www.thinkSRS.

com

Signal Enhancement
Application Note #6

Many experimental techniques rely on the quantitative Noise Sources

measurement of charged particles. Applications range from
the very simple, such as the measurement of a DC anode An understanding of noise sources in a measurement is critical
current by a picoammeter, to the complex, such as the to achieving signal-to-noise performance near theoretical
measurement of atomic state lifetimes using time-resolved limits. The quality of a measurement may be substantially
particle counting. degraded by a trivial error. For example, a poor choice of
termination resistance for an electron multiplier may increase
Often, the signal of interest is obscured by noise. The noise current noise by several orders of magnitude.
may be fundamental to the process: discrete charges are
governed by Poisson statistics which gives rise to shot noise. Shot Noise
Or, the noise may be from more mundane sources, such as
microphonics, thermal emf, or inductive pick-up. Light and electrical charge are quantized, and so the number
of photons or electrons which pass a point during a period of
This article will describe methods for making useful time are subject to statistical fluctuations. If the signal mean is
measurements of weak signals, even in the presence of large M photons, the standard deviation (noise) will be √M, hence
interfering sources, emphasizing the electronic aspects of the the S/N=M/√M=√M. The mean (M) may be increased if the
measurement. rate is higher or the integration time is longer. Short
integration times, or small signal levels, will yield poor S/N
The figure below details the elements of a typical values. The graph below shows the S/N which may be
measurement. In this experiment the photo-ionization cross expected as a function of current level and integration time for
section for a gas will be determined by passing a laser through a shot-noise limited signal.
the gas and by measuring the number of ions which are created.

Current Quanta (No./s)

Background

Signal 1A 18
Optics 1uS 10
Signal PMT Amp Detection 1 ns
1 mA
1 mS 1S

Bias
1 uA
10 3 S
1 nA 9
10 6 S 10
1 pA
Modulation
1 fA
10
1 aA
Prototye experiment 2 3 4 5
0.1 1 10 10 10 10 10 10 10 10 10
6 7 8 9

S/N
In many applications, the output of the detector must be
amplified, or converted from a current to a voltage, before the
signal may be analyzed. Selection criteria for amplifiers Signal-to-noise vs. flux and time
include type (voltage or transconductance), gain, bandwidth,
and noise. Often, the amplifier noise will be the limiting factor "Integration Time" is a convenient parameter when using time
in determining the S/N ratio in a measurementespecially in domain signal recovery techniques. "Bandwidth" is a better
situations where the charge detector has no gain. choice when using frequency domain techniques. The rms
noise current in the bandwidth ∆f Hz due to a "constant"
Signal Analysis current, I amps, is given by:

There are two broad categories of signal analysis: modulated Ishot noise = √(2qI∆f) where q = 1.6 × 10−19 C
and non-modulated sources. Modulating the source allows the
signal to be distinguished from the background. Often, source Johnson Noise
modulation is inherent to the measurement. For example,
when a pulsed laser is used to induce a current, the signal of The electrons which allow current conduction in a resistor are
interest is present only after the laser fires. Other times, the subject to random motion which increases with temperature.
modulation is "arranged", as when a CW source is chopped. This fluctuation of electron density will generate a noise
Sometimes the source cannot be modulated, or the source is so voltage at the terminals of the resistor. The rms value of this
dominant over the background as to make modulation noise voltage for a resistor of R ohms, at a temperature of T
unnecessary.

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com
 Signal Enhancement

degrees Kelvin, in a bandwidth of ∆f Hz is given by:

VJohnson,rms = √(4KTR∆f)

where K is Boltzman's constant. The noise voltage in a 1 Hz

Noise
bandwidth is given by: Power
Line VLF
AM

FM/TV
CRTs
RADAR
VJohnson,rms(per √Hz) = 0.13nV × √(R(ohms)) Year Day

Since the Johnson noise voltage increases with resistance, -8 -6 -4 -2 2 4 6 8 10

large value series resistors should be avoided in voltage 10 10 10 10 1 10 10 10 10 10

amplifiers. For example, a 1 kΩ resistor has a Johnson voltage Frequency (Hz)

of about 4.1 nV/√Hz. If detected with a 100 MHz bandwidth,

the resistor will show a noise of 41 µVrms, which has a peak- Simplified noise spectrum
to-peak value of about 200 µV.
The key features in this noise spectra are frequencies worth
When a resistor is used to terminate a current source, or as a avoiding: diurnal drifts (often seen via input offset drifts with
feedback element in a current-to-voltage converter, it will temperature), low-frequency (1/f) noise, power line
contribute a noise current equal to the Johnson noise voltage frequencies and their harmonics, switching power supply and
divided by the resistance. Here, the noise current in a 1 Hz CRT display frequencies, commercial broadcast stations (AM,
bandwidth is given by: FM, VHF and UHF TV), special services (cellular telephones,
pagers, etc.), microwave ovens and communications, to
Ijohnson rms (per √Hz) = 130 pA / √R (ohms) RADAR and beyond.

Since the Johnson noise current increases as R decreases, The best alternatives for avoiding these noise sources are:
small value resistors should be avoided when terminating
current sources. Unfortunately, small terminating resistors are 1) Shield to reduce pick-up.
required to maintain a wide frequency response. If a 1 kΩ 2) Use differential inputs to reject common mode noise.
resistor is used to terminate a current source, the resistor will 3) Bandwidth limit the amplifier to match the expected signal
contribute a noise current of about 4.1 pA/√Hz, which is about bandwidth.
1000× worse than the noise current of an ordinary FET input 4) Choose a quiet frequency for signal modulation when using
operational amplifier. a frequency domain detection technique.
5) Trigger synchronously with interfering source when using a
1/f Noise time domain detection technique.

The voltage across a resistor carrying a constant current will Common ways for extraneous signals to interfere with a
fluctuate because the resistance of the material used in the measurement are illustrated below. Noise may be injected via
resistor varies. The magnitude of the resistance fluctuation stray capacitance (figure (a)). The stray capacitance has an
depends on the material usedcarbon composition resistors impedance of 1/jωC. Substantial currents may be injected into
are the worst, metal film resistors are better, and wire wound low-impedance systems (such as transconductance inputs), or
resistors provide the lowest 1/f noise. The rms value of this large voltages may appear at the input to high-impedance
noise source for a resistance of R ohms, at a frequency of f Hz, systems.
in a bandwidth of ∆f Hz is given by:
C Stray B(t)

V1/f,rms = IR × √(A∆f/f)
VN Det Amp Det Amp

where the dimensionless constant A has a value of about 10 to

11 for carbon. In a measurement in which the signal is the
voltage across the resistor (IR), then the S/N = 3 × 105 √(f/∆f).
(a) (b)

Often, this noise source is a troublesome source of low- Det

frequency noise in voltage amplifiers. Amp

Det
R
R Amp
Cable
I(t) Ground
Non-Essential Noise Sources Plane

(c) (d)
There are many discrete noise sources which must be avoided
in order to make reliable low-level light measurements. The Permanent Sn Cu
Magnet
following graph shows a simplified noise spectrum on log-log Det
Amp ∆T Amp
Det
scales. S N

Sn Cu

(e) (f)

Coupling of noise sources

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com
 Signal Enhancement

Inductive pick-up is illustrated in figure (b). The current the PMT anode signal. Unfortunately, the small termination
circulating in the loop on the left will produce a magnetic resistance and wide bandwidth yield lots of current noise.
field, which in turn induces an emf in the loop on the right.
Inductive noise pick-up may be reduced by minimizing the It is important to choose an amplifier with a very high input
areas of the two loops (by using twisted pairs, for example), impedance and low input bias current when amplifying a
increasing the distance between the two loops, or by shielding. signal from a source with a large equivalent resistance.
Small skin depths at high frequencies allow non-magnetic Commercial amplifiers designed for such applications
metals to be effective shields. However, high-mu materials typically have a 100 MΩ input impedance. This large input
must be used to shield from low-frequency magnetic fields. impedance will minimize attenuation of the input signal and
reduce the Johnson noise current drawn through the source
Resistive coupling, or a "ground loop", is shown in figure (c). resistance, which can be an important noise source. Field
Here, the detector senses the output of the experiment, plus Effect Transistors (FETs) are used in these amplifiers to
the IR voltage drop from another circuit which passes current reduce the input bias current to the amplifiers. Shot noise on
through the same ground plane. Cures for ground-loop pickup the input bias current can be an important noise component,
include: grounding everything to the same point, using a and temperature drift of the input bias current is a source of
heavier ground plane, providing separate ground return paths drift in DC measurements.
for large interfering currents, and using a differential
connection between the signal source and amplifier. The bandwidth of a high input-impedance amplifier is often
determined by the RC time constant of the source, cable, and
Mechanical vibrations can create electrical signals termination resistance. For example, a PMT with 1 meter of
(microphonics) as shown in figure (d). Here, a coaxial cable is RG-58 coax (about 100 pF) terminated into a 1 MΩ resistor,
charged by a battery through a large resistance. The voltage on will have a bandwidth of about 1600 Hz. A smaller resistance
the cable is V=Q/C. Any deformation of the cable will would improve the bandwidth but increase the Johnson noise
modulate the cable's capacitance. If the period of the vibration current.
which causes the deformation is short compared to the RC
time constant, the stored charge (Q) on the cable will remain Bipolar transistors offer an input noise voltage which may be
constant. In this case, a 1 ppm modulation of the cable several times smaller than the FET inputs of high input-
capacitance will generate an AC signal with an amplitude of impedance amplifiers, as low as 1 nV/√Hz. Bipolar transistors
1 ppm of the DC bias on the cable, which may be larger than have larger input bias currents, hence larger shot-noise current,
the signal of interest. and so should be used only with low impedance (<1 kΩ)
sources.
The case of magnetic microphonics is illustrated in figure (e).
Here, a DC magnetic field (the Earth's field or the field from When AC signals from very low source impedances are to be
a permanent magnet in a latching relay, for example) induces measured, transformer coupling should be considered. The
an emf in the signal path when the magnetic flux through the transformer is used to step-up the input voltage by its turns-
detection loop is modulated by mechanical motion. ratio. The transformer's secondary is connected to the input of
a bipolar transistor amplifier.
Unwanted thermocouple junctions are an important source of
offset and drift. As shown in figure (f), two thermocouple Conventional bipolar and FET input amplifiers exhibit input
junctions are formed when a signal is connected to an offset drifts on the order of 5 µV/°C. In the case where the
amplifier. For typical interconnect materials (copper, tin), one detector signal is a small DC voltage, such as from a
sees about 10 µV/°C of offset. These extraneous junctions bolometer, this offset drift may be the dominant noise source.
occur throughout instruments and systems; their impact may A different amplifier configuration, chopper stabilized
be eliminated by making AC measurements. amplifiers, essentially measures the input offsets and subtracts
the measured offset from the signal. A similar approach is
Amplifiers used to "auto-zero" the offset on the input to sensitive
voltmeters. Chopper stabilized amplifiers exhibit very low
Several considerations are involved in choosing the correct input offsets with virtually no input offset drift.
amplifier for a particular application. Often, these
considerations are not independent and compromises will be The use of "true-differential" or "instrumentation" amplifiers is
necessary. The best choice for an amplifier depends on the advised to provide common mode rejection to interfering
electrical characteristics of the detector, and on the desired noise, or to overcome the difference in grounds between the
gain, bandwidth, and noise performance of the system. voltage source and the amplifier. This amplifier configuration
amplifies the difference between two inputs, unlike a single-
Charge counting and fast gated integration require amplifiers ended amplifier, which amplifies the difference between the
with wide bandwidth. A 350 MHz bandwidth is required to signal input and the amplifier ground. In high-frequency
preserve a 1 ns rise time. The input impedance to these applications, where good differential amplifiers are not
amplifiers is usually 50 Ω in order to terminate coax cables available or are difficult to use, a balun or common mode
into their characteristic impedance. When PMTs (which are choke may be used to isolate disparate grounds.
current sources) are connected to these amplifiers, the 50 Ω
input impedance serves as the current to voltage converter for

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com
 Signal Enhancement

Transconductance Amplifiers voltage noise, this term can dominate the noise performance
of the design.
When the detector is a current source (or has a large
equivalent resistance) a transconductance amplifier should be 3) Large Rf’s are desired to reduce the Johnson noise current;
considered. Transconductance amplifiers (current-to-voltage however, large Rf’s degrade the bandwidth. If low values of Rf
converters) offer the potential of lower noise and wider are used, the Johnson noise current can dominate the noise
bandwidth than a termination resistor and a voltage amplifier. performance of the design.
However, some care is required in their application.
4) To maintain a flat frequency response, the size of the
A typical transconductance amplifier configuration is shown feedback capacitance must be adjusted to compensate for
below. A FET-input op amp would be used for its low input different source capacitances.
bias current. (Op amps with input bias currents as low as
50 fA are readily available.) The detector is a current source, Io. As many undesirable characteristics of the transconductance
Assuming an ideal op amp, the transconductance gain is amplifier can be traced to the source capacitance, a system
A = Vout/Iin = Rf, and the input impedance of the circuit is Rin may benefit from integrating the amplifier into the detector,
to the op amp's virtual null. (Rin allows negative feedback, thereby eliminating interconnect capacitance. This approach is
which would have been phase shifted and attenuated by the followed in many applicationsfrom microphones to CCD
source capacitance at high frequencies to assure stability.) imagers.

Signal Analysis
Cf
A variety of noise sources are avoided by AC measurement of
the signal. When making DC measurements, the signal must
Rf compete with large low frequency noise sources. However,
Rin
– when the source is modulated, the signal may be measured at
+
Output the modulation frequency, away from these large noise
I Source C Source
sources.
FET
Op Amp
When the source is modulated, one may choose from gated
integrators, boxcar averagers, transient digitizers, lock-in
Typical transconductance amplifier amplifiers, spectrum analyzers, gated photon counters or
multichannel scalers.
Commercial transconductance amplifiers use Rs as large as
10 MΩ, with an Rin which is typically Rf/1000. A low input A measurement of the integral of a signal during a period of
impedance will insure that current from the source will not time can be made with a gated integrator. Commercial devices
accumulate on the input capacitance. allow gates from about 100 ps to several milliseconds. A gated
integrator is typically used in a pulsed laser measurement. The
This widely used configuration has several important device can provide shot-by-shot data which is often recorded
limitations which will degrade its gain, bandwidth and noise by a computer via an A/D converter. The gated integrator is
performance. The overall performance of the circuit depends recommended in situations where the signal has a very low
critically on the source capacitance, including that of the cable duty-cycle, low pulse-repetition rate, and high instantaneous
connecting the source to the amplifier input. Limitations count rates.
include:
The noise bandwidth of the gated integrator depends on the
1) The "virtual null" at the inverting input to the op amp is gate width: short gates will have wide bandwidths, and so will
approximately Rf/Av, where Av is the op amp's open loop gain be noisy. This would suggest that longer gates would be
at the frequency of interest. While op amps have very high preferred. However, the signal of interest may be very short
gain at frequencies below 10 Hz (typically a few million), lived, and using a gate which is much wider than the signal
these devices have gains of only a few hundred at 1 kHz. With will not improve the S/N.
an Rf of 1 GΩ, the virtual null has an impedance of 5 MΩ at
1 kHzhardly a virtual null. If the impedance of the source The gated integrator also behaves as a filter: the output of the
capacitance is less than the input impedance, then most of the gated integrator is proportional to the average of the input
AC input current will go to charging this capacitance, thereby signal during the gate, so frequency components of the input
reducing the gain. signal which have an integral number of cycles during the gate
will average to zero. This characteristic may be used to 'notch
2) The configuration provides high gain for the voltage noise out' specific interfering signals.
at the non-inverting input of the op-amp. At high frequencies,
where the impedance of the source capacitance is small It is often desirable to make gated integration measurements
compared to Rin, the voltage gain for noise at the non- synchronously with an interfering source. (This is the case
inverting input is Rf/Rin (typically about 1000). As FET input with time-domain signal detection techniques, and not the
op amps with very low bias currents tend to have high input case with frequency-domain techniques such as lock-in

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com
 Signal Enhancement

detection.) For example, by locking the pulse repetition rate to When ω1=ω2, there is a DC component (cos(φ)) at the mixer
the power-line frequency (or to any sub-multiple of this output. The output of the mixer is passed through a low pass
frequency), the integral of the line interference during the filter to remove the sum frequency component. The time
short gate will be the same from shot-to-shot, which will constant of the filter is selected to reduce the equivalent noise
appear as a fixed offset at the output of the gated integrator. bandwidth: selecting longer time constants will improve the
S/N at the expense of longer response times.
Trigger
Gate Reset
The simplified block diagram shown is for a single phase
Average lock-in amplifier, which measures the component of the signal
Reset
at a single phase with respect to the reference. A dual phase
C1 lock-in has another channel which measures the component of
Input + R1 the signal at 90 degrees relative to the first channel, which
allows simultaneous measurement of the amplitude and phase
+ R2 C2
– +
–
–
Output of the signal.
Gate Averaging
Digital Signal Processing Techniques
Buffer Integrator Exponential Average & Buffer
Digital signal processing (DSP) techniques are rapidly
Gated integrator and boxcar averager replacing the older analog techniques for the synchronous
detection of the signal. In these instruments, the input signal is
Shot-by-shot data from a gated integrator may be averaged to digitized by a fast, high-resolution A/D converter, and the
improve the S/N. Commercial boxcar averagers provide linear signal's amplitude and phase are determined by high-speed
or exponential averaging. The averaged output from the computations in a digital signal processor. To maintain the
boxcar may be recorded by a computer or used to drive a strip 100 kHz bandwidth of the analog designs, the DSP designs
chart recorder. The figure above shows a gated integrator with must complete a quarter million 16-bit A/D conversions and
an exponential averaging circuit. 20 million multiply-and-accumulate operations each second.
Many of artifacts of the analog designs are eliminated by the
Lock-In Amplifiers DSP approach; for example, the output drift and dynamic
range of the instruments are dramatically improved.
Phase-sensitive, synchronous detection is a powerful
technique for the recovery of small signals which may be Photon Counting
obscured by interference which is much larger than the signal
of interest. In a typical application, a CW laser (which induces Photon counting techniques (which are readily applicable to
the signal of interest) will be modulated by an optical chopper. particle counting measurements) offer several advantages in
The lock-in amplifier is used to measure the amplitude and the measurement of light: very high sensitivity (count rates as
phase of the signal of interest relative to a reference output low as 1 per minute can be a usable signal level), large
from the chopper. dynamic range (signal levels as high as 100 MHz can be
counted allowing a 195 dB dynamic range), discrimination
The figure below shows a simplified block diagram for a lock- against low-level noise (analog noise below the discriminator
in amplifier. The input signal is AC coupled to an amplifier thresholds will not be counted), and ability to operate over
widely varying duty cycles.

Amp Key elements of a photon counting system include: a high

Signal gain PMT (or charge multiplier) operated with sufficient high
Low Pass DC Amp
Filter voltage so that a single photoelectron (or charged particle) will
generate an anode pulse of several millivolts into a 50 Ω load,
Output a fast discriminator to generate logic pulses from anode
signals which exceed a set threshold, and fast gated counters
Reference
PLL to integrate the counts.

In situations where the time evolution of a light signal must be

Lock-in amplifier block diagram measured (LIDAR, lifetime measurements, chemical kinetics,
etc.), transient photon counters allow the entire signal to be
whose output is mixed with (multiplied by) the output of a recorded for each event. In these instruments, the
phase-locked loop, which is locked to the reference input. The discriminated photon pulses are summed into different bins
operation of the mixer may be understood through the depending on their timing with respect to a trigger pulse.
trigonometric identity: Commercial instruments offer 5 ns resolution with zero dead-
time between bins. The time records from many events may
Acos(ω1t + φ) × Bcos(ω2t) = ½AB[cos((ω1 + ω1)t + φ) be summed together in order to improve the S/N.
+ cos((ω1 − ω2)t + φ)]

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com
 Signal Enhancement

Which instrument is best suited for detecting signals from a

photomultiplier tube? The answer is based on many factors
including the signal intensity, the signal's time and frequency
distribution, and the various noise sources and their time
dependence and frequency distribution. In general, the choice
between boxcar averaging (gated integration) and lock-in
detection (phase sensitive detection) is based on the time
behavior of the signal. If the signal is fixed in frequency and
has a 50 % duty cycle, lock-in detection is best suited. This
type of experiment commonly uses an optical chopper to
modulate the signal at some low frequency. Signal photons
occur at random times during the 'open' phase of the chopper.
The lock-in detects the average difference between the signal
during the 'open' phase and the background during the 'closed'
phase.

To use a boxcar averager in the same experiment would

require the use of very long (50 % duty cycle) gates, since the
photons can arrive anywhere during the 'open' phase. Since
the gated integrator is collecting noise during this entire gate,
the signal is easily swamped by the noise. To correct for this,
baseline subtraction can be used: an equal gate is used to
measure the background during the 'closed' phase of the
chopper and subtracted from the 'open' signal. This is then
identical to lock-in detection. However, lock-in amplifiers are
much better suited to thisespecially at low frequencies (long
gates) and low signal intensities.

If the signal is confined to a very short amount of time, then

gated integration is usually the best choice for signal recovery.
A typical experiment might be a pulsed laser excitation where
the signal lasts for only a short time (100 ps to 1 µs) at a
repetition rate up to 10 kHz. The duty cycle of the signal is
much less than 50 %. By using a narrow gate to detect signal
only when it is present, noise which occurs at all other times
is rejected. If a longer gate is used, no more signal is
measured, but the detected noise will increase. Thus, a 50 %
duty cycle gate would not recover the signal well, and lock-in
detection is not suitable.

Photon counting can be used in either the lock-in or the gated

mode. Using a photon counter is usually required at very low
signal-intensities or when the use of a pulse height
discriminator to reject noise results in an improved S/N. If the
evolution of a weak light signal is to be measured, a transient
photon counter or multichannel scaler can greatly reduce the
time required to make a measurement.

Stanford Research Systems phone: (408)744-9040

www.thinkSRS.com