te-epc-lpc

08/05/2014
2
Thurel Yves,
CERN, for CAS Power Converters 2014 @ Baden CH
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
3
Goal and motivation
Thurel Yves, CERN, CAS Pwr Conv. 2014
My main motivation and goal for this presentation
You being able to:
Recognize when you actually NEED a 4-Quadrant converter.
Review load characteristics to OPTIMIZE solutions with.
Integrate a GLOBAL and SAFE technical solution, dealing with
Machine requirements like EMC, cycle types, installation requirements
Load PROTECTION system like energy extraction.
Get a general overview of technical solutions.
Strong focus on solutions in operation @ CERN.
Get a feeling about the known limitation.
Show you some CERN practical realizations and results.
Too heavy program !!
Needs some cut !
te-epc-lpc
08/05/2014
4 Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
5
Introduction
4-Quadrant Power Converter: when and for what?
A standard converter is often 1-Quadrant converter. Why?
Simply because we very often need power we immediately use.
Remember manufacturer communicate on power (Watts) standard
converters can transmit, rarely on energy (J).
Energy a 1-Quadrant converter has to manage itself is its losses.
A 4-Q converter is different from a standard one. Why?
A tiny 1-Quadrant converter can transmit high amount of energy
from energy source (mains) to the load (it just takes time).
A 4-Quadrant converter can modify the load energy level, increasing
or decreasing it. This means removing the load stored energy by:
Sending it back to network, storing it or
Dissipating it.


Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
6
4-Quadrant Converter & Load
Thurel Yves, CERN, CAS Pwr Conv. 2014
4-Quadrant Power Converter: Which load?
A 4-quadrant Power Converter can inject energy to the load
and can also received stored energy from the load.
2 types of loads - in electronic field - able to store energy:
Capacitor Ej
load
= 1/2 C V
load

Inductor Ej
load
= 1/2 L I
load

(accelerator natural preference one)
2 modes of operation from Converter side:
Generator Mode (energy flows from converter to load)
V
converter
I
converter
= Positive quantity.
Receptor Mode (energy flows from load to converter)
V
converter
I
converter
= Negative quantity.

Can a 1-quadrant-only
converter work with a
load storing energy ?
HumDid you
charge your
smartphone
recently?
te-epc-lpc
08/05/2014
7
Review of n Quadrant Converter operation
1-Quadrant Converter vs Inductive load (R-L)



1.Q Converter delivers Energy to load, but cannot absorb it.
Current control (its dI/dt) while load energy is increased, is
possible, only limited by V
converter max
and level of current.
Maximum voltage delivered by converter takes into account:
LdI/dt boost voltage directly applied to inductance.
RI
max
cable losses part, reducing the boost capacity.




Thurel Yves, CERN, CAS Pwr Conv. 2014
I got it ! Control of the current is easily
feasible when energizing the load !
0 20 40 60 80 100
I
conv
[A]
V
conv
[V]
I
conv
V
c
o
n
v

V
L

R

L
dI/dt = V
L
/ L
V
L
= V
conv
RI
conv

V
L.max
= [V
conv.max
; I
conv.min
]
V
L.min
= [V
conv.min
; I
conv.max
]
1
3 2
4
V
conv
I
conv
te-epc-lpc
08/05/2014
8
Review of n Quadrant Converter operation
1-Quadrant Converter vs Inductive load (R-L)



Removing energy from the load can be obtained choosing
who, among load or converter, will power cable resistance.
[dI/dt = 0; constant current] is obtained when V
conv
= R.I
conv
.
dI/dt to the load is given by 1 / L (V
conv
- R I
conv
).
Maximum rate for ramping down the current
is obtained when V
conv
= 0, and I
conv
is maximum.
As current indeed decreases, dI/dt constantly
decreases (exponential decay), with severe
constraints for overall current control loop.




Thurel Yves, CERN, CAS Pwr Conv. 2014
Yes, and when
current decreases
hummHumm
it depends?
0 20 40 60 80 100
I
conv
[A]
V
conv
[V]
I
conv
V
c
o
n
v

V
L

R

L
dI/dt = V
L
/ L
V
L
= V
conv
RI
conv

V
L.max
= [V
conv.max
; I
conv.min
]
V
L.min
= [V
conv.min
; I
conv.max
]
1
3 2
4
V
conv
I
conv
te-epc-lpc
08/05/2014
9
Review of n Quadrant Converter operation
1-Quadrant Converter vs Inductive load (R-L)



Lets just fix once for all the limitation to be cope with:
For converter being ultimately
controlled as a final current
source, major issue of 1-Q
converter is the limitation on
dI/dt when ramping down, with
severe limitation given simply by:
dI/dt
neg. critical
= - R / L x I
conv


So simple that ignoring it is not
allowed, and this would cost a lot !




Thurel Yves, CERN, CAS Pwr Conv. 2014
0 20 40 60 80 100
I
conv
[A]
V
conv
[V]
I
conv
V
c
o
n
v

V
L

R

L
dI/dt = V
L
/ L
V
L
= V
conv
RI
conv

V
L.max
= [V
conv.max
; I
conv.min
]
V
L.min
= [V
conv.min
; I
conv.max
]
1
3 2
4
V
conv
I
conv
Initial allowed negative
dI/dt = -1.2 A / sec

Allowed neg. dI/dt @
low current= 0 A / sec !!!

A 1-Q converter cannot follow this reference !!
I
conv
becomes too low to get negative voltage across inductor

te-epc-lpc
08/05/2014
10
Review of n Quadrant Converter operation
2-Quadrant Converter vs Inductive load (R-L)






Thurel Yves, CERN, CAS Pwr Conv. 2014
Ok. WAIT A MINUTE. Which 2
quadrants? 2 in Diagonal? 2 in
Vertical or horizontal?
Clarify before going
further !
te-epc-lpc
08/05/2014
11
Review of n Quadrant Converter operation
2-Quadrant Converter vs Inductive load (R-L)
System: 1-Quadrant converter + Polarity switch inverter example.








V and I can be both positive or negative, great! But, energy is
always and only delivered to the load. (V
conv
I
conv
> 0).
This type of system should be
assimilated to a 1-Q converter.







Thurel Yves, CERN, CAS Pwr Conv. 2014
This is a common trap for final
user ! Bad surprise in perspective !
1
3 2
4
V
conv
I
conv
1
3 2
4
V
conv
I
conv
1
3 2
4
V
load
I
load
1
3 2
4
V
load
I
load
0 20 40 60
I
conv

V
conv

I
load
[A]
V
load
[V]
0 20 40 60
I
conv

V
conv

I
load
[A]
V
load
[V]
te-epc-lpc
08/05/2014
12
Review of n Quadrant Converter operation
2-Quadrant Converter vs Inductive load (R-L)




V can be positive or negative, when current is only positive (typical
thyristor converter case, with energy sent back to mains).
Only? Yes, but energy can here be removed from the load in a
controlled way - as long as connection to network is present (thyristor).
Control of the current is then possible
when increasing or decreasing it,
current keeping same direction.







Thurel Yves, CERN, CAS Pwr Conv. 2014
Energy, matters?
Wellwhy not ?
Ok, what matters is
possible energy direction,
its flow ?
1
3 2
4
V
conv
I
conv
I
conv
[A]
V
conv
[V]
I
conv
V
c
o
n
v

V
L

R

L
dI/dt = V
L
/ L
V
L
= V
conv
RI
conv

V
L.max
= [V
conv.max
; I
conv.min
]
V
L.min
= [V
conv.min
; I
conv.max
]
0 20 40 60 80 100
te-epc-lpc
08/05/2014
13
Review of n Quadrant Converter operation
4-Quadrant Power Converter




Almost no topological limitation regarding any load.
Complex design, very different topologies with pro and cons.
Energy transfer possible at any point.
Choice of topologies mainly comes from:
Power level, Performances, Environment.
Machine use (Pulsed, or slow DC).
Load nature and time constant
(superconductive loads).





Thurel Yves, CERN, CAS Pwr Conv. 2014
1
3 2
4
V
conv
I
conv
I
conv
[A]
V
conv
[V]
I
conv
V
c
o
n
v

V
L

R

L
dI/dt = V
L
/ L
V
L
= V
conv
RI
conv

V
L.max
= [V
conv.max
; I
conv.min
]
V
L.min
= [V
conv.min
; I
conv.max
]
0 20 40 60 80 100
I confirm a 4-Q converter can
INDEED absorb the whole
energy contained in a Donut?
te-epc-lpc
08/05/2014
14
4-Quadrant Operation: Topologies

Goals:
Choosing adequate topology for removing defined amount of load
energy, Regulating magnet current (access to dI/dt).
Solutions:

Thurel Yves, CERN, CAS Pwr Conv. 2014
I>0
V>0
MAINS
Variable resistor (active), controlling the
dissipation level, i.e. the dI/dt.
Converter sending back energy to mains
controlling the Ej transfer.
Converter storing the magnet energy in a
capacitor, controlling the Ej transfer.
Alternative solutions which could be: rotative machine, or
superconductive inductornot common & not treated here.
Dont miss this step, even if all these
topologies tends to be each complex!
te-epc-lpc
08/05/2014
15
Machine Operation Impact on topology choice
Pulsed Machine





Slow Machine: LHC Type / Magnet or orbit Correctors

Thurel Yves, CERN, CAS Pwr Conv. 2014
[A]
[V]
Cycle Period
High Power
to be absorbed / Often
High Energy
12h or more run
Power to be
absorbed / sometimes on sharp correction
Low Energy
[A]
[V]
te-epc-lpc
08/05/2014
16
Real 4-Quadrant operation area
Load resistor impacts highly the receptive quadrant area
In case of dissipative 4-Quadrant solution, energy to be
evacuated directly in the converter costs money , weight,
reliability (number of power semiconductors being used)
CERN generally limits the power & energy its 4-Q. converter accept.
Typical operation & cycles from operational data are set once for all.





Result of optimization process is a non-full 4-Q converter, but a cost,
weight and technical effective compromise adapted to operation needs.

Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
17
Load Impact on design parameters
Impact on the resistance value on design & losses
Thurel Yves, CERN, CAS Pwr Conv. 2014
-30
-20
-10
0
10
20
30
-800 -600 -400 -200 0 200 400 600 800
I[A]
[V]
V [V]
P [kW]
Choice 1: R.cable = 20 mOhms











P
MAX
13.2 kW 100 %
P
ABSORBED
1.25 kW 100 %
Cable Losses 14.4 kW 100 %
-30
-20
-10
0
10
20
30
-800 -600 -400 -200 0 200 400 600 800
I[A]
[V]
V [V]
P [kW]
Circuit Characteristics
L = 1 H, dI/dt = 10 A/s
I = [-600 A; +600 A]
P
Cable
< 15 kW
-1.25kW
13.2 kW
22 V
MAX

Circuit Key Features
L di/dt = 10 V
10 mOhms < R
cable
< 21 mOhms
6 V < R
cable
I
MAX
< 12 V
Choice 2: R.cable = 10 mOhms











P
MAX
9.6 kW 73 %
P
ABSORBED
2.5 kW 200%
Cable Losses 7.2 kW 50 %

9.6 kW
16 V
MAX

-2.5kW
Max. power absorbed is divided by 2x, when power in generator
mode is only 1.37x higher, at the cost of higher cable power losses.
te-epc-lpc
08/05/2014
18 Thurel Yves, CERN, CAS Pwr Conv. 2014
Review?... Isnt it the
good moment to sleep
te-epc-lpc
08/05/2014
19
4-Quadrant Topologies: principles
2x Thyristor bridges in anti//
Very well known.
Load energy returned to mains.

Linear dissipative Stage
Basic background known.
4-Q operation to be explained later.
Dissipation in transistors, acting as
programmable resistor.

Switching Stage
Conventional 4-Q H bridge (+L-C Filter).
Power returned to capacitor or
dissipated in add. brake chopper.

Thurel Yves, CERN, CAS Pwr Conv. 2014

R
L
L
L
Bip. / Mos.
Transistor

te-epc-lpc
08/05/2014
20
4-Quadrant Topology Simplified Tree
A large panel of combinations
Several combinations.
From 50Hz to High Frequency
From medium complex to
insane complexity.
Not all solutions are
represented, only CERN
solutions of interest.
Thurel Yves, CERN, CAS Pwr Conv. 2014
4-Q. Thyristor
bridge
Filter
Filter
Rectifier
bridge
4-Q. AC-DC
Converter
Filter
Rectifier
bridge
Filter
DC-AC
Converter
4-Q. DC-AC
Converter
Storage Energy
or B. Chopper
Load
4-Q. DC-DC
H Bridge
Filter
4-Q. AC-DC
Converter
Filter
HF Transfo.
(Volt. Adapt.)
LF Transfo.
(Volt. Adapt.)
4-Q. Linear
Stage
Mains
te-epc-lpc
08/05/2014
21
4-Quadrant Topology Simplified Tree
with different results,
regarding load energy:
Some almost lossless.
some fully dissipative.
Someit depends
Thurel Yves, CERN, CAS Pwr Conv. 2014
4-Q. Thyristor
bridge
Filter
Filter
Rectifier
bridge
4-Q. AC-DC
Converter
Filter
Rectifier
bridge
Filter
DC-AC
Converter
4-Q. DC-AC
Converter
Storage Energy
or B. Chopper
Load
4-Q. DC-DC
H Bridge
Filter
4-Q. AC-DC
Converter
Filter
HF Transfo.
(Volt. Adapt.)
LF Transfo.
(Volt. Adapt.)
4-Q. Linear
Stage
Mains
Ej
Ej
Ej
Ej
te-epc-lpc
08/05/2014
22
1. Topology: 2x(2-Q) Thyristor
Schematic
[0,0] & [V] axis circulating current between the 2 bridges

Thurel Yves, CERN, CAS Pwr Conv. 2014
Active Filter
increase dynamic
performance (low
amplitude
bandwidth, low
frequency ripple,
perturbation)
B6C
.
.
.
.
.
.
alpha-refcircu
540u
2.16m
4.7
Rmag
Lmag
7.5m
540u
2.16m
4.7
7.5m
7.5m
7.5m
B6C
.
.
.
.
.
.

V
I
1
2
V
I
4
3
te-epc-lpc
08/05/2014
23
1. Topology: 2x(2-Q) Thyristor
Advantages
Very high power possible.
Load energy returned to mains.
High Efficiency.
Very well known topology, and rather simple design.
EMC relatively easy to handle (low frequency operation)

Drawbacks
Low Bandwidth (add. active filter can help for small signal range).
Poor AC mains perturbation rejection (limited by L-C filter).
Distortion (crossing axes, and point [0,0]) requires some additional
circuit for high precision converter: Circulating current.
Size, weight of transformer and filtering elements.


Thurel Yves, CERN, CAS Pwr Conv. 2014

te-epc-lpc
08/05/2014
24
2. Topology: 50Hz + 4-Q Linear
Schematic / Principle
Thurel Yves, CERN, CAS Pwr Conv. 2014
50 Hz transfo.
Optimal voltage
output
(Galvanic Isolation)
4-Q Linear Stage
(Dissipate Magnet
Energy)
I
LOAD
V
L
O
A
D


L F
L F L F
L F
Rectifier +
Double Output
Low Frequency
Filter

L
I
N
E
A
R
DC Fixed
DC Fixed
L F
te-epc-lpc
08/05/2014
25
2. 4-Q Linear Stage: basic principle
Highlight: using double DC fixed voltage outputs

Thurel Yves, CERN, CAS Pwr Conv. 2014

0..
100A

10V

R
L
L
L
12V

12V

2V

22V

0A..100A

0A


100A

2V

R
L
L
L
12V

12V

10V

14V

100A

0A


100A
..0A

-8V

R
L
L
L
12V

12V

20V

4V

100..0A

0A


0..
-100A

-8V

R
L
L
L
12V

12V

4V

0A

0A..100A

20V

GENERATOR GENERATOR RECEPTOR
-8V
2V
10V
-8V
0A
100A 100A
0A
-100A
20Vx100A in Q1!!!

Efficiency!!!

te-epc-lpc
08/05/2014
26
2. 4-Q Linear Stage: basic principle (optimized)
Highlight: using double DC variable voltage outputs
for better efficiency an losses reduced in receptor mode, adequate if DC Bus fast enough
Thurel Yves, CERN, CAS Pwr Conv. 2014

0..
100A

10V

R
L
L
L
12V

12V

2V

22V

0A..100A

0A


100A

2V

R
L
L
L
4V

4V

2V

6V

100A

0A


100A
..0A

-8V

R
L
L
L
0V

0V

8V

-8V

100..0A

0A


0..
-100A

-8V

R
L
L
L
10V

10V

2V

0A

0A..100A

14V

GENERATOR GENERATOR RECEPTOR
-8V
2V
10V
-8V
0A
100A 100A
0A
-100A
8Vx100A in Q1

Efficiency better

te-epc-lpc
08/05/2014
27
2. 4-Q Linear Stage: basic principle (variation)
Highlight: using single DC variable voltage output






Remark:
This solution is easier, from the inverter/input side, but leads to
polarity new transitions which can be source of distortion, and
finally lead to a complex 4-Quadrant stage.


Thurel Yves, CERN, CAS Pwr Conv. 2014

I
L
V
L
R
L
L
L
V
E+
V
E+

I
L
V
L
R
L
L
L
V
E
Polarity Switches
Single output inverter
te-epc-lpc
08/05/2014
28
2. Topology: 50Hz + 4-Q Linear
Advantages
Schematics and topologies are very well known.
Simple schematic.
Bandwidth (10 kHz relatively easily).
Drawbacks
Energy recovered from the load is entirely lost & has to be handled.
Efficiency very low in generator mode, sometimes increased using
thyristor instead of diode bridge providing a variable DC bus at a
cost of a strongly limited speed (50Hz).
Size, weight of transformer and filtering elements.
Distortion of the voltage @ 0A (circulating current can fix it).
4-Quadrant stage often realized with many transistors in parallel on
old designs (50Hz tagged) with MTBF and avalanche failure issues.



Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
29
2. Topology: 50Hz + 4-Q Linear
Drawbacks illustrated
Sometimes realized with
many transistors in //.



Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
30
3. Topology: 50Hz + 4-Q Switching
Schematic

Thurel Yves, CERN, CAS Pwr Conv. 2014
Y
400V ~
50 Hz
R
S
T
Ull
L
F
/2
L
F
/2
+
-
C
F
R
BC
T
BC
D
BC
Brake chopper
Rectifier
bridge
Chopper
H - bridge
Load
magnet
R
C
L
C
V
DCout
I
DCout
V
DCin
T
1
T
4
T
2
T
3
T
1
=T
4
=T
2
=T
3
v
0
(t)
HL
F
/2
HC
F
HL
F
/2
4-Q Switching
Stage
Additional
Brake Chopper
(Optional)
(Magnet Energy
Heat if in excess)

High Frequency
Output Filter
Low Frequency
Output Filter
(Magnet Energy
1/2 C V)
50 Hz transformer
Optimal voltage
output
(Galvanic Isolation)
Diode
bridge
te-epc-lpc
08/05/2014
31
3. 4-Q Switching Stage: basic principle
Highlight: Using natural reversibility of full (or H) bridge
Thurel Yves, CERN, CAS Pwr Conv. 2014
I
L
V
L
V
E
T1=T4=T2=T3
T1
T4 T2
T3
V
s
V
MEAN
T
I
MEAN

V
E
-V
E
H Bridge: different modes
It is possible to work in Buck
Mode so that both
- I is reduced
(HF output voltage ripple).
- Losses are reduced.
- transition between these 2
possible modes can be
source of distortions.
V
MEAN
=(2.-1).V
E
1-
te-epc-lpc
08/05/2014
32
3. Topology: 50Hz + 4-Q Switching
Advantages
High power, with part of magnet energy re-used.
Well known topology commercial elements.
High bandwidth (10 kHz) (good AC mains rejection).
No distortion (if H bridge not changing mode).

Drawbacks
4-Q H bridge switching load current
(conduction & commutation losses).
50Hz LC-Filter un-damped by H bridge neg. resistor.
Size, weight of transformer and filtering elements.
EMC issues with H bridge close to the load.


Thurel Yves, CERN, CAS Pwr Conv. 2014
150A
200V
te-epc-lpc
08/05/2014
33
4. Topology: Switching + 4-Q Linear
Schematic
Thurel Yves, CERN, CAS Pwr Conv. 2014
High Frequency Soft
commutation H Bridge
HF Transformer 300Hz L C Filter
HF Filter
+ Rectifier
4-Q Linear
Stage
To be noticed: Dual DC Bus must at least follow the output
voltage reference (speed) not to saturate the 4-Q linear stage !!
Dual DC
variable bus
te-epc-lpc
08/05/2014
34
4. Topology: Switching + 4-Q Linear
Advantages
Addition of 2 distinct well known topologies.
High bandwidth with good AC mains
perturbation rejection capability.
Addition of a fast input DC bus makes
possible to achieve correct efficiency:
generator mode optimized, with on-transistor being
almost transparent, 4-Q stage generating few conduction losses.
EMC OK (soft commutation & 1x high frequency switching stage).
Drawbacks
Energy recovered from the load is entirely lost & has to be handled.
Distortion of the voltage @ 0A (circulating current can fix it).
Dealing with Inverter, 4-Q Linear Stage and Circulation current loops
can lead to some complexity and some limitations.

Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
35
5. Topology: Switching + 4-Q Switching
Schematic

Thurel Yves, CERN, CAS Pwr Conv. 2014
High Frequency Soft
commutation H Bridge
HF Transformer 300Hz L C Filter
HF Filter
+ Rectifier
4-Q Switching
Stage
HF Filter
+ Rectifier
te-epc-lpc
08/05/2014
36
5. Topology: Switching + 4-Q Switching
Advantages
Part of magnet energy re-used, making
this suitable for pulsed application.
Very well known topology.
High bandwidth, with good AC mains
perturbation rejection capability.
No distortion of the signal.
Drawbacks
4-Q H bridge switching load current (conduction & commutation
losses), with a total of 2 switching stages in series.
EMC challenging, 2x high frequency switching stages.
2 high frequency filtering levels, number of components
(duplication of the switching stages with drivers, control etc)

Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
37
5.a Topology: Switching + 4-Q Switching (2)
Thurel Yves, CERN, CAS Pwr Conv. 2014
TH1
TH3
TH4
TH2
THbis1
THbis3
THbis4
THbis2
Iout1 iout2
Out -
Out +
Vdc +
Vdc -
Vin
off
off
off
off
TD1
TD2
TD3
TD4
IL

on
on
on
on





TH1
TH3
TH4
TH2
LOA
D
Iout

r
r
See Paper from
CERN & LEEI for
detailed
explanation.
Schematic (with Energy sent back to mains capability)
Exploring switching solutions for a fully reversible converter.
te-epc-lpc
08/05/2014
38
5.a Topology: Switching + 4-Q Switching (2)
Advantages
Theoretically possibility demo of fully reversible with energy sent
back to mains (AC thyristor bridge to be added to the prototype).
High frequency power converter (low volume and high dynamic
performance).
High bandwidth, with good AC mains perturbation rejection.
No distortion of the signal.

Drawbacks
Complex control & topology.
This converter stayed at design prototype
stage, and was not deployed in any machine
at CERN since judged too complex / risky.



Thurel Yves, CERN, CAS Pwr Conv. 2014
600A
12V
te-epc-lpc
08/05/2014
39 Thurel Yves, CERN, CAS Pwr Conv. 2014
Practical things?
Means photos? Cool !
te-epc-lpc
08/05/2014
40
LHC Machine, using 4-Quadrant converters
With its superconductive loads, LHC required around 1500
4-Quadrant converters ending up with 4 different types.
Correctors of a slow machine
in the range of [60A; 600A],
with no strong interest in
energy recovery.
Installation in a tunnel,
with a strong interest in
low level of losses.
(medium current < 600A)
Large quantities, difficult and
long access to the converters,
with strong interest in high
reliability.

Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
41
LHC Machine, using 4-Quadrant converters
Noise emitters and sensitive clients installed close to each other.
Installation of several sensitive to noise system
(Analog down to nanovolt quench systems)
in tiny areas (UA, RR), close to converters.
Low EMC noise emission is of great interest.
Load being magnet chain of more than 2 km long
Easy coupling (electrical and magnetic) between windings in cryostats
requiring low noise (EMC) generators.
Converters operating at current precision at the level of ppm.
Being precise at the one part per million level means achieving this
with every shot!

Low EMC noise emission is a key parameter for LHC.



Thurel Yves, CERN, CAS Pwr Conv. 2014
100 km
te-epc-lpc
08/05/2014
42
LHC Machine, using 4-Quadrant converters
Thurel Yves, CERN, CAS Pwr Conv. 2014
General approach each designer adopted was:
Switching stage + 4-Quadrant linear stage.






Main advantages initially expected and indeed obtained:
High Efficiency possible (soft commutation inverters, 4-Quadrant
stage not leading to high level of losses in generator mode).
Very low level of output noise, differential and common mode.

te-epc-lpc
08/05/2014
43
LHC Machine, using 4-Quadrant converters
Thurel Yves, CERN, CAS Pwr Conv. 2014
4 large families, 3 external companies: CIRTEM, EEI,
TRANSTECHNIK , in addition to CERN teams.

LHC120A-10V
4-Quadrant

300 Units
CERN Design
LHC600A-40V
4-Quadrant

40 Units
Transtechnik
LHC600A-10V
4-Quadrant

400 Units
EEI-CIRTEM
LHC60A-08V
4-Quadrant

730 Units
CERN Design
te-epc-lpc
08/05/2014
44
LHC 120A CERN Design & Results
Thurel Yves, CERN, CAS Pwr Conv. 2014


A CERN design in detail:
LHC 120A 10V Converter
120A 10V.
Air cooled converter.
50 kHz inverter switching frequency
1 kHz voltage bandwidth.
Load time constant range
[0.1; 1050] seconds.
300x installed
in LHC.


te-epc-lpc
08/05/2014
45
LHC 120A Topology Choice
AC-DC
3 phases + Neutral.
1400V Rectifiers (AC rectification) .
70Hz Input filter (damped with C-4C) to
give a flat 540V DC (around -20dB on 300Hz).
Soft start based on R + switch.

High frequency soft commutation inverter
Fixed switching frequency (50..100kHz).
ZVS operation through Phase Shifted.
ZCS for lagging leg possible (saturable inductor)
1.2kV IGBT Full Bridge.
Voltage or current regulation mode possible.


Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
46
LHC 120A Topology Choice
Rectifier + DC Filter
Schottky rectifiers.
High frequency ripple can be easily filtered.
Double LC cell at 5..10 kHz cut off frequency.

4-Quadrant Linear Stage (4-QLS)
4-Q Linear Stage based on several Power Mosfets
mounted in parallel for each leg (top and bottom).
Use of 4-QLS as an active filter.
Use of 4-QLS to pre-load inverter (continuous mode).
Circulating current always present but value
depending from load current (no mode change
transition, both DCDC outputs always minimum loaded.

Thurel Yves, CERN, CAS Pwr Conv. 2014
Rectifer + Filter
4 Quadrant
Linear Stage
VOUT
te-epc-lpc
08/05/2014
47
LHC 120A Topology Choice
Highlights on 4-QLS:
A double power source programmable
via inverter control (Dual VE source).

2x equivalent MOSFETS, Top & Bottom leg,
used as switch and/or programmable
resistor depending on generator /receptor
mode.

A command based on control rules :


Thurel Yves, CERN, CAS Pwr Conv. 2014
) I R V ( V
HIGH HIGH E OUT
+ =
) I R V ( V
LOW LOW E OUT
=

V
OUT
R
L
L
L
V
E
I
HIGH
I
LOW
V
E
I
OUT
V
GS LOW
V
GS HIGH
R
HIGH
& R
LOW
are equivalent MOSFET resistance
te-epc-lpc
08/05/2014
48
LHC 120A 4-QLS without circulation current
Limitation of the linear stage

Thurel Yves, CERN, CAS Pwr Conv. 2014
Main analog loop
cannot provide
needed step for
biasing MOSFET
(-3V to + 3V).
Playing with fixed
bias is a bad idea !!!
Discrepancies on
V
GS ON
1V !!!
+
-
V
GS LOW
V
GS HIGH
1
-1
PID
V
OUT
V
REF
Without Circulating Current
Without Circulating Current
V
OUT
R
L
L
L
V
E
I
HIGH
I
LOW
V
E
I
OUT
V
GS LOW
V
GS HIGH
V
DS
LOW
V
DS

HIGH
te-epc-lpc
08/05/2014
49
LHC 120A 4-QLS with circulation current
adding circulation current loop.
Thurel Yves, CERN, CAS Pwr Conv. 2014
V
OUT
R
L
L
L
V
E
I
HIGH
I
LOW
V
E
I
OUT
V
GS LOW
V
GS HIGH
V
DS
LOW
+
-
V
GS LOW
V
GS HIGH
-1
PID
V
OUT
V
REF
To avoid 0A distortion
crossing , a circulating
can help, biasing
MOSFET gates.
These additional
loops (one per leg)
must not perturb
main loop.
I
Circulation
can vary.
+
-
+
+
I
REF
I
HIGH
+1
PID
+
+
With Circulating Current
With Circulating Current
V
DS

HIGH
te-epc-lpc
08/05/2014
50
LHC 120A 4-QLS with circulation current
Different approaches:
Circulating current always present, its value changing with I
out
:
Easier control (reference is only smoothly changing with I
out
).
Inverter / DC Bus reference must always be set for this circulating
current being possible, leading to higher losses in receptor mode.
Continuous mode (Gate of each side MOSFET always biased).
Circulating current present when close to a potential transition:
This solution saves a lot of energy when
recovering, since inverter doesnt inject
power in MOSFET when absorbing power.
dI/dt Limitation exists, since if too fast,
circulating current doesnt have time
to appear and MOSFET Gate are not
well biased.

Thurel Yves, CERN, CAS Pwr Conv. 2014
[t]
Q2 Q1
ICC REF
VGS Q2
Fr. Main loop
Fr. Icc loop
Q1 & Q2
te-epc-lpc
08/05/2014
51
Dual DC Bus REF built according to:
V
BIAS MIN
not to saturate MOSFET
V
BIAS OPT
For 4-Q Linear stage with better
rejection of mains disturbances and 300 Hz
V
MIN
& I
circ. MIN
to ensure minimum load
(Conduction mode un-changed, 4-Q acting as a dummy load)
VE Loop to be faster / transparent than 4-Q stage loop

Thurel Yves, CERN, CAS Pwr Conv. 2014
+
-
V
GS
LOW
V
GS
HIGH
-1
PID
V
OUT
V
REF
+
-
+
+
I
REF
I
HIGH
+1
PID
+
+
V
E
+
-
PID
INVERTER
|V
REF
| +
+
V
MIN
V
BIAS
V
MIN
V
BIAS
LHC 120A 4-QLS Dual DC Bus Reference
te-epc-lpc
08/05/2014
52
LHC 120A 4-QLS MOS Model (FB180SA10)

Controlling a MOSFET as a variable resistance: a challenge
Thurel Yves, CERN, CAS Pwr Conv. 2014
MOSFET is difficult to
control directly from the
main loop to the gate, as
a variable resistance.
Indeed gain is from
virtually 0 if saturated to
thousands when almost
opened.

Circulating current and
V
BIAS
value allow to limit
the operating equivalent
resistance area, for an
easier control.
0.0
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
3.7 3.9 4.1 4.3 4.5 4.7 4.9
VGS [V]
R
d
s
o
n

[
O
h
m
s
]
Rdson at VDS= 5 V
Rdson at VDS= 2 V
Rdson at VDS= 1 V
Rdson at VDS= 0.6 V
Rdson at VDS= 10 V
Rdson at VDS= 20 V
te-epc-lpc
08/05/2014
53
LHC 120A Obtained and Measurable Results
Performances as a voltage source
No distortion @ 0A, thanks to circulating current.
High bandwidth (could be higher in small signal: 4-Q linear stage)
Good bandwidth in large signal (Dual voltage source speed +
4-Quadrant linear stage)
Thurel Yves, CERN, CAS Pwr Conv. 2014
Z=Vout / Vref
-40
-30
-20
-10
0
1 10 100 1000 10000
f [Hz]
G [dB]
[0V 0A]
[8V 100A]
te-epc-lpc
08/05/2014
54
LHC 120A Obtained and Measurable Results
Performances as a low-noise system
EMC is very well managed, with strong focus on cleaning inverter
side, being the only noise-generator of the topology presented.
EMC noise: common mode and differential very low, AC & DC Side.
Thurel Yves, CERN, CAS Pwr Conv. 2014
LHC, a Low Noise machine ?? Joking? Type
LHC on google and check it from yourself !!
Output voltage noise measured on LHC 120A 10V converter
(output dissymmetrical measurement polarity to ground)
3 mV
rms
@ f
switching Almost no noise at high frequencies

te-epc-lpc
08/05/2014
55 Thurel Yves, CERN, CAS Pwr Conv. 2014
Explore topics surrounding
4-Q converter, why not?
te-epc-lpc
08/05/2014
56
Some other concerns of interest
Magnet protection applied to 4-Q family
Magnet protection versus 4-Quadrant converters

Practical issues on 4-Q converters reception
Qualification of several design (even if close) for a final operation
on superconductive magnet, with a wide range of
time constant, without any early and easy
access to the costly cryogenic - loads !
Test of 1500 Converters @ CERN :
Need to find some efficient and
compact tests giving much
information on the
converter
integrity.
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
57
1. Magnet Protection and 4-Q Converter
Magnet protection widely spread principle @ CERN
Power Converter must always ensure a safe path for magnet
current for the Energy Extraction System to work whatever







its state: safe, faulty, even disconnected from mains!
Thurel Yves, CERN, CAS Pwr Conv. 2014
R.cable Magnet
L
I
load
R.EE
Energy Extraction
System: Active
V
conv
I
conv R.cable Magnet
L
I
load
R.EE
Energy Extraction
System: Inactive
V
conv
I
conv
te-epc-lpc
08/05/2014
58
1. Magnet Protection and 4-Q Converter
1-Q Converter typical protection
Safe Path ensured for Power Converter side with a diode only
Simple, efficient, high MTBF system: 1 component only !!!
Even redundant most of the time with Converter Components
Thurel Yves, CERN, CAS Pwr Conv. 2014
1-Quadrant Power
Converter
Power Supply
Rectifier Diodes
(Dr1+Dr2)
Power Supply
Free-Wheeling
Rectifier diodes
Dr1
Dr2
1
st
Path 2
nd
Path
R.cable Magnet
L
I
load
R.EE
Energy Extraction
System
Can I call a simple diode a complete
system ? Is that simple?
te-epc-lpc
08/05/2014
59
1. Magnet Protection and 4-Q Converter
4-Q Converter typical protection
Safe Path ensured for Power Converter side by a Crowbar
Active electronics which must be launched on purpose
Thyristor (Power Mosfet) self triggered by over-voltage condition
Converter OFF or FAULTY = non-conductive
Thurel Yves, CERN, CAS Pwr Conv. 2014
R.cable Magnet
L
R.EE
I
load
V
conv
I
conv
Definitively much more complex than a
simple diode !!!
Join the 4-Q family they said !!!
te-epc-lpc
08/05/2014
60
1. Magnet Protection and 4-Q Converter
Stressful period when commissioning circuits with crowbar
Depending on operating conditions, crowbar thyristor can be auto-
reset by natural oscillation (L
cabling
-C) leading to switching mode.
Thurel Yves, CERN, CAS Pwr Conv. 2014
They told me it was a switching
convertercrowbar included ?
Pwr
Module

L
R
cable
R
crowbar
V
circuit
V
CH1
C
crowbar
L
crowbar
V
CH2
Pwr
Module

L
R
cable
R
crowbar
V
circuit
C
crowbar
Test @ 15A
(600A nominal)
Test @ 10A
Is crowbar
dead?
One Spike
expected,
creating
higher voltage
to launch the
thyristor ctrl.
N Spikes
measured.
I
out
I
out
V
out
V
out
te-epc-lpc
08/05/2014
61
2. Testing 4-Quadrant Converter
How to test a 4-Quadrant converter?
In case of a superconductive magnet load, how to ensure the
converter will fit the requirements? In receptor mode?
Finding a superconductive load is not easy or can be available too
late in a project.
Some topologies are more sensitive to the quadrant
(generator or receptor) they operate regarding their control
It is mandatory to get a small signal behaviour with a good
confidence on a warm magnet despite low inductance value .
It is important to really and practically test the receptor
capability of the converter, like if operating on a superconductive
magnet with hundreds or thousands of sec of time constants to
validate thermal design.
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
62
2. Testing 4-Quadrant Converter
Method of the 2x Signal generators tests
With a moderate time constant load, it is possible to check the
control loop in receptive quadrant with a simple measurement.





Use 2 signal generators in series, once creating high frequency
small signal (sinus or square), with the other signal generator
creating slow frequency square signal (amplitude modulated) to
generate, at the transition, receptive quadrant conditions.
Thurel Yves, CERN, CAS Pwr Conv. 2014
R.cable Magnet
L
I
load
V
conv
I
conv
4-Q
V
ref
te-epc-lpc
08/05/2014
63
2. Testing 4-Quadrant Converter
Method of the 2x Signal generators tests
Of course, and zooming on the transition only, no real difference
between a warm load and a superconductive magnet (zoom).
Difference can appear and can come from thermal management.
This test ensure minimum stability of the converter while receiving
energy, working in receptive quadrants, without cryogenic load!






Test on warm magnet (to=0.1 s) Test on cold magnet (to=60 sec)
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
64
2. Testing 4-Quadrant Converter
Method of the 2x Back-to-Back Power converters
One converter simulate a large time constant load, using its 4-
Quadrant nature.
Very elegant and very efficient solution to test thermal management.
Issues can come if some instability exists in the converter to be
tested, since being used as a load as well.
Thurel Yves, CERN, CAS Pwr Conv. 2014
Vref1 Vload
Iload
Vref2
Converter 1 Converter 2

R+L*jw
Zserie
Simulate R,L
DCCT
Seems too nice, does it really work ?? If the case, better requires
2 prototypes instead of 1 in the contract, to validate the design
te-epc-lpc
08/05/2014
65 Thurel Yves, CERN, CAS Pwr Conv. 2014
2 minutes more !!
te-epc-lpc
08/05/2014
66
1. New Project of 4-Q converters
Switching + 4-Q switching
For pulsed applications, implementing digital regulations, and
changing 4-Q H Bridge operation mode for lower high
frequency ripple.
Performances regarding EMC
output signature tend to
be similar to switching 2
4-Q Linear Stage, once
well understood
and with adequate
filters in place.
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
67
2. New Project of 4-Q converters
Switching + 4-Q Linear
For LHC radiation tolerant converter upgrade.
A linear stage is less sensitive to radiation than H bridge stage
(Mosfet being always conductive, then suffered only from total dose
degradation, when switching transistor can die on 1x single event).
Implementing regulation at the level of the 4-Q linear stage, directly
controlling the Mosfet as current sources.
Proposing a redundant solution of 2x 4-Q power modules working in
parallel thanks to the current
control in the 4-Q
stage heart.
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
68
2. New Project of 4-Q converters
Switching + 4-Q Linear
Current sources
in 4-Q Linear Stage
are ideal for:
Implementing
circulation loop
easily.
Adding Mosfet
in parallel
inside each 4-Q
power module.

Thurel Yves, CERN, CAS Pwr Conv. 2014
VREF +
-
Voltage
Regulator
I
REF
>0
V
M
E
A
S
+
-
Current
Regulator
+
-
GATEREF
Current
Regulator
GATEREF
Driver
Driver
Driver
Driver
V
MEAS
CURRENT LOOP
CURRENT LOOP
VOLTAGE
LOOP
+
+
+
+
I
CC
I
CC
CURRENT LOOP
VREF +
-
Voltage
Regulator
I
REF
>0
V
M
E
A
S
V
MEAS
VOLTAGE
LOOP
CURRENT LOOP
CURRENT LOOP
CURRENT LOOP
CURRENT LOOP
CURRENT LOOP
Slides courtesy of Vicente Raul Herrero Gonzales, CERN
te-epc-lpc
08/05/2014
69
2. New Project of 4-Q converters
Switching + 4-Q Linear
Current sources in 4-Q Linear Stage are ideal for:
Implementing active sharing for redundancy between 2x 4-Q power
modules operating in parallel, sharing the load current information.
Thurel Yves, CERN, CAS Pwr Conv. 2014
Slides courtesy of Vicente Raul Herrero Gonzales, CERN
CURRENT LOOP
V
REF +
-
Voltage
Regulator
I
REF
>0
V
M
E
A
S
VOLTAGE
LOOP
CURRENT LOOP
CURRENT LOOP
CURRENT LOOP
V
MEAS
PM1
PM2
te-epc-lpc
08/05/2014
70 Thurel Yves, CERN, CAS Pwr Conv. 2014
Are you still
with me?
te-epc-lpc
08/05/2014
71
Conclusion
4-Quadrant converters are specific items
Understanding the final use is a key point to choose the
correct topology, and to be able to limit the requirements for
a suitable product.
All topologies present drawbacks, which can be managed if
the 1
st
step (clear final use) is correctly done.
These converters will always remain intrinsically more
complex and will require more attention: test/qualifying
phase is difficult and has to be addressed from the beginning.
Not a lot of companies are capable of designing a high
performance converter, since market is relatively small.
Thurel Yves, CERN, CAS Pwr Conv. 2014
te-epc-lpc
08/05/2014
72
And finally
Thurel Yves, CERN, CAS Pwr Conv. 2014
Thank you for
your attention !
This presentation : CAS INDICO N263328 or EDMS N1380014