Analog & digital signals

Analog Digital
Continuous function V of Discrete function Vk of
continuous variable t (time, discrete sampling variable tk,
space etc) : V(t). with k = integer: Vk = V(tk).

0.3 0.3
0.2 0.2
Voltage [V]

Voltage [V]
0.1 0.1
0 0
-0.1 -0.1 ts ts
-0.2 -0.2
0 2 4 6 8 10 0 2 4 6 8 10
time [ms] sampling time, tk [ms]
Uniform (periodic) sampling.
Sampling frequency fS = 1/ tS

Slides adapted from ME Angoletta, CERN

 Digital vs analog proc’ing
Digital Signal Processing (DSPing)

Advantages Limitations

• More flexible. • A/D & signal processors speed:

wide-band signals still difficult to
• Often easier system upgrade. treat (real-time systems).
• Data easily stored. • Finite word-length effect.
• Better control over accuracy • Obsolescence (analog
requirements. electronics has it, too!).
• Reproducibility.

Slides adapted from ME Angoletta, CERN

 Digital system example
V

DOMAIN
ANALOG
General scheme ms Filter
V Antialiasing

Sometimes steps missing

ms
- Filter + A/D A
A/D

DOMAIN
DIGITAL
(ex: economics);
k
- D/A + filter Digital
A
Processing
(ex: digital output wanted).
k
V
D/A

DOMAIN
ANALOG
ms
V Filter
Reconstruction
ms

Slides adapted from ME Angoletta, CERN

Digital system implementation
KEY DECISION POINTS:
ANALOG INPUT
Analysis bandwidth, Dynamic range

Antialiasing
Filter • Pass / stop bands.
1
• Sampling rate.
A/D
• No. of bits. Parameters. 2
Digital
Processing • Digital format. 3
What to use for processing?
See slide “DSPing aim & tools”
DIGITAL OUTPUT

Slides adapted from ME Angoletta, CERN

 1 Sampling
How fast must we sample * a continuous
signal to preserve its info content?

Ex: train wheels in a movie.

25 frames (=samples) per second.

Train starts wheels ‘go’ clockwise.

Train accelerates wheels ‘go’ counter-clockwise.

Why?
Frequency misidentification due to low sampling frequency.

* Sampling: independent variable (ex: time) continuous → discrete.

Quantisation: dependent variable (ex: voltage) continuous → discrete.
Here we’ll talk about uniform sampling.

Slides adapted from ME Angoletta, CERN

 1 Sampling - 2
1.2

1
__ s(t) = sin(2πf t)
0.8 0
0.6

0.4
s(t) @ fS
0.2

0
f0 = 1 Hz, fS = 3 Hz
-0.2 tt
-0.4

-0.6

-0.8
__ s (t) = sin(8πf t)
1 0
-1

-1.2
__ s (t) = sin(14πf t)
2 0

s(t) @ fS represents exactly all sine-waves sk(t) defined by:

sk (t) = sin( 2π (f0 + k fS) t ) , ⏐k ⏐∈

Slides adapted from ME Angoletta, CERN

 1 The sampling theorem
A signal s(t) with maximum frequency fMAX can be
Theo* recovered if sampled at frequency f > 2 f
S MAX .

* Multiple proposers: Whittaker(s), Nyquist, Shannon, Kotel’nikov.

Naming gets
confusing ! Nyquist frequency (rate) fN = 2 fMAX or fMAX or fS,MIN or fS,MIN/2

Example
s(t) = 3 ⋅ cos(50 π t) + 10 ⋅ sin(300 π t) − cos(100π t) Condition on fS?
F1 F2 F3

F1=25 Hz, F2 = 150 Hz, F3 = 50 Hz fS > 300 Hz

fMAX

Slides adapted from ME Angoletta, CERN

 1 Frequency domain (hints)
• Time & frequency:
frequency two complementary signal descriptions.
Signals seen as “projected’ onto time or frequency domains.
Example
Ear + brain act as frequency analyser: audio spectrum
split into many narrow bands low-power sounds
detected out of loud background.

• Bandwidth:
Bandwidth indicates rate of change of a signal.
High bandwidth signal changes fast.

Slides adapted from ME Angoletta, CERN

1 Sampling low-pass signals
Continuous spectrum
(a) (a) Band-limited signal:
frequencies in [-B, B] (fMAX = B).

-B 0 B f

(b) Discrete spectrum

No aliasing
(b) Time sampling frequency
repetition.
fS > 2 B no aliasing.
-B 0 B fS/2 f

Discrete spectrum
(c) Aliasing & corruption
(c) fS 2B aliasing !

Aliasing: signal ambiguity

0 fS/2 f
in frequency domain

Slides adapted from ME Angoletta, CERN

 1 Antialiasing filter
(a) Signal of interest (a),(b) Out-of-band noise can alias
Out of band
Out of band into band of interest. Filter it before!
noise
noise

-B 0 B f
(c) Antialiasing filter
(b) Passband: depends on bandwidth of
interest.

Attenuation AMIN : depends on

-B 0 B fS/2 • ADC resolution ( number of bits N).
(c) f
AMIN, dB ~ 6.02 N + 1.76
Antialiasing
• Out-of-band noise magnitude.
Passband filter
frequency
Other parameters: ripple, stopband
frequency...
-B 0 B f

Slides adapted from ME Angoletta, CERN

 2 ADC - Number of bits N
Continuous input signal digitized into 2N levels.
1113
Uniform, bipolar transfer function (N=3)
2

V FSR
1 Quantization step q =
0 2N
-4 -3 -2 -1 0 1 2 3 4
-1 V Ex: VFSR = 1V , N = 12 q = 244.1 µV
010
-2

001
-3 Voltage ( = q)
000
-4
VFSR LSB Scale factor (= 1 / 2N )
1

0.5
q/2
Percentage (= 100 / 2N )
0

-4 -3 -2 -1 0 1 2 3 4

Quantisation error
-0.5

-q/2
-1

Slides adapted from ME Angoletta, CERN

 2 ADC - Quantisation error
0.3

0.2 • Quantisation Error eq in

[-0.5 q, +0.5 q].
Voltage [V]

0.1

0 • eq limits ability to resolve

0 2 4 6 8 10 small signal.
-0.1

-0.2
• Higher resolution means
time [ms]
lower eq.

Slides adapted from ME Angoletta, CERN

 Frequency analysis: why?
• Fast & efficient insight on signal’s building blocks.
• Simplifies original problem - ex.: solving Part. Diff. Eqns. (PDE).
• Powerful & complementary to time domain analysis techniques.
• The brain does it?

General Transform as
problem-solving tool
analysis
time, t frequency, f
F
s(t), S(f) :
s(t) S(f) = F[s(t)] Transform Pair
synthesis

Slides adapted from ME Angoletta, CERN

 Fourier analysis - tools
2.5
Input Time Signal Frequency spectrum
2

1.5

1
T
1
Periodic c k = ⋅ ∫ s(t) ⋅ e − j k ω t dt
0.5

FS Discrete
0
0 1 2 3 4 5 6 7 8

time, t T
(period T) 0
2.5
Continuous − j2 π f t
+∞
FT S(f) = ∫ s(t) ⋅ e
2

1.5

1
Aperiodic Continuous −∞
dt
0.5

0
0 2 4 6 8 10 12

time, t

N −1 2πkn
−
DFS** Discrete
j
2.5

~ 1
ck = ∑ s[n] ⋅ e N
2

Periodic
1.5

1
N
0.5
n =0
0
0 1 2 3

time, tk
4 5 6 7 8
(period T)
+∞
Discrete S(f) = ∑ s[n] ⋅ e− j 2 π f n
2.5 DTFT Continuous
n= −∞
Aperiodic
2

2πkn
1.5

DFT** Discrete 1 N−1 −j

1

0.5

0
~
ck = ∑ s[n] ⋅ e N
time, tk
0 2 4 6 8 10 12

N
n =0

Note: j =√-1, ω = 2π/T, s[n]=s(tn), N = No. of samples Calculated via FFT

**
Slides adapted from ME Angoletta, CERN
 A little history
¾ Astronomic predictions by Babylonians/Egyptians likely via trigonometric sums.

¾ 1669:
1669 Newton stumbles upon light spectra (specter = ghost) but fails to
recognise “frequency” concept (corpuscular theory of light, & no waves).

¾ 18th century:
century two outstanding problems
→ celestial bodies orbits: Lagrange, Euler & Clairaut approximate observation data
with linear combination of periodic functions; Clairaut,1754(!) first DFT formula.
→ vibrating strings: Euler describes vibrating string motion by sinusoids (wave
equation).

¾ 1807:
1807 Fourier presents his work on heat conduction ⇒ Fourier analysis born.
→ Diffusion equation ⇔ series (infinite) of sines & cosines. Strong criticism by peers
blocks publication. Work published, 1822 (“Theorie Analytique de la chaleur”).

Slides adapted from ME Angoletta, CERN

 A little history -2
¾ 19th / 20th century:
century two paths for Fourier analysis - Continuous & Discrete.

CONTINUOUS
→ Fourier extends the analysis to arbitrary function (Fourier Transform).
→ Dirichlet, Poisson, Riemann, Lebesgue address FS convergence.
→ Other FT variants born from varied needs (ex.: Short Time FT - speech analysis).

DISCRETE: Fast calculation methods (FFT)

→ 1805 - Gauss, first usage of FFT (manuscript in Latin went unnoticed!!!
Published 1866).
→ 1965 - IBM’s Cooley & Tukey “rediscover” FFT algorithm (“An algorithm for
the machine calculation of complex Fourier series”).
→ Other DFT variants for different applications (ex.: Warped DFT - filter design &
signal compression).
→ FFT algorithm refined & modified for most computer platforms.

Slides adapted from ME Angoletta, CERN

 Fourier Series (FS)
A periodic function s(t) satisfying Dirichlet’s conditions * can be expressed
as a Fourier series, with harmonically related sine/cosine terms.

sis a0, ak, bk : Fourier coefficients.

he +∞
n s(t) = a0 + ∑ [ak ⋅ cos (k ω t) − bk ⋅ sin (k ω t)]
t
k: harmonic number,
sy
k =1 For all t but discontinuities
T: period, ω = 2π/T
s
y si
n al T
1
a a = ⋅ s(t)dt (signal average over a period, i.e. DC term &
T ∫
0 zero-frequency component.)
0
T
2
ak = ⋅ ∫ s(t) ⋅ cos(k ω t) dt
T Note: {cos(kωt), sin(kωt) }k
0
form orthogonal base of
T
2 function space.
- bk = ⋅ ∫ s(t) ⋅ sin(k ω t) dt
T
0 * see next slide

Slides adapted from ME Angoletta, CERN

 FS convergence
Dirichlet conditions
(a) s(t) piecewise-continuous;

In any period: (b) s(t) piecewise-monotonic;

T
(c) s(t) absolutely integrable , ∫ s(t) dt < ∞
0

T Rate of convergence
Example:
if s(t) discontinuous then
square wave
|ak|<M/k for large k (M>0)

s(t)

(a) (b) (c)

Slides adapted from ME Angoletta, CERN

 FS analysis - 1
2π
1.5
FS of odd* function: square wave.

square signal, sw(t)

1
T = 2π ⇒ ω = 1
0.5
⎧π 2π ⎫
1 ⎪ ⎪
(zero average)
0
a0 = ⋅ ⎨ ∫ dt + ∫ ( −1)dt ⎬ = 0
2π ⎪ ⎪⎭ 0 2 4 6 8 10
t
⎩0 π -0.5
⎧π 2π ⎫
1 ⎪ ⎪ -1
ak = ⋅ ⎨ ∫ cos kt dt − ∫ cos kt dt ⎬ = 0
π ⎪
(odd function)
⎩0 π ⎪⎭ -1.5

⎧π 2π ⎫
1 ⎪ ⎪
- bk = ⋅ ⎨ ∫ sin kt dt − ∫ sin kt dt ⎬ = ... =
2
⋅ { 1− cos kπ } = * Even & Odd functions
π ⎪ ⎪⎭ k ⋅ π
⎩0 π s(x)

⎧ 4 Even :
⎪ k ⋅ π , k odd
⎪ s(-x) = s(x)
= ⎨ x
⎪ 0 , k even
⎪
⎩
s(x)

Odd :
4 4 4
sw(t) = ⋅ sin t + ⋅ sin 3 ⋅ t + ⋅ sin 5 ⋅ t + ... x
π 3⋅π 5⋅π s(-x) = -s(x)

Slides adapted from ME Angoletta, CERN

 FS synthesis
Square wave reconstruction
from spectral terms
1.5

square signal, sw(t)

15911
7
3 0.5

sw1 (t)===∑
(t)
9(t)
7
3
5
11 ∑∑[[--[b-bkbkk⋅⋅sin(kt)
⋅sin(kt)
sin(kt) ]]]]
sin(kt)
0
kkk==1=11
-0.5

-1

-1.5
0 2 4 6 8 10
t

Convergence may be slow (~1/k) - ideally need infinite terms.

Practically, series truncated when remainder below computer tolerance
error BUT … Gibbs’ Phenomenon.
(⇒ error).

Slides adapted from ME Angoletta, CERN

 Gibbs phenomenon
1.5

Overshoot exist @ 1

each discontinuity

square signal, sw(t)

0.5

0
79
sw 79 (t) = ∑ [- bk ⋅ sin(kt)] -0.5

k =1
-1

-1.5
0 2 4 6 8 10
t

• First observed by Michelson, 1898. Explained by Gibbs.

• Max overshoot pk-to-pk = 8.95% of discontinuity magnitude.
Just a minor annoyance.
• FS converges to (-1+1)/2 = 0 @ discontinuities, in this case.
case

Slides adapted from ME Angoletta, CERN

 FS time shifting
FS of even function: 1.5
2π
π/2-advanced square-wave

square signal, sw(t)

1

0.5

0
a 0= 0 (zero average) 0 2 4 6 8 10
t
-0.5

⎧ 4 -1
⎪ k ⋅ π , k odd, k = 1, 5, 9...
⎪ -1.5
⎪
ak = ⎨ − 4 , k odd, k = 3, 7, 11...
⎪ k⋅π rk
⎪ 4/π
⎪

e
⎩ 0 , k even.

ud
it
4/3π

pl
- bk = 0 (even function) am
θk f1 3f1 5f1 7f1 f

π
Note: amplitudes unchanged BUT
phases advance by k⋅π/2.
e
as
ph

f1 3f1 5f1 7f1 f

Slides adapted from ME Angoletta, CERN

 Complex FS
Euler’s notation:
e jt + e − jt e jt − e − jt
e-jt = (ejt)* = cos(t) - j·sin(t) “phasor” cos(t) = sin(t) =
2 2⋅ j
s T
y si 1
a c k = ⋅ s(t) ⋅ e - j k ω t dt
l
an
T ∫
0
Complex form of FS (Laplace 1782). Harmonics
sis ck separated by ∆f = 1/T on frequency plot.
the ∞
jk ω t
n
sy s(t) = ∑ kc ⋅ e
k = −∞
jθ
Note:
Note c-k = (ck)* z=re
Link to FS real coeffs. r r = a2 + b2
b
θ θ = arctan(b/a)
c 0 = a0
a
1 1
ck = ⋅ (ak + j ⋅ bk ) = ⋅ (a −k − j ⋅ b −k )
2 2

Slides adapted from ME Angoletta, CERN

 FS properties
Time Frequency
Homogeneity a·s(t) a·S(k)

Additivity s(t) + u(t) S(k)+U(k)

Linearity a·s(t) + b·u(t) a·S(k)+b·U(k)

Time reversal s(-t) S(-k)

∞
Multiplication * s(t)·u(t) ∑ S(k − m)U(m)
T m = −∞
1
Convolution * ⋅ ∫ s(t − t ) ⋅ u( t ) dt S(k)·U(k)
T
0
2π k ⋅t
s(t − t ) −j
Time shifting e T ⋅ S(k)
2π m t
+j
Frequency shifting e T ⋅ s(t) S(k - m)
*

Slides adapted from ME Angoletta, CERN

 FS - “oddities”
Orthonormal base
Fourier components {uk} form orthonormal base of signal space:
T
*
∫
uk = (1/√T) exp(jkωt) (|k| = 0,1 2, …+∞) Def.: Internal product ⊗: uk ⊗ um = uk ⋅ um dt
o
uk ⊗ um = δk,m (1 if k = m, 0 otherwise). (Remember (ejt)* = e-jt )

Then ck = (1/√T) s(t) ⊗ uk i.e. (1/√T) times projection of signal s(t) on component uk

Negative frequencies & time reversal

k = - ∞, … -2,-1,0,1,2, …+ ∞, ωk = kω, φk = ωkt, phasor turns anti-clockwise.

Negative k ⇒ phasor turns clockwise (negative phase φk ), equivalent to negative time t,

⇒ time reversal.
Careful:
Careful phases important when combining several signals!

Slides adapted from ME Angoletta, CERN

 FS - power
T
1
Average power W : W = ∫ s(t) 2 dt ≡ s(t) ⊗ s(t)
T
o
Parseval’s Theorem • FS convergence ~1/k
∞ 1 ∞ ⇒ lower frequency terms
2
W= ∑ ck = a0 2 +
2
∑ ⎛⎜⎝ ak 2 + bk 2 ⎞⎟⎠ Wk = |ck|2 carry most power.
k = −∞ k =1
• Wk vs. ωk: Power density spectrum.
spectrum
Example 2 Wk/W0
1
Pulse train, duty cycle δ = 2 τ / T
10
-1
Wk = 2 W0 sync2(k δ)
s(t) 2τ

T 10
-2

10
-3 kf
t 0 50 100 150 200
bk = 0 a0 = δ sMAX W0 = (δ sMAX)2 ⎧⎪ ∞ W ⎫
⎪
W = W0 ⋅ ⎨1+ ∑ k ⎬
ak = 2δsMAX sync(k δ) sync(u) = sin(π u)/(π u) ⎪⎩ k =1 W0 ⎪⎭

Slides adapted from ME Angoletta, CERN

FS of main waveforms

Slides adapted from ME Angoletta, CERN

 Discrete Fourier Series (DFS)
Band-limited signal s[n], period = N. DFS generate periodic ck
with same signal period
DFS defined as:
is Orthogonality in DFS:
y s
N −1 2π k n
a l
1 − j
an ~ ck = ∑ s[n] ⋅ e N N −1 j
2π n(k -m)
N 1
n =0 N
∑ e N = δ k,m
~ ~ n =0
Note: ck+N = ck ⇔ same period N
i.e. time periodicity propagates to frequencies! Kronecker’s delta

s
esi
th N−1 2π k n
n j N consecutive samples of s[n]
sy s[n] = ~
∑
ck ⋅ e N completely describe s in time
k =0 or frequency domains.
Synthesis: finite sum ⇐ band-limited s[n]

Slides adapted from ME Angoletta, CERN

 DFS analysis
DFS of periodic discrete 1
s[n]

1-Volt square-wave

s[n]: period N, duty factor L/N -5 0 1 2 3 4 5 6 7 8 9 10 n

0 L N

e
ud
1 1
⎧ L ~

it
⎪ , k = 0, + N, ± 2N,... 0.6 0.6 ck 0.6 0.6

pl
N
⎪

am
⎪ 0.24 0.24 0.24 0.24
⎪ 0.2
~
ck = ⎨ π k (L −1) ⎛ π kL ⎞
⎪ −j sin ⎜ ⎟
⎪e N ⎝ N ⎠ 0 1 2 3 4 5 6 7 8 9 10 k
⎪ ⋅ , otherwise
N ⎛π k⎞
⎪ sin ⎜ ⎟
⎝ N ⎠ θk

e
⎩ 0.4π

as
0.4π

ph
0.2π 0.2π

Discrete signals ⇒ periodic frequency spectra.

0 2 4 5 6 7 8 9 10 n
Compare to continuous rectangular function
(slide # 10, “FS analysis - 1”) -0.2π -0.2π

-0.4π -0.4π

Slides adapted from ME Angoletta, CERN

 DFS properties
Time Frequency
Homogeneity a·s[n] a·S(k)

Additivity s[n] + u[n] S(k)+U(k)

Linearity a·s[n] + b·u[n] a·S(k)+b·U(k)

1 N−1
Multiplication * s[n] ·u[n] ⋅ ∑ S(h)U(k - h)
N h=0
N−1
Convolution * ∑ s[m] ⋅ u[n − m] S(k)·U(k)
m =0
2π k ⋅m
Time shifting s[n - m] −j
T
e ⋅ S(k)
2π h t
Frequency shifting +j S(k - h)
e T ⋅ s[n]

Slides adapted from ME Angoletta, CERN

 DFT – Window characteristics
• Finite discrete sequence ⇒ spectrum convoluted with rectangular window spectrum.
• Leakage amount depends on chosen window & on how signal fits into the window.

(1) Resolution: capability to distinguish different tones. Inversely proportional to main-

lobe width. Wish: as high as possible.

(2) Peak-sidelobe level: maximum response outside the main lobe.

Determines if small signals are hidden by nearby stronger ones.
(1)
Wish: as low as possible. (3)
(2)
(3) Sidelobe roll-off: sidelobe decay
per decade. Trade-off with (2).

Several windows used (application-

dependent): Hamming, Hanning, Rectangular window
Blackman, Kaiser ...

Slides adapted from ME Angoletta, CERN

 DFT of main windows
Windowing reduces leakage by
minimising sidelobes magnitude.

In time it reduces end-

points discontinuities.

Sampled sequence

Non
windowed

Windowed

Some window functions

Slides adapted from ME Angoletta, CERN

 DFT - Window choice
Common windows characteristics
Window type -3 dB Main- -6 dB Main- Max sidelobe Sidelobe roll-off
lobe width lobe width level [dB/decade]
[bins] [bins] [dB]
Rectangular 0.89 1.21 -13.2 20

Hamming 1.3 1.81 - 41.9 20

Hanning 1.44 2 - 31.6 60

Blackman 1.68 2.35 -58 60

Observed signal Window wish list

Far & strong interfering components ⇒ high roll-off rate.
Near & strong interfering components ⇒ small max sidelobe level.
Accuracy measure of single tone ⇒ wide main-lobe

NB: Strong DC component can shadow nearby small signals. Remove it!

Slides adapted from ME Angoletta, CERN

 DFT - Window loss remedial
Smooth data-tapering windows cause information loss near edges.

2 x N samples (input signal)

Solution:
sliding (overlapping) DFTs. DFT #1

• Attenuated inputs get next DFT #2

window’s full gain & leakage
reduced.
DFT #3
• Usually 50% or 75% overlap
(depends on main lobe width).

Drawback: increased
DFT AVERAGING
total processing time.

Slides adapted from ME Angoletta, CERN

DFT - parabolic interpolation
Rectangular window Hanning window
1.968 0.977

1.967

1.966 0.976

1.965

1.964
0.975

1.963

1.962
198 199 200 201 202 203 0.974
199 200 201 202 203 204

¾ Parabolic interpolation often enough to find position of

peak (i.e. frequency).
¾ Other algorithms available depending on data.

Slides adapted from ME Angoletta, CERN

Systems spectral analysis (hints)
System analysis: measure input-output relationship.

Linear Time Invariant

δ[n] h[n]
x[n] DIGITAL LTI y[n] 1 DIGITAL
SYSTEM LTI
SYSTEM
h[n] 0 n 0 n
h[t] = impulse response

∞
x[n] h[n] y[n] = x[n] ∗ h[n] = ∑ x[n − m] ⋅ h[m] y[n] predicted from { x[n], h[t] }
m =0

X(f) H(f) Y(f) = X(f) · H(f) H(f) : LTI transfer function

Transfer function can be estimated by Y(f) / X(f)

Slides adapted from ME Angoletta, CERN

 Estimating H(f) (hints)

G xx (f) = X(f) ⋅ X* (f) Power Spectral Density of x[t]

(FT of autocorrelation).

G yx (f) = Y(f) ⋅ X* (f) Cross Power Spectrum of x[t] & y[t]

(FT of cross-correlation).

Y(f) Y(f) ⋅ X* (f) G yx Transfer Function

H(f) = = =
*
X(f) X(f) ⋅ X (f) G xx (ex: beam !)

It is a check on
H(f) validity!

Slides adapted from ME Angoletta, CERN