to use a phase-matched crystal. The optical pump pulse train isgenerated from a 0.2 nm spectral slice of amplifiedspontaneous emission that is pulsed using a semicon-ductor optical amplifier. The pulses, with a width of10 ns and spaced by 5 �s, are filtered and amplifiedusing two erbium-doped fiber amplifiers (EDFAs) tothe operating power levels of near 1 W. The opticalpump is passed through two short-pass filters, whichreduce the noise level around the idler wavelength,1555.6 nm, facilitating its detection.

In this experiment instead of using a collimatedpump beam [10], we focused the pump in the centerof the bulk GaAs crystal by using a lens, L1, reducingits diameter from 0.86 mm to 0.59 mm. The overlap-ping optical pump and THz beams nonlinearly mix in

Fig. 1. (Color online) Schematic layout of the THz upcon-version experiment. The chopper is used only in conjunc-tion with the p-i-n diode and is removed when light is

coupled into the GM-APD via the optical fiber.

2008 Optical Society of America

2726 OPTICS LETTERS / Vol. 33, No. 23 / December 1, 2008

the crystal and generate the idler photons at1555.6 nm, given by energy conservation. The diverg-ing idler and pump beams exiting the GaAs crystalare recollimated using a second lens, L2. Two addi-tional lenses, L3 and L4, expand the beam back tothe original diameter, facilitating its coupling into amatched fiber collimator. The residual THz beam di-verges rapidly and is not collected by the 1550 nm op-tics. The optical pump is separated from the idler byusing three dielectric thin-film filters that each at-tenuate the pump by about 25 dB and transmit theidler beam with about 1 dB of insertion loss per filter.We measured a pump attenuation in excess of 70 dBthrough the three filters; the measurement was lim-ited by background noise. The idler photons are thencoupled into fiber by using a matched collimator.Once in fiber, the idler is routed via a JDS Uniphaseswitch/attenuator into a Princeton Lightwave GM-APD. Alternately the light can be coupled into anAndo optical spectrum analyzer (OSA) or to a powermeter.

We also retained the ability to route the idler beaminto a p-i-n diode, as in [10], by using a flip mirror.The large-area p-i-n diode is initially used to maxi-mize the upconversion process by optimally coalign-ing the THz and optical pump beams. Once this hasbeen achieved, the idler beam is coupled into fiber byremoving the flip mirror and adjusting lenses L3 andL4 and the six-axis fiber collimator mount to maxi-mize the signal. This technique enables independentoptimization of the conversion process and the cou-pling of the idler beam into the fiber. The THz beamis chopped only when the p-i-n diode is employed; thechopper is removed when idler light is coupled intothe GM-APD via the optical fiber.

Once in the fiber, the switch/attenuator can beused to either measure the average idler power or becoupled into an OSA. Figure 2 shows the optical spec-trum of the output light. When the BWO is turnedon, a discernible idler peak is visible, which is cen-tered at 1555.7 nm and has a 3 dB bandwidth of0.187 nm. The discrepancy between the expectedidler wavelength, 1555.6 nm, and that measuredhere was attributed to frequency settling of the

Fig. 2. (Color online) Optical spectrum of the light coupledin the fiber clearly shows the generated optical idler when

the THz source is on.

0.2 nm tunable filters in the optical pulse train setup(Fig. 1). The spectral peak expectedly disappearswhen the BWO is turned off. The power in the spec-tral peak is about −71.6 dBm and is less than ex-pected; it was attributed to poor fiber coupling. Afteroptimizing the fiber coupling we measured an aver-age idler power of −66 dBm that roughly corre-sponded to a coupling efficiency of 7 dB.

The idler is then routed into the GM-APD, which istriggered by a delayed copy of the electrical pulsethat drives the pulsed optical source. The trigger ini-tiates detection of idler photons in a 1 ns gate by theGM-APD; since the presence of the idler photons im-plies THz photons, this receiver achieves a THz tim-ing resolution of 1 ns. The GM-APD is equipped witha counter, which records the total number of 1 nsgates with photons over a 1 s time interval. The sen-sitivity or NEP of the GM-APD is limited by the darkcount rate (DCR) or the number of spurious countsper second in the absence of any signal; it is given by

NEP =h�

��DCR. �1�

� is the quantum efficiency of the GM-APD, and � isthe frequency of the incident radiation. The DCR ofthe GM-APD receiver is 20 kHz, which correspondsto an NEP�9.3�10−17 W/�Hz.

To experimentally measure the sensitivity of theTHz detector one can attenuate the THz power untilthe counts registered by the GM-APD are equal tothe fluctuations in the dark count over the 1 s inter-val. This attenuated THz power would then corre-spond to the minimum detectable power. Since cali-brated THz attenuators with a large dynamic rangeare not readily available, we chose to attenuate theidler signal instead. The linear relationship betweenthe idler and THz power justifies this approach. Fig-ure 3 shows a plot of the Geiger-mode counts as theattenuation of the idler beam was varied. The x-axisvalues correspond to the equivalent input THz power,for a given optical attenuation, in units of NEP

Fig. 3. (Color online) Plot of GM-APD counts versusequivalent input THz power, normalized by bandwidth ofgating signal. A linear fit to the signal counts (dashed line)

�

intersect noise level (solid line) at 4.5 pW/ Hz.

December 1, 2008 / Vol. 33, No. 23 / OPTICS LETTERS 2727

�W/�Hz�. The x-axis values are given by scaling theinput THz power, 2.5 mW, by the optical attenuationof the idler and normalizing it to the square root ofthe equivalent bandwidth of the gating signal. Theequivalent bandwidth of the 1 ns pulse train with atime period of 5 �s and integrated over 1 s is 5 kHz�5 �s/1 ns�1 Hz�. The mean number of dark countsin 1 s was measured to be 3.8 and was consistentwith the theoretically expected value of 4 �1 ns/5 �s�DCR�. To calculate the sensitivity of the resultingTHz detector we extrapolate a linear fit to the data tothe noise level. The fluctuation in the dark counts ornoise level is governed by Poissonian statistics and isgiven by the square root of the mean dark counts,�3.8. Using this method we obtained an NEP�4.5 pW/�Hz, as is shown in Fig. 3. This corre-sponds to a minimum detectable THz pulse energy of3.2�10−19 J in the 1 ns GM-APD gate; of course, weintegrate over 2�105 pulses in 1 s.

The relationship between the NEP of the GM-APDand the resulting THz detector is given by

�NEPGM-APD

�Idler� = �THz→optical�NEPTHz

�THz� �2�

where �THz→optical= �output idler photons� /�input THz photon� is the system photon conversionefficiency. �THz and �Idler are the THz and idler fre-quencies, respectively. Using the above equation wecalculate the photon conversion efficiency of the sys-tem to be 7.5�10−8; the corresponding system powerconversion efficiency is 2.1�10−5. The intrinsic pho-ton conversion efficiency is significantly higher andcan be calculated by taking into account the 2 dBeach of Fresnel loss at the uncoated crystal interfacefor the THz and the idler; 3 dB of insertion losses dueto the dielectric filters, and 7 dB of fiber couplinglosses. Alternately, the conversion efficiency can becalculated by using the following expression:

�THz→optical =PIdler

�p�

PTHz�

�THz

�Idler, �3�

where PIdler�p� is the peak idler power and is related to

the average idler power by the duty factor�5 �s/10 ns�. From the measured average idler powerof −66 dBm, we calculate a photon conversion effi-ciency of 1.8�10−7. The 3.8 dB difference betweenthis number and that calculated above is largely dueto the 3 dB insertion loss of the attenuator precedingthe GM-APD.

In summary, we demonstrated a room-temperatureTHz detector that has an NEP of 4.5 pW/�Hz. Its

sensitivity is at least 20 dB better than commercial

room-temperature THz detectors, such as Golay cellsand pyroelectrics; it is comparable to a liquid heliumcooled 4.2 K bolometer. We expect to improve thisperformance further by enhancing the THz-to-opticalconversion efficiency, which is the focus of ongoingwork.

This work is sponsored by the Department of theAir Force under AF contract FA8721-05-C-0002.Opinions, interpretations, recommendations, andconclusions are those of the authors and are not nec-essarily endorsed by the United States Government.

References

1. Y. C. Chen, T. Lo, P. F. Taday, B. E. Cole, W. R. Tribe,and M. C. Kemp, Appl. Phys. Lett. 86, 241116 (2005).

2. C. Baker, T. Lo, W. R. Tribe, B. E. Cole, M. R. Hogbin,and M. C. Kemp, Proc. IEEE 95, 1559 (2007)

3. A. M. Sinyukov, A. Bandyopadhyay, A. Sengupta, R. B.Barat, D. E. Gary, Z. Michalopoulou, D. Zimdars, andJ. F. Federici, Proc. SPIE 6373, 63730 (2006).

4. J. White and D. Zimdars, in Conference on Lasers andElectro-Optics/Quantum Electronics and Laser ScienceConference and Photonic Applications SystemsTechnologies, OSA Technical Digest (CD) (OpticalSociety of America, 2007), paper JWA112

5. J. C. Chen and S. Kaushik, IEEE Photon. Technol.Lett. 19, 486 (2007).

6. M. Tonouchi, Nat. Photonics 1, 97 (2007).7. P. Siegel, IEEE Trans. Microwave Theory Tech. 50, 910

(2002)8. X. Jiang, M. A. Itzler, R. Ben-Michael, and K.

Slomkowski, IEEE J. Sel. Top. Quantum Electron. 13,895 (2007).

9. S. Verghese, J. P. Donnelly, E. K. Deurr, K. A.McIntosh, D. C. Chapman, C. J. Vineis, G. M. Smith, J.E. Funk, K. E. Jensen, P. I. Hopman, D. C. Shaver, B.F. Aull, J. C. Aversa, J. P. Frechette, J. B. Glettler, Z. L.Liau, D. C. Oakley, E. J. Ouellette, M. J. Renzi, and B.M. Tyrrell, IEEE J. Sel. Top. Quantum Electron. 13,867 (2007).

10. M. J. Khan, J. C. Chen, and S. Kaushik, Opt. Lett. 32,3248 (2007).

11. R. W. Boyd, Nonlinear Optics, 2nd ed. (Academic,2003).

12. Y. J. Ding and W. Shi, Solid-State Electron. 50, 1128(2006).

13. R. Guo, T. Ikari, H. Minamide, and H. Ito, inConference on Lasers and Electro-Optics/QuantumElectronics and Laser Science Conference and PhotonicApplications Systems Technologies, OSA TechnicalDigest (CD) (Optical Society of America, 2007), paperQPDA7

14. V. G. Dmitriev, G. G. Gurzadyan, and D. N.Nikogosyan, Handbook of Nonlinear Optical Crystals,Springer Series in Optical Sciences, Vol. 64 (Springer-Verlag, 1997).

15. E. D. Pallik, Handbook of Optical Constants of Solids,

(Academic, 1998).