Ultimate Switching: Toward a Deeper

Understanding of Switch Timing Control in Power Electronics and Drives

P. T. Krein, Director Grainger Center for Electric Machinery and Electromechanics Dept. of Electrical and Computer Engineering University of Illinois at Urbana-Champaign

Outline
Fundamentals: power electronics control at its basic level Motivation False starts and model-limited control Small-signal examples Ultimate formulation Geometric control examples

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 2

Fundamentals
In any power electronic circuit or system, control can be expressed in terms of the times at which switches operate. The fundamental challenge is to find switching times for each device. Example:
For each switch in a converter, find switching times that best address a set of constraints. This is an optimal control problem of a sort. Might represent this with a switching function q(t).
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 3

Fundamentals
The general problem is daunting, so we simplify and address switch timing indirectly.
Averaging (address duty ratio rather than q) PWM (use d as the actuation, not just the control) Sigma-delta (make one decision each period based only on present conditions) Other approaches

We are researching to try and identify ways to address the timing questions more directly.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 4

Motivation
We believe that a new and more fundamental consideration of a switch timing framework has strong potential benefits. Motivated by our work on switching audio
Showed that sine-triangle PWM, used as a basis for audio amplifiers, provides nearly unlimited fidelity.

Motivated by past work on geometric and nonlinear control

Performance can be achieved in power converters that is unreachable with averaging approaches.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 5

False Starts
Many argue that space-vector modulation (SVM) gets more directly at switch timing. In fact, SVM addresses duty ratios and yields (at best) exactly the same result as a PWM process. It is usually worse because uniform sampling is involved. Small-signal analysis methods are even less direct. Sliding-mode controls confine the switching without getting to the timing challenge.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 6

Space Vectors in Time Domain

Space vector modulation Third-harmonic injection sine-triangle PWM

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 7

Model-Limited Control
Many control methods used in todays switching power converters are limited by the models of the systems. Model-limited control is an important barrier to improvement of converters.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 8

Model-Limited Control
Any type of PWM implies switching that takes place much faster than system dynamics. Dc-dc converters use controllers designed based on averaging. We often learn that bandwidths are limited to a fraction of the switching rate. We finally have the tools to interpret this rigorously.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 9

Model-Limited Control
Distortion in the low-frequency band can be computed as a function of switching frequency ratio. Distortion must be at least -40 dB (better -60 dB) to justify control loop design. Based on natural sampling: Frequency ratio In-band distortion
5 7 9 11 13 15 -9 dB -42 dB -70 dB -110 dB -154 dB -201 dB

10-10

This is consistent with signal arguments that yield 2 as the minimum ratio and rules of thumb about a ratio of 10 for best results.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 10

Model-Limited Control
These models are convenient and useful, but do not use the full capability of a conversion circuit. We gave up a factor of 10 on dynamic performance in exchange for precision. Consider an example:
Small-signal methods and models are powerful tools for analysis and design. They can only go so far toward the analysis of large-signals circuits and disturbances.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 11

Small-Signal Response Examples

Take a dc-dc converter, with a well-designed feedback control. Explore its response. In this case, a known sinusoidal disturbance is applied at the line input. Its frequency is 5% of the switching rate. Its magnitude is 10%. The controller is adjusted to cancel line variation completely the duty ratio tracks and cancels the disturbance based on smallsignal analysis.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 12

Buck Converter
In this example, a feedforward compensation is used to eliminate changes caused by line variation. iIN #1 L IOUT + + VIN vOUT RLOAD V OUT #2 ( ( V o ltag e V ) , cu rren t A )

i ( t)

v ( t)

Grainger Center for Electric Machines and Electromechanics

tm e i University of Illinois at Urbana-Champaign

13

Example Dc-Dc Converter Problem

10% disturbance around 80% reference value. Frequency is 1/20 of switching (e.g. 5 kHz on 100 kHz).
1.2

trip ( j k) s3lev( j k m) ref ( j m)

- 1.1 0 j 2048

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 14

Compensated PWM Output

Filter time constant about 1/10 of switching.
0.5

Cu r r e n t

0.5

500

1000

1500

2000

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 15

Result?
Is the disturbance rejected or not?
Yes and no.

Does this controller achieve the requested bandwidth?

In fact, the controller is completely eliminating linear aspects of the disturbance. But the output ripple has features that may not be preferred.

Now, ignore small signal limits.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 16

Example Dc-Dc Converter Problem

10% disturbance around 80% reference value. Frequency is 3/4 of switching.
1.2

trip ( j k) s3lev( j k m) ref ( j m)

- 1.1 0 j 4096

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 17

Output Ripple
10

Cu r r e n t
5

s3iiii

10

500

1000

1500

2000 iii

2500

3000

3500

4000

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 18

Result?
In several ways, the result is the same, although filtering is less effective because of the higher frequency. There is an aliasing effect (but there was previously as well). The disturbance frequency does not appear in the output.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 19

Quick Performance Check

Hysteresis control instead, 150 kHz disturbance.
12

10

Line input
Voltage
8

10

20

30 Time (us)

40

50

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 20

Hysteresis Method
Now the ripple is tied only to the switching rate. The disturbance has no noticeable influence on the output. This is true even though the disturbance is faster than the switching frequency! Does this mean the converter has a bandwidth greater than its switching frequency?

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 21

Comments
Frequency response and bandwidth imply certain converter models. Physical limits are more fundamental:
When should the active switch operate to provide the best response? How soon can the next operation take place? How fast can the converter slew to make a change?

Hysteresis controls respond rapidly. This is an issue of timing flexibility more than of switching frequency.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 22

Consideration of Disturbance Timing

In a buck converter, any line disturbance while the active switch is on will have a direct and immediate effect at the output. iIN #1 L IOUT + + RLOAD VIN vOUT V OUT #2 No line disturbance will have any effect if it occurs while the active switch is off. This means an impulse response cannot be written without a switching function.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 23

Consideration of Disturbance Timing

This indicates that the nonlinearity cannot be removed for impulse response. Impulse is not adequate information to determine the response. Average models cannot capture timing issues. Notice that similar arguments apply to step responses and others.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 24

The Ultimate Formulation

A converter has some number of switches. For each switch, there are specific times at which a device should turn on or off. The times represent the control action. Selection of the times is the control principle. For each switch i, find a sequence of times ti,j that produce the desired operation of the converter.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 25

The Ultimate Formulation

A converter with ten switches.

Time sequences t1,j through t10,j.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 26

The Ultimate Formulation

This is too generic -- there must be constraints and objectives. Example: for a dc-dc converter with one active switch, find the sequence of times ti that yields an output voltage close to a desired reference value.

t1

t2 t3

t4 t5

t6

t7

t8

t9

t10 t11

t12 t13
27

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign

The Ultimate Formulation

Example: boost dc-dc converter.
L C R +

VIN

VOUT
-

Find the best time sequence to correct a step load change and maintain fixed output voltage.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 28

The Ultimate Formulation

Still too generic no unique solution. Also limited in utility. The proposed constraint deals with steady-state output and only one specific dynamic disturbance. There were no constraints on switching rates or other factors.

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 29

The Ultimate Formulation

More practical: Given an objective that takes into account power loss, output steady-state accuracy, dynamic accuracy, response times, and other desired factors, find a sequence of times that yield an optimum result.

That is, find a set of times tk that minimizes an objective function.

Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 30

The Ultimate Formulation

This is a general formulation in terms of a hybrid control problem. Unfortunately, with results framed this way there are very limited results about existence of solutions, uniqueness, stability, and other attributes. Still very general, but with a well-formed cost function it might even have a solution. There is a control opportunity every time a switch operates.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 31

Implications
For steady-state analysis, this must yield familiar results. A dc-dc converter with loss constraints must act at a specific switching frequency with readily calculated duty ratio. For dynamic situations, the implications are deeper.
Should a converter operate for a short time at higher frequency when disturbed? How do EMI considerations affect times? Are our models accurate and complete enough?
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 32

Geometric Control Examples

Dc-dc buck converter, 12 V to 5 V nominal. L = 200 i uH, C = 10 uF, 100L kHz Iswitching.
i n ou t

+ V in_

+ #1 #2 v ou t _ R
l ad o

+ v ou t _

v ou t ( t) V
in

V o ltag e

v ou t ti e m 0 T 2T
University of Illinois at Urbana-Champaign 33

0
Grainger Center for Electric Machines and Electromechanics

Fixed Duty Ratio

Steady state, fixed duty ratio. This shows the inductor current and ten times the normalized capacitor voltage. The best solution given fixed 100 kHz switching.
1.1 1.05

i L (t)

0.95

v out (t) expanded

0 5 10 15 20 25 30 35 40

0.9

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 34

Result in State Space

Same data plotted in state space.
1.1

1.05
Inductor current

Steady state
0.95

0.9 4.99

4.995

5 Capacitor voltage

5.005

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 35

Hysteresis Control
Alternative: simply switch based on whether the output is above or below 5 V. No frequency constraint.
1.1

Hysteresis control on output voltage.

1.05 1 0.95 0.9

20

40

60

80

100

120

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 36

Hysteresis Control
Same result, in state space. These controls need timing constraints to prevent chattering.
1.1

State space.
1.05

0.95

0.9 4.99

4.995

5 Y2

5.005

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 37

Response to Step Line Input

Line step from 12 V to 15 V at 42 us. Duty ratio adjusts instantly to the right values. (This would happen in open-loop SCM.) Transient in voltage occurs.
1.1

0.9

50

100

150

200
Time (us)

250

300

350

400

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 38

State Space
State space plot shows how much the behavior deviates.
1.1
State space

1.05

iL i

0.95

0.9 4.98

4.99

5.01 vc i

5.02

5.03

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 39

Same Step Different Control

This is a current hysteresis control, with the switch set to turn off at a defined peak and on at a defined valley. Same line step.
1.1 1.05 1 0.95 0.9

20

40
Tim (us) e

60

80

100

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 40

State Space
The step is cancelled perfectly essentially in zero time.
1.1
State space

1.05

iL i

0.95

0.9 4.99

4.995

5 vc i

5.005

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 41

Boost Converter A Harder Test

What about a boost converter step? Example converter: L = 200 uH, C = 20 uF, 5 V input, 12 V output, 100 kHz switching
IIN + L iOUT + iC ILOAD + R

VIN

vL

vin

VOUT
-

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 42

Steady State Behavior

2.5 2

i L (t) v out (t) expanded

1.5

0.5

5
2.5

10

15

20

25

30

35

40

2.5
2.45

1.5

I ndu cto r cur r en t

2.4

State space
2.35

0.5

5
2.3 11.85

10
11.9

15
11.95

20
12 12.05 Capac itor voltage

25
12.1

30
12.15

35

40

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 43

Step Change Behavior

Step input from 5 V to 6 V at 42 us. Very slow transient even though the duty ratio values are set to cancel the change.
3
Current

1
Voltage

200

400

600

800

1000

1200

1400

1600

1800

2000

3 Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 44

State Space
Suggests a faster transition is possible.
State space
2.4

I ndu cto r cur r en t

2.2

1.8

1.6 11.4

11.6

11.8

12

12.2

12.4

12.6

12.8

13

13.2

Capac itor voltage

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 45

Ad Hoc Control
Short-term overshoot can be used to dramatically speed the response.

2.5

1.5

0.5

100

200

300

400

500
Tim (us) e

600

700

800

900

1000

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 46

State Space
Rapid move toward final desired result.
2.4

State space

I ndu cto r cur r en t

2.2 2 1.8 1.6 1.4 1.2 11.8

12

12.2

12.4

12.6

12.8

13

13.2

Capac itor voltage

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 47

Augmented Boost
Now alter the boost to achieve timing targets. This control eliminates the transient.
2.5

i L (t)
2

1.5

v out (t) expanded

0.5

50

100

150

200

250

300

350

400

450

500

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 48

State Space
The response never goes outside ripple limits.
Start

2.4
Inductor current

2.3 2.2 2.1 2

End

1.9 11.85

11.9

11.95 12 12.05 Capacitor voltage

12.1

12.15

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 49

More General Result

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 50

More General Result

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 51

More General Result

Grainger Center for Electric Machines and Electromechanics

University of Illinois at Urbana-Champaign 52

Research Topics
Find examples of high-performance converter controls, based on a timing control perspective. Develop design methodologies for them. Formulate sample optimization problems that address timing control directly. Seek controls that address system-level factors. Seek simplifications that reduce costs with little (or no) sacrifice in performance.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 53

Conclusion
The ultimate in power electronics control is to find a sequence of switching times that optimizes a specific objective function. Some test cases show that performance far outside the accepted range can be obtained. Good ways to specify constraints, quantify the problem, and optimize are issues for research. Examples show existence of such solutions. The objective is to identify and develop control concepts and methods that use the full physical capability of power electronics.
Grainger Center for Electric Machines and Electromechanics University of Illinois at Urbana-Champaign 54