UCK363E – AUTOMATIC CONTROL II
FALL ‘21
LECTURE 4
STATE SPACE MODEL FEATURES AND OBSERVABILITY
Instructor: Barış Başpınar
Zeros in State Space Models
Roots of transfer function numerator called the system zeros
Need to develop a similar way of defining/computing them using a state space model
Zero: generalized frequency 𝑠0 for which the system can have a non-zero input 𝑢(𝑡) =
𝑢0 𝑒 𝑠0𝑡 , but exactly zero output 𝑦(𝑡) ≡ 0 ∀𝑡
Note that there is a specific initial condition associated with this response 𝑥0 , so the state response is of
the form 𝑥(𝑡) = 𝑥0 𝑒 𝑠0 𝑡
Given 𝑥ሶ = 𝐴𝑥 + 𝐵𝑢, substitute the above to get:
Zeros in State Space Models
Also have that 𝑦 = 𝐶𝑥 + 𝐷𝑢 = 0 which gives:
So we must find the 𝑠0 that solves:
There is a zero at the frequency 𝑠0 if there exists a non-trivial solution of:
𝑥0
Key point: Zeros have both direction 𝑢 and frequency 𝑠0
0
Just as we would associate a direction (eigenvector) with each pole (frequency 𝜆𝑖 )
Zeros in State Space Models
Example:
so there is clearly a zero at 𝑠0 = −2, as we expected. For the directions, solve
gives 𝑥01 = −2𝑥02 and 𝑢0 = 2𝑥02 so that with 𝑥02 = 1
Zeros in State Space Models
Further observations: apply the specified control input in the frequency domain, so that
2
where 𝑢 = 2𝑒 −2𝑡 , so that 𝑈 𝑠 = 𝑠+2
say that 𝑠 = −2 is a blocking zero or a transmission zero
The response 𝑌1 (𝑠) is clearly non-zero, but it does not contain a component at the input
frequency 𝑠 = −2
That input has been “blocked”
Zeros in State Space Models
Note that the output response left in 𝑌1 (𝑠) is of a very special form – it corresponds to
the (negative of the) response you would see from the system with 𝑢 𝑡 = 0 and 𝑥0 =
−2 1 𝑇
So then the total output is 𝑌 𝑠 = 𝑌1 𝑠 + 𝑌2 (𝑠) showing that 𝑌 𝑠 = 0 → 𝑦 𝑡 = 0, as
expected
State-Space Model Features
There are some key characteristics of a state-space model that we need to identify
Will see that these are very closely associated with the concepts of pole/zero cancelation in transfer
functions
Example: Consider a simple system:
for which we develop the state-space model
But now consider the new state space model 𝑥ҧ = 𝑥 𝑥2 𝑇
which is clearly different than the first model, and larger.
But let’s looks at the transfer function of the new model ⇒
State-Space Model Features
This is a bit strange, because previously our figure of merit when comparing one state-
space model to another was whether they reproduced the same transfer function
Now we have two very different models that result in the same transfer function
Note that I showed the second model as having 1 extra state, but I could easily have done it with 99
extra states!!
So what is going on?
A clue is that the dynamics associated with the second state of the model 𝑥2 were eliminated when we
formed the product
because the 𝐴 is decoupled and there is a zero in the 𝐶 matrix
෨
Which is exactly the same as saying that there is a pole-zero cancelation in the transfer function 𝐺(𝑠)
* For this system we say that the
෨
Note that model #2 is one possible state-space model of 𝐺(𝑠) (has 2 poles) dynamics associated with the
second state are unobservable
using this sensor (defines C matrix)
State-Space Model Features
There is an analogous problem on the input side as well. Consider:
𝑇
with 𝑥ҧ = 𝑥 𝑥2
which is also clearly different than model #1, and has a different form from the second model
State-Space Model Features
Once again the dynamics associated with the pole at 𝑠 = −1 are canceled out of the
transfer function
But in this case it occurred because there is a 0 in the 𝐵 matrix
So in this case we can “see” the state 𝑥2 in the output 𝐶 = 3 2 , but we cannot
“influence” that state with the input since B = 2 0 𝑇
So we say that the dynamics associated with the second state are uncontrollable using
this actuator (defines the B matrix)
State-Space Model Features
Of course it can get even worse because we could have
So now we have
Get same result for the transfer function, but now the dynamics associated with 𝑥2 are
both unobservable and uncontrollable
State-Space Model Features
Summary: Dynamics in the state-space model that are uncontrollable, unobservable,
or both do not show up in the transfer function
Would like to develop models that only have dynamics that are both controllable and
observable
⇒ called a minimal realization
A state space model that has the lowest possible order for the given transfer function
But first need to develop tests to determine if the models are observable and/or
controllable
State-Space Model Features
Summary: Dynamics in the state-space model that are uncontrollable, unobservable,
or both do not show up in the transfer function
Would like to develop models that only have dynamics that are both controllable and
observable
⇒ called a minimal realization
A state space model that has the lowest possible order for the given transfer function
But first need to develop tests to determine if the models are observable and/or
controllable
Observability
Definition: An LTI system is observable if the initial state 𝑥 0 can be uniquely
deduced from the knowledge of the input 𝑢(𝑡) and output 𝑦(𝑡) for all 𝑡 between 0 and
any finite 𝑇 > 0
If 𝑥(0) can be deduced, then we can reconstruct 𝑥 𝑡 exactly because we know 𝑢 𝑡 ⇒ we can find
𝑥 𝑡 ∀𝑡
Thus we need only consider the zero-input (homogeneous) solution to study observability
Definition: A state 𝑥 ∗ ≠ 0 is said to be unobservable if the zero-input solution 𝑦(𝑡),
with 𝑥(0) = 𝑥 ∗ , is zero for all 𝑡 ≥ 0
Equivalent to saying that 𝑥 ∗ is an unobservable state if
Observability
For the problem we were just looking at, consider Model #2 with 𝑥 ∗ = 0 1 𝑇
≠ 0, then
so
So, 𝑥 ∗ = 0 1 𝑇
is an unobservable state for this system
But that is as expected, because we knew there was a problem with the state 𝑥2 from the
previous analysis
Observability
Theorem: An LTI system is observable iff it has no unobservable states
We normally just say that the pair (A,C) is observable
Pseudo-Proof: Let 𝑥 ∗ ≠ 0 be an unobservable state and compute the outputs from the
initial conditions 𝑥1 (0) and 𝑥2 (0) = 𝑥1 (0) + 𝑥 ∗
Then,
and
but
Thus 2 different initial conditions give the same output 𝑦(𝑡), so it would be impossible
for us to deduce the actual initial condition of the system 𝑥1 (𝑡) or 𝑥2 (𝑡) given 𝑦1 (𝑡)
Testing system observability by searching for a vector 𝑥(0) such that 𝐶𝑒 𝐴𝑡 𝑥 0 = 0 ∀ 𝑡
is feasible, but very hard in general
Better tests are available
Observability
Theorem: The vector 𝑥 ∗ is an unobservable state iff
Pseudo-Proof: If 𝑥 ∗ is an unobservable state, then by definition,
But all the derivatives of 𝐶𝑒 𝐴𝑡 exist and for this condition to hold, all derivatives must
be zero at 𝑡 = 0
Observability
Then
We only need retain up to the 𝑛 − 1𝑡ℎ derivative because of the Cayley-Hamilton
theorem
Observability
Simple test: Necessary and sufficient condition for observability is that
Why does this make sense?
The requirement for an unobservable state is that for 𝑥 ∗ ≠ 0
Which is equivalent to saying that 𝑥 ∗ is orthogonal to each row of
But if the rows of are considered to be vectors and these span the full 𝑛-dimensional space, then it
is not possible to find an 𝑛-vector 𝑥 ∗ that is orthogonal to each of these
To determine if the 𝑛 rows of span the full 𝑛-dimensional space, we need to test their linear
independence, which is equivalent to the rank test1
References
Jonathan P. How and Emilio Frazzoli. 16.30/31 Feedback Control Systems. Fall 2010.
Massachusetts Institute of Technology: MIT OpenCourseWare, https://ocw.mit.edu/.
License: Creative Commons BY-NC-SA.
Katsuhiko Ogata, Modern Control Engineering, 2010, 5th edition, Prentice Hall.
Norman Nice, Control Systems Engineering, 2011, Addison-Wesley.
