Design of an Active Differential Mode Current Filter for a Boost Power Factor Correction AC-DC Converter Rajib Goswami1, Shuo Wang2 and Yongbin Chu1 1 Power Electronics and Electrical Power Research Laboratory, University of Texas at San Antonio, TX 78249, USA 2 Power Electronics and Electrical Power Research Laboratory, University of Florida, Gainesville, FL 32611, USA Email: shuowang@ieee.org (Dr. Shuo Wang) Abstract - The aim of this paper is to introduce a methodology to design an active DM filter at the input side of AC-DC converters to cancel current ripples and DM EMI noise. The boost PFC AC-DC converter’s noise characteristic was first analyzed. The active filter topologies are designed based on the identified characteristics of the AC-DC converters. Several design issues such as thermal run away, DC offset, stability and compensation design, and the number of preamplifier stages are to be addressed in detail. Both simulations and experiments were conducted to validate the proposed methodology. It is found that the proposed design methodology can meet the design objective and the big stability challenge for the DM active filter of AC-DC converters. Index Terms—Boost PFC, DM noise, Electromagnetic Interference, Active DM current filter, AC/DC converters. I. INTRODUCTION Almost all the switching mode power conversion systems generate conducted electromagnetic interference (EMI) noise [1]. The EMI can be categorized to differential mode (DM) and common mode (CM) EMI. The noise flowing between the power conversion systems and ground is usually called CM noise and the noise flowing between two power carrying conductors is called DM noise [2]. Passive LC filters are widely used in industry to attenuate EMI. However, the size, weight and cost of the passive EMI filters are always a big concern for power electronics industry. Because filter inductors must carry the full load current and DM capacitors must withstand full AC line voltages, the passive DM filters can be very big. Hybrid CM filters have been proposed to reduce the EMI filter size, weight and cost [2]. In a hybrid CM filter, an active filter attenuates the low frequency (LF) CM noise and a passive filter attenuates the high frequency (HF) CM noise. Because the corner frequency of the passive filter is significantly increased, the CM EMI filter size could be significantly reduced. However, an EMI filter is composed of a CM and a DM functional units [3]. The hybrid CM filter can only reduce the size and weight of the CM functional unit. In order to further reduce the size of the whole system, the size and weight of the DM functional unit must be reduced. Most of the existing techniques address the CM active filter [4-8]. Some literature addresses the DM active filter for DC/DC applications [9]. On the other hand, the DM active filters designed for AC-DC converters is deemed to be very challenging [10] and there are few papers to address this topic. In this paper, a methodology will be developed to design This work was sponsored by The Boeing Company. 978-1-4673-7151-3/15/$31.00 ©2015 IEEE an active DM filter in the AC side of a boost PFC AC-DC converter. The DM noise model for the boost PFC is first developed. Based on the developed model, the current cancellation, voltage cancellation, feedback and feed-forward active filters are analyzed. The feedback current cancellation is proposed for DM active filter. Design issues such as thermal run away, DC offset, stability, compensation design, and the number of preamplifier stages are addressed in detail. Solutions are proposed to these issues. Both simulations and experiments were carried out to validate the proposed methodology. It is found that the proposed design methodology can meet design objective and stability challenge for the DM active filters of AC-DC converters. II. DM NOISE MODEL OF A BOOST PFC AC-DC CONVERTER Boost PFC AC-DC converters have been widely used as front end converters to offer high power factor to power grid. The converter works at switching mode which generates lots of DM and CM noise. Fig.1 shows a typical boost PFC circuit and an EMI measurement setup with line impedance stabilization networks (LISNs) connected between the power line and the converter. The path of DM noise current IDM is shown in the figure. Based on substitution theory, the DM noise model can be developed by replacing the MOSFET with a voltage source which has the same voltage waveform as the drain to source voltage VDS, and replacing diode with a current source which has the same current waveform as the diode current ID. Based on the superposition theory, the DM noise generated by VDS can be represented with Fig. 2. It is also found that ID does not contribute to the DM noise on the AC input side. So Fig. 2 is the DM noise model for a boost-PFC AC-DC converter at continuous conduction mode. The DM noise source impedance is determined by ZCBL//ZLB. For the prototype used in this paper, CBL is 0.22uF and LB is 400uH. Within the concerned frequency range from 150kHz to 30MHz, the noise source impedance ZN is determined by ZCBL. The noise load impedance is LISN’s impedance ZLISN or grid impedance ZGrid when LISNs are not present. D1 D2 LB LISNs DB i DM VAC CBL 50μH 50μH 0.1μF 1μF×2 50Ω ID RL VDS VL CB 0.1μF 50Ω D3 D4 Fig. 1 DM noise in a Boost PFC AC-DC converter. 4375 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. (a) Feedback, LB ZLISN or ZGrid iDM ZLISN or ZGrid (b) Feedforward. VDS CBL ZN ICancel(s) ZN ICancel(s) VN ZGrid ZN iDM VN AT(s) ZGrid AT(s) iDM Feedback (a) Fig. 2 DM noise model. Feedforward (b) Fig. 4 Feedback and feed-forward control. III. ACTIVE FILTER TOPOLOGY DESIGN In Fig. 3, to reduce DM noise, there are several ways to design active filters based on the criteria below: 1) Based on noise sensing and cancellation method: (a) Voltage sensing - voltage injection, (b) Current sensing - voltage injection, (c) Voltage sensing - current injection, (d) Current sensing - current injection. VCancel(s) ZGrid iDM VN ZGrid AT(s) V-V iDM ZN ICancel(s) (b) ICancel(s) VN V-I VN I-V AT(s) (c) ZN iDM A (s) T (a) ICancel(s) ZGrid VCancel(s) ZN For feedforward control, to achieve a good cancellation, the active filter's gain AT(s) should be as close to unity as possible. It has a strict requirement on the values of key components. For feedback control, the active filter's gain AT(s) should be as big as possible. Feedforward control does not improve the unity gain bandwidth of the active filter over feedback control due to the fixed unity gain-bandwidth for a given operational amplifier. It can improve the magnitude of gain at the pass-band, but that improvement is very much dependent on the accuracy of the component values used in the design [1]. Hence, due to the flexibility of the feedback control, it is selected in this paper. Because the current sensing - current injection is employed, for the convenience of analysis, it is better to represent the noise source with a Norton equivalence in Fig.5. ZN ZGrid VN ZGrid iDM AT(s) Fig. 5.Noise source with Norton equivalence. Fig. 3. DM active filter topology. For the active filter using voltage injection method, a voltage transformer is needed in series with AC power line. It must therefore carry full converter currents. Furthermore, the transformer magnetizing inductance must be big enough to reduce the magnetizing current supplied from the active filter so as to reduce the power loss of the active filter. Because of this, the transformer is big and heavy which contradicts active filter's objective: filter size and weight reduction. Because of this, voltage injection should be avoided. For the DM noise sensing, a current transformer is a simple way to directly sense DM noise current. It is also possible to eliminate the influence of CM noise current with a proper current transformer structure. DM noise voltage on power line can also be sensed for active filters. However, the sensed voltage includes not only the information of the DM noise of the boost PFC converter but also the information of grid impedance and the voltage noise flowing from the grid. It does not directly represent the DM noise information of the boost PFC converter. It complicates the filter design. Because of this, current sensing is the best choice. Based on analysis above, current sensing and current injection should be used for the active filter. In Fig. 4, the control strategy of the active filters incudes: IN Feedback I-I (d) 2) Based on control scheme: iDM ZN AT(s) The current gain GCL(s) from IN to iDM in Fig. 5 is given by (1): GOL ( s ) G (s) , (1) G (s) = = OL CL 1 + AT ( s )GOL ( s ) 1 + T ( s ) ZN where, G ( s) = and T ( s ) = AT ( s )GOL ( s ) . T(s) is OL Z N + Z Grid the loop gain. To ensure stability, T(s) must have sufficient gain and phase margin. The current insertion gain GIS(s) in Fig. 5 is defined as the iDM with active filter to that without active filter. It is given by (2) as, 1 1 . (2) G ( s) = = IS 1 + AT ( s )GOL ( s ) 1 + T ( s) GIS(s) should be as small as possible. Because of this, active filter’s gain AT(s) should be as big as possible. It should also have enough bandwidth. GOL(s) should be also as close to unity as possible to achieve small GIS(s). It means ZN>>ZGrid. Because in Fig. 2, ZN is small impedance, it is therefore not good for the active current filter implementation in Fig. 5. Because of this, a DM inductor LDM is added between the boost PFC converter and the active filter. Furthermore, a DM capacitor CR is added between the active filter and the power grid to reduce grid impedance so that new source impedance and new load impedance meet condition (ZLDM+ZN) >>ZCR//ZGrid. This means GOL(s)≈1. So GIS(s)≈1/(1+AT(s)). Fig. 6 shows this modification. 4376 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. CR ZGrid ICancel(s) IN ZN AT(s) iDM both EMI noise frequency component and power line frequency component. One of the biggest differences between the active filter of an AC-DC converter and a DC-DC converter is the active filter of an AC-DC converter must avoid to inject line frequency currents because the line frequency currents are not the DM noise currents to be suppressed. Because of this, a high pass filter is integrated with the CT. In Fig. 9, CHP is added between CT and the load resistor RCT. LCT, CHP and RCT forms a high pass filter with an I-V transfer function, LDM Feedback Fig. 6.Final optimal topology for a feedback, I-I DM active filter. IV. ACTIVE DM EMI FILTER DESIGN A. Insertion Gain GIS(s) and Current Gain GCL(s) with LISNs In a standard EMI noise test setup, there are two LISNs as shown in Fig. 1 between power grid and the power converter. Based on the CISPR 16 EMI standard, for DM noise from 150kHz to 30MHz, LISNs can be approximately modeled as a resistance RLISN=100Ω in parallel with an inductance LLISN=100μH as shown in Fig. 7. 50μH 100μH ZLISN(s) Fig. 7 LISN Equivalent Impedance Model for DM After the insertion of LISN model between power line and CR, the active filter topology can be shown as in Fig. 8. LDM LLISN R C LISN R iDM ICancel(s) ZN AT(s) IN Fig. 8. LISNs are inserted between power line and CR. The equivalent impedance on grid side 1 . As discussed in the last section, due to the || sLLISN || sC R small impedance of CR, GOL(s) is very close to unity and hence after LISNs' impedance is paralleled with CR, GOL(s) is still close to unity from 150kHz to 30MHz, 1 1 (3) G (s) ≈ G (s) ≈ = CL IS 1 + T ( s) · ¸¸ ¹ (4) 2 1 and Q = 1 RCT LCT CHP LCT C HP LCT CHP RCT Vin(s) The line frequency current usually has a high magnitude to deliver power to the AC-DC converter. ωc must be high enough so that the high pass filter has a high attenuation to line frequency current component. At the same time, to maintain a high loop gain T(s) or low GIS(s) within the concerned frequency range above 150kHz, ωc must be below 150kHz or the first noise spike after 150kHz. There is around 3.4 decades between 60Hz and 150kHz. The high pass filter also introduces phase shift which needs to be considered for the stability of the active filter. Hence, careful compensation is needed to shape the gain and phase of AT(s). 2) Cancellation of CM Noise while Current Sensing Feedback LISN § s s +¨ ω c Q ¨© ω c , Fig. 9. Current transformer with a 2nd order high pass filter. 100Ω 50Ω 50Ω 1+ 2 · RCT ¸¸ ¹ n IDM(s)/n 0.1μF 0.1μF is R where, ω = c LISN Model for DM in Impedance 150kHz-30MHz ZLISN(s) 50μH 1μF×2 Vin ( s ) = I DM ( s) § s ¨¨ © ωc In this paper, only the reduction of DM noise is considered so the performance of active filter in reducing DM noise is the focus. The AC-DC converter also generates CM noise which circulates between the power-line and the ground. Hence, when the current is sensed using a CT, it senses both the CM and DM currents. In Fig. 10, the current sensed by a typical CT is connected to a single power line. The sensed primary current iprim includes both DM and DM noise: iprim=iDM+iCM. Line 1 1 + AT ( s) Noise Receiving End Based on (3), the active filter's noise attenuation performance and stability are determined by T(s). T(s) should be analyzed and designed. B. Active Filter Circuit Design Toroidal current transformer Sensed Primary Current =iDM + iCM iDM iDM iCM Noise Source End iCM Line 2 1) Current Transformer Integrated with a High Pass Filter A toroidal current transformer (CT) with a turn ratio n is designed to sense the DM current. The magnetizing inductance of the secondary is LCT. The sensed current has Fig. 10 Noise current sensing using a current transformer. A simple current sensing technique can be applied to cancel the CM noise current from the sensed current. In Fig. 11, this CT arrangement technique has been shown. 4377 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. Line 1 Noise Receiving End Toroidal current transformer Sensed Primary Current =2×iDM iDM iCM Noise Source End iDM iCM The amplifier has a preamplifier stage (for voltage amplification) and a current amplifier stage. The preamplifier stage is composed of a wide bandwidth operational amplifier and the current amplifier stage is a class AB amplifier. b) Line 2 Fig. 11 Cancelling CM noise from the sensed noise current. In the arrangement as shown in Fig. 11, the line 1 goes through the CT's toroidal core in exactly the same manner as in a typical CT and the noise current flows through it is iDM+iCM . The difference in this arrangement is that the line 2 is also made to pass through it in the reverse direction so that iDM-iCM also passes through the toroidal core. Hence, due to this arrangement the total current sensed by the CT is, iprim=2×iDM. The sensed iCM is canceled. . The CT’s I-V transfer function of (4) is now modified as follows, 2 § s · R 2 × ¨¨ ¸¸ CT Vin ( s) (5) © ωc ¹ n = 2 I DM ( s) § s · s + ¨¨ ¸¸ 1+ ωcQ © ωc ¹ The first advantage of this arrangement is that, it does not sense CM noise which will be dealt with a dedicated CM hybrid filter. The second advantage is that it increases the DM current by twice (6 dB) which eventually increases the loop gain. The stability issue for both low and high frequencies needs to be considered carefully with this increased loop gain. Number of Preamplifier Stages More than one stage preamplifiers can increase the open loop gain of the amplifier. It leads to increased -3dB bandwidth. However, the open loop gain of an operational amplifier usually has a single pole within the concerned frequency range. One stage preamplifier introduce -π/2 phase shift. m stage cascaded preamplifiers introduce -mπ/2 phase shift. Because the output of CT has almost no phase shift within the concerned frequency range, with more than one stage preamplifiers, the AT(s) will have more phase shift which causes stability issues. Because of this, a single stage preamplifier is preferred. Class AB Amplifier Design c) A class AB amplifier is used in the output stage of the active filter circuit to have a higher current injection capacity. A typical configuration is shown In Fig. 13(a). +VCC RB1 ID Iin V1 3) Amplifier Design Fig. 12 shows the amplifier block diagram. It includes preamplifier stage and current amplifier stage. -VCC Current amplifier Preamplifier RB1 ID Iin V2 Iout V1 T2 IC IRB1 + T1 VBE - + RE1 V -D V2 Iout RE2 RB2 Zload T2 -VCC (b) Fig. 13 Class AB Amplifier (a) Typical, (b) with emitter resistors Cinj Rinj Rinj (b) Fig. 12. Amplifier configuration: (a) current feedback and (b) voltage feedback. a) T1 + V -D (a) Current amplifier Cinj (a) + VBE - Zload RB2 Preamplifier +VCC IC IRB1 DC Offset Issues In Fig. 12, the voltage amplifier (b) instead of current amplifier (a) is employed here because the current amplifier may have output DC offset issues. This is because the current amplifier regulates the output current so the output voltage may have DC offsets due to the unbalanced charge on the capacitor Cinj in the injection network or load impedance. The charge on the capacitor will build up and may finally saturate the output and then the active filter cannot work properly. The voltage amplifier regulates the output voltage, there is no output DC offset issues. Although voltage amplifier is used, it can still inject currents as the output voltage is added to the injection network and the load impedance. A class AB amplifier has an npn and a pnp transistor with equal base to emitter voltage. The two transistors should have a very high bandwidth and does not reduce the active filter’s bandwidth for the specified EMI frequency range. They should have enough current injection capacity and have sufficient power rating for the desired conditions. ZDT6753, npn-pnp complementary pair was chosen in the design. The two transistors in ZDT6753 have very similar features. It has a typical transition frequency (bandwidth) of 175MHz and hence does not have adverse effects in the concerned EMI frequency range. For biasing the amplifier, different biasing techniques has been analyzed. It is found that, as the supply voltage ±VCC can be fixed to a certain value, a simple diode bias would the advantage of having less number of components and simplicity of design. From the IC-VBE(ON) curve of the transistor as shown in Fig. 14, the turn-on baseemitter voltage, VBE(ON)=0.545V. It corresponds to a collector current, IC at 0.6mA. For having a good high frequency performance 1N4148 signal diodes are used for biasing. The voltage across diode, VD should be slightly higher than VBE, so it was chosen as VD=0.57V. From the IDVD curve of diode, as shown in Fig. 15, VD corresponds to a 4378 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. diode current, ID=0.3mA. To handle high injection current, the current in bias should be high enough, so two bias diodes are in parallel. Now, IRB1 ≈ 0.6mA. 0.01 0.009 0.008 IC (A) 0.006 0.004 0.003 X= 0.545 Y= 0.0006 0.001 0 0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.6 0.62 VBE(Volt) Fig. 14 IC-VBE(ON) curve of the transistor. -3 x 10 ID (A) 4 Rescaled and Redrawn From Diode Datasheet 3 Z Load 1 X= 0.57 Y= 0.0003 I inj Vout 0.45 0.5 0.55 0.6 0.65 0.7 ≈ 0.75 1 + Rinj sCinj 1 + Rinj sCinj (6) sC inj 1 = § 1 · 1+ s ¨ + Rinj ¸ ω inj ¨ sCinj ¸ © ¹ Here ωinj=1/(RinjCinj). VD (Volt) D. Component Parameter and Compensation Design Fig. 15 ID-VD curve of the bias diode. With the relationship, R = VCC − VD , the value of RB1 can B1 Active filter’s open loop transfer function is, I inj Vin Vout I inj . I RB1 be calculated. The value of RB1 is dependent on the bias voltage or supply voltage, VCC. Hence, the value of VCC needs to be chosen first. VCC needs to be chosen considering the highest magnitude of the injected current, so that the amplifier’s output voltage V2 is still smaller than VCC for that magnitude of injected current. d) § 1 · + Rinj ¸ ≈ Z Load = Z CR || ( Z LDM + Z N ) + ¨ ¨ sCinj ¸ © ¹ § 1 · = Z LISN || Z CR || ( Z LDM + Z N ) + ¨ + Rinj ¸ ≈ ¨ sC ¸ inj © ¹ The transfer function from the output voltage of class AB amplifier to injection current is therefore, (7) 2 0 0.4 IC The current injection network consists of an injection capacitor Cinj and a small resistor Rinj. The injection capacitor is needed for the isolation of active filter from high voltage AC power line and the resistor is needed to ensure stable impedance and a constant gain on pass-band. The load impedance of the class AB amplifier is the impedance of the injection network in series with the impedance between two power lines as shown in Figs 6 and 8. The load impedance is given by (6) for without LISNs and (7) for with LISNs. Due to the small impedance of CR, the load impedance of the class AB amplifier is mostly determined by injection network if the injection network is properly designed. X= 0.57 Y= 0.002 0.002 5 ID IS I C I SD , where VT is thermal C. Current Injection Network Design 0.005 6 the following relation, R = E VT ln voltage, ID is diode current, ISD is diode saturation current, IC is collector current and IS is BE junction saturation current, I The class AB amplifier configuration with the emitter resistors is shown in Fig. 13 (b). Rescaled and Redrawn From Transistor Datasheet 0.007 value of emitter resistors, RE1 and RE2 can be done based on Thermal Runaway Issue for the Class AB Amplifier For the class AB amplifier, the two transistors have power loss which generates heat and increases device temperature. When the transistor temperature increases, the transistor base to emitter (BE) junction voltage drops. However, the BE bias voltage is not reduced. This causes the transistor to enter linear region. As a result, there is more current flowing through two transistors and generating higher conduction power loss. This in turn further increases the device temperature until the transistors are killed. Two emitter resistors RE1 and RE2 are added to the class AB amplifier. They provide a negative feedback to stabilize the BE junction bias voltage. They were proved to be an efficient solution to thermal runaway issue. Choosing the AT ( s) = I DM = I DM × Vin × (8) Vout Hence, compensation can be designed for Vout/Vin to achieve desired gain and stability. n, LCT and RCT are designed based on primary currents, magnetic core characteristics and the desired loop gain. CT should be designed to make sure that the CT core does not get saturated for the designed current and desired inductance. In this paper, the current transformer was designed to have 10 turns and the chosen value for the high pass filter inductor is 1mH. The CT was made with ZW41610TC core from the Magnetics Company and has around 0.96mH with 10 turns. CHP can be determined based on the attenuation required at the line frequency. These three values determine ωc and Q in equation (4). The injection pole frequency, ωinj is determined by Rinj and Cinj. Rinj should be small to limit the power loss and also the conditions of (6) and (7) should be satisfied. From (7) and (8), Rinj also influence active filter’s loop gain 1/Rinj above ωinj. Hence, to obtain enough gain at the pass-band, Rinj should not be too big too. The choice of Cinj is dependent on the allowable 60Hz leakage current flowing through the Cinj. It is due to the 4379 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. fact that at line frequency (60 Hz for the case under consideration) the impedance of Cinj is typically much higher than Rinj (due to the small value of Rinj). Hence, the impedance of Cinj determines the leakage current at line frequency due to the drop of line voltage across the injection network. High leakage 60Hz current causes power loss to class AB amplifier. To limit the leakage current the value of Cinj needs to be small (high impedance). To keep a constant ωinj, a small Cinj requires a big Rinj. Hence, there is a tradeoff between the choices of these two values. The preamplifier can be designed to have the desired gain at pass-band and also to have sufficient phase and gain margin. It is assumed that the operation amplifier can be approximated with unity gain-bandwidth Aωop, where A is the open loop gain and ωop is the open loop single pole corner frequency. Another requirement for the operational amplifier is to have a high operating voltage. This is necessary to make sure that the operational amplifier is not saturated even when the maximum injection current is reached. This high operating voltage condition decreases the possibility of using a low voltage, high bandwidth operational amplifier. For the compensation design, a simple non inverting topology with a single amplifier stage has been chosen. Combining all the individual parts the chosen voltage amplifier current injection active filter circuit is shown in Fig. 16. In Fig. 16, for the simple non inverting topology, voltage gain of the amplifier is given by, Vout/Vin=1+ZF/ZG. For AT(s) in (8), the Vin/IDM in (5) and Iinj/Vout in (7) have three zeros at the origin. At high frequencies, these zeros at the origin are cancelled by the poles in (5) and (7).Operational amplifier’s pole will be the only pole at high frequencies. High frequency compensation design would be discussed in a later section. Low frequency instability could happen before 150kHz when loop gain is higher than unity and there is not enough phase margins. Hence, low frequency compensation design should focus on the frequency below 150kHz when loop gain is still smaller than unity. In the design, 1 , for implementing those 1 and Z F = RF + Z G = RG + sCF sCG compensation pole and zero at low frequencies. Hence, at low frequencies, the voltage gain of the amplifier becomes, s 1+ ωCMP1 Vout (9) = AOP × s Vin 1+ ωCMP 2 1 where, AOP = C F + C G ,and ω CMP 2 = , C G RG CF ω CMP1 = ω CMP 2 × AOP × RG RG + RF The zero at ωCMP1 needs to be placed very close to CT’s corner frequency ωc as shown in Fig. 17, so that the magnitude continues to rise at 40dB/dec and at the same time, the phase is compensated. This high rate is important if a high gain is needed for the EMI frequency range. The remaining pole at ωCMP2 has to be placed in between ωc and ωinj, so that there is sufficient gain and phase margin. In the high frequency range (>150kHz), for a more accurate model, the model is modified to include operational amplifier’s internal output resistance R0 and its open loop transfer function characteristics becomes as follows, Z 1+ F (10) Vout ZG = Vin s ­§ Z F · § Z F ·§ R0 · § R0 ·½ ¸ + ¨1 + ¸¨ ¸+¨ ¸¾ 1+ ®¨1 + Aω op ¯¨© Z G ¸¹ ¨© Z G ¸¹¨© Z Load ¸¹ ¨© Z G ¸¹¿ At low frequencies, (10) can be approximated using (9) and hence the pole placements hold alright. Considering (10) and plugging in all the different transfer functions of (8), the final loop gain of the active filter becomes, 2 AT ( s ) = I inj I DM § s · R 2 × ¨¨ ¸¸ CT © ωc ¹ n × sCinj × = 2 s s § s · s 1+ 1+ + ¨¨ ¸¸ 1 + ω inj Aωop ωc Q © ωc ¹ 1+ ZF ZG ­§ Z F · § Z F ·§ R0 ¸ + ¨1 + ¸¨ ®¨¨1 + Z G ¸¹ ¨© Z G ¸¹¨© Z load ¯© · § R0 ¸¸ + ¨¨ ¹ © ZG ·½ ¸¸¾ ¹¿ (11) +VCC T1 RB1 CHP Isec LCT RE1 Vin RCT ZG RE2 ZF RB2 RLISN Vout LLISN CT CInj T2 RInj -VCC IPrim LDM CR ZN ICancel IN(s) Fig. 16 Active filter circuit connected in the Feedback configuration. dB +20 |AT(s)| -60 ωCMP 2 20dB/dec ωinj ωop ' 40dB/dec 60dB/dec ωc , ωCMP1 Radian Frequency ωinj Fig. 17. Loop gain AT(s). Fig. 17 shows the loop gain with all the corner frequencies described above. ωop’ which is not directly shown but can be derived from (11), is caused by the pole ωop of the operational amplifier in (11). E. Design of Operational Amplifier and Related Component While choosing operational amplifier, the conditions as presented before needs to be satisfied. The operational amplifier can be selected based on its internal structures: voltage feedback amplifiers (VFAs) or current feedback 4380 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. F. Loop Gain Adjustment for High Frequency Stability The operational amplifier AD829 can be approximated using a single pole until a very high frequency. But if the pass band (150kHz-30MHz) gain needs to be increased, careful high frequency compensation is required. This compensation is needed due to the fact that the operational amplifier has parasitic poles and zeros at high frequency. Hence, the gain needs to be adjusted to fall below zero dB (unity gain) before the phase becomes affected by those parasitic poles and zeros. This can be done by adding a very small valued capacitor CFCMP in parallel with ZF. ZF is now modified as, § 1 · 1 . With this compensation method ¸¸ || Z F 1 = ¨¨ RF + sC F ¹ sCFCMP © a pole-zero pair is added and enough phase and gain margin can be ensured at high frequencies. With the stated 1 , and the configuration, the pole is placed at ω PCMPH = RF C FCMP § R + RF = ωPCMPH ¨¨ G © RG In this design a 10pF CFCMP capacitor is used. zero is placed at ω ZCMPH where, ω ZCMPH the low frequency spectrum with and without the active filter in section V below. Magnitude(dB) 30 With CFCMP 10 0 -10 5 10 6 10 Frequency 10 7 10 7 Phase(Degree) 100 0 -100 -200 5 10 6 10 Frequency Fig. 18 Comparison of Loop Gain without and with the Compensation Capacitor. V. EXPERIMENTAL RESULTS A prototype was developed with a pass band gain of 20dB following the methodology developed above. The prototype was first tested using a signal generator and an RF amplifier as the noise source. Then the prototype was used for a boost PFC AC-DC converter to reduce its DM noise. The simulations and the experiments fully validate the proposed design technique above. 120 Without Active Filter With Active Filter 100 80 60 40 20 ·. ¸¸ ¹ The measured loop gain with and without the CFCMP capacitor for the circuit in Fig. 16 are compared in Fig. 18. The loop gain was measured using network analyzer Agilent 4395A. The loop gain was calculated from the measured Sparameters. It is shown that the designed circuit has a good pass-band gain and has sufficient phase margin (no stability issue) when CFCMP capacitor is used. The frequencies lower than 100kHz is not shown in this measurement due to the frequency limitation of the network analyzer Agilent 87511. The low frequency stability was also ensured by using a square wave generator as the noise source, Vds and checking Without CFCMP 20 dbuA amplifiers (CFAs). Both of them have their advantages and disadvantages. The good thing about CFAs is that they can have higher bandwidth as compared to VFAs. But in CFAs, the condition of single pole approximation for the transfer function is not applicable, as their structure is more different. Moreover they tend to have more offset voltages as compared to VFAs. Hence, a VFA is preferable here. While choosing between different VFA's, the single pole approximation should hold for the chosen operational amplifier. The supply voltage of the operational amplifier needs to be high enough so the operational amplifier's output should not get saturated even for the highest amplitude of the output voltage. Another thing to consider is that the operational amplifier should have both positive and negative supply voltage due to the circuit topology chosen as shown in Fig. 16 (a single power supply with a DC bias to the amplifier may also be a choice). In this design AD829 VFA is chosen, as this operational amplifier has a high bandwidth and meets those criterions stated above. Based on the operational amplifier's performance the value of RG, CG and RF, CF needs to be chosen using (9). If the chosen resistor values are too big or too small the operational amplifier's performance becomes bad due to the internal circuit structure of the operational amplifier. In this work, RG=100, RF=1000, CG=69nF, CF=22nF are used. 0 5 10 6 10 Frequency(Hz) 10 7 Fig.19 Measured DM noise with and without active DM filter for a square wave noise source from 100kHz to 30MHz. In Fig. 19 the noise is generated using a signal generator and an RF amplifier. In the measurement, CR is 0.2μF and LDM is 180μH. The square wave noise which has high frequency harmonics was fed to the active filter. In Fig. 19, the noise spectrum without the active filter is shown by black lines and the noise after the addition of the active filter is shown by grey lines. It is seen that the active filter is working as expected and reduced the noise by around 20dB in the pass-band. After around 12MHz, it cannot cancel the EMI noise which is also expected from the measured loop gain of Fig. 18 as loop-gain is almost equal to the inverse of the insertion gain as predicted by (2). 4381 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply. To validate that the designed active filter will have reduced performance below 150kHz, the square wave’s frequency is adjusted to 10kHz. Fig. 20 shows the measured spectra from 10kHz to 500kHz. It clearly shows that due to the high pass filter integrated with the active filter, the active filter has reduced performance below 150kHz. It has negligible noise reduction performance for 10kHz noise. 90 Without Active Filter With Active Filter 80 70 60 dbuA 50 40 30 20 10 0 -10 4 10 5 Frequency 10 VI. CONCLUSION In this paper, a FB current cancellation active DM filter is proposed, implemented, and tested, and techniques to improve the performance of active filters are explored. This paper focuses primarily on the challenges of implementing the FB circuit topology for the case of AC-DC converters. It begins with introducing the DM EMI noise model for a boost PFC AC-DC converter. It describes the constraints and impedance requirements for the successful implementation of the active filter. It also discussed the choice of active filter topologies. Then it develops the steps for the active filter hardware design and describes the constraints for choosing the individual components. Based on the proposed design methodology of this paper, a prototype active filter circuit has been implemented. The developed active filter circuit can ensure stability as well as maintain a particular gain requirement, which are big challenges in case of active filter design for ACDC converters. The prototype was tested using a square wave noise source and also with a boost PFC AC-DC converter. The developed technique was validated with experiments. Fig. 20. Measured DM noise with and without active DM filter for a square wave noise source from 10kHz to 500kHz. In Fig. 21, the measured DM noise for a boost PFC ACDC converter is presented. The noise spectrum without the active filter is shown by black lines and the noise with the active filter is shown by broken grey lines. The PFC has a variable switching frequency due to its control scheme. In both measurements, all the CM and DM capacitors of the existing EMI filter of the boost PFC converter were removed (only the boost PFC capacitor CBL was not removed). The value of CR is 0.2μF and LDM, the leakage inductance of a CM inductor, is around 300μH. They meet the stated impedance condition as found from (2) and (6). 90 Without Active Filter With Active Filter 80 70 dbuA 60 50 40 30 20 10 0 5 10 6 Frequency 10 Fig. 21 Measured DM noise with and without active DM filter for a boost PFC AC-DC converter source For the boost PFC converter under consideration, the noise spectrum becomes very low at frequencies above 1MHz and after this frequency it becomes difficult to measure the improvement due to active filter, as it becomes close to background noise. It is shown that the results meet the designed pass band gain (not include LDM and CR’s attenuation) when the noise is measurable below 1MHz and it also operated free of instability. REFERENCES [1] Shuo Wang; Maillet, Y.Y.; Fei Wang; Boroyevich, D.; Burgos, R., "Investigation of Hybrid EMI Filters for Common-Mode EMI Suppression in a Motor Drive System," Power Electronics, IEEE Transactions on , vol.25, no.4, pp.1034,1045, April 2010. [2] Di Piazza, M.C.; Ragusa, A.; Vitale, G., "Design of Grid-Side Electromagnetic Interference Filters in AC Motor Drives With MotorSide Common Mode Active Compensation," Electromagnetic Compatibility, IEEE Transactions on , vol.51, no.3, pp.673,682, Aug. 2009. [3] Wenjie Chen; Weiping Zhang; Xu Yang; Zhiyong Sheng; Zhaoan Wang, "An Experimental Study of Common- and Differential-Mode Active EMI Filter Compensation Characteristics," Electromagnetic Compatibility, IEEE Transactions on , vol.51, no.3, pp.683,691, Aug. 2009. [4] W. Chen, Xu Yang and Z. Wang, "A Novel Hybrid Common-Mode EMI Filter With Active Impedance Multiplication," IEEE Trans. Ind. Electron. , vol.58, no.5, pp.1826-1834, May 2011. [5] K. Mainali and R. Oruganti, "Design of a current-sense voltage-feedback common mode EMI filter for an off-line power converter," in Proc. IEEE PESC, 2008, pp.1632-1638. [6] D. Shin, S. Kim, G. Jeong, J. Park, J, Park, K. J. Han and J. Kim, "Analysis and Design Guide of Active EMI Filter in a Compact Package for Reduction of Common-Mode Conducted Emissions," IEEE Trans. Electromagn. Compat. vol. PP, no.99, pp.1-12, 2015. [7] X. Chang, W. Chen, Y. Yang, K. Wang and Xu Yang, "Research and realization of a novel active common-mode EMI filter," in proc. IEEE APEC, 2015, pp. 1941-1945. [8] C. Zhu and T.H. Hubing, "An Active Cancellation Circuit for Reducing Electrical Noise from Three-Phase AC Motor Drivers," IEEE Trans. Electromagn. Compat., vol.56, no.1, pp.60,66, Feb. 2014. [9] Hamill, D.C., "An efficient active ripple filter for use in DC-DC conversion," Aerospace and Electronic Systems, IEEE Transactions on, vol.32, no.3, pp.1077, 1084, July 1996. [10] Farkas, T.; Schlecht, M.F., "Viability of active EMI filters for utility applications," Power Electronics, IEEE Transactions on, vol.9, no.3, pp.328-337, May 1994. [11] Chau, G.S.S.; Ziegler, S.; Iu, H.H.-C.; Daniyal, H., "Experimental verification of the linear current transformer model," Power Engineering Conference, 2008. AUPEC '08. Australasian Universities, vol., no., pp.1,6, 14-17 Dec. 2008. [12] Shuo Wang; Lee, F.C.; Odendaal, W.G., "Using scattering parameters to characterize EMI filters," Power Electronics Specialists Conference, 2004. PESC 04. 2004 IEEE 35th Annual, vol.1, no., pp.297-303 Vol.1, 20-25 June 2004. 4382 Authorized licensed use limited to: COVENTRY UNIVERSITY. Downloaded on March 25,2022 at 09:01:37 UTC from IEEE Xplore. Restrictions apply.