INTERNATIONAL JOURNAL OF CIRCUIT THEORY AND APPLICATIONS Int. J. Circ. Theor. Appl. (2017) Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/cta.2368
Direct synthesis technique (DST) for complex general Chebyshev filters
Fei Xiao*,†
School of Electronic Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China
SUMMARY Recently, we discussed the concept of direct synthesis technique (DST), in which real‐coefficient filtering polynomials containing all information of the filters to be synthesized are derived directly for realization, and they could find applications in the design of lumped‐element LC filters, active RC filters, and infinite impulse response digital filters. In this paper, another DST for complex general Chebyshev bandpass filters is discussed, which is based on a complex mapping relation and featured by complex‐coefficient filtering polynomials. It is called as complex DST in this paper. Compared with real‐coefficient filtering polynomials whose polarities are determined by the number of their zeros at zero frequency, the polarities of complex‐ coefficient filtering polynomials can be easily changed by multiplying imaginary unit j. Such advantage might make their realization more flexible. The analysis shows that conventional coupling matrix could be considered as narrow‐band approximation of network matrix derived by complex DST in the normalized frequency domain. In order to demonstrate the validity of complex DST in this paper, it is applied in the design of classic parallel‐coupled microstrip bandpass filters. Compared with conventional synthesis techniques, complex DST could find out better dimensions and provide more choices for realization and synthesize both even‐order and odd‐order parallel‐coupled microstrip bandpass filters. Copyright © 2017 John Wiley & Sons, Ltd.
Received 7 December 2016; Revised 25 April 2017; Accepted 27 April 2017
KEY WORDS: complex mapping relation; complex direct synthesis technique; direct synthesis technique; complex general Chebyshev filter; network matrix
1. INTRODUCTION
Filter synthesis has gone through at least 100‐year development since Karl Willy Wagner and Campbell independently proposed their filters in 1915 [1, 2]. Up to now, there arise various filter types such as Butterworth filters, Chebyshev filters, and Bessel filters [3–5]. Conventional filter synthesis techniques usually start from the derivation of filtering polynomials of lowpass prototypes according to the specifications of the filters to be synthesized. Lowpass prototypes are defined as the ones whose element values are scaled to make source/load resistance/conductance and the cutoff frequency equal to be unity. They are usually realized in ladder network forms. After applying lowpass to bandpass or other frequency transformations on ladder networks of lowpass prototypes, realization networks can be presented for the filters to be synthesized. Although conventional techniques are applied successfully in practice, there are still some problems with them. For example, they are only effective for the cases that frequency response of lowpass prototypes is symmetric about zero frequency. Otherwise, it is unlikely for them to be realized in
*Correspondence to: Fei Xiao, School of Electronic Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, 611731, China. †E‐mail: [email protected]; [email protected]
Copyright © 2017 John Wiley & Sons, Ltd. F. XIAO lumped LC networks [4]. Recently, we discussed the concept of direct synthesis technique (DST); there, we derive filtering polynomials containing all information of a filter in its own frequency domain for realization [6–11]. We therefore will call conventional synthesis techniques as indirect synthesis technique (IST) for comparison. However, they can also be extended to DST, as performed in [7]. DST for general Chebyhsev filters is characterized by its capability to place transmission zeros anywhere in order to control the performance of filters. Ordinary Chebyshev filters are actually a special case with all transmissions at zero frequency or infinite frequency. Meanwhile, Elliptic filters are also the special cases of general Chebyhsev filters with transmission zero at specific frequencies. At high frequency such as RF/microwave/optical frequency ranges, filters are realized through some transmission‐line structures with distributed‐element effect, such as waveguide, microstrip, stripline, suspended stripline, slotline, coplanar waveguide, and finline. Undoubtedly, electromagnetic theory can be used to describe them accurately but only certain simple solutions exist. In practice, lumped‐ element filter synthesis techniques are usually used to calculate initial dimensions. For example, the concept of coupling matrix is discussed in [12–15], which is used to describe filters in the same coupling topologies and provides the information of Q factor and coupling coefficients for the determination of their initial dimensions. In essence, these parameters are derived in the normalized frequency domain and should be transformed into the bandpass domain for use. Unfortunately, such approach is only accurate for narrow‐band filters [16]. In this paper, another DST for complex general Chebyshev bandpass filters is proposed. It is different from the one in [7] in that it is based on a real mapping relation, derives real‐coefficient filtering polynomials, and can be realized through lumped‐element LC networks. The DST in this paper is based on a complex mapping relation and is called as complex DST for distinction. Thus, its filtering polynomials have complex coefficients, and they can be represented by lumped‐elements CB or LX models where C is capacitor, B is frequency‐invariant susceptance, L is inductor, and X is frequency‐invariant reactance. The polarities of complex‐coefficient filtering polynomials can be easily changed by multiplying imaginary unit j, which makes their realization more flexible. A network matrix could be constructed to represent coupling schemes of a filter. Conventional coupling matrix is considered as narrow‐band approximation of network matrix obtained by complex DST. To demonstrate the validity of the proposed complex DST, it is applied in the design of classic parallel‐coupled microstrip bandpass filters. Several examples with different order and specifications are presented.
2. FILTERING FUNCTIONS
Note that only a lossless network is considered in this paper. The following scaled s‐plane to z‐plane complex mapping relation is used [17].
s−jω z2 ¼ u ; (1) s−jωd where z is a temporal complex variable. ωu and ωd are the upper and lower passband edge frequencies of the filter to be synthesized. For simplicity, a characteristic frequency ωc is used for scaling. Thus, s ¼ s=ωc ¼ jω=ωc, ωu ¼ ωu=ωc, and ωd ¼ ωd=ωc. Here, only transmission zeros at positive frequencies are taken into consideration. If there are Np transmission zeros at zero frequency, Nm at finite frequencies and Nl at infinity, the total number of transmission zeros considered should be
N ¼ N p þ N m þ N l: (2)
This is also defined as the order of the filter to be synthesized. These transmission zeros can be mapped in the z‐plane by using (1). For example, the kth transmission zero sk ¼ jωk is mapped to the point zk in the z‐plane. If the procedure in [7] is followed, a prototype function exhibiting the
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta DST FOR COMPLEX GENERAL CHEBYSHEV FILTERS required features is constructed in the z‐plane and mapped into the s‐plane through (1). Finally, filtering polynomials such as reflection polynomial FðÞs , transfer polynomialPðÞs , and common polynomial EðÞs are derived through the following equations.
N N 2 2k Ev ∏ ðÞzk −z ¼ ∑ d2k z ; (3) k¼1 k¼0
N k N−k β ⋅ FðÞ¼s ∑ d2k ðÞs − jωu ðÞs − jωd ; (4) k¼0
N m N p ε⋅PðÞ¼s P0ðÞ¼s s ⋅ ∏ ðÞs − sk ; (5) k¼1
where Ev means the operation of taking even part of a function. zk is the corresponding point of the kth ‐ fi ∏N ðÞ− 2 β transmission zero in the z plane. d2k is the coef cient obtained by expanding Ev k¼1 zk z . is used to make the coefficient of the highest‐order term in FðÞs to be unity. ε can be determined by the specified ripple or return loss at the edge frequency of the passband. The common polynomial EðÞs can be obtained through the following equality owing to the fact that only a lossless network is considered in the paper.
P⋅P þ F⋅F ¼ E⋅E : (6)
3. TRANSVERSAL NETWORK
We discussed how to derive complex‐coefficient filtering polynomials of a complex general Chebyshev bandpass filter in the previous section. In this section, how to realize them through lumped‐element networks will be addressed. Mathematically speaking, it is an inverse problem. Up to now, there are three approaches available, that is, the element‐based synthesis method, the block‐ based synthesis method, and the whole synthesis method [10]. Here, the whole synthesis method in [15] is applied. In Figure 1, a canonical N‐resonator transversal network consisting of lumped‐ element CB resonators and admittance inverters J is presented, which is called the CB model for simplicity. Its dual network is depicted in Figure 2, which consists of lumped‐element LX resonators and impedance inverters K. The network in Figure 2 can be called the LX model for simplicity. In the following, only the CB model in Figure 1 is discussed and its dual network in Figure 2 can be derived in a similar way. The CB model consists of N + 1 paths between the source and the load. One is the direct source/load coupling realized through the inverter JSL, and each of the N paths left is composed of a resonator consisting of a capacitor Ck and a frequency‐invariant susceptance jBk, an admittance inverter JSk between the source and the resonator, and an admittance inverter JLk between the resonator and the load. The susceptances BS and BL allow for some special cases such as antimetric cases. Different from conventional synthesis techniques in [13–15] that are actually the representation in the normalized frequency domain, the networks in Figures 1 and 2 are the representations in the bandpass domain. In addition, they are also different from that in [7], in which the LC model is used. If the circuit in Figure 1 is fed by a current source iS with internal conductance GS and the load conductance is GL, the relation of the voltage across the resonators vk, the voltage at the source vS, and the voltage at the load vL can be described by the nodal equations.
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta F. XIAO
Figure 1. N‐resonator transversal network consisting of lumped‐element CB resonators and admittance inverters J.
Figure 2. N‐resonator transversal network consisting of lumped‐element LX resonators and impedance inverters K.
N ðÞþ þ ∑ þ ¼ ; GS BS vS jJ Sk vk jJ SLvL iS k¼1 ðÞþ þ þ ¼ ; ¼ ; … ; sCk jBk vk jJ Sk vS jJ LkvL 0 k 1 N (7) N þ ∑ þ ðÞþ ¼ ; jJ SLvS jJ Lk vk GL BL vL 0 k¼1 where s is complex frequency variable. These equations can be rewritten in matrix form
½ A ⋅½ ¼v −j ⋅ ½ i : (8)
Here, [v] is the voltage vector, [i] the current vector, and [A] the network matrix.
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta DST FOR COMPLEX GENERAL CHEBYSHEV FILTERS
On the other hand, it is easy to obtain the admittance matrix of the transversal network.
2 3 N 2 N þ ∑ J Sk þ ∑ J Sk J Lk 6 BS jJ SL 7 6 ¼ jðÞω−ωk ¼ jðÞω−ωk 7 ½ y ¼ 6 k 1 k 1 7: (9) C 4 N N 2 5 þ ∑ J Sk J Lk þ ∑ J Lk jJ SL ðÞω−ω BL ðÞω−ω k¼1 j k k¼1 j k
Here, the capacitance value of the kth resonator is set to be unit for simplicity and thus the susceptance value Bk is determined by the resonant frequency ωk of the kth resonator.
4. SYNTHESIS PROCEDURE
According to [7], the polarities of both P(s) and F(s) should be opposite when a filter is to be realized in symmetric form. For DST in [7], F(s) is always even. The polarity of P(s) is determined by the number of its zeros at zero frequency if only imaginary transmission zeros are considered. When there is even number of transmission zeros at zero frequency, both P(s) and F(s) are even. Thus, it is unlikely for them to be realized through symmetric networks. Fortunately, filtering polynomials derived by complex DST in this paper are complex‐coefficient, which means that their polarities can be changed through the multiplication of the imaginary unit j. Then, we have the following notes to support these comments.
Note 1: If a function Q(s) is expressed as
n QsðÞ¼a ∏ ðÞs−jωk ; (10) k¼1 where s is complex frequency variable and both a and ωk are real, the polarity of Q(s) is determined by n. In other words, Q(s) is even if n is even or odd if n is odd.
Note 2: The polarity of Q(s) can be changed inversely by multiplying imaginary unit j.
Complex DST can be applied in the design of transmission‐line filters, and the synthesis flowchart is shown in Figure 3. According to filter specifications, the location of transmission zeros and return loss in passband are determined first. Then, filtering polynomials are derived by using (3)–(5). If symmetric networks are used to realize the desired frequency response, the polarities of both reflection and transfer polynomials should be opposite; otherwise, they can be changed by multiplying imaginary unit j in practice. Further, the corresponding admittance parameters can be derived, written in the form of partial fraction expansion and compared with those in (9) derived from the CB or LX models. Accordingly, we can determine those unknown parameters such as the resonant frequencies, and the values of the inverters. Finally, the following scaled transversal network matrix for the CB model is written as
2 3 −jGðÞS þ BS J S1 ⋯ J SN J SL 6 7 6 ω−ω ⋯ 7 6 J S1 1 0 J L1 7 6 7 A ¼ 6 ⋮ 0 ⋱ 0 ⋮ 7: (11) 6 7 4 5 J SN ⋮ 0 ω−ωN J LN
J SL J L1 ⋯ J LN −jGðÞL þ BL
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta F. XIAO
Figure 3. Synthesis flowchart of complex direct synthesis technique.
Similarly, the following transversal network matrix for the LX model in Figure 2 can be obtained. 2 3 −jRðÞS þ X S KS1 ⋯ KSN KSL 6 7 6 ω−ω ⋯ 7 6 KS1 1 0 KL1 7 6 7 A ¼ 6 ⋮ 0 ⋱ 0 ⋮ 7: (12) 6 7 4 5 KSN ⋮ 0 ω−ωN KLN
KSL KL1 ⋯ KLN −jRðÞL þ X L
The prototype network matrix is defined as the one with source/load resistance/conductance equal to be unity and scaled by a characteristic frequency ωc. This matrix is represented by global eigenresonances. By applying matrix operation such as similarity transformation, it can be transformed into the desired topology corresponding to transmission‐line filter to be deigned. Therefore, the equivalence between transmission‐line filter to be designed and its lumped‐element network is set up, and then we can calculate initial dimensions of the transmission‐line filter according to its specifications. Finally, its performance can be optimized to meet specification requirement by adjusting its dimensions in vicinity of their initial values.
5. SYNTHESIS EXAMPLES
In this section, we will present an example to demonstrate the synthesis procedure of complex DST discussed in the previous sections. It is a fourth‐order general Chebyshev bandpass filter whose passband covers the range from 4 to 6 GHz (40% fractional bandwidth) and return loss in the passband of −20 dB. Two transmission zeros are at zero frequency, the third one at 6.5 GHz, and the fourth one at infinity. The synthesis process starts with the derivation of filtering polynomials. Generally speaking, the characteristic frequency ωc could be chosen arbitrarily if only it can help simplification. 9 Here, ωc =2π ×10 rad/s. Then, the four transmission zeros ωzk (k = 1, 2, 3, and 4) are mapped into the z‐plane by (1), and then we can obtain
4 2 8 6 4 2 Ev ∏ ðÞz−zzk ¼ z þ 26:1686z þ 56:9607z þ 18:0370z þ 0:45: (13) k¼1
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta DST FOR COMPLEX GENERAL CHEBYSHEV FILTERS
By using (4) and (5), we can obtain PðÞs , FðÞs , and EðÞs as follows.
FðÞ¼s s 4 −20:1799 j⋅s3 −151:6734⋅s2 þ 503:1035 j⋅s þ 621:3091; (14a)
PðÞ¼s −0:0388⋅s3 þ 0:2521 j⋅s2; (14b)
EðÞ¼s s 4 þ ðÞ2:0991− j20:1799 ⋅s3 þ ðÞ−149:4712− j31:9230 ⋅s2 : (14c) þ ðÞ−159:0200− j480:5466 ⋅s þ ðÞ564:8060 þ j258:8808
Because PðÞs is odd and FðÞs is even, this fourth‐order bandpass filter can be realized through a symmetrical network. The derivation of transversal network matrix is similar to that in [15], and then we can determine the prototype network matrix of this example. 2 3 −jGðÞþ B J J J J J 6 S S S1 S2 S3 S4 SL 7 6 ω−ω 7 6 J S1 1 000 J L1 7 6 7 6 J 0 ω−ω 00 J 7 A ¼ 6 S2 2 L2 7: (15) 6 ω−ω 7 6 J S3 00 3 0 J L3 7 6 7 4 J S4 000ω−ω4 J L4 5
J SL J L1 J L2 J L3 J L4 −jGðÞL þ BL where GS = GL =1,BS = BL =0,J S1 ¼ 0:4153,J S2 ¼ 0:6413, J S3 ¼ 0:6017, J S4 ¼ 0:3224, ω1 ¼ 3:6709, ω2 ¼ 4:4465, ω3 ¼ 5:8243, ω4 ¼ 6:2381, J L1 ¼ 0:4153, J L2 ¼ −0:6413, J L3 ¼ 0:6017, J L4 ¼ −0:3224, and J SL ¼ 0. The frequency response of the network matrix is shown in Figure 4, which is indistinguishable from that obtained from the filtering polynomials. Interestingly, the bandpass frequency response is only formed in the positive frequency range, which is different from those in [7]. Thus, the filters synthesized by complex DST in this paper could be called complex filters while those by the DST in [7] could be called real filters, according to [17–19].
6. RELATION BETWEEN THE COMPLEX DST AND THE IST
In this section, we will compare complex DST with IST in [13–15] to reveal some interesting features underlying them. According to [13–15], conventional coupling matrix can be written as the following form.
Figure 4. Frequency response of the fourth‐order complex general Chebyshev bandpass filter.
Copyright © 2017 John Wiley & Sons, Ltd. Int. J. Circ. Theor. Appl. 2017 DOI: 10.1002/cta F. XIAO 2 3 − ⋯ jGS M S1 M SN M SL 6 7 6 M Ω þ M M ⋯ M 7 6 S1 11 12 L1 7 6 7 ½ ¼M 6 ⋮ M 12 ⋱ M 1N ⋮ 7; (16) 6 7 4 M SN ⋮ M 1N Ω þ M NN M LN 5 ⋯ − M SL M L1 M LN jGL
where Ω is the angular frequency variable in the normalized frequency domain and Mij the normalized coupling coefficient. The entries in this matrix are derived from lowpass prototypes and effective in the normalized frequency domain. In practice, they should be transformed into the bandpass domain for practical use by using the conventional frequency transformation.