EXTENDED STATE OBSERVER
Introduction
1. The Extended State Observer (ESO) [1] is a type of state and disturbance
observer. Apart from the states of the system, the ESO can also estimate
(a) Uncertainty in the states,
(b) Non-linearities in the system and
(c) External disturbances.
These uncertainties, non-linearities and external disturbances are treated as the
total uncertainty or disturbance.
Formulation
2. Consider an n-th order single-input-single-output (SISO) dynamical system
represented by
_____________________________(1)
________________________________________(2)
Where
is the coefficient matrix of the modelled part of
states,
is the matrix for uncertainty in states,
is the coefficient matrix for modelled part of input
is the uncertainty in input,
is the coefficient matrix for output,
w(t) is the external disturbance.=
3. Equation (1) can be rewritten as
____________________________(3)
where is called the total disturbance.
4. Now, define an extended state
_________________________________(4)
and __________________________________(5)
Thus equation (3) can be rewritten as
______________________(6)
___________________________________(7)
where
Design of Observer
6. Condition for Observability. Let the observer be of the form
____________________(8)
________________________________(9)
where L is the observer gain and
_______________________________(10)
Substituting equations (7) and (9) in (10) gives
_________________________(11)
Subtracting equation (8) from (6) and denoting
__________________________(12)
The observer error dynamics can be expressed as
__________________(13)
Using equation (11)and (12), the above equation can be rewritten as
__________________(14)
For a special case of h=0, the pair is observable. Thus it can be inferred
that for cases where rate of change of uncertainty asymptotically reduces to
zero, the pair is always observable. We now choose the observer
gain as
_____________________(15)
such that the observer dynamics is stable and has the desired transient
behaviour. In this case, the error, eo, will go to zero asymptotically, i.e.,
. This will ensure that .
7. To be more specific, the generalised form of the observer can be
expressed as
..........
____________________(16)
________________________________(17)
where gi(e) can be a linear function or a non-linear function. In case gi(e)is a
linear function then it is equated to e itself. In case gi(e)is chosen to be non linear
function, it has been selected heuristically based on experimental results as
_________(18)
Note:- In case of only one output,
and . Similarly, and .
Design of Control Law
8. Condition for Controllability. The controller to be designed is taken as
a pole-placement controller of the form
______________________________(19)
where . Substituting for u in equation (6) gives
__________________(20)
__________________(21)
For special case of h=0, the pair is controllable. Thus for all cases of h
reasonably small, the pair is always controllable and the above pole-
placement controller with gains K can be used in the feedback loop to achieve a
desired stable transient behaviour.
9. Observer-Controller Structure. In order to use the ESO estimated states
in the controller the choice of u is made as
______________________________(22)
where the controller gains, K, are designed using pole-placement as
________________________(23)
10. Corollary. Consider a second order pole placement problem were the
poles have to be placed as per the characteristic equation
____________________________(23)
______________________(24)
Equation (22) can be represented in the form of the coefficient matrix given by
________________(25)
For pole-placement, the controller gains have to be chosen such that
__________________________(26)
where
or
_________________________(27)
Thus, the control law given by equation (22) can be rewritten as
____________________(28)
11. In case of a tracking problem (regulator), the control law can be written as
______________(29)
where are the desired values of x.
Conclusion
12. This Extended State Observer has been found to be more robust and
superior in performance in comparison to other observers like high gain
observer, and sliding mode observer.[2].
[1] S.E. Talole and S.B. Phadke, ”Robust Missile Autopilot Design using Extended State Observer”, in Proc.
International Conference on Avionics Systems,(RCI, Hyderabad, India),pp.68-74, Feb 2008.
[2]W. Wang and Z. Gao, “A comparison study of advanced state observer design techniques,” in Proc. of the
American Control Conference, (Denver, Colarado, USA), pp. 4754–4759, June 2003.
