Frequency Response Analysis

Technical Report 10

tech10.pub 26 May 1999 22:05

page 1
 Frequency Response Analysis

Technical Report 10

N.D. Cogger BSc, PhD, MIOA

R.V.Webb BTech, PhD, CEng, MIEE
Solartron Instruments

Solartron Instruments
a division of Solartron Group Ltd
Victoria Road, Farnborough
Hampshire, GU14 7PW

1997

tech10.pub 26 May 1999 22:05

page 2
 Solartron Solartron Overseas Sales Ltd
Victoria Road, Farnborough Instruments Division
Hampshire GU14 7PW England Block 5012 TECHplace II
Tel: +44 (0)1252 376666 Ang Mo Kio Ave. 5, #04-11
Fax: +44 (0)1252 544981 Ang Mo Kio Industrial Park
Fax: +44 (0)1252 547384 (Transducers) Singapore 2056
Republic of Singapore
Solartron Transducers Tel: +65 482 3500
19408 Park Row, Suite 320 Fax: +65 482 4645
Houston, Texas, 77084 USA
Tel: +1 281 398 7890 Solartron
Fax: +1 281 398 7891 Beijing Liaison Office
Room 327. Ya Mao Building
Solartron No. 16 Bei Tu Chen Xi Road
964 Marcon Blvd, Suite 200 Beijing 100101, PR China
91882 MASSY, Cedex Tel: +86 10 2381199 ext 2327
France Fax: +86 10 2028617
Tel: +33 (0)1 69 53 63 53
Fax: +33 (0)1 60 13 37 06

Email: [email protected]
Web: http://www.solartron.com

For details of our agents in other countries, please contact our Farnborough, UK office.

Solartron Instruments pursue a policy of continuous development and product improvement.

The specification in this document may therefore be changed without notice.

tech10.pub 26 May 1999 22:05

page 3
 Frequency Response Analysis
P E Wellstead B.Sc. M.Sc. Ph.D C.Eng. MIEE
1. Introduction
Frequency response analysis is the technique whereby a sinusoidal test signal is used to
measure points on the frequency response of a transfer function or impedance function. The
basic set-up is shown in Figure 1 in which a sine wave u(t) is applied to a system with transfer
function G(s). After transients due to initial conditions have decayed away, the output y(t)
becomes a sine wave but with a different magnitude Y and relative phase Φ. The magnitude
and phase of the output y(t) are in fact related to the transfer function G(s) at the frequency (ω
rad/s) of the input sinusoid.

u = U sin ωt y = Y sin ( ωt + Φ)
G(s)

Fig. 1 Linear transfer function with sinewave input.

Y
Thus = G ( jω ) = gain at ω .......................................
U
1.
and Φ = ∠G( jω ) = phase at ω ...................................

U
INPUT u(t)

ω
Φ INPUT U
Y
Φ
RESPONSE y

RESPONSE

Y
Fig. 2 Sinewave input and steady-state sinewave response as displayed on an
oscilloscope. The corresponding vector notation is shown on the right.

The relative phase and magnitude of the input and the output waveforms can be directly
measured from an oscilloscope trace (Figure 2), although this is inherently of poor accuracy
and in practice more sophisticated methods are required to remove certain errors which occur
due to noise, non-linearity and the like.

tech10.pub 26 May 1999 22:05

page 4
 The great advantage of frequency response analysis lies in its frequency selective nature. Just
one component of the frequency spectrum is extracted, and the corresponding system response
at that frequency can be measured with great precision. This has significant advantages where
the system under consideration has resonant features. Specifically, the various resonances can
be excited and studies separately without disturbing other oscillatory modes.

Im

Fig. 3a

Re

locus of G (j ω)

Fig. 3b The frequency axis is often plotted to a log scale

log |G(ω)|
GAIN

Φ
ω

PHASE

Fig. 3 Presentation of frequency response information

By sequentially measuring the gain and phase at various frequencies, a picture of the system
frequency response can be built-up and plotted either as a diagram in the complex plane (Fig.
3a) or a joint graph of gain against frequency (Figure 3b). The latter are often termed Bode
diagrams because of their use in Bode’s control system design methods.

tech10.pub 26 May 1999 22:05

page 5
 In addition to its frequency selective nature frequency response analysis has the advantage that
since waves have well-defined amplitude characteristics. In particular the maximum amplitude
of a sine wave is only √2 times greater than the root mean square value. The ratio

peak value of signal

CF = root mean square value of signal

is termed the crest factor (CF) and is an important factor is designing test signals for systems
analysis. Specifically, one usually wishes to inject as much power (as determined by the mean
square signal level) into a system as compatible with a linear system response.

System linearity is usually related to the peak test signal excursion which can be tolerated
before non-linear effects become evident. The “best” excitation in this respect is the binary
waveform with crest factor of infinity. With a crest factor of 1.41 the sine wave is reasonably
close to the theoretical best, while retaining the great merit of frequency selective system
excitation.

2. Basic result
Here we present a simple demonstration of the result, quoted earlier, that the steady state
output of a linear system excited by a sinewave is itself a sine wave with gain and phase related
to the frequency response as per equation 1.

The transform output of the system in Figure 1 can be written as

Y(s) = G(s) U(s) …………………………………………………………………. 2.

Uω
where U(s) = L[Usinωt ] = …………………………………………………...
(s )
2
3.
+ ω2

Suppose the system G(s) is stable with distinct poles pi [ i = 1, - - - - n] then the equation 2
can be written as
 n
B1 * 
∑
Ai B1
Y( s ) = U  + +  …………………………………. 4.
 i=1
(s + p i ) (s − jω) (s + jω) 

with corresponding inverse Laplace transform

 n 
∑
y( t ) = U 
 i = 1
Ai exp[− pi t ] + 2 B1 cos(ωt + ∠B1) ………………………………….

5.

where |B1| is the magnitude of B1

∠B1 is the angle or argument of B1

G ( jω )
and, by Heavisides rule, B1 = …………………………………………………. 6.
2j

Now, after a sufficient time to allow the transients to decay, the “steady state” output is
obtained. Combining equations 5 and 6 this is

y(t) = U |G (jω)| sin ( ωt + ∠G(jω )) ……………………………………………. 7.

tech10.pub 26 May 1999 22:05

page 6
 3. Implementation
The direct relative gain and phase measurement procedure mentioned earlier has certain
practical draw-backs as a means of measuring Frequency Response. These relate to the
corruption or distortion of output signals as they occur in engineering systems. Signal
corruption is due to extraneous noise which infects the output measurements (as in Figure 4a)
and makes it impractical to accurately measure gain and phase. Signal distortion arises from
non-linear effects in a system, or measurement procedure, such that the measured output is no
longer sinusoidal (as in Figure 4b) and again direct measurement of gain and phase is difficult.

test signal drifting, noise-corrupted

Fig. 4a

Fig. 4b

test signal non-linearly distorted

Fig. 4 In practical systems, although the applied test signal may be sinusoidal, the
response is often corrupted due to noise, drift or non-linearity in the system being
tested

Both the problems of non-linear distortion and noise corruption are overcome in the
measurement scheme of Figure 5, in which the measured output y(t) is first multiplied by sine
and cosine respectively and then integrated. Notice the twin processing of multiplication and
averaging which are present in all noise suppressing identification and estimation techniques.

Effectively the set-up of Figure 5 correlates the output y(t) with different versions of the input,
hence the term correlation frequency response analyser.

Briefly, the correlation analyser functions as follows. The measured system output is multiplied
by both the sine and cosine of the test frequency ω. The results of the multiplications are then
fed to two identical integrators, where they are averaged over T seconds. As the averaging time
increases the contribution of all unwanted frequency components in y(t) go to zero, and the
integrator outputs R(T), I(T) become constant values which depend only on the gain and phase
of the system transfer function at the test frequency.

In practice, the averaging is conducted over a finite time interval T, and it is necessary, for
reasons to be explained later, that T be made an integer multiple of the test frequency period.

tech10.pub 26 May 1999 22:05

page 7
 In mathematical terms, the correlation analyser functions as follows:
u(t) = U sin ωt
y(t) = Y sin ( ωt + Φ)
where Y = U |G (jω)| and Φ = ∠G(jω)

Then the sine channel output R(T) is given by

T

∫ sin ωt sin(ωt + Φ)dt

U
R(T ) = G ( jω ) ……………………………………………. 8.
T
0

U   T sin2ωT   cos2ωT 1 
= G( jω) cos Φ  +  − sin Φ  − 
T   2 4ω   4ω 4ω  

Nπ
So that when T = , N = 1,2,3 - - - - the sine channel output is
ω
 Nπ  U
R  = G( jω ) cos Φ ………………………………………………………. 9.
 ω 2

Similarly the cosine channel output I(T) is given by

T

∫ cos ωt sin (ωt + Φ) dt

U
I(T ) = G ( jω ) ……………………………………. 10.
T
0

Nπ
which can be evaluated for T = as
ω

 Nπ  U
I  = G( jω ) sin Φ ………………………………………………………. 11.
 ω  2

SYSTEM
ON TEST
y(t)

sin ωt R(T)
U sin ωt
SINE/COSINE
GENERATOR
cos ωt I(T)

MULTIPLIERS AVERAGERS
Fig. 5 Implementation of a correlation

Thus the frequency response function can be recovered from the signals R(T), I(T) evaluated at
appropriate multiples of π/ω. In practice to avoid non-linear harmonic distortion of the results,
the averaging process is performed of whole cycles of the input sine wave, so that the average
time T is selected as,
2π
T = N , N = 1,2,3 - - -
ω
Note that the averaging time is given as multiples of 2π/ω. This is because averaging over a
complete cycle avoids certain measurement errors associated with offsets on the system output.

tech10.pub 26 May 1999 22:05

page 8
 4. Noise suppression
The key feature of the correlation frequency response analyser is that by increasing the
averaging time T, the influence of extraneous noise on the measurement process can be
reduced to very low levels indeed. To see how this works consider Figure 6 in which the true
system output is corrupted by additive noise n(t). Because.the noise is additive the sine and
cosine channels at analysis frequency ω, will each be in error an amount ∆R(T) and ∆I(T) given
by T
1
∆R(T ) =
T
0
∫
n (t ) sin ω1t dt ……………………………………………………….

12.
T
1
∆I(T ) =
T ∫
n (t ) cos ω1t dt
0
……………………………………………………….

Now consider the sine channel error only (since the cosine error behaves in a similar manner)
and interpret the action of the averaging process as a filtering operation.

n(t)
u(t) y(t)
SYSTEM

Fig. 6 System corrupted by noise

To do this notice that in equation 12, ∆R(T) can be written as

1
∆R(T ) =
T ∫ n(t ) sin [ω (T − t )] dt
1
……………………………………………. 13.

N 2π
[
Because sin ω 1t = - sin ω 1 (T − t ) , when T =] ω1
………………………………….

Equation 13 is a convolution between the disturbance n(t) and the filter with impulse response
(Figure7)
1
h ( τ) = sin ω1τ 0 < τ <T ………………………………………………………. 14.
T
= 0 elsewhere ……………………………………………………….

The frequency response of this filter is obtained by noting that it consists of the product of the
function l(τ) defined by
1
l (τ) = 0 < τ <T ………………………………………………………. 15.
T

= 0 elsewhere ……………………………………………………….

and the sine wave sin ω1τ. Now the sinewave can be written as
1
sin ω1τ =
2j
(
exp[ jω1τ] − exp[-jω1τ] ) …………………………………………. 16.

tech10.pub 26 May 1999 22:05

page 9
 Hence by the shift theorem in the frequency domain
1
H(jω) =
2j
[
L(jω + jω1) − L( jω - jω1) ] ………………………………………. 17.

+∞ T
1
where L( jω ) =
−∞
∫ l (t ) exp[-jωt ] dt =
T ∫
exp [-jωt ] dt
0
…………………………………….

T
h(τ)

(τ)
N complete cycles

Fig. 7 Impulse response of the averaging filter in the analyser.

 ωT  ωT
exp-j  sin
1
hence L( jω) =
jωT
(1− exp[-jωT ]) =  2 
ωT
2
……………………………. 18.

2
2πN
Combining 17 and 18, the filter transfer function for T = is
ω1
ω1(1- exp[-jωT ])
H( jω ) = …………………………………….. 19.
(
T ω 2 − ω 21 )

REJECTION CURVES FOR N CYCLES

0 OF INTEGRATION
N=1

-10
RESPONSE IN dB

-20 N=10

-30

N=100
-40

-50

-60
0.1 0.5 1.0 2.0 5.0 10

FREQUENCY RELATIVE TO FUNDAMENTAL

Fig. 8 Frequency response of the averaging filter in the analyser.

tech10.pub 26 May 1999 22:05

page 10
 The gain function of H(jω) is plotted in Figure 8. As can be seen from the plots, the averaging
associated with the correlating frequency response analyser acts as a band pass filter with
centre frequency ω1. As the average time T increases the bandwidth of the filter becomes
narrower. Thus the corrupting influence of wide band noise n(t) is increasingly filtered out as the
correlation time is increased. This feature of increasing averaging to reduce uncertainty in
measurements is a recurrent feature of dynamic measurement techniques and should be borne
in mind when considering more advanced procedures which are not amenable to the simple
filtering arguments used here.

5. Non linearities
The aim of frequency response analysis is to characterise the transfer function of an underlying
linear dynarnic system. However, in reality few systems are linear, and so we are really
measuring some linear approximation. In dynamic analysis the most troublesome forms of non-
linear behaviour are associated with activating (that is, getting signals into a system) and
sensing (that is, measuring a systems response). Often these forms of non-linear behaviour are
severe, and may have nothing to do with the system dynamics - only how we actuate them*
(Figure 9). Typically, actuation and measurement non-linearities contain no dynamics, but
constitute a static gain characteristic which distorts the input / output information. The first thing
the analyst tries to do is directly overcome this form of non-linearity. This is often possible by
putting local feedback loops around non-linear actuators or using the methods outlined in the
next section. Where this is not possible, intelligent instrumentation can sometimes compensate
for non-linearity by applying an inverse non-linear operation. Many commercial data logging
instruments include this feature.

When we can't get round the non-linearity, it must be accommodated in the analysis procedure.
To see how this might be done consider Figure 10, which shows a non-linear characteristic
which might be associated with an electric drive motor. The effect of feeding sine-waves
through this non-linearity is also illustrated.

The response of any such non-linearity to a sine wave can be expressed as a sum of sine and
cosine waves at the frequencies ω1, 2ω1, 3ω1, 4ω1, - - -. However, recall that the filter H(jω)
associated with the correlation frequency response analyser has zero gain at the frequencies
2ω1 , 3ω1 , - - -, since these correspond to zeroes of L(jω) when T = N2π /ω1. Thus the
frequency response analyser rejects all the harmonic components of output x(t) (Figure 11)
generated by non-linear distortion, and effectively measures the fundamental component of
response only. This has uses in control system design where the describing function method
hinges upon measurement of the fundamental component of a sinusoidal frequency response.
By correlating the output of a system with harmonic frequencies 2ω1 , 3ω1 , - - -, i ω1 , - - -, the
harmonic components of frequency response are obtainable. These higher frequency
components are useful in computing the exact frequency response and qualifying the level of
non-linear distortion. These harmonic measurements can also be used for exact describing
function measurements.

* This is not always true, since actuators and sensors often form an integral part of a system.

tech10.pub 26 May 1999 22:05

page 11
 test system system measured
signal stimulus response response

ACTUATOR SYSTEM SENSOR

Fig. 9 System actuation and sensing

input
voltage v speed η
POWER ELECTRIC
AMPLIFIER MOTOR

η
speed

input

Fig.10 Typical actuator non-linearity and system response.

∞

u(t) = U sin ω1 t
x (t ) = ∑ X sin(iω t + Φ )
i =1
i 1 i

NON-LINEARITY

RESPONSE
Envelope of FRA filter
averaging over one complete cycle
X1
X2
X3
X4
X5 X6
FREQUENCY

ω 2ω 3ω 4ω 5ω 6ω (rads/s)

Fig. 11 The harmonics which are present in the response from a non-linear system
are suppressed by the filtering action of the analyser.

tech10.pub 26 May 1999 22:05

page 12
 6. Closed loop and remote input measurement
Physical constraints often make the measurement scheme of Figure 1. impractical. In particular
the system to be tested might be an integral part of a larger system, or an unstable process
which must be maintained under closed loop control at all times. Most modern frequency
response analysers are modified to cater for such situations. Consider the two basic cases
shown in Figure 12 where the system G(jω) is either in closed loop or is down stream in a
cascade of transfer functions. It is assumed that the signals x(t),y(t) either side of the test
system can be measured and that a test sinusoid u(t) can be injected at some point which
casually influences x(t) and y(t). The modified transfer function layout of Figure 13 is then used.
In this scheme the frequency responses from r(t) to u(t) and r(t) to y(t) are measured and the
frequency response of G(jω) obtained by dividing the gain measurements and subtracting the
phase measurements. The system of referencing the measurements u(t), y(t) back to the test
sinusoid is most important. Theoretically it relates to the method of "Instrumental variables"
which grew out of econometric theory to deal with similar measurement problems in fiscal and
economic data. Practically, it enables us to test closed loop systems without breaking the loop,
and at the same time avoiding the ambiguity of feed-back information. The nature of this
ambiguity can be seen from consideration of Figure 14. Two relations can be written for y, thus

y ( s ) = G( s ) x(s) + n( s ) forward path ……………………………………….

1 20.
y( s ) =
F( s )
[− x( s ) + u( s )] feedback path ………………………………….

If the forward path noise n(t) is zero, the forward path relation is exact and a frequency
response analysis from x(t) to y(t) gives G(jω).

SYSTEM

u(t) x(t) y(t)

G(jω)

a) closed loop system

x(t) y(t)
A B G(jω) C D

a) cascade system

Fig. 12 Point to point frequency response analyser.

tech10.pub 26 May 1999 22:05

page 13
 However, if u(t) is very small compared with n(t) then the feedback path relation is exact and a
frequency response analysis from x(t) to y(t) gives -1 / F(jω).

For intermediate values of u(t) and n(t) an estimate is obtained which is a function of G, F and
the relative power in u(t) and n(t). The procedure of correlating back to u(t), as implemented in
Figure 13, ensures that no ambiguity occurs. Specifically, the modified implementation
measures the ratio of the transfer functions
y( s ) x(s)
and thus
u( s ) u(s)

y( s ) G( s )
u( s ) 1 + FG( s )
= = G( s )
x( s ) 1
u( s ) 1 + FG( s )

SYSTEM

G(jω)

x(t) y(t)
u sin ω1t
Rx(T)

X CHANNEL
Ix(T)
cos ω1t
SIN / COS
GENERATOR

sin ω1t Ry(T)

ω1
Y CHANNEL
FREQUENCY
CONTROL Iy(T)

Ry (T ) + jIy (T )
System frequency response at ω 1 = G( jω 1 ) =
Rx (T ) + jIx (T )
Fig. 13 A two channel point-to-point analyser.

n(t)
x(t)
u(t) y(t)
G

Fig. 14 A closed loop system.

tech10.pub 26 May 1999 22:05

page 14
 It is important to realize that the feedback between two signals need not be by an explicit closed
loop. The feedback may be implied by the nature of the system, as in the electrical network of
Figure 15, where the two variables v1, v2 are not related by a simple casual transfer function,
but are each joint functions of the source voltage v0.

R1 R2 L1

v0 C1 C2 v1 R3 v2

Fig. 15 Voltage variables in a circuit.

test
sinusoid
u(t) NON-LINEAR
ACTUATOR G(jω

system response x(t)

to be correlated
with u(t) y(t)

Fig. 16 Linear system with non-linear actuator.

The modified frequency response analyser can also avoid problems with non-linear distortion.
Say that block B in Figure 12 is non-linear, then by correlating both x and y back to u(t), only the
influence of the fundamental component of x is used and so G(jω) is correctly measured. Thus
in an open loop control system with actuator non-linearity then the test scheme of Figure 16 will
avoid the effects of non-linear distortion on the measurement, provided a measurement of x(t)
can be made.

tech10.pub 26 May 1999 22:05

page 15
 7. Drift and offsets
In most practical situations the output of a system with a sine-wave applied to the input will
consist of a normal output level, with the sinusoidal component of response superimposed.
Typically (Figure 17) the output level will consist of a dc offset, combined with a low frequency
drift. Moreover, the amplitude of offset and drift components maybe much larger than the test
signal component. It is therefore important to find ways of removing low frequency components
from the system output to prevent them distorting the measurement process.

RESPONSE
system output plus test sinusoid

normal system output

TIME

Fig. 17 Measured system output showing normal operating level

(comprising drift, offset etc.) and response to a test sinusoid.

The most direct practical procedure is to use ac coupling with a capacitor. This provides a high
pass filter which removes low-frequency components of response but leaves the sine-wave
component uncorrupted. However, it is interesting to notice that the frequency response
averaging filter H(jω) has a zero at ω= 0, so that constant components of output are always
exactly cancelled out. Moreover, as the averaging number N is increased (Figure 8), the
filtering out of low frequency drift becomes more effective. In practical situations the primary
method of removing offsets and low frequency drift is by high-pass filtering or subtracting an
estimate of the offset from the original signal. This removal at source of the major part of the
offset drift improves the dynamic resolution of the frequency response measurement, and allows
the natural filtering action of H(jω) to suppress residual effects.

tech10.pub 26 May 1999 22:05

page 16
 8. Applications
These brief notes outline some typical applications of the frequency response analyser.

(i) Harmonic Analysis

In practice one often meets engineering signals which are periodic with a well defined (i.e.
noise-free) shape. In such cases it is frequently useful to analyse the harmonic structure of
these waveforms, since the occurrence of certain harmonics may give insight into their source.
By the same token there may be a requirement that certain harmonic components must be less
than a pre-specified level. This is true in, for example, testing commercial signal generators
where part of the technical specification refers to permitted harmonic levels.

periodic signal

y(t)
sync
sin rωƒ t

SIN/COS
GENERATOR

cos rωƒ t
synchronise with frequency (ωƒ)
and phase of fundamental component

Fig. 18 The frequency response analyser used as a harmonic analyser.

The frequency response analyser can readily be adapted as a harmonic analyser by adding, to
the circuit of Figure 5, the facility to synchronise the frequency ωƒ and phase of the test sinusoid
with the fundamental period of the periodic signal. The periodic signal is then treated as if it
were the system output y(t) (see Figure 18) and correlated with the harmonic frequencies r ωƒ
(r =1, 2, 3, - - -) generated in the frequency response analyser. The sine and cosine output
channels then give the relative gain and phase of the harmonic components. The use of
harmonic frequencies in this way can also be applied to measure the harmonic content of
response of a non-linear system.

tech10.pub 26 May 1999 22:05

page 17
 ii) Transfer Function Analysis of AC Systems

In addition to conventional dc feedback control systems there is a class of servo-mechanisms

which employ amplitude modulated ac signals. The frequency response analyser is particularly
effective at response testing this kind of control system for three reasons:

a) It is particularly easy to modulate and demodulate the amplitude of an ac signal

with sinewaves.

b) The low crest factor of a sinewave test signal means that a relatively high level of
modulation can be achieved before non-linear affects occur.

c) If the demodulated signal still contains significant components of the carrier

frequency, the filtering action of the correlation frequency response analyser will
tend to suppress the carrier frequency components.

ω ωc - ω ωc ωc + ω
ωc
2

Fig. 19 Frequency spectrum of an amplitude modulated signal which has

been imperfectly demodulated.

A potential problem with ac modulated systems occurs when the frequency of analysis ω
approaches ωc /2, where ωc is the carrier frequency. The spectrum of the demodulated signal
(Figure 19) will contain contributions at ω, ωc and ωc ± ω. As ω is increased the fundamental
and the lower sideband ωc - ω converge, until at ω = ωc /2, they meet and both contribute to the
correlator output. Depending on the relative phases of ω and ωc - ω the transfer function
analyser at ωc /2 can take any one of a whole range of values. In practice, ac modulated servos
have a frequency response which falls to zero well before ωc /2, so that this kind of
phenomenum does not frequently cause problems.

tech10.pub 26 May 1999 22:05

page 18
 iii) Sampled Systems

W(t)

∆t
u(t) x(t) y(t)
G(s)

Fig. 20 A system with a sampler.

The modulation affect described above can cause problems in systems which included
sampling action. Specifically, consider a sampled data control system (Figure 20) in which the
test signal u(t) is sampled at uniform intervals ∆t seconds apart. The sampling action can be
written as

x (t) = u(t) W(t) ………......................................……...…. 22.

Where the sampler W(t) is a train of impulses spaced ∆t seconds apart. Now the sampler can
be considered as a series expansion in terms of harmonies of ωs, where ωs = 2π/∆t.

 ∞ 
∑
1
i.e. W (t ) = 1+ 2 cos kωst  ………………………………………………………. 23.
∆t  
 k =1 
If the input is a sinewave of frequency ω and relative phase Φ then the sampled output is

1  
∞
x (t ) = sin(ωt + Φ) + 2
∆t 

∑
k =1
sin(ωt + Φ) cos kω st 

…………………………. 24.

or
1  
∞

x (t ) = sin(ωt + Φ) +
∆t 
 k =1
∑[sin (kω s t + ωt + Φ) - sin (kω s t − ωt − Φ)  ……….

] 25.

ω ωs- ω ωs+ ω 2 ωs+ ω 2 ωs+ ω 3ωs- ω 3ωs+ ω

ωs ωs 3ωs 2ωs 5ωs 3ωs ω

2 2 2

Fig. 21 Frequency spectrum of the sampled signal x(t).

The frequency spectrum of this is shown in Figure 21. Note that when ω = ωs /2 the lower
sideband at ωs - ω coincides with the fundamental frequency ω and gives a component of x(t) at
frequency ω = ωs /2 demarked χ(t) and defined by
1
χ(t ) =
∆t
[
sin (ωt + Φ) − sin(ωt − Φ) ]
……………………………………. 26.

2
= cos ωt sin Φ ……………………………………………………….
∆t

tech10.pub 26 May 1999 22:05

page 19