3.

MEASUREMENT OF NOISE

3.1. Analog measurements of noise

Analog:

- real/measured one-to-one correspondence

- x=xm∆x: ∆x reference, xm real

- continuous in amplitude, no quantization

- analog measurement equipment

3.1.1. Measurements of <x>, RMS(x),

VAR(x)

Operation to be performed:

- integration over time

e.g. RC circuit, DC measurements

integrator: condensator

3-1
- filtering out high frequency components

<x> ∼ X(f=0)

RMS or VAR:

- the same techniques+squaring for

<x2>

- average power measurements

integrator: heat dissipation

3-2
p(x):

3-3
3.1.2. Measurement of spectral density

Problem:

measuring the average power in f±∆f/2

Method:

Bandpass filtering and RMS or VAR

measurement

3-4
3.1.3. Measurement of correlation

- calculation from S(f)

- direct measurement:

Problems:

- always finite time -> appriximate results

- won’t be exactly zero for independent

processes

3-5
3.2. Digital measurements and digital signal

processing

Why digital?

3.2.1. Basic elements of digital

measurements

3.2.1.1. A/D converters (ADC)

Basic function:

- convert quantities to integer numbers

3-6
3-7
Advantages:

- high precision

- easy and transmission

- easy storage and processing

- easy isolation

- no aging

- no temperature dependence, ...

Applications:

- digital instrumentation

- radar, medical instruments, digital control

- weight scales, image scanners, digital

thermometers, digital audio, video,

computer multimedia, cellular phones, ...

3-8
Resolution: 6..24 bits

Speed: 1Hz..1GHz

3-9
3.2.1.2. D/A converters (DAC)

Basic function:

- generate quantities ∼ integer numbers

3-10
Advantages:

- high precision

- digital format: transmission, storage,

processing, easy isolation, ...

Applications:

- digital instrumentation

- digital control

- digital waveform generation, digital audio,

video, computer multimedia, cellular

phones, monitors, ...

Resolution: 8..20 bits

Speed: 100kHz..500MHz

3-11
The R-2R ladder realization:

3-12
3.2.1.3. Digital signal processors (DSP)

What about having a very efficient, small

programmable signal processing element?

Basic function:

- special purpose single-chip microcomputers

Advantages:

- very efficient and high-speed signal

processing (20 MIPS for less than 10$, up

to 1600 MIPS)

- digital format:...

- no or small hardware changes -> new

function

3-13
3.2.2. Sampled data systems

Measurement of a time dependent signal x(t):

sampling

The sampled signal:

3-14
Shannon’s sampling theorem:

If the signal x(t) has no components over

the frequency fmax, than the signal can be

represented by its discrete set of values

x(k∆t) without loss of information, where

∆t < 1/2fmax.

Reconstruction of x(t):

3-15
Proof:

3-16
3-17
3.2.3. Quantization and aperture jitter noise

After A/D conversion, the value truncated:

quantization error: q(x) (sawtooth function)

For time dependent signal: quantization noise:

RMS=ULSB/√12

3-18
Aperture jitter noise

Aperture jitter:

random uncertainity of the sampling time

instants

Can be converted to amplitude uncertainity

(depending on the time derivative of the

signal)

3-19
3.2.4. Aliasing, antialiasing filters

What happens, if the signal contains

frequencies over fs/2?

For example:

f = kfs+∆f

The frequency of the measured signal:

3-20
Solution:

- oversampling

- filtering

3-21
Measurement system with anti-aliasing filter

Examples:

- fs=20kHz -> 1kHz, 19kHz, 21kHz all the

same after sampling

- digital audio: fs=44.1kHz, filter cuts off

between 20kHz..22kHz

- sigma-delta technique: 64x oversampling,

filter requirements relaxed

3-22
3.2.5. Using aliasing for frequency

conversion

Measured frequency fm versus the signal

frequency f (sampling frequency fs):

Undersampling: f > fs/2

-> frequency transformation occurs

3-23
3.2.6. Measurement of probability and p(x)

Must not apply anti-aliasing filters!

- No problem: the time structure is

unimportant.

The required formula: p(x)∆x ≈ Ni/N

3.2.7. Measurement of power density

spectrum

Data:

- sampled data (time quantization)

- amplitude quantization

- finite time samples

3-24
3.2.7.1. DFT, FFT

Method:

discrete Fourier transform (DFT):

For real xi: Xk=X*N-k

Fast version: FFT /n log(n) vs. n2/

Finite time analysis -> averaging required to

estimate power spectra

3-25
3-26
3-27
3.2.8. Window functions

If:

- Measurement time: T

- Periodic signal: T’

- T≠nT’, n integer

- periodic expansion (DFT does this)

-> sideeffects

Improving analysis: window functions

- improves detection of periodic components

- not for any signal, type depends of the

signal

- not recommended for noise

- DC component of x(t) should be removed!

3-28
- destroys resolution for T=nT’ correlated

sampling

Typical window functions for 0..T:

- rectangular

w(t) = 1, if 0<t<T

- triangular

w(t) ∼ 1-2 t/T-1/2

- Hann

w(t) ∼ 1-cos(2πt/T)

- Hamming

w(t) ∼ 1.85 (0.54-0.46 cos(2πt/T))

3-29
- Blackmann-Harris

q=2πt/T

w(t) ∼ 0.35875-0.48829 cos(q)

+0.14128 cos(2 q)-0.01168 cos(3 q);

- Gaussian

w(t) ∼ exp(-(6 t/T-3)2/2)

3-30
3-31
3.2.9. Time dependent spectral analysis

- finite time analysis (windowing)

- window swept in time

3.2.9.1. Wavelets

Frequency dependent window width

3-32
3.2.9.2. Windowed FFT

Frequency independent window width

3-33
3.2.10. Measurement of special quantities

E.g.:

- level crossing statistics

- conditional probability, second order

probability densities

- etc.

3-34
3.3. Small signal and low noise

measurements

Measurement of small signals and low noise

can be really challanging:

- very small quantities

- hard to isolate from other sources

- non-stationarity

Measurement equipments and transmission

channels always introduce noise:

- preamplifier noise

- radiated, capacitively or galvanically

coupled noise

- quantization noise

3-35
- thermal fluctuations

- etc.

3.3.1. Operational amplifier noise

Opamp:

- basic element for amplification and other

signal processing tasks

- has its own internal noise

- opamp noise may affect total signal-to-

noise ratio

Opamp noise sources?

3-36
 Noise voltage Multiplied by
Noise of Rs = Vs 1+R2/R1
Current noise in Rs 1+R2/R1
Voltage noise VN 1+R2/R1
Noise of R1 = V1 -R2/R1
Noise of R2 = V2 1
Current noise in R2 1

3-37
Total noise power spectral density referred to

the output:

where G=1+R2/R1.

3-38
Optimal choices:

- low source impedance: bipolar opamps

1..5 nV/√Hz, 1..4pA/√Hz

- high source impedance: low noise FETs

3..8 nV/√Hz, 1..10fA/√Hz

- matched, precision transistors

<1nV/√Hz, 1..4pA/√Hz

3-39
3.3.2. Bandwidth considerations

S(f) -> <x2>, RMS

For white noise:

<x2>=So BW=So (f2-f1)

For Lorentzian noise:

<x2>=So BW π/2

3-40
3.3.3. External noise sources

- radio, TV broadcast

- 50/60Hz power lines

- lightning

- computers, monitors

- electric motors

- ignition

How they coupled to our system?

- capatively: dV/dt -> noise current

- 1V/ns -> 1mA/pF

- inductively: di/dt -> noise voltage

- 1mA/ns -> 1mV/nH

3-41
- thermal effects (thermocouples)

- parasitic resistances, ground loops

- microphonics: dC/dt (cables,capacitors)

- leakage currents (PCB, air)

- long term changes (aging)

3-42
3.3.4. Noise reduction techniques

3.3.4.1. Limiting bandwidth

The RMS value of noise is a function of the

bandwidth -> reduction possibility

Example:

audio systems use 20Hz..20kHz

total noise with 10nV/√Hz:

3-43
3.3.4.2. Paralelling systems

- One signal processed by two independent

amplifiers

The output signal:

3-44
RMS value:

Cross spectrum of the two signals:

Vs+V1 and Vs+V2

If Vs, V1 and V2 are uncorrelated:

Sxy(f)=Sss(f)

3-45
3.3.4.3. Reducing source and other

impedances

Any impedance is a thermal noise source:

S(f)=4kTR

S(f)=4kTRe(X)

The input current noise of amplifiers:

S(f)=Sc(f)R

-> use as low values as possible in:

- source

- signal processing system

3-46
3.3.4.4. Using lock-in techniques

Very small bandwidth -> very small noise

- measurement of a periodic component

- AC excitations of a bridge

Lock-in amplifier: extremely narrow band

amplifier

Example:

- 10nV, 10kHz signal, 5nV/√Hz preamp

noise, 100kHz bandwidth, A=1000

SNR=10µV/1.6mV=0.00625

3-47
Using a very good bandpass filter:

10kHz, 100Hz BW (Q=100)

SNR=10µV/50µV=0.2

Using a lock-in amplifier:

10kHz, 0.01Hz BW (Q=105)

SNR=10µV/0.5µV=20

3-48
- analog realizations

- digital realizations (DSP)

3-49
3.3.5. Reducing external noise

Reducing capacitively coupled noise:

- reduce sources of high dV/dt

- proper grounding for cable shields

- reduce stray capacitance

- use grounded conductive Faraday shields

3-50
Reducing inductively coupled noise:

- careful routing of wiring

- conductive screens against HF magnetic field

3-51
- high permeability metals for LF fields

- use twisted pairs of wire

Reducing resistively coupled noise:

- remove large currents from signal paths

- ground to the same point (star grounding)

- use heavy ground plane

3-52
For long transmission lines:

- differential drivers/receivers

- current transmitters

- digitize first

3-53
3.3.6. Noise shaping

Before reducing bandwidth, shape the noise:

- move to outband, if possible

- non-linear transform required

Example:

reduction of quantization noise (Σ∆ A/D)

- quantization noise (QN): white noise

- oversampling: (note: fmax=fs/2)

<QN2>=const -> S(f) (fmax-fmin)=const

increasing fmax -> reducing QN in the ∆f of

interest

- noise shaping to higher frequencies: further

reduction

3-54
3-55
Σ∆ A/D converter:

- 1-bit A/D: comparator

- noise shaping

3-56