Title: Noises and Interferences limit the sensitivity and resolution of the signals from the sensors
1Noises and Interferences limit the sensitivity
and resolution of the signals from the sensors
- What are the tricks to fight against these
unwanted signals?
2Signal recovery techniques
- Amplitude Rectification and Phase Detection
- Lock-In Amplifiers
- Boxcar and Signal averagers
- Phase locked loops
3AC vs DC methods of sensor probing
- In many cases signals from sensors are
represented by relatively slowly changing signals
(voltages, currents). The question is what is
the best way to accurately measure these changing
signals? (If the changes are not slow the answer
is more obvious DC is ruled out) - DC sensing
- Source of excitation is DC
- Bandwidth considerations
- fL 1/tacq where tacq is the acquisition time.
- fH 1/2tconv where tconv is the ADC conversion
time - Advantages simplicity, no sophisticated
detection is needed - Disadvantages parasitic DC signals which are
hard to separate from useful signals and which
can exceed the useful signal (T-EMF, offsets,
drifts,etc), suffers from 1/f noise - AC sensing
- Source is AC signal
- Advantages
- No influence on the signal from DC parasitics,
- Narrow-bandwidth signaling allows to minimize
noise and coherent interference - Allows to measure differential parameters and
derivatives - Can be used to measure even small DC signals (by
using chopping) - Disadvantages more complicated circuitry (but
not always)
4Example How to measure Coulomb blockade
oscillationsin Single-Electron Transistors?
- Conductance oscillates as a function of gate
bias, Vg - Signals MUST be small, because of the small scale
of energies (eVdsltltEC1 meV), Vds100 mV - What to choose for measurements
- AC or DC ?
5DC measurement
- Conductance G (Vg)
- For small I and V
- Vds 100 mV
- Using IV-converter
- To measure DC Vout using DVM with averaging time
of 60 conversions per sec - Bandwidth estimation
- fL1/1sec 1 Hz
- fH 0.3146019 Hz for conversion time of 1/60
sec - BW18 Hz
- 1/f noise may be significant
6AC measurement
- If we use AC excitation
- Signal at the output of the IV converter
- Then after rectification and Low Pass Filter
- How to do rectification
- Amplitude detection
- Phase detection
RF
7Conversion from AC to DC
- Simplest and inaccurate use a diode
- Much better fill wave rectifier with opamps
- Whats missing?
- A low pass filter to get rid of ripple
8Ripple in the rectifier
- For a sawtooth the factor from P-P to RMS is
1/(2v3) (instead of 1/(2v2) for a sinewave) - The frequency is doubled (thanks to full wave
rectification), so Vripple(RMS) - For 1 sec time constant and 60 Hz
Vripple(RMS)VP 0.024 - For 1 sec time constant and 289 Hz
Vripple(RMS)VP0.0005
9Amplitude Detection in presence of Coherent
Interference
VOUT(t) abs(VS sin(wStf) VCI sin(wCItf2))
Signal
LPF removes high frequencies 2w
LPF
Signal CI
- Low-pass filter removes the ripple but does not
remove rectified CI signal! - fT2Hz (red)
- fT1Hz (blue)
- fT0.5Hz (magenta)
- fT0.25Hz (green)
Coherent Interference
10Amplitude Detection in presence of Noise
Vexc(t) abs(VS sin(wStf) VNoise)
Signal
LPF
Signal Noise
- Low-pass filter removes the ripple but does not
remove rectified NOISE signal! - fT2Hz (red)
- fT1Hz (blue)
- fT0.5Hz (magenta)
- fT0.25Hz (green)
Noise
11Amplitude detection summary
- If the SNR of the incoming signal is large AD can
be safely used - BUT
- Amplitude detector does not allow to recover
signal in presence of noise or CI. - The only possible improvement is to use a
narrowband filter before detector so there is no
CI or noise before signal is rectified - If the envelope function of modulating signal
changes sign, there is no way to figure it out
using AD because unwanted signals are rectified
as well - Is there a way to fix that?
12Phase Sensitive Rectifiers I (multiplication)
- Demodulation can be done by analog multiplier
- A reference signal (which is in-phase with
transducer excitation) - is multiplied with AC signal from the QUT
- Multiplying with the reference
- Using trig identities
- As a result VZ is the desired signal multiplied
by a scale factor - Unique phase information (f1-f2 ) is used to
decode the signal - If Df0 (between the reference VR and the signal
V0) - After putting it through LPF
Vout
Vref Asin(wctf2)
Vout f (Vg)sin(wctf1)
Vref Asin(wctf2) f (Vg)sin(wctf1)
VZ A/2f (Vg)(cos(f1-f2 ) - cos(2wctf1f2)
VZ A/2f (Vg)(cos(0) - cos(2wctf1f2)
VZ Af (Vg)1/2
13How phase detector rectifies a sinewave
- Output voltage after LPF is proportional to the
Magnitude of the input signal - Phase difference is between the reference and the
signal we want to demodulate
14Phase Sensitive Rectifiers II(flipping sign
technique)
- Switching gain technique
- Switch closed total gain is -1
- Switch opened total gain is 1
- Equivalent to multiplication of signal by the
function SGN(t) - Signal
- After multiplication by SGN
- Low-pass filter removes high-frequency components
2f and higher - The modulating signal is recovered
- Works for harmonics- one can use square wave
modulation
15Coherent Interference rejection in PSD
Signal
LPF
Signal CI
VoutVS?cos(Df)
Reference Signal (f0)
- Low-pass filter removes the ripples at parasitic
frequencies - fT2Hz (red)
- fT1Hz (blue)
- fT0.5Hz (magenta)
- fT0.25Hz (green)
Coherent Interference
16Noise rejection in PSD
Signal
LPF
Signal Noise
VoutVS?cos(Df)
- Low-pass filter removes the traces of noise
- fT2Hz (red)
- fT1Hz (blue)
- fT0.5Hz (magenta)
- fT0.25Hz (green)
Noise
17Comparing the results
- for acquisition time of 1 sec
- DC method BW 18Hz
- AC method with 300 ms time constant BW 0.8Hz
- SNR improvement for white noise in Vout
measurement - Far better in the presence of extra noises and
interferences - Amplitude rectification does not improve anything
unless one puts the filter BEFORE rectifier - Filtering the rectified signal in the amplitude
detector only reduces ripple, but does NOT change
the bandwidth for the signal recovery - But very same filter used with phase detector
will further improve the SNR by attenuating the
noises
18Example
- Signal from the sensor is a 10 nV sine wave at 10
kHz. Too small to be used for control/monitoring.
Amplification is needed! - A very good low noise amplifier adds 5 nV/vHz
of input noise. - For the amplifier bandwidth of 100 kHz and the
gain is 106, the output produces 10 mV of signal
(10 nV x 106) and 1.6 V of broadband noise (5 nV/
v Hz x v 100 kHz x 106). All you see is noise! - Add a high quality band pass Q100 centered at 10
kHz, bandwidth 100 Hz (10 kHz/Q). The noise
after the BPF will be 50 mV (5 nV/ v Hz x v 100
Hz x 106) and the signal will still be 10 mV. - Using PSD the bandwidth can be squeezed by LPF
down to 0.01 Hz In this case, the noise in the
detection bandwidth will be only 0.5 mV (5 nV/ v
Hz x v 0.01 Hz x 106) - The SNR is now 20 thus sensor signal can be
accurately measured.
19What is measured using detectors (both AD and
PSD)?
- For sinewave test signals
- Average value of fully rectified sinewave is
- Once we get rid of ripples, we can end up with
amplitude, A - So for sinewave to get RMS from A is to multiply
it by 0.707
20What is a lock-in Amplifier?
21Lock-In Amplifier Technique. How it works.
Carrier signal
- Time varying dependences can be measured
- Change in the sensor modulates the carrier signal
- PSD rectifies the signal
- After PSD the shape of the modulating curve is
nicely recovered - Note that amplitude rectifier will lose all the
information about the signature of the modulating
signal hidden in the phase - Lock-in detection amounts to bandwidth narrowing
again, with the bandwidth set by the
post-detection low-pass filter
Green - original function Red filtered signal
after PSD
22Lock-in applications. Small Light Intensity
measurements
- In many optical measurements the intensity of a
beam of light reflected or passed through the
sample is of interest. If the light beam is very
weak then the electrical signal from the
photo-detector is very weak and has to be
amplified. A continuous optical beam will create
a DC signal at the output of the photodetector - High-gain DC amplification is difficult (drift,
offset, noise). Also, stray light will also be
amplified and appear at the output - So conversion from DC to AC is needed. This can
be achieved by chopping - The most common technique is to pass the beam
through a rotating disk that has holes or slots
cut into it at regular intervals. As the disk
rotates it "chops" the beam producing an on/off
signal which when detected by a photodetector
will appear as an AC signal - The mechanical chopping of the light beam is very
precisely controlled by the chopper and therefore
the resultant AC signal due to the chopped light
is at a known and stable frequency which can be
monitored and amplified easily using an extra
photodetector
Chopper
Sample (for transmission)
Laser
Signal PD
Beam splitter
Reference PD
23Lock-In Amplifier application. Use of choppers
Sample is affected by external parameter
- Chopper replaces small DC signal from PD with AC
signal by chopping the beam - At the same time reference signal is created by
chopping direct beam - The signal from Signal PD is magnified with
narrow-band amplifier - Phase sensitive detector converts signal into DC
with magnitude proportional to the magnitude of
signal and phase shift between signal and
reference - DC signal is further magnified and filtered
- The functional dependence is recovered
24Lock-In applications using AD630 demodulator
Poor mans lock-in
25Dual channel Lock In amplifier
VR1
VC
VIN
2 phase LIA
- Very often both X and Y are of interest
- First PSD uses in-phase reference VL cos
(wtQref) - Second PSD uses out of phase (shifted by 90
degrees reference) - 1st PSD
26Application of the lock-in technique to the
measurements of function derivatives
- There are many situations in science and
engineering when the desired information is
hidden in the derivative of the QUT response to
external parameter (EP) rather then in its
functional dependence - How to measure the derivative?
- Applying a small modulating signal on top of the
changing EP - Dvmod V0cos(wt)
- DVout(dVout/dvmod)Dvmod.
- Total signal is Vout f(EP) DVout
- f(EP) is slowly changing signal, well below the
bandwidth of the modulation signal - After demodulation we get the derivative dV/dvmod
V0
27Example of derivative measurement using numerical
method
- IV measurement of nanoscale microstrips
- Conventional IV measurement shows straight line
superimposed with noise - With noisy signals numerical differentiation does
not produce good results signal is sunk into
noise
28Measurement of I-V 1st derivative using lock-in
technique
Excitation voltage 0.1 V
Lock-in Amplifier SR830
Internal Oscillator Signal Exc Ref Signal
input
Division by 103
Microstrip
VDS
IV converter
Ramp generator
- Experimental setup
- Simple resistive divider is used
- to provide ramping VDS
- and a small AC signal of constant amplitude
(100 mV) superimposed on - Internal oscillator (built into lock-in) is used
to provide both reference and excitation signals - IV converter is used to magnify small current
- Signal is amplified and demodulated in lock-in
amplifier - When lock-in technique is used, clear feature on
the dI/dV curve shows up
Instrumental
Numerical
29Measurements of higher order derivatives
30AC measurement techniques -II
- What to do if lock-in could not be used?
- Signal averagers
- Phase locked loops
- Frequency and time interval measurements
31Limitations of Lock-in technique
- The best method to recover a signal from noise
depends on the nature of the signal of interest
and the required representation of the result - Lock-in Amplifiers are extremely powerful signal
recovery instruments if the signal can be made to
be an amplitude modulated AC waveform, where the
envelope of the modulation is the required output - However, where this cannot be done, or where the
rise/fall time of the signal exceeds the
available bandwidth, or where the signal is short
lived, other techniques are needed - Why? Narrow bandwidth is bad for signals for
which Fourier spectrum is not a single peak a
significant portion of information is lost
32Signal Averaging
- By repeating the experiment and averaging the
results, SNR can be greatly improved - Signal-to-Noise Improvement ratio (SNIR)vn
- Instruments available for averaging which must be
synchronized with the source of pulses - Digital Oscilloscopes
- Fast ADC capable of fast data transfer to the
storage - Boxcar Averagers
- Signal Averagers
SNRIN
ADC
Averager
SNROUT
Averager
Pulse Generator
QUT
Detector
Sync
33Boxcar averager (aka boxcar integrator, gated
integrator)
- The Boxcar Averager uses analog electronics,
supported by digital control, to monitor one
discrete point in time on a repetitive signal. - Gated Integrator circuit is the key. S1 sampling
gate RC storage circuit - Choice of RC V0 is close to V1 by the end of
gating - High frequency components are removed Equivalent
noise bandwidth - 1.57/(2pRC)
- It builds up an average of that point over many
cycles before recording it as a value - Two types of averaging. Note that outputs of
averager (not gated integrator!) are shown.
Outputs of the gated integrator look like pulses
of almost equal height - Linear (good for digital storage)
- Acquire a sample, pass it to averager, reset the
charge stored in C. Note that each step is
nothing but a stored voltage across C by the end
of gating time. - Exponential (analog storage) RCltltR2C2
- Acquire a sample, pass it to the exp
averager(2nd), reset the the charge stored in C
of the first integrator - After averaging is finished the reset is done to
null the averager storage (analog or digital)
Note that the steps Are not of the same height.
For exp integrator the output is reaching the
height of single step in linear integration
method only by the end of the averaging cycle
34Use of Boxcars
- Two modes of operation
- Static gate pulse recovery
- Waveform recovery.
- The groups of gated signal corresponding to fixed
delay are averaged, then delay is incremented,
and process is repeated for a new delay. - Usually the gate interval is shorter compared to
static pulse recovery - The number of datapoints representing the signal
is m - In either case make sure your bandwidth is
adequate!
35Some useful info about boxcars
- Boxcar are the best for averaging a single point
in time repetitively. - Example the amplitude of one peak of a
spectrum, from a repetitively swept
monochromator, could be averaged easily and
recorded as a function of time using a Boxcar
system. - This technology can also give good time
resolution, with better than 1 ns being possible.
36Signal Averagers vs Boxcar
- The Signal Averager uses digital techniques to
record all of the waveform on each cycle. This
makes it much more time efficient than Boxcar
systems. Nonetheless the time taken to do the
summation does limit the maximum data throughput
unless a dedicated hardware averager is included - In the case of Gaussian noise, the improvement in
signal-to-noise ratio gained from this process is
approximately equal to the square root of the
number of summed cycles. Hence averaging 100
records of an identical event will improve the
signal-to-noise ratio by 10 times. - Signal Averagers can provide maximum time
resolutions of a similar level, but are better
suited to waveform recovery and to monitoring
short lived phenomena due to their better time
efficiency.
37Phase Locked loop basics
- PLL is a frequency selective circuit designed to
synchronized with the incoming signal and
maintain synchronization inspite of noise or
variations in the incoming signal frequency - Consists of phase detector (PD), loop filter and
a VCO (voltage-controlled oscillator) - VCO generates a free standing frequency if no
input is applied - If input frequency changes an error voltage
develops to adjust the frequency which will then
minimize the phase shift - VE can be used as output to detect changes in
input frequency - VCO frequency can be used to recover signal
buried in the noise and for frequency synthesis
38Frequency multiplier locked to the reference
- Inserting frequency divider in PLL feedback is
similar to inserting voltage divider in feedback
loop of op-amp voltage swing increases for
op-amp, frequency swing increases for PLL - Phase detector converts phase to voltage
- VCO converts voltage to time derivative of the
phase (frequency) - Fixed input voltage error produces a linearly
rising phase error at the VCO output - Generated frequency is 60 Hz 102461440 Hz.
This results in 7.5 measurements/sec (4096 pulses
per ramp)
39FM demodulator using PLL
- Change in input frequency leads to change in VE,
VE in turn will adjust the local oscillator
frequency - Popular chip 4046 is used (contains buffers, PC
and VCO) - VE is used as an output for frequency
demodulation - Input is 1 MHz10kHz
- For VCO, free standing frequency is set to 1MHz
(by choosing 100pF cap) - The bandwidth for a LPF (connected after phase
comparator PC1) is set to 10 KHz
40Frequency and Period Measurements
- Input signal is converted into pulses
- 1 second pulse comes from divided oscillator
- The signal starts BCD (binary coded decimal)
counter - Result is latched and displayed
- Counter is reset between counting intervals
- Limitations it is hard to accurately measure
frequencies close to 1/Tgate. - Example with a gate time of 1 sec 10 Hz will be
read as 9,10, or 11, whereas 1 MHz can be
measured in the same setup with 1ppm resolution.
41Phase Locked loop basics
- PLL is a frequency selective circuit designed to
synchronized with the incoming signal at f0 and
maintain synchronization inspite of noise or
variations in the incoming signal frequency - Consists of phase detector (PD), loop filter and
a VCO (voltage-controlled oscillator) - VCO generates a free standing frequency if no
input is applied - If input frequency changes an error voltage
develops to adjust the frequency which will then
minimize the phase shift - VCtrl can be used as output to detect changes in
input frequency - VCO frequency can be used to recover signal
buried in the noise and for frequency synthesis
42Frequency multiplier locked to the reference
- Inserting frequency divider in PLL feedback is
similar to inserting voltage divider in feedback
loop of op-amp voltage swing increases for
op-amp, frequency swing increases for PLL - Phase detector converts phase to voltage
- VCO converts voltage to time derivative of the
phase (frequency) - Fixed input voltage error produces a linearly
rising phase error at the VCO output - Generated frequency is 60 Hz 102461440 Hz.
This results in 7.5 measurements/sec (4096 pulses
per ramp)
43FM demodulator using PLL
- Input is 1 MHz10kHz how to detect it?
- Change in input frequency leads to change in VE,
VE in turn will adjust the local oscillator
frequency - Chip 4046 is used (contains buffers, PC (phase
comparator) and VCO) - For VCO, free standing frequency is set to 1MHz
(by choosing 100pF cap) - PC gives 0 after filter if f in 1 MHz
- vo is used as an output of FM demodulator
- The bandwidth for a LPF (connected after phase
comparator PC1) is set to 10 KHz - The output signal is free from noise, because of
the squeezed bandwidth
44Summary for signal recovery
- Always try to minimize the bandwidth to improve
SNR - Noisy signals could be recovered by carefully
choosing the appropriate instrumentation - If the information from the sensor is encoded in
the magnitude of the signal and probing signal is
a sinewave or, more generally, a waveform where
most of the power is concentrated within first
few harm onical components (meander, symmetrical
triangular, wave, clipped sinewave, etc), then
use lock-in amplifiers for signal recovery - If the signals of interest are repetitive, but
have broad spectrum (e.g., pulses with duty cycle
Dt/T0ltlt1 (where T0 is a period and Dt is a pulse
length), use boxcar averagers ofr signal
averagers - If the information is encoded in the frequency
change around the carrier (tone modulation, FM )
or there is a need to create a signal with rigid
phase attachment to some standard, then use Phase
Locked Loops