ADSP-05-TFA-MathPrelims-EC623-ADSP (UandiStar - Org)

for more :- http://www.UandiStar.
org
Time Frequency Analysis

Mathematical Preliminaries
S. R. M. Prasanna
Dept of ECE,
IIT Guwahati,
[email protected]
100% free SMS:- ON<space>UandiStar to 9870807070 for JNTU, Job Alerts, Tech News , GK News directly to ur Mobile
Time Frequency Analysis – p. 1/2
for more :- http://www.UandiStar.org
Development of Relations
Time domain description of signals
Frequency domain description of signals
Frequency domain in terms of time domain
Time domain in terms of frequency domain
Time-frequency description
Time Description of Signals

Energy Density or instantaneous power
How much energy a signal has?
s(t) is given signal
Energy Density: |s(t)|2 energy per unit time at time t
Fractional Energy: |s(t)|2 ∆t fractional energy in the
interval ∆t at time t
Total Energy: E = |s(t)|2 dt
R
Time Description (contd.)

Characterizing time waveforms
x is a randomRvariable and p(x) its density function
Mean: µ = xp(x)dx
Variance: σt = (x − µ)2 p(x)dx
2
R
Mean (µ) gives gross characterization of density

i.e., where density is concentrated
Std. Dev. (σt ) indicates how the density is
concentrated around the mean.
t is a random variable and |s(t)|2 itsR density function
Mean or Average Time: < t >= t|s(t)|2 dt
Variance orR Duration:
T 2 = σt2 = (t− < t >)2 |s(t)|2 dt =< t2 > − < t >2
< t2 > is defined similar to < t >

Description of Time Waveforms
Duration of Signal
In 2σt most of the density is concentrated
T or σt is an indication of the duration of signal
If σt is small then most of the signal is concentrated
around mean
Signal is of short duration
If σt is large then most of the signal is spread over a
long time
Signal is of long duration
A figure of signal w/f to illustrate < t > and T = σt

Illustration of < t > and σt computation
1/4 −(t−t0 )2 /2
s(t) = (α/π) e
1/4 −(t−t0 )2 /2 2 1/2 −(t−t0 )2
|s(t)|2
= |(α/π) e
= | (α/π) e
p R −α(t−t )2
< t >= α/π te 0
dt = t0
2
p R 2 −α(t−t )2
< t >= α/π t e 0
dt = 1/2α + t0
2
σt2 = < t2 > − < t > = 1/2α
R ′ R ′
Note: Use uv = uv − u v
Frequency Description of Signals

s(t) is given signal and S(ω) its spectrum
Energy Density Spectrum: |s(ω)|2 energy per unit
frequency at frequency ω
Fractional Energy: |S(ω)|2 ∆ω fractional energy in the
interval ∆ω at frequency ω
Total Energy: E = |S(ω)|2 dω
R
Parceval’s Theorem: E = |s(t)| dt = |S(ω)|2 dω

2
R R
Frequency Description (contd.)

ω is a random variable and |S(ω)|2 its density function in
frequency
Mean or Average Frequency: < ω >= ω|S(ω)|2 dω
R
Variance orR Bandwidth:

B 2 = σω2 = (ω− < ω >)2 |S(ω)|2 dω
=< ω 2 > − < ω >2
< ω 2 > is defined similar to < ω >
Frequency Description (contd.)

Description of Spectrum of the signal
Bandwidth of Signal
In 2σω most of the density is concentrated
B or σω is an indication of the bandwidth of signal
If σω is small then most of the signal is concentrated
around mean
Signal is of small bandwidth
If σω is large then most of the signal is spread over a
large bandwidth
Signal is of large bandwidth
A figure of signal spectrum to illustrate < ω > and
B = σω
Average and Density Functions

t and ω are R.Vs. Any function of these R.Vs.
The average
R of any time function, g(t), is
< g(t) >= g(t)|s(t)|2 dt
The average
R of any frequency function, g(ω), is
< g(ω) >= g(ω)|S(ω)|2 dω
Energy density
Signal amplitudes present in the total bandwidth of
the signal.
No information about the different frequency
components present.
Energy density spectrum
Frequency components present in the total duration
of the signal.
No information about when those frequencies

Mathematical and physical ideas needed to
understand and describe how the frequencies are
changing in time
Alternative Relation to Avg. Freq.

Frequency in terms of time
R 2
R ∗ 1d
Avg. Freq.: < ω >= ω|S(ω)| dω = s (t) j dt s(t)dt
Avg. Freq. can be computed directly from s(t) without
computing its S(ω)
−jωt dt and S (ω) = t s (t)ejωt dt
∗
R R ∗
WKT, S(ω) = t s(t)e
|S(ω)|2 = S(ω)S ∗ (ω)
< ω >= ω|S(ω)|2 dω
R
′
′ −jωt dt t s (t)ejωt dtdω
′
R R R ∗
< ω >= ω ω t′ s(t )e
′
′ ∗ jω(t−t ) dtdt′ dω
R R R
< ω >= ω t t′ ωs(t )s (t)e
d jω(t−t′ ) jω(t−t ′
) jω
dt e = e
Avg. Freq. Relation (contd.)

′ ∗ 1 d jω(t−t′ ) ′ dω
R R R
< ω >= ω t t ′ s(t )s (t) j dt e dtdt
R jω(t−t′ ) ′)
ω e dω = δ(t − t
R ∗ 1d R ′ )dt′ dt
t s (t) j dt t ′ s(t)δ(t − t
R ∗ 1d
< ω >= s (t) j dt (s(t))dt
Proceeding on the similar lines
< ω >= ω |S(ω)| dω = s (t)( 1j dt
2 d 2
R 2 2
R ∗
) s(t)dt =
R d
| dt s(t)|2 dt
< ω >= ω |S(ω)| dω = s (t)( 1j dt
n d n
R n 2
R ∗
) s(t)dt =
R d
| dt s(t)|n dt
Frequency Operator
For frequency function in time domain only
1 d
Frequency operator is defined as: W = j dt
Repeated use of W denoted by W n

n 1 n dn
W s(t) = ( j ) dtn s(t)
Average of a frequency function calculation from time
function: R
< g(ω) >= g(ω)|S(ω)|2 dω
R 2
R ∗ 1d
Using < ω >= ω|S(ω)| dω = s (t) j dt s(t)dt
R ∗
< g(ω) >= s (t)g(W)s(t)dt
d
< g(ω) >= s (t)g( 1j dt
R ∗
)s(t)dt
1 d
Take g(ω), replace ω by j dt , operate on s(t), multiply by
s∗ (t) and integrate.
Time Operator
For time function in frequency domain only
d
Time operator is defined as: T = − 1j dω
Repeated use of T denoted by T n
n 1 n dn
T S(ω) = (− j ) dωn S(ω)
Average of a time function calculation from frequency
function: R
< g(t) >= g(t)|s(t)|2 dt
R ∗
< g(t) >= S (ω)g(T )S(ω)dω
< g(t) >= S (ω)g(− 1j dω d
R ∗
)S(ω)dω
d
Take g(t), replace t by − 1j dω , operate on S(ω), multiply
by S ∗ (ω) and integrate.
Complex Time and Frequency Functions

We are aware of Complex Frequency Function:
S(ω) = A(ω)ejφ(ω) = Sr + jSi
Can we have similar thing in time domain?
Useful in time frequency descriptions
Complex Time Function: s(t) = A(t)ejψ(t) = sr + jsi
What should be the values of A(t) and ψ(t)
Equivalently values of sr and si
Mean Freq. using Freq. Operator

Mean Frequency:
Ws(t) = WA(t)ejψ(t)
1 d jψ(t) ′ A′ (t)
= j dt A(t)e = (ψ (t) − j A(t) )s(t)
Therefore, the mean frequency is
R 2
R ∗ 1d
< ω >= ω|S(ω)| dω = s (t) j dt s(t)dt
A′ (t)
= (ψ (t) − j A(t) )A2 (t)dt
R ′
Since < ω > is always real, second term should be zero.

Second term is zero because it is a perfect differential
that integrates to zero.
< ω >= ψ (t)|s(t)| dt = ψ (t)A2 (t)dt
2
R ′ R ′
Concept of Instantaneous Frequency

ψ ′ (t)|s(t)|2 dt ψ ′ (t)A2 (t)dt
R R
< ω >= =
Avg. frequency is obtained by integrating ψ ′ (t) with
density over all time.
ψ ′ (t) is instantaneous value of the quantity for which we
are calculating the average.
Hence it is termed as instantaneous frequency,
ωi (t) = ψ ′ (t)
Bandwidth Equation
B2 σω2
= (ω− < ω >)2 |S(ω)|2 dω
R
=
B2 = σω = s (t)( 1j dt
2 d
− < ω >)2 s(t)dt
R ∗
R 1d
B2 = |( j dt − < ω >)s(t)|2 dt
R 1 A′ (t)
B2 = | j A(t) + ψ ′ (t)− < ω > |2 A2 (t)dt
R A′ (t) 2 2
B2 = ( A(t) ) A (t)dt + (ψ (t)− < ω >)2 A2 (t)dt
R ′
B.W. is the avg. of two terms, one depending on the

amplitude and other depending on phase.
What is the significance of these two terms?
It will be apparent in joint time and frequency
description
AM and FM Contributions to B.W.

A′ (t) 2 2
B2 < ω >)2 A2 (t)dt
R R ′
= ( A(t) ) A (t)dt + (ψ (t)−
First term averages an amplitude term over all time
Second term averages a phase dependent term
It is natural to define AM and FM contributions by
2
R ′2
BAM = A (t)dt
BF M = (ψ (t)− < ω >)2 A2 (t)dt
2
R ′
With B 2 = BAM
2 + BF2 M
BF M BAM
Fractional contributions: rF M = B and rAM = B
Duration and Mean Time

Mean Time: < t >= − φ′ (ω)|S(ω)|2 dω
R
Duration: R ′
B (ω) 2 2
T = σt = ( B(ω) ) B (ω)dω + (φ (ω)+ < t >)2 B 2 (ω)dω
2 2
R ′
Group Delay: tg (ω) = −φ′ (ω)

Amplitude and frequency variations of the spectrum will
contribute to the duration.
Spectral Amplitude Modulation (SAM)
Spectral Phase Modulation (SPM)
2
R ′2
TSAM = B (ω)dω
TSP M = (φ (ω)+ < t >)2 B 2 (ω)dω
2
R ′
With T 2 = TSAM
2 2
+ TSP M
Covariance of a Signal
To find how time and instantaneous frequency are
related?
< tψ (t) >= tψ (t)|s(t)|2 dt
′
R ′
Average of time multiplied with the instantaneous

frequency
If time and frequency are not related then
< tψ ′ (t) >=< t >< ψ ′ (t) >=< t >< ω >
So excess of < tψ ′ (t) > over < t >< ω > is a good
measure of how time is correlated with instantaneous
frequency.
Covtω =< tψ ′ (t) > − < t >< ω >
Covtω
Correlation coefficient: r = σt σω
Covariance in terms of Spectrum

In this case let tg represent time and ω represent
frequency.
Covtω =< tg ω > − < t >< ω >
ωφ′ (ω) >= − ωφ′ (ω)|S(ω)|2 dω
R
with < tg ω >= − <
When the two identities will be equal?
tψ (t)|s(t)| dt = − ωψ ′ (ω)|S(ω)|2 dω
2
R ′ R
When Covariance will be zero?

Covariance is an indication of how inst. freq. and time
are related
When inst. freq. does not change, then covariance
should be zero
Example: s(t) = A(t)ejω0 t
where the amplitude mod is arbitrary, but no change
inst. freq.
Covtψ′ (t) = tω0 |A(t)|2 dt = ω0 < t >
R
But since < ω >= ω0 , we have < ω >< t >= ω0 < t >
Therefore covariance and correlation coefficient are
equal to zero
Characteristic Function
F.T. of a density is called the characteristic function
Chr. Fn.R for the energy density spectrum is
R(τ ) = |S(ω)|2 ejτ ω dω = s∗ (t)ejτ W s(t)dt
R
ejτ W is Rthe translation operator and therefore

R(τ ) = s∗ (t)s(t + τ )dt
R(τ ) compares or correlates the signal at two different
times and hence it is termed as autocorrelation function
2 1
Inversely we have: |S(ω)| = 2π R(τ )e−jωτ dτ
R
Similarly, Chr. Fn. in freq. domain is

R(θ) = |s(t)| e dt = S (ω)ejθτ S(ω)dω =
2 jθt
R R ∗
R ∗
S (ω)S(ω − θ)dω
2 1
Hence |s(t)| = 2π R(θ)e−jtθ dθ
R
Non Additivity of Spectral Operators

Conceptual difficulties in TFA are due to basic
properties of signals and spectra
Freq. content is not additive in TF plane
In T or F plane, s = s1 + s2 and S = S1 + S2
However, for energy density
|S|2 = |S1 + S2 |2 = |S1 |2 + |S2 |2 + 2Re{S1∗ S2 } =
6 |S1 |2 + |S2 |2
Thus the freq. content is not the sum of the frequency
content of each signal
When two signal are added, the waveforms may add
and interfere in all possible ways to give different
weights to original frequencies
Mathematically, energy density spectrum is the
absolute square of the sum of the spectra, which results
in nonlinear effects

ADSP-05-TFA-MathPrelims-EC623-ADSP (UandiStar - Org)

Uploaded by

Copyright:

Available Formats

ADSP-05-TFA-MathPrelims-EC623-ADSP (UandiStar - Org)

Uploaded by

Document Information

Original Description:

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

ADSP-05-TFA-MathPrelims-EC623-ADSP (UandiStar - Org)

Uploaded by

Copyright:

Available Formats

for more :- http://www.UandiStar.

Time Frequency Analysis

Time Description of Signals

Time Description (contd.)

Mean (µ) gives gross characterization of density

Time Description (contd.)

Time Description (contd.)

Frequency Description of Signals

Parceval’s Theorem: E = |s(t)| dt = |S(ω)|2 dω

Frequency Description (contd.)

Variance orR Bandwidth:

Frequency Description (contd.)

Average and Density Functions

Time Frequency Analysis

Alternative Relation to Avg. Freq.

Avg. Freq. Relation (contd.)

Repeated use of W denoted by W n

Complex Time and Frequency Functions

Mean Freq. using Freq. Operator

Since < ω > is always real, second term should be zero.

Concept of Instantaneous Frequency

B.W. is the avg. of two terms, one depending on the

AM and FM Contributions to B.W.

Duration and Mean Time

Group Delay: tg (ω) = −φ′ (ω)

Average of time multiplied with the instantaneous

Covariance in terms of Spectrum

When Covariance will be zero?

ejτ W is Rthe translation operator and therefore

Similarly, Chr. Fn. in freq. domain is

Non Additivity of Spectral Operators

You might also like