Controls Systems and Mechantronics

back to home

Contents
1

2nd Order System behavior

Chapter 5 Transfer function, block diagram, PID controls

Chapter 6 s-Plane and Root Locus

Chapter 7 Root Locus Based Analog Design

Chapter 8 Use of Frequency Response in Analog Controller Design

11

Chapter 9 Digital Control

14

Controls Lab

16

7.1

Lab 9 Op-amp Circuits and servo-system Open Loop Transfer functions . . . . . . . . . . . . . . . . . . . . . . . . . 17

7.2

Lab 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

Good to know

18

2nd Order System behavior

Remark: Behaviour of 2nd order System

A canonical 2nd order system has the transfer function of: G(s) =

n2
s +2 n s +n2
2

Under Step input:

Overdamped: meaningless, eigenvalue of system real, no overshoot(oscillation), behaves like 1st order. Overshoot is
pseudo, merely a superposition of 1st order.
Critically damped: = 1 threshold.
Underdamped: < 1 eigenvalue complex, overshoot and oscillation(however slight) ALWAYS occur.

Under Sinudoidal input:

For this case, we only concern about Steady State: Y = A|G( j )| sin(t +). Since if eigen complex, |G( j )| = n2 / (n2 2 )2 + (2
we have:
Heavily Damped p1 < < 1, |G( j )| decrease monotonously for ANY input .
2

Critical point = p1 , |G( j )| = n2 / n2 + 2 , threshold for resonance, arguably the fastest to steady state
2
p
Moderatly Damped < p1 , resonace! |G( j )| peaks at p = n 1 2 Not n ! If = 1, then p = n !
2

Remark: the 0.707 damping

y =1 p

e n t sin( 1 2 n t + )

12

i
n e n t h
y0 = p
sin( 1 2 n t + ) 1 2 cos( 1 2 n t + )
12
a perfect chance to apply trig:

i
n e n t h
y0 = p
sin( 1 2 n t + )
12

i
n e n t h
y0 = p
sin( 1 2 n t )
12
thus we have critical pts at

1 2 n t = k where k = 1, 2, 3...

p
we want the solution to be fully enveloped by the error band of a fixed percentage. The error at first peak t1 = / 1 2 n
is the key:
y1 = 1 p
=1+ p

1
1
1

e n t1 sin( + )
e n t1 sin()

use the definition of :

=1+ p

1
1

e n t1

12

thus error from steady solution of 1 is:

= e n t1 = e
=e

1 2

n p

1 2 n

Chapter 5 Transfer function, block diagram, PID controls

Defn: Back EMF

The voltage induced on the motor in the linear graph of power interface for the motor transudcer; opposes the current
in the motor inner circui; distinct from voltage due to winding and coil resistor and inductance
Remark: stall the motor smell burning
Assume source voltage constant, power conserved over transducer P = M TM , then by transducer eqn iM =
dropping M to zero will blow up iM , resistor heats up, burns the resistive coating, and shorts the circuit.
Convention: Positive direction of motor velocity
Clockwise when viewing from the "shaft end".
2

Tm

P
M

Defn: Block Diagram

Arrows and blocks: Signals and Operator: operator can be summation junction or multiplicative block. All blocks
must be linear operator! Basic assumption of block diagram: responses of signal to downstream block does not affect
the repsonse up stream, i.e, ideal blocks require no power. This idealization collapse when the downstream block
draws a large amount of energy from upstream, a condition termed loading of the upstream block. Buffer amplifier
(operational amplifer in the case of controller circuits) with high input impedence" and low output impedence" is
needed when loading occurs.
Rule: Manipulating Cascades/ Blocks
If loading, or non-ideal power draw, can be neglected, we will be able to apply all the simplifying block diagram algebras,
namely: blocks in cascade(series) can be interchanged, combined , and distributed (to several branches). Thus, blocks can
be moved through branches, and conversely, branching points can be moved past blocks, suitable with compensations.
Defn: Equivalence of block diagram
Two block circuits are equivalent iff identical input yields identical output
Note: Operational Calculus
Integrating and differentiating in a pipeline fashion is invented by Heaviside
Remark: Ambiguity in s-domain Operational Calculus
1
can be either step signal or integration operation; a polynomial ratio can be either a signal or a transfer function, yet
s
the context makes it clear.
Property: Summation Junction
Inflows to a summation junction can be "inverted";
Junctions can be interchanged IF theres no blocks between them;
Inflow must have same units!
Rule: Branching of signal on Block Diagram
Put one dot per branching even if the branched signals are identical.
Property: I/V Signals
Voltage signal from sensors (4-20mV, 0mV for fault) are amplified to current signals for transport, then converted back
with 250V precision resistor.
Defn: Feed Forward/Backward Loop
If a branching feeds the branched signal in the direction of the information flow, then its Feedward Loop; if in the
opposite of the information flow, its Feed Backward Loop. Both Loops can be collapsed into one equivalent block.
Remark: Postive Feedback
Leads to instability: the screeching of loudspeakers when output(loudspeaker) was brought too closed to the amplifier(microphone) and amplified again
Remark: Feed Backward Loop
G
For a simple GF B (s) = 1+H
feedback loop, input R is the Reference signal or some other physical quantity that is
G
mapped to desired output", output C is the Controlled signal, the feedback branch is the actual output, the summation
junction compiles E the Error signal and G(s) operates on the error, increasing output if positive, decreasing output if
negative, and maintaining it if zero.
Remark: Summation junction interchange
possible if theres no block or branch between the two junctions.
Defn: Gain vs. Attenuation
Both are modification of input signals. Gain is just a multiplicative constant", not necessarily larger than one, Attenuation however, implies the magnitude of output is less than input.
Defn: Error

H
(a) Before

G
1+GH

(b) After

Figure 1: Negative Feedback Formula

Error is the error signal E(s ) = H C (s ) R(s) rather than nominal errorc(t ) r (t )
Defn: Noise and disturbance
Noise is extraneous electronic signals;Disturbance is extraneous energy variable. Influence of these additional input
signal can calculated using Superposition, i.e zero out other inputs and reduce the block algebra just for that input.
Beware of inverting signals and check if the feedback formula is truely applicable.
Defn: Unity Feedback
A feedback block of 1 impies the signal used for feedback has the same unit with that of the reference. We do this,
however, in order to implement Matlab step() function:
R

C
R

1
H

H
(a) Before

(b) After

Figure 2: Converting to Unity feedback

Rule: Reduction Rules

1. Start from the innermost weird looking loop or branch or summation junction;
2. Use , , , to represent signal when checking the effect of moving branches or nodes;
3. Adding/subtracting signal can also be used to compensate the effect of moving, in addition to multiplying with
another block;
4. Remember to switch sum junction only when theres no block or branch in between;
5. Separate the functionality of branch as much as possible;
Defn/Remark: PID Control
K

Integral( sI ) and Derivative(KD s) control are almost always used with Proportional(KP ) control, in a positive feedforward
loop. Note gains could be negative in sense! But in PID, we restrict it to postive realvalue
Defn: Open/Closed Loop control
Open if the control system has no feedback loops; closed otherwise.
Rule: Inverse Laplace tranformation
1.Reduce the block diagram to a single transfer function;
2.Find intial condition(final value thm) of system for the use of superposition;
3.Solve for eigenvalue, ,n (the denominator);
4.Split numerator/ factor denominator;
5. Perform partial fraction(use if numerator order lower than denominator, watch for repeated root);
6. Inverse Laplace by table.

Thm: Final Value Theorem

lim{s F (s)} = lim { f (t )} if such limit exist.
s 0

Pf: Since
L { f 0 } = s F (s ) f (0)
we have:
Z
lim{s F (s)} = lim{L { f 0 } + f (0)} = lim{
e s t f 0 (t ) d t + f (0)}
s 0
s 0
s 0 0
Z


=
1 f 0 (t ) d t + f (0) = f (t ) + f (0)
0

= lim { f (t )}
t

Remark: warnings about Final Value Thm

1. f (t ) must have a steady limit value to begin with; i.e for oscillatory system, input is not unsteady and system is
damped
2. Dont forget to multiply s to the output Laplace function;
3. lim {c(t )} = lim{s R(s)G(s)}.
t

s0

Remark: State Space requirement

State space representation of a high order ODE cannot have inputs/outputs higher than 1 on the right side!! Thus,
when input has high order, use the method below.
Remark: Time domain block diagram: scalar and state space
Given b0 u 00 + b1 u 0 + b2 u 00 = a0 y 00 + a1 y 0 + a2 y, wanna express y as the output of input u as state space canonical form.
Rearrange the highest order term of output into y 00 = ..., integrate to eliminate all derivative, express the result in nested
integrals; the result block diagram will have u at top row, integrator/sum junction in the middle row, and branches of
y at bottom row. Note that the output of every integrator is a new state space variable. Finally, express the derivative
vector of new state space variable, introduction the expression for y and simplify.
If input is only u and output is a vector of desired variables: write state equations, express the desired output(real physical energetic variable) as multiple interlocking rows of output and a input row. Here well have summation junction
representing the derivative and a integrator to get back to the variable
Remark: Scalar ODE to system of ODE(state space)
# of variable in the system=Order of ODE given
Defn: non-causal
If a scalar ODE input functions order is higher than the order of ODE, then output is trying to predict the input, a
condition term non-causual.

Chapter 6 s-Plane and Root Locus

Remark: Synonyms and classification for controls

Controls are divided into two categories:
Classical Control=Analog Control=Closed-Loop Control=Feedback Control=Continuous-Time Control
Modern Control=Digital Control
Defn: Bode Plots
A plot of transfer function G( j ) versus input angular frequency , usually of its magnitude and phase angle.[2, 2].

Remark: maximum possible Phase Angle of G( j ) in Bode plot

max = 2 (nd nn ),
where nd is denominator order, nn is numerator angle. Cycle through [, ].
Defn: Root Locus Plot
Root Locus Plot is the plot of all closed loop poles as a function of adjustable parameter. Closed loop means the transfer
function is KG/(1 + KGH ); every closed loop has a companion open loop, or just 1 + KGH .
Remark: open loop pole vs closed loop pole
Theyre different, every point in Root Locus is a closed loop pole, corresponding to a non-negative K input, but only
X , Os are the open loop pole and zeros! They are used to aid us in the construction of root loci.
Defn: Angle/Magnitude criterion
always use K1 = GH not K: both sides need to have same complex angle and magnitude, thus the criteria. Magnitude
criterion is used to find corresponding K for a specific root.
Remark: complex # notation
Z = |Z|Z = |Z|e j Z
Remark: s-domain t-domain link
j

e n t sin(d t + )

= cos
n d

Remark: effect on t domain parameters when moving in s plane

In s domain:
stay on a vertical line: n fixed, same decay envelope in s-plane; on a circle in z-plane.
stay on a ray: fixed, same damping ratio.
stay on a horizontal line: d fixed, fixed damped frequency in s plane; on a ray in z.
stay on a circle: n fixed, same natural frequency.
Steps: Manual Root Loci Sketching
Derive open loop transfer function, in GH = K1 form
plot open loop poles(X) and zeros(O). Rule: PTZ, branch go from Pole to Zero/; one branch per pole/zero;
identify real roots on real axis. Rule: count P/Z to the right, odd go, even no go. Pf: sign flipping and angle criterion
check
 break in/out:
 if (OO), (XX), (O)occurs

d
ds

GH = 1/K to get break pts: branching is always perpendicular: proof: Let s = x +y j , so GH (s) = u(s)+v(s ) j .

On the real line, s = x, GH (s) = u(s ), thus d GH

= u x and v = 0. But by Cauchy Riemann theorem on complex
ds
differentiability, at differentiable points( non pole pts on the complex plan) we have u x = vy . Thus if d GH
= 0, then
ds
vy = 0, meaning the imaginary component v is constant(remain 0) in the direction of a small increase in y, i.e. an
increment away from real axis to the imaginary axis. Thus the point will yield purely real function value, which implies
it will be on root loci.
obtain asymptotes:
P
P
intercept a = ( ( p o l e) (z e r o)/(n m)) and
6

angle = (2k + 1)/(n m): draw half-lines! cycles as k > n m

if asymptote intersect imag axis, calculate crossing frequency C and critical gain KC
AoD, AoE: angle of departure/entry
Remark: Multiplicity, break in/out point
Any open loop pole/zero with multiplicity 2 is a potential break in/out point
Remark: effect of OZ(s)
Open loop Zeros bend the branch towards them: thus limits and increase the gain required while changing systems
response pattern.
Code: rlocus, Matlab Addon
use s = t f (0 s 0 ), and then GH = 32/(s + 3) to input transfer functions. r l oc u s(G) always input GH together.
Code: time domain response, Matlab Addon
Matlab command: i m p u l s e( s12 G, t f i na l ) can simulate unit ramp, by cheating Matlab with impulse function. s t e p(G, t f i nal )
will produce unit step.
Defn: Reduction of Order
Eliminating a pole or poles from a transfer function
Remark: Reduction of order
If a > 10 b or more, then (s + a ) a , (s 2 + 2|a |s + a2 + a2 ) a2 + a2 The reasoning is that the contribution
from large time term decays too fast and their residue often too small. Error is largely due to time shift errorshape of
response close with reduced model leading the original. Works better for purely real or a >> b oscillatory transfer
functions.

Chapter 7 Root Locus Based Analog Design

Remark: Purpose of Controller

For a dynamics system, we want to minimize transient error, steady state error, and increase response speed.
Remark: Types well look at
Analog controllers(Feedback Controllers): PID and Lead-Lag Controller.
Remark: PIDs
Input/Output refers to closed-loop transfer functions.
PID is the most widely used 1-input,1-output control algorithm.
power flow is the key regulated variable.
Info Flow: error signal PID G Control signal out to Plant sensor summation and error signal ...
R

GP I D

G p l an t

GP I D

G p l an t

H
(a) PID paradigm: H6= 1 R may not be desired output, merely
reference

(b) Unity feed back: step input yields step output

Figure 3: Converting to Unity feedback

Remark: PID control transfer function

Table 1: PID transfer function

Type
P
P-D

transfer function

n-m order

G(s) = KP

move on branch

+any zero, move on branch

+1

+any zero,+pole at origin, move on branch

+2zeros, +1 O pole, KD adjustment

-1

G(s) = KD (s + K P )
D

P-I

G(s) = KP

s+ KI

s
K

P-I-D

G(s) = KD

s 2+ KP s + K I
D

Remark: Connection to Root Loci

From Root loci, the next step is to figure out, after finding or given desired d , how to figure out the root s, and
more importantly, the K.
Remark: Crux of PID design...
is the placement of controller zeros.
Remark: Angle Defficiency method
Given a target closed loop zero value (s), applying the angle criterion can yield a equation that the open loop zero must
satisfy. For PD control, this will determine the zero.
Remark: Target cases
1. Pick arbitrary s.
2. pick s that minimize the slowest response factor
3. critically damped case
Remark: P-D controls
P-D controls add ZEROS to the open loop root loci, attracting branches that would otherwise sink at infinity. shape of
root loci is modified. Compare P-D to uncompensated or P system.
Remark: P-D procedure
Once a target s is set, angle criteria determines the unique zero that must be added, which fixes the entire branch, from
this point,we can still vary s along the branch by adjusting gain KD , which will be found according by the final s and
magnitude criterion.
Remark: P-I controls
PI adds an arbitrary zero, and a pole at origin, which tend to slow system down by shifting branch toward unstable
region. The advantage of PI is that it overcomes the non-linearities of the system..?
Defn: Pole-Zero Cancellation
Placing a controller s zero on a open system pole to cancel it is call Pole Zero Cancellation.This is highly risky, since
in actuality, parameters are never precise and the real effect is introducing a short branch near the pole, or fixes the pole,
often slowing the system down. One exception is the PZC at origin, where integration (1/s) cancels differentiation (s).
Remark: Unit Step Steady State Error
P/PD controller under unity feedback always has an Unit step Steady State error.
PI/PID controller under unity feedback always has no Unit Step Steady State Error!
1
Reason: for integrator introduce 1s and use final value thm on Error ER = 1+GH
Remark: PID control
Compared to PD, or P, the order n m is reduced, thus asymptotes are less likely to head to unstable zone.
Remark: Root Loci: multiple break-in out on a branch
In the Root loci, one single segment on the real axis bound by open loop pole can have multiple break-out break-in

points! these are due to complex poles sucked into the real axis instead of the asymptotes. One pair of such poles brings
1 break-in+1 break-out.
Remark: PID design steps
1.choose the PID permutations, I eliminates S.S error signal; D is easy.
2.Given desired closed loop zero, place your controller zero by angle deficiency method. Note this sd e s i r e is best to fall
on = 0.707 line. which sd e s i r e d ?
3.To improve accuracy of root loci, iterate with angle criterion.
4.Select a target closed loop point: most negative real, = 0.707, 5.Calculate Gain required to attain that point. Limit
to 100 6.Produce step response
Remark: Limit Controller max gain to 100
K

s 2+ KP s + K I

D
D
when attempting to limit the max controller gain, go to G(s) = KD
and plug in the zeros placed in your proto
s
design and get the ratio between KP , KI and KD to set the target bound of 100. Why 100? too high a gain degrades the
signal to noise ratio, and increase power required.

Warning: +correlation for Controller application

When deriving G p l an t from linear graph, transformer and transducer signs must be such that a positive reference signal
yields a positive controlled signal, or the system will be unstable
Remark: Lead Lag Controller
This type of control, in contrast to PID, has no time domain interpretations.I: time domain integral of error,D: time
domain change rate of error. Instead, LL controller is based on phase shift or lead lag in steady frequency response.
Defn: Lead Lag controller general form
LL control has the following general form:
GLL (s) K

s+z
s+p

recall that a frequency response of Asin(t ) to a LL controller:

c(t ) = AK|
=

j + z

j + p
j + z

| sin(t + )

j + p

if > 0 Lead control; if < 0 Lag control.

Remark: in LL controller
LL as a function of varying can be thought of as a pair of vectors(one from controller zero and one from pole)
sweeping the imaginery axis. Thus the one thats closer to origin has the bigger angle. Zero closer means lead; pole
closer means lag. Also, we have:
0 l ead 90 ; 90 l a g 0
Remark: Cause and effect
Lead and lag of controlled signal doesnt violate cause and effect since this is steady state response.
Remark: Crux of LL controller design
place the Controller zero and pole to obtain desired closed loop pole.
1. angle deficiency + rule of bisection gives zero and pole(note, keep the controller net angle between 0 and 90 for lead,
-90 to 0 for lag.). Graphical or trig can be used.
2.calculate gain
9

Remark: LL vs PID
1. LL is generally faster than Integral control, since no poles are placed near origin.
2. LL has a steady state error that is a function of Gain. Thus a compensation of input signal is required! In fact, it would
take infinite gain to reach zero error signal.
Remark: Potentiometer
Pot" is essentially Uninsulated high resistance wire winding around a circular or helical frame and a rotating Wiper.
Pot can be used as Variable resistor, or a voltage divider, which is a analog rotational position encoder. Thus angular
position and voltage of wiper has a one-to-one correspondence.
Remark: Digital position encoder
Two types of Boolean" or 0/1 position encoders: incremental and absolute. Incremental use 2 or 3 Boolean variables:
one or two for flagging an increment and the index" for flagging the completion of a revolution. If we have two sensor
for increment, separated by a quadrature, then the encoder has 4 times the resolution(line number) than without.
Absolute type has fixed zero and measures the absolute position relative to that zero, requires more binary digit as
resolution of revolution becomes finer.
Remark: Quadrature incremental encoder
We have a fixed resolution of lines and clear area, say 500 lines and clears. By using 2 sensors placed a quadrature apart
we increased the resolution from one line-clear cycle to quadrature of a line-clear cycle or 2000 increments. To convert
digital to analog voltage, we use Encoder Ditial to Analog Converter (EDAC), that spans 0 to 4000 Boolean count with
9.962 1r e v
0 to 0.962 Volts. Thus the volt per rad position decoder constant is Ke nc od e r = 2r
e v 2
Operation: Op-Amp
Op-Amp is a differential amplifier that can be operated in two modes: without feedback(comparator) or with feedback
signals( various amplifier).
Naming:
V + , i + non-inverting terminal.
V , i inverting terminal.
Vou t output
V s + positive power supply, rail voltages
V s negative power supply
Comparator Mode:
no output to input feedback, any difference from the two terminals will produce a fixed magnitude voltage difference,
whose sign switches as terminal inputs are switched. Amplifier Mode:
If output-input feedback exist, we follow these rules for non-saturated Ideal amplifier:
V =V+
i = i+ = 0
Gai n =
Remark: Saturation
Opt-Amp reaches Saturation if output reaches 90% rail voltage, at which point output is constant and additional input
yields no additional output.
Remark: Op-Amp DIP
8 pin Dual In Line Package means two rows of 4 spaced 0.1 inch apart (through hole), as opposed to "surface mount".
Note: Dot marks pin 1.
Remark: Having multiple op Ampt in one IC
Dual Op-Amp 8 Pin DIP has two op Amp in one Integrated circuit.
Remark: Single Op-Amp Circuits, Amplifier mode
Remark: Complex Impedance
Laplace transform of Elemental equation of circuit elements yields its complex impedance.
In series and in parallel of impedance follows identical rules with resistors.
10

 For single input voltage, when deriving the transfer function of a op Amp network, use lumped inverting input
V
impedance Z with lumped feedback impedance Z f b and a ground on non-inverting input, we have: Vo u t = G(s) =
in

Zfb .
For multiple inputs, one cannot derive a single impedance, like the above case, use superposition if linear! Eg: inverting
summer before saturation.
Remark: flaws in Ideal integrator, Ideal differentiator
1
but is prone to self saturating, thus a non saturating integrator is used in stead near its linear
Ideal Integrator has Rs
region; the cost is the introduction of a zero at negative real axis. Ideal differentiator has C s/R but is unstable due
to internal Amp oscillations, thus a stable non-ideal differentiator is used: the cost associated is the introduction of a
zero at negative real axis. However, we do have control on where is zero or pole is placed by changing resistance and
capacitance.
Remark: Non-ideal is pretty good though
The result is better if time constant of the non-ideal integrator far exceeds time constant of the plant see figure bellow:
Figure 4: how non-ideal integrator approach ideal as i nt e g rat o r increase

Chapter 8 Use of Frequency Response in Analog Controller Design

Remark: Noise and disturbances

These come from magnetic field induction-by AC power source, motor magnetic field-on conductors in sensor, control
circuitry, cables etc.
Remark: Frequency response review

11

Steady state: when systems natural/unforced/homogeneous response vanished and systems excited(driven)/forced/particular
response dominates. For sinusoidal steady state, theres no change between cycles in time, not no change in time.
Remark: Decibels
2 schemes of definition exist!
Scheme A: {x} d B s = 10 log(x)
Scheme B: {x} d Be = 20 log(x), we use scheme B in controls and ECE.
eg: 1 0 dB, 105 100dB, 0.707 -3.01 dB
if plotting log graph, y minus -1 corresponds to -20 dB
Remark: Interpolating on log scale
10
2 0.3
3.16 0.5
4 0.6
5 0.7
8 0.9
10 1
accurate to 2 decimals.
Interpolating on log scale with any base between point a < p < b :, use formula:
  L pa
L
x{ p} = a ba b a
Remark: Phase angle Bode plot
the following rules apply for phase angle Bode plot:
use rad/s for degrees for phase shift.
at = 0, = 0 for every factor.
view the net phase shift as the sum(and difference) of all phase angle contributor from numerator and denominator
factors.
identical factors on numerator and denominator are symmetric around = 0 axis.
asymptotic phase angle rule is 180 for 2nd order factor on numerator, 180 on denominator, 90 for 1st order
factor on numerator.
negative phase shift is lag, positive means lead.
pass through half the change to asymptotic phase angle at = corner frequency(1/, or n ). But magnitude response is not max or min at corner frequency, but at d !
differentiator adds a lead of 90 ; integrator adds a lag of 90 . they are straight lines on phase plot.
For 2nd order factor,fix natural frequency, and reduce damping ratio pinches the phase response steeper toward c
at which point must be half the asymptotic frequency. Pf: use circle for constant natural frequency and use low
rays cluster near imaginary axis, = 1 is real axis. Show = n implies the angle sums up to 90 and < n high
damping ratio has bigger phase angle in magnitude and > n lower damping ratio has bigger phase angle.
for 2nd order no damp, we have the limiting case of sharp cuspy corner. or a step transition from 0 to 180
Procedure for phase bode plot for transfer function: put in pole zero form, identify the numerator factors, denominator factors and the order and corner frequency for each factor. Then plot each and superimpose.
Remark: Asymptotic Approximation of Log-Mag Bode Plot
Using this technique, the log magnitude of Bode is accurate except near c , which is related to n and most often near
d
for 2nd order system, asymptotic approx is more accurate at corner freq if damping is larger.
2
2 s
asymptotic form: convert all factors into the following 1st or 2nd order form: s 0 + 1 and s 2 + 2n + 1
n

the corner frequency c is n or the pseudo natural frequency for first order: n0 = 1/.
what is the corner frequency and what its not? its the point of intersection of low freq approx and high freq approx;
its the at which the phase is half the asymptotic phase shift in phase Bode. Its NOT the frequency at which resonance
or max mag occurs!! = d is! However, when 0, the two coincides.
Bandwidth of a frequency response of a transfer function is the frequency range where log mag is above -3dB or an
attenuation of 0.707 and above.
12

 gain factor is the gain extracted from all factors after converted to asymptotic form. This will be a constant offset is
the Asymptotic approx.
Integrator and differentiator are 1 slope straight lines crossing (1, 0)
Remark: 1st order pseudo frequency and 2nd order term natural frequency
A
When using frequency response, for 1st order term s+a
, the a is 1 and can be treated as pseudo frequency n of the 1st
B
order term. This frequency is the corner frequency of the log magnitude bode plot.For 2nd order term s 2 + s
has
+ 2
corner frequency at the natural frequency of this term, namely n

Remark: Procedures for Asymp approx

convert to asymptotic form with gain factor and corner frequencies of all factors. Note overdamped must be factored
to two factors.
get gain factor offset
determine frequency range, +2, -1 decade of crossing frequency max, min.
on paper 1, draw individual factor contributions, (order slope, rising or falling)
calculate Log mag at lower and upper frequency limits.
slope is piecewise varying at corner frequencies.
one paper 2, piece everything up.
Remark: Correction at corner frequency
for 2nd order underdamped factor at corner frequency, true log mag differs from low freq asymp approx by log(2 )
Note if damping is zero, d = n we have blow up.
p
for 1st order log mag at corner frequency, the true value deviates from low freq approx by log( 2) = 0.151.
Remark: identify transfer function from Log mag response
initial slope at (1,0) indicates integrator or differentiator.
every change of slope is a corner frequency.
locate gain factor preferably at = 1 where you have precisely log K.
2nd order factor can be underdamped or overdamped, the two cases must be identified(by damping ratio fit) and
treated differently
underdamped, get corner freq and compare with normalized family for .
p
overdamped, utilizing the fact that c = c1 c2 , we introduce separation factor s.t.
c1 =

, c2 = c , and then obtained the appropriate from normalized family

Remark: Non-minimum phase Systems

Is when a zero is misplaced in the unstable region resulting in the phase response shifting upward by 180 . Detectable
only in phase diagram.
Defn: marginally stable
A closed loop transfer function with poles on imag axis is called marginally stable.
Remark: Stability of closed loop TF and open loop freq resp
Stability of closed loop TF depends on both root loci and crossing frequency into unstable region. A quick thought
would find that crossing frequency and frequency response has a convenient connection: at the intersection of root
loci and imag axis, the root loci s has purely imaginary form, j . In the frequency response of the open loop transfer
function, we evaluate that function with s = j . These are the same! The frequency response can be viewed as a
restriction of s to only imaginary number. For a particular K, we can of course input any s = j to Go.l . but only the
that makes KGH = 1 hold for that gain K will it be on the root loci and be on imag axis producing instability.
Using mag/ angle criteria, this means on the Nyquist plot of KGH , the curve crosses 180 and pass 1. On Bode
of KGH , this means when = 180 , log mag is 0. However, for K that are already in the unstable region, when we
attempt to satisfy the angle criterion at 180 , the log magnitude of KGH |180 will be bigger than 0 (due to gain even
bigger than the critical gain), or on Nyquist, the curve goes to the left of 1.
Remark: Gain margin/ Phase margin
Gain margin tell if the closed loop system will go unstable at particular gain with open loop freq resp at that gain.

13

 For frequency response on open loop transfer function Go.l = KGH , if at the frequency input(180 ) where phase
Bode passes 180 , the log mag bode point exceeds 0, then crossing frequency exists and its value is the input frequency.
The critical gain is the gain is the K that make the log mag bode passes exactly 0.
The phase bode plot for any K is the same. But to check if a particular gain is stable or not, we need the log mag Bode
with K equal to the gain of interest. If log(|Go.l (180 )|) 0 we are in stable region, and =" implies marginally stable.
The Log Gain Margin is the log margin from reaching marginally stable at crossing frequency.
Phase margin is defined similarly but reverses the order of utilizing phase and log mag bode. At log mag equal to 0,
the phase shift relative to 180 is the phase margin.
Remark: The following are equivalent
closed loop response of step input blows up
root loci head to marginally stable or unstable region.
bode gain margin is 0 or negative
bode phase margin is 0 or negative
Nyquist plot encloses -1 on real axis
Remark: Transport lag stability
for transfer function of the form KG(s)e s T where T is the lag, we can make the following observation:
bode gain margin and phase margin criteria still works. i.e. the gain is too large if at 180 the log mag exceeds 0, or
K Kc
the prominent trait of transport lag is phase heading to
thus if G(s) s log magnitude slope upward eventually, the lag sys always blows up regardless of gain K or lag T .
effect of reducing lag period:phase plot stretches to the right and 180 increases, system is safer in this case if
log |KG(s )| exceeds 0 at some point; else if none exceeds 0, were always safe whatever the lag time T is.
Reducing lag time(system respond more swiftly) will play a role (stablize system) if G(s) and K is such that the log
magnitude exceeds 0 at some point(however small a ) in the bode mag plot.
K > Kc will always make system unstable under a lag.
e sT in Laplace domain is a time shift (i.e lag) is the time domain: L { f (t T )} = e s T F (s) and {u s (t T ) f (t T )} =
F (s)
e sT s .
Remark: Nyquist
Read magnitude increase and phase angle from Nyquist plot. Based on these info, transfer functions the order(phase
angle range), resonance(increase in mag at some j ), stability(enclosed -1 or not) can be determined.

Chapter 9 Digital Control

Remark: Digital control is all about lag and quantization

sz
continuous time sampling quantized time sampling at T
stability region of half plane stability region of circle of radius 1
Laplace transform Z transform.
integration summation.
differentiation difference.
Remark: Digital Analog
error analog signal A-D converter(Sampler) Digital Controller
D-A converter(ZOH) plant
Remark: 3 types of signals

continuous time signal : cont voltage , cont time sampling

discrete time signal :
cont voltage , quantized time sampling

digital signal :
quantized voltage , quantized time sampling

14

(1)

Remark: whys lag?

Since sampling is discrete, the controller algorithm uses data from old sampling and also requires a finite amount of time
to compute, thus total sample lag time must account for both computation time and input, output time lag.
Defn: z unit delay, transport lag of 1 lag period T
z 1 e sT where T is the lag period
z 0 = e s 0 no lag
z 1 = e sT a shift of 1 lag period
z 2 = e s (2T ) a shift of 2 lag period
Remark: T is also the sampling period of controller, a design variable
Remark: time is a sequence in Z
t = kT , k = 0, 1, 2, ...
Remark: indistinguishable sum of pulse and unit stepdigit weirdness
In digital time sampling, a infinite sequence of time lagged impulse is indistinguishable from a unit step.
using this fact z, or digital version, of a step function is a series of time lagging impulses:
L 1

nX

o X
X
X
(t nT ) = u s (nT ), n = 0, 1, 2...
L 1 {1 e s(nT ) } =
L 1 {z n } =
z n =

(2)

n=0

n=0

n=0

n=0

which is a series of lagging impulses(not Dirac delta impulse) starting at t = 0 and is indistinguishable with a step in
digital space.

P
z transform of step is
z n
n=0

Defn: Z transform

P
Z { f (kT )} =
f (kT )z k this is simply the discrete version of Laplace transform.
k=0
R
L { f (t )} = 0 f (t )e s t d t
Remark: Z transform pairs
here the is not the Dirac delta function.
Z {(t )}
Z {u(t )}
Z {u(t L)}

all z
z k =

k=0

X
k=k

z k =

1
1 z k
1

1 z k

z >1
z > 1, k > L, mi n

Remark: s, z conditions necessary for transform?

NOt really, the assumption of s and z in order to perform transform is that for some s and z, the transform is valid.
Remark: identity
since 1s = 1z1 1 , we have:
s = 1 z 1 This can be used to represent integration and differentiation using Z transform. However, the result using
backward rule and trapezoid will be different from using Z transform. To derive trapezoid, use series cancellation.
1 z 1 represents the difference between a variable and its previous value
Remark: Derive Z transform pairs
Use power series, munipulation, and existing pairs.
15

Remark: Final value theorem

Recall:
lim f (t ) = lim{s F (s)}

(3)

lim f (kT ) = lim{(1 z 1 )F (z)}

(4)

s0

z1

(5)

Remark: zero-order hold

s T
1
ZOH u s (t ) u s (t T ), thus we have L (ZO H ) = 1es = 1zs , Z(ZO H ) = 1, doesnt change digital signal, digitize
analog signal
ZOH is a digital to analog signal converter
digital controller G= Sampler + controller + ZOH. if we input a discrete sampling signal to ZOH, we are essentially
feeding it a scaling! a scaling for impluse of width T. Thus we thus a bunch of discrete dot-sample signal to impulse
sample signal. ZOH change nil if input to sampler is already discrete.
L { f (t T )} means use previous value: time shift the sequence down one.
Remark: reduction to difference equation
Digital controller, in discrete math, is done in difference equation, meaning the output is a weighted average of previous
time ouput and inputs.
step1: write out digital controller, ZOH (sampler has no explicit block).
step2: move s of ZOH to plant transfer function.
step3: Z -transform plant transfer function and express ck as an average of cki , rki .
pad the difference table with as many 0s as the highest negative order of z
Remark: T effect and gain effect
Increasing T, or reducing sample frequency destablize a system. For some reason, increasing gain has the same effect of
destablizing a system.
Remark: Pole zero cancellation
valid for (1 z 1 ), but in general, no since in practice they are never precisely the same.
Remark: z 1 and z form of transfer function
z 1 is used for deriving difference equation,representing a delay of one sampling period; z for pole zero determination
and stability analysis
Remark: various ways of approx integration or differentiation
F (s )
The result in z of s can have different forms!

Controls Lab

Defn: Main Types of Motor

Variable(Switched Reluctance) motor: non-magnetized, soft iron rotors
Permanent Magnet (can,can stack,claw pole) motor: magnetized but not castellated
Hybrid magnetized and castellated
Description: Permanent Magnet
Motor= Stator +rotor;
rotor=permanent magnet shell(with/w.o. Transition region)+ Non-ferromagnetic core/yoke. The shell has N,S pole
fanning out along the circumference.
Stator=static casing of a stack of 2 cans, each made up of a pair of Bell and Plate,each has a row of claw poles. The two
rows will have opposite polarity.The can assembly contains wound coils that magnetize the claw poles.
16

remark: PM motor step size and stator pole

motor step size=1/4 magnet domain on shell=width of a stator pole
Description: Hybrid Motor
rotor=PM cores encased in two halves of laminated ferromagnetic rotor caps.
Big Stator poles are castellated at the tip:castellation describes the alternating saw tooth shaped land or high portions,
and grooves, or low portion. teeth refers to the castellation teeth not the stator pole. Note the rotors face that points
toward the stator teeth are also castellated. Pitch of the teeth controls the real step of motor:this is the beauty of Hybrid!
the castellation teeth allow fine control of rotor angular position.
Remark: Hybrid step size
Recall that rotor are castellated as well and secret that lies behind the fine step size of Hybrid Motors comes from the
1/2 pitch offset in castellation between the 2 rotor caps! This constrains that only one set of cap castellation on the
rotor can perfectly align with stator castellation. As the poles(group A and group B) are alternatingly energized, the
unaligned cap will be align as the previously energized cap falls off of alignment. Thus the rotor moves by a angular
distance of 1/2 pitch. If we allow A B to be energized at the same time, the step size is reduced in half again. Thus Step
size =1/4 of castellation pitch.
Remark: Diodes
Made with P-N junction semiconductors. At the time of manfacturing, the junction between P and N will neutralize
and form a depletion layer that act as insulator. act as conductor in forward voltage before. The reduction of depletion
layer by applying forward biased E field to counter the built in biased field is the key to increasing the conductivity
of the transistor: once the depletion layer gets thin enough, the flow of electron to holes is smooth: forward direction
positive on P side, negative on N side. In the reverse bias, higher voltage widens the depletion layer further and further
insulates. However, when the layer grows to certain level, the strength of the layer field causes the charges at junction
to breakdown, like sparks in the air.
Remark: BJTs
Bipolar(two charge carriers) Junction Transistors: NPN, Collector/ emitter (of carriers)and Base. It can be used to
amplify a current(Collector to Emitter current) with a base current. Mechanism: the emitter P-N junction is forward
baised, conductivity is easily triggered( the thin depletion layer reduce in size). as the electron drift towards the base,
it can traverse a typical distance called diffsuion distance(1m) before the settling recombination with holes. THus a
thinner P layer will cause the electron to drift across the base layer. As it does that, the biased field(draw it out to see!) in
the Collector N-P junction sweeps it into the Collector N side, triggering a amplification in ic current. This diffusion
process is aided by the higher Doping concentration on the Emitter N side(thus not merely two side by side N-P diode).
Usually 100i b = ic under certain modes. Cutoff Mode transitior is a diode ic = 0, Saturation, not effective, i b too large
for Vc e , Forward Active Mode good amplification varying with i b . A useful equation in the Active Forward Mode:
VE = VB VB E(t h r e s)
Remark: FET
Field Effect Transistors: Enhancement(like Normally open switch) and Depletion mode(normally closed),based on
widening or narrowing the conductive channel based on semiconductor field effects(external E field pushes or attract
P/N)

7.1

Lab 9 Op-amp Circuits and servo-system Open Loop Transfer functions

Op amp:
1. Operational amplifier amplifies the differential voltage input. Two ways of using it include (1) open loop and
(2) closed loop. The open loop(no feedback) gain is usually extremely large: 10,000-100,000. Closed loop use,
however, keeps the gain to much more practical levels.
2.
Ideal Opt-Amp are feedback circuit controllers such that the following rules holds:
- if voltage information is fed back, then V + = V
17

- else, if inputs are not identically the same then opt amp gives discrete digital outputs that are at saturated voltage(90% of
supply voltage).
- infinite input complex impedance to current
- infinite gain
- from the above rules
Remark: Complex impedence
can be used to conveniently obtain the transfer function of opt-amps

7.2

Lab 11

Remark: Overshoot doesnt mean oscillatory

superposition of 1st order tfs can produce fake overshoot. A true overdamped system must have a root loci branch that
leaves real axis and then returns

Good to know
COMPLEX IMPEDENCE In the steady state signal of a circuit with complex compnents under sinusoidal voltage
forcing. The steady state current signal is also a sinusoid scaled by a (complex) resistance and a phase shift. Such two
changes can be conveniently captured by the so-called complex impedence Z of the equivalent circuit. For example, a
resistor has Z = R, so it provides the resistance scaling, but no phase shift on the current. A capacitor Z = j /C ,
on the other hand, provides no scaling if used alone, but a phase shift whose influence becomes greater at lower
frequencies. The inductor, Z = j L provides phase shift at higher voltage frequenies.
The current thus has, for Z = Z0 e j and V = V0 e j t :
I=

V0
Z0

e j (t )

Note that the complex resistance Z0 is a combination of resistance, capacitance and inductance since it is the modulus
of the complex impedence Z rather than its real part which is R.
Remark: Series Parallel
The rule for computing equivalent complex impedence for elements in series and paralle follows the resistor rule.
Remark: Complex impedence
can be used to conveniently obtain the transfer function of opt-amps

18