Thermal Joys
Jonah Shoemaker1
1 Arizona State University, Tempe, AZ, USA, email [email protected]
October 18, 2018
In this study, three researchers with searing hot minds and smoke rising from their ears
explored the effects of noise on voltage measurements. Shot noise and flicker noise were
observed in power spectral densities obtained from several combinations of frequency band-
width and resistance. The Boltzmann constant was obtained to be kexp = (1.3 ± 0.02) ∗ 10−23
J/K from taking several measurements of voltage across a resistor for several different resis-
tances.
1
Introduction
Noise is inherent to all fields of electronic instrumentation and measurements. Most noise
can be explained by statistical fluctuations in current across resistors. In this study, various
types of noise, including white noise, shot noise, and flicker noise, were studied by measuring
power spectral densities for several frequency bandwidth and resistance combinations, and
by measuring voltage fluctuations for various resistances.
Background and Theory
There are three types of noise considered in this lab project: shot noise, Johnson-Nyquist or
white noise, and flicker or 1/f noise. Shot noise is caused by random fluctuations of current
through the resistor. These fluctuations are a direct consequence of the discrete nature of the
electron. For example, if an average of 1018 electrons move through the resistor each second
(a current on the order of 1 ampere), our intuition alone is enough to tell us that, obviously,
this does not necessarily imply that 1018 electrons are moving across the resistor each second
– maybe only 1015 electrons move across over one second, while the next, 1021 move across
instead. We describe these fluctuations with the term "shot noise", named for our dear Wal-
ter Hans Schottky. It is clear that even large fluctuations will not significantly impact our
measured current when it is on the order of something like 1018 , but in the case of extremely
small resistances, the current becomes small enough that these fluctuations become signifi-
cant. Over large time intervals, these fluctuations would still be expected to largely cancel one
another, but over shorter time intervals, we should see a significant contribution from shot
noise. Since time and frequency are Fourier conjugates, shorter time intervals is equivalent
to broader frequency intervals, so we should expect to see significant shot noise contribu-
tions at low resistances and high frequency bandwidths. From the lab document, we have
the expression for the current fluctuations, given by
δi2 = 2i0 e∆f
As we can see from the expression above, as the frequency bandwidth ∆f increases, the
current fluctuations δi2 will also increase.
Johnson-Nyquist noise, or white noise, is caused by the random movement of electrons across
a resistor. Unlike shot noise, no current is needed to observe this phenomenon. The voltage
fluctuations associated with white noise are given by the expression
hVR2 i
= 4T ∆f k
R
Here, T is the temperature in Kelvin, ∆f is the frequency bandwidth, and k is the Boltzmann
constant. Once again, we can see that broader bandwidths will give higher fluctuations across
the resistor, since these broader bandwidths correspond to shorter time intervals, over which
we would expect to see more significant fluctuations. In addition, note that white noise is
temperature- and resistor-dependent. If the resistor is heated, we should expect to see the
2
electrons "loosen up", so to speak, and fluctuate more easily. Similarly, if the resistance is
increased, we should expect to see less of the electrons managing to skip across the resistor,
and so we should expect less white noise for higher resistances.
Flicker noise, or 1/f noise, is similar to shot noise in that it usually results from the flow
of current across a resistor. However, flicker noise is instead thought to be caused by ran-
dom temperature fluctuations in the resistor that cause its resistance to randomly fluctuate.
These fluctuations then go on to cause voltage fluctuations that we can then pick up on our
instruments. Since these temperature fluctuations within the resistor grow with the amount
of current through the resistor, we expect to see flicker noise predominate at high currents
(and therefore high resistances).
Experimental Procedure
The basic experimental procedures were followed as outlined in the lab description, except
that the researchers made sure to have a hell of a good time.
3
Results
Figure 1: For 30 Hz to 3.3 kHz with R = 1 Ohm, log A = (−3.2±0.02) dB, B = 2±0.7 Hz, and
C = (6.8 ± 0.09) ∗ 103 Hz. For 30 Hz to 3.3 kHz with R = 100 kOhms, log A = (−2.7 ± 0.01)
dB, B = 4 ± 0.4 Hz, and C = (3.8 ± 0.02) ∗ 103 Hz. For 100 Hz to 10 kHz with R = 1 Ohm,
log A = (−3.17 ± 0.004) dB, B = 5 ± 0.3 Hz, and C = (1.10 ± 0.003) ∗ 104 Hz. Lastly, for
100 Hz to 10 kHz with R = 1 MOhm, log A = (−2.9 ± 0.01) dB, B = (5 ± 0.3) ∗ 101 Hz, and
C = (5.2 ± 0.04) ∗ 103 Hz. A represents the overall amplitude of the PSD. B represents the
low-pass frequency, which is the frequency at which the signal drops 3 dB below its plateau
value on the left side of the plateau. C represents the high-pass frequency, which corresponds
to the same drop in dB, but on the right side of the plateau instead.
4
Figure 2: For R = 1 Ohm, log A = (−4.259 ± 0.002) dB, B = 1.3 ± 0.01 Hz, and C =
(1.90 ± 0.007) ∗ 105 Hz. For R = 100 Ohm, log A = (−4.248 ± 0.002) dB, B = 1.3 ± 0.01
Hz, and C = (1.92 ± 0.007) ∗ 105 Hz. For R = 100 kOhms, log A = (−4.018 ± 0.004) dB,
B = 1.8 ± 0.03 Hz, and C = (1.11 ± 0.002) ∗ 105 Hz. Lastly, for R = 1 MOhm, the parameters
obtained were log A = (−3.85±0.004) dB, B = 2.0±0.28 Hz, and C = (4.8±0.02)∗104 Hz.
The increase in noise for decreasing frequency for the low resistances observed in the graphs
is likely caused by either shot noise or flicker noise – this rise in noise could be explained by
the random oscillations in the current taking place over the small resistance (ie shot noise),
or by random temperature fluctuations that cause more of a drastic impact on the voltage
measured across the resistor. The decrease in noise for increasing frequency for the large
resistors is likely due to the decrease in white noise – since the voltage across the resistor is
divided by the resistance, at large resistances, this white noise contribution possibly becomes
overwhelmed by the amplifier noise.
5
Figure 3: The fit parameters obtained for 100Hz-10kHz were k1 = (1.407 ± 0.0003) ∗ 10−23
J/K with hVA2 i0 = 1 ± 0.1) ∗ 10−20 JΩ/K. The parameters for 10Hz-100kHz were k2 = (1.1 ±
0.02) ∗ 10−23 J/K with hVA2 i0 = (6 ± 0.7) ∗ 10−20 JΩ/K. The parameters for 30Hz-3.3Hz were
k3 = 1.426 ± 0.0009) ∗ 10−23 J/K with hVa2 i0 = (5.2 ± 0.04) ∗ 10−20 JΩ/K. The best estimate for
Boltzmann’s constant obtained from this experiment is the average of the three fit results,
kexp = (1.3 ± 0.02) ∗ 10−23 J/K. To obtain better fits for the data, the measurements obtained
for R = 1 MOhm were neglected, since the signal tends to drop off at high frequencies for
high resistances, as can be seen from the earlier figures. Since white noise, the predominant
noise observed in the data, is due to random fluctuations of electrons across the resistor, as
the resistance increases, we should expect to see much less of these fluctuations.
6
Discussion
The power spectral densities obtained with different bandwidths and resistances were fitted,
along with the frequencies corresponding to the densities, to the following function:
4
A ∗ Bf
S(dB) = log 4
(1 + Bf ) ∗ (1 + f 4
C
)
B represents the low pass frequency, C represents the high pass frequency, and A represents
the amplitude of the power spectral density function. For the sake of fitting more easily, the
base-10 log of A was fitted as a parameter instead. The curve fit function from SciPy was
used to fit the data to this function. Since there were about 105 measurements per power
spectral density obtained, attempting to fit the data directly using curve fit ran into memory
issues, so the median of every ten consecutive power spectral densities was used instead,
which reduced the number of measurements down to 104 . A fortuitous consequence of this
method was the elimination of anomalous measurements in the data, which resulted in lower
errors for the fit parameters obtained. The curve fit function from SciPy was not robust
enough to fit functions very closely to all of the data, however, but due to time constraints,
the researchers were unable to attempt to use another curve-fitting utility in order to attempt
to compensate for these poor fits. In addition, although the frequencies were directly fitted
to the function, the log of the frequencies was taken for graphing purposes, as this log of
frequencies is what shows us the physical features that arise due to the various types of noise
present. These features are still present otherwise, but shamefully lurk within the shadows
of the data, unobserved by any casual passerby.
For the measurements of hV 2 i for various resistances, the results were plotted against the
function
2 G2 hVR2 i G2 hVA2 i
hV i = +
10 10
This function can be expressed using the formula for white noise as
10hV 2 i hVA2 i
2
= kR + 2
= kR + hVA2 i0
4G T ∆f 4G T ∆f
When fitting to this function, it was necessary to exclude the measurements taken for R = 1
MOhm, since this tends to be lower than expected due to the drop in white noise observed
at high frequencies for large resistances.
Conclusion
Although the curve fits using the SciPy module were unable to fit perfectly to the data, the
parameters obtained were still somewhat reasonable, and the data was accurate enough to
observe features due to the various types of noise. In addition, the data obtained from mea-
suring hV 2 i against varying resistances yielded a very close value to the expected value for the
7
Boltzmann constant, reliably confirming the formula for white noise. The next steps would
be to try to find some way to compensate for the drop in PSD at increasing frequencies for
large resistors, and to find some way to model the rise in PSD at decreasing frequency for
smaller resistors. Hopefully this research project exposition has not been so dense as to fall
upon the reader’s ears like a bunch of white noise. It is possible that the researcher’s ability
to explain the phenomena observed was overwhelmed by the explanation noise caused by
the researcher’s lack of complete understanding.
1 References
Ralph Chamberlin, Office Sessions, ASU
Melissinos, Experiments in Modern Physics, 2nd edition
Wikipedia
