Fundamentals of DSP: Introduction To Digital Signal Processing

Fundamentals of DSP
Part 1
Introduction to Digital Signal Processing
Unrestricted © Siemens AG 2017

Fundamentals of Digital Signal Processing
Content
Part 1: What is a signal?

Time and frequency domain
• Fourier transformation
Part 2: Digitizing signals
• Sampling
• Aliasing
Part 3: Effects to be aware of when
converting to digital
• Quantization
• Leakage
Counter-measures to ensure
your digital data is valid

Page 2 Siemens PLM Software
Signals
Signal: measurable quantity carrying information about some physical

phenomenon
• Pressure, displacement, acceleration, …
• Temperature, voltage, biomedical potential (EKG, EEG, ...)
The signal is generated by a structure and detected by a sensor or transducer
• Accelerometer: acceleration  voltage Analog signal

• Microphone: pressure  voltage
• Strain Gauge: strain (deformation)  voltage
• Thermocouple: temperature changes  voltage
The signal is what you want to analyse in view of a particular problem

Signals

Signal Processing
Signal processing: specific manipulations of the

measured signal to
• Extract key information

• Understand physics
• Provide input data for specific analysis
• Confront simulation results with reality Time
• Modify the signal for specific applications domain
Signal processing transforms the signal to different domains
• Time domain
• Frequency domain
• Amplitude domain
Frequency Amplitude
• Laplace domain
domain domain
• …
Time, frequency and amplitude domains
Time
domain
Each domain = different coordinate Each domain highlights a
system that is used to view or particular aspect of the
describe the characteristics of a characteristics of a system
system or event or event
Frequency Transformation
Amplitude
domain domain
• The time domain is usually the basis for a description of a system’s dynamic behavior.
e.g. differential equation of motion. Events are measured as a function of time
• The frequency domain highlights the periodic characteristics of the system or event
• The amplitude domain represents looks at the probability distribution of the amplitudes

Fourier transformation
Joseph did help us a lot…
4
-1
Joseph Fourier -2
(º1768 - †1830)
Time
-3
Théorie analytique de -4
la chaleur (1822)
domain
2.5
Fourier’s law of heat

2
1.5
conduction 0.5
-0.5
u   2u  2u 
-1
-1.5
k  2  2 
-2
-2.5
t  x y  Any signal can be described as a combination of sine

Analysed in terms of infinite waves of different frequencies
mathematical series Frequency
domain
For mathematicians … 
X     x t  e  j t
dt
• Convert from time to frequency domain

and back 
xt   
1
• Fourier integral
  d
j t
 X e
• No information is lost when converting! 2 
For engineers … Time Frequency

domain domain
4
2.5
3 2
1.5
2
1
1
0.5
0 0
-1 -0.5
f [Hz]
-1
-2
-1.5
-3
-2
-4 -2.5
Detect sine waves Draw line at sine frequency

Some definitions of sine waves
Time domain
T0 1 rad
t [s]
Period: T0 [s]
 2
Frequency domain
f [Hz]
Frequency: f0 = 1/T0 [Hz]
f0
 [rad/s]
Pulsation / circular frequency: 0 = 2f0 = 2/T0 [rad/s]
0

Basics of sine waves
Sine Wave Equation x(t )  A sin( 2ft   )

Amp A = Amplitude
f = Frequency
θ = Phase
time
t = Time
  2f
1 second
f is in Hz
ω is in radians/sec

Frequency
2 Hz 4 Hz
Amp Amp
time time
1 second 1 second

Amplitude
1g 0.5 g
Amp Amp
time time
1 second 1 second

Amplitude
5 g Peak
5
3.5 g RMS (.707 of Peak)
Amp
0
time
-5
10 g Peak-to-Peak (2xPeak)
Scaling can cause amplitude difference!

Phase
Phase is the amount of shift in time relative to

another reference
(another signal, start of FFT block, etc.)
Amp
0
time
Phase is measured as an angle
Orange signal “lags” the green by about
45° or /4 radians
Green signal “leads” the orange by about
315° or 7/4 radians
x(t )  A sin( 2ft   )

Q: What is the output of a Fourier transformation?
A: A spectrum in the frequency domain. It represents a series of sines and

cosines in the form of complex numbers. When these numbers are
summed, they form the original signal in the time domain.
Solution: a + jb = complex number

a = real part
b = imaginary part
j = -1
Note: You will get a complex number for each point in your spectrum

Frequency spectrum
Complex numbers
Q: How do I make sense of this complex numbo-jumbo?
A: Spectrum most commonly viewed as Magnitude/Phase
Log
Magnitude
Phase
Phase
Hz
AKA: Bode Plot

Frequency spectrum
Time history
Selection of domain depends on the application aims

Equivalence of time and frequency domain: no loss of information
f
Time Frequency Time Frequency

Signals

To be continued….
PART 2: Digitizing signals

• Sampling
• Aliasing
PART 3: Effects to be aware of when

• Quantization
• Leakage
Thank you Counter-measures to ensure

Fundamentals of DSP
Part 2

Content

• Sampling
• Aliasing
• Quantization
• Leakage

Nice theory…
but we need to do this on a computer
Digital Signal Processing: apply manipulations using a
computer-based system
• Most transducers output an analog (continuous) signal

• Computers are digital devices (0/1; on/off)
• Convert the sensor signal into a discrete stream of digital
information
• Discretization in time and in amplitude
• Massive loss of information when sampling

Sampling analog signals
10 Hz sine wave, sampled at 512 Hz: digital representation looks like a perfect sine

10 Hz sine wave, sampled at 128 Hz: digital representation still looks OK

10 Hz sine wave, sampled at 64 Hz: digital representation starts looking strange…

10 Hz sine wave sampled high (512 Hz, red) and low (64 Hz, green) on a digital oscilloscope

Exploring the limits…
Sampling frequency = sine wave frequency Sampling frequency = 2 x sine wave frequency
fs = fsine fs = 2 x fsine
Observed frequency = 0 Hz (DC) Observed frequency is correct, but it is borderline

(sampling frequency cannot be lowered)

When pushing further, you get aliasing
Sine wave frequency = 20 Hz

Sample rate 21.3 Hz
Observed frequency is wrong: 20 Hz sine wave sampled at 21.3 shows as 1.3 Hz signal

When pushing further, you get aliasing
Amplitude Observed Correct
Aliasing
Sample rate
0 1.3 20 21.3
Frequency
fs/2 Nyquist
“Observed” Usable
frequency
Information to be discarded frequency
frequencies No usable data
range
True fs
frequencies f max 
0 fs/2 fs 2fs 3fs 2

How to prevent aliasing?
Select sample rate to cover full signal

bandwidth
If there is no frequency content above Nyquist Low pass filter to prevent aliasing:
Frequency, then there is no Aliasing Removes frequencies that violate
This is not always practical or possible: Nyquist from analog signal before
Amplitude
• Large files sizes digitizing
• Limitations of data acquisition equipment
1
Limit signal bandwidth using Anti-Aliasing Anti-aliasing filter

Filter Ideal case
• Analog and/or Digital low-pass filters
0
Frequency fs/2=fmax
How to prevent aliasing?
Low pass filter

Extremely sharp toanalog
prevent aliasing:
filter, but
Removes frequencies
“brick wall” that
effect is not violate
possible
Make sure the signal does not contain Nyquist from analog
Roll off point starts @signal
80% of before
Amplitude
frequencies above half the sample frequency digitizing
bandwidth
Do this by applying a sufficient performing 1

low-pass filter
Limit signal bandwidth using Anti-Aliasing Anti-aliasing filter

Be aware that the amplitude of the last
Filter filterIdeal case
practical
portion of the spectrum is
• Analog and/or Digital low-pass filters
80% of fmax
Automatically
• attenuated done infilter
by the anti-alias good data
acquisition
• contaminated hardware
by out-of-band signal 0
Frequency fs/2=fmax
Aliasing demonstration
Sine sweep from 45 to 82 Hz, sampled at 128 Hz

To be continued….
PART 3: Effects to be aware of when

• Quantization
Thank you • Leakage

Fundamentals of DSP
Part 3

Content

• Sampling
• Aliasing
• Quantization
• Leakage

Digitizing analog signals
Quantization resolution
Sampling in time domain:

Store an amplitude value
on a periodic basis (fs)
Amplitude is determined
with a discrete resolution
Real values are rounded

or truncated to discrete
levels
This process causes

quantization noise

Digitizing analog signals
Some formulas
• Assume ADC with M bits at the output

• The number of voltage intervals N is given by
𝑁 = 2𝑀 − 1 M N ΔV SQNR
• The voltage resolution of an ADC is equal to # of bits # of voltage Voltage Signal to
its overall voltage measurement range steps resolution quantification
divided by the number of discrete values:
(for +/- 10V range) noise ratio
4 16 1.24 V 24 dB
𝑉𝑟𝑎𝑛𝑔𝑒
∆𝑉 = 8 255 78.1 mV 48 dB
2𝑀
12 4 095 4.88 mV 72 dB
• The Signal-to-quantization-noise ratio is given
16 65 535 0.305 mV 96 dB
by
24 16 777 215 1.19 µV 144 dB
𝑆𝑄𝑁𝑅 = 20 𝑙𝑜𝑔10 2𝑀 ≈ 6.02 ∗ 𝑀 𝑑𝐵

Quantization demonstration



Mitigating quantization errors
Some terminology
+V Headroom or
Overhead
Input range
or
Full scale range
or Time
ADC range
 Analog Digital 
-V

Mitigating quantization errors
Demonstration: 10 Hz, 2 mV sine wave
16 bit ADC, 10 V input range

 0.305 mV discretization level
2 mV signal / 0.305 mV = 13
 truncated to 13 possible levels
 Large quantization errors!
2 possible solutions:
Use ADC with higher number
of bits
Apply gain to the signal prior
to ADC

Select the right range,
but watch out for overloads!

We don’t have all day…
Finite observation period
Periodic observation: correct amplitude level

at correct spectral line
Fourier

If signal is not periodic in observation window
 Leakage
Periodic observation: correct amplitude level

at correct spectral line
A-periodic observation: up to 63% amplitude

error at spectral line closest to correct
frequency
Remaining 37% amplitude spread out over
entire frequency spectrum
Fourier
Leakage = severe distortion of spectrum

if the signal is not periodic in the
observation window

Leakage demonstration
Slow sweep from 10 to 11 Hz, showing spectrum with 0.5H resolution

Effect of finite observation time
Sine
frequency
10.7 Hz
Leakage when signal is not periodic Leakage when sine wave frequency
within the observation window falls between the spectral lines
Why is this happening?
Periodicity assumption
Observation window is assumed to repeat

itself, introducing ‘spikes’ in the signal
Fourier

Reduce the effect of leakage
‘Force’ the input for the Fourier

transformation to be periodically expandable
Practical implementation: multiply signal with
time domain window to eliminate
discontinuities
Effects of time window:

Fourier
Improved amplitude estimate

 flatten central lobe
Reduce frequency range of smearing
 lower side lobes
Local smearing of spectral energy due
to wider central lobe
 lower effective spectral resolution

Window types
most popular ones
Rectangular, uniform Hanning Flat top

Time domain
AKA
No Window
Freq. domain

Thank you

Fundamentals of DSP: Introduction To Digital Signal Processing

Uploaded by

Copyright:

Available Formats

Fundamentals of DSP: Introduction To Digital Signal Processing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamentals of DSP: Introduction To Digital Signal Processing

Uploaded by

Copyright:

Available Formats

What are the different domains that signals can be transformed to?

What are the different domains that signals can be transformed to?

What is the difference between the time domain and frequency domain?

What is the difference between the time domain and frequency domain?

Fundamentals of DSP

Unrestricted © Siemens AG 2017

Part 1: What is a signal?

Unrestricted © Siemens AG 2017

Signal: measurable quantity carrying information about some physical

The signal is generated by a structure and detected by a sensor or transducer

• Accelerometer: acceleration  voltage Analog signal

The signal is what you want to analyse in view of a particular problem

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

Signal processing: specific manipulations of the

• Extract key information

Signal processing transforms the signal to different domains

Unrestricted © Siemens AG 2017

Fourier’s law of heat

t  x y  Any signal can be described as a combination of sine

For engineers … Time Frequency

Detect sine waves Draw line at sine frequency

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

Sine Wave Equation x(t )  A sin( 2ft   )

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

Scaling can cause amplitude difference!

Unrestricted © Siemens AG 2017

Phase is the amount of shift in time relative to

Unrestricted © Siemens AG 2017

Q: What is the output of a Fourier transformation?

A: A spectrum in the frequency domain. It represents a series of sines and

Solution: a + jb = complex number

Unrestricted © Siemens AG 2017

Q: How do I make sense of this complex numbo-jumbo?

A: Spectrum most commonly viewed as Magnitude/Phase

AKA: Bode Plot

Selection of domain depends on the application aims

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

PART 2: Digitizing signals

PART 3: Effects to be aware of when

Thank you Counter-measures to ensure

Unrestricted © Siemens AG 2017

Part 1: What is a signal?

Unrestricted © Siemens AG 2017

• Most transducers output an analog (continuous) signal

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

10 Hz sine wave, sampled at 128 Hz: digital representation still looks OK

Unrestricted © Siemens AG 2017

10 Hz sine wave, sampled at 64 Hz: digital representation starts looking strange…

Unrestricted © Siemens AG 2017

Unrestricted © Siemens AG 2017

Observed frequency = 0 Hz (DC) Observed frequency is correct, but it is borderline

Unrestricted © Siemens AG 2017

Sine wave frequency = 20 Hz