Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Design of a Resonant Power Source to Drive HPS Lamps Christian Brañas, Francisco J. Azcondo, Salvador Bracho Dept. of Electronic Technology, System Engineering and Automation University of Cantabria Ave. de los Castros (s/n) 39005. Santander, SPAIN {branasc, azcondof, brachos}@unican.es Acknowledgement The Spanish Ministry of Science and Technology supports this research under the project reference DPI 2001-1047. Keywords Electronic ballasts, Resonant power converters, Soft switching. Abstract The aim of this paper is to define a new criterion for the design of resonant inverters applied to the control of high-pressure sodium (HPS) lamps, which is extended to metal halide (MH) lamps. Once the effect of the ageing on the lamp is determined, the analysis is focused on the selection of the most suitable value of the LCsCp resonant network in order to achieve a good repeatability of the circuit performance. The study is based on the analysis of the lamp power sensitivity regarding tolerance of circuit components. The proposed design is validated with experimental results and a statistical simulation by the Monte Carlo method. Introduction Nowadays, electronic ballasts for fluorescent lamps are gaining market share. However, since production cost is one of the most important issues in the success of a new product, the still high cost of electronic ballasts for HPS lamps limits their wide application [1]. The high voltage needed to achieve the lamp ignition, the lamp ageing and the acoustic resonance phenomenon increase the complexity and design effort. This paper is focused on the design of high-frequency resonant-inverterbased electronic ballasts that lead to circuit simplifications. The lamp ignition is achieved using the large voltage gain, presented by the inverter, as the switching frequency approaches the unloaded resonant frequency, avoiding the use of an auxiliary circuit. In steady state, the ballast must ensure that the power supplied to the lamp, Plamp, is maintained within the operational area defined by the trapezoidal diagram [2] of the lamp. The design sequence proposed in this paper minimises the Plamp variation, for an open loop operation at a constant switching frequency. Practical prototypes easily achieve a maximum Plamp deviation lower than 20% throughout the lamp lifetime. This characteristic simplifies the design, since no additional feedback control circuit is needed to stabilise Plamp. Additional constraints to the design sequence are introduced when considering the tolerance of the inductance and capacitor of the resonant tank. The analysis of sensitivities allows the definition of limits to the component tolerances to guarantee that a large enough number of circuits meet the quality specification. Since the acoustic spectrum shifts to lower values of frequency as the lamp size increases then, for medium to high power HPS lamps, it is possible to set the switching frequency above the highest frequency of acoustic arc resonance, ensuring an operation free of arc instabilities. In [3] the acoustic resonant phenomenon is measured in MH lamps, showing that for lamps over 100W at frequencies higher than 50kHz the instabilities seem to be damped. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.1 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Output Power Sensitivity of the LCsCp Resonant Inverter The class D LCsCp inverter is shown in Fig. 1. Some considerations have been assumed in order to make the circuit analysis easier. First, the circuit analysis is based on the fundamental approximation, i.e. taking into account only the first harmonic of the square voltage vAB. Also, the capacitor CZ is neglected during the circuit analysis. This capacitor is short-circuited during most of the switching cycle, thus its effect on the input impedance of the resonant circuit can be assumed to be negligible. Cp Cp M1 + Vdc + v1 - L Cs Rlamp A M2 + v2 + L + B vlamp ii CZ - A Cs - ii B ilamp Zi (a) - vlamp vAB1 ilamp Rlamp + (b) Fig. 1 (a): Class D LCsCp resonant inverter and (b) simplified circuit for analysis purposes. The LCsCp resonant circuit is correctly described by the following parallel parameters: Table I Parallel Resonant Frequency 1 ωp = LC p Characteristic Impedance Z p =ωpL = Parallel Quality Factor Rlamp Qp = Zp 1 ω pC p According to these assumptions, the module Zi, and phase φi, of the input impedance of the resonant circuit is calculated with (1) and (2). 2 Cp C + k − p Q p2 1 − k 2 + C s kC s Zi = Z p ⋅ 2 1 + k 2Q p φ i = Tan −1 k − ( C p 1 + k 2Q p kC s Qp 2 2 , (1) ) − Q k , p (2) where k is the normalised switching frequency, k=ω/ωp. The resonant frequency of the loaded LCsCp network dependent on the parallel quality factor is expressed in (3) kr = C 1 + p − 1 + 2 Cs Q p 2 4C p C 1 + p − 1 + 2 2 C Q Q s p pCs 2 . (3) As is well known, at any switching frequency above resonance (k >kr), the transistors of the inverter section turn on at zero-voltage (ZVS). The ZVS mode at turn-on is preferred because fast recovery diodes are not needed [4], even when there are switching losses during the turn-off transition, which are reduced by CZ. Making Qp→• in (3), the unloaded resonant frequency is obtained as k r ( off ) = 1 + Cp . (4) Cs The input current to the resonant circuit is obtained from (5). 2Vdc ⋅ Iˆi = π ⋅Zp EPE 2003 - Toulouse 1 + k 2Q p Cp Q p2 1 − k 2 + Cs (5) 2 2 C + k − p kC s ISBN : 90-75815-07-7 2 P.2 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian The maximum value of input current to the resonant circuit is the maximum current stress of transistor M1 and M2. On the other hand, the power supplied to the lamp, Plamp, is given by (6) Plamp = 2V dc2 π Zp 2 ⋅ . Qp 2 Cp C +k − p Q p2 1 − k 2 + C kC s s (6) 2 Using (6) and making use of the sensitivity definition, given in (7), it is possible to obtain the sensitivity of Plamp regarding the different circuit elements. S xP x= x o = ∂P ( x ) xo ⋅ ∂x P ( xo ) (7) Sensitivity of Plamp vs. Rlamp The lamp equivalent resistance, Rlamp, is the most variable element in a ballast circuit, increasing to up to 200% of the nominal value of a new lamp throughout the lamp lifetime. For this reason, the analysis of the Plamp sensitivity regarding Rlamp is of interest. It is calculated with (8), which is represented in fig.2. P lamp S R lamp Cp Q p2 1 − k 2 + Cs =− C p Q p2 1 − k 2 + Cs 2 C −k − p kC s 2 C +k − p kC s 2 2 (8) P lamp S Rlamp 1 Qp=0.2 0.5 Qp=0.5 0 ∆S -0.5 Qp=1.5 -1 0 0.2 0.4 0.6 0.8 1 1.2 1.4 k Fig. 2: Sensitivity of Plamp regarding Rlamp for three values of quality factor Qp and Cp/Cs=0.1. The lamp ageing leads to an increment of Qp, which produces the sensitivity variation indicated in fig.2. The values of Qp that make the sensitivity of Plamp regarding Rlamp equal to zero are relevant: Cp Q PM = ± k − kC s C 1 − k 2 + p . C s (9) At Qp=QPM, the maximum value of Plamp is reached, and it has no dependence on Rlamp, PM = ± Vdc2 . (10) C p C 1 − k 2 + p π 2 Z p k − kC s C s In order to achieve a straightforward implementation of the soft start-up sequence [5], only the frequency range k<kr(off) is of interest. From (10), note that for k≠kr(off), the Plamp vs. Qp curves have a maximum value. This feature may allow us to maintain the power supplied to the lamp within the operation area defined by the trapezoidal diagram during the whole lamp life. Accordingly, the curve Plamp vs. Qp for the proposed LCsCp inverter is designed to meet the characteristic depicted in Fig.3. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.3 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Plamp(W) Point of Maximum Power Operation Area Plamp(Max) PM ∆Plamp Plamp(N) Lamp at the end of its life time New Lamp Ballast Curve Plamp(Min) Qp(Min) Qp(N) QPM Qp(Max) Qp Fig. 3: Proposed ballast curve. The aim of the design is to achieve that the maximum deviation of Plamp, regarding the lamp ageing, fulfils the trapezoid limit. The power increment due to the ageing process, ∆Plamp (see Fig.3), is normalised in (11) with the nominal power supplied to the new lamp, Plamp(N), and the resulting parameter is defined as the power regulation index. For the proposed ballast curve, this index defines the amplitude of the maximum power deviation and thus how much the inverter approximates to a power source. The power regulation index is calculated as follows: 2 ∆Plamp Plamp ( N ) = PM − Plamp ( N ) Plamp ( N ) Cp C + k − p Q p2 ( N ) 1 − k 2 + C kC s s =± C p Cp 2 1 − k + Q p 2 k − kC s C s 2 −1 . (11) The value of Qp(N) is calculated to adjust the initial Vlamp. Eq. (12) calculates Qp(N) as a function of the normalised switching frequency, k. Cp Q p ( N ) = k − kC s 2V dc π 2V lamp 2 2 − 1 − k 2 + C p C s (12) Working with (12), (11) and (10), expression (13) is obtained, where the resulting value of k defines the on-steady-state switching frequency, kωp. C 2Vdc 1 − k 2 + p = C s π 2Vlamp Π (13) where Π becomes a design parameter calculated from the power regulation index ∆Plamp/Plamp(N). 2 Π= ∆Plamp 2 ∆Plamp + Plamp ( N ) Plamp ( N ) . 1 − 2 ∆Plamp 2 + 1 Plamp ( N ) (14) Sensitivity of Plamp vs. L In the circuit under study, for the frequency range of interest k<kr(off), the highest sensitivity corresponds to the inductance variation [5]. P S Llamp Cp 2 Cp − 2 k − 2 k 2 Q p2 1 − k 2 + C s Cs = 2 2 Cp C + k − p Q p2 1 − k 2 + C s kC s (15) Expression (15) is depicted in the graph of fig.4. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.4 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian P S L lamp 1 0 -1 ∆S Qp=1.5 Qp=0.5 -2 Qp=0.2 -3 0 0.2 0.4 0.6 0.8 1 1.2 1.4 k Fig. 4: Sensitivity of Plamp regarding L for three values of quality factor Qp and Cp/Cs=0.1. From fig. 4, the highest values of sensitivity of Plamp regarding L takes place for k>kr(off). Note that for k<kr(off), the sensitivity is greatly reduce as Qp increases. Thus, the worst case of sensitivity takes place at the beginning of the lamp lifetime. Therefore, to achieve a good yield in industrial production using a circuit without a feedback control loop, the sensitivity of the design with regard to the inductance tolerance must be incorporated as a design input parameter. Load range and power balance Due to the effect of the high variability of the HPS lamp upon Plamp and the switching mode, the load range of the inverter should be defined in the first step of the design process. The maximum value allowed for Rlamp is defined by Qp(Max) (see fig. 3). A larger value of Rlamp will lead to the loss of the ZVS mode. Qp(Max), is Q p ( Max ) = 1 k 2 Cp k − Cs C 1 − k 2 + p Cs . (16) Thus, the load range, where the ZVS mode is maintained, is calculated according to: ∆Q p Q p( N ) 2V dc = π 2Vlamp 1− Π2 ( ) 2V dc 1 − π 2Vlamp Π Π − 1 . (17) Finally, working with (16), (12) and (6), the condition such that Plamp is equal at the end and beginning of the lamp lifetime is obtained: 2V dc π 2V lamp ( ) 1+ 1− 4 1− Π 2 Π 2 . = 2Π (18) Expression (17) and (18) are depicted in fig. 5. 3 ∆Q p 2.5 Qp( N ) 2 1.5 2Vdc π 2Vlamp 1 0.5 0 0 0.05 0.1 0.15 0.2 0.25 ∆Plamp/Plamp(N) Fig. 5: Load range and voltage ratio as a function of the power regulation index EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.5 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Figure 7 shows the relationship between the load range and the supply to lamp voltage ratio. The load range increases with the voltage ratio. Higher voltage ratio increases the initial phase of the input impedance, φi, of the resonant circuit as can be deduced from (19). 2V dc π 2Vlamp φ i = Tan −1 − Π 1− Π2 . (19) The phase displacement between vAB and ii implies a penalty in terms of reactive energy in the resonant circuit. Fig. 6 shows the power balance at the input of the resonant tank. v AB V dc/2 ii Îi φi P lamp PA t Pr Îi V dc/2 I III II IV Fig.6: Power balance of the resonant inverter for ZVS mode. In figure 6, Pr is the power that returns to the Vdc supply from the resonant circuit. On the other hand, PA is the power drawn by the resonant circuit from Vdc. Considering a resonant circuit free of losses, the power supplied to the lamp can be calculated as Plamp=PA-Pr. The expressions are given below, PA = Vdc Iˆi [1 + Cosφ i ] , 2π (20) Pr = Vdc Iˆi [Cosφ i − 1] , 2π (21) Plamp = Vdc Iˆi Cos φ i π (22) . The power transfer index, T, defined as the initial Plamp to PA ratio is, T= Plamp PA = 2Cosφi . 1 + Cosφi (23) Expression (23) is depicted in figure 7 as a function of the power regulation index ∆Plamp/Plamp. T 1 0.95 0.9 0.85 0.8 0.75 0.7 0 0.05 0.1 0.15 0.2 ∆Plamp/Plamp(N) 0.25 0.3 Fig. 7: Power transfer index as a function of the power regulation index EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.6 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Using (23) and fig.7, it is possible to evaluate the amount of reactive energy in the resonant tank at the beginning of the lamp lifetime. The selection of the exact supply to lamp voltage ratio, as is derived from (18) and (23) implies that Cosφi=1 for Qp(Max) i.e. at the end of the lamp lifetime. We call this situation quasi-optimum mode [6], given that the amount of reactive energy in the resonant tank is just enough to assure ZVS until the end of the lamp lifetime. Design for manufacturing In section II, it was commented that the inductance is the element with highest value of sensitivity regarding Plamp. In order to find a design criterion to minimise the Plamp sensitivity regarding L, it is of interest to rewrite expression (15) as a function of the input design parameters. P S L lamp 2 2V dc 1− π 2Vlamp Π π 2V lamp 2Vdc Π − 1 + 2Vdc π 2Vlamp Π − 1 = C p Cs (24) Expression (24) is represented in fig.8. Cp/Cs 1 0.8 0.015 0.6 0.4 0.03 0.05 0.07 0.2 0.1 0.15 -2 -1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 P S Llamp Fig. 8: Capacitor ratio Cp/Cs as a function of the sensitivity of Plamp regarding L and ∆Plamp/Plamp as parameter. Using (24), this paper suggests calculating Cp/Cs in accordance with the value of sensitivity of Plamp that ensures high design repeatability with the available inductance tolerance ∆L/L. The proposed design sequence is as follows: 1) The steady state electrical data for a new 250W HPS lamp at high frequency were experimentally obtained: Plamp=250W, Vlamp=100Vrms, Ilamp=2.5Arms and Rlamp=40Ω . A value of power regulation index of ∆Plamp/Plamp =0.16 results in Π=0.496. 2) For Π=0.496, using expression (18), the voltage ratio is 1.517, then Vdc=337V for quasi-optimum operation. Also, using (17), the load range is ∆Qp/Qp(N)=2.05, which means that the inverter accepts an increment of 200% in Rlamp. This load range is high enough to ensure the ZVS mode during the whole lamp lifetime. 3) The Plamp sensitivity with regard to ∆L/L was fixed at –1.4, considering a worst case of tolerance for the inductance of ∆L/L=±15%. From (24), the capacitor ratio is calculated: Cp/Cs=0.01. 4) The input current is Îi=3.59A. The phase of the input impedance to the resonant circuit is calculated from (19), φi=49.6º. This value of phase results in an initial power transfer index of T=0.78, which indicates that almost 80% of the apparent power is transferred to the lamp. 5) From expression (13), k is obtained: k=0.506. 6) From (12), the parallel quality factor for nominal conditions is calculated Qp(N)=0.37. 7) Knowing Qp(N), the characteristic impedance Zp is: Zp=Rlamp/Qp(N)=108.3Ω . 8) The switching frequency is fixed above the highest value of arc acoustic resonance. kωp=2π(150kHz), so ωp=2π(296.4kHz). 9) Finally, from table I, the reactive components are: L=Zp/ωp=58µH, Cp=1/ωpZp=4.95nF and Cs=495nF. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.7 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian The ballast curve of the designed LCsCp resonant inverter is shown in fig. 9. This curve was obtained using the parametric sweep together with the performance analysis of pSpice. From the simulation results, the maximum lamp power is PM=291.5W, which is reached for a value of lamp equivalent resistance of RPM=71Ω. Plamp 300W 290W (71Ω,291.5W) 280W 270W 260W (126.0Ω,254.7W) (40.6Ω,250.313W) 250W 240W 40Ω 50Ω 60Ω 70Ω 80Ω 90Ω MAXr(avg(-i(rlamp)*v(rlamp:2)), 150us, 200us) 100Ω 110Ω 120Ω 130Ω Rlamp Fig. 9: Simulated ballast curve. The design repeatability was evaluated using the Monte Carlo simulation. The Monte Carlo method was applied making use of the electrical simulator pSpice [7][8] to analyse the effect of the L, Cs and Cp tolerances in the circuit performance. The normal distribution was assumed for all values. A value of ±15% of tolerance was chosen for the inductance L and ±5% (J) for Cs and Cp. Figure 10 shows the histogram obtained for Plamp in the designed prototype after 400 cycles of Monte Carlo. P c t 10 o f S a m p l e s 5 0 160 n samples n divisions mean 180 = = = 200 400 30 249.417 220 240 260 MAXr(avg(i(rlamp)*i(rlamp)*40), sigma minimum 10th %ile = = = 17.8911 192.909 227.941 160us, 280 200us) median 90th %ile maximum 300 = = = 320 340 247.846 274.429 308.719 Fig. 10: Plamp histogram considering a tolerance in the value of L equal to 15% and 5% for Cs and Cp. In order to evaluate the result, the upper and lower specification limits (USL and LSL) were established according to the trapezoidal diagram for 250W HPS lamps. In this case, LSL=180W and USL=290W. The parameters of the resulting distribution are µ=249.41W and σ=17.89W. According to these values, a quality level equal to 3σ is expected, which means a very low defect rate [9]. Experimental results A practical implementation of the class D LCsCp inverter was carried out according to the proposed design sequence. The selected MOSFET transistor was the low-charge, fast-switching IRF840LC with the IR2111 circuit as driver. The practical values used in the experimental prototype were Vdc=340V, L=55µH, Cs=470nF and Cp=4.7nF. Fig.11 shows the experimental waveforms of the instantaneous lamp voltage, power and current for a new 250W LUCALOX lamp manufactured by General Electric, as well as for a more than 10,000hour old lamp of the same type. The ageing was estimated considering a voltage rate increment of 2V per 1000 operation hours. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.8 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian (b) (a) Fig. 11: Experimental waveforms measured in (a) new LUCALOX lamp and (b) old LUCALOX lamp. In both cases: Upper trace:Vlamp, Middle Trace: Plamp, Bottom Trace: Ilamp. For the new lamp, the measured Rlamp, is 39Ω. The measured lamp current crest factor is CF=1.33, below the maximum value, 1.7, recommended by [2]. For the old lamp, the power rises up to Plamp=288W, Rlamp being 67Ω. Note that for a variation of ∆Rlamp/Rlamp=72%, the measured increment of Plamp over the nominal value was only ∆Plamp/Plamp=15%, in accordance with the standard. The measured CF of the lamp current was CF=1.15. Metal Halide Lamps Since the design accepts a large variation of the load, its performance has also been verified with MH lamps. The deviation foreseen in MH lamps is from 2V to 4V per 1000 hours of operation. Considering 10,000 hours of operation, a Rlamp equal to 105Ω can be considered at the end of the lamp lifetime, which is in the range covered by this design. Two types of MH lamps were tested with an initial Rlamp=36Ω and 54Ω respectively. The experimental waveforms are shown in fig.12. (b) (a) Fig. 12: Experimental waveforms: (a) measured in a new HQI-T-250W Osram lamp. (b) measured in a new HSI-T-250W Sylvania lamp. Upper trace: Vlamp, Middle Trace: Plamp, Bottom Trace: Ilamp. The Sylvania HSI-T-250W lamp operating at 150kHz, free of acoustic resonance, is shown in the photograph of fig. 13. Fig. 13: Arc at 150kHz. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.9 Design of a Resonant Power Source to Drive HPS Lamps CHRISTIAN BRANAS REY Christian Turn-off switching losses The calculation of capacitor CZ is carried out assuming that its charging time is much shorter than the switching cycle. Thus, the capacitor CZ can be obtained from (25), tD ≥ Vdc C Z , Iˆi Sinφ i (25) where tD is the dead time of the driver signals of transistors M1 and M2. The driver IR2111, with a typical value of 600ns, imposes the dead time. Taking a safe margin in order to prevent short-circuit of Vdc through M1 and M2, a maximum time of charge for CZ of 100ns was assumed. Substituting in (25) the value of Cz is obtained, CZ=800pF. Finally, the experimental value used for CZ was 470pF. The waveforms obtained in transistor M1 are shown below. (a) (b) Fig.14: (a) Transistor waveforms vds and ids. (b) Switching losses. Conclusions A study of the sensitivity of the power supplied to a HPS lamp by LCsCp inverters and a new design sequence for this circuit has been presented. The most variable element in the ballast circuit is Rlamp. The design sequence limits the expected variation of Plamp regarding Rlamp below the standard restriction. The sensitivity analysis allows us to evaluate how critical the components of the design are, showing that the inductance L is the element whose variation produces the highest deviation in Plamp. The experimental results verify a low distortion of the lamp current, with CF<1.7. Furthermore, the design evaluation by the Monte Carlo method, in this case, predicts a good yield for industrial production of the circuit. References [1]- G. Trestman, "Minimizing Cost of HID Lamp Electronic Ballast", Proceedings of the 28th Ann. Conf. of the IEEE Industrial Electronics Society (IECON'02), Vol.2, pp.1214-1218. [2]- "High Pressure Sodium Vapour Lamps", European Standard EN60662, June 1990. [3]- J. Oslen and W. Moskowitz, "Optical Measurement of Acoustic Resonance Frequencies in HID Lamps", IEEE Industry Applications Society Conference, Oct. 1997. [4]- Marian K. Kazimerczuck, Wojciech Szaraniec, “Electronic Ballast for Fluorescent Lamps”, IEEE Transactions on Power Electronics, Vol. 8, No. 4, October 1993, pp. 386-395. [5]- Ch. Brañas, F. Azcondo, S. Bracho, “Study of Output Power Variation due to Component Tolerances in LCsCp Resonant Inverters Applied to HPS Lamp Control”, Proceedings of IECON'01, pp. 1021-1026. [6]- Ch. Brañas, F.J. Azcondo, S. Bracho, “Contributions to the Design and Control of LCsCp resonant Inverters to Drive High Power HPS Lamps”, IEEE Trans. on Ind. Elect., Vol. 47, No. 4, August 2000, pp. 796-808. [7]- Yiyoung Sun, “Using pSpice to Determine Lamp Current Variation Due to Electronic Ballast Component Tolerances”, IEEE Trans. on Industry Applications, Vol. 33, No. 1, pp. 252-256, January/February 1997. [8]- HSPICE User's Manual H9001, Meta-Software Inc. 1990. [9]- R.V. White, "Component Tolerance and Circuit performance: A case of study", Proceedings of the 8th Ann. Appl. Power Elec. Conf. (APEC'93), pp.922-927. EPE 2003 - Toulouse ISBN : 90-75815-07-7 P.10