On-line Calibration of Lossless Current Sensing

Transcription

1 On-line Calibration of Lole Current Sening Yang Zhang, egan Zane, Dragan Makimovi Colorado Power Eletroni Center ECE Department, Univerity of Colorado Boulder, CO {yyzhang, zane, Abtrat n thi paper, we deribe a lole urrent ening method with on-line alibration where an auxiliary with and a preiion reitor are onneted in parallel with a main power with to ahieve auray omparable to the reitor method, together with the advantage of eentially no additional power lo. The propoed urrent-ening iruit and the alibration method are partiularly well uited for digital ontroller implementation where the required ontrol and alibration funtion an be eaily aomplihed. Experimental reult with a digitally ontrolled 1.5, 15 A ynhronou buk onverter demontrate funtionality of the on-line alibration approah, howing a ignifiant improvement in auray over voltage ening aro the power MOSFET on-reitane. Keyword - urrent ening; alibration; withing power upplie. NTODUCTON Swithing power onverter require urrent ening for a ombination of reaon, inluding: (a) urrent-mode ontrol implementation, (b) urrent haring among paralleled module, () output voltage poitioning (a in voltage regulator module (M)), and (d) over-load protetion. n high-frequeny DC- DC withing onverter, the mot ommonly ued urrentening method are ummarized in Table [1-5]. TABLE. Method name in thi paper S ON L W SFET CT SUMMAY OF CUENT-SENSNG METHODS. Current-ening method Sening the voltage aro a erie reitor Sening the voltage aro the on reitane on of a power MOSFET Sening and filtering the voltage aro an indutor Sening the voltage aro the PCB wire trae having w reitane Uing a power MOSFET with built-in urrent-ening mirror ( SeneFet ) Uing a urrent tranformer in erie with a power with Alekandar Prodi ECE Department, Univerity of Toronto 10 King' College oad, Toronto, Canada, M5S 3G4 prodi@power.ele.utoronto.a The urrent-ening method differ in how they addre the trade-off among the oppoing requirement reulting from the following key onideration: 1. Power lo due to the enor 2. Auray with repet to parameter tolerane and temperature variation 3. Operating range and linearity 4. Bandwidth 5. Senitivity to withing noie 6. Complexity and ot of implementation The method of hoie uually depend on the peifi appliation requirement. For example, the S method i often preferred in power-fator-orretion or off-line appliation beaue of it auray and impliity. However, in lowvoltage, high-urrent point-of-load appliation (uh a M), the ON, L or W method are the mot ommonly applied beaue they do not indue any additional power loe, and do not require any additional omponent. But, thee lole urrent-ening method have relatively poor auray due to tolerane and temperature variation of the ening reitane ( on, L, or w ). Furthermore, beaue of the deire to maintain high effiienie even a the regulated output voltage are dereaing (approahing 1 or le), the voltage drop allowed aro the ening reitane beome lower, whih further adverely affet the urrent-ening auray and enitivity to noie. n thi paper, we propoe a lole urrent ening method with on-line alibration where an auxiliary with Q a and a reitor are onneted in parallel with a main power with Q a hown in Fig. 1, to ahieve ombined advantage of the aurate S method and the lole ening tehnique. n normal withing yle, the main with Q i turned on to ondut the urrent i w. Any of the lole urrent ening i w Q a Q Figure 1. Current-ening alibration iruit oniting of an auxiliary with Q a and a preiion ening reitor in parallel with a main power with Q. The auxiliary with i turned on infrequently, only to perform on-line alibration of a lole urrent ening method. Thi work wa upported by Arteyn Tehnologie through Colorado Power Eletroni Center /04/$17.00 (C) 2004 EEE 1345

2 method (uh a ON, L, or W ening), or enorle urrent etimation [6], an be performed ontinuouly. n a alibration yle, whih i performed only oaionally, the main with Q i kept off, while the auxiliary with Q a i turned on to ondut the urrent i w. The voltage aro the preiion urrent- reitor i d and ued to alibrate the lole urrent ening or the enorle urrent etimator. Sine the alibration yle i performed only infrequently, e.g. one in every 1000 normal yle, the auxiliary with an be rated at a lower urrent, and the additional lo aoiated with the auxiliary iruit i negligible. Depending on the appliation, the auxiliary iruit an be ued in parallel with any of the power withe, and the propoed approah an be applied in onjuntion with any of the lole urrent-ening method, or in onjuntion with a enorle urrent etimation. n thi paper we fou on appliation of the on-line alibrated urrent ening in low-voltage point-of-load d power upplie or M baed on a ynhronou buk onverter. n thi appliation, it i onvenient to add the auxiliary urrent-ening iruit aro the ynhronou retifier. The paper i organized a follow. Appliation of the propoed urrent-ening method in a low-voltage ynhronou buk onverter i deribed in Setion, together with everal alibration tehnique that an be implemented in the propoed method. t i noted that the method i partiularly well uited for digital ontroller implementation where the required ontrol and alibration funtion an be eaily aomplihed. Setion preent an experimental tet iruit ued to validate the propoed approah. Experimental reult that demontrate ignifiant improvement in auray over the ON method without penalty in the onverter effiieny are ummarized in Setion and.. CUENT-SENSNG CCUT WTH ON-LNE CALBATON Our approah i baed on ombining the S method a a alibration tage together with any of the lole urrent ening approahe. n thi etion, we deribe a ombination of the S and ON method, a hown in Fig. 2 for a ynhronou buk onverter with urrent ening aro the ynhronou retifier. During normal operation, the onverter run in the ON mode with Q 2 on over the time interval (1 D)T, while Q 3 i off. n the normal yle, the d voltage i ued to meaure the indutor urrent a: g Figure 2. Q 1 Q 2 L Q 3 C out Synhronou Buk onverter with urrent-ening iruit for on-line alibration - =, (1) on where on i initially etimated from the MOSFET dataheet (or ome other bet etimate). The ON approah provide the benefit of no additional omponent or power lo, but reult in poor auray due to the omponent tolerane and enitivity to temperature. We propoe to gain the auray benefit of a reitor while maintaining the ON benefit of no additional power lo by oaionally withing to the auxiliary MOSFET Q 3 for alibration, while the main ynhronou retifier Q 2 i kept off. n a alibration yle, the indutor urrent flow through Q 3 and the aurate reitor, reulting in a more preie urrent meaurement: alib =. (2) The alibration yle i performed infrequently a required for aurate alibration, allowing a relatively mall devie Q 3 to be ued. Here, we deribe three approahe for alibrating on baed on the onfiguration of Fig. 2. A. Bai alibration of urrent ening The mot traightforward appliation of (2) i to aume that the indutor urrent i not affeted by the alibration yle, reulting in a alibrated value of the on reitane given by: on alib =. (3) Then in following normal yle, thi onalib i ued in (1) intead of the nominal on to obtain a urrent ening auray omparable to the S method. Calibration an be performed infrequently, e.g. every thouand of yle, to maintain the benefit of lole ON ening, while keeping trak of low hange in operating ondition, uh a temperature. Note that the approah an alibrate out the effet of temperature variation without ening the temperature itelf, and an alo remove other unertaintie, uh a omponent tolerane and aging. The auray of the method depend on the reitor, and the aumption that the indutor urrent i not diturbed during the alibration yle. n the iruit of Fig. 2, although the auxiliary with Q 3 an have a lower urrent rating ompared to the main ynhronou retifier Q 2, the ombined on-reitane of Q 3 and mut be low enough o that the body diode of Q 2 i not turned on during alibration yle. Another pratial onern i that a alibration yle hould be performed only when the iruit i operating in teady tate, but not in tranient. B. Etimation of the error indued in during alibration The alibration yle itelf an indue error in the indutor urrent due to an inreae in the voltage drop aro Q 3 and ompared to the voltage drop aro the main ynhronou retifier Q 2. Thi reate a larger indutor voltage during alibration, reulting in an inreaed urrent lope a hown in 1346

3 dnormal /dt t d dalib /dt error alib D Normal yle Fig. 3. The indued error an be partiularly large in onverter (uh a M) with low output voltage and low indutane value. Thi error an be etimated and anelled auming that the indutane value L and the time delay t d for ampling are known: i L error on alib Calibrate yle ( = td = normal alib ) ( L i 1 Normal yle Figure 3. Error indued in the indutor urrent during alibration. yle ), 1, normal L error where normal and alib are the voltage at in Fig. 2 during normal and alibration yle, repetively. The additional ot of implementing (4) over (3) inlude ening and imultaneouly during alibration yle, and the added omputational requirement in (4). The approah i well uited for appliation with digital ontrol, where the additional proeing requirement an be provided at minimal ot. The primary drawbak of (4) i the dependene on the indutane value, whoe tolerane and variation may not fall within the requirement for preiion urrent limiting, preiion urrent haring, or other appliation that may require aurate abolute urrent ening. Additional auray an be gained by uing multiple ample during the alibration yle to meaure the indutor urrent lope and atively etimate the indutane value. When even greater auray i required, the ative anellation approah of the following etion an be ued. C. Ative anellation of error indued in The next improvement i to atively remove the error indued by the alibration yle from the meaurement by adjuting the duty yle. Thi approah an alo be ued to atively remove any other perturbation in the waveform reulting from the alibration yle. One ative anellation option i a 3-yle alibration equene. The firt yle i a preparation yle (Q 3 on, Q 2 off), the eond i a ompenation yle (Q 2 on, Q 3 off) and the third yle i the atual ening yle (Q 3 on, Q 2 off), a hown in Fig. 4. n the firt yle, alib i d to gain the knowledge of lope. The purpoe of the ompenation in the eond yle i to fore the urrent at the ening point of the 3 rd yle to reah the ame value that the indutor urrent ha in normal yle at t (4) Calibration yle Normal yle Calibration yle Preparing Compenating Sening Figure 4. Ative anellation of the error indued in the time the ampling i performed. The duty yle i inreaed by D in the 2 nd yle to ompenate for the urrent error indued in the 1 t and the 3 rd alibration yle: (( alib ) ( normal )) t d D = (1 D), (5) g T reulting in a meaured in the 3 rd yle that an be ued diretly in (3) to alibrate on. One key benefit of (5) i that it doe not depend on the indutane value, thu having the potential for failitating very aurate lole urrent ening. n Setion, and, an experimental iruit and experimental reult are preented for the alibration method baed on (3) and (4).. EXPEMENTAL TEST CCUT An experimental tet iruit wa built to validate the propoed urrent ening and alibration approahe. The tet iruit i hown in Fig. 5. The power tage i a ynhronou buk onverter operating at 100 khz withing frequeny. The input voltage an be 5 or 12. The output voltage an be 1.2, 1.5, or 1.8. The load urrent i between 0 A and 15 A. Two ynhronou-buk MOSFET driver (TPS2838) are ued to drive the main ontrol with Q 1, the main ynhronou retifier Q 2, and the auxiliary g PMW Buk Driver MainDrv Ctrl Normal Q 1 A/D Q 2 SynDrv ref PMW /2 /2 /2 Q 4 Q 5 CalibDrv /2 PMW Digital Controller Calib Buk Driver Ctrl Q 3 L ref L C A/D Figure 5. Experimental tet iruit for the on-line alibrated urrent ening method in a ynhronou Buk onverter. out t 1347

4 alibration retifier Q 3. The nominal on reitane for Q 2 i 2.9 mω (two Si4888DY MOSFET in parallel). The nominal indutane value = 3.0 µh and the output apaitane i C = 750 µf. The reitor i = 10 mω. An FPGA-baed digital ontroller i ued to generate the gate-drive ignal and to ontrol the urrent-ampling A/D onverter. Two operational amplifier (OPA350) and the urrounding iruitry perform differential ening of and, a well a level hifting of the d voltage by ref. The output of the operational amplifier and atify: = = ref ref,. The mall MOSFET withe Q 4 and Q 5 are ued to protet the ening operational amplifier from the high voltage (5 or 12 ) when the main ontrol MOSFET Q 1 i on. Compared with ening and diretly, ening and ha the following benefit: ha muh maller pulating omponent than ; and have an adjutable ommon-mode voltage ref, whih an be ued to fit the d analog ignal into the ommon mode input range of the analog to digital onverter; C filter an be added to filter out high-frequeny noie from and ; Differential amplifier rejet the ommon mode noie at the ground of the power tage.. EXPEMENTAL ESULTS For the experimental reult reported in thi paper, the ynhronou buk onverter operate at 100 khz withing frequeny, with 12 input voltage and 1.5 output voltage. Four load ae have been teted from no load to full load. n order to tet and diplay alternating normal and alibration yle eaily, the iruit i ontrolled to have one alibration yle every three normal yle. A onverion tart ignal i alo generated for A/D onverter to ample and hold and with approximately the ame time delay in eah yle. Fig. 6 how the three gate-drive ignal and the indutor urrent waveform when the iruit i operating at full load. The average value of the indutor urrent i 15.7 A (10 A/div). MainDrv, SynDrv and CalibDrv are the gate drive ignal for the main ontrol with Q 1, the main ynhronou retifier Q 2, and the auxiliary alibration retifier Q 3, repetively. Fig. 7 how the voltage,, the indutor urrent and the onverion tart ignal at full load. The large pulating omponent an be oberved in, whih goe to approximately 12 when the main ontrol with i on, and loe to zero when one of the ynhronou retifier (main or alibration) i on. n normal yle, when the main ynhronou retifier i on, the voltage aro the ynhronou retifier i negative, but very loe to zero, beaue of the very low on reitane of Q 2. Sine Q 3 i off during normal yle, i zero. During (6) MainDrv SynDrv CalibDrv Figure 6. Top-to-bottom: the gate drive ignal for the main ontrol with Q 1; gate drive ignal for the main ynhronou retifier Q 2; gate drive ignal for the auxiliary alibration retifier Q 3 and the indutor urrent at maximum load (15 A). Figure 7. Top-to-bottom: (Ch. 3), (Ch. 1), indutor urrent (Ch. 4) and onverion tart ignal (Ch. 2) at maximum load (15 A). alibration yle, Q 3 i on and Q 2 i off. A a reult, ha a more viible negative value beaue of the larger voltage drop aro the reitor. The voltage aro Q 3 and i alo larger than in normal yle. A mall diturbane produed by the alibration yle an be oberved in the indutor urrent waveform. Fig. 8 how the voltage waveform, and at the output of the differential ening amplifier, together with the indutor urrent and the onverion tart ignal at full load (15 A). Compared to the waveform of Fig. 7, ha a muh maller pulating omponent. During DT, when the main ontrol with Q 1 i on, i omewhat lower than ref beaue of the voltage drop aro the on-reitane of the mall protetion withe Q 4 and Q

5 TABLE. EXPEMENTAL ESULTS FO BASC CALBATON Load (W) Atual urrent (A) normal (m) Error (%) in (urrent d by ON method only) (m) Error (%) in alib (urrent d by on-line alibrated ON method) Figure 8. Top-to-bottom: (Ch.3), (Ch.1), the indutor urrent (Ch.4), and the onverion tart ignal (Ch.2) at maximum load (15 A). A. Experimental verifiation of the bai alibration of urrent ening The ening delay t d i meaured between the time a ynhronou retifier (main or alibration) i turned on and the time the riing edge of the onverion tart ignal ample the d ignal. n the experiment reported in thi etion, the ening delay i t d = 6.7 µ. At full load, in normal yle, we have the following reult for ening: normal = ref = 52m (7) f only the ON ening method i ued, without alibration, the nominal value of the Q 2 on reitane yield the following etimate for the d urrent: 52m = = = 17. A (8) 2.9mΩ 9 on n the alibration yle, we at the riing edge of the onverion tart ignal to obtain: (9) = ref = 143m Uing (2) and (3) we get the alibrated value for the d urrent: 143m = = = 14. A (10) 10mΩ alib 3 The urrent value obtained in (8) and (10) an be ompared to the atual urrent atual = 14.6 A. The ening (8) baed on the nominal on reitane value reult in 23% error, while the alibrated value (10) i off by only 2.1%. Following the ame proedure, four load ae have been teted: 15 A, 9 A, 4.5 A and no load. The reult are hown in Table. We an ee that the alibrated d urrent ha ignifiant improvement in auray ompared to the urrent d baed on the nominal on of the main ynhronou retifier. The bai alibration method an have very good performane when the indutane value and the output voltage in the onverter iruit are not too low, o that the indutor urrent error indued during alibration yle an be ignored. f thi i not the ae, the more aurate alibration method of Setion.B or.c an be employed. B. Experimental verifiation of the etimation of the error indued in By adding the omplexity of ampling and imultaneouly during alibration yle and performing the omputation of (4), the improved on-line alibrated ening reult of Table were obtained. The reult are hown for two ae: (a) urrent error were ompenated baed on the nominal indutane value, L = 3.0 µh, and (b) urrent error were ompenated baed on the ative etimation of the indutane value. The reult baed on the nominal L how an improvement over the reult of Table at maximum load. Additional auray wa ahieved by ampling at two point during the alibration yle to atively etimate the indutane value for ue in (4). Additional auray may be required in appliation uh a M where very low indutane value are ued with high urrent, whih may lead to larger error than thoe hown in Table and. n uh appliation, the method of Setion.C an be ued to anel out the dependene on preie knowledge of the indutane. TABLE. EXPEMENTAL ESLUTS FO ON-LNE CALBATED CUENT SENSNG WTH NDUCTO CUENT EO COMPENSATON Load (W) L nominal (µh) (a) Error (%) in alib with nominal L ued in (4) L etimated (µh) (b) Error (%) in alib with etimated L ued in (4)

6 . LOAD CUENT SENSNG APPLCATON EXAMPLE n thi etion, we onider an appliation example where the purpoe of urrent ening i to etimate the average indutor urrent, whih i equal to the load urrent in the ynhronou buk onverter. To aomplih thi, we hooe the ening delay t d to be equal to one half of the ynhronou retifier on time, t d = 1 D) T / 2. (11) ( Fig. 9 how the waveform,, and onverion tart ignal at full load, when the ening point i in the middle of the ynhronou retifier on time. The firt yle hown i a normal yle and the eond yle i a alibration yle. and are d with the ame delay t d = 4.3 µ. The urrent-ening reult for the four load ae are hown in Table. alib i the urrent d by the on-line alibrated ON method uing (2) and (3). t an be oberved that the propoed method allow aurate load urrent ening. Fig. 10 how the,, and onverion tart ignal when the onverter i operated at zero load. and waveform go from negative to poitive when the indutor urrent goe from poitive to negative. When the ening point i in the middle of the ynhronou retifier on time, the d voltage and are nearly zero, reulting in a large relative error in the alibration proe. Option to avoid large error at light load inlude reverting bak to the ON method until larger urrent are deteted, or hanging the delay t d to avoid the zero roing of the indutor urrent.. CONCLUSONS Thi paper deribe a lole urrent ening method with on-line alibration where an auxiliary with and a preiion reitor are onneted in parallel with a main power with to ahieve auray omparable to the reitor method, together with the advantage of eentially no TABLE. EXPEMENTAL ESLUTS FO THE LOAD CUENT SENSNG APPLCATON EXAMPLE Atual urrent (A) Error (%) in (urrent d by ON method only) Error (%) in alib (urrent d by on-line alibrated ON method) / / Figure 10. Top-to-bottom: (Ch. 3), (Ch. 1), the indutor urrent (Ch. 4) and onverion tart ignal (Ch. 2) at zero load (0 A). additional power lo. The propoed urrent-ening iruit and the alibration method are partiularly well uited for digital ontroller implementation where the required ontrol and alibration funtion an be eaily aomplihed. Experimental reult with a digitally ontrolled 12 -to-1.5 ynhronou buk onverter operating at 100 khz withing frequeny demontrate funtionality of the propoed ening and alibration, and a ignifiant improvement in auray over voltage ening aro the power MOSFET on-reitane. Figure 9. Sening and alibrating the average indutor urrent. Top-to-bottom: (Ch. 3), (Ch. 1), (Ch. 4) and onverion tart ignal (Ch. 2) at full load (15 A). The vertial uror are hown at the urrent ening point in the normal (firt) and alibration (eond) yle. EFEENCES [1] H.P. Forghani-zadeh, G.A. inón-mora, Current-ening tehnique for DC-DC onverter, The 45th EEE Midwet Sympoium on Ciruit and Sytem, 2002, pp [2] E. Dallago, M. Paoni, G. Saone, Lole urrent ening in lowvoltage high-urrent DC/DC modular upplie, EEE Tranation on ndutrial Eletroni, ol.47, No.6, De 2000, pp [3] D. Grant,. William, Current Sening MOSFET for Protetion and Control, EE Colloquium on Meaurement Tehnique for Power Eletroni, 1992, pp. 8/1-8/5. [4] Current-ening power MOSFET, ON Semiondutor appliation note, AND8093/D, July 2002, ev.5. [5] on Lenk, Optimum urrent ening tehnique in CPU onverter, Fairhild Appliation Bulletin AB-20.G. [6] P.Midya; M.Greuel, P.T.Krein, Senorle urrent mode ontrol - an oberver-baed tehnique for DC-DC onverter, EEE Tranation on Power Eletroni, ol.16, No.4, Jul 2001, pp