You are on page 1of 9

LOW LIGHT SIGNAL MEASUREMENTS

TREVOR DOLINAJEC

PARTNER: YOUNG PYO HOUNG

Abstract: The power output of a low light source, in particular a red LED, was measured
from across a room in the presence of myriad noise types. The measurement was made pos-
sible by the use of a lock-in amplifier in combination with an optical chopper which turned
the DC signal of the LED into an oscillating signal at a known frequency. The chopper spun
at a frequency of 750Hz and a large time constant was chosen for the Lock-In. This allowed
for the minimization of many types of noise. The experimental measurement made for the
amount of power received from the LED was 0.542nW and the theoretical value was 1.51nW.

Introduction: This type of low light experiment has applications in astronomy. If, for
example, a pulsar can be observed through a telescope then theoretically a photodiode could
be used along with a preamp and a fast Fourier transform spectrum analyzer or a lock-in
amplifier to identify the frequency of pulsation. This process can obviously be difficult due
to the small magnitude of the signal. Knowledge of noise sources and appropriate filtering
can help with this identification. Using a lock-in amplifier with the reference frequency set
on internal can allow for precise identification of the signal frequency through scanning fre-
quencies. This method was used in this experiment to determine the frequency at which he
LED was being chopped by the optical chopper. The method proved successful in identifying
the frequency of the flickering LED at a distance of 5.8m. In this sense the experiment is
a simulation of pulsar investigation and as such a frequency of 750Hz was chosen for the
chopper which is a frequency just above the maximum frequency that pulsars pulsate. Using
the lock-in amplifier to scan for the frequency of interest works best if a relatively low time
constant and slope is used so that the lock-in is responsive (more on this in the Apparatus
and Procedure section). Low time constants and slopes, however, result in mote noise (see
next paragraph) thus once a frequency is determined another procedure is used to determine
the power output of the light source (LED in the experiment’s case) as accurately as possi-
ble. This procedure involves choosing a higher time constant and slope and using the lock-in
interface program on the computer to collect data automatically over the long time spans
necessary for large time constants and over multiple consecutive runs. In this experiment
the experimental power output and the theoretical power output of the LED were only off
by a factor of two which probably says more about the method of approximation used in the
theoretical value than the accuracy of the experiment.
In the exploration of low light signals noise is a very large factor. Relevant forms of noise
1
are f noise, Johnson noise, microphonic noise, 60/120Hz noise, shot noise and instrument
input noise (more on specifics of these noise forms is found in the Theory section). Johnson
noise, shot noise and instrument input noise can and often are measured in units of √AHz

Date: 9/20/09.
1
2 TREVOR DOLINAJEC PARTNER: YOUNG PYO HOUNG

where Hz is the bandwidth ∆f. This allows us to express these noise values independent of
bandwidth. When working with a fixed bandwidth, however, these noise forms can be mea-
sured in Amps. In particular, the photodiode in this experiment sourced 530µA of current
as a result of the LED. The Johnson, shot and preamp instrument input noise were 305 √fHz A
,
876 √fHz
A
and 600 √fHz
A
respectively while the bandwidth concerned was the Equivalent Noise
Bandwidth (ENBW) of the lock-in amplifier. The ENBW had an inverse relationship to
the slope (measured in dB/oct) and time constant of the lock-in amplifier (more on this in
the Apparatus and Procedure section). At the highest slope of 24dB/oct the ENBW was
5/(64T) where T was the time constant. These numbers and calculations are indicative
of the importance of choosing an appropriate time constant on the lock-in amplifier. For
example, if a time constant of 300µs was chosen then these three noise sources alone would
almost completely obscure the LED signal. In this experiment a appropriately large time
constant of 30s was chosen. The lock-in amplifier also had instrument input noise of 6 √nV
Hz
so
the time constant of 30s was also appropriate in regards to that noise. Capacitive coupling
noise of 60/120Hz was avoided by avoiding multiples of these frequencies and thus the noise’s
harmonics while f1 noise was avoided by choosing a rotational frequency of the chopper that
was well out of the low frequency range where f1 noise dominates.

Apparatus and Procedure: This experiment primarily used a SR830 Lock-In Ampli-
fier, SR760 Fast Fourier Transform (FFT) Spectrum Anayzer, SR570 Low-Noise Current
Preamplifier and SR540 Optical Chopper as well as a PIN-10DP Photodiode, MV5753 red
LED, 12V/50W lamp, DS345 Synthesized Function Generator, HP 3465A Multimeter and
a Remote Control Box to run the Optical Chopper the lamp and the LED as well as allow
us to check the voltage across the resistor in the LED circuit from ease of our lab station.
The FFT Spectrum Analyzer uses a fast method of Fourier transform to take signal
data in the time domain and transform it to data in the frequency domain. This method
allows for real-time Fourier transforms but the digital aspect of the FFT analyzer means
that resolution of the transform is span dependent. Connecting the Function Generator to
the FFT Analyzer and setting up a 400Hz sine wave shows that the FFT analyzer reads a
frequency of 398.44Hz when set at a span of 1.56kHz and a frequency of 400.391 at a span
of 780Hz.The FFT does, however, clearly display the harmonics of the 400Hz wave coming
out of the Function Generator which shows that the Function Generator is not perfect and
that the FFT is indeed a very precise instrument.
The Lock-In Amplifier works by using the fact that the product of two sine waves is zero
unless the frequencies are exactly the same. In other words, the Lock-In gives a DC output
that is proportional to the frequency that matches the reference frequency by multiplying all
components of the input signal by the reference signal simultaneously. The Lock-In displays
the input signal in Vrms not Vpk−pk . Although the Lock-In does not exactly display Vrms ;
for example, if a function generators creates a square wave with an amplitude of 1V the
Lock-In does not display 1Vrms but rather 0.9141Vrms . This is because the Lock-In actually
displays the amplitude of the fundamental Fourier component of the reference frequency.
The sensitivity setting on the Lock-In should always be set greater than the Vrms of the
input signal to avoid overload.
A important feature on the Lock-In Amplifier is the time constant setting. By using
the Lock-In Interface Program and a signal from a function generator it can be shown that
LOW LIGHT SIGNAL MEASUREMENTS 3

if the amplitude of the signal is suddenly and drastically changed, say two orders of magni-
tude, that by counting the bins and noting the sample rate the time for the signal amplitude
reading on the Lock-In to settle down can be calculated. This settle down time is directly
related to the slope setting and the time constant setting of the Lock-In. If the slope is
set at 18dB/oct then the wait time for the signal to settle down is 9T where T is the time
constant; 9 seconds for a 1 second time constant for example. The four slope setting on the
Lock-In basically correspond to 1 through 4 pole type filters. The following table from the
user manual of the Lock-In amplifier shows the relationship between slope and wait time.
The table also shows the relationship between slope and Equivalent Noise Bandwidth as
described in the introduction.

The time constant clearly determines the ENBW (more on ENBW in the Theory section).
The Low-Noise Current Preamplifier basically converts the current signal from the pho-
todiode to a voltage signal that then feeds into the The preamp sensitivity setting was used
to maximize the voltage going into the Lock-In amp while not exceeding the operational
guideline of having a source impedance greater than (sensitivity)−1 . Given that the PIN-
10DP photodiode had an impedance of approximately 177.8kΩ this allowed a sensitivity
of 1µA/V although in practice 10µA/V was used so that the Lock-In would not overload.
The Preamp effectively multiplied the incoming current by (sensitivity)−1 . For example, if
the photodiode produced 2.4µA of current then the preamp would produce approximately
2.4mV if the sensitivity was set to 100µA/V.
In the LED measurement experiment the chopper speed was set at 750Hz. This rate
was measured by connecting the frequency out cable from the chopper control to the Lock-In
amplifier and setting the reference frequency to external. The tuning of the frequency was
done using the Remote Control Box. The chopper did not maintain a precise frequency but
the Lock-In adjusted accordingly (the variance was about 1.3Hz). The frequency of 750Hz
was chosen because it was not a multiple of 60/120Hz noise nor was it low enough to be in
the frequency range dominated by f1 noise. A cone was attached around the photodiode to
prevent the Lock-In amplifier from unexpectedly overloading. The basic set up is shown in
the following schematic.
4 TREVOR DOLINAJEC PARTNER: YOUNG PYO HOUNG

On three separate nights three runs were made. On all three runs the Low-Noise Pream-
plifier was set to a sensitivity of 10µA/V and the Lock-In Amplifier sensitivity was set to
50µV. Also a 30s time constant and slope of 25dB/oct was used on all three runs. The only
setting that was changed from the first to the second and third runs was the Gain Mode on
the preamp was changed from High BW to Low Noise (the results of this change our outline
in Results and Conclusions). The Filter Type was set to none on the preamp for all three
runs. Each run consisted of seven runs as set in the Lock-In Interface Program.

Theory and Supporting Data: One of the predominate theoretical concepts in this lab
was the idea of maximizing the time constant so as to minimize noise that was dependent
on bandwidth. When considering time constants in general, however, there is more at play
than simply minimizing noise. For example, if a signals amplitude was varying with time a
large time constant would be detrimental to recovering the signal because it would average
out the signal. The following graph shows that a time constant of 100ms does not keep up
with toggling the amplitude of the signal between 2Vpk−pk and 1Vpk−pk . The sample rate
was 128Hz.

This graph shows data using a slope of 6dB/oct which has the shortest of the wait times
but we clearly see that the Lock-In does not have time to get back to 0.707Vrms or down to
0.354Vrms . A time constant of 10ms allows for much more accurate data in this case. The
amplitude of the triangle wave created by the chopper and the LED, however, was quite
constant so this consideration did not have to be made in that experiment. Choosing a time
constant T such that T−1 is significantly larger than the reference frequency can also create
very inaccurate data.
The theory behind noise is also a major part of this experiment. There is not much to
say about f1 noise that the name doesn’t say. It is only relatively strong at low frequencies
due to its simple mathematical definition and it diverges at zero. Its ubiquitous and there is
no universal theory to explain it. Capacitive coupling noise is arguable more complex. In this
experiment frequencies above kHz were unexplored thus the only relevant capacitive coupling
noise is that of 60/120Hz and harmonics. It can be shown (see Calculations (i)) that the
minimum stray capacitance needed to induce a 1mVrms voltage across a 1MΩ resistor from a
120Vrms power line with f =60Hz is 0.02pF. This is small necessary capacitance to produce
a relatively high voltage, especially when one considers that one foot of RG/58 coaxial cable
has 33pF of capacitance. Perhaps this explains the large magnitudes of 60/120Hz noise in
many experiments. The following graph shows the spectrum of a unshielded cable attached to
LOW LIGHT SIGNAL MEASUREMENTS 5

the FFT analyzer and laying on the table. We can see f1 noise to the far left and a large peak
at 60Hz. We can also see many harmonics of 60/120Hz noise. Some are larger than others,
such as 180Hz which is second only to the fundamental. The harmonics gradually decrease in
magnitude with higher frequencies until they are essentially lost among miscellaneous noise.

The harmonics on the fundamental of 60Hz noise exist because 60Hz noise is not perfectly
60Hz and the Fourier transform exposes its components. It is unfortunate that the vertical
scale of this graph is dBV so for clarity the 60Hz peak has an amplitude of 14mVrms . When
a much shorter cable was used the amplitude of the 60Hz peak fell drastically to 125µVrms .
When the full length cable is hanging rather than laying the 60Hz peak is at 580µVrms .
When the cable is coiled inside of the shielding tube the peak drops to 395µVrms and when
that shielding is grounded the peak drops further to 180µVrms . The following graph shows
the grounded and shielded case.

The horizontal range is much shorter on this graph than the last but we can make out the
attenuation of our the noise peaks. The peak at 75Hz was from the computer monitor and
would completely disappear when the monitor was turned off. Interestingly, this graph does
not represent the greatest attenuation of the 60Hz peak. When the cable was coiled but
not in the shielding the 60Hz peak was the lowest of all at 90µVrms . This suggests that the
shielding was acting more as an antenna than shielding. The theory behind coiling is that
when coiled the cable has inductance across the output and thus acts as a High-Pass filter.
Microphonic noise can be caused by shaking in a cable that either has voltage through
it or a cable hanging. A hanging cable when shaken produces low frequency ”humps” that
can approach the magnitude of the 60Hz noise peak. The ”humps” are only noticeable up to
about 60Hz. If shaking of a cable is estimated to cause a certain change in capacitance than
6 TREVOR DOLINAJEC PARTNER: YOUNG PYO HOUNG

microphonic noise in a cable with voltage across it can be estimated quantitatively. If, for
example one is to shake one meter of RG/58 coaxial cable at a frequency of 10Hz with 1V
across it and 33pF/ft of capacitance leading into a scope with 1MΩ impedance, then given
an estimated change of capacitance of 1pF there would result 10µV of microphonic noise
(see Calculations (ii)). Microphonic noise can be minimized by having shorter cables that
are less likely to vibrate or shake or even by having stiffer wires.
Shot noise is yet another form of ambient noise and the first of the three bandwidth
dependent noises mentioned in this section. Like Johnson noise and Instrument noise shot
noise can be measured in √AHz . Shot noise occurs when current flows through electronic
devises such as diodes. Shot noise is basically given by the standard deviation of the√ current
over a given bandwidth. More precisely the formula for shot noise is Inoise (rms)= 2qI∆f .
Where q is the electron charge of 1.6×10−19 Coulomb and ∆f is the ENBW. For example,
the photodiode used in this experiment had a measured current during standard operation
of approximately 2.4µA thus the shot noise of the photodiode at a given frequency would be
8.76 × 10−13 √AHz . As mentioned before, the larger the time constant the smaller the ENBW,
i.e the smaller the ∆f and hence the smaller amount of noise.
Johnson noise is similar to shot noise in many ways, but instead of being associated with
the current flow it is associated with thermal activity. Johnson noise is√the root mean square
of the voltage across a resistor. Specifically, Johnson is given by V= 4kT ∆f R where k is
the Boltzmann qconstant and T is the temperature. Johnson nose can also be described as a
4kT ∆f √V √A . The PIN-
current as I= R
. Thus Johnson noise can be given in units of Hz
or Hz
10DP photodiode had a resistance of 177.8lΩ so its Johnson noise was 54.2 √nV
Hz
or 305 √fHz
A
.
If we were looking at a current source we would want high impedance according to the above
equation and if we were looking at a voltage source we would want low impedance according
to the appropriate equation above. In the case of our photodiode we are looking at it as a
current source. The reason that we look at the photodiode as a current source and not a
voltage source is because the PN junction behavior of the photodiode is such that voltage as
a function of incoming power is nonlinear whereas current as a function of incoming power
is linear.
In order to experimentally confirm the theoretical values for Johnson noise discussed
above an experiment was performed with 10K and 100K resistors attached to the Lock-In
amplifier via a short stiff cable used in order to minimize microphonic noise. The Lock-In
was set to low noise with a slope of 24dB/oct. The Lock-In Interface Program was used and
the number of runs was set at seven and the BEST CHOICE sampling mode was chosen.
The internal reference frequency on the Lock-In was set to 500Hz, chosen because its not
so low as to be dominated by f1 noise or possible microphonic noise and because it is not
a multiple of 60/120Hz noise. With these settings one run each was conducted using time
constant of 30ms, 100ms, 300ms, 1s and 3s for each resistor. The analyzed data outputted
by the Interface Program contained a column of seven values (one for each run) that repre-
sented the RMS of the voltages in the first column. This value, of course, represented the
Johnson noise. For each time constant for each√resistor these seven values were averaged.
This averaged value was generally off by about 2 from the √ theoretical value. The reason
that the experimental value was seemingly off by a factor of 2 was because the Interface
program when calculating the values in the < dR2 >−1/2 column took the RMS of Johnson
noise which is essentially already an RMS value. Thus, in the following comparison of the
LOW LIGHT SIGNAL MEASUREMENTS 7

two vales, the experimental numbers are always multiplied by a factor of 2 to match them
to the theoretical numbers. The Lock-In amplifier also had instrumental noise of its own so in
order to compare the theoretical Johnson noise to the experimentalr Johnson noise of the re-
q
2 5
sistors the following concept is helpful: (Total Theoretical Noise) = (VJ,theo ) + 6nV 64T .
The following two graphs show all the experimental values of noise compared with the total
theoretical noise values.

We can see that to about a factor of two the experimental and theoretical values agree.
However, the results appear to be more inaccurate for the 10k then for the 100k for some
reason. Also, the 3s time constant produces odd results experimentally for both the 10k and
100k resistors. This may be do to some sort of ambient behavior at 3s intervals that is being
picked by the cable holding the resistor as if it was an antenna.
The last significant theoretical concept in this experiment is the approximation methods
used to determine theoretically the amount of power emitted by the LED which would be
received by the PIN-10DP photodiode. Firstly, the circuit containing the LED was analyzed.
This circuit consisted of a 5V power supply, a 150Ω resistor and the LED itself. The voltage
across the resistor was measured at 1.4024V and thus the current was calculated at 9.34mA.
Since the voltage across the LED was 5V-Vresistor the power output of the LED can be cal-
culated as WLED =VI=33.6mW. This number, however, represents the total power output
not that which is discernible as visible light. To calculate the power output of the LED as
red light the standard eyeball curve along with the specifications of the LED were used. The
LED according to its specifications sheet has a dominant wavelength of 635nm so using the
curve this corresponds to 175lumens/watt. The LED has a millicandela rating of 15mcd. To
convert these candela to lumens we multiplied by 4π. Then to convert lumens to watts we
used the curve and divided by 175 to get a power output of 1.07mW in terms of red light.
This value is appropriately small compared to the 33.6mW total power output. The portion
of this
Z power that hits the photodiode across the room is given by the integral
P= I(~r) dA.
photodiode0 ssurf ace
cos2 (θ)
I(~r) can be approximated as C .
r2
3P
Completing the integral gives us C= .

This integral works out this way because we are only interested in the hemisphere with the
photodiode o the z axis. Using the power we calculated using the standard eyeball curve
8 TREVOR DOLINAJEC PARTNER: YOUNG PYO HOUNG

we get a value for C of 511µW. Using a small angle approximation and the location of the
photodiode on the z axis we can estimate the Pphotodiode as I(θ = 0, r)×Area of Photodiode.
The specification of the photodiode tells us that the area of the diode is 1cm2 . Thus the
power received by the photodiode from the LED can be estimated at 1.51nW. This is the
so-called theoretical value that is compare with the measured value.

Results and Conclusions: While conducting the experiment to determine the power
output of the LED using the set up described in Apparatus and Procedure we found that if
the preamp was set on High BW then the signal from the LED was indiscernible on the FFT
but when the preamp was set on Low Noise or Low Drift the signal was visible on the FFT.
That is to say, there was a small peak of magnitude 13.2µV that was significantly larger
than surrounding noise at the frequency of approximately 750Hz which was the frequency
the chopper was set at.
Three data runs were done to measuring the power output of the LED. The first was
done using the High BW gain mode. No filter, input offset or bias was used on the Low
Noise Preamplifier. The preamp was set to a sensitivity of 10µA/V as it was it on on all
three runs. On all three runs the Lock-In was set on a slope of 24dB, time constant 30s and
sensitivity of 50µV. The average voltage given by the Lock-In on this first run was 13.1µV.
Using Low Noise gain on the second and third run resulted in average voltages of 13.2µV
and 13.3µV. It’s worth noticing that changing from High BW to Low Noise had little effect
on the amplitude of the signal but a large effect on the spectrum as seen on the FFT.
The time constant of 30s was used to minimize Johnson, shot and instrumental noise. A
larger time constant could have been used but the greater time necessary for data collection
may have not been worth the slight loss in noise. Using our theoretical knowledge of noise
we would expect about 15.6pV of instrumental noise from the Lock-In and the Preamp had
approximately 156pV of instrumental noise. Likewise, we would expect about 8.27nV of
Johnson noise and 252pV of shot noise (assuming we can use the impedance of the photodi-
ode to calculate voltage shot noise). This noises sum to 9.18nV of noise whereas the Lock-In
Interface Program reads a RMS of the voltage around 19-28nV. The unexplained noise likely
comes from using the impedance of the photodiode for calculating Johnson and Shot noise
as registered by the Lock-In. Microphonic noise could have been present considering the
long cable leading from the PIN-10DP photodiode to the Preamp but this noise like f1 would
have had a much lower frequency than the 750Hz.
Using the voltage values given by the Lock-In and the theory described above the ex-
perimental power picked up by the photodiode from the LED was 0.754nW as compared to
the theoretical wattage of 1.51nW using the specifications of the LED and the the approx-
imation methods described above. This values are within a factor of two from each other
but this does not necessarily say that the experiment was inaccurate in its measurements.
In fact, since the cos2 (θ) approximation is an over approximation it is to be expected that
the theoretical values are larger than the experimental. What we can conclude is that the
power measurements are quite accurate and that this accuracy was made possible by the
Lock-In amplifier in conjunction with the chopper and that our use of a large time constant
minimized some of the most troublesome noise.
LOW LIGHT SIGNAL MEASUREMENTS 9

1mV
Calculations: (i) VR =IR ⇒ 1mV=I1MΩ ⇒ I= =1nA
aM Ω
q
VT =IZ=I R2 + ( ωC ) ⇒ ( VIT )2 = R2 + ( ωC
1 2 1 2
)
1 q 1 1
= ( VIT )2 − R2 ⇒ C= q = q =0.02pF
ωC ω ( VIT )2 − R2 2π60 ( 120V )2 − 1M Ω2
1nA

˙ = C(t)V
˙ dC(t)
(ii) 1m has 108.27pF capacitance ⇒ I = Q(t) ⇒I= V
dt
dC(t) C τ δC 1M Ω108.27pF 1pF
Thus Vn =IRs =Rs V = V= ≈ 10µV
dt C dt C .1s 108.27pF

You might also like