TSDT14 Signal Theory Lecture 11 Multi-Dimensional Processes – Primarily 2-D Mikael Olofsson Department of EE (ISY) Div. of Communication Systems
TSDT14 Signal TheoryLecture 11
Multi-Dimensional Processes – Primarily 2-D
Mikael Olofsson
Department of EE (ISY)
Div. of Communication Systems
Example: 2-D Filtering
2016-10-06 2
TSDT14 Signal Theory - Lecture 11
Source: http://www.grand-illusions.com/opticalillusions/angry_and_calm/
Mr.
Angry
Mrs.
Calm
Example: 2-D Filtering
2016-10-06 3
TSDT14 Signal Theory - Lecture 11
Source: http://www.grand-illusions.com/opticalillusions/angry_and_calm/
Mr.
Angry
Mrs.
Calm? ?
Multi-Dimensional Signals & Systems
2016-10-06 4
TSDT14 Signal Theory - Lecture 11
Two-Dimensional Signals
2016-10-06 5
TSDT14 Signal Theory - Lecture 11
),( 21 aax
A function of two variables
(space-discrete)
(space-continuous)
],[ 21 nnx
Separable signals:
][][],[ 221121 nxnxnnx ⋅=
)()(),( 221121 axaxaax ⋅=
−1.5
−1
−0.5
0
0.5
1
1.5
−1.5
−1
−0.5
0
0.5
1
1.5
−1
−0.5
0
0.5
1
2a1a
1a−1.5 −1 −0.5 0 0.5 1 1.5
−1.5
−1
−0.5
0
0.5
1
1.5
2a
Two-Dimensional Stochastic Processes
2016-10-06 6
TSDT14 Signal Theory - Lecture 11
{ }),(E),( 2121 aaXaamX =Mean:
Auto-correlation: ( ) { }),(),(E,;, 221121221121 babaXaaXbabaaarX ++=++
),( 21 aa
Wide-sense stationarity:
Both ),( 21 aamX( )221121 ,;, babaaarX ++and independent of .
Simplified notation: Xm ( )21,bbrXand
Classification of Systems
2016-10-06 7
TSDT14 Signal Theory - Lecture 11
LSI: Both linear and space-invariant.
Similarily for space-discrete systems
( )211 , aaxLinearity: Input gives output ( )211 ,aay
( ) ( )212211 ,, aaxbaaxa ⋅+⋅
( ) ( )212211 ,, aaybaaya ⋅+⋅
( )212 ,aaxInput gives output ( )212 ,aay
Then input
gives output
.
.
.
Space-invariance: ( )21,aaxInput gives output ( )21,aay
( )2211 , babax −−Then input
gives output ( )2211 , babay −−
.
.
Two-Dimensional Convolution
2016-10-06 8
TSDT14 Signal Theory - Lecture 11
222222111111 )()()()( dbbahbxdbbahbx −−= ∫∫∞
∞−
∞
∞−
))(())(( 222111 ahxahx ∗⋅∗=
For LSI systems, the output is given by a two-dimensional convolution ofthe input and the impulse response of the filter.
Separable signals: ( )( ) ( ) ( ) ( ) ( ) 21222111221121, dbdbbahbahbxbxaahx −−= ∫ ∫∞
∞−
∞
∞−
*
Definition: ( )( ) ( ) ( ) 2122112121 ,,, dbdbbabahbbxaahx −−= ∫ ∫∞
∞−
∞
∞−
*
Space-discrete: ( )[ ] [ ] [ ]22112121 ,,,1 2
knknhkkxnnhxk k
−−⋅= ∑∑*
Non-Separable Convolution 1(2)
2016-10-06 9
TSDT14 Signal Theory - Lecture 11
( ) ( ) <+
==elsewhere,0
1,1,,
2
2
2
1
2121
aaaahaax
−2
−1
0
1
2
−2
−1
0
1
2
0
0.2
0.4
0.6
0.8
1
Non-separable signals−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
Non-Separable Convolution 2(2)
2016-10-06 10
TSDT14 Signal Theory - Lecture 11
−2
−1
0
1
2
−2
−1
0
1
2
0
0.5
1
1.5
2
2.5
3
3.5
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
<+=−−
=elsewhere,0
4,)2/(1)2/arccos(2 2
2
2
1
22aaaaaa
( ) ( )( )2121 ,, aahxaay = *
Separable Convolution 1(3)
2016-10-06 11
TSDT14 Signal Theory - Lecture 11
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2−2
−1
0
1
2
−2
−1
0
1
2
0
0.2
0.4
0.6
0.8
1
Separable signals
( ) ( ) == 2121 ,, aahaax
<<
=elsewhere,0
21and2/1,1 21 /aa
( ) ( )21 azaz ⋅=
( ) ( ) <
==elsewhere,0
2/1,1rect
aaazwith
Separable Convolution 2(3)
2016-10-06 12
TSDT14 Signal Theory - Lecture 11
{ } { }21 1,0max1,0max aa −⋅−=
( ) ( )21 triangletriangle aa ⋅=
−2
−1
0
1
2
−2
−1
0
1
2
0
0.2
0.4
0.6
0.8
1
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
( ) ( )( )2121 ,, aahxaay = *
Separable Convolution 3(3)
2016-10-06 13
TSDT14 Signal Theory - Lecture 11
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2
−2
−1.5
−1
−0.5
0
0.5
1
1.5
2
Scaledgraphic
−2
−1
0
1
2
−2
−1
0
1
2
0
0.2
0.4
0.6
0.8
1
{ } { }21 1,0max1,0max aa −⋅−=
( ) ( )( )2121 ,, aahxaay = *
( ) ( )21 triangletriangle aa ⋅=
2-D Space-Continuous Fourier Transform
2016-10-06 14
TSDT14 Signal Theory - Lecture 11
( ) ( ){ } 21
)(2
2121212211),(,, dadaeaaxaaxffX
afafj +−
∞
∞−
∞
∞−
∫ ∫==π
FDefinition:
Inverse: ( ) ( ){ } 21
)(2
2121212211),(,, dfdfeffXffXaax
afafj +
∞
∞−
∞
∞−
∫ ∫==π1-
F
Properties: ( ){ } ( ){ }{ } ( ){ }{ }2112212121 ,,, aaxaaxaax FFFFF ==
( ) ( ){ } ( ){ } ( ){ }2221112211 axaxaxax FFF ⋅=⋅
*( ) ( ){ } ( )212121 ,)(,, ffHXaahaax =⋅F
( ){ } ( ) ( )212121 ,,,)( ffHffXaahx ⋅=F *
2-D Space-Discrete Fourier Transform
2016-10-06 15
TSDT14 Signal Theory - Lecture 11
[ ] [ ]{ } [ ] )(2
2121212211
1 2
,,,nnj
n n
ennxnnxXθθπ
θθ+−∑∑==FDefinition:
Properties:
Inverse: [ ] [ ]{ } [ ] 21
)(2
21
1
0
1
0
21212211,,, θθθθθθ
θθπddeXXaax
nnj +
∫ ∫==1-
F
[ ]{ } [ ]{ }{ } [ ]{ }{ }2112212121 ,,, nnxnnxnnx FFFFF ==
[ ] [ ]{ } [ ]{ } [ ]{ }2221112211 nxnxnxnx FFF ⋅=⋅
*[ ] [ ]{ } [ ]212121 ,)(,, θθHXnnhnnx =⋅F
[ ]{ } [ ] [ ]212121 ,,,)( θθθθ HXnnhx ⋅=F *
[ ] [ ]221121 ,, mmXX ++= θθθθ for integers and .1m 2m
(periodic conv)
Power and Power Spectral Density
2016-10-06 16
TSDT14 Signal Theory - Lecture 11
( ) ( ){ } ( ) 21
)(2
2121212211,,, dbdbebbrbbrffR
bfbfj
XXX
+−
∞
∞−
∞
∞−
∫ ∫==π
FPSD:
Mean: )0,0(),( 2121 Hmdadaaahmm XXY ⋅== ∫ ∫∞
∞−
∞
∞−
Output: ),)((),( 2121 aahXaaY = *
ACF: ),)(~
(),( 2121 bbrhhbbr XY = * * with ),(),(~
2121 aahaah −−=
Output of LSI-system if input is WSS:
PSD: ( ) ( ) ( )21
2
2121 ,,, ffRffHffR XY =
Power: ( ){ } ( ) ( ) 212121
2 ,0,0, dfdfffRraaXEP XXX ∫ ∫∞
∞−
∞
∞−
===
Two-Dimensional Sampling
2016-10-06 17
TSDT14 Signal Theory - Lecture 11
[ ] ( )221121 ,, AnAnxnny =Sampling:
Probabilistic (for rectangular grid):
Deterministic (for rectangular grid):
Spectrum: [ ] ∑∑
−−=
1 2 2
22
1
11
21
21 ,1
,k k A
k
A
kX
AAY
θθθθ
1A 2ASampling periods: and
Sampling: [ ] ( )221121 ,, AnAnXnnY =
PSD: [ ] ∑∑
−−=
1 2 2
22
1
11
21
21 ,1
,k k
XYA
k
A
kR
AAR
θθθθ
Two-Dimensional PAM
2016-10-06 18
TSDT14 Signal Theory - Lecture 11
Probabilistic (for rectangular grid):
Deterministic (for rectangular grid):
Spectrum: ( ) ( ) [ ]22112121 ,,, AfAfYffPffZ ⋅=
PSD: ( ) ( ) [ ]2211
2
21
21
21 ,,1
, AfAfRffPAA
ffR YZ =
PAM: ( ) [ ] ( )∑∑ −−=
1 2
2221112121 ,,,n n
AnaAnapnnyaaz
PAM: ( ) [ ] ( )∑∑ Ψ−−Ψ−−=
1 2
222211112121 ,,,n n
AnaAnapnnYaaZ
1Ψ 2Ψuniform on [ )1,0 A [ )2,0 Auniform onand
both independent of and of each other.[ ]21,nnY
www.liu.se
Mikael Olofsson
ISY/CommSys
Rounding Up the Course
2016-10-06 20
TSDT14 Signal Theory - Lecture 11
Stochastic processes: Stationarity, ergodicity, mean, ACF, PSD…
LTI filtering: Mean, ACF, PSD.
Cross-correlation and cross-spectrum. Joint stationarity.
Poisson processes.
Prediction.
Non-linearities: Squaring and such, saturation, quantization.
Modulation: AM, FM, PM, noise.
Estimation (only on laborations).
Linear mappings: Sampling, PAM, reconstruction.
Two-dimensional: Signals, systems,…
Written Examination
2016-10-06 21
TSDT14 Signal Theory - Lecture 11
When: Friday 2015-10-30, 14.00-18.00. Sign up!
Allowed aids:
Olofsson: Tables and Formulas for Signal Theory
Henriksson/Lindman: Formelsamling i Signalteori
Pocket calculator with empty memory
A German 10 mark note of the fourth series (1991-2001)
What:
A three-part introductory task (simple, 2/3 must be OK).
Five problems – 5 points each, pass is 10 points.
Written Examination – cont’d
2016-10-06 22
TSDT14 Signal Theory - Lecture 11
A German 10 mark note of the fourth series (1991-2001)
Good Practices at Exams
2016-10-06 23
TSDT14 Signal Theory - Lecture 11
Rules according to the exam cover:
• Only one task on the same piece ofpaper.
• Use only one side of the paper.
• Number the pages. (see common sense → )
• Do not use a red pen(cil). (that’s my color)
Common sense:
1. Solve the exam problems.
2. Sort the papers according totask numbering.
3. Number the pages last!
4. Now hand in your exam.
Do not do it in any other order!
Let me add:
• Hand in readable solutions.
• Do not hand in scriblings!
Finally:
• Always provide solid arguments for steps taken in your solutions.