320-channel dual phase lock-in optical spectrometer

P. S. Fodor, S. Rothenberger, and J. Jevy

Citation: Review of Scientific Instruments 76, 013103 (2005); doi: 10.1063/1.1830013
View online: http://dx.doi.org/10.1063/1.1830013
View Table of Contents: http://scitation.aip.org/content/aip/journal/rsi/76/1?ver=pdfcov
Published by the AIP Publishing
Articles you may be interested in
Band alignment tuning of InAs quantum dots with a thin AlGaAsSb capping layer
Appl. Phys. Lett. 102, 173104 (2013); 10.1063/1.4803013
Temperature dependent and time-resolved photoluminescence studies of InAs self-assembled quantum dots
with InGaAs strain reducing layer structure
J. Appl. Phys. 106, 013512 (2009); 10.1063/1.3159648
Fabrication and optical properties of polymeric waveguides containing nanocrystalline quantum dots
Appl. Phys. Lett. 85, 4469 (2004); 10.1063/1.1818723
Femtosecond pump-probe near-field optical microscopy
Rev. Sci. Instrum. 70, 2758 (1999); 10.1063/1.1149841
Time resolved magneto-optical spectroscopy on InGaAs nanostructures grown on (311)A and (100)-oriented
substrates
Appl. Phys. Lett. 74, 676 (1999); 10.1063/1.122984

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

REVIEW OF SCIENTIFIC INSTRUMENTS 76, 013103 (2005)

320-channel dual phase lock-in optical spectrometer

P. S. Fodor,a) S. Rothenberger, and J. Jevy
Department of Physics and Astronomy, University of Pittsburgh, 3941 OHara Street, Pittsburgh,
Pennsylvania 15260

(Received 6 May 2004; accepted 11 October 2004; published online 20 December 2004)
The development of a multiple-channel lock-in optical spectrometer (LIOS) is presented, which
enables parallel phase-sensitive detection at the output of an optical spectrometer. The light intensity
from a spectrally broad source is modulated at the reference frequency, and focused into a
high-resolution imaging spectrometer. The height at which the light enters the spectrometer is
controlled by an acousto-optic deflector, and the height information is preserved at the output focal
plane. A two-dimensional InGaAs focal plane array collects light that has been dispersed in
wavelength along the horizontal direction, and in time along the vertical direction. The data is
demodulated using a high performance computer-based digital signal processor. This parallel
approach greatly enhances (by more than 100x) the speed at which spectrally resolved lock-in data
can be acquired. The noise performance of a working system optimized for the 1.3 m wavelength
range is analyzed using a laser diode light source. Time-resolved absorption traces are obtained for
InAs quantum dots embedded in a GaAs matrix, and for dispersed films of PbSe nanocrystals.
2005 American Institute of Physics. [DOI: 10.1063/1.1830013]

I. INTRODUCTION

Optical spectroscopy is a pervasive tool across physical,

chemical, and biological sciences. With a typical spectrometer, light focused on the entrance slit is dispersed by a diffraction grating which spatially separates the light according
to wavelength. Spectroscopic information is obtained by
scanning the angle of the diffraction grating, allowing different wavelengths to pass through an exit slit. The availability
of charge-coupled device (CCD) and focal plane array (FPA)
detectors has greatly reduced the amount of time required for
the acquisition of optical spectra. Further parallelization has
been achieved by using so-called imaging spectrometers
that are optimized to preserve the vertical point-spread function of the spectrometer, allowing multiple input fibers to
be imaged onto different horizontal portions of the detector
array without interference.1 The increased throughput of such
parallel techniques has become critical for applications such
as Raman spectroscopy2 and near-field scanning optical
microscopy,3 where the signals of interest are small and
background free.
For many applications, the signal of interest is superimposed upon a large noisy background. Applications that fit
this description include spectrophotometry, and a wide variety of linear and nonlinear spectroscopies in which information about the sample is encoded in the intensity or polarization of light.4,5 A simple background subtraction can be
achieved by synchronizing the frame acquisition to an optical chopper or similar modulator. However, the spectral response of such a technique is functionally equivalent to a
comb filter and thus is less effective than bandwidth narrowing methods that use true lock-in amplification.6,7 Furthera)

Electronic mail: [email protected]

0034-6748/2005/76(1)/013103/7/$22.50

more, the range of modulation frequencies is limited by the

camera frame rate, which is usually quite low. Laser noise
usually falls off significantly above a few kilohertz, higher
than the frame rates of most high-sensitivity photometric
imaging devices like CCDs and FPAs. As a consequence,
true phase-sensitive detection or lock-in amplification810 is
the method of choice in the above-mentioned situations,
when modulation at frequencies higher than a few kilohertz
is desired. However, due to the high photon fluxes and large
backgrounds, it is not straightforward to parallelize lock-in
techniques.
Here we present an implementation of a multiplechannel lock-in optical spectrometer (LIOS) which enables
phase sensitive detection methods to be applied in parallel,
greatly enhancing the speed (by more than 100x) with which
optical spectra can be acquired. The operating principle of
LIOS is described, and operation is demonstrated using a
commercial 1310 nm laser diode. Also, time-resolved absorption measurements are presented for InAs quantum dots
embedded in a GaAs matrix, as well as for dispersed films
containing PbSe nanocrystals.
II. PRINCIPLE OF OPERATION
A. Overview

A straightforward implementation of a LIOS requires a

one-dimensional detector array and a minimum of one
lock-in channel per detector. Lock-in amplifiers that employ
digital-signal-processing (DSP)11 methods have rapidly replaced the original analog lock-in amplifier design invented
by Dicke.7 Digital lock-ins have superior phase stability and
dynamic range, making them favorable for most demanding
applications except perhaps milikelvin transport experiments
where digital circuitry is usually avoided. Banks of up to 32
digital lock-in channels are commercially available,12 but for

76, 013103-1

2005 American Institute of Physics

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-2

Fodor, Rothenberger, and Jevy

Rev. Sci. Instrum. 76, 013103 (2005)

FIG. 1. Schematic drawing of the lock-in spectrometer optical setup. (FC)

fiber coupler; (GL)GlanLaser polarizer; (AO)acousto-optical deflector;
(L)collimating lens; (DG)diffraction grating part of a CzernyTurner
monochromator; and (DSP)digital signal processor.

a reasonable number of channels (e.g., 256 or 512), this approach quickly becomes financially burdensome and unwieldy.
Here an alternative approach is described that combines
the parallelism of an imaging device with the DSP capabilities of modern lock-in amplifiers. The optical diagram of
LIOS is depicted in Fig. 1. The key elements are a highefficiency acousto-optic deflector (AOD) (A-A Optoelectronic, France), a high-resolution imaging spectrometer
(McPherson Inc., Chelmsford, MA), a two-dimensional InGaAs FPA (Phoenix NIR Camera, Indigo Systems, Goleta,
CA) and a high-performance computer-based DSP. The FPA
collects spectral information along the horizontal direction
(dispersed by the spectrometer), and temporal information
along the vertical direction (dispersed by the AOD). The
time-domain information is Fourier-decomposed using a
computer-based DSP to produce X and Y lock-in outputs for
every wavelength channel. A more detailed description of the
operation of LIOS is given below.
Light from the experiment is coupled into a polarizationpreserving single-mode optical fiber, and expanded and collimated into a 6 mm diameter beam. The AOD chosen for the
LIOS has an efficiency of 85% at 1.3 m, a deflection range
of 40 mrad, and a bandwidth of 30 MHz at 1000 nm. Generally, better vertical imaging is achieved with longer focal
length F spectrometers, so a F = 2.0 m spectrometer f / #
= 14.1 is chosen. A ray-tracing analysis of the spectrometer
indicates that a 10 m diameter spot incident on the entrance
slit forms an approximately 32 m spot at the detector plane.
The (FPA) consists of a two-dimensional array (320
horizontal256 vertical) of InGaAs square pixel detectors,
30 m in size, with sample-and-hold circuitry behind each
pixel. The maximum diffraction efficiency of the AOD covers only about half of the vertical range of the camera, so a
subset of the camera, consisting of nv = 128 vertical pixels
nh = 320 horizontal pixels, is used. The FPA is sensitive in
the near-infrared region (0.82.6 m) and has a specified
noise-equivalent irradiance 1.5 107 W cm2 for the gain
setting employed. The camera is equipped with four 14-bit
analogtodigital converters operating from a 40 MHz bus,
allowing data transfer rates of up to 80 MB/s. Optimal
throughput is achieved using integrate-while-read mode, in
which data from the previous frame is transmitted while the
current frame is being acquired.
The AOD diffracts the light, scanning it vertically along

FIG. 2. Triggering diagram for the data acquisition and acousto-optical deflector control. The signals controlling the amplitude modulation are shown
both for normal operation () and calibration (---).

the entrance slit of the spectrometer. Thus, at the exit port the
light is dispersed by the high resolution imaging spectrometer along the horizontal axis according to wavelength, and
by the AOD along the vertical direction according to time
(Fig. 1). The AOD is controlled by a radio-frequency (rf)
source that can be amplitude-modulated (AM) and
frequency-modulated (FM) to control the intensity and angle
of the diffracted beam, respectively. Both AM and FM control signals are produced by synchronized arbitrary wave
form generators. For broadband spectral applications, an
AOD deflection provides 50%70% diffraction efficiency,
but its flexibility, stability, and programmability outweight its
efficiency limitations. Additionally, the AM capability can be
used for calibration purposes, as described below. The encoded optical information in both the spectral and time domains, is collected by the FPA. The acquisition is triggered
by a TTL signal that is phase-locked to the reference frequency f R, enabling phase-sensitive detection. Subsequently,
the data is demodulated using a high performance computerbased digital signal processor.
B. Digital signal processing

The sequence of images obtained by the FPA is transferred in real time to a PC (Pentium Xeon dual processor at
2.2 GHz, 4 GB RAM) which performs the DSP tasks. To
discuss the signal processing, we restrict ourselves to the
case of monochromatic light, so that only a single spectral
channel is involved. The generalization to multiple channels
is straightforward. We adopt the convention that functions of
time (treated as a continuous index t) are represented with
tildes [e.g., ft], and occasionally are written without an
explicit time index (i.e., f). In a LIOS measurement, the
signal of interest is modulated at a fixed reference frequency
f R = 1 / TR. The signal S St , t0 t t0 + mTR, where m is
an integer, represents the time-dependent intensity measured
by a fixed detector. This signal is encoded by LIOS in the
following manner. As seen in Fig. 2, the frequency of the
AOD is ramped linearly in time, triggered by the phase of
the reference signal. The diffracted light from the AOD is
focused by a lens into the spectrometer and subsequently
imaged onto a one-dimensional vertical array of pixels from
the FPA. The signal image is obtained as the difference of

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-3

Multi-channel lock-in spectrometer

Rev. Sci. Instrum. 76, 013103 (2005)

FIG. 3. Signal image extraction using the difference method to eliminate large backgrounds.

two separate frames A and B obtained after triggering on the

positive and negative edge of the reference signal. This provides a straightforward reduction of the background arising
from the dark count of the camera detector and from the
unmodulated scattered light, allowing the extraction of small
relevant signals (Fig. 3).
In principle, it should be possible to calibrate the spectrometer in order to extract lock-in information from the
measured array. While it is possible to model the entire system accurately, taking into account nonlinearities and nonuniformities from the AOD, and variations in pixel gain and
offsets from the FPA, there is a much more straightforward
and accurate method of extracting lock-in signals from the
measured response.
The LIOS detector can be regarded as a linear mapping
C from the input signal S = St onto an array of pixel inten-

present system, the PSF 3 with a 10 10 m aperture.)

For lock-in detection, it is necessary to define an appropriate scalar product or metric for applying Fourier decomposition methods to the image of the original signal. The
following metric is useful for determining the Fourier coefficients of a signal St that is periodic in TR : AB
2/TRT0 RAtBtdt. With this definition, a natural basis
for the space of square-integrable functions is given by
X0t 1 / 2, Xkt cos2kf Rt, k = 1 , 2 , , and Y lt
sin2lf Rt, l = 1 , 2 , ,. These basis functions have the
property XkY l = 0 and XkXl = Y kY l = kl, where kl is the
Kronecker delta function. Fourier decomposition of an arbi-

sities S = Sk , k = 1 , , nV : SS. Because there are a finite

number of pixels, this mapping is not invertible; however,
it preserves the spectral information of interest, namely, the
Fourier components at the reference frequency as well as
higher harmonics, provided the number of effective vertical
pixels exceeds the Nyquist criterion: nVeff 2m. (The number
of effective vertical pixels per reference period T is given
by neff
v = nv tscan f R PSF, where tscan is the time it takes to
scan the FPA with the AOD, and PSF is the point-spread
function full width at half maximum (FWHM). For the

FIG. 4. Graphic representation of the metric M extraction from the calibration signals for a single wavelength channel.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-4

Fodor, Rothenberger, and Jevy

Rev. Sci. Instrum. 76, 013103 (2005)

FIG. 5. (a), (b) Images of the

sin2 f Rt and cos2 f Rt, respectively, reference vectors. (c), (d) line
cuts along the wavelength channel 54.

trary function St = k=0

akXkt + k=1
bkY kt is given by an
= SXn and bn = SY n.
Because there is no direct access to St, the task is to
find the best estimate for an and bn, given the measured array
of pixel values Sk. Formally, this can be done by defining a
scalar product for the domain of C : A , B C1AC1B.
The mapping C is singular (noninvertible), but singular value
decomposition (SVD) may be used to find the best estimates
for an and bn.
The first step is to calibrate the system by measuring
images of the basis functions Xn and Y n. These functions are
measured by amplitude modulating the AOD at the appropriate frequency and phase. A matrix C can be constructed as
follows: C = X0X1Y 1X2Y 2T. This matrix is a representation of the mapping C, and can be used to construct a
metric. Using SVD, we may write C = U VT. A metric
M can be generated from this decomposition, with the property that: CT M C = I. Simple mathematics gives: M
= U 2 UT. The determination of M for a single wavelength channel is graphically represented in Fig. 4. An identical algorithm is used to determine the corresponding metrics for all the wavelength channels employed. The
calibration procedure must be repeated if the modulation frequency or the central wavelength (grating position) are
changed, or if the overall light intensity (not the ac modulation) varies significantly. Once the M and C matrices are
known for all the wavelength channels, the Fourier coefficients of interest ab a0 , a1 , b1 , a2 , b2 , , can be quickly
calculated from a measurement S using the metric M as fol-

lows: ab = C , S = CT M S. Typically, the components of

the signal are calculated up to the second harmonic. However, if desired higher harmonics can be easily included, with
an associated increase in computational resources. The range
of available frequencies f R is limited only by the bandwidth
of the AOD, which in our case is a few megahertz. For high
frequencies, a multiple-coat method is used in which the
beam is scanned vertically an integer number of times across
the FPA, before the data is transferred to the DSP.
Figure 5 shows sine and cosine reference images obtained by acquiring 1000 individual frames for a modulation
frequency f R = 42 kHz. The 5 shear observed in the images
comes from intrinsic dispersion in the AOD. This shear does
not impact in any way the performance of the design, since

FIG. 6. Diode laser spectrum obtained after demodulation.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-5

Multi-channel lock-in spectrometer

Rev. Sci. Instrum. 76, 013103 (2005)

and shot noise monotonically increase with the intensity following different dependencies: linear and square root, respectively. For the present setup, the noise data is accurately
fitted with a linear function added to a constant background
(Fig. 7), indicating that the performance of the system is
limited only by the laser noise and the intrinsic noise of the
camera, with the former becoming dominant above 1000
counts/wavelength channel.
III. EXPERIMENTAL RESULTS

FIG. 7. Noise dependence on the laser intensity.

each wavelength channel is demodulated using its own

unique metric, independent of the other channels. Vertical
cuts along the wavelength channels give periodic functions
as expected. The number of periods observed is determined
by tscan f R, where tscan is the vertical scanning time.
The functionality of the lock-in spectrometer has been
tested using a 1310 nm diode laser (Thorlabs Inc., Newton,
NJ) driven close to the threshold current, with the intensity
modulated at 42 kHz. After demodulation of the acquired
signal image, a high-resolution spectrum (8 nm in range)
of the diode output is obtained (Fig. 6), with a spectral resolution identical with that of a scanned diffraction grating
setup. However, the speed at which spectrally resolved data
can be acquired is enhanced by more than two orders of
magnitude.
C. Noise characterization

The system noise is characterized using an optical parametric oscillator (Spectra Physics Opal, Mountain View,
CA), tuned at 1250 nm with a FWHM of 17 nm. In particular, the rms noise was measured as a function of the input
intensity. Typically, three major noise sources have to be
considered: camera noise (itself comprised of thermal noise
and electronic circuit noise), laser noise, and shot noise.13
The different noise sources can be distinguished by measuring their different dependence on the light intensity. While
the camera noise is independent on the intensity, both laser

The implemented system, optimized for 1.3 m wavelengths, was used to study carrier dynamics in two semiconductor systems: dispersed films of PbSe nanocrystals prepared through an organometallic route (Evident
Technologies, Troy, NY) and InAs quantum dot arrays grown
epitaxially on GaAs substrates. Both systems are optically
active in the 1.21.5 m range in which the efficiency of the
lock-in spectrometer peaks.
Time-resolved signals are obtained from the samples using a pumpprobe experiment in which ultrashort pump
pulses (120 fs) create well-defined carrier populations in the
sample, which are subsequently probed using time-delayed
probe pulses. A typical optical setup employed in the experiment (Fig. 8) allows both degenerate pump = probe and nondegenerate pump probe experiments to be performed. A
mechanical optical delay stage with a retroreflector is used to
change the delay between the pump and probe beam.
The time-resolved absorption is measured at T = 70 K
from PbSe nanocrystalline films deposited on fused silica
substrates, shown in Fig. 9(a). The 1.3 m output FWHM
= 17 nm of the optical parametric oscillator was used both
to pump and probe the system. Both the pump and the probe
are intensity modulated using photoelastic modulators at
f 1 = 42 kHz and f 2 = 47 kHz, respectively, and lock-in detection is performed at the difference frequency f R = f 1 f 2
= 5 kHz. For each wavelength channel, the measured signal
corresponds to contributions from only a few nanocrystals
and is affected by the noise sources mentioned above. A
statistical average for an ensemble of nanocrystals is obtained by spectrally averaging the time-resolved signal over
the 320 channels of the lock-in spectrometer [Fig. 9(b)]. The
data collected over a similar time span using an identical
experimental setup and a similar averaging method, but with

FIG. 8. Optical setup for a pumpprobe experiment.

(M) Mirror; (BS) beam splitter; (GLP) GlanLaser polarizer; (OB) 40 objective; (FC) fiber coupler. The
flipper mirror (FM) allows switching from a degenerate
to a nondegenerate experiment.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-6

Fodor, Rothenberger, and Jevy

Rev. Sci. Instrum. 76, 013103 (2005)

FIG. 9. (a) Time-resolved absorption on PbSe films obtained by precipitation of colloidal nanocrystals; (b) averaged data over the 320 channels of the lockin
spectrometer; and (c) data averaged over all the wavelengths using a InAs detector.

a single InGaAs detector and conventional lock-in [Fig. 9(c)]

shows a deteriorated signal-to-noise ratio, arising from the
less efficient data collection and the lack of demodulation
capabilities along individual wavelength channels. The oscillations observed around zero delay for the lock-in spectrometer data arise due to the energy-time uncertainty relation and correspond to the 50 eV spectral resolution of the
spectrometer.
Time-resolved absorption experiments are also performed on self-assembled InAs quantum dots systems grown
GaAs. The quantum dots are probed using a nondegenerate
setup, with the pump tuned to 775 nm and the probe tuned to
1240 nm. Carriers are created by the pump pulse in the
GaAs, and relax quickly into the wetting layer and the quantum dots, where they eventually reach the ground state. Because the pumpprobe experiment is nondegenerate, interference and scattering from the pump laser is eliminated, and it
is possible to modulate the pump beam (at f R = 47 kHz ) in
order to provide the reference for phase-sensitive detection.

FIG. 10. Pumpprobe time-resolved absorption on InAs quantum dots.

Data acquired in parallel on 320 channels covering a wavelength range
of 8 nm.

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48

013103-7

Rev. Sci. Instrum. 76, 013103 (2005)

Multi-channel lock-in spectrometer

Time-resolved absorption reveals long exciton decay times

1 ns along all wavelength channels (Fig. 10). The overall
magnitude of the time-resolved signal is wavelength dependent, being determined by the number of quantum dots that
couple with the probe field at each wavelength. For the systems studied the quantum dot density is about 5
1010 cm2, meaning that more than 200 dots are in the field
of view of the in-cryostat 40 objective. Consequently, more
than one quantum dot contributes to each wavelength channel and the envelope of the signal is uniform rather than
discrete.
We have presented the theory and implementation of a
multiple-channel lock-in optical spectrometer, capable of acquiring spectrally resolved data simultaneously over 320
wavelength channels, with a signal to noise ratio limited only
by the laser noise, for intensity levels sufficient to overcome
the intrinsic noise of the camera detector employed. The system implemented was optimized for the 1.3 m wavelength
range, although with a proper choice of acousto-optic modulator the performance of the system can be optimized for
other wavelength ranges.

ACKNOWLEDGMENTS

The authors are grateful to Gilberto Medeiros-Ribeiro

for providing the InAs quantum dots samples, and to
Yuichiro Kato for a careful reading of the manuscript. This
work was supported by DARPA (DAAD 19-01-1-0650).
1

McPherson Inc. http://www.mcphersoninc.com/systems/mfi mfi.htm

A. Feofanov et al., Rev. Sci. Instrum. 66, 3146 (1995).
3
W. E. Moerner and D. P. Fromm, Rev. Sci. Instrum. 74, 3597 (2003).
4
M. J. Snelling et al., Phys. Rev. B 44, 11345 (1991).
5
J. J. Baumberg, S. A. Crooker, and D. D. Awschalom, Phys. Rev. B 50,
7689 (1994).
6
M. L. Meade, Lock-In Amplifiers: Principles and Applications (Peter Peregrinus, 1983).
7
R. H. Dicke, Rev. Sci. Instrum. 17, 268 (1946).
8
O. Breitenstein et al., Rev. Sci. Instrum. 71, 4155 (2000).
9
Z. He, N. Mukohzaka, and K. Hotate, IEEE Photonics Technol. Lett. 9,
514 (1997).
10
T. Kawasaki et al., J. Phys. D 27, 1646 (1994).
11
P. A. Probst and A. Jaquier, Rev. Sci. Instrum. 65, 747 (1994).
12
Signal Recovery, Inc.
13
D. Marcuse, Principles of Quantum Electronics (Academic, New York,
1980).
2

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitationnew.aip.org/termsconditions. Downloaded to IP:
150.212.39.64 On: Fri, 03 Apr 2015 20:45:48