Professional Documents
Culture Documents
1
Frequency Response Transfer Function
r e u y B A
Σ C(s) P(s
L(s)
)
-1
-1
a b
Figure 1. The Loop Transfer Function
Course Module
Feedback and Control Systems
2
Frequency Response Transfer Function
Shown in Figure 1are the method to determine the stability of the system. Take Figure 1a as
the system to determine its stability. By traditional method, the closed loop characteristic
polynomial is determined if all its roots are in the left half plane. If the process and controller
have rational transfer functions P(s) = np(s)/dp(s) and C(s) = nc(s)/dc(s) then the closed loop
system ha s a transfer function
𝑃𝐶 𝑛𝑝 (𝑠)𝑛𝑐 (𝑠)
𝐺𝑦𝑟 (𝑠) = =
1+𝑃𝐶 𝑑𝑝 (𝑠)𝑑𝑐 (𝑠)+𝑛𝑝 (𝑠)𝑛𝑐 (𝑠)
To determine stability, the roots of the characteristic polynomial are computed and
determine if they each have negative real part. This method is direct but it cannot be a design
guide and does not give how the controller be modified to stabilize a system.
Nyquist’s idea was to determine conditions under which oscillations can occur in a feedback
loop. The loop transfer function L(s) = P(s)C(s), the transfer function obtained by breaking
the feedback loop is introduced as shown in Figure 1b. The loop transfer function is just the
transfer function from the input in position A to the output at position B.
The conditions for having a periodic oscillation in the loop are first determined. It is assumed
that a sinusoid of frequency ω0 is injected at point A. In steady state, the signal at point B will
also be a sinusoid with the frequency ω0. An oscillation can be maintained if the signal at B
has the same amplitude and phase as the injected signal, because A can then be connected to
B. After the signals around the loop were traced, the signals at A and B were found to be
identical if
Eq. 1. L(iω0) = -1
which provides a condition to maintain oscillation. The important idea of Nyquist stability
criterion is to understand when this can happen in a general setting.
An example is to determine the loop transfer function for operational amplifier as shown in
Figure 2.
Source: (275)
Z1 and Z2 are the transfer functions from voltage to current of the feedback elements. There
is feedback because the voltage v2 is related to the voltage v through a transfer function –G
describing the op amp dynamics and the voltage v is related to the voltage v 2 through the
transfer function Z1/(Z1 + Z2). The loop transfer function is then
Course Module
Feedback and Control Systems
3
Frequency Response Transfer Function
𝐺𝑍1
𝐿= 𝑍1 +𝑍2
The current I is assumed to be zero, the current through the elements Z 1 and Z2 is the same
which implies
𝑣1 −𝑣 𝑣−𝑣2
=
𝑍1 𝑍2
The block diagram is shown in Figure 2b. It follows from Equation 1 that the condition for
oscillation of the op amp circuit is
𝑍1 (𝑖𝜔)𝐺(𝑖𝜔)
𝐿(𝑖𝜔) = = −1
𝑍1 (𝑖𝜔)+𝑍2 (𝑖𝜔)
One of the concepts in Nyquist approach to stability analysis is that it allows the study of the
stability of the feedback system by looking at the properties of the loop transfer function.
The advantage of this is it is easy to see how the controller should be chosen to obtain the
desired transfer function. An example is if the gain of the controller is changed, the loop
transfer function will be balanced accordingly. The simple way to stabilize an unstable
system is to reduce the gain so that the -1 point is avoided. Another way is to introduce a
controller with the property that it bends the loop transfer function away from the critical
point.
Course Module
Feedback and Control Systems
4
Frequency Response Transfer Function
then the portion of the contour “at infinity” maps to the origin. Also, the portion of the
plot corresponding to ω < 0 is the mirror image of the portion with ω>0.
Source: Feedback Systems: An Introduction for Scientists and Engineers(276), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
There is restraint with the Nyquist plot when the loop transfer function has
poles on the imaginary axis because the gain is infinite at the poles. To solve
this problem, the contour Γ is modified to include small deviations that avoid
any poles on the imaginary axis as in Figure 3a (assuming a pole of L(s) at the
origin). The deviation consists of a small semicircle to the right of the
imaginary axis pole location.
The condition for the oscillation as given in Equation 1 implies that the Nyquist
plot of the loop transfer function goes through the point L = -1, which is called
the critical point. Assume ωc represents a frequency at which ∠L(iωc) = 1800,
which corresponds to the Nyquist curve crossing the negative real axis. It is
equitable that the system is stable if |L(iωc)| < 1, which means that the critical
point, -1 is on the left hand side of the Nyquist curve as in Figure 3b. This
means that the signal at point B will have a similar amplitude than the injected
signal.
Theorem 1. (Simplified Nyquist criterion). Let L(s) be the loop transfer function
for a negative feedback system and assume that L has no poles in the closed right
half plane (Res ≥0), except for single poles on the imaginary axis. Then the closed
loop system is stable if and only if the closed contour given by Ω = {L(iω) : - ∞ <
ω <∞} ⊂ ℂ has no net encirclements of s = -1.
In the theory of complex functions, it is customary to encircle the Nyquist
contour in the counter clockwise direction, which means that the imaginary
Course Module
Feedback and Control Systems
5
Frequency Response Transfer Function
The plot is as shown in Figure 4 with the points corresponding to ω > 0 drawn
as solid line and ω < 0 as a dashed line. It can be noted that these curves are
mirror images of each other.
Source: Feedback Systems: An Introduction for Scientists and Engineers(278), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
For the completion of the Nyquist plot, L(s) is computed for s on the outer arc
of the Nyquist D contour. This arc has the form 𝑠 = 𝑅𝑒 𝑗𝜃 for R→∞ and gives
1
𝐿(𝑅𝑒 𝑗𝜃 ) = 3 → 0 𝑎𝑠 𝑅 → ∞
(𝑅𝑒 𝑗𝜃 +𝑎)
Therefore the outer arc of the D contour maps to the origin on the Nyquist plot.
Another method of computing the Nyquist plot is to determine the plot from
the frequency response (Bode plot), which gives the Nyquist curve for s = iω,
ω = 0. G(iω) is first plotted from ω = 0 to ω = ∞, which is read off from the
magnitude and phase of the transfer function. Then 𝐺(𝑅𝑒 𝑗𝜃 ) is plotted with θ
∈ [0,π/2] and R →∞, which always maps to zero. The remaining parts of the
plot can be determined by taking the mirror image of the curve. The plot can
Course Module
Feedback and Control Systems
6
Frequency Response Transfer Function
Where the gain has the minimal value k = 1. The Bode plot is shown in Figure
5a. The system has a single pole at s = 1 and a double pole at s = -1. The gain
curve of the Bode plot then has the slope -1 for low frequencies and at the
double pole s = 1 the slope changes to -3. For small s the result is L≈k/s which
means that the low frequency asymptote intersects the unit gain line at ω = k.
The phase curve starts at -900 for low frequencies, it is -1800 at the break point
ω =1 and it is -2700 at high frequencies.
After getting the Bode plot, the Nyquist plot is sketched as shown in Figure 5b.
It starts with a phase of -900 for low frequencies, intersects the negative real
axis at the breakpoint ω = 1 where L(i) = 0.5 and goes on to zero along the
imaginary axis for high frequencies. The small half circle of the Γ contour at
the origin is mapped on a large circle enclosing the right half plane. The
Nyquist curve does not encircle the critical point and it follows from the
simplified Nyquist theorem that the closed loop is stable. Since L(i) = -k/2 the
system is found to become unstable if the gain is increased to k = 2 or beyond.
Source: Feedback Systems: An Introduction for Scientists and Engineers(279), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
The Nyquist criterion does not require that |L(iω)| < 1 for all ω corresponding
to a crossing of the negative real axis. It rather says that the number of
encirclements must be zero which allows the possibility that the Nyquist curve
could cross the negative real axis and cross back at magnitudes greater than 1.
One advantage of Nyquist criterion is that it tells how a system is influenced
by changes of the controller parameters. An example is that it is easy to
Course Module
Feedback and Control Systems
7
Frequency Response Transfer Function
visualize what happens when the gain is changed since this scales the Nyquist
curve.
Another example is the congestion control system in an Internet. Assuming
there are N identical sources and a disturbance d representing an external data
source as shown in Figure 6a. Time delay between the router and the senders
are included, representing the time delays between the sender and receiver.
To analyze the stability of the system, the transfer functions used are
𝑓
𝑁𝑤𝑒 𝑒 −𝜏 𝑠 𝑁
𝐺𝑏𝑤 (𝑠) = 𝑓 , 𝐺𝑤𝑞 (𝑠) = 𝑞𝑒 (𝜏𝑒 𝑠+𝑞𝑒 𝑤𝑒 )
, 𝐺𝑝𝑏 (𝑠) = 𝜌
𝜏𝑒 𝑠+𝑒 𝜏 𝑠
Where (we, be) is the equilibrium point for the system, N is the number of
sources, τe is the steady state round trip time and τf is the forward propagation
time.
Source: Feedback Systems: An Introduction for Scientists and Engineers(280), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
2𝑁 2𝑁3 𝑏𝑒 𝜏𝑒 𝑐
It is known that 𝑞𝑒 ≈ 𝑤 2 = (𝜏 2 and 𝑤𝑒 = = , it shows that
𝑒 𝑒 𝑐) 𝑁 𝑁
𝑁 𝑐 3 𝜏3
𝑒 −𝜏𝑒 𝑠
𝐿(𝑠) = 𝜌. 𝑓 . 2𝑁3 (𝑐𝜏2 𝑠+2𝑁 2) 𝑒
𝜏𝑒 𝑠+𝑒 𝜏 𝑠 𝑒
The exponential term representing the time delay gives significant phase
above ω = 1/τ and the gain at the crossover frequency will determine stability.
To check stability it is required that the gain is sufficiently small at crossover.
If it is assumed that the pole due to the queue dynamics is sufficiently fast that
the TCP dynamics are dominant, the gain at the crossover frequency ω is given
by
Course Module
Feedback and Control Systems
8
Frequency Response Transfer Function
𝑐 3 𝜏3 𝜌𝑐 2 𝜏𝑒
|𝐿(𝑖𝜔𝑐 )| = 𝜌. 𝑁. 3 𝑒2 =
2𝑁 𝑐𝜏 𝜔 𝑒 𝑐 2𝑁𝜔𝑐
Using the Nyquist criterion, the closed loop system will be unstable if this
quantity is greater than 1. In particular, for a fixed time delay, the system will
become unstable as the link capacity c is increased. This means that the TCP
protocol may not be scalable to high capacity networks.
Conditional Stability
There are instances that unstable systems can be stabilized by simply
reducing the loop gains. However, there are also instances that systems
cannot be stabilized by increasing the gain. This is common in the
design of feedback amplifiers calling this as conditional stability. An
example is a feedback system with the loop transfer function as
3(𝑠+1)2
𝐿(𝑠) = 𝑠(𝑠+6)2
Source: Feedback Systems: An Introduction for Scientists and Engineers(281), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
Course Module
Feedback and Control Systems
9
Frequency Response Transfer Function
Stability Margins
In reality, it is not enough that the system is stable. There must also be margins of stability
that describe how stable the system is and its strengths to agitations. The most common way
to show this is to use gain and phase margins from the Nyquist’s stability criterion. This is
because it is easy to plot the loop transfer function L(s). An increase of controller gain
expands the Nyquist plot radially and an increase in phase of the controller twists the
Nyquist plot clockwise. Thence from the Nyquist plot, it can be pointed out the amount of
gain or phase that can be added without causing the system to be unstable.
Assume that ωpc as the phase crossover frequency, the smallest frequency where the phase
of the loop transfer function L(s) is -1800. The gain margin is defined as
1
Eq. 2 𝑔𝑚 =
|𝐿(𝑖𝜔𝑝𝑐 )|
This indicate how much the controller gain can be increased before reaching the stability
limit.
Also assume that ωgc is the gain crossover frequency, the lowest frequency where the loop
transfer function L(s) has unit magnitude. The phase margin is
Eq. 3 ∅𝑚 = 𝜋 + arg 𝐿(𝑖𝜔𝑔𝑐 )
Course Module
Feedback and Control Systems
10
Frequency Response Transfer Function
The amount of phase lag required to reach the stability limit. These margins have simple
geometric interpretations in the Nyquist diagram of the loop transfer function as shown in
Figure 8a.
Source: Feedback Systems: An Introduction for Scientists and Engineers(285), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
There is a disadvantage with gain and phase margins, it is necessary to give both of them to
guarantee that the Nyquist curve is not close to the critical point. An alternative method is to
express margins by a single number, the stability margin, sm, which is the shortest distance
from the Nyquist curve to the critical point.
Gain and phase margins can be determined from the Bode plot of the loop transfer function.
To determine the gain margin, the phase crossover frequency ω pc where the phase is -1800
is solved. The gain margin is the inverse of the gain at that frequency. On the other hand, to
find phase margin, the gain crossover frequency ωgc, where the gain of the loop transfer
function is 1 is determined. The phase margin is the phase of the loop transfer function at
that frequency plus 1800. Figure 8b shows how the margins are determined in the Bode plot
of the loop transfer function.
Course Module
Feedback and Control Systems
11
Frequency Response Transfer Function
𝜋 ∞ 𝑑𝑙𝑜𝑔|𝐺(𝑖𝜔)| 𝜋 𝑑𝑙𝑜𝑔|𝐺(𝑖𝜔)|
Eq. 4 arg 𝐺(𝑖𝜔0 ) = ∫ 𝑓(𝜔) 𝑑𝑙𝑜𝑔𝜔 ≈
2 0 𝑑𝑙𝑜𝑔𝜔 2 𝑑𝑙𝑜𝑔𝜔
The phase curve is then a weighted average of the derivative of the gain curve. If the gain
curve has constant slope n the phase curve has the constant value π/2.
Equation 4, Bode’s relation, is used for systems with no poles and zeros at the right half plane.
Such systems are called minimum phase system since systems with poles and zeros in the
right half plane have larger phase lag. Minimum phase systems are easier to control than
systems with larger phase lag.
The transfer function of a time delay of Td units is 𝐺(𝑠) = 𝑒 𝑠𝑇𝑑 . This transfer function then
has unit gain, |G(iω)| = 1, and the phase is arg G(iω) = -ωTd. The corresponding minimum
phase system with unit gain has the transfer function G(s) – 1. The time delay then has an
additional phase lag of ωTd. It can be noted that the phase lag increases linearly with
frequency as shown in Figure 9.
Source: Feedback Systems: An Introduction for Scientists and Engineers(289), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
Figure 9b shows a system with the transfer function G(s) = (a-s)(a+s) with a>0, which has a
zero s = a in the right half plane. The transfer function has unit gain, |G(iω)| = 1, and the phase
is arg G(iω) = 2 arctan (ω/a). the corresponding minimum phase system with a unit gain has
the transfer function G(s) = 1. While Figure 9c shows the analysis of the transfer function G(s)
= (s + a)/(s – a) with a > 0, which has a pole in the right half plane, shows that its phase is arg
G(iω) = -2 arctan (a + ω).
Having poles and zeros in the right half plane introduces severe limitations on the quality of
performance. This can be avoided by redesigning the system whenever possible. While poles
are intrinsic properties of the system and do not depend on sensors and actuators, the zeros
depend on how the inputs and outputs of the system are coupled to the states. Thus, zeros
can be changed by moving sensors and actuators or by introducing new sensors and
actuators. However, non-minimum phase systems are common in practice.
Course Module
Feedback and Control Systems
12
Frequency Response Transfer Function
The eigenvalues of the matrix A*A are called singular values of the matrix A and the
largest singular value is denoted 𝜎̅ (A).
To generalize this case of an input/output dynamical system, the inputs and outputs
should be thought of not as vectors of real numbers but the vectors of signals.
Consider the case of scalar signals and the signal space L2 as square integrable
functions with the norm
∞
‖𝑢‖2 = √∫0 |𝑢|2 (𝜏)𝑑𝜏
This definition can be generalized to vector signals by replacing the absolute value
with the vector norm in Equation 5. The gain of the system can be defined by taking
inputs 𝑢 ∈ 𝐿2 and producing outputs 𝑦 ∈ 𝐿2 as
‖𝑦‖
Eq. 6 𝛾 = sup ‖𝑢‖
𝑢=𝐿2
Where sup is the supremum, the smallest number that is larger than its argument.
Supremum is used because the maximum may not be defined for 𝑢 ∈ 𝐿2 . This
definition of the system gain is general and can be used for some classes of nonlinear
systems. The norm described in Equation 6 has good properties in the case of linear
Course Module
Feedback and Control Systems
13
Frequency Response Transfer Function
system. Given a stable linear system with transfer function G(s), it can be shown that
the norm of the system is given by
𝛾 = 𝑠𝑢𝑝|𝐺(𝑖𝜔)| = ‖𝐺‖∞
𝜔
Conversely, the gain of the system corresponds to the peak value of the frequency
response. This relates to the idea that an input produces largest output when the class
of system is at resonant frequency. ||G||∞ is called the infinity norm of the transfer
function G(s).
This idea can be generalized to the multi-input, multi-output case. For a linear
multivariable system with real rational transfer function matrix G(s), the gain is
defined as
Eq. 7 𝛾 = ‖𝐺‖∞ = sup 𝜎̅(𝐺(𝑖𝜔))
𝜔
If systems H1 and H2 are linear it follows from the Nyquist stability theorem that the
closed loop is stable, because if 𝛾1 𝛾2 < 1 the Nyquist curve is always inside the unit
circle. Therefore, the small gain theorem can be said to be an extension of the Nyquist
stability theorem.
Aside from the fact that the small gain theorem can be utilized also for nonlinear
input/output systems, its main limitation is that it does not consider the phasing of
signals around the loop, so it can be very conservative. To define the phase scalar
product is needed thus square integrable functions can be defined as
∞
(𝑢, 𝑦) = ∫0 𝑢(𝜏)𝑦(𝜏)𝑑𝜏
The phase ø between two signals can then now be defined as
(𝑥, 𝑦) = ‖𝑢‖‖𝑦‖ cos ∅
Passive systems are those systems where the phase between inputs and outputs is
900 or less for all inputs. It follows then from the Nyquist stability theorem that the
closed loop linear system is stable if the phase of the loop transfer function is between
Course Module
Feedback and Control Systems
14
Frequency Response Transfer Function
–π and π. The result is also true for nonlinear systems and this theorem is called the
passivity theorem which is closely related to the small gain theorem.
Describing Functions
A generalization of Nyquist’s stability criterion can be obtained based on describing
function for special nonlinear system as illustrated in Figure 10 which consists of a
feedback connection of a linear system and a static nonlinearity.
Source: Feedback Systems: An Introduction for Scientists and Engineers(294), by K.J. Anstrom and R.M.
Murray, 2014, Sta. Barbara, California, https://www.cds.caltech.edu/~murray/amwiki. Retrieved
December, 2018.
By following the approach of the Nyquist stability condition the conditions for
maintaining an oscillation in the system can be analyzed. If the linear subsystem has
low pass character, its output is approximately sinusoidal even if its input is highly
irregular. The condition for oscillation can then be found by exploring the
propagation of a sinusoid that corresponds to the first harmonic.
It is first analyzed how a sinusoidal signal propagates through a static nonlinear
system particularly how the first harmonic of the output of the nonlinearity is related
to its input. Let F represents the nonlinear function, F(e-iωt) is expanded in terms of
harmonics as
𝐹(𝑎𝑒 −𝜔𝑡 ) = ∑∞
𝑛=0 𝑀𝑛 (𝑎)𝑒
𝑖𝑛𝜔𝑡−∅𝑛 (𝑎)
Where Mn(a) and øn(a) represent the gain and phase of the nth harmonic which
depend on the input amplitude since the function F is nonlinear. Describing function
is defined to be the complex gain of the first harmonic.
Eq. 8 𝐿(𝑖𝜔)𝑁(𝑎) = −1
This equation means that if a sinusoid is injected at A in Figure 10, the same signal
will appear at B and an oscillation can be maintained by connecting the points.
Equation 8 gives two conditions to find the frequency ω of the oscillation and its
amplitude a: the plane must be 1800 and the magnitude must be unity. To solve the
equation it is advised to plot L(iω) and -1/N(a) on the same diagram as shown in
Figure 10b. The graph is the same as the Nyquist plot where the critical point -1 is
replaced by the curve -1/N(a) and a ranges from 0 to ∞.
Course Module
Feedback and Control Systems
15
Frequency Response Transfer Function
Course Module