Application Note AN-1038

Low Voltage DC Supply

Dimmable Ballast for 1 x 36W T8 Lamp
By Peter Green, Senior Lighting Systems Engineer

Table of Contents

Page
Introduction ..........................................................................................1
Functional Description..........................................................................1
Selection of the Output L and C Values ...............................................2
Preheating and Cathode Heating .........................................................2
Control IC Auxiliary Component Selection ...........................................3
Layout Issues .......................................................................................6
Step-Up Transformer Design ...............................................................7
Output Inductor Design ........................................................................8
Bill of Materials.....................................................................................9

It is possible to design an effective dimmable ballast based around the IR21592, that is powered
from a low voltage DC supply instead of the AC line. A non-dimmable version based around the
IR2156 is also possible using the same basic configuration as described here. This application
note shows a ballast for a single 36W T8 lamp driven from a 30V DC supply. Lower supply
voltages are possible, however the IR21592 control IC requires up to 13V supplied to the VCC
pin before it will operate, also as the current is high large conductors are needed to keep losses
at an acceptable level (for a 36W ballast operating from a 30V supply the input current is around
1.25A).

www.irf.com AN-1038 cover

 AN-1038 revB
International Rectifier • 233 Kansas Street, El Segundo, CA 90245 l USA

Low Voltage DC Supply Dimmable Ballast for 1 x 36W T8 Lamp

by
Peter Green, Senior Lighting Systems Engineer

I. INTRODUCTION which produces a high voltage switching voltage at

the secondary that can be connected to a conven-
It is possible to design an effective dimmable ballast tional series inductor and capacitor ballast output stage
based around the IR21592, that is powered from a to the lamp.
low voltage DC supply instead of the AC line. A non-
dimmable version based around the IR2156 is also In this system both power switching MOSFET sources
possible using the same basic configuration as de- are connected to COM. In order to obtain the required
scribed here. The following example shows a ballast current level and phase information a sense resistor
for a single 36W T8 lamp driven from a 30V DC sup- must be added from the source of the LO driven
ply. Lower supply voltages are possible, however the MOSFET to COM. The current will be much larger at
IR21592 control IC requires up to 13V supplied to the this point than in a mains powered ballast circuit and
VCC pin before it will operate, also as the current is consequently a appropriately lower resistor value is
high large conductors are needed to keep losses at necessary, in this example 0.1R. Since the step up
an acceptable level (for a 36W ballast operating from transformer introduces no significant phase shift, the
a 30V supply the input current is around 1.25A). waveform detected at the current sense CS input to
the control IC IR21592 is almost identical to a current
II. FUNCTIONAL DESCRIPTION sense signal in a half bridge ballast circuit. This signal
can be used in the same way to detect the output cir-
The ballast control IC and associated circuitry is pow- cuit phase shift for dimming control.
ered from the low voltage DC supply by means of a
simple dropper resistor that provides the required In this system as well as in non-dimming designs the
15.6V VCC supply from the 30V input, clamped by CS pin is used to monitor the current and detect faults
the internal zener diode of the IC. This can supply allowing the ballast to shutdown if the lamp fails to
sufficient current for the ballast control circuitry, dissi- ignite correctly as in half bridge ballast circuits.
pating only a small amount of power and therefore no
charge pump is needed to maintain the supply. The It is not necessary for the output section to be isolated
circuit topology consists of a push-pull power switch- from the input section and so one side of the second-
ing stage as opposed to the usual half-bridge employed ary can be referenced back to COM. We can there-
in mains supplied ballasts. This simplifies the circuit fore connect one side of the lamp to COM allowing
since a high side driver is not needed so the VB pin of the SD pin of the control IC to detect lamp removal or
the IC can be connected directly to the VCC, the VS an open circuit in the lower filament. During dimming
pin can be connected to COM (0V) therefore no boot- the lamp is prone to produce striations (dark rings that
strap diode or capacitor are necessary. The output appear to move along the tube). We can remove these
section has a step-up transformer with a split primary, by adding a small DC offset to the lamp voltage through
www.irf.com 1
AN-1038

R16, which is connected back to the 30Vdc bus. A In advanced mode the BDA software generates a curve
snubber network consisting of R15 and C10 is also showing the ballast operating points:
added to reduce ringing overshoot voltages that oc-
cur when each MOSFET switches off. The snubber
will also increase the commutation time at switch off
so that soft switching can be achieved using the
IR21592, which has a fixed dead time of typically
1.8uS. The MOSFETs used in this example are type
IRF540, which have a Vdss rating of 100V and Rds(on)
of 0.044? at 25ºC. The peak drain voltage is 60V plus
the transient produced by the leakage inductance of
the step up transformer when at switch off which is
comfortably less than 100V limited by the snubber.

I II. SELECTION OF THE OUTPUT L AND C

VALUES

Using the lamp parameters supplied by International

Rectifiers Ballast Design Assistant software we can
calculate the preheat, ignition and running frequen-
cies for a bus voltage of 300V (Revision 4 released
this year is recommended and superceeds all older
releases). The output from the step up transformer Fig. 1
will be a square wave of 300Vp-p as would be ob-
tained using a half bridge connected to a 300V DC These points are good because the preheat frequency
bus. The output circuit will be the same as in a half is more than 5kHz higher than the ignition frequency,
bridge topology. A DC blocking capacitor must also which will prevent the possibility of premature lamp
be included so that the step up transformer does not ignition during preheat caused by the lamp voltage
drift into saturation in one direction if the primary volt- during preheat being too large. The run frequency is
second products are not perfectly balanced in each as required and is well below the ignition frequency,
flux direction. which will allow a smooth lamp startup sequence.

In this example the step up transformer is designed to IV. PREHEATING AND CATHODE HEATING
operate at 40kHz minimum frequency, where the bal-
last will be at maximum output. The core needs to be Current mode pre-heating may be used in a dimming
larger to cope with the same throughput power at a ballast designed for a 36W T8 lamp as it is able to
lower frequency so in this case in order to limit the produce the correct preheat current of 0.6A with this
size to EF25 we have chosen a 40kHz running fre- configuration. The required preheat frequency can be
quency. By iterating the values of the output L and C obtained from the formula
in the software we are easily able find values that
produce the desired running frequency. The values i ph
f ph =
are L=1.6mH and C=6.8mH.
2πCV ph

2 www.irf.com
 AN-1038

and well as the IR21571 which could also be used in a

non dimming low voltage design. It can calculate ap-
− Vin V L ⋅ I ph
2 2
 proximate values of all external resistors and capaci-
V ph = +  in +  tors using the procedures described below. These
π  π C  values may be used to make an initial breadboard type
  ballast setup. The final refining and obtaining of the
exact component values must be carried out in the
where Vin will be 300V. lab by experimentation.

In a dimming ballast it is also very important that

the cathode current is sufficient when the lamp IR2156 based system (non-dimming):
is dimmed. The Cathode current at minimum can
For non-dimming design based around an IR2156, the
be calculated with the formula selection of components is straightforward. CT should
be selected to provide a dead time of approximately
I cath = Vlamp ( 2%) ⋅ 2πf 2%C 1.8uS as close as possible to the fixed dead time of
the IR21592. This can be calculated from the formula:
The lamp voltage at 2% light output, is 165Vpk and
DT
the frequency is 71kHz therefore the cathode current CT = (Farads)
is 0.5Apk, which is 0.35Arms. 1475
A general rule is that the lamp filament (Cathode) re- The value obtained is 1.2nF, in practice 1nF would be
sistance over the range of dimming levels should be acceptable. The next step is to determine the value of
between 3 and 5.5 times the resistance when cold. RT which can be calculated from the formula :
For a T8 36W lamp the cold resistance is around 3?.
The resistance of the cathode at maximum brightness
is not critical, as the arc current flowing in the lamp 1
RT = − 2892 (Ω)
will serve to keep the temperature at a sufficient level. 1.02 ⋅ CT f run
At minimum output the cathode voltage will be around
3Vrms so the resistance will be 9?, which is 3 times
the cold resistance. This will not be the case for many For Frun = 40kHz the value of RT is 22K. *
other types of lamp and consequently voltage mode
preheating is often needed. The value of RPH can be calculated from the formula
:
V. CONTROL IC AUXILIARY COMPONENT  1 
SELECTION RT ⋅  − 2892 
RPH =  1.02 ⋅ CT f PH  (Ω)
The quickest and easiest method for doing this in each  1 
case is to use version 4 (or higher) of the International RT −  − 2892 
Rectifier Ballast Design Assistant software which can  1.02 ⋅ CT f PH 
be downloaded from IR’s website at www.irf.com. The
BDA supports both the IR2156 and the IR21592 as

www.irf.com 3
AN-1038

The preheat time is determined by the value of CPH Schematic Diagram

calculated from the formula : +VIN
F L
1 RDC
1

−6
C PH = TPH ⋅ 0.385 × 10 (Farads)
RS RVDC

CDC LRES

0.33uF is the value typically used to give a 1s preheat NC

1 14
VB

RHO M1
CF CBUS VCC HO

time. 2 13

IR2156
CVDC VDC VS
3 12
RSNUB
RT RT LO
4 11 CRES
RPH RPH CS T
In this circuit RCS is used to shut down the circuit in a CT
CT
5

6
10

9
SD
M2
1 CSNUB

fault condition. The shutdown threshold is 1.3V there- CPH CPH

7 8
COM
RLO

RLIM1

fore the value can be calculated from the modified R1 RSD

formula :
CVCC1 CVCC2
CSD CCS RCS

1 .3 N P
RCS = (Ω) 0V

I IGN N S

where Np is the transformer primary turns (center to IR21592 based system (dimming):
one side) and Ns is the secondary turns. IIGN in this
case is defined as the maximum worst case ignition In a dimming system the selection of external compo-
current that the ballast may produce before shutting nents is more critical and the procedure more compli-
down. cated.
In this case, taking a value of 2A for the ignition The values of Fph, Fign, F(100%) and F(2%) can be
current the value is 0.13Ω so we scale this down to calculated by hand using the procedure described in
the nearest preferred value 0.1Ω. the Lighting Ballast Control IC Designer’s Manual
2001* based on known values of lamp voltage and
The VDC pin can be utilized by connecting it to the power. However the BDA software is able to do this
DC bus via a resistor so that if the supply voltage falls far more quickly using lamp parameters from its own
the output frequency will increase preventing the pos- database.
sibility of hard switching, which would cause overheat-
ing and possible failure of the MOSFETs. A value * Please note that there is an error on p.213 of the
should be chosen that will start to take effect at around Lighting Ballast Control IC Designer’s Manual (2001)
25V in a system designed to run at 30V in this case the formulae for f(100%) and f% should be as stated
150K is recommended. on p250 and p.251

The following procedure can be used to determine the

values for the IR21592 external resistors. Firstly Fmin
must be set, which limits the minimum frequency that
the oscillator will run at. This must be lower than
F(100%) or Fign, whichever is lower.

4 www.irf.com
 AN-1038

25 × 10 −6 − ( f MIN − 10000)⋅10 −10 (Ω)

Note 1
RFMIN = For a preheat current of 0.6Arms which is correct for
( f MIN − 10000)⋅ 2 ×10−14 this lamp the result is 22K.
It is a good idea at this point to verify this value on the
In this case we can select Fmin to be 40kHz this will bench by connecting pin CPH to COM, thus keeping
limit the maximum ballast power. This gives a value the ballast in permanent preheat and ensuring that
of 36K. there is sufficient cathode heating taking place but that
the lamp voltage is low enough to prevent the possi-
The next step is to determine RCS from the formula bility of premature ignition, which in the case of the
IR21592 does not allow the dimming control loop to
close correctly and inhibits the dimming function.
1 .6 N P
RCS = (Ω) Note that for the IR21592 to operate correctly the ig-
nition detection function must be completed correctly,
I IGN N S
which requires that the current at ignition is more than
10% greater than the current during preheat and then
For an ignition current of 2A the value will be 0.16Ω,
falls when the system enters run mode.
which can be scaled down to 0.15Ω or 0.1Ω. The over
current shutdown will obviously be more sensitive for
In order to program the MIN and MAX settings of
0.15Ω so if no problems of false tripping are experi-
the dimmer interface, the phase of the output current
enced this would give better protection.
stage at minimum and maximum lamp power must
be calculated. This is a very complicated calculation
If using the BDA software to calculate the resistor via
requiring the lamp voltage and power to be known
the Program IC function, it should be noted that the
at minimum and maximum dim settings. The follow-
step up function of the transformer is not be taken into
ing method avoids the need for this by assuming that
account. Therefore the value given for RCS will need
the phase will be very close to –90º at minimum bright-
to be multiplied by Np/Ns to provide the correct value
ness and using this value to calculate Rmin from the
to use in a low voltage system. This is unlikely to re-
formula
sult in a preferred E24 resistor value. The easiest way
RFMIN  φ2% 
around this is to choose the nearest preferred value
RMIN = ⋅ 1 −  (Ω)
and multiply Ns/Np. Make a note of the result and then 4  45 
adjust the ignition current in the software and recalcu-
late until the RCS is the same as the value required.
This gives the result 27K.
Bear in mind that the values of Riph, Rmax and Rmin
will change as RCS changes. In this example the value
The BDA software can calculate the value of Rmax by
of 0.15Ω is multiplied by 125/25 giving 0.75Ω. The
first calculating the phase shift based on its database
ignition current is changed from 1.8A to 2.2A and the lamp data parameters, however to get a rough esti-
resistor values recalculated to give RCS of 0.75Ω. mate of what Rmax should be we can use an approxi-
mation of the phase shift. We know from graph of Fig
The next step is to calculate Riph from the following 1 that the frequency at 100% power is below reso-
formula nance and so the phase shift must be between 45º
2 RFMIN RCS I PH N S and zero. To obtain a starting point we can estimate a
RIPH = (Ω) phase shift of 30º at maximum brightness and use the
NP formula

www.irf.com 5
AN-1038

0.86 RFMIN RMIN The procedure is as follows:

RMAX = (Ω) The value of Rmin must ensure that the lamp does
 φ 
4 RMIN − RFMIN 1 − 100%  not extinguish when Vdim is set to 0V so gradually
 45  reduce the DIM input voltage to 0V, adjusting Rmin if
necessary to prevent the lamp from extinguishing.
This gives the result 18K. When DIM is at 0V adjust to the desired minimum level.
Finally adjust the dimming control input voltage to 5V
Having determined the values the circuit can then be and adjust Rmax to give the correct maximum light
assembled and tested. By measuring the input cur- output and adjust between minimum and maximum
rent to the ballast it is possible to know whether the levels tweaking the values of Rmin and Rmax until
maximum power is correct (i.e. the current is 1.25A) both limits are satisfactory.
and if not to change the value of Rmax accordingly. It
is possible to observe the lamp brightness at mini- (Notes 1,2, and 3 also apply to setting up an AC line
mum and adjust the value of Rmin to obtain the de- supplied dimming that utilizes a IR21592)
sired result.

* If necessary values can be altered slightly to achieve Schematic Diagram

fine adjustment of the frequencies to obtain the opti-
mum preheat level and lamp running power. +30VDC F1

2A
L1 R16

22K
1W

Note 2 R1 R8
24K 1K
0.5W
C11 L2
100nF 400V 1.6mH

At this point it is also necessary to verify the values of C1 C2

R2
5K6
IC1 R13
Q1
18R
C3100nF

Rmin and Rmax on the bench. In order to do this the

VDC HO
1uF 220uF 1 16
50V 50V C4 10nF VCO VS IRF540
2 15
R15

ballast can be switched on and allowed to preheat and

IR2159

C5 330nF CPH VB C8 C9 1K5

3 14 100uF 100nF 1W C12 36W
R3 10K 25V 6n8
DIM VCC T8
1500V

ignite the lamp, with the dimming control input to be

4 13 T1 C10 LAMP
Control Input R4 12K 100nF
MAX COM Q2
0 to 5Vdc 5 12 R9 400V
(w.r.t. 0V) R14 IRF540
2M2

set to 5Vdc.
R5 27K MIN LO 18R
6 11 25+25 : 125

R6 36K FMIN CS R11

7 10
R10 1K
R7 27K IPH SD
8 9
680K

The procedure is as follows: R12

0R15

Increase the voltage at the DIM input further if neces- C6

470nF
C7
470pF

sary until minimum frequency is reached. It will go no

lower if the dimming level is raised further. 0V

NOTE : All capacitors are rated at 50V DC unless otherwise

stated.

The minimum frequency limit set by Rfmin should be

Fig. 2
verified as being 40kHz as required. The calculations
will not always give very accurate results for resistor
VI. LAYOUT ISSUES
values and these always need to be verified. Trim the
Rfmin value to obtain the correct result.
When laying out the PCB for this type of ballast it is
important to allow the high current carrying traces to
Note 3
be as wide and as short as possible. The control IC
The next stage is to set the dimming control voltage to
COM should be a star point connected to the COM
0V and adjust the value of Rmin to give the required
end of R12. The IC COM should be connected directly
minimum dimming level.
to this point via a very short track. The negative side
of C2 should also be connected as close to this point
6 www.irf.com
 AN-1038

as possible and the positive side should be connected In this design we have chosen the turns ratio of the
as close to the center tap of T1 as possible. The C9 transformer to give 300Vp-p at the secondary, which
decoupling capacitor should be connected directly can be fed into the ballast resonant output circuit. The
between VCC and COM and C3, C4, C5, R4, R5, R6, turns ratio required can be determined as follows :
R7, C6, C7, C8 should all be connected back to the
star point. Tracks around the IC should be kept short 300 / 120 = 2.5
as far as possible except the gate drives and VS and
VB, which can be a little longer if necessary. It is also 2 x 2.5 = 5
important to keep traces that are carrying high switch-
ing currents away from sensitive components around Therefore the turns ratio will be 1 + 1 : 5.
the IC as much as possible.
The core size should be selected for a throughput
VII. STEP-UP TRANSFORMER DESIGN power of 36W at 40kHz. We have used an EF25 (E25/
13/7) core of 3C85 or N27 material, which ungapped
The oscilloscope traces in fig. 3 show the voltage at has an Al value of 1900nH and an effective area Ae of
the drain of each of the switching MOSFETS. The drain 52mm².
voltage rises to 60V when the MOSFET switches off.
This is because the primary winding is center tapped Primary Volt-Seconds = 60V x 12.5uS = 750V-uS
and the center point is connected to the 30V DC bus.
When one MOSFET is switched on the voltage be- We have chosen 25 + 25 : 125 turns.
tween the center point and the drain is 30V therefore
the voltage across the entire primary winding will swing This gives a primary inductance of 50² x 1900nH =
from 60V in one direction to 60V in the other direction 4.75mH.
the result being 120Vp-p. Therefore the magnetizing current will be 750 x 10-6 /
4.75 x 10-3 = 0.16A (from V=L.dI/dT).
Voltage across the primary winding (from drain to drain)
The peak flux will be
N P AL I PK
Ae
= 50 x 1900 x 10-9 x 0.16 / 52 x 10-6 = 0.29T
(2900 Gauss).

This calculation shows that the core is being pushed

close to saturation in each direction but will not satu-
rate at high temperature (see manufacturers B-H curve
for the Ferrite material).
The winding wire sizes should be chosen such that
they fill the winding area. The primary should have
approximately twice the diameter of the secondary as
the primary RMS current will be 1.25A and the sec-
ondary RMS current will be 0.25A.
Fig. 3
www.irf.com 7
AN-1038

Voltage across the secondary winding

A current density 3A/mm² can be used to calculate
to minimum cross sectional area of conductor that
will be required. In this case the result is 0.12mm².

The skin effect must now be considered. For Copper

conductors the penetration depth at a given frequency
can be calculated by the formula
65
∆= (mm)
f
Using the maximum frequency of 70kHz this gives
the result 0.24mm therefore the strands should be
less than 0.24mm diameter.

The conductor area for a wire of 0.24mm is

Fig. 4 4D 2
A=
π
VIII. OUTPUT INDUCTOR DESIGN
This gives the result 0.073mm².
The BDA software will design the output inductor if
required. It will suggest a wire diameter for a single A practical solution would be to use 4 strands wire
strand, however a multi stranded wire that has an that has a diameter much smaller than 0.24mm. The
equivalent total cross sectional area will produce lower area for each strand would have to be 0.03mm² this
copper losses. equates to AWG 32, which has an area of 0.046mm²
including the insulation.
Alternatively the following procedure may be used :
2. Select the core size
1. Select the winding wire
The BDA uses an iterative process that attempts the
In a dimming design because the frequency goes as design procedure using a range of core and gap sizes
high as 70kHz it is necessary to use multi stranded finally selecting the smallest size that can contain the
wire in order to keep losses due to the skin effect to a winding wire without saturating during lamp ignition.
minimum. If single stranded wire is used the inductor This is extremely important because if the core does
will run at an increased temperature when the lamp saturate the resulting current pulse will be detected at
output is low. the CS pin of the IC causing the ballast to shut down.
A common design error is to fail to allow for a hot re-
Consider the RMS running current of the lamp. This strike condition (i.e. when the ballast has been run-
can be easily estimated by dividing the maximum lamp ning and is switched off and back on again) where the
power by the RMS lamp voltage. The RMS lamp Ferrite core is already at increased temperature and
voltage can be approximated by dividing the peak the saturation point of the material is reduced result-
lamp voltage by v2 in this case 100V giving 0.36A. ing in saturation at a lower current.

8 www.irf.com
 AN-1038

To follow the procedure of the BDA by hand is time possible to minimize copper losses when the lamp is
consuming and therefore it is easier to pick an option running. The BDA does this automatically.
based on experience. For a 36W ballast a reasonable
starting point would be to design an inductor based IX. BILL OF MATERIALS
on an EF25 (E25/13/7) core with a standard gap size Description Reference
of 1mm made of standard power grade Ferrite (type
3C85 or N27). Power MOSFET Q1,2
Ballast Control I.C. IC1
For this the Al value is 63nH and Ae is 52mm². The Fuse 2A F1
inductance required is 1.6mH therefore Capacitor 1uF 50V 105ºC C1
Radial Electrolytic
L Capacitor 220uF 50V 105ºC C2
N= Radial Electrolytic
AL Capacitor 100nF 50V C3,8
The number of turns required is 159. Capacitor 10nF 50V C4
Capacitor 330nF 50V C5
Capacitor 470nF 50V C6
The maximum ignition current is 2A so the peak flux
Capacitor 470pF 50V C7
density will be
NAL I PK Capacitor 100uF 25V 125ºC C9
BMAX = Radial Electrolytic
Ae Capacitor 100nF 400V C10,11
Polyester
Which gives the result 0.39T (3900 Gauss). By look- Capacitor 6.8nF 1500V C12
ing at the manufacturers curve of B against H we can Polypropylene
see that the material will saturate at around 0.42T at Resistor 24K 0.25W R1
25ºC and 0.35T at 100ºC. When the ballast is cold Resistor 5K6 0.25W R2
there is no possibility of saturation at ignition and dur- Resistor 10K 0.25W R3
ing a hot re-strike situation the core is unlikely to be Resistor 12K 0.25W R4
as hot as 100ºC. Therefore this solution is acceptable Resistor 27K 0.25W R5,7
as in reality the ignition voltage of the lamp will be Resistor 36K 0.25W R6
Resistor 1K 0.5W R8
somewhat less than 2A if the lamp is correctly pre-
Resistor 2M2 0.25W R9
heated. The inductor should be built and tested under
Resistor 680K 0.25W R10
worst case conditions to ensure that the lamp will strike. Resistor 1K 0.25W R11
If there are problems then a larger gap or larger core Resistor 0R15 0.25W R12
will be required. Resistor 18R 0.25W R13,14
Resistor 1K5 0.25W R15
The available winding area in an EF25 bobbin is Resistor 22K 1W R16
56mm². The winding area required is Filter Inductor L1
Inductor 1.6mH EF25 L2
0.046 x 4 x 159 = 29.3mm² Transformer 1+1:5 EF25 T1

Allowing for empty spaces there will be sufficient wind- X. REFERENCES

ing space. It is always an advantage to increase to
Lighting Ballast Control IC Designer’s Manual 2001 -
wire size or better still add more strands as much as
International Rectifier
11/22/04
www.irf.com 9