Chris Basso APEC Seminar 2021

Designing Compensators for the Control of
Switching Power Supplies

Christophe Basso – Technical Fellow
IEEE Senior Member
Public Information
Public Information
Feb 2021 – Rev. 0.3
Agenda
 Introduction to Control Systems
 Compensation Strategies with an Operational Amplifier
 The TL431 at Work in Compensators
 Building a PID
 Continuous-Time to Discrete-Time Domains
 Mapping and Compression
 Digital Compensation
 Practical Implementation of Filters
 Conclusion
Public Information
Agenda
 Building a PID
 Conclusion
Public Information
What is a Control System?
 An open-loop system links the output to the control variable
H s
U s control output
Y s  U s H s
Plant
 A control system observes the output and minimizes errors

Compensator
 s Verr  s 
+
U s Y  s   Verr  s  H  s 
-
Plant
G s U  s   Y  s   G  s 
H s
input output
Public Information
Shaping the Loop with the Compensator
 The compensator builds the error variable and ensures stability
 Insert poles and zeros to build the compensation strategy
 Choose how to cross over at fc with phase and gain margins
Compensator dynamic response
G s G  f 
  s Error
Vref Verr  s 
0 processing

k  Vout  s  Verr  s 
G s  G f 
Scaling factor Vout  s 
1 10 100 1k 10k 100k
 The block amplifies and shapes the error  between Vref and Vout
 Minimize the error between the setpoint and the output
Public Information
For a comprehensive analysis see APEC 2012 seminar: The Dark Side of Loop Control Theory
Building the Compensator the Analog Way
 Associate active and passive components to form the compensation chain
  s Verr  s  D s

Vref  s  G s Fm  s  H s Vout  s 
 compensator modulator plant
scaling factor
kD
1. Select poles/zeroes placement

2. Calculate components values
3. Solder resistors and capacitors  s
Change in strategy requires

new components values G s
6 Public Information
How do you Build a Compensator?
 The compensator can be implemented with analog components
Vout Vout Vout
verr
verr 

Op amp TL431 OTA
 Passive components suffer drawbacks:  Active components are not perfect!

1. Tolerance, aging 1. Open-loop gain, bias requirements
2. Sensitivity to temperature, humidity 2. Limited in bandwidth, slew-rate
3. Temperature drift
Compensating the Digital Way
 A digitally-controlled system is a discrete-time system
 It contains continuous- and discrete-time components
Digital comparator
  Vc D
Vref  Vout
 Digital controller Digital modulator plant
000
N bits N-bit scaling factor
001
010 kD
011
100
quantizer
Quantization Sampling
(ADC)
v  n
00101 v  n  1 v  n  1 v t 
10011
10111 Ts
t t
 n  1 Ts nTs  n  1 Ts
Binary values Discrete time Continuous time
* designates a discrete variable M. Jovanović, Introduction to Digital Control of Switch-Mode Converters, in-house course, PHX 2014
Why do we Need to Close the Loop?
 We want to compensate the power stage deficiencies to obtain:
 Speed  High bandwidth

 Precision  Large dc gain
 Robustness  High gain below fc
overshoot precision
2.5 V 5V
vout  t 
0V vref  t  0V
speed
Stimulus Response
What compensation strategy?
Choosing the Crossover Frequency fc
 Selecting fc depends on the topology and its control mode
Topology Voltage Mode Current Mode
F * F *
Buck 3  f 0  f c  sw f c  sw
2 2
Boost 3  f 0  f c  0.3  f RHPZ f c  0.3  f RHPZ
Buck-boost 3  f 0  f c  0.3  f RHPZ f c  0.3  f RHPZ

Continuous conduction mode * Theoretical upper limit
Too high a crossover frequency affects susceptibility to noise!
f0 is the LC network resonant frequency
Crossover Frequency Impacts Response Speed
 Adjust fc to meet the transient response you want
5.20 vout  t  Vout 

I out
f c  5 kHz 2 f c Cout
5.10 f c  10 kHz
(V) 5.00
45 mV f c  20 kHz
67 mV
4.90 I out 1
fc 
130 mV Vout 2 Cout
4.80 CCM buck
Depends on Cout and fc
converter
990u 1.17m 1.35m 1.53m 1.71m
11 Public Information rC  Z Cout @ f c

Transient Response Depends on Phase Margin
 Phase margin affects overshoot and recovery time
5.05
vout  t  m  30 Crossover is constant to 5
m  45
kHz
 m  60
5.00
Step
m  90 load
(V) 4.95
10 A
4.90
m  80 7A iout  t 
 m  70
4.85
Iout = 3 A in 3 µs
Marginally depends on PM
990u 1.17m 1.35m 1.53m 1.71m
Open-Loop Phase Margin Affects Closed-Loop Response
cos  m   m Open-loop phase margin It is an approximation
Qc  Qc Closed-loop quality factor for a 2nd-order system!
sin  m 
20
 m  10
Qc 10  m  20
8 Qc
 m  30
6  m  45
 
Q m m (dB) 0
-3 dB
4
 10
2
T s
1 T s  m  90
0.5
0  20
0 20 40 60 80 100 10 100 1 kHz 10 kHz 100 kHz
 m
360
76° Closed-loop response
2 
C. Basso, “The Dark Side of Loop Control Theory”, APEC 2012 Professional Seminar
Agenda
 Building a PID
 Conclusion
Public Information
Meeting the Selected Crossover Frequency
 Extract magnitude and phase of the power stage transfer function at fc
° dB
180 40.0
Vout  f  Attenuation at 10 kHz
VFB  f 
90.0 20.0
G fc  12 dB
0 0
Vout  f 

VFB  f 
-90.0 -20.0
-180 -40.0
PS  140
10 100 1k 10k 100k 1Meg
Shift the Magnitude Curve Up or Down to Meet Crossover
 Tailor the compensator to offer a 12-dB gain at the selected crossover
° dB
180 40.0 Vout  f  Shift up by 12 dB
VFB  f 
90.0 20.0 f c  10 kHz
0 0
-90.0 -20.0
The compensator is tuned to

-180 -40.0
offer a 12-dB gain at 10 kHz
10 100 1k 10k 100k 1Meg
A Permanent 270° Phase Lag
 The compensator is typically built around an inverting op-amp
Permanent Log f
Zf phase lag
Z f s
Vin Zi G s   -180°
Vout Zi  s 
Vref +
G  f 
 A compensator typically includes a pole at the origin: G  0   

1
Zf 
sC f Add another
-90° Log f
1 1 -90°
G s  
Vin Zi sC f Z i  s 
Vout -270°
Vref + pole s = 0
G  f 
Stay Away from the 360° Limit: Build Phase Margin
 Add up plant and compensator phase responses to total less than -360°
Compensator: integrator Compensator fc
Log f Log f
G  f  G  f 
-145°
Phase boost
Decrease compensator   125
-270° lag near crossover -270°
+ +
Plant Plant
Log f Log f
H  f    90
H  f  H  f c   145
  145
-180° -180°
H  f c   270  boost  360  m
Loop phase
= Loop phase
=
Log f Log f
-180° -180°
f c1 f c2 m  70
boost   m  H  f c   90
-360° -360° 0°
0° PM <0 PM
T  f  signal returns T  f 
in phase
How to Create and Adjust Phase Boost
 Combining a zero and a pole lets you adjust the boost from 0° to 90°
z fc
1 peaks at Apply k factor fz 
s fz f
G  s   G0 G  f c     tan 1  tan 1 c fc  fz f p k
s fc fp
1 f p  k  fc
Mid-band p
gain magnitude phase
180°
k 1 k  25
G f  G  f 
20 dB f c  1 kHz
0 dB
k  25
G0
-20 dB boost
f c  1 kHz
k  1, 2, 5, 10, 25
k 1
90°
10 Hz 100 kHz 10 Hz 100 kHz
D. Venable, The k-Factor: a New Mathematical Tool for Stability Analysis and Synthesis, Proceedings of Powercon 10, 1983
Boosting the Phase up to 180°
 By placing a double zero and a double pole, the boost increases up to 180°
  z1  s 
1    1  

 s    z2  f z1 fc f f
G  s   G0 G  f c     tan 1  tan 1  tan 1 c  tan 1 c
 s  s  fc f z2 f p1 f p2
1   1  
 p  p 
magnitude  1  2  phase
270°
k 1 f z1  f z2 k  25
G f  G  f 
k  25 f p1  f p2
20 dB f c  1 kHz
0 dB 1 k2 1 Boost up
G0   to 180°
1 k  G fc
G fc k 2 1
k2
-20 dB
f c  1 kHz
k  1, 2, 5, 10, 25
k 1
90°
10 Hz 100 kHz Public Information
10 Hz 100 kHz
20
There are Three Compensator Types
Gain 
z
  z1 s 
1 Inverted zero 1    1  

1 s  s    z2 
G  s   G  s   G0 G  s   G0
s s  s  s 
1 1   1  
 po p  p  p 
 1  2 
Type 1 Type 2 Type 3
G f  G f  G f 
0 to 90 0 to 180
G  f  G  f  G  f 
0° phase boost Up to 90° phase boost Up to 180° phase boost
Switching Topologies and Compensators
 Select the adequate compensator based on the plant phase lag
clock VCC
Voltage mode clock VCC Current mode

S DRV
driver
Q
S DRV
R
Q
CMP verr  t  R CMP
FB FB
+
-
CS
-
+ LEB kdiv LEB
vCS  t  CS verr  t  vCS  t 
+
- VCS ,max
Vref Vp Vref
VCS ,max
vsaw  t 
PWM block
 in CCM, 2nd-order response: type 3  in CCM, 3rd-order response: type 2

 in DCM, 1st-order response: type 2  in DCM, 1st-order response: type 2
Heavily-damped 2nd-order
Provide Gain or Attenuation at fc: no Boost
 Type 1 is a simple integrator without phase boost
20

Vout  s  G fc  20 dB @ f c  10 Hz G fc  10 20
 0.1
C1 fc
R1 f po   100 Hz
G fc
60 0
V    s   0
Virtual ground fc
VFB  s  40 G f 
20
(dB) (°)
Vref Rlower 0
G  f  -270°
1 1
G s    po  -20
s R1C1
-40
 po No role 0.1 100
23
in ac Public Information
The Finite Open-Loop Gain Limits the dc Gain
 The op-amp characteristics affect the frequency response
 Virtual ground is lost in dc AOL kdiv
G s  
 s  AOL kdiv  s 
Vout  s  1   1 
C1   po    p AOL 
kdiv AOL
100
R1 38 kΩ
80 dB – 30 Hz 66.4 fc G f 
Rlower 50
kdiv 
V    0   kdiv  Vout Rlower  R1
VFB  s  -13.6 dB
(dB) 0
AOL
Rlower 10 kΩ
-50 G  f 
p
1st pole 2nd pole
-100
 Voltage divider enters the picture at dc 1 mHz 1 Hz 100 Hz 100 kHz
Simulating Large-Open-Loop-Gain Compensators?
 If you use an op-amp model, use another E-source to fix the operating point
E1
LoL -100 dB °
G f 
1G
ok 12.0V
5
12.0V
4
40 270
C1
R1 2.50V
0.1uF Rdum
2.62V 38k 8
3
Vcc
2.50V 12.0V 0 180
1 6
12.0V CoL
1G
VCC
VFB
VEE 2.50V 0V V3
-40 90
-12.0V
Vee
2
7
V2
AC = 1
Vcc Vee 2.5
G  f 
Vref
X2 2.5
R2 V4 V5 -80 0
10k 12 12
TL081M
AOL  106 dB -120 -90
1 10 100 1k 10k 100k

 E1 fixes the operating point to 2.5 V
 12 V at R1 confirms regulation is ok
Boosting the Phase at Crossover with Type 2
 By adding another zero and pole to the type 1, phase boost is created
z
1
s R2 C1
Vout  s  G  s   G0 G0 
s R1 C1  C2
1
C2 p
1 1
C1 R2 p  z 
R1 CC R2C1
R2 1 2
C1  C2
2
 fc 
   1
VFB  s  R1 f p  fp 
R2 
G fc  f p  f z   f  2
z
  1
Vref Rlower  fc 
1 C1
C1  C2 
2 R2 f z 2 f p C1 R2  1
C. Basso, Designing Control Loops for Linear and Switching Power Supplies, Artech House, 2012
Design Example with a Type 2 Compensator
 Stabilizing a buck converter operated in current-mode control
524mV
dc
Vout
rL L1
duty -cy cle 100m 47u 5.00V
5.01V 5.00V
a c
2 1 4
X1
PWMCM
L = 47u vc
Fs = 100k PWM switch CM p C3
Ri = Ri
Se = Se {C2}
695mV
231mV R6 rC
9.00V 12
3 10k 50m
2.50V
Vin Verr 7 16 5.00V Rload
{Vin} R2
GAIN
8
XPWM C4 {RL}
{C1} {R2}
GAIN
K = 0.36 LoL
1k
695mV 695mV
C2
6 17 2.50V 180u
parameters CoL 2.50V
1k 5
0V 10
19 X3 R5
Vin=9 AMP384X Vref
V5 2.5 10k
RL=5 AC = 1
27 4/1/2021 Public Information

C. Basso, Switch-Mode Power Supplies: SPICE Simulations and Practical Design, McGraw-Hill, 2014
Plot the Control-to-Output Transfer Function
 Read the Bode plot and extract the plant attenuation and lag at crossover
parameters
° dB Control-to-output transfer function Automate the
Rupper=10k calculation for easily
180 20 Hf fc=10k
Gfc=-24 trying different
90 0 H  f c   24 dB pfc=-61 strategies
pm=70
H  f  Se=20k Slope comp.
0 -20 Ri=600m
boost=pm-(pfc)-90
-90 -40
Vout  f  G=10^(-Gfc/20) BOOST = 4.10e+001
Hf H  f c   61 boost=pm-(pfc)-90 G = 1.58e+001
-180 -60 Vc  f  pi=3.14159 PI = 3.14e+000
K=tan((boost/2+45)*pi/180) K = 2.19e+000
10 100 1k 10k 100k
C2=1/(2*pi*fc*G*k*Rupper) C2 = 4.58e-011
C1=C2*(K^2-1) C1 = 1.75e-010
R2=k/(2*pi*fc*C1) R2 = 2.00e+005

Immediately Check the Resulting Performance
 It is easy to verify the strategy impact on the transient response
° dB Loop gain transfer function (V) Step-load response

Tf 5.08
80 180
m = 63°
40 90 5.04 f c  10 kHz,  m  63
T  f 
0 0 5.00
fc = 10 kHz f c  5.2 kHz, m  77

-40 -90 4.96
1.5 A
1A
-80 -180 4.92 vout  t  1 µs
10 100 1k 10k 100k 1.44m 1.73m 2.01m 2.29m 2.58m
 Perform parametric sweeps with Monte Carlo analysis to check robustness

Automated Type 2 Calculations with SIMPLIS®
G f 
G  f 
30 Press F11 Public Information

Changing the Phase Boost on the Fly
 How to keep magnitude at fc but increase the boost by a certain amount?
fc
fc Increase the
boost by k1 72°
60°
fz fp k1=20% f z1 f p1
 
 f   f   x y 
tan 1  c   tan 1  c   k1  boost tan 1  x   tan 1  y   tan 1  
  fz   fp   1  xy 
 
 
f z1   f z
tan  boost  k1   
2
4 f p fz  f p fz  f p fz
 2
fc  2 f p f z  2
    fc 
2 fz  tan  boost  k1   f c  fc  fp
  f p1 

Increase the Boost by 20%
 Spread pole and zero while keeping mid-band gain untouched
0 0
fc G f  fc G f 
40 40
20 k1=20% 20
(dB) (°) (dB) (°)
0 0
boost boost
-20 -20
fz fp f z1 f p1
G  f  G  f 
-40 -360 -40 -360
10 100 1k 10k 100k 10 100 1k 10k 100k
boost  60 boost  72

f z  268 Hz   0.5911 f z1  158 Hz
f p  3.73 kHz f p1  6.3 kHz
Watch for Boost Distortion Brought by the Op-Amp
 The response must be imposed by the pole and the zero, not by the op-amp!
100
G f 
Op amp
50
20 dB
(dB) GBW
Ideal type 2
0
fc
20fc
-50
1 10 100 1 kHz 10 kHz 100 kHz 1 MHz 10 MHz
 Limit phase/gain distortion by selecting an op-amp still having gain at 20fc

C. Basso, Understanding Op-Amp Dynamic Response in a Type 2 Compensator, www.How2Power.com
Increasing Phase Boost up to 180°
 The type 3 compensator adds another pole/zero pair to the type 2
  z1   s 
Vout  s  1   1  
s  
  z 2  R C1
G  s   G0 G0  2
 s  s  R1 C1  C2
C2 1  1  
  p   p 
R1 R3  1  2 
1 1 1 1
C1 R2  z1   z2   p2   p1 
R2C1  R1  R3  C3 R3C3 CC
R2 1 2
C1  C2
C3 2 2
 f   f 
1  c  1  c 
VFB  s  R1 f p1  fp   fp 
 1  2
R2 
G fc f p1  f z1 2
 fc 
2
 f z1 
1   1 
Vref  fc   f z 
Rlower  2
C1 1 f p2  f z2 R1 f z2
C2  C1  C3  R3 
2 f p1 C1 R2  1 2 f z1 R2 2 Rupper f p2 f z2 f p2  f z2
Design Example with a Type 3 Compensator
 A voltage-mode-controlled buck converter is a typical usage for a type 3
PWMVM
L = 47u Vout
Fs = 100k rL L1
100m 5.00V 47u 5.00V
a c 5.01V
3 14 1
444mV d
9
PWM sw itch VM p Verr
C4
{C2} R3
592mV {R3} rC
2 R6
4
C1 R4 10k 5.00V 50m
Vin {C1} {R2} 15 C3
{Vin} Rload
{C3} 8 5.00V
GAIN
7
XPWM 2.50V {RL}
GAIN
K = 0.75 LoL
1k
592mV
10.0V 592mV
C2
5 17
2.50V 190u
CoL 10 2.50V
parameters 1k 16
0V X2
19 AMPSIMP Vref R5
Vin=10 V5 2.5 10k
RL=5 AC = 1
Extract Data from the Control-to-Output Transfer Function
 Choose fc beyond f0 but watch for the op-amp gain-bandwidth product
parameters
dB ° Control-to-output transfer function Poles and zeroes can
Rupper=10k
40 180 fc=10k be placed
H  f c   12 dB Gfc=-12
Hf pfc=-144 independently unlike
pm=60
20 90 boost=pm-(pfc)-90
with k-factor
G=10^(-Gfc/20)
H  f  pi=3.14159
0 0 fz1=1k
fz2=1k
fp1=fc/tan((2*atan(fc/fz1)-tan(fc/fp2))-boost*pi/180)
-20 -90 fp2=50k
C1=1/(2*pi*fz1*R2)
Vout  f  C2=C1/(C1*R2*2*pi*fp1-1)
C1 = 2.63e-008
Hf C3=(fp2-fz2)/(2*pi*Rupper*fp2*fz2)
C2 = 2.70e-009
-40-180 Vc  f  H  f c   144 R3=Rupper*fz2/(fp2-fz2)
C3 = 1.56e-008
R3 = 2.04e+002
a=sqrt((fc^2/fp1^2)+1) A = 1.37e+000
10 100 1k 10k 100k b=sqrt((fc^2/fp2^2)+1) B = 1.02e+000
c=sqrt((fz1^2/fc^2)+1) C = 1.00e+000
d=sqrt((fc^2/fz2^2)+1) D = 1.00e+001
R2=((a*b/(c*d))/(fp1-fz1))*Rupper*G*fp1 R2 = 6.06e+003
Assess Compensation Effects with Transient Steps
 The buck transitions from a 2nd- to 1st-order model when going to DCM
dB ° Loop gain transfer function (V) Step-load response

Tf  5.08
40 180
m = 60°
20 90 5.04 CCM
T  f 
0 0 5.00
fc = 10 kHz
-20 -90 4.96 DCM DCM CCM
0.1 A 1A
0.05 A 0.5 A
-40 -180 4.92 vout  t  1 µs 1 µs
10 100 1k 10k 100k 1.20m 3.60m 6.00m 8.40m 10.8m
 One of the zeroes could go to a lower frequency and improve DCM response
Agenda
 Building a PID
 Conclusion
Public Information
The TL431 in Compensators
 The TL431 is the most popular choice in adapters designs
 It associates an open-collector op amp and a reference voltage
 The internal circuitry is self-supplied from the cathode current
 When the R node exceeds 2.5 V, it sinks current from its cathode
K
R
K
TL431A R
A R
A
2.5V K
A
Many Different Variations Around the Part
Part number Max voltage (V) Min operating Reference voltage Packages
current (µA) (V)
TL431 37 1000 2.5 TO92, SO8, µ8,

DIP8
TLV431 18 100 1.24 TO92, TSOP5, SOT-
23
NCP431 37 60 2.5 TO92, SO8, SOT-23
NCP100 7 100 0.9 TO92, TSOP5
The TL431 Type 2 is Associated with the Optocoupler
 A type 2 based on the TL431 requires a single capacitor C1
Vout Vdd Type 2 TL431
pole Vout
C2
R pullup RLED R1
R1
C1 Fast Slow
lane lane
VFB zero
R2
Rbias
VFB pole
C2 C1 zero
Rlower
Vref
TL431 Rlower
Type 2 op amp
The Optocoupler Adds its Own Dynamics
 The optocoupler hosts a low-frequency pole: characterize it
 Reproduce the controller internals where the opto connects
Vdd  5 V Frequency
Response
Tweak to the same bias Analyzer
R pullup
2.8 V VB VA Req
FB
PWM
R1 reset 20 kΩ 100 kΩ 4.7 µF
c a
Req
R2
5V
GND vCS  t  e k
10-100 kHz
Adjust
bias
The Original Pole is Modified by the Optocoupler
 A simple 1st-order model includes an output capacitor
1
Copto   2 nF
2 f p R pullup
c a
Ic CTR Vf
Copto If
Ic
e CTR  k
If
Simplified 1st-order model
The LED Resistance Influences Bias Point and Gain
 Select RLED so that the optocoupler always pulls VFB low in worst-case
Vout
dB ° R pullup
G0  CTR
Vdd
40.0 180
Not ok RLED
RLED
Requires
Minimum less
f c  500 Hz
gain!
20.0 90.0 H  s than 17 dB
R pullup I bias of gain
VFB 0 0
Vf
-17 dB
VCE , sat -20.0 -90.0
VTL 431,min arg H  s 
-40.0 -180
ok
Vout  V f  VTL 431,min
RLED ,max  R pullup CTR min
Vdd  VCE , sat  I bias CTR min R pullup 10 100 500 1k 10k 100k
Public Information
Apply the Compensation to a CM-LLC Converter
 The current mode LLC can be stabilized with a type 2 compensator
A Current-Mode LLC Converter Operates in Free-Running
 The on-time duration is precisely memorized and mirrored for the off-time
Exact 50% duty ratio
Next cycle
Counter
reset
Hi-frequency + vCS  t 
clock stop
- verr  t 
D0 Dn
Digitized ton
duration
Simplified ton Replication with SIMPLIS®
 Using digital counters with SIMPLIS® is not an option if small-signal analysis is wanted
 A simple capacitor-based circuit does the job well and remains compatible with POP
 Capacitor Ct is charged by a
constant 1-µA current.
 When the CS comparator sends the (V) vCt  t 
reset signal, capacitor stops
charging: end of ton.
 The Ct capacitor is now discharged
with the same 1-µA source.
 When its voltage touches ground,
this is the end of toff. (µA) iCt  t 
Coming from the

CS comparator
(V)
Q
(V) Q
Exact 50% duty ratio
No Change in Plant Response at Different Inputs
Hf H  f 
Vin  400 V
Vin  400 V
H  fc  V  15 dB
in  370 V
Fsw  74 kHz
H  f c V  370 V
 85
in
 One benefit of the CM operation

is the 1st-order response.
For a 1-kHz crossover
 The control-to-output transfer frequency, a simple type-2
function remains the same from compensator will do
370 V dc to 400 V dc. Public Information
Type 2 Compensator with Optocoupler
 Component values are calculated by SIMPLIS® to meet gain and phase boost targets
.GLOBALVAR Gfc=-24 * magnitude at crossover *
.GLOBALVAR PFC=-87 *plant phase at crossover *
.GLOBALVAR PM=70 * phase margin goal *
.GLOBALVAR boost=pm-(pfc)-90 * required boost *
*
*GLOBALVAR fc=1k * targeted crossover *
*
* Pole-zero calculations *
*
.GLOBALVAR k=tan((boost/2+45)*pi/180)
.GLOBALVAR fp=k*fc
.GLOBALVAR fz=fc/k
.GLOBALVAR Vout=12
.GLOBALVAR Ibias=250u
.GLOBALVAR Vref1=2.5
.GLOBALVAR Rlower=Vref1/Ibias
.GLOBALVAR Rupper=(Vout-Vref1)/Ibias
.GLOBALVAR Rpullup=18k
*
* Optocoupler specifications *
*
.GLOBALVAR Fopto=15k
.GLOBALVAR Copto=1/(2*pi*Fopto*Rpullup)
.GLOBALVAR CTR=0.3
*
 Automated calculations account for the optocoupler pole contribution
CM-LLC Compensated Loop Gain
 Plot the loop gain to check crossover frequency and phase margin at this point
(dB) (°)
 m  70
f c  1 kHz
Tf T  f 
 You can now assess parasitics impacts by sweeping these terms in Monte Carlo runs
Compensated Transient Response
 CM makes transient response stable and immune to input voltage variations
vout  t 
Vin  400 V
I out  10 to 20 A in 1µs
OUT / V
Vin  385 V
I out  10 to 20 A in 1µs
Agenda
 Building a PID
 Conclusion
Public Information
The PID Controller in the Parallel Form
 The block combines three actions: proportional, integral and derivative
P k p  t 
+  t  vc  t 
vref  t   I ki    t   vout  t 
- Plant
d t 
D kd
dt
Compensator
 Each processing block serves a purpose:

P adjusts the correction I makes sure the static D tailors reaction strength
signal to the disturbance error is reduced to zero in relationship to
amplitude disturbance speed
C. Basso, The Dark Side of Loop Control Theory, APEC 2012 Professional Seminar
Many Methods to Tune a PID Compensator
 A PID can be adjusted through different approaches:
Manual
Settling
Parameter Rise time Overshoot Static Error Stability
time
Small
kp Decrease Increase Decrease Degrade
Change
ki s 1  Decrease Increase Increase Eliminate Degrade
Increase
Minor Improve if
kd s  Decrease Decrease No effect
change small kd
ki kp
G s  kp   skd Parallel form i  s 
1 s ki
H s  2
Factor kp
s s  Tune k p ki kd
1    1  k
0  0  G  s   k p 1   s d  Standard form  d  d s 
 s i  kp
The plant The compensator
https://en.wikipedia.org/wiki/PID_controller
Each Coefficient Affects the Transient Response
G s H s 1
 Calculate the closed-loop response to a 1-V step input:   invlaplace  vout  t 
1 G  s H  s s
k p  1 ki  1µ kd  0 k p  10 ki  1k kd  0 k p  10 ki  1k kd  1m
1.5 1.5 1.5

1.0
target 1.0 1.0
(V)  (V) (V)

0.5 0.5 0.5
vout  t  vout  t  vout  t 

1.5 1.5
k p  10 ki  1µ kd  0 k p  10 ki  10k kd  0 k p  10 ki  1k kd  4m

1.0  0 1.0
(V) (V) (V)

0.5 0.5
vout  t  vout  t  vout  t 

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
t (ms) t (ms) t (ms)

0  1 krd s Q  0.5
Watch for the High-Frequency Gain
 The kd term does not roll the gain off at high frequency and can bring noise issues
ki k p  10 ki  1k kd  4m
G s  kp   skd 100
s
 z   G  f 
G  s   G0 1  1  1 
s
 80 100 lim s  kd  

 s    z2  s 
kp k p  k p 2  4 k d ki
(dB) 60 +1 0 (°)
The derivative term cannot
z  
1
kd 2kd be physically implemented
2
40
-1
G f   100
k p  k p  4 k d ki ki 0
z  G0 
2
2k d z 1
20
10 100 1k 10k 100k 1Meg
 Add a low-pass filter to force a flat gain at high frequency

 1  s d
G  s   k p 1   s d  N
s PID with pole  p  Filtered PID
 s i  1 d d
N
Transform PID Coefficients in Poles and Zeroes
 If you add a 3rd pole, you end-up with the type 3 compensator
    z1  s 
  1    1  
k skd s 
GFPID  s    k p  i   1 GT3  s   G0
  z2 
 s 1 s 
1 
s  s  s 
1    1   1  
p   p1
N  kd   gain
  p1 
1
p2
Filtering Extra  p2 
pole hi-frequency pole
Identical responses
 Assume the following specifications: 40
f c  3 kHz kd 
 z2   p1  k  G    510 µs
i 0 p1
30
G  f  100
2
 p z1 2
G fc  20 dB  f 
2
 f 
2
(dB) (°)
1  c  1  c  20 0
f z1  200 Hz  fp   fp  ki  G0z1  2.51 ks 1
 1  2
G0  2
f z2  600 Hz  fz 
1  1 
2
 f 
1  c 
2
 z   z2 z1 
10 G f  -100
f p1  21 kHz  fz 
 fc   2 k p  G0  1
 z
   2.643

 2
 p1 
f p2  21 kHz 100 1k 3k 10k 100k

Integrator Saturation Issue
 The Integral term accumulates the error over time
vout  t  1. Assume a large output current step
2. Control asks for maximum duty ratio, e.g. 80%
3. Output cannot keep up, control voltage is flat
R1 C1
t
kp
vc  t       d kp
 t  vc  t  i
0
vc  t    t
i
   is constant
R2 Vref 1
G s   max
s
 po vc  t 
(V)
1
 po   t 
R1C1
58 4/1/2021 Public Information t
Event
SPICE Simulates the Saturation Effect
 An average model coupled to a PID simulates a voltage-mode buck converter
L1 R6
vout
vout  t 
75u 100m
a c Vout2
parameters 2 12
fc=10k d Resr
Gfc=-20
3
PWM switch VM p 70m Large Integral C6
0.1u
Vin
Vin=10
Vpeak=2
{Vin}
AC = 0
X7
PWMVM 16
I1 current voltage 10 9
I
L = 75u
Fs = 100k
Cout
220u step X5
R10
IntFilt AMPSIMP
{Ti/0.1u}
G=10^(-Gfc/20)
pi=3.14159
13
fz1=1.2k
fz2=1.2k
5
vint  t  DeriveFilt
XPWM R9 vout
GA IN
fp1=11.3k GAIN R100 {Td/0.1u}

fp2=50k K = 1/Vpeak
X6
10k
15
11
X9
28 D
POLE C7 -
AMPSIMP
Wz1=2*pi*fz1 4
FP = fp2 0.1uF K1
K = -1 R12
Wz2=2*pi*fz2 Verr
{10k/kpf}
K1 S UM2
Wp1=2*pi*fp1 SUM3 K2
K
K2
+
19 17 S +A 20
K
K3 Vref
S +A 18 8 X3
i=(1+fc^2/fp1^2)*(1+fc^2/fp2^2) B2 R8 SUM2 5
err2 X4
Voltage 10k K1 = 1
j =(1+fz1^2/fc^2)*(1+fc^2/fz2^2) SUM3 X1 K2 = -1
V(err2)>1.99 ? X8 K1 = 1 POLE
Wpi=sqrt(i/j )*G*fz1*2*pi 1.99 : AMPSIMP K2 = 1 FP = (N/Td)/(2*pi) R7
V(err2)<10m ? 10m : V(err2) VHIGH = 5 K3 = 1
10k
VLOW = -5
e=(Wp1-Wz1)*(Wp1-Wz2) Control voltage 25
f=Wp1*Wz1+Wp1*Wz2-Wz1*Wz2
vc  t 
Vdirect
Td=e/(f*Wp1)
N=((Wp1^2)/f)-1
Ti=((Wz1+Wz2)/(Wz1*Wz2))-(1/Wp1)
X10
AMPSIMP P
kpf=(Wpi/Wz1)-(Wpi/Wp1)+(Wpi/Wz2)

Windup Phenomenon in the Integral Path
 The small-signal response shows a 10-kHz bandwidth and a good phase margin
° dB
1.11 saturation
Tf 906m vint  t 
180 60.0
(V) 706m
506m
 m  71 306m Full recovery
90.0 30.0
6.00
T  f  4.00
vc  t 
0 0 (V) 2.00
0
f c  10 kHz -2.00 Control swings negative
-90.0 -30.0 6.00
regulation regulation
5.00
(V) 4.00
-180 -60.0 3.00
Overshoot vout  t 
2.00
10 100 1k 10k 100k 710u 1.11m 1.51m 1.91m 2.30m
 The transient response shows overshoot due to saturation and recovery time
Anti-Windup Methods
 Insert a clamping circuit at the PID output
2V verr  t  PWM 2.5 V

verr 2  t  2V d t 
0
100 mV – 5 V
Limiter Max dc = 80%
Normal operation: verr 2  t   verr  t   0

Clamp active: verr 2  t   verr  t   0
 Scale the difference to reduce the integral term input

P
+ verr 2
 I ki    t   Limiter verr
-
kb verr 2  verr D
The PID with Anti-Windup Back Calculation
 The injection in the integral path reduces its voltage excursion
P k p  t 
vc  t 
2V
+  t  +
vref  t    I ki    t   vout  t 
- - Plant
d t 
D kd 
dt + -
Compensator
kb
 Adjust the k coefficient to meet the overshoot specs
Transient Response with Back Calculation Anti-Windup
C6
 The overshoot is gone with almost no recovery time 10

0.1u
9
X5
AMPSIMP R10
6.50 vint {Ti/0.1u}
kb  0 6
5.50
SUM2
X12
SUM2
R9 K1 = 1 vout
K2
K1
(V) 4.50 {Td/0.1u} K2 = 1
3.50
kb  1 11
X9
28
INTIN
-
AMPSIMP
vout  t 
K1
K1 SUM2
5
2.50 SUM3 K2
17
K
S+A
C7 K2
+
8
K3
0.1uF
13 X3 Vref
R8 SUM2 5
X4
1.20 kb  0 SUM3
K1 = 1
X1
POLE
10k K1 = 1
K2 = -1
K2 = 1 FP = (N/Td)/(2*pi) R7
K3 = 1 10k
800m 25
(V) 400m
0 kb  1 X6
vint  t 
15
POLE R100
-400m limiter FP = fp2

K = -1
10k
R12
{10k/kpf}
D err2
300u 900u 1.50m 2.10m 2.70m K
2
S+A 18
B2
Voltage
X11 X8
SUM2 AMPSIMP
 Increasing k too much degrades response K1 = 1

K2 = -1
VHIGH = 5
VLOW = -5
K1
V(err2)>1.99 ?
1.99 : SUM2
14
GAIN INTIN
V(err2)<10m ? 10m : V(err2) K2
X13
GAIN
63 Public Information K=1
Transient Response with Integral Reset Anti-Windup SW
 A switch can reset the integral path when saturation occurs X7

PSW1
C6
0.1u
10 9
X5
R10
6.50 Without reset AMPSIMP
{Ti/0.1u}
6
5.50
SUM2
X12
(V) 1
2
R9
SUM2
K1 = 1 vout
K2
K1
4.50 {Td/0.1u} K2 = 1
11 28
3.50 With reset X9

AMPSIMP
INTIN
K1
-
2.50 vout  t  SUM3

K1
K2
17
K
S+A
C7
0.1uF
5
SUM2
K2
+
8
K3 Vref
13 X3
R8 SUM2 5
X4
SUM3 10k K1 = 1
X1
1.00 Without reset K1 = 1
K2 = 1
K3 = 1
POLE
FP = (N/Td)/(2*pi) R7
10k
K2 = -1
(V) 700m
25
400m
X6 15
100m With reset limiter

POLE
FP = fp2
R100
10k
vint  t  K = -1 R12
{10k/kpf}
D err2 2
-200m S+A
K
18
B2
Voltage
300u 900u 1.50m 2.10m 2.70m X11
SUM2
X8
AMPSIMP
K1 = 1 VHIGH = 5
K2 = -1 VLOW = -5
K1
V(err2)>1.99 ?
1.99 : SUM2
12 + SW
V(err2)<10m ? 10m : V(err2) K2
14 -
Sat
V7
64 Public Information 1 X14
COMPARHYS
Turn off the Pole at the Origin in a Type 2
 The added current source can be turned on or off, removing the pole at the origin
C2 Z f s
1 1 1
Z f  s   R2 ||  R2 p 
sC2 s R2C2
1
p
 
R2  
R1
G  s     g m k1  s 
R2
 G0
1  G0  R2
Vin  s   s s  R1
Vout  s   1  1  
 p p 
R2 
1 g m k1  s  1 z
g m k1  s  1 G0 s
k1  s   G0  G0
s s s
z 1 1
 po   po p p
g m R1
New equation Type 2
 The technique can be extended to a type 3 compensator
Example with the Pole at the Origin Toggling Circuit
 The added current source can be turned on or off, removing the pole at the origin
R3 C4
{R1} {C2}
parameters R7 R5
10k {R2}
R4
R1=38k
fc=1k
{R1} C3
{Ck}
2
Pole/zero
Vout1
Gfc=-20
pfc=-90
gm 5
on demand
pm=70
4 8
E2
G=10^(-Gfc/20) 100k
boost=pm-(pfc)-90
pi=3.14159
E3 G2 C2
K=tan((boost/2+45)*pi/180) {C2}
100k {gm}
fz=fc/k integrator C1 R2
fp=k*fc {C1} {R2}
gm=10u
R1
{R1}
6
Classical type 2
R2=G*R1 vin Vout2
C2=1/(2*pi*fp*R2) 1
3
compensator
C1=1/(2*pi*R2*fz) Vin
R6
Ck=gm/(2*pi*fz) AC = 1
10k
E1
Approximate formulas 100k
for C2 << C1
On-Demand Pole at the Origin and Zero
 By setting the transconductance gm to zero, the pole and zero are removed
(dB °) (dB °)
40.0 170
G f  40.0 170
gm  0 G f 
30.0 150 30.0 150

Type 2
20.0 130 20.0 130

gm  0
10.0 110 10.0 110

Type 2
0 90.0 0 90.0
G  f  g m  10 µS G  f 
10 100 1k 10k 100k 10 100 1k 10k 100k
 Turn the current source off to improve overshoot and recovery time
Application in a High-Current Dc-Dc Converter
 The output current changes from 61 A to 1 A in 1 µs: undershoot and ringing are gone
Pole at the origin is active Pole at the origin is disconnected and back in place
vout  t  vout  t 
4 µs
verr  t  verr  t 
NCP81111
Agenda
 Building a PID
 Conclusion
Public Information
Continuous-Time and Discrete-Time Signals
 The switch samples the input signal at a Ts switching interval
u t  u  t  v  n  1 Sample
v  n
Sampler
v  n  1
0 t 0 t
Time-continuous waveform
Ts  n  1 Ts nTs  n  1 Ts no u  n  1.5
Sampling period
Sampling clock Ts
 n  1 Ts  n  2  Ts
 There is no discontinuity in u along the time axis  There is no in-between sample value
 u can take on any value: u  t   U  sin t   n is an integer: u   n   U  sin  nTsw 
B. White, Digital Control for Power Supply Engineers, APEC 2013, Professional Seminars
Fourier, Laplace and z-Transforms
 Laplace and Fourier transforms map continuous-time functions
U   Two-dimensions U   
Fourier transform


frequency domain
Time-continuous  j t  
U     u t   e dt
 
bilateral U   U  
Laplace transform

s-domain
u t  U  s    u  t   e dt st
0
s    j  
unilateral
j j
Third dimension
 The z-transform maps discrete-time functions Im  z 
z-domain

U  z    u  n z n
U  z   1  z 1  z 2  z 3  ... Re  z 
n0 1
u  t  Discrete sample value

Laplace and z-Transforms
 Convert a continuous-time signal into a complex frequency function
Time-continuous
Time-continuous
Im  s 
U  s    u  t  1
Re  s  H s  y  t   1  H  s  U  s 
Laplace 1  s 1
transform s    j Inverse-Laplace transform
Analog compensator y t 
u t 
Map to the s-plane
 Convert a discretized signal into a complex frequency function
Discrete-time
Im  z 
Discrete-time
U ( z )  {u[ n]} y  n   1
H  z U  z 
Re  z  H  z   1  z 1
Inverse-Z
Z-transform
z  e sTs transform
Digital compensator y  t 

u t 
Map to the z-plane
Step-Response via Laplace- and z-Transforms Approaches t
  
y  t   u  t  1  e  
 The stimulus is continuous in time 1
 
0.8
step
1 0.6 63%
1 1 1 1 
H s  y  t   1  
0 1  s 1 0.4
s  s 1  s 
u t 
s-domain Transfer function in s 0.2
Time-continuous
Continuous time
y t 
0 
0 10 ms
n n
 The stimulus is a discretized signal y  n   a 1  a  u  n   1  a  u  n   u  n 
1
0.8
step 0.6
1
z a  z a 
y  20
H z  y  n    1  1 
0.4
0 z 1 1  1  a  z 1  z  11  1  a  z 
u  t  0.2
Discrete time
z-domain Transfer function in z y n
Discrete time
0
0 20 40 60 80 100
From Laplace to z-Transform
 Take the Laplace transform of the sampled signal

U 
 s    u  t    u  nTs  e snT
 s
 u  0  u Ts  e  sTs  u  2Ts  e 2 sTs  ...  u  nTs  e  nsTs
n 0
 We can introduce the variable z and remove the exponentiel terms
This is
z  e sTs or z 1  e  sTs a delay

U  z    u  nTs  z  n  u  0  u Ts  z 1  u  2Ts  z 2  ...  u  nTs  z  n
n0
 This is the definition of the unilateral z-transform
Z-Transform of a Time Delay
 The signal y*(t) is u*(t) shifted by 3 clock cycles y  n  u  n  k  k 3
u  n  3 u t  
n2 n3
Y  z    u  n  k    u  n  k   z  n
n 0
n  3 n  2 n 1 n n 1
3 clock cycles Y  z   U  z   z k
y  n y  t 
n  3 n  2 n 1 u n z 3 u  n  3
n n 1 n  2 n  3
u n z 1 z 1 z 1 u  n  3
B. White, Digital Control for Power Supply Engineers, APEC 2013, Professional Seminars
Assemble Blocks to Form a Transfer Function
 Assume the following difference equation
y[n]  u  n   u  n  1 y n  k   Y  z   z k
 Involve the delay operator to z-transform the expression y[n]  Y  z 
1 y n  k   Y  z   z k
Y  z  U  z U z z
Y z
 Factor and rearrange:  1  z 1
U z
 Describe this equation in a flow-graph style
u n y  n
This is a simple 1 G0  2
z 1
+ low-pass filter
G  s   G0
s p 
2
1 Ts
u  n  1 p
How do you Practically Build a Delay?
 A register delays the information by one clock cycle
One cycle
data D Q output Two cycles

data D Q D Q D Q Three cycles
clock clk
clock clk clk clk
/ No delay
vclk  t  8
data / One cycle

8
8
delay /
vD  t  Two cycles
output
8
delay 1 1 1
data z z z / Three cycles
vQ  t 
Modeling the Delay with SPICE
 A delay line in SPICE can be used to model a delay
u t  UTD
y t  G  s   e  sTs  Ae j 
Ts = 1 µs G    1
(dB) G    Ts (°)

(° dB)
10 180
180 4.00
5
90.0 2.00
G f  G f 
0 0 0 0
-90.0 -2.00 G  f  -5 G  f 
-180 -4.00 -10  180

1k 100k 10Meg
1k 2k 5k 100k 10Meg
S. Ben-Yaakov, D. Adar, Generic Average Modeling and Simulation of Discrete Controllers, APEC 2001
Approximating the Delay (1)
 A delay can be defined by associating a RHP zero and a LHP pole (all-pass filter)
s Gd    1 Pole/zero cancel
1 each other

Gd  s    
s 1    1   
1 Gd   tan 1    tan    2 tan  
         
 What value of  satisfies this equality?
 1  s    
arg e  sTs
  arg   Ts  2 tan 1  
 1  s     
0
 Use the arctangent Taylor series equivalent: 3 5
     
     
x3 x5    
1
tan x  x    ...   2          
3 5  3 5
  
Public Information  
79
 
Approximating the Delay (2)
 Solving for  gives us the 1st-order Padé approximant of the exponential
Ts 1  s
1 s
 sT 2  z 2
e s
with z   p 
Ts s Ts
1 s 1
2 p
180
The phase deviation

0
error between both
Ts
1 s formulas increases
(dB) 0
 2 0 (°)
1 s
Ts with frequency
2
e  sTs
-180
1k 10k 100k
Public Information
Agenda
 Building a PID
 Conclusion
Public Information
Discretizing Continuous-Time Transfer Functions
 How to move from one domain (continuous time) to the other one (discrete time)?
Rf Cf
00101
10011
10111
Ri
Vout  s  ?
Verr  s 
Vref s  f  z
Digital control
Rf 1
sVerr  s    s Vout  s   Vout  s  b0Verr  z   b1 z 1Verr  z   a0Vout  z   a1 z 1Verr  z 
Ri Ri C f
z  e sTs
Vout  s     Verr  s  Vout  a0  a1 z 1 Verr 

G  s   G0 1  z  G  z 
 s  b0  b1 z 1

Mapping s in the z-domain
 Assume a simple integrator transfer function
X s 1 t
Y s  y t   x  t  dt
s  0
 In continuous-time domain, the integral is the area under the curve
x t  t1
A   x  t  dt
0
A
t
0 t1
 How do we evaluate the integral in the discrete-time domain?
x t  x  n
Sum discrete areas
x  n  1
A  A1  A2  A3
A1 A2 A3
t
0
 n  2  Ts  n  1 Ts nTs
Three Integration Options x  nTs 
 The area can be approximated in three ways x   n  1 Ts   x  nTs 
Forward x   n  1 Ts  2
x   n  1 Ts 
x  nTs  A2  x   n  1 Ts  Ts
Ai A2  x  n  1Ts Ai
x   n  1 Ts 
0
Ai ?  n  1 Ts nTs  n  1 Ts nTs
x  nTs  Backward x   n  1 Ts   x  nTs 
A2  Ts
2
0 A2  x  nTs  Ts
 n  1 Ts nTs x  n  1  x  n 
Ai A2  x  n Ts A2  Ts
2
 Trapezoidal integration
 Flat-top approximation
84
 n  1 Ts nTs
Discretizing in Different Ways
 Forward-Euler method
Y  z   Y  z  z 1  X  z  z 1Ts
y  n  y  n  1  x  n  1Ts Y z z 1Ts z Ts Y s 1
   
X  z  1  z 1 z z  1 X s s
Integral value Integral value at Approximate integral
z 1
at sample n previous sample between samples H  z   H  s  s  z 1 s
Ts Ts
 Backward-Euler method
Y  z   Y  z  z 1  X  z  Ts
y  n   y  n  1  x  n Ts Y z Ts z zTs Y s 1
   
X z 1  z 1 z z  1 X s s
Integral value Integral value at Approximate integral
z 1
at sample n previous sample between samples H  z   H  s  s  z 1 s
zTs zTs
Tustin and Backward/Forward Euler Methods
 Tustin method
2Y  z   2Y  z  z 1  X  z  z 1Ts  X  z  Ts
x  n  1  x  n
y  n   y  n  1  Ts 2Y  z  1  z 1   X  z  Ts 1  z 1 
2
Ts 1  z  z
1
Y s 1
Integral value Integral value at Approximate integral Y  z  X  z  
at sample n previous sample between samples 2 1  z 1  z X s s
2 z 1
H  z   H  s  s  2 z 1 s 
 Other approaches Ts z 1 Ts z  1
Discretization Method What is Important?
Zero-Order Hold (ZOH) Transient response with a staircase input
 Tustin brings a good
First-Order Hold (FOH) Transient response with a PWL input
matching between frequency-
Transient response with impulse train
Impulse-Invariant Mapping
input and discrete-domains
Good matching between frequency- and
Pole-Zero Matching
discrete domains
https://www.mathworks.com/help/control/ug/continuous-discrete-conversion-methods.html
Viewing Tustin, Backward- and Direct-Euler Differently
 It is possible to approximate the exponential using Taylor’s series expansion
1
e x  1  x  x 2  ...  x n 1 ex  1  x
 n  1!
Neglecting
high-order terms
 Forward-Euler  Tustin transform

z 1 Ts
ze sTs
 1  sTs s 1 s
Ts 2 z 1 z 2
s
Ts z  1 T
1 s s
2
 Backward-Euler
1 1 z 1 Use the Padé
z  s approximant of
e  sTs 1  sTs zTs
the delay operator
Public Information Henri Padé

87 1863-1953
Applying Tustin’s Method to a Low-Pass Filter
 We have the following low-pass filter transfer function
1
Ga  s   2 z 1 1  z 1 2 Ts p  2
s s Gd  z   G0 G0  1  b1 
1 Ts z  1 1  b1 z 1 Ts p  2 Ts p  2
p
 Test the response in the frequency domain with z 1  e sT s
10 10
Fs  40 kHz Fs  40 kHz
f p  fd
f p  1 kHz f p  10 kHz f p  10 kHz
0 0
-3 dB -3 dB
(dB) (dB)
Ga  f 
-10 -10
1  e  j
f d  8.5 kHz
G2    G0
1  b1e j
fd is the discrete pole
Gd  f 
10 100 1k 10k 100k 10 100 1k 10k 100k
f p  fd f p  fd
 0.2%  15.2%
88 fp Public Information fp
From the s-Plane to the z-Plane – Exact Mapping
 The stable pole region – left half-plane – entirely maps inside the unit circle
j z z Im  z 
s magnitude angle
 unstable
j
Ts region
z  e
  j Ts
j

0  e Ts e jTs  k 2
2Ts    0
Ts Ts 1
 stable Re  z 
-10 -1  k 2 region k 2
j   
2Ts s0 z 1 Ts Ts Ts
 0 z 1
Left half-plane Right half-plane    z 0
z 1
Ts  1 s  0
Imaginary
The imaginary part defines the angle
The real part defines the radius
 The stable poles (LHPP) in s are confined inside the unit circle
 Unstable poles (RHPP) in s lie outside the unit circle
Watch Brian Douglas’s videos at https://www.youtube.com/channel/UCq0imsn84ShAe9PBOFnoIrg
The Bilinear Transform
 Tustin’s approximation maps the entire j axis on the unit circle in the z-plane
j Im  z 
T
 1 s s unstable
j 2
Ts z  region
Ts d 
 Ts
j
2Ts
0 1 s
2 a   stable  0
region
 1
Re  z 
-10 -1
s0 z 1 a  

 0 z 1 d  
Ts
Left half-plane Right half-plane    z 1
z 1
Ts  1 s  0
Imaginary
 The stable poles (LHPP) in s are confined inside the unit circle
 Unstable poles (RHPP) in s lie outside the unit circle
Stable in the Stable in the
s-domain z-domain
a  analog world
d  discrete world
Compression in the Frequency Response
 Map the s-domain transfer function in the z-domain with bilinear transform
1 Map to z 1
Ha  s  Ha  z  
1  s  2 z 1 
 In general:
1   Euler
j T  T s z  1 
ze d s
 2 z 1   2 e j T  1  d s
 2  d Ts 
d   a  T e j T  1  a  T  2
j T
Hd  z   Ha   H e 
d s
H  H j tan 
T
 s z  1   s 
d s
 s 
discrete analogue
 The discretized pole d is not at its original frequency-domain position

1 What you have What is expected
analogue
2  T  0.9 1
d  tan 1  a s  fd fd  tan 1  f a Ts 
Ts  2  0.8  Ts
fa
discrete
0.7 f a  1 kHz f d  0.997 kHz
discretized
0.6
0 0.1 0.2 0.3 0.4 0.5 f a  10 kHz f d  8.5 kHz
Frequency warping f a Fs
Fs  40 kHz
Magnitude and Phase Distortion of Mappings
Fs
 Tustin offers the best magnitude tracking while phase diverges as you approach
2
4
reference e sTs
Exact
z 3 Forward Euler z1( s )  e
s T s
z 150
180
z 1( 2   i F)
Ts
 arg z 1( i 2  F)    
  8 Tustin
Tustin 1  s  2 180
z 2( 2   i F) reference
z2( s )   arg z 2( i 2  F)     100
z 3( 2   i F)
2 Ts
180
  8
1  s  arg z 3( i 2  F)    
z 4( 2   i F) 2
e sTs Forward
z3( s )  1  s  Ts
 arg z 4( i 2  F)   
180
Tustin 
50
1
Backward
1
Backward and
z4( s ) 
1  s  Ts forward Euler
0
Backward Euler 0
0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5
F 0.13 F
Fs Fs
 Tustin provides same phase-mapping accuracy with lower sampling frequency

 Acceptable results for Fs  5  f while Euler would require Fs  7  f
M. Jovanović, Introduction to Digital Control of Switch-Mode Converters, in-house course, PHX 2014
Account for Warping when Positioning Poles and Zeroes
 Frequency compression occurs because of nonlinear relationship linking d and a
j Im  z    a  
2  T  Entire continuous
d  tan 1  a  d frequency range
a x T  2 
d   Ts x d  0
 Re  z 
 T 
a     Discrete-time
2
a  tan  d 
a  0
  d   frequency
T  2  Ts Ts
 You can anticipate the compression and prewarp your poles and zeroes
Where you should 10 10 90
set the pole What you want 1  e  j
G2    G0
in discrete time 1  b1e  j
2  T  50 45
p'  tan  d s  5 5
Ts  2  f p  10 kHz
(dB) G2  f  (°)
0 0 0 0
-3 dB
2  2 10k  Ts  1
f p'  tan   2  12.7 kHz -5 5
G2  f  -45
25u  2  
-10 10 -90
2 Ts p '  2 10 100
G0  1  b1  1 kHz 10 kHz 100 kHz
Ts p '  2 Ts p '  2 Public Information
Agenda
 Building a PID
 Conclusion
Public Information
From the z-Equation to Practical Implementation
 Start with the z-transform expression
Y  z 1  z 1 Expand
 G0 Y  z   Y  z  b1 z 1  U  z  G0  U  z  G0 z 1
U  z 1  b1 z 1
Rearrange
Y  z   U  z  G0  U  z  G0 z 1  Y  z  b1 z 1
y  n   u  n  G0  u  n  1 G0  y  n  1 b1
 Assemble blocks according to the obtained equation
2
1
How can we
u n G0 z 1
+ y  n test this
3
configuration?
z 1 b1
SPICE Simulates Delays Efficiently
 We know z 1 can be modeled by a delay line in SIMextrix®
Model from Basil AlMukhtar – ON Semiconductor
 By reproducing the flow-graph blocks, you can build the transfer function
ba_z^-1
U1
*
u  n u  n  1 G0
y n .param Fs=40k
.param Ts={1/Fs}
.param fp=10k
u  n  G0 .param wp={2*pi*fp}
.param wpw={(2/Ts)*tan(wp*Ts/2)}
y  n  1 b1 y  n  1 .param G0={1-2/(Ts*wpw+2)}
.param b1={(Ts*wpw-2)/(Ts*wpw+2)}
*
Continuous-time
Laplace TF
Plot the Continuous-Time Response
 Pole frequency pre-warping offers an excellent matching
-3 dB
Phase / degrees
Gain / dB
-45°
Tustin
Tustin
G f  G  f 
 Phase extends beyond -90° with discrete-time implementation

 Flowchart structure and behavior is validated before coding begins
SIMPLIS® Include 1st -Order Filters Blocks
 A pole-zero equation implemented in SIMPLIS® is the following:
N 0 1
1 z Expand
N1  N 0 z 1 N1
T z   N1 Y  z   Y  z  D0 z 1  U  z  N1  U  z  N 0 z 1
1  D0 z 1 1  D0 z 1
Rearrange
Y  z   U  z  N1  U  z  N 0 z 1  Y  z  D0 z 1
y  n   u  n  N1  u  n  1 N 0  y  n  1 D0
 Assemble blocks according to the obtained equation

N1
u n N0 z 1 + y  n
? ?
z 1  D0
Determining Coefficients Values
 Replace z by its value in the s-domain: z 1  e sT s
Tsw
1 s
e  sTsw 2
N1  N 0 z 1 N1  N 0 e  sTsw 
H z  H s  T
1  D0 z 1 1  D0 e  sTsw 1  s sw
2
If we plug this expression into the original transfer function, we have:
Tsw 2 N1  N 0 H 0 p Ts z  2 
1 s z  N0 
2 Tsw N1  N 0
N1  N 0
T s  z Ts p  2 
1  s sw 1
2 rearrange z 2 1  D0 H 0 p Ts z  2 
H s  H  s  H0 p  N1 
T s
1  s sw 1 Tsw 1  D0  z Ts p  2 
1  D0 2 p
T N  N1 4
1  s sw H0  0 D0  1 
2 1  D0 Ts  p  2
The Implementation Requires a Clock Generator
 This is a transient simulation with a POP block
 Automate the z-coefficients calculations

*
*
.GLOBALVAR fp=10k * pole position *
.GLOBALVAR N10=H00*wp*(Tsw*wz+2)/(wz*(Tsw*wp+2))
.GLOBALVAR fz=1k * zero position *
.GLOBALVAR N00=H00*wp*(Tsw*wz-2)/(wz*(Tsw*wp+2))
.GLOBALVAR H0=0 * dc gain in dB *
.GLOBALVAR D00=1-4/(Tsw*wp+2)
.GLOBALVAR Fsw=100k * sampling frequency *
* .GLOBALVAR N00=0
*
* if no zero, set N00= 0 and update N10 with below line *
.GLOBALVAR Tsw=1/Fsw
* .GLOBALVAR N10=H00*(wp*Tsw)/(1+(Tsw*wp/2))
.GLOBALVAR H00=10^(H0/20)
*
.GLOBALVAR wp=2*pi*fp
.GLOBALVAR wz=2*pi*fz
Sampling Frequency Affects the Filter Response
 The phase drops faster at a 100-kHz sampling frequency
20 20
G f  G f 
(dB) 10 (dB) 10
1
s
z
G  s   H0
s
f z  1 kHz 1
Fs  100 kHz 0 p
0 f p  10 kHz
90 90
G  f  Fs  1 MHz G  f 
45 45
(°) 0 (°) 0
-45 -45
-90 -90
10 100 1k 10k 100k 1M 10 100 1k 10k 100k 1M
Agenda
 Building a PID
 Conclusion
Public Information
Implementing a Type 2 Compensator
 You start from the low-entropy form Laplace transfer function Divide by z2
z
1 G0Ts p 1  z  2 z  Ts z  Ts  z z  2 
s 2 1  z 1
G  s   G0 s G z 
1
s Ts 1  z 1 4 z 2  8 z  2Ts p  2Ts p z 2  4
p Divide by z2
Biquad
filter
a0  a1 z 1  a2 z 2 Specifications:
G z 
1  b1 z 1  b2 z 2 a0  0.0857
G fc  20 dB f c  1 kHz
G0Ts p Ts z  2  a1  1.957 10 4
a0  Boost  50
2  2  Ts p 
 boost   a2  -0.0855
k  tan     2.74
 2 4
G0Ts 2 p z 2 b1  -1.9829
a1  b1   f z  364 Hz
2  Ts p 1  0.5Ts p
f p  2.74 kHz b2  0.9829
G0Ts p Ts z  2  2
a2  b2  1 Fs  1 MHz
2Ts  p  4 1  0.5Ts  p
The Biquad Filter is a Recursive Digital Structure
 A part of the output is fed back to form the transfer function
Feedforward
coefficients
Feedback
coefficients
Reduce to 1st
order filter
y[n]  a0 x[n]  a1 x[n  1]  a2 x[n  2]  b1 y[n  1]  b2 y[n  2]
a2  b2  0
x n + y  n
a0 a0
z 1 z 1 x n + y n
a1
+
1 b1 1
z 1
z z
a2 a1 b1
+ +
2nd-order b2
 This configuration is called Direct Implementation I
 Can be transposed in different forms (DI II) so as to minimize calculation errors
Where are Poles and Zeroes?
 If you read coefficients from the code or a circuit: where are poles and zeroes?
Tsw
1 s
e  sTsw  2 Double zero
T G0   a0  a2
2
1  s sw T  a0  a1  a2
2 1  sTs  s2  s  a a a
a0  a1 z 1  a2 z 2 a a a a0  a1  a2 2  0 1 2
Gz  G s  0 1 2 2
1  b1 z 1  b2 z 2 b1  b2  1 1  b2 T  b2  b1  1
1  sTs  s2  s  b  b 1
b1  b2  1 2  1 2
 Numerator and denominator follow the form: Double pole
2
T 2
N  s   a0  a1  a2  sTsw  a0  a2   s  s   a0  a1  a2 
2
2
N  s   1  s  z1   z 2   s  z1 z 2 1  s  z1   z 2   s 2 z1 z 2
T 
2 G  s   G0
D  s   1  b1  b2  sTs 1  b2   s 2  s   b2  b1  1 1  s  p1   p 2   s 2 p1 p 2
2 Normalized low-entropy form
D  s   1  s  p1   p 2   s  p1 p 2
2

C. Basso, Linear Circuit Transfer Function: an Introduction to Fast Analytical Techniques, Wiley, IEEE-Press, 2016
Link Time Constants and Coefficients
 Solve a system of equations to unveil the poles and zeroes
2
a0  a2 a a a T 
 z1   z 2  Ts  z1   z 2  0 1 2 s
a0  a1  a2 a0  a1  a2  2  a0  a1  a2
G0 
1  b2 b  b 1  T 
2 1  b1  b2
 p1   p 2  Ts  p1  p 2  2 1  s
1  b1  b2 1  b1  b2  2 
 The second zero does not have a physical meaning in a type 2 compensator
1 1
f z1  fz2 
2 

 T a  a  a 2  4a a
sw 0 2 1 0 2   Infinite value
2  
sw
 T a  a  a 2  4a a
2 0 1 0 2  
2  a0  a1  a2   2  a0  a1  a2  
  a a a  
G0  0 1 2
1 1  b11  b2  1 1
f p1  f p2 
T
2 
sw  
b112  4b2  b2  1
b11  b1  1 p 2  
sw
 T b  1  b 2  4b
2 11 2  
2 1  b11  b2   2 1  b11  b2  
   
Testing Coefficients with Mathcad®
 Back-calculate poles and zero values: s
1
f p1 G0 z
f z  363.9 Hz f p1  1µHz f p2  2.74 kHz H 0  4 108 or 172 dB  20.008 dB G  s   G0
f z1 s  s 
Pole at the origin Infinite OL gain Mid-band gain
1  
 p1   p2 
0 0
1. Reference
50
G f  2. Tustin
50 G f 
3. Back-calculated  20  20
20 dB
 40  40
(dB) 0 (°) (dB)0 (°)
 60  60
Tustin
 50
G  f   80  50
G  f   80
10 100 1 kHz 10 kHz 100 kHz 10 100 1 kHz 10 kHz 100 kHz
Fs  1 MHz Fs  100 kHz

Testing the Filter with SPICE
 Parameters calculation is automated with SIMetrix®
20 dB
G f  Fs  1 MHz
OUT
n1
n2
n3
Fs  100 kHz
OUT
n1
n2
G  f 
*
.PARAM Fsw=100k .PARAM a0={G0*Tsw*wp*(Tsw*wz+2)/(2*Tsw*wp+4)}
.PARAM Tsw={1/Fsw} .PARAM a1={G0*Tsw^2*wp*wz/(Tsw*wp+2)}
.PARAM Ts=Tsw .PARAM a2={G0*Tsw*wp*(Tsw*wz-2)/(2*Tsw*wp+4)}
.PARAM fz=364 .PARAM b1={-8/(4+2*Tsw*wp)}
.PARAM fp=2.74k .PARAM b2={(4/(Tsw*wp+2))-1}
* *
.PARAM wz={2*pi*fz} { '*' } a0 = {a0}
.PARAM wp={2*pi*fp} { '*' } a1 = {a1}
.PARAM G0=10 { '*' } a2 = {a2}
* { '*' } b1 = {b1}
{ '*' } b2 = {b2}
20 dB *
Implementing a Type 3 Compensator
 You can now repeat the exercise for a 2-zero/3-pole compensator
 2 1  z 1
  z1 s  s
1    1   Ts 1  z 1
 s   z2 
G  s   G0
 s  s 
1   1  
  p1   p2 
 
N  z   G0Ts p1  p2  z  1 2 z  Ts z1  Ts z1 z  2 2 z  Tsz2  Tsz2 z  2 
Divide by z 3
 
D  z   2 z2  z  1 2 z  Ts p1  Ts p1 z  2 2 z  Ts p2  Ts p2 z  2 
a0 
 
G0Ts p1  p2 Tsz1  2 Tsz2  2  a3 
 
G0Ts p1  p2 Ts z1  2 Tsz2  2 
1
a0  a1 z  a2 z  a3 z2 3 
2 4 z2  2Ts  p1 z2  2Ts p2 z2  Ts 2 p1  p2  z2  
2 4 z2  2Ts  p1 z2  2Ts p2 z2  Ts 2 p1  p2 z2 
G z   2
G0Ts  p1  p2 2Ts z1  2Ts  z2  3Ts  z1  z2  4  Ts  p2  2 4 16
1  b1 z 1  b2 z 2  b3 z 3 a1 

2 4 z2  2Ts p1  z2  2Ts p2  z2  Ts 2 p1  p2  z2
b1  
Ts p2  2 Ts p1  2
b2 
T 
s p1 
 2 Ts  p2  2 
a2  

G0Ts  p1  p2 2Tsz1  2Ts z2  3Ts 2z1 z2  4
b3  
T s p1 
 2 Ts p2  2 
 2
2 4z2  2Ts p1 z2  2Ts  p2 z2  Ts  p1  p2  z2  T s p1 
 2 Ts p2  2 

Watch for the Computation Flow
Feedforward Feedback
 The zeroes are computed first then the poles coefficients coefficients
y[n]  a0 x[n]  a1 x[n  1]  a2 x[n  2]  a3  n  3  b1 y[n  1]  b2 y[n  2]  b3 y[n  3]

a0
 Computing zeroes first reduces overflow issues
x n + y  n
 Poles tend to amplify variables when they occur
 Zeroes attenuate the variable when they occur
z 1 a1 z 1
b1
+ 10 10
x 1 G1=50 0 G2=0.1 0 y 1
1 1 50 10
z z Clamped to 10
Wrong result
a2 b2
+ 10 10
x 1 G1=0.1 0 G2=50 0 y5
z 1 z 1 0.1 0.1
a3 b3 Not clamped
Accurate result
+
Zeroes calculation Poles calculation
M. Jovanović, Introduction to Digital Control of Switch-Mode Converters, in-house course, PHX 2014
Going Digital – Converting Samples in Words
 The analog-to-digital converter converts the sample in a binary word
Sampler u  n   u  nTs 
ADC v ADC  t 
N bits
u t 
1 1 1 u t 
Ts
 The conversion process brings a quantization error Quantization

error
111 ADC out
110 Range: 0 to 1 V
101 011
100 Resolution:
011 V  125 mV
1 1 010
010   125 mV
3-bit ADC 001 23 8
000
8 quantization Ts
111 levels Public Information
SIMetrix® Lets you Test Coefficient Values
 Before implementation, it is interesting to test the filter in SPICE
continuous continuous
Zero-order hold (ZOH) *
ba_z^-1
* Enter Design Goals Information Here *
U7
*
discrete .PARAM fc=3k ; targeted crossover
*
V(N1)-V(N2)
ARB3
.PARAM Fsw=1Meg ; sampling frequency
.PARAM Tsw={1/Fsw} ; sampling period
.PARAM Ts=Tsw
OUT
*
*
n1
n2
n3
.PARAM fz1=200
.PARAM fz2=600
Discrete- to .PARAM fp1=21k
continuous- .PARAM fp2=21k
OUT
.PARAM Gfc=-20 ; magnitude at crossover
time *
n1
n2
n3
conversion .PARAM G1={10^(-Gfc/20)} ; needed gain at fc

.PARAM G0={G1/SQRT((1+(fz1/fc)^2)*
Reference TF +(1+(fc/fz2)^2)/(1+(fc/fp1)^2)*(1+(fc/fp2)^2))}
*
OUT
in Laplace .PARAM wz1={2*pi*fz1}

.PARAM wp1={2*pi*fp1}
n1
n2
.PARAM wz2={2*pi*fz2}
.PARAM wp2={2*pi*fp2}
Automated calculation

S. Ben-Yaakov, D. Adar, Generic Average Modeling and Simulation of Discrete Controllers, APEC 2001
SIMPLIS® Offers a Sampled PID Block
From type
 Add the third pole via a RC filter tuned at p2 3 conversion
Ki K d Ad  z  kd
G z  Kp   Kp  kp K i  kiTs Kd 
Ai  z   K d Ad  z   1 Ts Forward-Euler
1 1
  0   p1 ,z  1  kd  0 Backward-Euler
Internal z transfer functions  p1 kd Pole must lie
computed with forward- or
in unity circle
0.5 Tustin
backward-Euler or Tustin
f c  3 kHz G0  1.99
Ts  1 µs
G fc  20 dB K p  k p  2.643
K i  kiTs  2.509m
f z1  200 Hz
kd
f z2  600 Hz Kd   510
Ts
f p1  21 kHz 1 You can choose and
   15m
f p2  21 kHz  p1 k d combine mapping
Externally
functions
implemented

https://www.simplistechnologies.com/documentation/simplis/dp_discrete_filters/topics/piddiscretefilter.htm
The Sampled PID Parameters Can be Automated
 Because of infinite quasi-static gain, a biasing circuit is necessary
Second pole

You Can Also Derive PID Blocks Yourself
 We can map the continuous-time representation in the z-domain
P k p  t 
+  t  vc  t 
u t   I ki    t  
-
d t 
y t  D kd
dt
 The proportional block is straightforward
Vc  s  k  k 0
 k p  s  Vc  n   k p   n    n kp k p  n
i d

Backward-Euler Applied to the Integral Path
 Map the integral block defined in the s-domain
kp 1 1  z 1 kp 1
Vc  s  k   s s
Ts
Vc  z  k  1
  z
p  kd  0
i s Backward-Euler
p  kd  0
i 1 z
Ts
 Develop and rearrange this expression:
k pTs k pTs
Vc  z  1  z 1
   z Vc  n   Vc  n  1    n
i i
 Assemble the blocks to form the integral path:
kp +
  n Ts + Vc  n 
i +
z 1

Backward-Euler Applied to the Derivative Path
 Map the derivative block defined in the s-domain
1  z 1 1  z 1
Vc  s  k  ki  0
 k p d s  s  s Vc  z  k  k p d  z
p Ts p  ki  0
Ts
Backward-Euler
 Develop and rearrange this expression:

k p d k p d
Vc  z  
Ts
1  z    z 
1
Vc  n  
Ts
  n    n  1
 Assemble the blocks to form the derivative path:
k p d +
  n + Vc  n 
Ts -
1
z

A Filter is Needed for the Derivative Term
 Start from the low-pass filter expression
1 1  z 1 1 1
s
G s  Ts G s  
s d d  d  d 1 
1 N  z 1  1  1   1  z 
p p  NTs NTs  NTs    d  NTs 
d
 Factor and rearrange this expression considering 2 as the input and Vc the output
a Ts
G s  a Vc  n   a   2  n   Vc  n  1 1  a 
1  z 1 1  a  d
 Ts
N
 Insert the block in series with the derivative term
k p d +  2  n +
  n + a + Vc  n 
Ts - +
1
z z 1 1 a

Final Filtered PID Implementation
 The model can be enriched with an anti-windup structure
kp Proportional
kp + Integral
  n Ts + + Vc  n 
i -
z 1
k p d Derivative + + Filter
+ a +
Ts - +
1
z z 1 1 a
i 
z  z
1 2

1
N
p 2 1
1 kp 
 po  po  po
  d 
 p1   z1  p1   z2 
z z1 2
p 1
 p z   p z  z z
1 1 1 2 1 2
z  p z
1 1 2  p1 z   p z  z z
1 1 2 1 2
 p1

Use SIMetrix® to Validate the Structure
Second
 Simulation lets you verify the blocks are well assembled together pole
ba_z^-1
U3
V(N1)-V(N2)
ARB5
V(N1)+V(N2)+V(N3)
ARB2
Zero-order hold (ZOH)
Automated
V(N1)+V(N2)
ARB3
calculation
 d  193 µs
f c  3 kHz
N  26.43
G fc  20 dB
ba_z^-1
U5  i  1.053 ms
f z1  200 Hz
k p  2.64
  z1  s  V(N1)-V(N2)
f z2  600 Hz
V(N1)+V(N2)
ARB4
ARB1
1    1   Ai  2.51m
 s    z2  f p1  21 kHz
G  s   G0 Ap  k p
 s  s 
 1   1
  p  f p2  21 kHz
  p1  2  Ad  509.6
Reference TF type 3
Testing PID Responses
 The SIMPLIS® PID and SIMetrix® discrete version agree well with each others
Laplace type 3
(dB) SIMPLIS
SIMextrix
G f 
Fs  200 kHz
Laplace type 3
(°)
SIMPLIS
SIMextrix
G  f 

Practical Application to a Buck Converter
 The PID compensator is applied to a 250-kHz VM buck converter
IL
Vref  k  Vout
Loop delay
FB
Anti-aliasing
filter

Automate All Calculations with SIMPLIS®
 You can quickly check different compensation strategies and see what it does
*
.var Gfc=-8 * magnitude at crossover * .globalvar Wtau=2/Ts ; transport delay
.var PS=-148 * phase lag at crossover * .globalvar R={1/(10n*Wtau)}
* *
* Enter Design Goals Information Here * .var Req={Rupper*Rlower/(Rupper+Rlower)} ; equivalent resistance driving Ca
* .var Ca={1/(pi*Fs*Req)} ; anti-aliasing filter
.var fc=10k * targetted crossover * * Position the anti-aliasing filter
.var PM=60 * choose phase margin at crossover * .var wz1={2*pi*fz1}
* .var wz2={2*pi*fz2}
* Enter the Values for Vout and Bridge Bias Current * .var wp1={2*pi*fp1}
* .var wp2={2*pi*fp2}
Display
.var Vout=5 * coefficients in
.var Ibias=250u .var a=sqrt(1+(fc/fp1)^2) the netlist
.var Vref1=2.5 .var b=sqrt(1+(fc/fp2)^2) *
.var Rlower=Vref1/Ibias .var c=sqrt(1+(fz1/fc)^2) { '*' }
.var Rupper=(Vout-Vref1)/Ibias .var d=sqrt(1+(fc/fz2)^2) { '*' }
.var kdiv=Rlower/(Rlower+Rupper) .var G0=(a*b/(c*d))*G { '*' } G0 = {G0}
* .var Cwp2={1/(wp2*1k)} { '*' } fp1 = {fp1}
* Capture the double zero position and one of the pole position * * { '*' } ki = {ki}
.var fz1=1.5k .var ki=G0*wz1 { '*' } kd = {kd}
.var fz2=1.5k .var kd=((wz2-wp1)*(ki-G0*wp1))/(wp1^2*wz2) { '*' } kp = {kp}
.var fp2=50k ; pole is brought by the Tustin mapping at Fsw/2 .var kp=G0*((wz1+wz2)/wz2-wz1/wp1) { '*' } kpz = {kpz}
* Account for the * { '*' } kiz = {kiz}
* Do not edit the below lines * divider network .var kpz=kp { '*' } kdz = {kdz}
.var boost=PM-PS-90
attenuation .var kdz=kd/Ts PID block { '*' } gam = {gam}
.var G=(10^(-Gfc/20))/kdiv .var kiz=ki*Ts coefficients { '*' }
.var fp1=fc/tan((2*atan(fc/fz1)-tan(fc/fp2))-boost*pi/180) * .var gam=1/(wp1*kd) { '*' }
.var Fs=250k * *
.var Ts={1/Fs}
* 123 4/1/2021 Public Information
Cycle-by-Cycle Simulation Shows a Stable Operation
 Simulation shows the staircase waveform on the PID output
vFB  t 
iL  t 
iout  t 
vPID  t 
vDRV  t 
vout  t 
vSW  t 

Open-Loop Gain is Obtained in a few Seconds
 Crossover and phase margin are good at a 250-kHz sampling frequency
f c  10 kHz
GM  11 dB
Tf
Fs  250 kHz
m  60
T  f 

Excellent Transient Response
 The recovery is good and fast with a 1-to-5-A transient step (1 µs)
vFB  t 
iL  t 
iout  t 
vPID  t 
vDRV  t 
vout  t 
vSW  t 

Agenda
 Building a PID
 Conclusion
Public Information
Conclusion
 To close the loop, you need a compensator
 The compensator is designed to stabilize and shape the response
 This filter can be implemented using an op-amp, a TL431, an OTA…
 An op-amp is not a perfect element and it impacts final results
 Digital control implements a discrete-time compensator
 You analyse the stability of the converter in the z-domain
 Simulation tools are useful to verify architectures before coding
 Going from the s-domain to the z-domain warps the response
 Pre-warping poles-zeroes position is the way to go
Merci !
Thank you!
Xiè-xie!

Chris Basso APEC Seminar 2021

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chris Basso APEC Seminar 2021

Uploaded by

Copyright:

Available Formats

Designing Compensators for the Control of

Switching Power Supplies

 A control system observes the output and minimizes errors

  s Verr  s  D s

1. Select poles/zeroes placement

Change in strategy requires

 Passive components suffer drawbacks:  Active components are not perfect!

 Speed  High bandwidth

Topology Voltage Mode Current Mode

Buck-boost 3  f 0  f c  0.3  f RHPZ f c  0.3  f RHPZ

Too high a crossover frequency affects susceptibility to noise!

5.20 vout  t  Vout 

11 Public Information rC  Z Cout @ f c

The compensator is tuned to

 A compensator typically includes a pole at the origin: G  0   

Type 1 Type 2 Type 3

Voltage mode clock VCC Current mode

 in CCM, 2nd-order response: type 3  in CCM, 3rd-order response: type 2

AOL  106 dB -120 -90

1 10 100 1k 10k 100k

27 4/1/2021 Public Information

28 4/1/2021 Public Information

° dB Loop gain transfer function (V) Step-load response

fc = 10 kHz f c  5.2 kHz, m  77

 Perform parametric sweeps with Monte Carlo analysis to check robustness

30 Press F11 Public Information

boost  60 boost  72

 Limit phase/gain distortion by selecting an op-amp still having gain at 20fc

33 4/1/2021 Public Information

dB ° Loop gain transfer function (V) Step-load response

TL431 37 1000 2.5 TO92, SO8, µ8,

NCP100 7 100 0.9 TO92, TSOP5

Exact 50% duty ratio

Coming from the

Exact 50% duty ratio

 One benefit of the CM operation

 Automated calculations account for the optocoupler pole contribution

 Each processing block serves a purpose:

(V)  (V) (V)

vout  t  vout  t  vout  t 

(V) (V) (V)

vout  t  vout  t  vout  t 

55 4/1/2021 Public Information

 Add a low-pass filter to force a flat gain at high frequency

57 4/1/2021 Public Information

fp1=11.3k GAIN R100 {Td/0.1u}

59 4/1/2021 Public Information

2V verr  t  PWM 2.5 V

Limiter Max dc = 80%

Normal operation: verr 2  t   verr  t   0

 Scale the difference to reduce the integral term input

 Adjust the k coefficient to meet the overshoot specs

 The overshoot is gone with almost no recovery time 10

6.50 vint {Ti/0.1u}

-400m limiter FP = fp2

 Increasing k too much degrades response K1 = 1

 A switch can reset the integral path when saturation occurs X7

3.50 With reset X9

2.50 vout  t  SUM3

100m With reset limiter

 The technique can be extended to a type 3 compensator

30.0 150 30.0 150

20.0 130 20.0 130

10.0 110 10.0 110

in Laplace .PARAM wz1={2pifz1}