White Paper

Principles of lock-in detection Zurich

and the state of the art Instruments

Release date: November 2016

Introduction on how signal modulation can be exploited in order to

improve on signal-to-noise ratio (SNR) while keeping
Lock-in amplifiers were invented in the 1930’s [1, 2, 3] acquisition time low. Finally, recent innovations are
and commercialized [4] in the mid 20th century as discussed and the state of the art is described.
electrical instruments capable of extracting signal am-
plitudes and phases in extremely noisy environments
(see Figure 1). They employ a homodyne detection Lock-in amplifier working principle
scheme and low-pass filtering to measure a signal’s
amplitude and phase relative to a periodic reference. Lock-in amplifiers use the knowledge about a signal’s
A lock-in measurement extracts signals in a defined time dependence to extract it from a noisy background.
frequency band around the reference frequency, effi- A lock-in amplifier performs a multiplication of its in-
ciently rejecting all other frequency components. The put with a reference signal, also sometimes called
best instruments on the market today have a dynamic down-mixing or heterodyne/homodyne detection, and
reserve of 120 dB [5], which means they are capable of then applies an adjustable low-pass filter to the re-
accurately measuring a signal in the presence of noise sult. This method is termed demodulation or phase-
up to a million times higher in amplitude than the sig- sensitive detection and isolates the signal at the fre-
nal of interest. quency of interest from all other frequency compo-
Over decades of development, researchers have found nents. The reference signal is either generated by the
many different ways to use lock-in amplifiers. Most lock-in amplifier itself or provided to the lock-in ampli-
prominently they are used as precision AC voltage fier and the experiment by an external source.
and AC phase meters, noise measurement units, The reference signal is usually a sine wave but could
impedance spectroscopes, network analyzers, spec- have other forms, too. Demodulation with a pure sine
trum analyzers and phase detectors in phase-locked wave enables selective measurement at the funda-
loops. The fields of research comprise almost every mental frequency or any of its harmonics. Some instru-
length scale and temperature, such as the observa- ments use a square wave [9] which also captures all
tion of the corona in full sunlight [6], measuring the odd harmonics of the signal and, therefore, potentially
fractional quantum Hall effect [7], or direct imaging of introducing systematic measurement errors.
the bond characteristics between atoms in a molecule To understand lock-in detection, we will look at both
[8]. Lock-in amplifiers are extremely versatile. As es-
sential as spectrum analyzers and oscilloscopes, they
are workhorses in all kinds of laboratory setups, from input signal Vs (t)
lock-in
physics to engineering and life sciences. As with most amplifier
powerful tools, only a solid understanding of the work-
amplitude noise amplitude
ing principles and features enables the user to get the phase
phase
most out of it and to successfully design experiments.

This document provides a quick introduction to the reference signal Vr (t)

principles of lock-in amplification and explains the
most important measurement settings. The lock-in Figure 1. Lock-in amplifiers are capable of measuring the amplitude
detection technique is described both in the time and and the phase of a signal relative to a defined reference signal, even
in the frequency domain. Moreover, details are laid out if the signal is entirely buried in noise.
a a b c
Vs (t) R
DUT lock-in ωs
amplifier
Vr (t) ϴ
sine wave
generator ωs 2ωs
y
b x
mixer LP filter
input signal Vs (t) X
Figure 3. Demodulation process represented in the complex plane.
(a) The input signal Vs (t) can be expressed as the sum of two counter-
oscillator rotating vectors. (b) The projections onto the x-axis add up whereas
R
reference the projections to the imaginary y-axis cancel each other out. (c) In
the rotating frame the counter-clockwise vector is standing still, the
signal Vr (t)
+90° ϴ clockwise moving vector rotates at twice the observer’s angular ve-
locity. Note that by convention, ϴ is positive if the counter-clockwise
vector is ahead of the reference.
Vr (t) Y
mixer LP filter
Signal mixing in the time domain
Figure 2. (a) Sketch of a typical lock-in measurement. A sinu- Complex numbers provide an elegant mathematical
soidal signal drives the DUT and serves as a reference signal. The
response of the DUT is analyzed by the lock-in which outputs the formalism to calculate the demodulation process. We
amplitude and phase of the signal relative to the reference signal. use the elementary trigonometric law
(b) Schematic of the lock-in amplification: the input signal is multi-
plied by the reference signal and a 90◦ phase-shifted version of the 1 +ix 1 −ix
reference signal. The mixer outputs are low-pass filtered to reject cos(x) = e + e (2)
2 2
the noise and the 2ω component, and finally converted into polar
coordinates.
to rewrite the input signal Vs (t) as the sum of two
√ vec-
tors in the complex plane, each one of length R/ 2 ro-
tating at the same speed ωs , one clockwise and the
the time and the frequency domain, first for mixing and
other counter-clockwise:
then for the filtering process.
√
Vs (t) = 2 R · cos(ωs t + ϴ)
Dual-phase demodulation R +i(ωs t+ϴ) R
= √ e + √ e−i(ωs t+ϴ) . (3)
In a typical experiment, the device under test (DUT) 2 2
is stimulated by a sinusoidal signal, as shown in Fig-
ure 2 (a). The device response Vs (t) as well as the refer- In the graphical representation given in Figure 3 (a) and
ence signal Vr (t) are used by the lock-in amplifier to de- (b) one can see that the vectors’ sum projected on the x-
termine the amplitude R and phase ϴ. This is achieved axis – the real part – is exactly Vs (t), whereas the vector
using a so-called dual-phase demodulation circuit, as sum projection onto the y-axis – the imaginary part –
illustrated in Figure 2 (b). The input signal is split and is always zero.
separately multiplied with the reference signal and a The dual-phase down-mixing is mathematically ex-
90◦ phase-shifted copy of it. The outputs of the mix- pressed as a multiplication of the input signal with the
ers pass through configurable low-pass filters, result- complex reference signal
ing in the two outputs X and Y, termed the in-phase √ √ √
and quadrature component. The amplitude R and the Vr (t) = 2 e−iωr t = 2 cos(ωr t) − i 2 sin(ωr t). (4)
phase ϴ are easily derived from X and Y by a transfor-
mation from Cartesian coordinates into polar coordi- The complex signal after mixing is given by
nates using the relation
Z(t) = X(t) + iY(t) = Vs (t) · Vr (t)
√ [ ]
R = X 2 + Y2 , = R ei[(ωs −ωr )t+ϴ] + e−i[(ωs +ωr )t+ϴ] , (5)
ϴ = atan2 (Y, X). (1)
with signal components at the sum and the difference
Note that in order to have an output range for the of the signal frequency and the reference frequency. In
phase angle that covers all four quadrants, i.e. (−π, π], the picture of Figure 3 (c), the complex mixing is equiv-
atan2 is used instead of atan. alent to an observer located at the origin and rotating
Figure 2 (b) shows that the lock-in amplifier has to split in a counter-clockwise direction with frequency ωr .
up the input signal in order to demodulate it with two In the eyes of this observer, the two arrows appear
different phases. Contrary to analog instruments, dig- to rotate at different angular velocities ωs − ωr and
ital technology overcomes any losses in SNR and mis- ωs + ωr , with the arrow ωs + ωr rotating much faster
match between the channels when splitting the signal. if the signal and reference frequencies are close.

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 2
a 1 b 1 ter mixing and filtering as they would appear on an os-
cilloscope. Figure 4 (a) shows the sinusoidal example
0.5 0.5
signals Vs and Vr over time having exactly the same
amplitude (V)

amplitude (V)
0 0
frequencies ωs and ωr . The signal after mixing, blue
trace in Figure 4 (b), is dominated by the 2ω compo-
-0.5 -0.5 nent. After filtering, green trace, only the DC compo-
Vr signal after mixing
Vs signal after filter nent remains, which is equal to the in-phase ampli-
-1 -1
0 1 2 3 0 1 2 3 tude X of Vs . If the signal frequency and the reference
time (s) time (s)
frequency deviate, as shown in Figure 4 (c), the result-
ing signal after mixing is no longer a simple sine wave
c 1 d 1
and averages out to zero after filtering, as shown in
0.5 0.5 Figure 4 (d). It is the perfect example of synchronous
detection, which exclusively extracts signals coherent
Amplitude (V)
amplitude (V)

0 0
with the reference frequency and discards all others.
-0.5 -0.5
Vr
Vs
signal after mixing
signal after filter
Signal mixing in the frequency domain
-1 -1
0 1 2 3 0 1 2 3 To switch between the time domain and the frequency
time (s) time (s)
domain picture, we use the Fourier transform [10]. The
Figure 4. (a) An input signal Vs (red) with peak amplitude of 0.5 V is Fourier transform is linear and converts a sinusoidal
multiplied with the reference signal Vr (blue) at the same frequency. function with frequency f0 in the time domain into a
(b) The resulting signal has a DC offset and a frequency component Dirac delta function δ(f-f0 ) in the frequency domain, i.e.
at twice the frequency of Vs and Vr . The DC value is 0.17 V, which is
the in-phase component X of the input signal. (c) The input signal a single peak at frequency f0 in the spectrum. As any
Vs is multiplied by a reference Vr at a different frequency. (d) The periodic signal can be expressed as a superposition of
resulting signal has frequency components at fs − fr and fs + fr . The sines and cosines [11], transformations of signals con-
average signal is always zero.
sisting of only a few spectral components can often be
intuitively understood.
The subsequent filtering is mathematically expressed Figure 5 (a) shows a noisy sinusoidal represented in
as an averaging of the moving vectors over time, in- the time domain, which is then Fourier transformed
dicated by the angle brackets ⟨· · · ⟩. Filtering strips into the frequency domain in Figure 5 (b). The sinu-
away the fast rotating term at |ωs + ωr | by setting soidal signal shows up as a peak both at +fs and at −fs
⟨exp [−i (ωs + ωr ) t + iϴ]⟩ = 0. The averaged signal af- in the spectrum. The smaller peak at zero frequency is
ter demodulation becomes caused by the input signal’s DC offset. The blue trace
in Figure 5 (c) represents the time domain signal after
Z(t) = R · ei[(ωs −ωr )t+ϴ] . (6) mixing. The associated spectrum shown in Figure 5 (d)
is essentially a copy of the one in (b) shifted by the ref-
In the case of equal frequencies ωs = ωr , this further erence frequency fr towards lower frequencies.
simplifies to Low-pass filtering is indicated as a dashed red trace
in (d) and selects the frequencies up to a certain fil-
Z(t) = R · eiϴ . (7) ter bandwidth fBW . The output signal, red trace in (c),
is the DC component of the spectrum visualized in (d)
Equation 7 is the demodulated signal and the main plus the noise contribution within the filter bandwidth
output of the lock-in amplifier: with the absolute value |f| < fBW . It is evident from this picture that a filter band-
|Z| = R given as the root-mean-square amplitude of the width significantly smaller than the signal frequency fs
signal and its argument arg(Z) = ϴ given by the phase is required to efficiently suppress offsets in the input
of the input signal relative to the reference signal. signal. In the next sections, we’ll discuss further crite-
The real and imaginary parts of the demodulated sig- ria for choosing suitable filter characteristics in a given
nal Z(t) are the in-phase component X and the quadra- experimental situation.
ture component Y. They are obtained using Euler’s for-
mula exp(iωs t) ≡ cos(ωs t) + i sin(ωs t) as:
Low-pass filtering in the frequency domain
X = Re(Z) = ⟨Vs (t) cos (ωs t)⟩ = R cos ϴ, For the low-pass filtering we start by considering the
Y = Im(Z) = − ⟨Vs (t) sin (ωs t)⟩ = R sin ϴ. (8) frequency domain because for most filters there is
a simple relationship between the incoming signal
In the graphical view, ωs = ωr means that the arrow ro- Qin (ω) and the filtered signal Qout (ω) given by
tating counter-clockwise will appear at rest. The other
arrow is rotating clockwise at twice the frequency, i.e.
−2ωs , and is often called the 2ω component. The low- Qout (ω) = H(ω) Qin (ω). (9)
pass filter usually cancels out the 2ω component com-
pletely. H(ω) is called the transfer function of the filter. Qin (ω)
Figure 4 illustrates the different signals before and af- and Qout (ω) are the Fourier transforms of the time do-

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 3
 a 1
b a First-order RC low-pass filter

FFT amplitude (dB, a.u.)

40
amplitude (V)

0.5
DC Qin(ω) Qout(ω)
0 20

–0.5
0
fr
–1
0 0.5 1 1.5 2 –20 –10 0 10 20 b Higher-order RC low-pass filter
time (s) frequency (Hz)
c 1
d Stage 1 Stage 2 ... Stage n
2f BW
FFT amplitude (dB, a.u.)

40
...
amplitude (V)

0.5
Qin(ω) Qout(ω)
0 20

–0.5 –2fr –f r
0
–1
0 1 2 3 4 –20 –10 0 10 20
time (s) frequency (Hz) Figure 6. (a) First-order RC filter and its transfer function for-
mula. (b) Steeper roll-offs towards higher frequencies are achieved
Figure 5. Relationship between time and frequency domain repre- by stacking multiple RC filters. The transfer function results from a
sentation before and after demodulation. (a) Sinusoidal input signal multiplication of each filter’s transfer function.
superimposed with noise displayed over time. (b) Same signal as in
(a) represented in the frequency domain. (c) After mixing with the
reference signal (blue trace) and low-pass filtering (red trace), the
half. The amplitude, proportional to the square root of
signal spectrum up to fBW remains. (d) In the frequency representa-
tion, the frequency-mixing shifts the frequency components by −fr . the power, is reduced by 1/√2 = 0.707 at f−3dB .
The filter then picks out a narrow band of fBW around zero. Note the
component at frequency −fs , which comes from offset and 1/f noise For the filter described by Equation 10, the cut-off fre-
in the input signal. To obtain accurate measurements this compo- quency is f−3dB = 1/(2πτ ). From Figure 7 (b) we see that
nent has to be suppressed by proper filtering. the low-pass filter also introduces a frequency depen-
dent phase delay equal to arg[H(ω)].
Compared to the idealized brick-wall filter, the first-
main input signal Qin (t) and output signal Qout (t) re- order filter has a fairly poor roll-off behavior. To in-
spectively. crease the roll-off steepness it is common to cascade
To perfectly reject unwanted parts of the spectrum, several of these filters. For every filter added the filter
one might think that an ideal filter should have full order is increased by 1. Since the output of one filter
transmission for all frequencies below fBW , i.e. the becomes the input to the following one, we can simply
passband, and zero transmission for all other frequen- multiply their transfer functions. From Equation 9 we
cies, also called the stop band. Unfortunately such thus get the following transfer function of an nth order
idealized “brick-wall filters” are impossible to realize filter:
since their impulse response extends from −∞ to +∞
in time, which makes them non-causal. As a basic ap- ( )n
proximation, we consider the RC filter model, see Fig- 1
Hn (ω) = H1 (ω)n = . (11)
ure 6. This type of filter is easy to implement both in 1 + iωτ
the analog and the digital domain. The transfer func-
tion of an analog RC filter is well approximated by Its attenuation is n times the attenuation of a first-
order filter, with a total roll-off of n × 20 dB/dec. The
frequency responses of a 1st , 2nd , 4th and an 8th order
1 RC filter are shown in Figure 7 (a) and (b). The higher
H(ω) = , (10)
1 + iωτ the filter order, the closer the amplitude transfer func-
tion gets to a brick-wall filter behavior. At the same
where τ = RC is called the filter time constant with
time, the phase delay increases with filter order. For
the resistance R and capacitance C. The blue traces
applications where the phase is used to apply a feed-
in Figure 7 (a) and (b) show this transfer function in
back to a system, for example phased-locked loops,
Bode plots, 20log|H(2πf)| and arg[H(2πf)] as functions
any additional phase delay can limit the stability and
of log(f).
bandwidth of the control loop.
From the blue curve in Figure 7 (a) we can infer that
the attenuation grows ten times every tenfold fre- Figure 8 (a) and (b) show the Bode plots for filters of dif-
quency increase above f−3dB . This equals 6 dB/octave ferent orders with the same bandwidths f−3dB but dif-
(20 dB/decade) corresponding to an amplitude reduc- ferent time constants. Table 1 provides the numerical
tion by a factor of 2 every doubling of the frequency. relationship between corresponding filter properties.
The cut-off frequency f−3dB is defined as the frequency
at which the signal power is reduced by −3 dB or one

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 4
 a 0 a 0
attenuation (dB)

−3 dB −3 dB

attenuation (dB)
−10 −10
−20 −20
−30 −30
−40 n=1 −40 n=1
n=2 n=2
−50 n=4 −50 n=4
−60 n=8 n=8
−60
b 0 b 0
−π/2 −π/2
−1 −1
phase (π)

phase (π)
−2 −2

−3 −3

10 100 1000 0. 1 1 10
frequency (Hz) frequency (f–3dB)

c c
1 1
0.99 0.99
0.8 0.8 1
step response

step response

0.6 0.6
0.995
0.4 0.4

0.2 0.2 1/f–3dB

0.99
0 0

0 2 4 6 8 10 12 14 16 18 0 1 2 3 4 5 6 7 8
time (τ ) time (τ1 )

Figure 7. The blue traces in (a) and (b) show the transfer function Figure 8. Same set of plots as for Figure 7 but this time all filters
H(ω) of an RC filter in the form of a Bode plot. The transfer func- have the same cut-off point f−3dB but different time constants τ =
tions for higher-order filters (n = 2, 4, 8) with the same filter time 0.16, 0.10, 0.069, 0.048. (a) Higher-order filters show a steeper roll-
constant τ are also plotted and clearly have much lower signal band- off towards higher frequencies. (b) Higher-order filters have larger
width f−3dB . (c) Associated step response functions in the time do- phase delays, which can be detrimental for feedback applications.
main. Cascading multiple filters leads to a significant increase in (c) Step response as a function of time in units of the time constant
settling time to achieve the same level of accuracy. This is related to τ 1 of the first-order filter. Though lower-order filters respond more
the larger phase delay that is inferred from (b). One additional nice quickly to changes of the input signal at the beginning, this advan-
feature of the cascaded RC or integrator filter is that it has no over- tage decreases over time and at some point higher-order filters even
shoot in the time domain, which is an issue with Butterworth filter “overtake” lower-order filters, as seen in the inset.
for instance.

Order Time Roll-off Bandwidth in units of 1/τ Settling times in units of τ

n constant τ dB/oct dB/dec f−3dB fNEP fNEP /f−3dB 63.2% 90% 99% 99.9%
1 1 6 20 0.159 0.250 1.57 1.00 2.30 4.61 6.91
2 1 12 40 0.102 0.125 1.23 2.15 3.89 6.64 9.23
3 1 18 60 0.081 0.094 1.16 3.26 5.32 8.41 11.23
4 1 24 80 0.069 0.078 1.13 4.35 6.68 10.05 13.06
5 1 30 100 0.061 0.069 1.12 5.43 7.99 11.60 14.79
6 1 36 120 0.056 0.062 1.11 6.51 9.27 13.11 16.45
7 1 42 140 0.051 0.057 1.11 7.58 10.53 14.57 18.06
8 1 48 160 0.048 0.053 1.10 8.64 11.77 16.00 19.62

T able 1. Overview of the filter properties of nth order RC filters with the same time constant. Dynamic applications usually take into consideration
f−3dB and settling times, whereas for noise measurements taking into account the correct fNEP is key to achieve accurate results. With the
relations given above one can easily calculate filter time constants for filters of the same bandwidth but different order.

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 5
For noise measurements, it’s often more useful to 1+h
specify a filter in terms of its noise equivalent power 1.5
1 + h cos(ωm t)
bandwidth fNEP , rather than the 3 dB bandwidth f−3dB . 1.0

signal amplitude (V)

The noise equivalent power bandwidth is the cut-off 1–h
0.5
frequency of an ideal brick-wall filter that transmits
the same amount of white noise as the filter we wish 0.0
to specify. For cascaded RC filters, the conversion fac- –0.5
tor between fNEP and f−3dB is listed in Table 1. –1.0
After mixing
√ the input signal Vs (t) with the reference
–1.5
signal 2 exp (−iωr t), the input signal spectrum is
shifted by the demodulation frequency ωr and be-
0 4 8 12
comes Vs (ω−ωr ). Low-pass filtering further trans-
time (ms)
forms the spectrum through a multiplication by the fil-
ter transfer function Hn (ω). The demodulated signal Figure 9. Amplitude modulated signal: the green trace is the car-
Z(t) contains all frequency components around the ref- rier input signal (displayed at a lower frequency for clarity). The blue
erence frequency, weighted by the filter response trace indicates the signal amplitude, which is the envelope of the in-
put signal.

Z(ω) = Vs (ω − ωr )Hn (ω). (12)

represented in Figure 9 as an example to discuss how
requirements for different experimental questions can
This equation clearly shows that demodulation be-
be met. The signal amplitude R(t) = 1 + h cos(ωm t),
haves like a bandpass filter in that it picks out the fre-
the blue trace in the Figure 9, is modulated at a fre-
quency spectrum centered at fr and extending on each
quency fm = ωm /2π around the average value 1, where
side by f−3dB . Moreover, it shows that one can recover
the modulation index h characterizes the modulation
the spectrum of the input signal around the demod-
strength. For this example we choose carrier and mod-
ulation frequency fr by dividing the Fourier transform
ulation frequencies of fc = 2 kHz and fm = 100 Hz, re-
of the demodulated signal by the filter transfer func-
spectively.
tion. This form of spectral analysis is often used by
Using the complex representation introduced with Fig-
FFT spectrum analyzers and sometimes referred to as
ure 3, Figure 10 (a) shows the AM signal after mix-
zoomFFT [12].
ing. Its modulus |1 + h cos(ωm t)| is time-dependent
but its angle φc is constant. The term cos(ωm t) is the
Low-pass filter in the time domain sum of the two counter-rotating vectors exp(iωm t) and
The time domain characteristics of a filter is best vi- exp(−iωm t). These two vectors represent the upper
sualized by its step response, as shown in Figure 7 (c) and lower sidebands of the frequency spectrum of an
and Figure 8 (c). These plots correspond to a situation amplitude modulated signal, as seen in Figure 10 (d).
where the input of the filter is changed in a step-like Figure 10 (b) and (c) show the quadrature and in-phase
fashion from 0 to 1. A certain amount of time will be component, respectively.
needed before the filter output settles at the new value. Most applications require measuring one of the follow-
In order to measure a signal that has passed through ing quantities:
a filter accurately, the experimentalist must wait for a
settling time long enough before taking the measure- 1. the time dependence of the amplitude
ment. R(t) = 1 + h cos(ωm t)
Table 1 lists the times to reach 63.2%, 90%, 99% and 2. the average value of the amplitude ⟨R(t)⟩
99.9% of the final value for filters of different orders 3. the modulation index h
but identical time constant τ . Assume we have a
In the first situation, we would like the demodulated
1 MHz signal and want to use a 4th -order filter with a
signal to follow amplitude changes at a rate fm . This
bandwidth of 1 kHz around 1 MHz. From the numbers
requires a filter bandwidth significantly larger than fm .
given in Table 1 we can derive that the time constant is
Consider for instance a 4th -order filter with a band-
69 μs and the settling time to 1% error is 0.7 ms.
width of f−3dB = 500 Hz. With this choice, the trans-
mission at fm = 100 Hz (that is 100 Hz away from the
carrier fc ) is about 98.5% and the phase delay is about
Signal dynamics and demodulation 20◦ as one can calculate from Equation 11 and Table 1.
bandwidth In other words, the modulation signal is only slightly
Setting the demodulation bandwidth is often a trade- affected by the filter. The demodulated signal is dis-
off between time resolution and SNR. Let’s consider an played as the dashed black line in Figure 10 (b) and (c).
amplitude modulated (AM) input signal with carrier fre- Apart from the desired sideband suppres-
quency fc = ωc /2π, sion/admission and phase delay, the amount of
noise in the measurement is an important criterion
Vs (t) = [1 + h cos(ωm t)] cos(ωc t + φc ) (13) in the choice of a filter. Figure 11 illustrates this

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 6
 a b a
quadrature component Y (V)

1.2 1.2 0.9

0.8
0.8 0.8 0.7
0.6

in-phase component X (a.u.)

0.4 0.4
b
0.9
0.0 0.0
0.8
0.0 0.4 0.8 1.2 0 10 20 30
in-phase component X (V) time (ms) 0.7
c 0 d 0.6

0.8 c
10 0.9
time (ms)

0.8
0.4 0.7
20
0.6

30 0.0 0 10 20 30 40 50
0.0 0.4 0.8 1.2 −200 −100 0 100 200
in-phase component X (V) frequency (Hz) time (ms)

Figure 10. (a) An amplitude modulated signal in the rotating frame Figure 11. (a) A noisy input signal will produce a noisy demodulated
of reference is a vector with a time dependent length. The instan- signal, blue trace. The underlying signal without the noise is plotted
taneous signal is represented by the thick blue arrow; the thinner as a black dashed trace. (b) Applying a filter with bandwidth f−3dB =
arrows display the two sidebands of the AM signal. (b) and (c) the fm = 100 Hz will eliminate most of the noise but will also affect the
quadrature and in-phase components of the demodulated input sig- detected signal. (c) Same as (b) but with f−3dB = fm /5 = 20 Hz.
nal: the blue trace is the unfiltered signal, the dashed black, red and
cyan traces are the filtered signals with f−3dB = 500 Hz, 100 Hz and
20 Hz, respectively. (d) The frequency spectrum of the demodulated
ulation frequency accessible with this method can’t
signal after filtering with three different bandwidths (black, red and
cyan curves). be larger than the maximum demodulation bandwidth
of the first lock-in unit. In direct sideband demodula-
tion, the signal is demodulated at fc ± fm in a single
with an AM signal with relatively strong noise after step, and the accessible modulation frequencies are
demodulation in (a). Panel (b) shows the same signal only limited by the frequency range of the lock-in am-
after filtering with a cutoff frequency equal to the plifier. Also, direct sideband demodulation works with
modulation frequency. While this filter eliminates a single lock-in amplifier instead of two and is there-
most of the noise, it introduces systematic changes in fore usually the preferred choice.
the amplitude and phase that need to be corrected to
get accurate results.
For the second set of requirements, frequency compo- Achieving high SNR
nents corresponding to the sidebands are rejected by
reducing the filter bandwidth to a value smaller than Reducing the filter bandwidth generally leads to higher
fm . A 4th -order filter with f−3dB = 20 Hz, dashed cyan SNR at the expense of time resolution. What other
line in Figure 10 (d), suppresses the sidebands by 0.03 measures can be taken to improve SNR?
or 30 dB. Figure 11 (c) illustrates the effect of such a If the signal strength cannot be increased, the noise
strong filter on the measurement. has to be reduced or avoided as much as possible.
In the third case, we want to know the modulation in- However, noise is always present in analog signals and
dex h but don’t need to resolve the full signal dynam- arises from different sources, some of which are of fun-
ics. This is used, for instance, in Kelvin probe force damental origin, for example Johnson-Nyquist (ther-
microscopy, where h is a measure of the electrostatic mal) noise, shot noise and flicker noise, while others
force between a probe and a sample in response to an are of technical origin, as for example ground loops,
alternating voltage at fm . Since the modulation index interference, cross-talk, 50–60 Hz noise or electro-
is proportional to the amplitude of the sidebands, this magnetic pick-up. The magnitude of a random volt-
measurement can be performed by applying narrow fil- age noise Vnoise (t) is specified by its standard deviation.
ters around the sidebands at fc −fm and fc +fm . There In the frequency domain, noise is characterized by its
are two ways to do this: by so-called tandem demodu- power spectral density |vn (ω)|2 in units of V2 /Hz, or by
lation or by direct sideband demodulation. |vn (ω)| in units of V/√Hz.
In tandem demodulation, we first perform a wide-band The qualitative spectrum in Figure 12 shows that dif-
demodulation around the center frequency. The re- ferent noise sources have different frequency depen-
sulting signal, typically looks similar to the one in Fig- dencies: while Johnson-Nyquist noise has a flat spec-
ure 11 (a), is then demodulated again at fm . The mod- trum for all practical frequencies and therefore con-

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 7
 50–60 Hz noise, a mixer LP filter
acoustic and other input signal VS(t) X
ADC
interferences
oscillator
reference
filter signal VR(t)
+90°
Y
radio, ADC
amplitude

mobile
mixer LP filter

1/f-noise b
input DSP mixer LP filter
signal X
ADC
white VS(t) oscillator
noise reference
f1 frequency f2 signal VR(t) +90°
Y
Figure 12. Qualitative noise spectrum of a typical experiment. The
measurement frequency should be chosen in a region with small mixer LP filter
background, avoiding any discrete peaks coming from technical
sources. In the example, f2 will yield better results than f2 for the
same filter bandwidth, since it is located in a clean white noise re- Figure 13. (a) Analog lock-in amplifier: the signal is split into two
gion above the 1/f noise at low frequencies. paths, mixed with the reference signal, filtered and then converted
to digital. (b) Digital lock-in amplifier: the signal is digitized and then
multiplied with the reference signal and filtered.
tributes to the “white noise”, flicker noise has a 1/f fre-
quency dependence (“pink noise”). If there is some
freedom in the choice of modulation frequency, we can 100, because the noise amplitude is proportional to
zoom in to a part of the spectrum where the noise the square root of the bandwidth. The settling time to
level is lowest. Often higher frequencies where the 1% then increases to more than 2 minutes. The lock-
spectrum consists of white noise characteristics work in technique can support such long measurements be-
best. Figure 12 illustrates this approach: the amount cause it is insensitive to DC offset drift in the input
of noise inside a filter, indicated by the blue and gray signal. Nonetheless, other sources of drift such as
filled area, is larger for example in the lower frequency changes in the DUT resistance, or in amplifier gain,
1/f noise region. Hence, the SNR at f2 is higher than at may affect long measurements. Maintaining stable
f1 using the same filter bandwidth, because the noise conditions and especially constant temperature are
density is lower as long as other noise sources, such then crucial.
as as radio and wireless transmission are avoided.
To give a more quantitative example, let us assume we
want to measure a sinusoidal signal with amplitude State of the art
of 1 μV across a 1 MΩ resistor with a SNR larger than
Since the early 1930s lock-in amplifiers have come a
10. Such a resistor R exhibits a thermal noise with a
long way. Starting from vacuum tubes as basic instru-
power spectral
√ density of vn2 = 4kB T R, which amounts
√ ment technology, we note the transition to digital is
to about vn2 = 0.127 R nV/√Hz =127 nV/√Hz at T well underway but not yet complete. In digital lock-in
= 300 K room temperature1 . In this example, thermal amplifiers, the input signal is immediately converted
noise is identified as the dominant noise source. It is to the digital domain by an analog-to-digital converter
clearly stronger than the lock-in input noise of typically (ADC) and all subsequent steps are then carried out nu-
less than 10 nV/√Hz. We can thus calculate the SNR as merically by digital signal processing (DSP), as shown
in Figure 13 (b). In contrast, analog lock-in amplifiers
use analog elements like voltage-controlled oscilla-
1 μV
SNR = √ √ = 10 (14) tors, mixers and simple RC filters for signal process-
127 nV/ Hz · fNEP ing. There are also hybrid versions [9], as sketched in
Figure 13 (a), which digitize the signals only after the
By solving this equation for fNEP , we calculate that we
analog mixing stage before or after filtering.
need to select a NEP filter bandwidth of 620 mHz or
The transition from analog to digital was fueled by
less to achieve a SNR of 10. We choose a 4th order fil-
the availability of ADCs and DACs with ever increas-
ter. From Table 1 we can calculate the corresponding
ing speed, resolution and linearity. This development
cutoff frequency f−3dB = 549 mHz, the time constant τ
helped to push the frequency range, input noise and
= 126 ms, and the settling time to 1% is 1.26 s.
dynamic reserve to new limits. In addition, digital sig-
To further increase the SNR by a factor of 10, we would
nal processing is much less prone to errors introduced
need to decrease the filter bandwidth by a factor of
by a mismatch of signal pathways, to cross-talk and
1 Boltzmann constant k
B = 1.381×10−23 V2 /(Ω Hz K) to drifts, caused for instance by temperature changes.

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 8
Figure 14. Zurich Instruments UHFLI Lock-in amplifier representing the state of the art of lock-in technology. The 600 MHz signal input band-
width as well as the 5 MHz demodulation bandwidth make it by far the fastest lock-in amplifier on the market today. In addition, the 19 inch
wide instrument integrates the greatest amount of functionality, see Figure 16, while providing the most advanced instrument control software
LabOne® (see Figure 15).

This is particularly critical at higher frequencies. But Clearly, the wealth of functionality indicated in Fig-
the biggest advantage of the digital approach is prob- ure 16 cannot be controlled and utilized with a few
ably the ability to analyze the signal in multiple ways knobs and buttons on the front panel. Instead, the
simultaneously without loss of SNR. As mentioned ear- UHFLI is entirely controlled from a computer running
lier, this enables not only better dual-phase demodu- LabOne® , an instrument control software using the
lation, but also the analysis of several frequency com- latest browser technology that provides a graphical
ponents of a signal directly, without the need to cas- user interface to any device with a web browser, see
cade multiple instruments with all the accompanying Figure 15. High-level tools such as the Parametric
detrimental effects. Sweeper, the Software Trigger, or the PID Advisor, ex-
After the transition from analog to digital, another sig- ploit the available processing power of the host com-
nificant step of innovation was sparked by the avail- puter for measurement tasks, which improves con-
ability of field programmable gate arrays (FPGA) with fidence in the measurement results, and enables a
high computing power, abundant memory and speed. more efficient workflow. In addition LabOne also of-
FPGAs are well understood as digital clockworks that fers programming interfaces for LabVIEW® , MATLAB® ,
can be flexibly programmed to carry out almost any Python and C# to conveniently integrate the measure-
desired signal processing task in real time. The natu- ment instrument into existing experiment control envi-
ral extension of the lock-in is to add time domain and ronments.
frequency domain analysis before and after demodu-
lation, something that would otherwise be done with a
separate scope and spectrum analyzer. Furthermore,
a single instrument can contain boxcar averagers to
analyze signals with low duty cycle, PID and PLL con-
trollers for feedback loops and arithmetic units to pro-
cess measurement data in real time. The measure-
ment signals can then be transferred to a computer for
further analysis. If an analog interface to another in-
strument is needed, measurement data from different
functional units are easily converted back to the ana-
log domain using high-resolution DACs.
The most advanced instrument today regarding speed
and level of integration is Zurich Instruments’ UHFLI
[13], introduced in 2012. Figure 14 shows the instru-
ment front panel. The UHFLI has a signal input band-
width of 600 MHz and a maximum demodulation band-
width of 5 MHz, which makes it by far the fastest lock-
in amplifier on the market today. Despite high speed,
it still provides exceptional input noise performance of
only 4 nV/√Hz and a dynamic reserve of 100 dB. The
high level of integration is illustrated in Figure 16 show- Figure 15. The LabOne® user interface of the UHFLI Lock-in ampli-
fier uses the latest web browser technology. The instrument can be
ing the main functional components of the UHFLI and controlled from multiple browser sessions on multiple PCs, tablets,
their interconnections. Functionality that used to re- etc. at the same time. Every signal analysis and control tool has a
quire an entire rack of instruments is now housed in a dedicated tab. Some of the functionality is intuitively displayed in
single instrument no larger than a shoe box. form of block diagrams.

Zurich Instruments – White Paper: Principles of lock-in detection and the state of the art Page 9
 Analog and Fast digital signal Data link LabOne data
digital interface processing on FPGA to PC server & toolkit

Auxiliary 14 MSa/s Spectrum

Outputs
2x
AWG
14 MSa/s 2 Ch
Signal 1.8 GSa/s Sweeper
Outputs 1.8 GSa/s USB
2x OSC PID
8x 4x LAN
Num
Trigger 1.8 GSa/s
Outputs
2x Demo- AU
dulator 2x Plotter
1.8 GSa/s
Trigger 8x
Inputs
2x Boxcar SW Trigger
2x
Auxiliary 100 kSa/s
Inputs PWA
4x 2x USB Harm
CNT LAN 2x
4x
Signal 1.8 GSa/s
Inputs Scope FFT
2x 2 Ch 128 MS

Figure 16. Block diagram showing the Zurich Instruments UHFLI’s main functional entities and the signal flow between them. Fast digital signal
processing takes place inside the instrument’s FPGA clocked at 450 MHz but also on the computer connected by USB or 1GbE running the instru-
ment control software LabOne® . The main functional components inside the instrument are the 8 dual-phase demodulators, an oscilloscope
(Scope) with digitizer functionality (DIG) and FFT , PID modules with PLL capability, an arithmetic unit (AU), a boxcar averager with periodic wave-
form analyzer (PWA) and a pulse counter module (CNT). For signal generation the instrument provides sinusoidal signal generators (OSC) and
arbitrary waveform generators (AWG) for complex signal shapes. The LabOne control software running on the PC adds a parametric sweeper, a
spectrum analyzer, a numerical parameter display (Num), a plotter, a software trigger for time domain analysis and a harmonic analyzer (Harm).

References
[1] C. R. Cosens. A balance-detector for alternating- [7] D. C. Tsui, H. L. Stormer, and A. C. Gossard. Two-
current bridges. Proceedings of the Physical So- dimensional magnetotransport in the extreme
ciety, 46:818, 1934. quantum limit. Phys. Rev. Lett., 48:1559, 1982.
[2] W. C. Michels. A Double Tube Vacuum Tube Volt- [8] L. Gross et al. Bond-Order Discrimina-
meter. Rev. Sci. Instrum., 9:10, 1938. tion by Atomic Force Microscopy. Science,
337(6100):1326, 2012.
[3] W. C. Michels and N. L. Curtis. A Pentode LockIn
[9] Stanford Research SR844. Product web page.
Amplifier of High Frequency Selectivity. Rev. Sci.
Accessed: 2016-11-21.
Instrum., 12:444, 1941.
[10] Wikipedia Article: Fourier Transform.
[4] Interview of Robert Dicke by Martin Hawrit. Accessed: 2016-11-21.
Niels Bohr Library and Archives, College Park,
[11] Wikipedia Article: Fourier Series.
MD: American Institute of Physics, 1985, web
Accessed: 2016-11-21.
link. Accessed: 2016-11-21.
[12] N. Thrane. Zoom-FFT. Brüel & Kjær Technical
[5] Zurich Instruments HF2LI. Product web page. Review, (2):3, 1980.
Accessed: 2016-11-21.
[13] Zurich Instruments UHFLI. Product web page.
[6] A. M. Skellett. The Coronaviser, an Instrument Accessed: 2016-11-21.
for Observing the Solar Corona in Full Sunlight.
Proc Natl Acad Sci USA, 26(6):430, 1940.

Zurich Technoparkstrasse 1
CH-8005 Zurich
www.zhinst.com
[email protected]
Disclaimer: The content of this document is provided by Zurich Instrumenty ‘as is’.
Zurich Instruments makes no warranties with respect to the accuracy or completeness of

Instruments Switzerland +41 44 515 04 10

the content of this document and reserves the right to make changes to the specification
at any time without notice. All trademarks are the property of their respective owners.