11. discrete-time control: background1,2pcsl.ecs.umass.edu/cbs/handouts/ch11_discrete_time... ·...
TRANSCRIPT
3/22/2005 11-1 Copyright ©2005 (Yossi Chait)
11. Discrete-Time Control: Background1,2
Discrete-time control system are hybrid: part continuous time and part continuous time.
∑ PCDI
Y-FR
DOUE
∑ ∑
In studying how to analyze such systems we’ll visit: • Impulse sampling and zero-order hold
• z transform
• Stability
• Design 1. Ogata, K., Discrete-Time Control Systems, Prentice-Hall, Inc., 1987.2. Houpis, CH., and Rasmussen, SJ., Quantitative Feedback Theory Fundamentals and Applications, Marcel
Dekker AG, 1999.
3/22/2005 11-2 Copyright ©2005 (Yossi Chait)
11.1. Sampling and Hold
An ideal sampler comprises of a switch that closes to admit an input signal every sampling period T. The finite duration of the sampling is assumed infinitesimal. This is called analog-to-digital(A/D) conversion. A digital-to-analog (A/D) conversion converts to sampled-data signal back to a continuous-time signal – typically via a zero-order hold (ZOH) circuit.
tt kT
3/22/2005 11-3 Copyright ©2005 (Yossi Chait)
For a signal zero at t < 0, h(t) is related to x(t) as follows
( ) ( ) ( )[ ] ( ) ( )[ ]( ) ( )[ ]
(0) 1 1 ( ) 1 1 2
(2 ) 1 2 1 3
h t x t t T x T t T t T
x T t T t T
= − − + − − −
+ − − − +…
The Laplace transform of a delayed step is
and for the ZOH we jave
( )[ ] ( )( )1
0
( )k TskTs
k
e eh t H s x kT
s
∞ − +−
=
−= = ∑L
3/22/2005 11-4 Copyright ©2005 (Yossi Chait)
We define
The sampled signal is a train of impulses (the strength of each equals x(t) at kT)
where ( ) 0 unless .t kT t kTδ − = =
Nyquist (Shannon’s) sampling theorem says:
A function f(t) which contains no frequency components greater than ωc (i.e., band limited) can be represented by x(kT) with T < π/ωc.
3/22/2005 11-5 Copyright ©2005 (Yossi Chait)
In addition to assuming no aliasing, we assume that quantization, saturation, and conversion errors in the A/D conversion are negligible.
11.2. The z Transform
Define
so we can write
The machinery developed for Laplace transform domain, such frequency response, root locus, and stability analysis is readily applicable in the z domain for discrete-time control system having sampling as discussed above.
3/22/2005 11-6 Copyright ©2005 (Yossi Chait)
Examples of z transforms:
( ) ( )1, 0
00, 0
kx t
k=⎧
= δ = ⎨ ≠⎩
so
The unit step function
( ) ( )1x t t=
has a z transform of
3/22/2005 11-7 Copyright ©2005 (Yossi Chait)
A table of z transforms is provided.
Important Properties.
•
•
• Final Value Theorem: if X(z) has only stable poles (inside the unit circle) with the exception of one at z = 1, then the final value is
A table of important z transform properties is provided.
3/22/2005 11-8 Copyright ©2005 (Yossi Chait)
Inverse z transform.
• direct division
• partial fraction expansion
• other methods
( )( ) ( )
1 2
1 210 5 10 5
.1 .2 1 1.2 0.2
z z zX z
z z z z
− −
− −+ +
= =− − − +
Example. Find x(k) for
1 2 1 2
1 2 3
2 3
2 3 4
1 1.2 0.2 10 5 10 12 2 17 2 17 20.4 3.4
z z z zz z z
z zz z z
− − − −
− − −
− −
− − −
− + +− +
−− +
3 4
3 4 5
4 5
18.4 3.4 18.4 22.08 3.68 18.68 3.68
z zz z z
z z
− −
− − −
− −
−− +
−4 5 6 18.68 22.416 3.736z z z− − −− +
1 2 3 410 17 18.4 16.86z z z z− − − −+ + + +
3/22/2005 11-9 Copyright ©2005 (Yossi Chait)
Since
( )0
( ) ,k
k
X z x kT z∞
−
== ∑
direct comparison of this infinite series with the long division gives
Example. Find x(k) for ( )( ) ( )
101 .2
zX z
z z=
− −
If X(z) has one or more zeros at the origin, use this trick
3/22/2005 11-10 Copyright ©2005 (Yossi Chait)
From z transform tables
Note: if we did not divide by z, the expansion of x(z) would yield terms not appearing in the table.
3/22/2005 11-11 Copyright ©2005 (Yossi Chait)
11.3. The Pulse Transfer Function
Here we study transfer functions in the z domain. Consider the system shown below.
The continuous-time convolution integral becomes a convolution integral in discrete time
∞
( ) ( ) ( )
( ) ( ) ( ) ( )
0
0
, 0.
h
h
y kT g kT T x hT
x kT hT g hT x k h g k h
=∞
=
= −
= − < = < =
∑
∑The z transform of y(kT) is
( ) ( ) ( ) ( )0 0 0
k
k k h
Y z y kT z g kT hT x hT∞ ∞ ∞
−
= = == = −∑ ∑ ∑
3/22/2005 11-12 Copyright ©2005 (Yossi Chait)
( ) ( ) ( ) ( ) ( )0 0 0 0
( ) ( ), .
m h m h
m h m h
g mT x hT z g mT z x hT z
G z X z m k h
∞ ∞ ∞ ∞− + − −
= = = == =
= = −
∑ ∑ ∑ ∑
The pulse transfer function is
( )( )
.Y z
GX z
=
Back to the original system.
To relate sampled signals, we note that
( ) ( )* * , 0,1,2,sX s X s j k k= ± ω = …
so
3/22/2005 11-13 Copyright ©2005 (Yossi Chait)
and the starred Laplace transform with becomessTz e=
( ) ( ) ( ).Y z G z X z=
Note that in this system
That is
How do we obtain a pulse TF for a system? Typically, we use basic block diagram algebra and z-Transform Tables (see attachments).
Example. Find the PTF of
( )1 1
( )1
TseG s
s s s
−−=
+
3/22/2005 11-14 Copyright ©2005 (Yossi Chait)
( ) ( )( )
( )
( )( )
( ) ( )( )( )
2 2
1 2
1 12 1 11
1 1 1 1 1 11 11 11 11
1 11 11 1 .1 11 11
( )
T T T
TT
Tsez z
s s s s ss s s
T e z e Te zTzz
z e zz e zz
G z
− − − − −
− − −− − −−
− ⎡ ⎤⎡ ⎤− ⎡ ⎤− −− = − + +⎢ ⎥⎢ ⎥ ⎢ ⎥⎣ ⎦+ ++⎣ ⎦ ⎣ ⎦
⎡ ⎤ − + + − −⎢ ⎥−− + + =⎢ ⎥ − −− −−⎢ ⎥⎣ ⎦
=
=
=Z Z Z
Example. Find the PTF of a closed-loop discrete-time control system shown below.
( )e t∑
( )r t ( )e kT
T( )G s
( )c t
−
( )H s
*E R HC R HGE= − = −
Starring both sides
[ ]
*
* *1
1
ER HG
=+
3/22/2005 11-15 Copyright ©2005 (Yossi Chait)
Combining the last 2 equations gives
Note: we have assumed all samples have same sampling period and are synchronized.
Some systems do not have a PTF. This occurs when the input signal dynamics cannot be decoupled from the dynamics of the system.
∑
( )r t
T
( )G s( )c t
−
( )H s
3/22/2005 11-16 Copyright ©2005 (Yossi Chait)
Example.
E∑
R *ET
KC
−( )ZOH s 1
1s+
plantzero-order
holddigital
controller
*C kGEE R C== −
( ) ( ) ( ) 11 1
1 1 1
11 1
1 1 1 1
T
T T
k e zk k k kz z
s s z e z e z
− −− −
− − − − −
−⎡ ⎤ ⎡ ⎤= − − = − − =⎢ ⎥ ⎢ ⎥+ − − −⎣ ⎦ ⎣ ⎦Z
( )( )( )
1
1
1.
1 1
T
T
k e zCz
R k k e z
− −
− −
−⇒ =
⎡ ⎤− − +⎣ ⎦
3/22/2005 11-17 Copyright ©2005 (Yossi Chait)
Example. E∑
R *ET
PIDC
−( )ZOH s ( )
11s s+
( ) ( ) ( )1 111
11
1d
p i d psi
T KPID s K K K s PID z K z
K Tz− −
−= + + ⇒ = + + −−
( ) ( ) ( )[ ] ( ) ( ) ( ) ( )1 2
12 1 1
1 .3697 .26421 .
1 1 .3697 1z z
G z ZOH s G s zs s z z
− −−
− −
⎡ ⎤ += = − =⎢ ⎥+ − −⎣ ⎦Z Z
1 0, 1.d i pK K K−= = =Let
( ) ( )( )
1 2 1
1 2 2 1.3697 .2642 .3697 .2642
.1 1 .6321 .6321
p
p
K G zC z z zz
R K G z z z z z
− −
− −+ +
= = =+ − + − +
3/22/2005 11-18 Copyright ©2005 (Yossi Chait)
Step Scope
1
Gain
num(z)
([1 -.3679],[1 -Discrete
Transfer Fcn1
0 5 10 15 200
0.2
0.4
0.6
0.8
1
1.2
1.4
sec
c(kT
)
3/22/2005 11-19 Copyright ©2005 (Yossi Chait)
For a unit step
( )1 2
1 2 1
1 2 3 4 5 6
.3697 .2642 11 .6321 1
.3679 1.3996 1.3996 1.1469 .8944
z zC z
z z zz z z z z z
− −
− − −
− − − − − −
+=
− + −= + + + + + +…
Since z-1 implies time shift by one sampling period, taking inverse z transform gives
Note: it is possible to compute the response between sampling instances.
3/22/2005 11-20 Copyright ©2005 (Yossi Chait)
11.4. s-plane to z-plane Mapping
Stability and performance of continuous-time (CT) systems depends on pole location. Since s and z are related by z = eTs, we can study these discrete-time properties of a PTF by related z domain pole location via the map.
Poles and zeros in the s plane are mapped to the z plane via z = eTs. Denote the complex number s = σ+jω so
( ) ( )2 , 0, 1,T j j T kT jT Tz e e e e e kσ+ ω ω+ πσ ω σ= = = = ± …
Note that s plane frequencies with integer multiple of ωs difference are mapped into the same z plane location.
Stable systems have al their poles in the open left-half s plane, or in the z plane
The jω-axis in the s plane maps into |z| = 1 circle.
3/22/2005 11-21 Copyright ©2005 (Yossi Chait)
The phase of z, ωT, varies from -∞ to ∞ as ω varies from -∞ to ∞. In particular, along the jω axis, as ω varies from -½ωs to -½ωs we have |z|=1 and ∠z varies CCW from –π to π. From -½ωs to -½ωs the phase ∠z varies again CCW from –π to π. This is depicted below.
planez
2sj ω−
jω
σ
planes
2sj ω
2sj ω−
sjω
32
sj ω
Im
Re
primary2
sj ω
−∞
jω
σ
Im
Re
3/22/2005 11-22 Copyright ©2005 (Yossi Chait)
planezplanes
ω
1−σ 2σ0
j Im
1Reσ
0 1
Im
Reσ
jω2
sj ω
2sj ω−
2jω1jω
-1 -0.5 0 0.5 1-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0
jω
2sj ωsjω
σ
3/22/2005 11-23 Copyright ©2005 (Yossi Chait)
-1 -0.5 0 0.5 10
0.2
0.4
0.6
0.8
1
planez
3/22/2005 11-24 Copyright ©2005 (Yossi Chait)
11.4. Stability Analysis
Discrete-time stability is determined from the roots of the characteristic equation 1+L(z)=0.
3/22/2005 11-25 Copyright ©2005 (Yossi Chait)
11.4.1 Stability tests
• Jury Stability Test an algebraic test using the characteristic equation.
• Routh test via bilinear transformation.
jω
σ
planesIm
planez planew
j∞
Re
Im
j− ∞
2sj ω
2sj ω−
Re11−
( ) 1 11 1 00 1 n n
n nL z a z a z a z a−−= + = + + + +1 1
1 1 01 1 1
01 1 1
n n
n nw w w
a a a aw w w
−
−+ + +⎛ ⎞ ⎛ ⎞ ⎛ ⎞+ + + + =⎜ ⎟ ⎜ ⎟ ⎜ ⎟− + − + − +⎝ ⎠ ⎝ ⎠ ⎝ ⎠