Content

1. Introduction
2. Characteristics of signals
3. Analog signals in frequency domain
4. Analog LTI systems
5. Sampling theorem and reconstruction
6. Discrete time signals in frequency domain
7. Discrete time LTI systems
8. Digital filters
9. Correlation
10. Advanced topics

465
 Content - Correlation

9.1 Autocorrelation / cross-correlation

- Energy and power of signals
- Correlation functions for energy signals
- Transformations of power signals
- ACF and spectral power density
- Special codes

467
9.1 Cross- and autocorrelation
The Fourier transform is a mathematical tool used in signal processing to analyze the frequency content of a signal. When applied to deterministic signals, which are signals with known and
predictable characteristics, the Fourier transform allows us to decompose the signal into its constituent frequency components.

§ For the spectral analysis of deterministic signals, one can use the Fourier transform.

§ If the time-limited part of a time variable or stochastic signal is Fourier transformed, then the result is much
harder to interpret since neighboring spectral ranges are in an arbitrary phase relation with each other. The
consequence are strong notches in the amplitude of the signal. If the signal is delayed, the rough structure
of the spectrum will be maintained, however the notches are now somewhere else.

§ Correlation, however, eliminates the phase relations between the spectral domains,
but it keeps the correct amplitude information.

§ Correlation produces the correct values for the mean value <̅ and the variance D² of a
non-deterministic signal.

468
9.1 Energy and power of signals
§ The energy of a signal is in general defined as the integral of the squared signal, e.g.,
)" )"
1
˚.W = T # ! % d! = ã T f ! % d!
ã
)G )G

§ When using standardized (also named normalized) values for the energy,
the following will apply in system theory

)"

˚! = T . ! % d!
)G

469
9.1 Energy and power of signals
§ If the energy ˚.W of a signal .(!) within its complete definition domain is a finite value, then one calls such a
signal an energy signal.
+

˚.W = T . ! % d! < ∞ energy signal

*+

§ A signal .(!) infinite in time with a finite energy per time interval is called a power signal.
,
1
^4 = lim T . ! % d! < ∞ power signal
,→+ 2)
*,

470
9.1 Comparison of two signals
§ To compare two signals .(!) and à(!), one has to compare the corresponding values at the same time points.
.(!)
!
à(!)
!
§ If the difference of the corresponding values is very large, then the similarity of the signals will be rather small.
Whereas the similarity will be very large, if the difference of the corresponding values is rather small.

§ Thus, the difference of two signals represents a measure of the similarity: ∆ ! = . ! − à(!)

§ The energy of the difference signal ∆ ! is a measure for the similarity of both signals. The squaring of the
amplitude in the formula for the energy causes only positive values to appear, where larger values correspond to
large deviations and smaller values correspond to small deviations. in energy function we square the value therefore only positive values will appear

471
9.1 Comparison of two signals
§ The energy ˚∆ of a difference signal ∆ ! :
+ +
% % d!
˚∆ = T ∆ ! d! = T . ! − à !
+ *+ + *+ +
% %
= T . ! d! + T à ! d! − 2 T .(!) ( à ! d!
*+ *+ *+

= ˚! + ˚P − 2˚!P Cross−term of both signals

§ The difference energy ˚∆ consists of the sum of the energies of the individual signals (i.e., ˚! + ˚P ),
and the subtraction of a cross term ˚!P , which is a measure of the similarity of both signals.
§ The result will reduce to zero if both signals are equal.

472
9.1 Comparison of two signals

§ For the standardized (also named normalized) signals .̃ ! and àœ ! , the energies of the signals is 1:
. ! à(!)
.̃ ! = , àœ ! =
˚! ˚P
§ The standardized (or normalized) energy ˚∆ê of the difference, thus, is:
+
%
˚∆ê = T [.̃ ! − à(!)]
œ d!
*+
+ + +
∫*+ . ! % d! ∫*+ à ! % d! ∫*+ .(!)à(!)d!
= + −2
˚! ˚P ˚! ( ˚P
+
∫*+ .(!)à(!)‡! Standardized cross-term
=2−2
˚! ( ˚P of both signals

473
9.1 Comparison of two signals
§ The standardized energy of the difference, therefore, depends only on the standardized cross-term
of both signals. This factor is called the standardized (or normalized) cross-correlation coefficient
+
N
∫*+ .(!) ( à ! d! ˚∆ê
é!P = =1−
˚! ( ˚P 2
§ The standardized difference energy ˚∆ê represents a measure for the relative similarity of two functions and
does not depend any longer on the absolute signal amplitude.

N reduces to zero, then the following will apply

§ If the value of the cross-correlation coefficient é!P
+
N
é!P = 0 ⇒ T . ! ( à ! d! = 0
Orthogonality
*+
§ Signals, which satisfy this condition are called orthogonal signals.

474
9.1 Comparison of two signals
§ Definition of the standardized correlation coefficient for energy signals
+
N
˚∆," ∫*+ < ! | ! d!
é0h =1− =
2 ˚0 ˚h

§ Examples:
N
§ é0h =1 for < ! = | !
N =1
§ é0h for <(!) = l|(!)
N = −1 for
§ é0h < ! = −|(!)
N
§ é0h =0 Definition for orthogonality

N N
§ Greatest similarity will occur when é0h = 1, greatest dissimilarity (i.e., orthogonality) will occur when é0h = 0.

475
9.1 Example
)
.9 ! = Ö9 ( rect ,G
)
.% ! = Ö% ( rect
,"

§ Energy calculations:
,Gë
+ % +

˚!G = T .9% ! d! = Ö9% T 1 d! = Ö9% )9 , ˚!" = T .%% ! d! = Ö%% )%

*+ , *+
* Gë%

§ Resultant standardized signals:

1 ! 1 !
.̃9 ! = rect , .̃% ! = rect
)9 )9 )% )%

476
9.1 Correlation –
Standardized correlation coefficient for energy signals
§ In order to define the similarity of time shifted signals, we can expand the correlation coefficient.

§ Definition of the standardized (cross)-correlation function:

+
N (7) =
∫*+ < ! | ! + 7 d!
é0h
˚0 ˚h

§ Definition of the standardized autocorrelation function:

+
N
∫*+ < ! < ! + 7 d!
é00 (7) =
˚0

477
9.1 Correlation –
correlation functions for energy signals
§ Cross-correlation function and autocorrelation function:
+ +
N 7 = T < ! | ! + 7 d!
90h N 7 = T < ! < ! + 7 d!
900
*+ *+
N
§ Symbolic form: 90h 7 = <(7)×|(7)

+ *+
§ Relation with convolution: N
90h 7 = T < ! | ! + 7 d! = T < −# | −# + 7 (−d#)
< 7 ×| 7 = < −7 ∗ | 7 *+ +
< 7 ∗ | 7 = < −7 ×| 7 +

= T < −# | 7 − # d#
*+

N 7 = 9 N −7
§ Symmetry characteristic: 90h h0

478
9.1 Correlation –
correlation functions for energy signals
§ Characteristics of autocorrelation functions:
- Autocorrelation functions (ACFs) are always even functions, i.e.,
N 7 = 9 N −7
900 00

- ACFs always exhibit the maximum value at ! = 0 and the value of ACF at the point 0 corresponds to
the energy of the signal.
N 0 ≥ |9 N 7 |
900 00

N 0 =˚
900 0

- The value of the ACF at infinity corresponds to the square of the mean value.
N ∞ = <̅
900 %

- For time-limited signals, the ACF has twice the length of the signal.

479
9.1 Correlation -
Fourier transform of pulse correlation functions
§ Fourier transform of the autocorrelation function:
< 7 ×< 7 = < −7 ∗ < 7 >∗ : > : = > : %

N %
⇒ 900 7 > :

- ACF is a function only of the magnitude of the spectrum of the function <(!),
but not on the phase of the spectrum.
- > : % is called the energy density spectrum.
§ Calculation of the energy from the energy density spectrum:
+ +
N 7 = T > :
900 % er%&(> d: N 0 = T > :
⇒ 900 % d: = ˚0
*+ *+
+ +
N %
˚0 = 900 0 = T > : d: = T <(!)% d! …Parseval Theorem
*+ *+

480
9.1 Energy density spectrum
%
§ The Fourier transform of the ACF gives the energy density spectrum > : . It shows the energy
distribution as a function of frequency. autocorrelation function
Energy density spectra: § Noise signal with
+
ú00 = > ; N 7 e*$H> d7
= ∫*+ 900 large bandwidth
è Smaller ACF-peak
§ Noise signal
with smaller
bandwidth
èLarger ACF-peak
§ Harmonic signal
è ACF = harmonic
signal
§ Wide-banded noise
superimposed with a
harmonic oscillation
481
9.1 Correlation – ACFs and LTI systems
§ It can be shown that the following relation is true:

! V = U! V ∗ U! V
Uˇˇ "" ##

W - ô = X - ô Y 1 - ô

482
9.1 Correlation – transform of power signals

483
9.1 Correlation – example application of the cross-correlation
Path / Velocity measurement
§ 2 sensors (e.g., photo detectors, ,(-)
Cross-Correlator
capacitive/inductive sensors)
positioned in series along the belt. ,(-) 8#$ (')
§ Statistic surface signal is measured. .
v̄ =
)#
í(ì) ≈ ï(ì − ó, ) !

s 8(-)

rolls
v

v
Paper web

484
9.1 Correlation -
pulse correlation functions for discrete signals
.
§ Cross-correlation function:
7 ° = J ! Y .[Y + °]
†Cb
41&.

.
§ Autocorrelation function:
7 ° = J ! Y ![Y + °]
†CC
41&.

§ Time-frequency domain relation: 7 G

⇒ †CC 2 3 !

. (
§ Energy of the signal ![°]:
7 ° = J ! Y
†CC ! = E2 3 ! d3
41&. &(

485