Design and Implementation of an Active EMI Filter for Common-Mode Noise Reduction

Abstract

This paper presents the analysis and design of an active electromagnetic interference (EMI) filter (AEF) for the common-mode (CM) noise reduction of switching power converters. The features of the several types of AEFs are discussed and compared in terms of implementation. The feed-forward AEF with a voltage-sensing and voltage-cancellation (VSVC) structure is implemented for an LLC resonant converter to replace a multiple-stage passive EMI filter and thereby reduce CM noise. The characteristics and performance of the VSVC-type AEF are investigated through theoretical and experimental works.

I. INTRODUCTION

The reduction of electromagnetic interference (EMI) is a significant issue in the design and implementation of high-power density switching power converters. A power-line EMI filter is a traditional solution to reduce the conducted emissions of converters [1]. A passive filter with a common-mode (CM) choke and X and Y capacitors is generally used for this purpose. However, this filter is bulky, and the manufacturing of the filter components, such as the CM choke, is too laborious.

In recent years, active EMI filters (AEFs) have been considered as an alternative to passive filters. AEFs perform an active cancellation of noise instead of LC filtering. Hence, they do not require bulky passive components. Moreover, they show great potential for integration into EMI filters in a small chip or package.

Research on AEF can be found in the literature [2]-[10]. The active cancellation of noise currents using AEFs was studied in [2]-[7]. Voltage cancellation and hybrid methods were reported in [8]-[10]. Differential mode (DM) AEFs were also considered in [5], [10]. Despite the extensive study on the usefulness of AEFs, further research on AEF circuits and components should still be performed for practical design and implementation.

This paper presents the design and implementation of active EMI filters for the CM noise reduction of LLC resonant converters. The characteristics of several AEF topologies are discussed and compared. The control loop characteristics of feedback and feed-forward AEFs are also investigated. A feed-forward AEF with a voltage-sensing and voltage-cancellation (VSVC) structure is finally implemented to replace the multiple-stage passive EMI filter. The performance of the implemented AEF in terms of EMI reduction is investigated through experimental works.

 

II. STRUCTURE OF ACTIVE EMI FILTER

A. Active EMI Filter Topologies

Fig. 1 shows the structure of the AEF for the CM noise reduction of switching power converters. The conducted noise of the power line is measured and is actively cancelled by the compensating voltage or current generated from the control amplifier.

Fig. 1.Structure of active EMI filter for CM noise reduction.

Four types of AEF topologies can be considered for the combination of the sensing and cancellation methods. The simplified CM equivalent circuits of four AEF types are shown in Fig. 2. In terms of implementation, voltage sensing is simpler than current sensing because it does not require a current transformer. Meanwhile, the current capability of control amplifiers is crucial for the selection of a cancellation method. The current type is difficult to implement because of the lack of a high-current and wide-bandwidth operational amplifier (OP amp). The possibility of a high CM current occurring is another problem for this method because it makes a bypassing path to the ground. As a result, the VSVC type is considered for the implementation of the AEF in this study.

Fig. 2.Types of AEFs.

B. Characteristics of Feedback and Feed-forward AEFs

In view of the control loop, two AEF structures can be considered according to the point of the measurement of noise signals. A noise signal is measured at the LISN terminals (control target) in the feedback control and at the converter input (noise source) in the feed-forward control.

Fig. 3 shows the CM equivalent circuits of the VSVC-type feedback and feed-forward AEFs. The transfer functions of both AEFs are derived to investigate the characteristics of the control loops. The transfer function of the feedback AEF is given as

where Vs, Vo, Zs, RL, and A denote the voltage of the noise source, voltage of the LISN terminal, noise source impedance, resistance of the LISN, and forward gain of the control loop, respectively. The measurement of noise source impedance is discussed in [11]. The total impedance ZL of the LISN side can be represented as

or

for the circuits with or without the additional CM choke, respectively, where LCM is the inductance of the additional CM choke. The small CM choke is generally combined to reduce the EMI in the high frequency range over few tens MHz because of the frequency limitation of the filter components, such as the OP amp and injection transformer.

Fig. 3.CM equivalent circuit of feedback and feed-forward AEFs. (a) Feedback AEF (b) Feed-forward AEF.

The transfer function of the feed-forward AEF can also be derived from Fig. 3(b) as

As indicated in (1), an extremely high value of the forward gain A is required to minimize the noise voltage (Vo) at the LISN terminal for the feedback AEF. However, the noise voltage tends to drop to zero as the forward gain A approaches unity (A = 1) for the feed-forward AEF, as shown in (4). Therefore, the feed-forward structure is chosen for the implementation.

 

III. DESIGN AND IMPLEMENTATION

The CM circuit diagram of the VSVC type feed-forward AEF is shown in Fig. 4. The diagram consists of a sensing circuit, a control amplifier, and a voltage injection transformer. The forward gain A can be represented as follows using the three blocks:

where GT, Gc, and Hs are the transfer functions of the injection transformer, control amplifier, and sensing circuit, respectively. The characteristics and design of each part are discussed in the next section.

Fig. 4.CM circuit of VSVC-type feed-forward AEF.

A. Sensing Circuit

The sensing circuit in Fig. 4 is a first-order high-pass filter. The transfer function of this circuit can be given as follows using the concept of virtual ground:

Its cut-off frequency is given as fh =1 / (2π R1Cs ) . The cut-off frequency of the sensing circuit should be lower than the lowest frequency component of the noise voltage. It is thus determined to be the value that is sufficiently lower than the lowest switching frequency fs of the LLC resonant converter ( fs = 80kHz in this design). Fig. 5 shows the simulated result for the frequency response of the sensing circuit, where R1 =10kΩ , Cs = 20nF , and fh = 796Hz .

Fig. 5.Simulated frequency response of sensing circuit ( R1 =10 kΩ , Cs = 20 nF and fh = 796 Hz ).

B. Control Amplifier

The OP amp is generally used for a control amplifier that generates a cancellation voltage. The gain of the inverting amplifier shown in Fig. 4 can be represented as

where R1, R2, and ωc ( = 2π fc ) are the feedback resistor, input resistor, and −3 dB frequency of the OP amp, respectively. The phase response of the OP amp circuit is more important than the gain because the phase delay of the control amplifier severely degrades the performance of the feed-forward AEF, especially in the high frequency range. In the worst case, a large phase delay may amplify the noise by summing up the noise and cancellation voltages. Fig. 6 shows the measured gain/phase responses of the inverting amplifier using the OP amp OPA847, where fc = 22MHz . Several important parameters of OPA847 are summarized in Table 1. The frequency response of the OP amp in the low gain (|Gc| < 1) is important for the high frequency range, but it is not a major concern in practice because the injection transformer exhibits a relatively slow response, which is dominant in the control loop.

Fig. 6.Measured gain/phase response Gc(s) of inverting amplifier using OPA847 ( R1 =10kΩ , R2 = 5kΩ ).

TABLE IOP AMP SPECIFICATIONS (OPA847)

The output voltage swing and current capability of the OP amp are critical parameters in real implementation. The OP amp should supply sufficient current to magnetize the injection transformer, and the input voltage of the transformer should be within the maximum output voltage swing Vop,max. These requirements are discussed in the next section.

C. Injection Transformer

The output voltage of the OP amp is applied to the primary terminal of the transformer, and the CM noise voltage Vn can be canceled by the secondary voltage Vc connected in series to the CM path. The CM equivalent circuit of the transformer is shown in Fig. 7(a); its simplified form is shown in Fig. 7(b) [12], where

and n = Ns / Np . The symbols C1, C2, C12, Ll1, Ll2, and Lm denote the capacitance of the primary terminal, capacitance of the secondary terminal, inter-winding capacitance, leakage inductance of the primary winding, leakage inductance of the secondary winding, and magnetizing inductance, respectively.

Fig. 7.Circuit model of injection transformer. (a) CM equivalent circuit model. (b) Simplified model.

The transfer function of the injection transformer can be derived from Fig. 7(b) and for Lm >> Llk as

where

As shown in (11) and (12), the leakage inductance Llk and equivalent winding capacitance Ceq are the parameters that determine the bandwidth of the injection transformer. Thus, these parameters should be minimized to reduce high frequency noise.

The maximum output voltage swing and current capability of the OP amp should also be considered in the transformer design. The peak value of the primary voltage Vp, peak should be within the maximum output swing Vop,max of the OP amp; that is,

where Vs,peak denotes the peak voltage of the transformer output. The input current of the transformer is also limited by the maximum output current Iop,max of the OP amp, as shown in (14).

where ZTi denotes the input impedance of the transformer. The primary turns of the transformer can be determined as

where

and μ0 , μr , Ac, and lc denote air permeability, relative permeability, cross-sectional area, and mean length of the magnetic core, respectively.

The selection of core materials is extremely important in miniaturizing transformer size and improving high frequency characteristics. The large number of turns of the transformer winding increases the leakage inductance and winding capacitance, which in turn degrade the frequency response. Hence, the core material should exhibit high permeability to reduce winding turns. A nanocrystalline core with μr = 14,000 is used for the implementation. The parameters of the implemented injection transformer are listed in Table II.

TABLE IIPARAMETERS OF IMPLEMENTED TRANSFORMER

Fig. 8 shows the measured gain/phase response GT(s) of the injection transformer, the poles of which are located at 10.9 MHz. The measured total response Gc(s)GT(s), including the OP amp and transformer, is shown in Fig. 9. Both figures show that the frequency response of the control loop is governed by the transformer characteristics.

Fig. 8.Measured gain/phase response of the transformer GT(s).

Fig. 9.Measured gain/phase response of the OP amp and transformer Gc(s)GT(s).

D. Discussion on Stability

The feed-forward AEF does not have any feedback signals in the control loop. However, we note that in (4), the natural feedback loop is made by the impedance of the noise source Zs. We can rewrite (4) for the voltage VL shown in Fig. 4 as

If the source impedance Zs = 0, then the transfer function is (1 – A), and only the feed-forward loop exists. However, for the case in which Zs ≠ 0 , a feedback loop is made with the forward and feedback gains of (1 – A) and Zs/ZL, respectively. Consequently, the operation of the AEF becomes unstable for a certain condition of gain A.

Figs. 10 and 11 show the Bode diagrams of the loop gain (1 − A)·Zs/ZL for the DC gain of Gc(s)GT(s), (R2/R1)·n = 0.98 and 1, respectively, where Zs =1/ j(2πf ·1nF) , CL = 0.1 uF, and RL = 50 Ω. As shown in Fig. 10, the gain and phase margins are sufficient, and the feedback loop is stable. However, these margins decline rapidly for (R2/R1)·n = 1, as shown in Fig. 11. Therefore, this condition should be avoided for the stable operation of the feed-forward AEF.

Fig. 10.Bode plot of loop gain (1 − A)Zs/ZL for n(R2/R1) = 0.98.

Fig. 11.Bode plot of loop gain (1 − A)Zs/ZL for n(R2/R1) = 1.

 

IV. EXPERIMENTAL RESULTS AND DISCUSSION

Figs. 12 and 13 show the implemented AEF and experimental setup, respectively. The LLC resonant converter with a power rating of 500 W is used for the test, in which the AEF replaces the three-stage passive CM filter. The tested conditions are given in Table III.

Fig. 12.Experimental setup.

Fig. 13.Photograph of implemented AEF.

TABLE IIIEXPERIMENTAL CONDITIONS

Fig. 14 shows the CM noise spectrum without any EMI filter, with the reference line being the EN55022 limit. The CM noise is severe in the frequency range of 150 kHz to 3 MHz. The CM noise spectrum with the three-stage passive filter is shown in Fig. 15, in which the CM noise is below the limit line.

Fig. 14.CM noise spectrum without an EMI filter.

Fig. 15.CM noise spectrum with a three-stage passive EMI filter.

Fig. 16 shows the experimental waveforms of the AEF and CM noise spectrum for (R2/R1)·n = 1, where R1 =10kΩ and R2 = 5kΩ . As shown in Fig. 16(a), the output voltage Vop of the OP amp is saturated to Vop,max and cannot fully cancel the noise voltage. The low frequency spectrum exceeds the desired EMI limit, as shown in Fig. 16(b). As discussed in the previous section, the feedback loop is unstable for this condition, and the output voltage of the OP amp is saturated. This problem can be solved by applying the slightly reduced OP amp gain.

Fig. 16.Experimental results for (R2/R1)·n = 1, R1 =10kΩ , and R2 = 5kΩ . (a) Experimental waveforms of AEF. (b) CM noise spectrum.

Fig. 17 shows the experimental results for (R2/R1)·n = 0.97, where R1 =10.3kΩ and R2 = 5kΩ . The figure clearly shows the significant improvement in the EMI performance. The peak output voltage of the OP amp is within the maximum voltage swing Vop,max , and the sensed noise voltage Vns and OP amp output Vop exhibit nearly the same shape, as shown in Fig. 17(a). Thus, the CM noise voltage is successfully canceled, and an extremely small LISN voltage Vo is observed. The CM noise spectrum for this condition is shown in Fig. 17(b). The noise margin of over 10 dB for the EN55022 limit line can be obtained for the frequency range of 150 kHz–20 MHz. As predicted in the frequency response of Gc(s)GT(s) shown in Fig. 10, the AEF performance declines at a frequency of over 15 MHz because of the limited bandwidth of the injection transformer. In the frequency range below 20 MHz, the performance of the implemented AEF is comparable to that of the three-stage passive filter shown in Fig. 15. A noise voltage above this range can be reduced by employing a small high-frequency CM choke.

Fig. 17.Experimental results for (R2/R1)·n = 1, R1 =10.3kΩ , and R2 = 5kΩ . (a) Experimental waveforms of AEF. (b) CM noise spectrum.

 

V. CONCLUSIONS

This study presented the design and implementation of an AEF for the CM noise reduction of switching power converters. A VSVC-type feed-forward AEF was considered for the implementation and used to replace the three-stage passive EMI filter. The practical considerations for the implementation of the filter were provided. Such considerations include the output voltage swing and current capability of the OP amp, the frequency characteristics of the filter components, and the stability of the feed-forward AEF. The operation and performance of the implemented AEF were investigated through the experimental works for an LLC resonant converter with a power rating of 500 W. The experimental results show that a noise margin of over 10 dB for the EN55022 limit can be obtained for the frequency range of 150 kHz–20 MHz by employing the implemented AEF. The designed AEF can replace bulky passive filters and be successfully used to reduce low-frequency EMI.