Constant current paralling controller for mid-power LEDs

Heidinger, Michael, KIT, Germany, michael.heidinger@kit.eduSimon, Christoph, KIT, Germany, christoph.simon@kit.eduDenk, Fabian, KIT, Germany, fabian.denk@kit.eduKling, Rainer, KIT, Germany, rainer.kling@kit.eduHeering, Wolfgang, KIT, Germany, wolfgang.heering@kit.edu

Abstract

A simple, cost-effective current sharing controllerintegrated in a LED module is presented. By thistechnology it is possible to drive high luminous flux,SELV-compliant LED arrays, using a large numberof mid-power LEDs arranged in a series-parallelconfiguration. Still a conventional constant currentpower supply can be used.

This novel current sharing concept uses one dis-sipative, cost-effective, balancing current sink perstring, implemented by using a single operationalamplifier, and one global averaging circuit. The doc-ument describes the circuit and corresponding de-sign equations. Operation is demonstrated by mea-surements.

1. Introduction

The nominal output power of LED floodlights is ris-ing with recent development. For optimal perfor-mance, LED strings connected in parallel shouldbe driven with matched current [3]. Currently par-alleling is avoided by series-wiring all LEDs [4]. Butthat approach has safety limits: SELV limits the to-tal DC voltage to 120 V [5]. To increase safety andlower touch protection requirements, some LEDlamp manufacturers apply the lower SELV limit of60 V. Assuming 3 V forward voltage per LED, themaximum number of standard series LEDs is hencelimited to 20. Another industry trend in LED floodlighting is to use mid-power LEDs [6], instead ofhigh power LEDs, which tend to glare and are lesscost-effective.

As the luminance per area of mid-power LEDs islower, they promise to meet anti-glare requirementswithout using lenses. Omitting the lenses reducescost, increases efficacy and avoids yet another fail-

ure mechanism. To achieve high luminous flux withhundreds of mid-power LEDs, current single chan-nel LED drivers require high output voltages andare therefore not SELV compliant.

In this paper a simple, safe SELV-compliant andcost-effective solution for paralleling LEDs is pro-posed, which allows to use a standard, single chan-nel, constant-current LED driver to parallel connectLED strings.

2. State of the art

The state-of-the-art approach is to separate theLED driver, into input and output stage, shown inFig. 1. The input stage converts mains AC to a fixedDC output. The output stage is responsible for pro-viding the appropriate currents to each LED string.

~

LEDCurrrentSharing

Controller

n=1

n=2

Grid

ACto

DC

Fig. 1: Typical generic LED driver topology for paralleldriving LEDs.

2.1. Resistors

The resistor method paralleling approach in Fig. 2is widely used, e.g. in LED stripes. This parallelingapproach can compensate the LED string currents,however LED string currents can only be compen-sated, but not matched. To achieve a decent andreliable equalization high resistances must be used.However using a high resistance value introduceslosses, lowering the modules efficacy.

U in

R R

Fig. 2: Paralleling LEDs can be achieved in the sim-plest manner by using resistors. While being simple andstraight forward, that solution has high electric losses.

2.2. Buck Converter

Another option for paralleling is by using a regulatedbuck converter for each string. In case one or sev-eral LED fail to short circuit, the buck converter cancompensate for the lower string voltage. Using thisapproach the LEDs do not need to be matched byforward voltage – thus the driver is more versatile.This flexibility comes at increased cost – especiallywhen tens or hundreds of parallel channels are re-quired.

To replace a 2.8 A high power LED by 120 mAmid-power LED, 24 parallel strings are required.That’s why the state-of-the-art constant current ap-proach is highly costly and impracticable for mid-power LEDs. Additionally this approach is only pos-sible if a higher entity, like a microcontroller, definesthe desired forward current.

3. Self averaging topology

3.1. Overview

A new averaging regulation topology targeted forlow cost is proposed in Fig. 3. The schematic isshown for two LED strings, but may be extended ton ∈ N. The paralleling controller aims to distribute aconstant current source equally to the LED strings.Therefore it measures all string currents, averagesthem, and provides that value as a reference to thedissipative constant-current sink of each string.

If an LED fails to short circuit, the string current

Rs

R f R f

CfCf

M1

Rs

R f R f

CfCf

M1

Ra

Ra

Ca

Rb

Rstartup

ConstantCurrent #1

Startup Circuit

Averaging Circuit

Q1

Q2

D1

LEDs

ConstantCurrent #2

LEDs

Fig. 3: Proposed novel current matching topology, basedon a dissipative current limiter with self-determining setcurrent. For startup an additional startup circuit is re-quired.

does not change, as the additional voltage is dis-sipated in the constant current sink. The averagecurrent will also not change.

This simple paralleling approach is easy to use anddoes not require a predefined reference value bythe circuit or from a microcontroller. Hence it canfollow the linear dimming of the primary constantcurrent source while achieving good equalization ofthe led string currents at any operating point.

3.2. Constant current regulation

Formula derivation

The main constant current functionality is imple-mented using an operational amplifier combinedwith a mosfet, as shown in Fig. 4. The operationalamplifier adjusts URS that is equal to UIset.

UIset = RSILED (1)

The second order low pass filter reduces the gainat high frequencies. The additional gain GCS of themosfet and sense resistor can be calculated to:

GCS = gM1RS (2)

For a simplified calculation it is assumed that theadditional gain is frequency independent. For astable operation the following equation must be ful-filled.

GCSGOPV(f)Gfilter(f) < 1 (3)

The additional required damping of Gfilter at highfrequencies is done by using a second order lowpass filter, however also lower and higher orderscould be used. The second order low pass fil-ter Gfilter has 40dB attenuation per decade. Thecompensated constant current source is shown inFig. 4.

UIset

Rs

R f R f

CfCf

M1

Fig. 4: Compensated constant current source, in this ex-ample using a second order low pass filter.

The cutoff frequency fB of the second order lowpass is calculated, so that the systems gain at theoperational amplifiers GBW frequency is less then1.

fB = GBW

(10

− GCS[dB]

GFilter[dB]

)(4)

The cutoff frequency fB of a coupled second orderlow pass filter can be calculated using the follow-ing formula. By this, the values of Rf and Cf arechosen.

fB =1

4π√2RfCf

(5)

The mosfet M1 switches only in dissipative modeif the individual string current is above the averagecurrent, to maintain a safe current. However, when

all LED strings have the same forward voltage, neg-ligible conduction loss by the current measurementand by the mosfet can be observed.

Constant current filter design example

For M1 the NTTFS4930 is chosen. From the data-sheet a forward transconductance of S = 19A

V isdenoted. We determine the sense resistor RS = 1Ωand calculate the gain.

G = RSS = 19 (6)

The amplification is converted into log-scale:

20 log(G) = 26 dB (7)

To determine the number of decades of dampingthe system gain is set in relation to the used secondorder low pass filter, with an attenuation of 40dB perdecade

26 dB

40 dB= 0.65 dec (8)

As an operational amplifier we choose LM324. Fromthe datasheet of the LM324 we know, that the Gain-Bandwidth-Product (GBP) is 1MHz. So we can cal-culate the boarder frequency for our damping net-work.

fB = 1MHz 10−0.65 = 223 kHz (9)

The frequency for boarder frequency for the con-verter can be calculated using (5). Choosing Cf =1nF, Rf = 281Ω. For a practical value Rf = 330Ωis chosen.

3.3. Averaging Circuit

Conventionally a set current, represented by UIset,is determined by a higher entity. This approach hastwo drawbacks: First, a higher instance is requiredto determine the current. Second, a connection isneeded for transporting this information.

Therefore a new method is proposed to determinethe set current. Each LED-String current is mea-sured at RSx and produces the voltage URx. Thesecurrents are averaged over the samples and filteredto determine the UIset voltage. The filter cutoff fre-quency ff of the averaging circuit should be chosento be lower than the cutoff frequency fB to preventoscillation, recommended is a factor of 5 to 10.

The filter cutoff frequency can be tuned with the ca-pacitor Ca. It can be calculated with the followingformula, where n is the number of parallel strings.

Ca =n

2πffRa(10)

Fig. 5 shows the averaging circuit. The circuitshown is implemented in this setup as first orderaveraging circuit, but a higher order is also possi-ble. The calculated average current is then used asset value UIset.

Note that the AC/DC power supply from Fig. 1 mustoperate in constant current mode, to determine theto led module current.

Ra

Ca

RaU

S1

US2

Uavg

Fig. 5: Averaging circuit determining the set current UIset

using the individual sense voltages.

3.4. Startup Circuit

As Uavg is fed back to UIset, the circuit has two sta-ble operating points. The initial operating point isat 0 A and the second at equally shared current.The startup circuits task is to force a slightly highercurrent than that the averaging circuit determines.This is done by adding a voltage offset, which isproduced by a current source. The required currentcan be calculated to:

Istartup =UOffsetn

RA(11)

In case an increased mismatch current should beaccepted by the circuit the following formula can beused.

Istartup =RSImismatchn

RA(12)

To generate the startup current, two possible solu-tions are analyzed, a dual transistor approach anda zener diode approach.

Transistor approach

The transistor approach is shown in Fig. 6. Thisapproach uses two PNP transistors. The followingformula can be used to calculate the startup resis-tor:

Rstartup =UBE

Istartup(13)

The bias resistor Rb is chosen to provide always aminimum operating current for the startup circuit.

Rb

Rstartup

Istartup

Fig. 6: Transistor constant current source, using twotransistors. This constant current approach is known tobe temperature sensitive.

This approach is highly temperature sensitive, asthe bipolar transistors base emitter voltage UBE

changes with temperature.

Zener / Transistor approach

As LED lamps generate heat, it’s important to havean improved temperature stability of the startup cir-cuit. The zener constant current approach, shown

in Fig. 7, is designed for advanced temperature sta-bility.

Q1 and D1 produce a voltage offset, where Rb isused to bias Q1 and D1. Q2 mirrors the voltage ofonly D1, which translates the voltage to a current.As the transistor Q2 has a fairly high amplification,the following equation holds true:

ISRC ≈ IRstartup (14)

A bipolar transistor has a temperature coefficientof −2 mV

K [1]. By subtracting this temperature co-efficient again, the transistors temperature drift iscompensated. As the dual transistors are fabricatedon the same die, both temperature coefficients andtemperature are likely to be matched. The addi-tional required zener diode should ideally have notemperature variation, thus a 5.6V zener diode [2] isideal, as temperature drift is centered around 0 mV

K .

For the design the following equations hold true:

Rsrc =UD1

Isrc(15)

Rb =Usupply − UQ1 − UD1

Ibias(16)

Ibias is the minimum cathode current of the zenerdiode specified by the datasheet.

Rb

Rstartup

Isrc

Q1

Q2

D1

Fig. 7: Zener constant current source, which is temper-ature compensated using matched Q1 and Q2 same dietransistors.

Comparison

The following simulation results of Fig. 8 comparethe two solutions. It can be observed that thedual transistor solution has a significant tempera-ture drift, while the zener transistor solution offersdecent temperature stability. Therefore the zenertransistor solution is favored in the complete designand verified in the section 4.6.

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

-60 -40 -20 0 20 40 60 80 100 120

Sourc

e C

urr

ent

(mA

)

Temperature (°C)

Constant current temperature dependence

ApproachDual transistor

Zener transistor

Fig. 8: Temperature dependence of the two constant cur-rent sources (spice simulation). It can be observed thatthe zener transistor solution offers improved temperaturestability.

4. Measurements & Validation

The design and its equations are validated by a pro-totype to prove the proper operational behavior ofthe circuit.

4.1. Measurement conditions

The measurements where done under the condi-tions stated in tab. 1, unless not otherwise noted.

The used measurement equipment is stated in table2.

4.2. Current Distribution

The current distribution over one module is depictedin Fig. 9. It can be observed that the LED currentsare matched within a distribution of 7mA.

As a fairly low current sensing resistor is used tosave power (1Ω), the current distribution is withinallowable limits. The typical operational amplifiers

Parameter ValueNumber of parallel LED strings 24

Number of series LEDs 6LED forward current 0.12A

LED type used Lumiled 2835COp-amp LM324mosfet NTTFS4930Rsense 1Ω

Rf 330Ω

Cf 1nF

Ra 4700Ω

Ca 1µFDual transistor (Q1,Q2) BC857S

Zener diode MM3Z5V6Rb 150k

Rstartup 220kTamb 20°C

Tab. 1: Measurement conditions

Type Model numberMultimeter Pico Test M3500A

Power Supply Korad KA3005D

Tab. 2: Used measurement equipment for experiments.

offset (5mV) can be observed. Further a voltagedifference between different ground points can beobserved: Though a solid ground plane is used,potential differences of up to 3mV at the groundplane could be measured. Therefore strong atten-tion must be given to obtain a solid strong groundplane.

Also the startup circuit allows an additional mis-match current for the LEDs.

For a better matching of LED currents the senseresistors could be increased, compromising effi-ciency. Using lower offset operational amplifierscan reduce mismatch.

4.3. Voltage Drop over mosfet

The dropout voltage over the mosfet without thesense resistor voltage is depicted in Fig. 10. It canbe observed that approx 1/3 of the mosfet operatenearly free of losses. The better the LED strings arematched, the less power loss can be observed. The

0

2

4

6

8

10

123 124 125 126 127 128 129 130 131

m

ILED [mA]

Fig. 9: LED current distribution in the range of 123 mAto 131 mA.

average voltage drop in this sample was 105mV.

0

2

4

6

8

10

0 50 100 150 200 250 300

m

Mosfet drop voltage [mV]

Fig. 10: The voltage drop measured over the mosfetwithout the sense resistor voltage is depicted. It can beobserved that even though one batch is used, a signifi-cant difference in voltage drop can be observed.

4.4. Quiescent Current

As additional circuits are introduced, current for theregulation is required that intentionally should beused for lighting. The currents are listed in Tab.3. However the quiescent currents are fairly low(<4.4 mA) compared to 2.88 A total LED drive cur-rent. Therefore the efficiency loss by quiescent cur-rents can be neglected.

4.5. Efficiency

The table 4 depicts the losses. The sensing resis-tors are the most significant loss source followed bythe equalization losses.

Type Quiescent current Total1/4 LM324 175 µA 4.2 mA

Startup circuit 140 µA 0.14 mATotal 4.4 mA

Tab. 3: Quiescent current of each individual sections andthe sum of all loss sources.

Loss source ValueQuiescent losses 90 mWEqualizing losses 302 mWSensing losses 345 mW

Total 750 mW

Tab. 4: Losses calculation of each functional circuits sec-tion. The sensing resistors are the most significant losssource followed by the equalization losses.

4.6. Constant Current Source temperaturedrift

The constant current source for the startup circuitdrifts by heat generated by the LEDs. For a stableperformance, the constant current source shouldhave a minimal sensitivity to the LEDs temperature.Fig. 11 demonstrates the temperature dependencyof the constant current source ISRC. A typical driftis 2.5% over 60 K, which equals to a temperaturedrift of 417 ppm/K.

0.995

1

1.005

1.01

1.015

1.02

1.025

20 30 40 50 60 70 80

I(T) src

/I(2

0°C

) src

T [°C]

Fig. 11: ISRC temperature drift of the dual transistor tem-perature compensated constant current startup circuit.(Section 4.6).

4.7. Reliability

So called catastrophic failures of LEDs are rare [7],and if so, typically a short circuit can be observed.

When a short circuit failure occurs the additionalvoltage drop is taken by the mosfet, demonstratedin Fig. 12.

Fig. 12: A short circuit of a LED is compensated withoutaffecting the LED drive current of the other LEDs.

Open circuit failure is not covered by the presentedcircuit diagram in Fig. 3, as it’s not likely. However,if open circuit failure should be covered additionalzener diodes anti-parallel to the LEDs may be used.

5. Design considerations

5.1. PCB Layout

LED PCBs are typically fabricated using single sidedaluminum PCBs, therefore a layout can be quitecomplex. A solid ground plane is strongly advisedto avoid different ground potentials. Even 3 mV dif-ference can create a current mismatch in the mArange.

Also it’s strongly recommended to add overvoltageprotection to the system and the mosfet gate. Thiscan be done using TVS or zener diodes.

5.2. Glare requirement

To reduce glare the mid-power LEDs should be po-sitioned close to another, a distance of less then5mm is recommended.

5.3. Legal

The legal patent situation should be checked in theindividual countries for applied patents, especiallyfor Germany.

6. Results

The LED driver can be integrated on a single sidedaluminum PCB. It uses minimal board space, asshown in Fig. 13. A minimal PCB space require-ment of only AEQ = 110mm2 per series LED chan-nel has been demonstrated.

Fig. 13: Aluminum LED pcb board with integrated cur-rent sharing controller. The control section is highlightedin green.

A cost of approx. 20ct per series LED channel couldbe demonstrated, compared to 60ct of a buck con-verter solution. If the LED strings are constructedequally, minor losses can be observed.

Therefore the dissipative current sharing solution issuperior in terms of simplicity, size, cost effective-ness, efficacy and efficiency compared to existingsolutions.

7. Summery

In this document an easy paralleling approach forLEDs is demonstrated. In normal operation, thecircuit consumes minimal power and balances dif-ferences between the individual LED strings accu-rately within ±2.6%. In case of an LED short cir-cuit failure the remaining LEDs are protected, asthe mosfet of the constant current sink will takethe additional voltage drop. The losses of the cur-rent sharing circuit are measured to be minimal andnarrow current distribution among the parallel LEDstrings are observed.

As LED produce heat, a temperature stable startupcircuit was proposed, allowing tight tolerances.

By this technology high luminance LED arrays withhundreds or thousands of mid-power LEDs can beconstructed cost effectively and conventional safetyextra low voltage (SELV) LED drivers may still beused.

As mid-power LEDs are typically far more efficient,a significant amount of energy can be saved. Byusing mid-power LEDs, thermal and glare require-ments are easily fulfilled.

8. References

[1] BC857 Datasheet NXP.

[2] Zener Diode BZX84 datasheet.

[3] APPLICATION NOTE 3256. Why drive whiteleds with constant current? Technical report,Maxim Integrated, 2004.

[4] Avago. Driving high power, high brightness leds.Technical report, Avago Technologies, 2010.

[5] Schroeder Hoerman. Schutz gegen elek-trischen Schlag in Niederspannungsanlagen.VDE-Verlag, Berlin, 2010.

[6] LED inside. Low and mid-power led chip be-comes major trend in the industry. Technical re-port, LED inside, 2015.

[7] Lumleds. Evaluating the lifetime behavior ofledsystems: The path to a sustainable luminairebusiness model. Technical report, Lumieds,2016.