Professional Documents
Culture Documents
Fourier Transform of Discrete Time Aperiodic Signals If the innite sum converges, x(n)ejn
n=
=
n=
x (n )
ej(mn)d
x(n)ej (+2k)n
n= j
= 2 x(m) +
n=,n=m
x (n )
ej(mn)d
= X (e ) X (ej ) is periodic with period 2 . So, it has a Fourier Series. x(n) can be calculated by integrating both sides
= 2 x(m) +
n=,n=m
ej(mn) x (n ) j (m n )
X (e )e
jm
1/2
=
n=
x(n)ejn ejmd
2.
y (n ) =
k =
g (k ) u (n k )
0.5
1.5
2.5
3.5
G e
jw
=
n=
Arg (G) =
0 0 w < 23 23 w<
|G ejw
w = 0 : 0 . 0 1 : pi ; subplot ( 2 , 1 , 1 ) p l o t (w , abs (1+2 cos (w ) ) / 3 ) , g r i d , y l a b e l ( Magnitude ) subplot ( 2 , 1 , 2 ) p l o t (w , angle (1+2 cos (w ) ) ) , g r i d , x l a b e l ( w ) , y l a b e l ( Phase )
3.
2 3
= =1e
1 2 3
4.
G e
jw
=
n=
g (n)e
jwn
=
n=
g (n) cos wn j
n=
g (n) sin wn
Symmetry of magnitude and phase angle for real valued sequences |G ejw | = G ejw G ejw
1/2 1/2
G ejw =
n=
g (n)ejwn
=
n=
g (n) cos wn + j
n=
This shows that the magnitude is an even function. In a similar way, Arg G ejw = Arg G ejw
Comparing the above two equations, we get Re G ejw Im G ejw = Re G ejw = Im G ejw
5.
u(n) = ua (nTs ),
LHS =
1/2
U (f )ej 2f n df 1 Fs
Fs /2
ua (t)e
j 2F t
U
Fs /2
F Fs
ej 2nF/Fs dF
ua (t) =
Ua (F )e u(n)e
n= 1/2
j 2F t
Ua (F )ej 2nF/Fs dF
(k +1/2)Fs (k 1/2)Fs Fs /2 Fs /2 Fs /2 Fs /2 Fs /2 Fs /2
U (f ) = u(n) =
1/2
j 2f n
Ua (F )ej 2nF/Fs dF
k = j 2f n
U (f )e
df
(3)
=
k =
Ua (Q + kFs )ej 2n(Q+kFs )/Fs dQ Ua (Q + kFs )ej 2nQ/Fs dQ Ua (F + kFs )ej 2nF/Fs dF Ua (F + kFs ) ej 2nF/Fs dF
=
k =
U (f )ej 2f n df =
1/2
Ua (F )ej 2F nTs dF Ua (F )e
j 2nF/Fs
=
k = F s /2
dF =
Fs /2
k =
6. U F Fs
= Fs
k =
ua (t)
U
Fo + Fs F
F0 Fs
= FsUa(F0)
t ua(nT ) = u(n)
Fo B
F Fs
Fs
U is scaled version of Ua - shape not aected by sampling. Can recover Ua from U . U (F/Fs) is periodic in F with a period Fs: = = U F0 + kFs Fs ,
t Ts 3 2 Fs
s F 2
F
Fs 2 3 2 Fs
F0 Fs
F0 + Fs Fs k = 1, 2 =U
=U
F0 Fs Fs
7. U
= Fs
k =
t B U
F Fs
F B
Consequence of aliasing: F0 Fs = Fs[Ua(F0 Fs) + Ua(F0)] = Fs[overlapping value + Ua(F0)] U Cannot recover Ua from U , as the high frequency components have changed.
Fs
t Ts
s F 2
Fs 2
U u(n)
F Fs
Fs
t Ts
s F 2
Fs 2
8.
1 Fs U
F Fs
U a ( F + Fs )
Ua (F ) Ua (F )
U a ( F Fs )
1 Fs U
F Fs
U a ( F + Fs )
Ua (F )
U a ( F Fs )
9.
Sampling Theorem To calculate analog signal at t, we need sampled values for all future times, corresponding to n > 0 Hence for control purposes, this reconstruction is not useful. Provides absolute minimum limit of sampling rate. If sampling rate is lower than this minimum, no lter (whether causal or not) can achieve exact reproduction of the continuous function from the sampled signals.
Suppose highest frequency contained in an analog signal ua(t) is Fmax = B . It is sampled at a rate Fs > 2Fmax = 2B . ua(t) can be exactly recovered from its sample values:
sin ua(nTs)
ua(t) =
n=
Ts (t Ts (t
nTs)
nTs)
10.
Frequency Domain Interpretation of ZOH Recall normalized frequency: f = F/Fs, fmax = B/Fs
sin f f
Minimum sampling rate Fs,1 = 2B fmax,1 = 0.5 Sample at twice the minimum Fs,2 = 4B fmax,2 = 0.25 Maximum deviation = 10% Sample at 4 times minimum Fs,3 = 8B fmax,3 = 0.125
0.9
0.8
0.7
0.6 SH
0.5
0.4
0.3
0.2
0.1
0.05
0.1
0.15
0.2
0.25 f
0.3
0.35
0.4
0.45
0.5
0 0.5
10
11.
Minimum sampling rate = twice band width Problems Not really band limited Systems are generally nonlinear Shannons reconstruction cannot be implemented, have to use ZOH Solution: sample faster Number of samples in rise time = 4 to 10 Sample 10 to 30 times bandwidth Use 10 times Shannons sampling rate cTs = 0.15 to 0.5, where, c = crossover frequency
CL 692 Digital Control, IIT Bombay
11
12.
Filtering - Motivation
Measurements are often corrupted by high frequency noise, which have to be ltered before further processing. Systems that transmit the low frequency information while removing the eect of high frequency noise are known as low pass lters and this action is known as low pass ltering. Sometimes we are interested in monitoring of a transient response so as to take early corrective action. Often this requires a derivative action that works on the basis of the slope of the response curve. We will see later that this requires the usage of the high frequency content of the response. Indeed we may be interested in ltering of the frequency content in some arbitrary frequency range while passing the others. Will demonstrate that such things can be achieved by suitable choice of pole and zero locations.
12
13.
Filter Design
Apply input u(k ) = ak 1(k ) If e1 is large, input is present in the output y . to an LTI system with with If e1 is small, eect of u is removed. z z transfer function G(z ). G (z ) = e0 + e1 za za Want to know what happens + { terms due to the poles of G(z )} to frequency content of u by z za G (z ) |z =a = G(a) e1 = G(z ). Let a be of the form z za ej and let G(z ) not have a If we want to pass the input signal ak in pole at a. the output, choose G(a) large z za z Y (z ) = e0 + e1 za + {terms due to poles of G(z )} Y (z ) = G ( z ) A small G(a) would result in the reduction of this input signal in the output Large G(a) can be achieved if G(z ) has a pole close to a A zero of G(z ) near a will ensure the rejection of the eect of u on the output.
13
14.
If input frequency w0 is to be passed through G(z ), place poles of G(z ) near w0 If it is to be ltered, place zeros of G(z ) near w0 Unique frequency values are in (, ] w close to 0 corresponds to low frequencies while w close to corresponds to high frequencies
High Low Pass
High Frequency Low Frequency
Notice that e with w (, ] denes the unit circle. As a result, we can mark the low and high frequency regions as in the gure:
CL 692 Digital Control, IIT Bombay
jw
Pass
14
15.
To pass signals of frequency w0, we should place poles near w0 To reject w0, we should place zeros near w0
Im(z)
Im(z)
O O O
Re(z)
O Re(z) O O
Low pass lter Place the poles inside unit circle for stability If complex, choose in conjugate pairs
15
16.
G1 (z ) =
0.5 z 0.5
G1 (z )|z =1 = 1, so that its steady state gain is 1. Substituting z = ejw , we get 0.5 G1 (e ) = jw e 0.5
jw
0.5
1.5
2.5
50 Phase
0.5 (cos w 0.5) + j sin w (cos w 0.5) j sin w = 0.5 (cos w 0.5)2 + sin2 w 0.5 |G1 (ejw )| = 1.25 cos w sin w G1 (ejw ) = tan1 cos w 0.5 = This lter magnies the signal frequencies near w = 0 in relation to other frequencies
100
16
17.
0.5
1.5
2.5
G2 (ejw ) ejw + 1 cos w + j sin w + 1 = jw = K e 0.5 cos w + j sin w 0.5 [(cos w + 1) + j sin w][(cos w 0.5) j sin w] = (cos w 0.5)2 + sin2 w (cos2 w + 0.5 cos w 0.5) + sin2 w G2 (ejw ) = 0.25 sin2 w + cos2 w + 0.25 cos w j sin w(cos w 0.5 cos w 1) + sin2 w + cos2 w + 0.25 cos w (0.5 + 0.5 cos w) 1.5j sin w G2 = 0.25 1.25 cos w G1 solid line, G2 broken line |G2 (ejw )| < |G1 (ejw )| w > 0. Thus, G2 is a better low pass lter.
50 Phase
100
17
18.
Calculate the response of G3 (z ) = z+1 for input u(n) = (1)n 1(n) z1 G3 (z ) = z 1 + z1 z1 g3 (n) = 1(n) + 1(n 1)
y (n) = 2
k =n
y (n) =
i=
= =
i=0 n
1 (1)n+1 =2 1(n) (1)n 1(n) 1 (1) = 1(n) 1 (1)n+1 (1)n = 1(n) This shows that (1)n has been ltered. This is because the lter has a zero at (1, 0).
u(n i) +
i=1
u(n i)
=2
i=0 n
= 2
18
19.
y (n ) =
k =
g (k ) u (n k )
0.5
1.5
2.5
3.5
G e
jw
=
n=
Arg (G) =
0 0 w < 23 23 w<
|G ejw
w = 0 : 0 . 0 1 : pi ; subplot ( 2 , 1 , 1 ) p l o t (w , abs (1+2 cos (w ) ) / 3 ) , g r i d , y l a b e l ( Magnitude ) subplot ( 2 , 1 , 2 ) p l o t (w , angle (1+2 cos (w ) ) ) , g r i d , x l a b e l ( w ) , y l a b e l ( Phase )
20.
2 3
= =1e
1 2 3
20
21.
x[t] =
X [F ]e
j 2F t
dF
X [F ] =
x[t]ej 2F tdt
X (e ) =
n=
x (n )e
jn
1 x (m ) = 2
X (ej )ejmd
G (k ) =
n=0
g (n)ej 2nk/N
1 g (n ) = N
N 1
21