Scribd
Upload
21 views4 pages

Sol 1

1) The step response of a first-order system can be used to estimate the time constant T. The time at which the tangent to the step response reaches the steady state value equals T plus the time delay. 2) Estimating the impulse response hk of a system from its step response yt is approximated by the difference (yk - yk-1) divided by the step input a. This provides a rough estimate of the transfer function that is approximately equal to the estimate from spectral analysis. 3) For a second-order AR model with feedback input ut = -Kyt, the parameters a0 and b0 cannot be separately estimated from input-output data. Only their sum a0 + b

Uploaded by

anıl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
21 views4 pages

Sol 1

1) The step response of a first-order system can be used to estimate the time constant T. The time at which the tangent to the step response reaches the steady state value equals T plus the time delay. 2) Estimating the impulse response hk of a system from its step response yt is approximated by the difference (yk - yk-1) divided by the step input a. This provides a rough estimate of the transfer function that is approximately equal to the estimate from spectral analysis. 3) For a second-order AR model with feedback input ut = -Kyt, the parameters a0 and b0 cannot be separately estimated from input-output data. Only their sum a0 + b

Uploaded by

anıl
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Chapter 11

11.1 Dynamic Models


11.1.1 Exercises + Solutions
Exercise 1.0: Hello world
K
1. Given a discrete system G(z) = τ −z with K = 1 and τ = 2. Is it BIBO stable? Why/not?

Solution : No. The system has a pole outside the unit circle and therefore is unstable.
2. Given a system which outputs positive values for any input. Is it LTI? Why/not?
Solution: No. For LTI system if the input u(t) produces the output y(t) which is strictly
positive then −u(t) produces the output −y(t) which is negative.
3. Can you solve a least squares estimate for θ for a system satisfying xi θ = yi for any {(xi , yi )}i ?
Why/not?
Solution: No. The least square solution for the system xi θ = yi given {(xi , yi )}ni=1 is
Pn
xi yi
θ̂ = Pi=1
n 2
i=1 xi
Pn
in case i=1 x2i 6= 0.
4. Is the median estimate optimal in a least squares sense? Why/not?
Solution: No. The mean value (sample average) is optimal in a least square sense. The median
estimate is optimal for the absolute objective value.
5. If we are to model a certain behavior and we know some of the physics behind it - should we
go for a black box model? Why/not?
Solution: No. But we don’t have to. We better go for the white box model. The white
box model is formalized in terms of physical laws, chemical relations, or other theoretical
considerations.
6. If we have a very fast system (time constants smaller than O(10−2 )s). Can we get away with
slow sampling? Why/not?
Solution: No. A fast system does need a fast sampling (In general).

1
2 CHAPTER 11.

7. Does a non-causal model allow an impulse representation? Why/not?


Pn
Solution: No. From the impluse response of a system yt = k=0 hk ut−k one can see that the
output does not depend on future inputs.

8. Is a sequence of two nontrivial LTIs identifiable from input-output observations? Why/not?


Solution: No. Consider the following sequence of two LTIs

u z y
S1 S2

u z0 1
y
cS1 c S2

where c is a constant parameter. The two LTIs have the same input-output behavior but
different systems. Thus it is not identifiable from the input u and output y observations.

9. Is an ARMAX system linear in the parameters of the polynomials? Why/not?


Solution: No. For example the ARMAX model

y(t) + ay(t − 1) = bu(t − 1) + e(t) + ce(t − 1)

is not LIP, since c and e(t − 1) are unknown.

10. Is an OE model LIP? Why/not?


Solution: No. The OE model

B(q −1 )
y(t) = u(t) + e(t)
F (q −1 )

which can be written as F (q −1 )e(t) = F (q −1 )y(t) − B(q −1 )u(t). A simple example is

y(t) + f1 y(t − 1) = u(t − 1) + e(t) + f1 e(t − 1)

since f1 and e(t − 1) are unknown, f1 e(t − 1) is not LIP.

Exercise 1.2: Least Squares with Feedback


Consider the second-order AR model

yt + a0 yt−1 = b0 ut−1 + et

where ut is given by feedback as


ut = −Kyt .
Show that given realizations of this signal we cannot estimate a0 , b0 separately, but we can estimate
a0 + b0 K.
Solution: (Book p. 26)
11.1. DYNAMIC MODELS 3

Exercise 1.3: Determine the time constant T from a step response.


A first order system Y (s) = G(s)U (s) with
K
G(s) = e−sτ
1 + sT
or in time domain as a differential equation
dy(t)
T + y(t) = Ku(t − τ )
dt
derive a formula of the step response of an input ut = I(t > 0).

Solution: The step response of the system T dy(t)


dt + y(t) = Ku(t − τ ) is

o t<τ
y(t) =
K(1 − exp(−(t − τ )/T ))
the tangent at t = τ is given as
K
y 0 (t) =
(t − τ )
T
The tangent reaches the steady state value K at time t = τ + T .

Exercise 1.4: Step response as a special case of spectral analysis.


Let (yt )t be the step response of an LTI H(q −1 ) to an input ut = aI(t ≥ 0). Assume yt = 0 for
t < 0 and yt ≈ c for t > N . Justify the following rough estimate of H
yk − yk−1
ĥk = , ∀k = 0, . . . , N
a
and show that it is approximatively equal to the estimate provided by the spectral analysis.
Solution:
The response yt of the LTI system to the input ut is
t
X t
X
yt = hk ut−k = a hk
k=0 k=0

and since yt remains constant for values t > N it follows that


yt − yt−1
ht =
a
for t = 0, 1, 2, . . . , n. Since ht ≈ 0 for large n, thus the possible estimate of the transfer function is:
n
X
Ĥ(eiω ) = hk exp(−iωk)
k=0
n
1 X
= (yk − yk−1 ) exp(−iωk)
a
k=0
n
1 X 1 1
≈ yk exp(−iωk) − yk exp(−iωk)exp(−iω) = Yn (ω)(1 − exp(−iω))
a a a
k=0
4 CHAPTER 11.

Now
∞ ∞
X X a
U (ω) = uk exp(−iωk) = a exp(−iωk) =
1 − exp(−iω)
k=0 k=0

You might also like