Frequency response measurement and SNR improvement ofthree infrared detectors
Olivier N. Kuttel and Rolf Philipona
A comparison of three different infrared detectors in the 10.6-,um region is presented. A method formeasuring the frequency response from dc up to 100 MHz in a continuous way is developed. We show that theflatness of the frequency response does not only depend on the electronic circuit following the detector,matching the impedance between the detector and the preamplifier, but also on the parallelism of the localoscillator and the signal beam. The signal-to-noise ratio of the detector signal is improved and reaches thetheoretical limit by chopping only the signal beam and detecting the spectral analyzed intermediate frequen-cy signal with the lock-in technique. Detecting the signal in this new way suppresses externally detectednoise, which is omnipresent in weak signal detection.
1. Introduction
Infrared heterodyne detection of light beating is apowerful technique with many uses. The basic phys-ics is well known. 1-3 The theory shows that an infra-red detector may in principle approach the quantumdetection limit of 2hvB/ for photoconductive detec-tors,2 but in practice a number of factors significantlyreduce the sensitivity.
Several authors have investigated the signal-to-noise ratio (SNR) in homodyne and heterodyne detec-tion systems. Jakeman et al. 4 reviewed optical homo-dyne detection. Foord et al.5 reported a comparison ofexperimental and theoretical SNRs in CO2 laser het-erodyne systems, while Abbas et al.6 examined thesensitivity limits of an infrared heterodyne spectrome-ter for astrophysical applications. Mandel and Wolf7show the existence of an optimum receiver size whichmaximizes the SNR.
For heterodyne detection with large local oscillatorpower, the main noise contribution is due to shot noise.Other external noise sources are often present, reduc-ing the sensitivity of detection. We show that applica-tion of the lock-in technique in an unusual way im-proves the SNR and suppresses all externally detected
When this work was done both authors were with University ofFribourg, Physics Department, CH-1700 Fribourg, Switzerland; R.Philipona is now with University of California, Electrical Engineer-ing Department, Los Angeles, California 90024.
noise. The theoretical detection limit may beachieved in most cases.
One important characteristic of detectors is the fre-quency response, which can be described theoreticallyas the spectral response to a short laser pulse. Thismethod is not very suitable for an experimental deter-mination. A better means to determine the frequencyresponse is to measure the beating between a localoscillator and a frequency shifted signal beam. Theamount of the frequency shift is produced with an AO(acoustooptic) modulator. The disadvantage of thissystem is the necessary readjustment of the optics foreach frequency, following the Bragg angle conditionand the nonconstant frequency shifted power over therequired bandwidth. Only continuous measurementusing a constant laser power source could give informa-tion about the kind of frequency response.
In the following we present a method to measure thefrequency response in a continuous way up to 100MHz. This experimental setup allows the examina-tion of the Siegman antenna theorem.- 10 We showthat a violation of this theorem reduces the sensitivityof the detection and significantly changes the order ofthe transfer equation of the detection system. Otherfactors, such as impedance matching between detec-tors and electronics and the manner of picking up theintermediate frequency (IF) signal from the detector,can influence the frequency response.
11. Experiment
A. Experimental Setup
The basic idea is given by Teich.2 3 The beam from aCO2 waveguide laser (10.6 gm, linearly polarized) withan output power of 2.5 W is incident on a Michelson
interferometer (Fig. 1). In one leg of the interferome-ter a 0.15-m diam rotating aluminum wheel is fixed,built in a closed vacuum cavity, and evacuated down to10 Pa. Reducing the air resistance in this way, themaximum rotating speed becomes -34,000 rpm. Toreduce acoustic vibrations, the wheel is separated fromthe optical table and mounted on a strengthened rack.
The purpose of lens Li is twofold: First it serves tofocus the radiation through a ZnSe window to a singlepoint on the sandblasted rim of the rotating wheel, andsecond it collects the reflected and frequency shiftedsignal. The iris, maintains the angular alignment ofthe wavefronts of the two beams. Optimum SNR atthe output of the detector is obtained by carefullyadjusting mirror Ml. The chopper can be placed ei-ther at the output of the laser, chopping the signal andlocal oscillator beam together, or in the signal or localoscillator leg. The local oscillator power is measuredwith a bolometer in front of the detector. Blocking thelocal oscillator, the reflected, weak power is measuredin video with a lock-in amplifier, working at a choppingfrequency of -200 Hz. This technique allows us todetect in video absolute laser power down to 10-9 W.
The frequency shift of the incident radiation is givenas Af = (2VI/X), where V11 is the velocity componentparallel to the laser beam. Focusing the incidentbeam on the top of the wheel, maximum frequencyshift is achieved with a maximum velocity of V11Vwheel = 270 m/s. This corresponds to a rotating speedof 34,000 rpm and leads to a frequency shift of -50MHz. Measurements of the frequency response aremade by sweeping the spectrum analyzer as fast aspossible after the propulsion of the wheel has beenswitched off and photographing the scope until thewheel has come to a standstill (-i min). Examples areshown in Fig, 5.
The great advantage of this layout is that the opticsremains unchanged while the frequency is shifted fromdc up to the desired value. Also, the reflected lightpower is constant and independent of the wheel speed,a necessary condition to measure the frequency re-sponse and to realize an absolute calibration of thedetector senstivity over the required bandwidth.
B. Detectors and Signal Processing
Three different detectors are used: a HgCdTe pho-tovoltaic (PV) and a HgCdTe photoconductive (PC)(77 K), both from New England Research Center, withincorporated preamplifier and a GeHg cooled down to
Table 1. Detector Data
Operation TypicalOperation Area Bandwidth tempera- LOpower
20 K by a closed cryocooler system (Table I). Theelectronics necessary for the GeHg detector as well asthe Dewar have been constructed in our workshop. Toallow a maximum choice of different electronic param-eters a layout as shown in Fig. 3 was built. By short-circuiting one or several connectors, B1 ... B6, one canchoose the cooled load resistor RL and therefore thefrequency bandwidth B. Exchanging another pack ofsix resistors can easily be done. Connecting the biasvoltage UB = 30 V to point Al (A2) and short-circuitingA2 (Al) gives the output voltage Us over load resistorRL (detector). Thus, an examination of the best pa-rameter set for a given bandwidth B is possible.
The signal Us is preamplified according to the detec-tor in use (Fig. 4). An HP 4847 preamplifier (50-Qinput impedance) and a Tektronix TE 7A26 verticalplug-in amplifier (10-Mg input impedance), both ofwhich show good noise characteristics, are employed.The video output of the Tektronix spectrum analyzeris fed to the lock-in amplifier, working at the choppingfrequency of the signal beam -200 Hz. The frequencyreference is produced with a He-Ne laser, crossing thechopper beside the CO2 beam. Sweeping the spec-trum analyzer very slowly (50 s/sweep) over the desiredfrequency range significantly improves the SNR. Anexample of a spectrum drawn by an X-Y plotter isshown in Fig. 9. Frequency response measurementsare made without a lock-in amplifier by photographingthe scope. Details are described in Sec. II.A.
Ill. Results and Discussion
A. Frequency Response
Our purpose is to measure the frequency responseup to a maximum of 50 MHz. Figure 5 shows thefrequency response for different detectors. Note thedifference for the GeHg detector in Figs. 5(c) and (d)achieved by choosing another load resistor. It is re-markable that measurements with the preamplifierssupplied by the manufacturer do not work. TheHgCdTe (PV) detector is impedance mismatched,which results in a significant decrease of sensitivity,while the HgCdTe (PC) detector oscillates for somediscrete frequencies. Therefore, measurements weredone with the TE 7A26 and the HP 4847 preamplifiers(Fig. 4).
Siegman's theorem predicts a decrease in sensitivityif the parallelism between the local oscillator (LO) andthe signal beam exceed a phase variation of more thanone wavelength across the aperture: AO > /d (d =diameter of the aperture). Figure 6(b) shows that notonly a drop in sensitivity but also an important changein frequency response results from a misalignment ofthe phase fronts. If measurements with a large band-width are done, careful control of parallelism of LOand signal beam is necessary. A similar result is shownin Fig. 6(c). Impedance mismatching between the de-tector and the preamplifier generates this undulatingresponse. While the output signal is detected acrossthe load resistor in Figs. 6(a)-(c), Fig. 6(d) shows avoltage detected across the detector, resulting in im-
Fig. 1. Optical setup to measure the frequency response with arotating wheel continuously up to a maximum frequency shift of Af= 50 MHz: A = calibrated attenuator; B = bolometer; BS = beamsplitter; CH = chopper; D = detector; L = lens; M = mirror. Thechopper position can be changed in the local oscillator or signal
beam.
HgCdlefPVf referencei f~~~~~~rom chopper
Hg~d~efP~l IE analyzer video apiir Vl
"g(PC) A _ J Track gen.
Preamplifier
Fig. 4. Electronic assembly. The signal is preamplified accordingto the detector in use. Frequency response measurements are doneby photographing the scope of the spectrum analyzer. The lock-inamplifier and X-Y plotter are only employed for the SNR and theprin determination. The lock-in amplifier is locked to the chopperin the signal beam. Accurate video power measurements down to10-9 W are done by connecting the output of the preamplifier direct-ly to the lock-in amplifier and by chopping the beam incident to the
detector.
Fig. 2. Frequency response measurements up to 100 MHz. TheAO cell shifts the frequency by an amountf, (for example, 50 MHz);the rotating wheel increases the shift continuously to fl + Af, accord-ing to the rotation direction of the wheel. For clarity, attenuators in
the local oscillator and the signal beam are omitted.
Al A1
GeHg__Detector
Signal_out
A2 C B B2 B3 B& BS B6
Fig. 3. Electronic circuit of the GeHg detector. The bias voltageUB = 30 V is either connected to Al (and short-circuiting A2) whichgives the signal out across the load resistor or to A2 (short-circuitingAl) which gives the signal out across the detector. With the connec-tors B1 ... B6 the cooled load resistor is chosen, determining the
spectral bandwidth of the detector.
pedance mismatching. The a priori assumption of alinear transfer equation for detection systems is rarelysatisfied. Consequently, care must be taken to assurea correct mode of operation.
If one is interested in bandwidths larger than 50MHz, the setup shown in Fig. 2 is recommended. In afirst step the frequency response is examined in themanner described above. With the AO cell a frequen-
, _ _ .e P
MUIFig. 5. Frequency responses of detectorsHgCdTe (PV), (c) GeHg (1 MHz, RL = 47
MHz, RL = 51 T).
(a) HgCdTe (PC), (b)kg), and (d) GeHg (50
cy shift of 50 MHz is then fixed. Rotating the wheelfrom a standstill to its maximum speed (fD = 50 MHz)gives a frequency bandwidth from 50 to 100 MHz.Joining together both spectra determines the responsefrom dc up to 100 MHz. Let us point out that the AOcell can produce larger frequency shifts than 50 MHzresulting in a larger bandwidth. The reflected powerfrom the wheel is constant for each frequency interval,so that no problem arises in putting the different spec-tra together.
B. SNR and Psin Measurements with the Lock-InAmplifier
Before looking at the results we must clarify thevalue of bandwidth B in the formula below for SNR(the reader is referred to the paper by Foord et al. 5 ):
SNRleffPsSNR = p'hvB (1)
where Jeff is the effective quantum efficiency of the
Fig. 6. GeHg detector, RL = 51 Q. Dependence of the frequencyresponse on (b) parallelism of signal beam and local oscillator, (c)impedance matching, and (d) influence of the detected IF signal; (a)
shows the unperturbed spectral response.
,.B
cm
0.1
- -Fig. 8. Bandwidth B, of the scattered signal for four different filterresolutions BF; B, - 330 kHz (-3 dB). As long as BF < B, the SNRremains constant (a)-(c). If BF > B8, the signal amplitude is con-stant but the SNR decreases by increasing filter bandwidth BF
(c),(d).
as_0
-CL
*G9. _LO
0.01I I -I ,',, ,, , , ,,, I0.1 1 10
Frequency stiff of sinal beam [MHz
Fig. 7. Frequency shift of the signal beam vs the bandwidth of thesignal. The result is in good agreement with the theory B8 - fDoppler
detector, P8 is the power, and B is the appropriatespectral bandwidth, either the filter bandwidth BF Ofthe spectrum analyzer or the bandwidth B, of thesignal reflected from the wheel; whichever has thelargest bandwidth gives the value of B. Teich 2 gives aformula for bandwidth B, of a scattered laser beam:
BVI B5~~~~~ do ~~~~(2)d
where V± is the velocity component perpendicular tothe laser beam, V = V- coso (Fig. 1), and d is the spotsize of the focused beam:
FX\d~ DFX (3)D
with F the focal length of the lens and D the diameterof the beam.
For the bandwidth B, of the signal we getV1D
B.- F (4)FX
While the factor D/(FX) is a constant for a fixed optical
20 MHz 20 MHzfrequency
Fig. 9. Weak signal detection (a) with only the spectrum analyzerand (b) with the lock-in amplifier. The signal power is 8 X 10-1 2 W
(-50-dB trace) and 8 X 10-13 W (-60-dB trace).
setup, V increases with the rotating speed of thewheel and therefore with the Doppler frequency. Fig-ure 7 shows a measurement of this dependence. If wewant to compute the SNR we must know the values ofB, and BF to choose the larger bandwidth. Figure 8provides the IF signal scattered from the wheel for fourdifferent filter resolutions. The Doppler frequency is18 MHz. We can measure the bandwidth B, of thesignal (-3 dB) in Fig. 8(a) to -330 kHz. A larger filterbandwidth does not change the SNR [see Figs. 8(a)-(c)], because B, is the adapted value for B in Eq. (1).Let us remark that, although the reflected signal isconstant, the amplitude of the IF signal on the scope ofthe spectrum analyzer is not, as long as B, is larger thanBF. This is very important to know for absolute mea-surements of the signal. Also, one does not gain inSNR by reducing the filter bandwidth of the spectrumanalyzer, as long as bandwidth B, of the IF signal islarger.
Let us now calculate the minimal detectable powerPnin The reflected video power is measured with the
Table 11. Minimum Detectable Power for Different Detectors
HgCdTe HgCdTe GeHg(PC) (PV) f 1 MHz f ' 50 MHz
prnin/B (W/Hz) 1.5 X 10-18 3 X 10-16 2.5 X 10-19 1.8 X 10-8without lock-in
Prnin/B (W/Hz) 5 X 10-19 8 X 10-17 2.5 X 10-20 1.8 X 10-19with lock-in
Bandwidth B is either the filter bandwidth BF or the signal band-width B,, whichever is larger.
lock-in amplifier to 8 X 10-7 W. Table II summarizesthe values obtained for operation with and without alock-in amplifier. The GeHg detector reaches, for the1-MHz bandwidth in combination with the lock-inamplifier, the theoretical limit of 3.8 X 10-20 W/Hz fora photoconductor and a quantum efficiency ij of unity.In reality is -0.5 which emphasizes the excellentoperation of these detection systems. It is still notclear why the photovoltaic detector gives a poorer pmin.
Use of the lock-in amplifier in the described mannergives an improvement of the SNR (Table II). Tworeasons are responsible: First, all externally detectednoise is eliminated, a factor which is important fordetecting very weak signals in a noisy environmentand, second, the lock-in amplifier is capable of improv-ing the SNR by 5-10 dB; Fig. 9 shows an example. Thetraces at the left are the result of measuring the signalwith the spectrum analyzer alone. Comparison withthe results on the right shows this improvement. It isimportant to bear in mind that only the signal beam ischopped.
IV. Conclusions
The spectral response of a detector depends on theimpedance matching between detector and preampli-fier and on the parallelism of local oscillator and signalbeam. This conclusion goes beyond Siegman's theo-
rem which only predicts a decrease in sensitivity formisaligned beams. The frequency response of a detec-tor is made flat only by a careful check of these points.
The SNR is significantly improved by chopping onlythe signal beam and detecting the video output of thespectrum analyzer with a lock-in amplifier. This newmethod is reasonable if one detects very weak signalsin a perturbed environment. An increase in SNR upto 10 dB is observed in comparison with usual meth-ods, and all the externally detected noise is eliminated.
We are greatly indebted to D. Aeby for assistancewith this work. We thank P.-A. Paratte, EPF-Lau-sanne, for many valuable discussions. This work wassupported by the Swiss National Science Foundation.
References1. R. H. Kingston, Detection of Optical and Infrared Radiation
(Springer-Verlag, New York, 1978).2. M. C. Teich, "Detection in the Infrared," in Semiconductors
and Semimetals, edited by R. K. Willardson and A. C. Beer(Academic, New York, 1970), Vol. 5, Chap. 9.
3. M. C. Teich, "Infrared Heterodyne Detection," Proc. IEEE 56,37 (1968).
4. E. Jakeman, C. J. Oliver, and E. R. Pike, "Optical HomodyneDetection," Adv. Phys. 24, 349 (1975).
5. R. Foord, R. Jones, J. M. Vaughan, and D. V. Willetts, "PreciseComparison of Experimental and Theoretical SNRs in CO2Laser Heterodyne Systems," Appl. Opt. 22, 3787 (1983).
6. M. M. Abbas, M. J. Mumma, T. Kostiuk, and D. Buhl, "Sensitiv-ity Limits of an Infrared Heterodyne Spectrometer for Astro-physical Applications," Appl. Opt. 15, 427 (1976).
7. L. Mandel and E. Wolf, "Optimum Conditions for HeterodyneDetection of Light," J. Opt. Soc. Am. 65, 413 (1975).
8. A. E. Siegman, "The Antenna Properties of Optical HeterodyneReceivers," Proc. IEEE 54, 1350 (1966).
9. V. J. Corcoran, "Directional Characteristics in Optical Hetero-dyne Detection Processes," J. Appl. Phys. 36, 1819 (1965).
10. W. S. Read and D. L. Fried, "Optical Heterodyning with Non-critical Angular Alignment," Proc. IEEE 51, 1787 (1963).
NASA continued from page 2948
fied, tracking filtered, and converted by phase-lock loop to squarewaves. When a positive-going edge of the squared version of wave-form B occurs while squared A is in the low state, the latch output isset high. When a positive-going edge of squared A occurs whilesquared B is low, the latch output is then set low.
Thus the latch output alternates between the high and low statesas frequently as the occurrence of the positive-going edge of squaredA alternates between the high and low states of squared B. Thefrequency of this alternation, averaged over a large number of cycles,is just the difference in frequency between signals A and B. By useof two more basic digital circuits shown in Fig. 6, the digital-differ-ence processor can detect if A > B or A < B. This indicates targetpath, and a low Doppler difference indicates a target collision path.
The number of cycles in the latch output is counted during 213cycles of signal B, which serves as the frequency-and-time reference.The count thus equals 213 times the, ratio of the difference (orDoppler) frequency to the frequency of signal B. The count isdisplayed, and the counter is reset every 213 cycles for repeated
updating. By measuring digital difference for distance traveled, thesame difference readout is obtained for all speeds.
Other uses for the digital-difference processor are for frequencydiscrimination in frequency demodulation, digital phase detection,frequency-shift-keying demodulation, and the like. The techniquedetermines the target path mathematically, which makes its re-sponse faster than that of the tracking antenna. The technique,therefore, can be used for tracking cars, missiles, bullets, baseballs,and other fast-moving objects.
This work was done by Paul Shores, Chris Lichtenberg, Herbert S.Kobayashi, and Allen R. Cunningham of Johnson Space Center.This invention is owned by NASA, and a patent application has beenfiled. Inquiries concerning nonexclusive or exclusive license for itscommercial development should be addressed to the Patent Coun-sel, Johnson Space Center, E. K. Fein, Mail Code AL3, Houston, TX77058. Refer to MSC-20865.
