The Dark Side of Loop

Control Theory
Christophe Basso

IEEE Senior Member

1 • Chris Basso – APEC 2012
 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
2 • Chris Basso – APEC 2012
 What is the Purpose of this Seminar?
 There have been numerous seminars on control loop theory

 Seminars are usually highly theoretical – link to the market?

 Control theory is a vast territory: you don't need to know everything!

 This 3-hour seminar will shed light on some of the less covered topics:

 PID compensators and classical poles/zeros compensation

 Output impedance considerations in a switching regulator

 Understanding delay and modulus margins

 In a 3-hour course, we are just scratching the surface…!

3 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
4 • Chris Basso – APEC 2012
 What is a Closed-Loop System?
 A closed-loop system forces a variable to follow a setpoint
 The setpoint is the input, the controlled variable is the output
 French term is "enslavement": the output is slave to the input
 A car steering wheel is a possible example:


u t  y t 
control


The input u The output y

5 • Chris Basso – APEC 2012

 Representing a Closed-Loop System
 A closed-loop system can be represented by blocks
 The output is monitored and compared to the input


sensor
+  Error vc
processing °
° V - V
The compensator The plant
G H 

 Any deviation between the two gives birth to an error 

 This error is amplified and drives a corrective action

6 • Chris Basso – APEC 2012

 A Servomechanism or a Regulator?
 Airplane elevator control is a servo
servo--mechanism
mechanism:
• The pilot imposes a mechanical position via the yoke


y t 
Variable u t 
setpoint 
Output follows
the input

 Car cruise control is a regulator

regulator:
• The speed is set, the car keeps it constant despite wind, etc.
v

u t  v  constant
Fixed
setpoint Output is constant

7 • Chris Basso – APEC 2012

 Processing the Error Signal
 The error signal is processed through the compensator G
 We want the following operating characteristics:
 Speed
 Precision 1.40
overshoot precision
 Robustness u t 
1.00

Since
600m
1930

200m speed

PID -200m y t 
12.0u 36.0u 60.0u 84.0u 108u

8 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
9 • Chris Basso – APEC 2012
 Where do You Shape the Signal?
 The compensator is the place where you apply corrections

 Error vc
processing

The compensator: G
 The compensator is built with an error amplifier:

vc 
vc 

Op amp TL431 OTA

10 • Chris Basso – APEC 2012

 The PID Compensator
 A PID welcomes a Proportional block

 t  kp vc  t   k p   t 

240m
vc  t  248 mV
kp is high:
120m
100 mV  reaction speed
0
 t   risks of overshoot
kp is low:

k p  2.48  sluggish response

4.29u 12.9u 21.5u 30.1u 38.7u

11 • Chris Basso – APEC 2012

 The PID Compensator
 A PIID includes an Integrating block
t
kp kp
 t  vc  t       d
i  i
0

240m -2.92
vc  t  Integral action:
120m -2.94

 t   no static error
0 -2.96
-100 µV  slow response
  decreases stability
-2.98 S
i  large overshoots
-3.00
-100u . 30k = -3 V  i  107 μs
8.59u 25.8u 42.9u 60.1u 77.3u

12 • Chris Basso – APEC 2012

 The PID Compensator
 A PID
D offers a Derivative block

d d t 
 t  k p d vc  t   k p d
dt dt

240m 2.00 k p d S  1.63 V

Derivative action:
120m 1.00 S
 t   perturbation variation
0 0
is known: anticipation
vc  t 
-1.00
 stabilizes the response
o no static effect
-2.00
 d  163μs
3.66u 11.0u 18.3u 25.6u 32.9u

13 • Chris Basso – APEC 2012

 The PID Compensator
 The Derivative block anticipates the signal evolution and speed
y

 t   d   t 
  t   t   d    t 

 t  S t d
t
t
t t d

d  t 
 t   d    t    d
dt
if  d small
later now

14 • Chris Basso – APEC 2012

 Combining the Blocks
 You can formulate the PID transfer function in different ways:
d t 
 differentiation: vc  t    Vc  s     s  s
dt
 s
 integration: vc  t     t  dt  Vc  s  
 s
 The standard form:
The derivative term cannot
 1  be physically implemented:
G  s   k p 1   s d 
 s i  lim s d  
s 
 The parallel form:
Need a pole
ki s d
G  s   k p   skd s d 
s s d
1
N
15 • Chris Basso – APEC 2012
 Combining the Blocks
 The transfer function becomes a filtered PID:
 
 1 s d 
G  s   k p 1   
 s i 1  s d 
 N 
 If we develop, we obtain a more familiar expression:

    
1  s  d   i   s 2  d i   d i 
N   N  A double zero
G s  An origin pole
i  d 
s 1  s  A high-frequency pole
kp  N 

16 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
17 • Chris Basso – APEC 2012
 Practical Implementation
 Sum up the output of each individual block:

u(t) P kp

+
(t) kp
 I
i   vc(t)
-
d
D k p d
dt

y(t)
 This is the parallel form of the PID
18 • Chris Basso – APEC 2012
 Practical Implementation
 This is the filtered standard form of the PID

u(t)

+
(t) 1
 I
i   P kp vc(t)
-
d 1

D d
dt 1 s d N
y(t)

19 • Chris Basso – APEC 2012

 Bridging a PID to a Type 3
 A type 3 is implemented around an op amp

G s 
1  s  1  s  
z1 z2

s  s  s 
1   1
  p 
 po   p1  2 

Added
pole

 1 1  2 1 
1 s   s 
  z  z    z1 z 
G s   1 2   2 
s  s  s 
1   1  
 po   p1    p2 

20 • Chris Basso – APEC 2012

 Bridging a PID to a Type 3
 Identify the terms and write the equations:

      1 1  2 1 
1  s  d   i   s 2  d i   d i  1 s   s 
  z  z    z1 z 
N   N1   2   
G s  G s  1 2

i  d  s  s 
s 1  s  1  
kp  N   po   p1 

 Four unknowns, four equations:

d 1 1  d i 1
i     d i 
N z 1
z 2 N z z
1 2

i 1 d 1
 
kp  po N p 1

21 • Chris Basso – APEC 2012

 Bridging a PID to a Type 3
 From Type 3 to PID:

d 
 p1   z1  p1   z2  N
p 2 1
1
 p1 z   p z  z z
1 1 2 1 2
 p1
 p z   p z  z z
1 1 1 2 1 2

z  z 1  po  po  po
i  1 2
 kp   
z z 1 2
p 1
z  p z1 1 2

 From PID to Type 3:

 d  4 N 2 d i  N 2 i 2  2 N d i   d 2  N i N
fz  f p1 
1
2 d i 1  N  2 2 d

 d  4 N 2 d i  N 2 i 2  2 N d i   d 2  N i kp
fz  f po 
2
2 d i 1  N  2 2 i

22 • Chris Basso – APEC 2012

 Testing the Conversion
 Assume a type 3 compensator calculated for:
G fc  1 at a crossover of f c  2740 Hz
f z1  200 Hz f z2  600 Hz f p1  21400 Hz f p2  21400 Hz
 The "0-dB crossover" pole is placed at: High-frequency
pole
2 2
 fc   fc 
1  1 
 f p   f p 
 1  2
f po  G fc f z1  43.4 Hz  272 rd s
2 2
 f z1   fc 
1  1 
f
  f z 
 c   2

 d  194u  i  1.05m N  25.6 k p  0.287

23 • Chris Basso – APEC 2012

 What is the "0-dB Crossover" Pole?
 s appears as an isolated term in N(s), it is an origin pole
1
G s  s = 0 is the origin pole

s 1  s  p1 
 If s is affected by a coefficient it is the "0-dB crossover pole"
s
1  s sz1 sz1

sz1 s  1  
 po sz1 s  1 
sz1 s  1  
G s     G0

As 1  s  p1  s

1  s  p1 
 z1 1  s  p1
 
1  s  p1 
 po
G s Dimension
of a gain
Crossover
pole
 po
0 dB log10  f 

24 • Chris Basso – APEC 2012

 Testing with Mathcad®
 We wanted a magnitude of 1 at 2.7 kHz

20

Filtered PID G s 100

10
180

20 log G1( i 2  f ) 10   
arg G 1( i 2  f ) 

0 0
20 log G2( i 2  f ) 10 180
 
arg G 2( i 2  f ) 
G  s  
 10
 100
Type 3

 20
3 4 5
10 100 110 110 110
f
2.7 kHz, 0 dB
Maximum phase boost

25 • Chris Basso – APEC 2012

 Testing with SPICE
parameters parameters

fc=2740 TD = 1.929e-004
i=(1+fc^2/fp1^2)*(1+fc^2/fp2^2) Gfc=0
j=(1+fz1^2/fc^2)*(1+fc^2/fz2^2) Integration N = 2.594e+001
Wpi=sqrt(i/j)*G*fz1*2*pi G=10^(-Gfc/20)
pi=3.14159 C6
0.1u
TI = 1.054e-003
e=(Wp1-Wz1)*(Wp1-Wz2) 4.00V -500uV

f=Wp1*Wz1+Wp1*Wz2-Wz1*Wz2 fz1=600
7 6 KPF = 2.871e-001
fz2=200
Td=e/(f*Wp1) fp1=21.4k X5
R10
N=((Wp1^2)/f)-1 fp2=21.4k IntFilt AMPSIMP
{Ti/0.1u}
Ti=((Wz1+Wz2)/(Wz1*Wz2))-(1/Wp1)
kpf=(Wpi/Wz1)-(Wpi/Wp1)+(Wpi/Wz2) Wz1=2*pi*fz1
Wz2=2*pi*fz2
Wp1=2*pi*fp1 Vstim
DeriveFilt X3
R9
R11x {Td/0.1u} SUM2
0V stim
10k K1 = 1
X6
115uV
11
Filtered 0V 8 9 K2 = -1
X9
POLE derivation AMPSIMP C7 5.00V -
FP = fp2 X8 0.1uF K1
K = -1 AMPSIMP R12 K1 SUM2
{10k/kpf} 4.00V 5
SUM3 K2
0V K 5.00V
K2
+ V3
Vout 12 10 S+A -500uV 16 unknown
K 500uV
K3 5.00V
2 S+A 3 13
17
1.15V -1.15V X4 R8
SUM3 10k Vrefx Vbias
X1 5 4.9995
K1 = 1 -50.0nV
POLE
kp K2 = 1
K3 = 1
FP = (N/Td)/(2*pi) R11
10k
14

 This is the filtered -1 X10

AMPSIMP

form implementation
26 • Chris Basso – APEC 2012
 Testing with SPICE
0 1

dB 
20.0 360

10.0 180
G  s 

0 0
G s

-10.0 -180

2.7 kHz, 0 dB
Maximum phase boost
-20.0 -360

10 100 1k 10k 100k

27 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
28 • Chris Basso – APEC 2012
 Stabilizing a Buck with a PID
 We will use a PID to stabilize a voltage-mode Buck converter
L1 R6
75u 100m vout
10.0V 5.01V
a c
2
5.21V
12
5.21V
Vout2 Out
d Resr
3
PWM switch VM p 70m
Vin Rload
{Vin} X7 5.01V
16 2.5
AC = 0 PWMVM
L = 75u Cout
Fs = 100k 220u

523mV
5
GAIN

XPWM
GAIN
K = 1/Vpeak

1.04V
H s
4

Vstep The Plant

V3
In unknown

29 • Chris Basso – APEC 2012

 Small-Signal Response of the Buck
 The transfer function shows a resonance at 1.2 kHz
30.0

10.0

-10.0

-30.0 Resonance
-50.0 H s at 1.2 kHz 5

10 100 1k 10k 100k

90.0

-90.0

-180

-270 H  s 
10 100 1k 10k 100k

30 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We will explore three different methods for compensation:

1. Shape closed-loop gain to make it a 2nd-order system

2. Place poles and zeros to crossover at 10 kHz
3. Shape the output impedance only

 Method 1 – derive the open-loop gain first

G s H s TOL  s   G  s  H  s 

31 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The loop gain expression is that of the PID and H(s)

d  2   d i  N(s)
1 s  i   s    d i 
 N   N  1  s  z1
TOL  s   H0 2
i  d   s  s
s 1  s 
kp  N     1
 0  Q00 D(s)

 To simplify the expression, we can neutralize D(s) by N(s)

Place a double zero at the double pole position:

2
d  2   d i   s  s
1 s  i   s    d i      1
N   N   0  Q00

32 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The loop gain expression is now well simplified:

TOL  s  

H 0 1  s  z1 
i  d 
s  1 s 
kp  N 

 For a unity feedback control system, the closed-loop gain is:


H 0 1  s  z1 
i  d 
s  1  s
kp  N  1  s  z1
TCL  s   
1

H 0 1  s  z1   1
1 s  
i  2   d i
  s 


 z H 0 k p
s
i  d 
1 s   1   NH 0 k p 

kp  N 

33 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We want a damped second-order response:
2
 1   2   d i  s  s 
1 s   i   s    1   
 z H 0 k p  c Qc  c 
 1   NH 0 k p 

Closed-loop denominator Second-order canonical form

 Choose a crossover frequency and a quality factor:

Qc  0.5 Non-ringing response c  27.3rd s 10 kHz

34 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 Four unknowns, four equations:
1 Ti 1 Td Ti 1 Td 1 Td Ti 1
   2  Ti   Td Ti  2
z1
H 0 k p c Qc NH 0 k p c N 0 Q0 N1 0

Q0 Qc 2 z1 20 2  Qc 2 z1 0c 2  Q0 Qc 2c 4  Qc z1 20c  2Q0 Qc z1 c 3  Q0 z1 2c 2

Td   1.116 ms
0c  Qcc   z 1
 Qcc 2   z1 c  Q0 Qc z1 0 
Q0 Qc 2 z1 20 2  Qc 2 z1 0c 2  Q0 Qc 2c 4  Qc z1 20c  2Q0 Qc z1 c 3  Q0 z1 2c 2
N   72.4
0 Qc z  Qcc   z c  Q0 Qc z 0 
1
2
1 1

Ti  
Qcc 2   z1 c  Q0 Qc z1 0
 14.6μs kp  

Qc z1 Qcc 2   z1 c  Qc Q0 z1 0   0.178
Q z1 0c  Q0 Qc0c 2 
H 0 Q00  z1  Qcc 
2

35 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We can now compute our transfer functions in Mathcad®
60
G s
40 G  s  100

180
 
20 log G1( i 2  f ) 10 20 0  
arg G 1( i 2  f ) 


0
Compensator  100

response G(s)  20
3 4 5
10 100 110 110 110
f

60

10 kHz
40 100

TOL  s  crossover
20
180
 
20 log T OL( i 2  f ) 10 0 
arg T OL( i 2  f )  

0

 100
Open-loop gain  20
TOL  s  m = 80°
response TOL(s)  40
3 4 5
10 100 110 110 110
f

36 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The closed-loop system is perfectly compensated

TCL  s  11.8 kHz

-3 dB 100
 10
180

20 log T 1( i 2  f ) 10  0  
arg T 1( i 2  f ) 

 20 TCL  s 
 100

 30
3 4 5
10 100 110 110 110
f

TOL  s 
TCL  s  
1  TOL  s 

37 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We can test the compensation with SPICE
L1 R6
vout
parameters a 75u 100m
10.0V c
5.10V
2
5.10V
12
5.00V
Vout2

fc=10k
Gfc=-20 d Resr C6
3
Vin=10 PWM switch VM p 70m 0.1u
Vin Rload 1.15V -1.27mV
Vpeak=2 {Vin} X7 10 9
5.00V 16 2.5
AC = 0 PWMVM
G=10^(-Gfc/20) L = 75u Cout
Fs = 100k 220u X5
pi=3.14159 R10
IntFilt AMPSIMP
{Ti/0.1u}

fz1=1.2k
fz2=1.2k 511mV
fp1=10k 5

fp2=50k DeriveFilt X3
R9 vout
GAIN

XPWM R11x SUM2

{Td/0.1u}
GAIN 0V 0V K1 = 1
10k 1.12mV
11 13
Wz1=2*pi*fz1 K = 1/Vpeak X6 15 K2 = -1
X9
Wz2=2*pi*fz2 1.02V POLE
AMPSIMP
C7 -
1.02V X8 K1
4 FP = fp2 0.1uF
Wp1=2*pi*fp1 K = -1
AMPSIMP R12 K1
-1.27mV
SUM2
23
LoL {10k/kpf} 0V K 5.00V+
SUM3 K2 K2
18 17 20
1kH K -1.02V 1.26mV S+A
1.15V K3
6 8 14
i=(1+fc^2/fp1^2)*(1+fc^2/fp2^2) S+A
X4 R8
j=(1+fz1^2/fc^2)*(1+fc^2/fz2^2) R13 Vref
T SUM3 X1 10k
100m VPIDFiltered
Wpi=sqrt(i/j)*G*fz1*2*pi K1 = 1 POLE
5

22 K2 = 1 FP = (N/Td)/(2*pi) R11
e=(Wp1-Wz1)*(Wp1-Wz2) CoL K3 = 1 -1.39uV 10k
f=Wp1*Wz1+Wp1*Wz2-Wz1*Wz2 1kF 25

Td=e/(f*Wp1) V1
AC = 1 Vdirect
N=((Wp1^2)/f)-1 X10
Ti=((Wz1+Wz2)/(Wz1*Wz2))-(1/Wp1) AMPSIMP
kpf=(Wpi/Wz1)-(Wpi/Wp1)+(Wpi/Wz2)

Tdd=1.116m
Tii=14.6u
kpff=0.178
NN=72.4

38 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We can test the compensation with SPICE

° dB
TOL  s 
180 60.0

m = 80°
90.0 30.0
TOL  s 

0 0

Perfect
10-kHz compensation
-90.0 -30.0 crossover dude!

-180 -60.0

10 100 1k 10k 100k

39 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We have a stable but oscillatory response!
1
5.08
   0.128
2
 
 2 
45 mV 1  
5.04   20  
 ln  45  
  
20 mV
Q  1 2  3.9
5.00

Wasn't it
supposed to
4.96 67 mV be perfect,
806 µs uh?

1.2 kHz
4.92
Vout  t 
500u 1.50m 2.50m 3.50m 4.50m

I out  1 A in 100 μs

40 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 Bode or Nyquist do not predict the oscillatory response
m
4.00 TOL  s 

2.00

1, j 0
1/GM
0
m
c

-2.00

-4.00

-4.00 -2.00 0 2.00 4.00 e

41 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We compensated the Vout to Vref path only!
Vin  s  GVin  s 
+ I out  s  Z out  s 
+ (s) d(s) -
Vref  s  GPWM H(s) +
Vout  s 
 T  s

G(s)

TCL  s  100
 10

 20 TCL  s 
0
?
 100
Vref  t   30 3 4 5
Vout  t 
10 100 10 10 10

42 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The output response is as expected

5.35

5.25

5.15

5.05

4.95 Vout  t 
300u 900u 1.50m 2.10m 2.70m

 Where is the issue coming from then?

43 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
44 • Chris Basso – APEC 2012
 Compensating the Buck – Method 1
 In reality, Vref is fixed: "we have a regulator, stupid!"
I out  s  Z out ,OL  s 
-
+
Vref  s  G(s) H(s) +
Vout  s 

 Because the system is linear, superposition applies

T s
Vout1  s   Vref  s  OL Vout 2  s   I out  s  Z out ,OL  s   Vout  s  TOL  s  I think he's
1  TOL  s  right…
TOL  s  Z out ,OL  s 
Vout  s   Vout1  s   Vout 2  s   Vref  s   I out  s 
1  TOL  s  1  TOL  s  Z out ,CL

 During the load step, vˆref 0: Zout fixes the response!

45 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 What matters is the output impedance Z out ,CL

model I2
Rth I out
Vth
I1

 What is the output impedance of a buck converter?

Vout  s 
It follows
L
rC I out  s  the form N s
Z out  s   R0
Rload D s
rL Cout
Ω Ω
No dimension

46 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The output impedance is a transfer function
Vout  s  response
Z out  s  
I out  s  excitation

 Let's find the term R0 in dc: open caps, short inductors

I out  s 
rL Rload R0  rL || Rload

 The zeros cancel the response

Vout  s   0?
 L  rL
L
rC I out  s  sL  rL  0 rL  s  1  0 z 
 rL 
1
L
Rload 1
1 z 
rL  rC  0 1  srC Cout  0 2
rC Cout
Cout sCout

47 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 The denominator is solely dependent on the structure

 It is independent from the excitation: set it to zero!

 There are two storage elements: this is a 2nd–order network
2
2 s  s 
D  s   1  a1s  a2 s  1   
0 Q  0 
1 2
 D must be dimensionless thus: a1   Hz  a2   Hz 

 The two possible terms for a1 are 1   2

L
 or   RC
 1 '2 R
 The two possible terms for a2 are
 '1  2

48 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 For a1 look at the resistance R driving L and C

 Look at the driving impedance at L while C is in its dc state

 Look at the driving impedance at C while L is in its dc state

R? R?
 L 
Rload Rload ... s   C  rL || Rload   rC  
 rL  Rload 
rL rC rL rC

1 2

R  rL  Rload R   rL || Rload   rC

49 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 how 1 (involving L) combines with '2 (involving C)? a2
 how 2 (involving C) combines with '1 (involving L)?
 Look at the driving impedance at C while L is in its HF state
 Look at the driving impedance at L while C is in its HF state

R? If we chose R? If we chose
Rload L
1  Rload  2  C  rL || Rload   rC 
rL  Rload
rL rC rL rC

Same result
L
 '2  C  rc  Rload   '1 
 1 '2   '1  2 rL   Rload || rC 

50 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We have our denominator!
 L   r  Rload 
D s  1 s   C  rL || Rload   rC    s 2  LC C 
r
 L  Rload   rL  Rload 

 The complete transfer function is now:

 L
 1  s  1  srC Cout 
 rL 
Z out  s    rL || Rload 
 L  2 rC  Rload 
1 s   C  rL || Rload   rC    s  LC 
 rL  Rload   rL  Rload 

See "Fast Analytical Techniques" from Vatché Vorpérian, Cambridge Press

51 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 It can be put under the following form:

Z out  s   R0
1  s  1  s  
z1 z2
2
s  s 
1  
0 Q   0 

 Where we can identify the terms:

rL 1
R0  rL || Rload  z1   z2 
L rC Cout

1 rL  Rload LCout0  rC  Rload 

0  Q
LCout rC  Rload L  Cout  rL rC  rL Rload  rC Rload 

52 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 If we now plot the output impedance, we see peaking
 For an non-oscillatory response, the peaking must be damped!
dBΩ f0
3.00
Z out ,max

-7.00

Z out ,OL
-17.0

rC
-27.0
Z out ,OL
-37.0 1  TOL  s 
10 100 1k 10k 100k

 It's not, this is where the problem comes from

53 • Chris Basso – APEC 2012
 Compensating the Buck – Method 1
 We organized a gain deficit right at the resonance!
60
G s
40 G  s  100

180
 
20 log G1( i 2  f ) 10 20 0  
arg G 1( i 2  f ) 


0
 100

G s  20
3 4 5
10 100 110 110 110
f

To tame the peaking, we must have gain at f0

 How much do we peak at f0?
2 2

Z out ,max 0   R0


1  0  z1  
1   0  z2 
2 2
 2 rC  Rload     L 
 1   0  LC out  
  0   C out  rC  rL || R 
load  
  rL  Rload   r
  L  Rload 

54 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 We can impose a magnitude to stay below rC
 evaluate the needed gain to fulfill this goal:

Z out ,max  f 0   Z out ,max  f 0  Z out ,max  f 0 

 rC  rC TOL  f 0  
1  TOL  f 0  TOL  f 0  rC
Closed-loop
output impedance

 Applying the numerical values of the buck:

1.12
TOL  f 0    16 or 24 dB
70m
 Is this enough to obtain a ringing-free response?

55 • Chris Basso – APEC 2012

 Compensating the Buck – Method 1
 No, ringing is reduced but not eliminated
dBΩ V
-10.0 5.08

-30.0 5.04

-50.0 5.00

-70.0 4.96
Z out ,OL Less than 0.4%
Increased
-90.0 OL gain 1  TOL  s  4.92 Vout  t 
10 100 1k 10k 100k 500u 1.50m 2.50m 3.50m 4.50m

 The peaking in the output impedance is still there!

 The notched zeros are the cause for the gain dip at f0
 We must find a different compensation method
56 • Chris Basso – APEC 2012
 Another (Bad) Example
 Can we crossover at 10 Hz according to this plot?
dB °
40.0 180 H 10 Hz   14 dB

20.0 90.0

0 0

-20.0 -90.0
arg H 10 Hz   0

-40.0 -180
1
10 100 1k 10k 100k
 We have no phase lag at 10 Hz, a type 1 could do?
57 • Chris Basso – APEC 2012
 Rolling-off the BW at Low Frequencies
 SPICE gives us the open-loop gain snapshot
rLf L1
10m 100u vout
a c vout
7 3
parameters

Vout=5V d R10
11
Rupper=10k PWM switch VM p 1m
Vin R11
fc=10 10 X3
PWMVM 16 1
Gfc=14 L = 100u
Fs = 100k Vcp(t) C5
1m
G=10^(-Gfc/20)
pi=3.14159

C1=1/(2*pi*fc*G*Rupper)
fp0=1/(2*pi*C1*Rupper) vout
8

-6 dB
GAIN

X1
GAIN
K = 0.5 PWM gain C1
{C1}
Rupper
2
10k

LoL
1kH
Vin

Verr Rlower
CoL X2 10k
1kF V2
AMPSIMP 2.5
9

Vstim
AC = 1

58 • Chris Basso – APEC 2012

 The Open-Loop Gain Looks Good…
 The type 1 confirms our 0-dB crossover frequency
dB °
T s
40.0 180

argT  s 
20.0 90.0

m  90
0 0
GM  24 dB

-20.0 -90.0
f c  10 Hz

-40.0 -180

100m 1 10 101 1k 10k

59 • Chris Basso – APEC 2012

 As Expected: It is Ringing!
 The load step reveals a ringing ac output
V
5.60
I out  2 A
Vout  t  1 f0
5.20

48
47

4.80

4.40

4.00 Vcp  t 
Vcp(t) is first order
5.50m 16.5m 27.5m 38.5m 49.5m Munch

60 • Chris Basso – APEC 2012

 Good dc Coupling, Weak ac Coupling
 H1 is stable per Bode analysis, but H2 is out of the loop…

Vin Vout(s)
H1(s) H2(s)

T(s)
d(s)
dc
Loose coupling
f < fc in ac, no signal
G(s) transmission >> fc…
ac
f >> fc
 The dc is fed back via the loop but not the ac…
 Oscillations are NOT due to the loop!
“Fast Analytical Techniques for Electrical and Electronic Circuits”, V. Vorpérian, Cambridge Press, 2002

61 • Chris Basso – APEC 2012

 Again, an Undamped RLC Network…
 No gain at resonance: the RLC network runs open loop
dBΩ
20.0

output filter f0
0

Gain action
-20.0

Zout,OL
-40.0

Zout,CL Properly compensated, fc = 4 kHz

-60.0
Zout,CL

1 10 100 1k 10k 100k 1Meg

 The system cannot reduce the Q at the resonant frequency

62 • Chris Basso – APEC 2012
 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
63 • Chris Basso – APEC 2012
 Compensating the Buck – Method 2
 In this method, we will focus on two parameters:
 crossover frequency fc
 phase margin m
 First, look at the ac response of the power stage:
dB °
40.0 180
resonance
H s
20.0 90.0

0 0
-20 dB
H  s  -132 °
-20.0 -90.0

-40.0 -180
ESR zero
10 100 1k 10k 100k

64 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 The peaking in H brings a severe phase lag at f0
 stay away from f0 , pick fc at least 10 times above (10 kHz)
 extract the phase/magnitude of H at 10 kHz:
H 10 kHz   20 dB H 10 kHz   132

 The compensator G must shape the loop gain TOL by

 providing a high dc gain for precision: place an origin pole

 reducing the phase lag at 10 kHz to provide a m of 70°

What compensation type do we need?

65 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 The origin pole brings a permanent phase lag of 90°
 added to the op amp inversion of -180°, we have -270°
 total phase (op amp and H) must be -360 + 70° = -290°
 the needed phase boost at fc is thus:
boost   m  H  f c   90  70  132  90  112

G s G s G s

boost

boost
270
G  s   270 270
G  s  G  s 
1 2 5 10 20 50 100 200 500 1k 10 100 1k 10k 100k 10 100 1k 10k 100k

Type 1 – no boost Type 2 – up to 90° Type 3 – up to 180°

66 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 type 3: an origin pole, a double zero and 2 poles
 this is our PID compensator!
 s  s    z1  s 
1   1   1   1  
  z1    z2   po  s    z2 
G s   
s  s  s   z1  s  s 
 1   1    1   1  
 po   p1    p2    p1   p2 

 The magnitude is derived as:  The argument is found to be:

2
 f 
2
arg G  f   arg N  arg D
 f z1 
1   1 
f po  f   f z   f z1   f 
 2 arg N  arctan       arctan  
G f   f  f 
f z1  f 
2
 f 
2    z2 
1  1   f   f 
 f p   f p  arg D  arctan    arctan  
 1  2  fp   fp 
 1  2
67 • Chris Basso – APEC 2012
 Compensating the Buck – Method 2
 Place the double zero at f0 ,the second pole at Fsw/2

 The 0-dB crossover pole is adjusted to provide +20 dB at fc

 The first pole is adjusted to provide the right m

 f z1   fc   fc   fc 
arg G  f c   arctan       arctan    arctan    arctan  
 fc       
 f z2   f p1   f p2 

fc
f p1    10.8 kHz
 1
 fc  f
1  z1  1
 fc  
tan  arg G  tan   tan    tan   
  f p  f
 c   f 
  2  z2  

68 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 The 0-dB crossover pole is adjusted to give 20 dB at 10 kHz
2 2
 fc   fc 
1  1 
 f p   f p 
 1  2
f po  G  f c  f z1  2 kHz
2 2
 f z1   fc 
1   1 
 fc   f z 
 2
 The final configuration is as follows:

|G(s)| 10.8 kHz 50 kHz

-1
20 dB
+1
0
-1 TOL 1.2 kHz   G 1.2 kHz  H 1.2 kHz   37 dB
0 dB
1.2 kHz
double
It should be ok to damp Zout
zero (+2) 2 kHz

69 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 Enter the PID coefficients from the poles/zeros positions

 d  55.5u  i  250u N  3.76 k p  3.1

dB ° dB
60 180 -10.0

30 90.0
TOL  s  -30.0
Z out ,CL
0 0
m = 70°
-50.0
-21 dB

-30 -90.0 -70.0

TOL  s  Z out ,OL
-60 -180 -90.0 1  TOL  s 
10 100 1k 10k 100k 10 100 1k 10k 100k

 More than 30 dB at f0 and absolutely no peaking in Zout

70 • Chris Basso – APEC 2012

 Compensating the Buck – Method 2
 The transient response with a 0.1-A load step is excellent

5.000 Vout  t 

4.998

4.996

8 mV
4.994

4.992
Z out ,CL 10k   21dB  84 mΩ
683u 906u 1.13m 1.35m 1.58m

71 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
72 • Chris Basso – APEC 2012
 Compensating the Buck – Method 3
 The capacitor stray elements affect the transient response
5.000
4.994
4.988 Depends on
Cout fc m
rC 4.982
Vout  t 
4.976
ESR 80.0u zoom 240u 400u 560u 720u

4.998
4.994 SlC Vout  t 
4.990
C 4.986 rC I out
4.982
SlC
94.7u 102u 109u 117u 124u
lC 1.10 t
900m
ESL 700m I out S  I out
500m t I out  t 
300m
94.7u 102u 109u 117u 124u

R. Redl et al."Optimizing the Load Transient Response of the Buck Converter", APEC 1998

73 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 During a step load, the converter fights the current change
Vmax
Vmax  Vmin
Vmax  Vmin
2
5.000 5.000

Vout  t  Vout  t 
Vmin
936u 1.16m 1.39m 1.61m 1.84m 500u 1.50m 2.50m 3.50m 4.50m

 Traditional compensation:  Adaptive Voltage Positioning

 inductive output impedance  resistive output impedance

Limited excursion Full-span excursion

74 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Whatever the gain, Zout meets the open-loop value beyond fc
dBΩ
0
lim Z OL ,CL
s 
Z out ,OL
-20.0
rC

-40.0

-60.0

-80.0
Z out ,CL
10 100 1k 10k 100k

 Make the output impedance equal to rC along the freq. range

75 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 How to force the output impedance to be resistive?

 s  s 
 1    1  
Z out ,OL   z1    z2  1
Z out ,CL   R0 2
1 T s  s  s 1
s
   1 z
 0  0 Q 1 H0 2
1
G s
 s  s
   1
  0  0 Q
Extract G(s)
to have:

Z out ,CL  rC

76 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Once G(s) is extracted, what filter is that?
40

 s   s 2 s
30
f p1
R0 1      1
 z
G s   2    0  Q0 dB 20

rC G0  s  f z1
G0 1 
  z  10
 1  G s
0
3 4 5
1 10 100 1´ 10 1´ 10 1´ 10

s
1
z
G  s   K0 G

s
1
p G

77 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Some parameter identification is now needed:
r r rL rC  1 1  rC
K 0  L C  1.8 a   47.7 p b  rL     74.2u
H 0 rC z z 0 2   z  z  Q0
1 2  1 2 

b  b 2  4ac
c  rL  rC  0.27 f pG  f z1  24 kHz f zG   580 Hz
4 a
40

30

dB 20
approximated

10

G s
0
3 4 5
1 10 100 1´ 10 1´ 10 1´ 10
YAO et al. ,"Design Considerations for VRM Transient Response Based on the Output Impedance", IEEE Proceedings, 2003

78 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The following op amp architecture will do the job
Vout  s 
vc  s  R
Z1  1
Vout  s  Z1
C1
 1 
R1 R2  R3  
R2 R3  sC1 
Z1 
 1 
R2   R3  
vc  s   sC1 

R2 1  sC1  R2  R3  
G s  
R1 1  sR3C1

R2 1 1
K0  z  p 
R1 1
C1  R2  R3  1
R3C1

79 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 A test fixture is assembled using a buck averaged model
L1 R6
75u vout
10.0V 0.3 5.00V
Vin a c Vout2
2 12
5.30V 5.30V

parameters d Resr
3
PWM switch VM p 30m
Vin
Vin=10 {Vin} X7
Rload
5.00V 2.5
Vpeak=2 AC = 0 PWMVM 16

L = 75u Cout
Fs = 100k 220u
pi=3.14159

fz1=580
fp1=24k
531mV
5 vout
Wz1=2*pi*fz1
Wp1=2*pi*fp1

GAIN
kv=1.8 XPWM Rf C2
GAIN {R1} {C1}
2.50V 2.50V
K = 1/Vpeak
8 10
R1=10k
1.06V
R2=R1/kv 4
X8 R12 R8
AMPSIMP {R2} {R3}
R3=R1*Wz1/(Kv*(Wp1-Wz1))
C1=(Kv*(Wp1-Wz1))/(R1*Wp1*Wz1) LoL
1kH 1.06V
Vin 6
2.50V
9
R13
100m Vout Vref R7
2.5 {R2*1.47}
22
CoL
1kF
7

V1
AC = 1

80 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 After stabilization we have good margins with a 20-kHz fc
dB ° dBΩ
40 0 40 Z out ,CL  s 
T s
20 -90 20

0 -180 0

T  s 
-20 -270 -20
30 dBΩ  31mΩ
-40 -360 -40

10 100 1k 10k 100k 10 100 1k 10k 100k

 The output impedance is purely resistive!

 But the dc gain is low: line and load regulation problems!

81 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Also, the gain expression K0 can be a problem:
rL  rC
K0   0  rL  rC
G0 rC

5.005
30 mV
4.995

4.985

4.975

4.965 Vout  t 
I out  1A
500u 1.50m 2.50m 3.50m 4.50m

 VM, fixed frequency, is not the best for Zout resistive shaping

82 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Going current-mode, fixed frequency is one way to go

duty-cycle L rL vout  s 
a c

rC
vc Rload
Vin PWM switch CM p

Cout

vc  s 
G s

 Use the PWM switch model in current-mode for Zout

83 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The large-signal model combines two current-sources

duty-cycle Ia Ic
a c a c
vc I L
vc dI c I1 I2 
PWM switch CM p Ri 2

vc p
Ri Ic
Vout L
I L vc 1  D  Tsw
I2   Vout
Ri 2L
I L t  toff

84 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 For ac study, we must obtain a small-signal model
 Vout 
1   Tsw
Vc 1  D  Tsw Vc  Vin 
I 2 Vout , Vc    Vout   Vout
Ri 2L Ri 2L
 Calculate the partial derivative coefficients to vout and vc

I 2 I 2 vˆc Tsw  2Vout 

vˆc  vˆout     1 vˆout
Vc Vout Ri 2 L  Vin 
gm
 Update the schematic with this linear source
 as R, L and C are also linear, Laplace applies

85 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 We look at the output impedance closed-loop
L rL vout  s 

rC
I2 Rload
vc  s  iout  s 
 gm  vout  s 
Ri
Cout

vc  s 
Tsw  2Vout 
gm    1
G s 2 L  Vin 
86 • Chris Basso – APEC 2012
 Compensating the Buck – Method 3
 We can apply the superposition theorem:
 vc  s    1 
vout  s     gm  vout  s    Rload ||  rC    iout is 0
R  sCout
 i   
  1 
vout  s   iout  s   Rload ||  rC    I2 is 0
 sCout  
 
 Considering vc  s   G  s  vout  s  we have:
 G s   1    1 
vout  s   vout  s   gm    Rload ||  rC     iout  s   Rload ||  rC   
 Ri    sCout    sCout 
  G s   1    1 
vout  s  1   gm    Rload ||  rC      iout  s   Rload ||  rC   
 Ri    sCout   sCout 
  

87 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The output impedance is thus:
  1 
 Rload ||  rC   
vˆout  s    sCout  
Z out ,CL  s    
ˆiout  s   G s   1 
1   gm   R ||
  load  C r   
 R i   sCout 
 Giving a small massage, we obtain:
Rload Ri 1  srC Cout
Z out ,CL  s  
Ri  G  s  Rload  Ri Rload  gm  R R  Ri rC  G  s  Rload rC  Ri Rload rC gm 
1  sCout  i load 
 Ri  G  s  Rload  R R
i load  gm 
"There is a
massage
for you"

88 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Factoring and re-arranging, we have:
1  s  z1 Rload Ri 1
Z out ,CL  s   R  s  R s  z1 
1  s  p1 Ri  G  s  Rload  Ri Rload  gm rC Cout
1
p  1
 Ri Rload  Ri rC  G  s  Rload rC  Ri Rload rC gm 
Cout  
 Ri  G  s  Rload  Ri Rload  gm 

 Now make Zout,CL(s) = rC and extract G(s):

Rload rC 2 gm  rC 2
1  sCout
Ri  Rload  rC  Rload rC gm  Rload  rC  Rload rC gm
G s 
rC Rload 1  srC Cout

89 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The compensator brings a single pole/zero response
1  s  z1 Ri  Rload  rC  Rload rC gm  R 1
G  s   G0 G0   i p 
1  s  p1 rC Rload rC
1
Cout rC

1 1 Very high frequency

z  
1
Rload rC 2 gm  rC 2 rC 2 can be neglected
Cout Cout
Rload  rC  Rload rC gm Rload

dB °
30
Ri  0.6 Ω
20
G s 100 rC  30 mΩ
Cout  220μF
10 0
G  s 
0  100

 10
3 4 5 6
10 100 10 10 10 10

90 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Sub-harmonic oscillations at Fsw 2 can cause peaking
 Place a zero at  Fsw as recommended by YAO and al.
Vout  s 
Z1 vc  s  R
 1
C1 Vout  s  Z1
 1 
R2  R3  
R1  sC1 
R2 R3 Z1 
 1 
R2   R3  
vc  s   sC1 

R2 1  sC1  R2  R3  
G s  
R1 1  sR3C1

R2 1 1
G0  z  p 
R1 1
C1  R2  R3  1
R3C1
YAO et al. ,"Design Considerations for VRM Transient Response Based on the Output Impedance", IEEE Proceedings, 2003

91 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 Run the SPICE simulation with the CM PWM switch model
d
499mV

L1 R6
duty-cycle 75u vout
10.2V 0.02 5.05V
parameters Vin a c Vout2
2 12
5.09V 5.09V

Vin=10.18 Slope
1.42V
vc Resr
3 4
se=20k PWM switch CM p 30m

Fsw=100k
comp. Vin
X3
{Vin} 5.05V R1 AC = 1
16
PWMCCMCM
pi=3.14159 AC = 0
L = 75u
2.5 I2
Cout
Fs = 100k 220u
Ri = 0.6
pi=3.14159 Se = Se
Ri=0.6
Cout=220u
rC=30m vout
wp=1/(rC*Cout)
Rfx C2
Wz1=pi*Fsw {R1} 2.50V 2.50V
{C1}
Wp1=1/(rC*Cout) 6 10

G=Ri/rC X8 R12 R14

AMPSIMP {R2} {R3}

R1=10k
R2=R1/G 2.50V
R3=R1*Wz1/(G*(Wp1-Wz1)) 9

C1=(G*(Wp1-Wz1))/(R1*Wp1*Wz1) Vrefx R7
2.5 {R2}

Automated calculations

92 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The output impedance is resistive, the system is stable.
Ω °
40.0m 180

20.0m 90.0 Z out ,CL  s 

0 0

-20.0m -90.0 Z out ,CL  s 

-40.0m -180

10 100 1k 10k 100k 1Meg

180 40.0
 m  66
90.0 20.0 TOL  s  TOL  s 
0 0

-90.0 -20.0 f c  40 kHz

-180 -40.0

10 100 1k 10k 100k 1Meg

93 • Chris Basso – APEC 2012

 Compensating the Buck – Method 3
 The output response is a square signal as expected

5.090
Vout  30 mV

5.080

f z1 @ Fsw
5.070

5.060

5.050 Vout  t 
I out  1A
500u 1.50m 2.50m 3.50m 4.50m

94 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
95 • Chris Basso – APEC 2012
 Output Impedance and Quality Factor
 How to get Q and m from the available signals?
9.00 V peak 1  8.21 V
6.33  5
k  0.414 You need
8.21  5
7.00 V peak 3  6.33 V enough signal
and ringing to
extract the data
5.00

3.00
td  841.5 μs
Must remain
1.00
small-signal!
Q = 3.6
Vout  t 
500u 1.50m 2.50m 3.50m 4.50m

2
   1 1 1
Q      f0 
 ln k  4  2 
2
td 1   2
  1
 ln k 

96 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Extraction is difficult with flat ac 2nd-order responses
7.00

5.00

3.00

1.00

-1.00 5-V step response Vout  t 

 dB 500u 1.50m 2.50m 3.50m 4.50m

180 80.0
Ac analysis
90.0 40.0

0 0
H s
-90.0 -40.0
H  s 
-180 -80.0

10 100 1k 10k 100k 1Meg

97 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 The phase drops at a different pace as Q changes
dB
20.0 Q = 10 H s
Q=5
10.0
Q=3
0

-10.0 Q=1
-20.0
Q = 0.6
10 100 1k 10k 100k 1Meg
° Q = 10
-10.0
Q=5 H  s 
-50.0 Q=3
-90.0
Q = 0.6

-130
Q=1

-170

10 100 1k 10k 100k 1Meg

98 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Is there a link between Q and the phase rate of change?
Group delay
d  
g   [s]
d
 Let's apply the definition to a 2nd-order network:

1 s  j 1
H s  2
H  j  
s  s  2 
1   1 2  j
0 Q   0  0 0 Q

a b
 
 
1  b   1
H    a 2  b 2  H    tan 1    tan 1  
2 2
a  0 Q   2 
 2      1  2 
1  2       0  
 0   0 Q 

99 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 We can apply g definition to the argument

 
 
 1
d tan 1  
 0 Q   2 
 1  2 

  0   Q0  2  0 2 
g   
d Q 2 4  2Q 2 20 2  Q 20 4   20 2

  0
2Q
g 
0

 At 0 the following formula links Q to the group delay

 g 0
Q   g  f0
2

100 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Let's apply the theory to a classical case, the RLC filter
Vin
R1 L1 parameters
{R} out
{L}
3 2 Vout
Vp=5
V1 C1 f0=1.2k
AC = 1 {C}
L=10u
C=1/(4*3.14159^2*f0^2*L)
w0=(L*C)^-0.5
Q=0.6
R=L*w0/Q
R2=1/(Q*C*w0)
Q1=(sqrt(L/C))/R
 plot the ac response Dzeta=(R/2)*sqrt(C/L)
Dzeta1=R/(2*L*w0)
 calculate the group delay Q3=1/(2*Dzeta)
per=1/(f0*sqrt(1-Dzeta^2))
 see if we can find Q tp=1/(2*f0*sqrt(1-Dzeta^2))

101 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
1
0

-10.0

-50.0 f 0  1207 Hz
-90.0

-130

-170
Z out  s 

260u

200u
 g  158 μs
140u

80.0u

20.0u Group delay

10 100 1k 10k 100k 1Meg

Q   g  f 0  1207  158u  3.14159  0.599

102 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Knowing Q can also lead us to the phase margin m
 dB
180 80.0 TOL  f  Approximation of T(s)
TOL  f  around fc
90.0 40.0 m

0 0

fc 1
TOL  s  
-90.0 -40.0
-1
 s  s 
0   1  
0 dB 
 0   2 
-180 -80.0 2
-2

10 100 1k 10k 100k

Open-loop ac analysis

103 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 If we consider the open-loop gain around fc only…
Closed-loop
1 TOL  s  1
TOL  s   
 s  s  1  TOL  s  s2 s
  1    1
 0    2  Unity return 02 0
Identify
1 1 0
 Qc  c  02
s2 s s2 s 2
 1  1
02 0 c 2 c Qc
Closed-loop Open-loop
data data

 We want to link Qc and the crossover frequency

1 1
TOL  s   TOL  s  
 s  s   s  s 
  1   0  Qc 22  2  1  

 0   2  Q 
 c 2   2 

104 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Calculate the TOL magnitude at crossover with Qc in:

Solve c
1
 jc   

Qc 22 2
2 2 c 
2  1  4Qc 4  1 
jc    c 4
 2  1   c 2
2
Q 
 c 2   2  TOL c   1

 Derive the argument of TOL, simplify it:

 
 jQc 22 2   
1  1 
arg TOL  s   arg  arg    arg
 s  s   c2  jc 2    c 2 c 2 
  1     j Q 2 2  Q 2 2 
 0    2   c 2 c 2 
  c 2 
 
1  Qc 22 2  1   2 
arg TOL  s   arg  1  tan      tan  
 c 2   c 
 Q 2 2 
 c 2 

105 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 If we substitute c by its definition, we have:
   
2 2
arg TOL  s     tan 1       tan 1  



1  4Qc 4   1  


1  4Qc 4   1 
 dB
180 60.0
TOL  f c    m  
TOL  s 
90.0 30.0
fc  m  TOL  f c   
0 0 TOL  f c 

-90.0 -30.0
TOL  s  m  
2
-180 -60.0   m  tan 1  

1 10 100 1k 10k 100k



1  4Qc 4   1 

106 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 We can now extract the closed-loop quality coefficient:
4
1  tan  m 
2
cos  m   4Q 4  1  1 
1c
Qc    m  cos  2

tan  m  sin  m   2Qc 
 
dB
20
 m  10
Qc
8
10 Qc  m  20
6  m  30
 
Q m m 0  m  45
4 -3 dB

 10
2 T s
0.5 1 T s  m  90
0
0 20 40 60 80 100  20
10 100 1 kHz 10 kHz 100 kHz
360
 m 76°
2  Closed-loop gain

107 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 The formula considers the vicinity of fc only: precision?
 dB
TOL  s  5.16 128 mV
80.0 120  m  23
40.0 60.0
TOL  s  5.08 47 mV

0 0 5.00

fc 4.92
-40.0 -60.0 Double zero
-80.0 -120
close to fc 4.84 Vout  t 
Step-load response
1 10 100 1k 10k 100k 869u 1.15m 1.43m 1.70m 1.98m

2
 
   1  4Q 4  1  1 
1 c
Qc      3.1  m  cos  2
  18
  47  4  2Qc 
ln
  128    
  
"Revisiting the Response of Closed Loop of PWM Converters ", S. Ben-Yaakov, IEEE Apec 2008

108 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 We can ac sweep the output impedance also and check g
 dB 3.5 kHz
140 0

100 -20.0 Z out  s 

60.0 -40.0 Z out  s 
20.0 -60.0

-20.0 -80.0

240u

120u

-120u

-240u Group delay  g  253 μs

1 10 100 1k 10k 100k

Q   g  f 0  3.5k  253u  3.14159  2.8 m  20.4

109 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 We can apply the technique to simple linear regulators
H s

Z out  s 
Vout  s 
+
Vin rC
vc  s 
Rload

Cout

 You have no means to perform open-loop analysis

 Plot its output impedance with a network analyzer…
S. Sandler, C. Hymowitz, "New Technique for Non-Invasive Testing of Regulator Stability", PET Magazine, September 2011

110 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 Plot magnitude and phase of Zout and compute g
 dBΩ
120 0 Z out  s 
80.0 -20.0
Z out  s 
40.0 -40.0

0 -60.0

-40.0 -80.0

500u

300u

100u

-100u
 g  574.9 μs
-300u Group delay
10 100 1k 10k 100k

Q   g  f 0  1k  574.9u  3.14159  1.8

 4Q 4  1  1   4  1.84  1  1 
1 1
 m  cos    cos    cos 1  857 m   31
 2Q 2
  2  1.8 2 
   

111 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 The LDO circuit is made of the following elements
Iout

Q1
2N3055 5.00V
7 Vout
10.0V
Q2
2N2219 5.69V
11
C2 L2
{C2} 13 20n
R1
10k
R4
C3 R2 10k 12
2 1 {C1} {R2} C1
Vin 10
1mF
10 2.50V 0V I1
2.50V
3 unknown
4
6.38V R3
2.50V 20m
6

Verr
X1 Vref R5
AMPSIMP 2.5 10k
GAIN = 30000

 We can open its loop and plot the open-loop gain TOL(s)

112 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 The open-loop plot confirms the experimental results

80.0 180

TOL  s 
40.0 90.0  m  30

0 0

f c  1 kHz
-40.0 -90.0

TOL  s 
-80.0 -180

10 100 1k 10k 100k

113 • Chris Basso – APEC 2012

 Output Impedance and Quality Factor
 The phase margin is increased to 60°
dBΩ °
0 120 Z out  s  1.3 kHz
-20.0 80.0 Z out  s 
-40.0 40.0

-60.0 0

-80.0 -40.0

500u

300u

100u

-100u
 g  185.5 μs
-300u Group delay
10 100 1k 10k 100k

Q   g  f 0  1.3k  185.5u  3.14159  0.76

 4Q 4  1  1 
1 1
 4  0.764  1  1 
 m  cos    cos    cos 1  457 m   63
 2Q 2
  2  0.76 2 
   
114 • Chris Basso – APEC 2012
 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
115 • Chris Basso – APEC 2012
 Considering a Delay in the Loop
 Before a decision is actually executed, a delay occurs
 The delay can be digital (computation time) or analogue
 A typical delay is the duty-ratio conversion in a VM converter
ton
G 5V

+ d
ton
Verr  t  0V Tsw
+ Tsw
- d t 
Prop. delay in
compensator - the 50 – 100 ns
Vramp  t 
range
Vramp  t 
Vp
Verr  t 
Tsw

116 • Chris Basso – APEC 2012

 A Delay is a Time-Domain Shift
 The output signal is the input signal that occured  s before

u t 
advance delay

y t 
t y t   u t  


t
t1 t2

u t 
Time delay y t 


117 • Chris Basso – APEC 2012

 Deriving the Delay
 To account for the delay, we need its Laplace expression

The plant H  s 

1  s  z1
vc  s  1 d s
Verr  s  s  s 
2
Vout  s 
V peak 1  
0 Q  0 

  delay   ?

118 • Chris Basso – APEC 2012

 Deriving the Delay
 To account for the delay, we need its Laplace expression

  y  t     u  t    

 Let's start with a sinewave phasor expression

Time delay
 y t 

u  t   Ae jt y  t   Ae j  t  

e a eb  e a  b y  t   Ae j t    Ae jt e  j  u  t  e  j

119 • Chris Basso – APEC 2012

 Deriving the Delay
 Let's take the Laplace transform of the new expression

 s Y s  s
s  j Y  j 
Y s  U se e  e  j
U s U  j 
 Euler phasor formula uses the argument as the exponent

arg e  j  

e  j  e j
e  j  1

 A delay block is a simple delay line!

u t  UTD y t 
K = 1, td = 250 ns

120 • Chris Basso – APEC 2012

 Building the Delay
 A delay line can time-shift the input signal

Delay block
parameters

tau=250n
T1
2 Vout
TD = {Tau}
50 = Zo
3 1

R1
E1
50
4 1
Vin
AC = 1 E2
1

 Best simulation practice is to buffer the input and the output

D. Adar, S. Ben-Yaakov, "Generic Average Modeling and Simulation of Discrete Controllers ", APEC Anaheim, 2001

121 • Chris Basso – APEC 2012

 Building the Delay
 A Bode plot confirms the 0-dB magnitude over frequency

4.00

2.00 e  s
Magnitude
0

-2.00
is flat
-4.00

-20.0 e  s
-40.0
-9°
-60.0

-80.0
10 100 1k 10k 100k 1Meg

180
100 kHz    100k  6.28  250n   9

122 • Chris Basso – APEC 2012
 Adding the Delay in the Laplace Domain
 We can now update our transmission chain with the delay

The plant H  s 

1  s  z1
vc  s  1 d s
Verr  s  e  st
s  s 
2
Vout  s 
V peak 1  
0 Q  0 

e  s 1  s  z1
H s  2
V peak s  s 
1  
0 Q  0 

123 • Chris Basso – APEC 2012

 Adding the Delay in the Laplace Domain
 st
 How do you deal with the term e in the transfer function?
 You don't: replace it with a poles/zeros combination!
 A pole will bring phase shift as frequency increases
1  
1
arg   tan  
1  s    
 But you still need to compensate the transmittance decrease
 A zero will do but now, all is neutralized!
2
 
1  
1  s       1   
mag  1 arg  tan 1    tan  0
1  s   
2
     
1  
  

124 • Chris Basso – APEC 2012

 Calling the RHP Zero for Help
 Replacing the LHP zero by a RHP zero does the job
 Both pole and zero magnitude neutralize each other
 The RHPZ phase lags and cumulates with that of the pole
dB °
20

10 1  s  100
 st 1  s 
e  1  s 
1  s  0 0

 1  s  
 10 arg    100
 1  s  
 20
3 4 5 6
10 100 10 10 10 10

125 • Chris Basso – APEC 2012

Mapping the Delay to the Pole/Zero Position
 Both arguments must be equal:
 1  s  
arg  e  s   arg     arg 1  s    arg 1  s  
 1  s  

 Replacing s by j:
   1    1   
  tan 1     tan     2 tan  
   
     

 Use the arctangent Taylor series equivalent: 0

3 5
     
     
x3 x5    
1
tan x  x    ...   2          
3 5  3 5
  
 
 

126 • Chris Basso – APEC 2012

 We Have the Padé Approximation
 Solving for  gives us…
2
 

 Substituting  in our first expression
s Henri Padé
1 1863-1953
e  s  2
s
1
2
 This is the 1st-order Padé approximant of an exponential
x 2nd-order 1 1
1 1 x  x2
ex  2 ex  2 12
x 1 1
1 1  x  x2
2 2 12
http://en.wikipedia.org/wiki/Padé_approximant

127 • Chris Basso – APEC 2012

 Padé Approximant Frequency Response
 When frequency increases, a phase deviation occurs
dB °
0

2  1  s 2 
arg    20
 1  s 2 
e  s  40
0
1  s 2
 60
1  s 2
2
e  s
 80

10 100 110 3 110 4 110 5 110 6

Use higher order approximants to improve precision

128 • Chris Basso – APEC 2012

 How to Avoid the Delay Line?
 A delay line adds computational burden in simulations
 Is there any simpler circuit that could be used?

parameters
R1
{1/(10n*Wtau)}
tau=250n
2 3
Wtau=2/tau C1
K1

10n SUM2
5 Vdelay1
K2
1
R3
{1/(10n*Wtau)} V1  s  X2
SUM2
7 6 K1 = 1
V1 C2 K2 = 1
10n E4 E3
AC = 1
1 1k
RHPZ

R1
Vout  s   V1  s   V1  s   V1  s  1  sR1C1 
1 sC1

129 • Chris Basso – APEC 2012

 How to Avoid the Delay Line?
 Good phase response of the analogue circuit
° dB
Op amp phase
0 4.00
response

-20.0 2.00

Op amp output magnitude

-40.0 0 12
11

Delay line output magnitude

-60.0 -2.00
Delay line phase
response

-80.0 -4.00

10 100 1k 10k 100k 1Meg

130 • Chris Basso – APEC 2012

 Delay Margin versus Phase Margin
 The characteristic equation is affected by the delay:

 s  1 T s  0   s   1  e  s T  s   0
Unity return denominator T(s)

 The stability condition for the magnitude does not change:

e  s max T c   1 e  s  1 T  c   1

 The stability condition for the argument does change:

 
  arg e  s max  arg T c    max  arg T c 

m    arg T c 
m
The max acceptable
 max 
delay in the loop c Solving
for  max

131 • Chris Basso – APEC 2012

 Delay Margin versus Phase Margin
 The delay margin is thus defined by:
   max   Current delay
Maximum acceptable
delay

1. A buck converter features a 250-ns internal delay

2. At a 100-kHz crossover, the phase margin is 49.5°
3. The maximum delay, accounting for 250 ns, is:
m 49.5 
 max    1.375 μs
c 2  100k 180
  1.375  0.250  1.125µs Maximum acceptable extra delay

132 • Chris Basso – APEC 2012

 Checking via Simulation
 The delay is simply added in series with the PWM block
L1 R6
vout
10.0V {L} 100m 5.00V
a c Vout
3 2 12
5.20V
5.20V

524mV
d Resr
9
PWM switch VM p {Resr}
Vin Rload
10 X1 5.00V 16 2.5
PWMVM2
L=L Cout
Fs = 1Meg {Cout}

parameters

tau=250n T1
TD = {Tau} out
50 = Zo
X4
out

3 1
Config_1
R1
E1
in 50
in

1
524mV vout E2
6 C2 1
{C2}
R3
GAIN

XPWM {R3}
R2 C1
GAIN Rupper 5.00V
{R2} {C1}
K = 0.5 {Rupper} 13
15 C3
1.05V 1.04V 2.50V
5 10 {C3}
LoL 1.05V
1G
4
2.50V
8 Rlower
CoL Verr

0V
7
1G
X2
AMPSIMP
V2
2.5
{Rlower}
Delay built with an adapted
V1
AC = 1 transmission line

133 • Chris Basso – APEC 2012

 Checking via Simulation
 With a 250-ns delay, the phase margin is acceptable
dB °

40.0 180
T s

20.0 90.0 m
49.5°
T  s 

0 0

fc
100 kHz
-20.0 -90.0

-40.0 -180

100 1k 10k 100k 1Meg

134 • Chris Basso – APEC 2012

 Checking via Simulation
 If we add 1.125 µs to 250 ns, no phase margin at all!
dB °
40.0 180 T s

20.0 90.0
m
T  s  0°

0 0

fc
100 kHz

-20.0 -90.0

-40.0 -180

100 1k 10k 100k 1Meg

135 • Chris Basso – APEC 2012

 Course Agenda
 Introduction to Control Systems
 Shaping the Error Signal
 How to Implement the PID Block?
 The PID at Work with a Buck Converter
 Considering the Output Impedance
 Classical Poles/Zeros Placement
 Shaping the Output Impedance
 Quality Factor and Phase Margin
 What is Delay Margin?
 Gain Margin is not Enough
136 • Chris Basso – APEC 2012
 Gain Margin Defines the Robustness
 GM defines the robustness of a system to gain variations

TOL  s  Conditions for

oscillations:
TOL  s   0

TOL  s  drift TOL  s   0 dB

0 dB 0°

-38 dB 1
20 dB/div GM   38dB
T  f 
T  f 
10 100 1k 10k 100k 1Meg

 If the gain drifts up by 38 dB, we have oscillations

137 • Chris Basso – APEC 2012
 Gain Margin Defines the Robustness
 In Bode representation but also in Nyquist
m

2 T  s    or  
II I

1 Im T  s   0
ReT  
1 GM
-1,j0
0 e
 m 0 rd 1
h
GM 
2 2
T  s  Re T  f   Im T  f 
c
1

 
III  IV
2 1
2  Re T  f 
2 1 0 1 2 GM

138 • Chris Basso – APEC 2012

 Gain Margin in Nyquist
 What really matters is the distance to the "-1" point
m
-1 ReT  
2 1  ReT  

1
1 ReT c  h
-1,j0 ReT c 
0 e
h
ImT c 
ImT  
c
1 2 2
h    1  Re T      Im T  

2
2 1 0 1 2 h  1 T s

139 • Chris Basso – APEC 2012

 Rejecting the Perturbation
 A closed-loop system rejects the incoming perturbation u1
perturbation u1(s)

+ (s)
u2(s) G(s) H(s) y(s)
+


T s 1
y  s   u2  s   u1  s 
1 T s 1 T s 1
h  1 T s 
S
Sensitivity function S

140 • Chris Basso – APEC 2012

 Watch for a Peak in S
 The trajectory must keep away from the "-1" point
 A 0.5-radius circle detects a peak in S: modulus margin
m m

2 2

1 1

1/GM 1/GM
-1,j0 -1,j0
0 M e 0 M e

c c
1 1
 

2 2
2 1 0 1 2 2 1 0 1 2
Spec limit Margin violation
141 • Chris Basso – APEC 2012
 Bode can also Show the Modulus Margin
frequency = 26.3k hertz
800m real = -649m
imag = -111m

400m
M 1/GM
imag

0
1 S min
-400m

-800m 

-1 -800m -400m 0 400m 800m

real
dB
20.0
1
S 
10.0 1 T s
6-dB limit

-10.0

-20.0 frequency = 26.3k hertz

sens = 8.67 dB

10 100 1k 10k 100k

142 • Chris Basso – APEC 2012

 Conclusion
 Switching or linear power supplies are regulators
 Applying a pure mathematical compensation brings problems
 Engineering judgment found output impedance guilty
 Lack of sufficient gain at resonance brings oscillations
 Standard poles/zeros placement method gives good results
 For high-speed dc-dc converters, resistive shaping rules
 Q to phase margin approximation requires engineering judgment
 Less known delay and modulus margins are useful figures!

Merci !
Thank you!
Xiè-xie!

143 • Chris Basso – APEC 2012