Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 Rise time analysis of pulsed klystron-modulator for e$ciency improvement of linear colliders J.S. Oh!,*, M.H. Cho!, W. Namkung!, K.H. Chung", T. Shintake#, H. Matsumoto# !Pohang Accelerator Laboratory, POSTECH, San 31 Hyoja-dong, Nam-ku, Pohang, South Korea "Department of Nuclear Engineering, Seoul National University, Seoul, South Korea #National Laboratory for High Energy Physics, Tsukuba, Japan Received 20 April 1999; received in revised form 21 August 1999; accepted 18 October 1999 Abstract In linear accelerators, the periods during the rise and fall of a klystron-modulator pulse cannot be used to generate RF power. Thus, these periods need to be minimized to get high e$ciency, especially in large-scale machines. In this paper, we present a simpli"ed and generalized voltage rise time function of a pulsed modulator with a high-power klystron load using the equivalent circuit analysis method. The optimum pulse waveform is generated when this pulsed power system is tuned with a damping factor of &0.85. The normalized rise time chart presented in this paper allows one to predict the rise time and pulse shape of the pulsed power system in general. The results can be summarized as follows: The large distributed capacitance in the pulse tank and operating parameters, < ]¹ , where < is load voltage and ¹ is the pulse 4 1 4 1 width, are the main factors determining the pulse rise time in the high-power RF system. With an RF pulse compression scheme, up to $3% ripple of the modulator voltage is allowed without serious loss of compressor e$ciency, which allows the modulator e$ciency to be improved as well. The wiring inductance should be minimized to get the fastest rise time. ( 2000 Elsevier Science B.V. All rights reserved. 1. Introduction An e#e-linear collider at 500 GeV (center of mass energy) has been proposed as a future accelerator. It is a large-scale machine. For instance, the C-band scheme requires more than 4000 units of klystrons and matching modulators [1]. The initial design value of the wall plug power for the whole RF system in the C-band scheme was 150 MW. However, the previous e$ciency analysis of the "rst C-band system shows that the present system re- * Corresponding author. Tel.: #82-562-279-1143; fax: #82562-279-1199. E-mail address: jsoh@postech.ac.kr (J.S. Oh). quires 270 MW AC power due to the conservative design [2]. Therefore, the e$ciency improvement of the pulsed power system is a key concern to reduce operational cost in future large-scale accelerators. The pulse tail is a result of stored energy decay in the distributed elements of the pulse generator. In a short pulse generator, most of this energy comes during the rising period. Therefore, the rise time of the modulator pulse has to be minimized to achieve high e$ciency. In this paper, we derive a simple and generalized form of the pulse rise time characteristics of a modulator with a high-power klystron load. As a result, the practical limits on the rise time and reasonably achievable values are described. 0168-9002/00/$ - see front matter ( 2000 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 8 - 9 0 0 2 ( 9 9 ) 0 1 1 7 1 - 7 224 J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 An RF pulse compressor such as SLED increases the peak RF power for a "xed number of joules per pulse [3}5]. The RF phase ripple and the energy storage e$ciency of the pulse compressor are analyzed in detail, and the voltage ripple requirement is revaluated. The rise time analysis of the C-band system is included as an example case. 2. Optimum pulse waveform In a conventional modulator, a line-type PFN (pulse forming network) generates a high-voltage pulse through a step-up pulse transformer. This pulsed power system can be modeled by the circuit shown in Fig. 1. The parameters in the "gure are referenced to the secondary side of the pulse transformer. We can neglect the e!ect of the shunt resistance R and the shunt inductance ¸ during the % P short rise time in the leading edge analysis. The analytical solution of the load voltage < (t) L is described elsewhere [6}8]. The normalized load voltage y(t) de"ned as A < (t) 1#m y(t)" L < m G is given by B (1) CA B y(t)"1!exp(!at) D a sinh(kt)#cosh(kt) k Fig. 2. Normalized rising pulse waveform as a function of damping factor, p. (p'1) (2) CA B y(t)"1!exp(!at) D a sinh(ut)#cosh(ut) u (p(1) (3) where 2pp a" , q Fig. 1. Equivalent circuit of a step-up pulse system. < is the G PFN voltage. S is the main switch. R is the PFN impedance. G G ¸ is the wiring inductance. ¸ is the leakage inductance of the W L pulse transformer. R is the shunt resistance of the pulse trans% former. ¸ is the shunt inductance of the pulse transformer. P C is the distributed capacitance of the pulse transformer. C is D L the distributed capacitance of the load including the klystron. R is the load impedance. L 2pJp2!1 2pJ1!p2 k" , u" q q A S B m 1 1 p" cm# , q"2p J¸ C T T m#1 c 2Jm(m#1) S ¸ R Z T, m" L , c" T , Z " T C R R T G L C "C #C . T D L The expanded view of the leading edge of y(t) near the #at top is shown in Fig. 2 as a function of the damping factor p. In this "gure, the normalized time S is de"ned by S"t/q. The pulse shape and the rise time are sensitive to the damping factor p near the #at top. For a given value of ¸ C , small T T p gives fast rise time but generates large overshoot. From Eq. (3), the maximum overshoot is located at ut"n, and the corresponding peak value is given by AB y L "¸ #¸ T W L A A p a "1#exp ! p u u B B p "1#exp ! p . J1!p2 (4) J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 225 In general, a pulse #at-top with less than $0.5% ripple is required to produce high-power RF with small phase variations in high-energy linear accelerators. The damping factor p has to be larger than 0.86 to limit the overshoot to less than 0.5% during the #at-top. So the optimum pulse waveform is generated when the pulsed power system is tuned with a damping factor of 0.86. 3. Rise time of the pulse Fig. 3 shows the normalized rise time S as 3 a function of p for three amplitude fractions: 10}90% (S ), 0 to 95% (S ), and 0 to 99% (S ). 31 32 33 The rise time S is the standard de"nition that 31 characterizes the leading edge. However, the rise time ¹ given by S is more suitable to estimate 3 33 the rise time e$ciency for a given #at-top width ¹ : 3& ¹ 3& . rise time e$ciency" ¹ #¹ 3 3& Fig. 3. Normalized rise time as a function of p. (5) The PFN impedance is normally made equal to the load resistance to get the maximum energy transfer e$ciency [6]. For a matched PFN (m"1), the rise time ¹ is given by 3 ¹ "S (p)q"J2pS (p)J¸ C . 3 33 33 T T (6) The klystron load is a nonlinear device since the klystron current I is given by I "k<1.5 where k is k k k the perveance and < is the klystron voltage. This k means that the klystron impedance is continuously changing during the pulse rise time. Fig. 4 shows the rising edge of the waveform for a speci"c case, p"0.9. These results were obtained using the PSPICEt circuit simulation code. The klystron voltage is rising faster than the resistor voltage, but it approaches the #at-top more slowly with less overshoot. As a result, the rise time for the klystron load is longer. With the klystron load, a damping factor of 0.8 yields the fastest rise time for less than 0.5% overshoot. We denote this optimum value by p . 015 Fig. 5 shows the correction factor g(p) of the rise time with a klystron load, that is, the ratio of the Fig. 4. Rising waveform for a resistor and a klystron load. rise time with a klystron load to the rise time with a resistor equal to the klystron impedance at full voltage. These values of g(p) were found by simulating a range of p values with the PSPICE code. The correction factor is rather slow varying and changes less than 20% over the practical range of p. A third-order polynomial "t to the correction factor yields g(p)"!1.918p3#1.057p2#3.370p!1.331. (7) The rise time with a klystron load, ¹ , is thus k ¹ "g(p)¹ "J2ng(p)S (p)J¸ C . k r 33 T T (8) 226 J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 Fig. 5. Rise time correction factor for a klystron load. Fig. 7. Example of pulse rise time versus damping factor. Fig. 6. Equal rise time curves and damping factors. Equal rise time curves as a function of total shunt capacitance C and total series inductance ¸ are T T plotted in Fig. 6 using (8). There are two constant lines for each p in Fig. 6 because these are two possible values of c for each value of p: A B 1 1 p" c# . c 2J2 (9) Since the discrete PFN is not a perfect transmission line, it generates ringing in the leading edge. Therefore, the relatively small series inductance in a low c system does not help much to reduce the oscillation amplitude. Also, in a high-voltage pulser, it is hard to acheive low c due to the large leakage and wiring inductance associated with the high step-up ratio. For these reasons, we ignore the lower c solutions in Fig. 6 which are the lines with smaller slopes. In order to generalize the rise time curves, shunt capacitance and series inductance are normalized using the load resistance R . The range of the L horizontal and vertical axes correspond to the practical range of series inductances (300}1200 lH), shunt capacitances (100}300 pF), and load resistances (700}1,400 )). For given values of R , C , L T and ¸ , one can easily evaluate the rise time of T a pulsed power system and the magnitude of overshoot. For example, system &A' in Fig. 6 has a rise time of 1.5 ls with p"0.9. From this point, one could reduce the rise time to 1.1 ls by decreasing the series inductance by about 30% (&B'). However, making the inductance even smaller to shorten the rise time (&C') yields too large of an overshoot (greater than 0.5%). Fig. 7 shows typical pulse waveforms for the C-band modulator. The waveform with the slower rise time (1.9 ls) was measured initially. The rise time was reduced to 1.2 ls by reducing the wiring inductance. This also yielded a faster fall time. 4. Minimum rise time The equivalent circuit parameters of the pulse transformer are related to geometrical and material J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 parameters. The constant gradient or tapered bi"lar secondary winding technique is widely used in high-voltage pulse transformers [9,10]. The leakage inductance, distributed capacitance, and shunt inductance of the secondary side are given by pN2*; # ¸ " S L l # [nH] A B n!1 2; l ## C "0.0885 e D 3 n * N2A ¸ "4pk S . P 3 l . (10) [pF] (11) [nH] (12) where N is the number of turns in the secondary 4 coil, * is the insulation distance between layers in cm, ; is the average circumference of the layers # in cm, l is the winding length in cm, e is the # 3 relative dielectric constant of the insulating material between the layers, n is the step-up ratio, k is 3 the relative pulse permeability of the core, A . is the cross-sectional area of the core in cm2, and l is the mean magnetic path length of the . core in cm. The pulse magnetization of the core is given by *BN A "108< ¹ , 4 . 4 1 (13) where *B is the average increment of the magnetic #ux density of the core in gauss, < is the load 4 voltage in volts, and ¹ is the pulse width in sec1 onds. Using Eqs. (10), (11) and (13), the transformer time constant g is given by A BA B ; # g"J¸ C "1.67]10~3Je L D 3 A . A B 1 ](< ¹ ) . 4 P *B n!1 n (14) The time constant is proportional to operating parameters, < ]¹ , geometric parameters, ; ]A~1, 4 1 # . and material parameters, e0.5]*B~1. For a given r c, total series inductance is given by ¸ "¸ #¸ "(C #C ) (cR )2. T W L D L L (15) 227 Using Eqs. (14) and (15), ¸ and C are given by L D 1 ¸ " [(a2C !¸ ) L L W 2 AB #J(a2C !¸ )2#4a2g2] L W 1 C " [!(a2C !¸ ) D L W 2a2 (16) A B (17) #J(a2C !¸ )2#4a2g2] L W where a"cR . Using (15), (8) can be written as L ¹ "J2n g(p)S (p)cR (C #C ). 33 L L D k Substituting Eq. (17) yields (18) C p ¸ ¹ " g(p)S (p) aC # W k J2 33 L a SA # B D 2 ¸ aC ! W #4g2 . L a (18a) With optimum parameters, p ("0.8) and 015 c ("1.66), Eq. (18) can be simpli"ed as follows: 015 ¹ "4.77R (C #C ). (19) k L L D With "xed C and a, C "0 yields minimum rise L D time ¹ "4.77R C . (20) .*/ L L The actual rise time is always larger than the value given by Eq. (20). This equation gives a useful measure of the minimum rise time. In general, to get a fast rising pulse for given values of a and C , ¸ and g should be minimized. L W This follows from Eq. (18a). The primary wiring inductance is practically limited to &1.0 lH due to the dimensions and layout of the high-voltage components. From Eq. (14), g can be reduced by increasing core cross-sectional area because ; is # proportional to A0.5. Using Eqs. (12) and (13), the . pulse droop D is given by A BA B R ¹ R 10~7 *B2 L (A l ). D" L P " (21) .. 2¸ k <2¹ 8p P S P 3 A large cross-sectional area produces a large pulse droop so it has to be limited by the droop requirement. 228 J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 Table 1 Rise time parameters of the C-band modulator Load parameters Matching parameters Rise time factors Insulator parameter Transformer parameters Minimum rise time Fig. 8. Distributed capacitance and potential lines in the pulse tank with E3746 C-band klystron. To estimate the correct rise time, it is very important to have a realistic measure of the load capacitance C . Fig. 8 shows the results of the L distributed capacitance obtained using the DENKAI2 code [11] for the pulse tank and klystron assembly of the C-band system [12]. The right half of the "gure shows electrostatic potential lines in R "1104 ), C "120 pF, L L < "350 kV, ¹ "3.5 ls S P m"1, p "0.8, c "1.66 .*/ .*/ S (p )"0.61, g(p )"1.06 r3 .*/ .*/ e "2.2 r n"15, *B"1.8 T ¹ "0.63 ls .*/ the region of the klystron tube and between the coils of the pulse transformer in the tank. The calculated capacitance of each section for the C-band system is shown in the left half of the "gure. The total distributed load capacitance C is L 115 pF. If one adds the capacitance between the corona ring and other structures, which are not included in this estimation, it becomes approximately 120 pF. Table 1 lists the rise time parameters for the C-band klystron-modulator. It is assumed that the transformer is immersed in the insulation oil (e "2.2). Due to the distributed load capacitance 3 of 120 pF, the rise time is larger than ¹ ("0.63 ls) given by Eq. (20). Table 2 shows .*/ the calculated result of the rise time and corresponding transformer parameters. It is possible to get the rise time less than 1 ls with a primary wiring inductance of 2 lH. A rise time e$ciency of about 75% with a #at-top of 2.5 ls is achievable with 1 lH wiring inductance. If the wiring inductance is relatively large, smaller leakage inductance and larger distributed capacitance are required (see Table 2 Rise time evaluation of the C-band modulator Core dimension g [ls] ¸ [lH] W ¸ [lH] L C [pF] D ¹ [ls] k 4 cores (5]5 cm2) A "100 cm2 . ; "72 cm C 4 cores (7]7 cm2) A "196 cm2 . ; "92 cm C 0.11 1.0]n2 314 41 0.85 0.07 2.0]n2 185 69 1.00 1.0]n2 251 22 0.75 2.0]n2 114 48 0.89 J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 Eqs. (16) and (17)). From Eqs. (10) and (11), this can be done by increasing the winding length although this increases C . It means that the transformer D size has to be increased to meet the requirement. However, the rise time cannot be improved signi"cantly even with a core twice as large. Therefore, the rise time is determined by the distributed load capacitance and the transformer time constant which is proportional to the operating parameters, < ]¹ . 4 1 5. Relaxation of ripple requirement The tight ripple speci"cation of the #at-top can be relaxed if a pulse compressor is used. This is a common technique in electron linear accelerators to increase peak power. For purpose here, it can be thought of as a narrow band RF storage device. The klystron output RF can be described by v(t)"A(t)sin(2pf t#*/(t)), (22) # where f is the operating frequency, */(t) is the # phase ripple due to the voltage ripple, and A(t)2 is proportional to the instantaneous RF power. The maximum RF phase ripple $*/ of the klystron . tube is given by A B A BA B */ ¸ e< *< . "! 0 (c2!1)~1.5 4 . , (23) % 2p j m c2 < 0 % 4 where c "1#eV /(m c2), ¸ is the tube length % 4 % 0 from input cavity to output cavity of the klystron, j is the free space wavelength of the klystron 0 operating wave, m is the electron rest mass, and % $*< is the maximum allowable #at-top ripple. . Then, Eq. (22) can be written as v(t) u(t)" "sin(2pf t#*/ sin(2pf t)) # . 3 A(t) (24) where f is the fundamental ripple frequency. Using 3 the following relations, sin(2pf t#*/ sin(2pf t)) # . 3 "sin(2pf t)cos(*/ sin(2pf t)) c . 3 #cos(2pf t)sin(*/ sin(2pf t)) # . 3 229 cos(*/ sin(2pf t)) . 3 "J (*/ )#2J (*/ )sin(4pf t) 0 . 2 . 3 #2J (*/ )sin(6pf t)#2 4 . 3 sin(*/ sin(2pf t)) . 3 "2J (*/ )sin(2pf t) 1 . 3 #2J (*/ )sin(6pf t)#2. 3 . 3 Eq. (24) can be written as u(t)"J (*/ )sin(2pf t) 0 . # #J (*/ )[sin(2p(f #f )t) 1 . # 3 #sin(2p(f !f )t)] c 3 #J (*/ )[sin(2p(f #2f )t) 2 . # 3 #sin(2p(f !2f )t)]#2 (25) # 3 where J is the Bessel function of the "rst kind with n order n. The #at-top oscillation is caused by the wave re#ections between the primary and the secondary transformer coils [6]. The fundamental frequency of this oscillation is usually larger than 1 MHz. The pulse compressor has a high unloaded Q of &105, a loaded Q of &104, and a narrow o L bandwidth of less than a few 100 kHz, which means that the energy in the side band terms in Eq. (25) cannot be stored in the pulse compressor. Therefore the reduction of the storage e$ciency of the pulse compressor is simply given by J (*/ )2. 0 . Typical parameters of the C-band system are summarized in Table 3. As shown in Table 3, even though there is $3% voltage ripple and $8.63 phase ripple, an energy storage e$ciency of 99% is achievable. Of course, the ripple after the phase #ip that is used to discharge the pulse compressor should be less than 1% because the beam acceleration is done during this period. However, it is not di$cult to adjust the #at-top near the pulse tail using tunable inductors in the PFN. In the matched PFN, the maximum possible overshoot is 4.3% which occurs for p"0.71 (i.e., c"1). Thus one can reduce the rise time without serious degradation of the energy e$ciency by using a pulse compressor and having the pulsed power system tuned to as small a damping factor as possible. Table 4 shows the rise time assuming a maximum ripple less than $3%, and using 230 J.S. Oh et al. / Nuclear Instruments and Methods in Physics Research A 443 (2000) 223}230 Table 3 RF phase ripple and pulse compressor e$ciency Operating frequency, f # Cavity distance, ¸ 0 Operating voltage, < 4 Voltage ripple, $*< Phase ripple, $*/ Pulse compressor e$ciency, J (*/ )2 0 . 5,712 MHz 30 cm 350 kV $3% $8.63 0.99 < ]¹ , are the main factors determining the pulse 4 1 rise time in the high-power RF system. With the pulse compression scheme, up to $3% voltage ripple is allowable without serious loss in e$ciency, which allows the modulator e$ciency to be improved as well. References Table 4 Rise time with a pulse compressor (p "0.75) .*/ Core dimension g [ls] ¸ [lH] W ¸ [lH] L C [pF] D ¹ [ls] k 4 cores (5]5 cm2) A "100 cm2, ; "72 cm . C 0.11 1.0]n2 208 61 0.65 2.0]n2 111 145 0.84 S and the parameters, p "0.75, c "1.4, 32 .*/ .*/ S (p )"0.53, g(p )"0.98. With the looser 32 .*/ .*/ ripple constraints, the rise time decreases from 1.0 to 0.84 ls for 2.0 lH wiring inductance. Thus, for a #at-top of 2.5 ls, a rise time e$ciency of about 80% is achievable with 1 lH wiring inductance. 6. Conclusion The generalized rise time chart presented in this paper is useful for predicting the rise time and pulse shape of a pulsed power system. The minimum rise time and corresponding circuit parameters can be evaluated using the simpli"ed equations with reasonable accuracy. The large distributed capacitance in the pulse tank and the operating parameters, [1] T. Shintake et al., C-band main linac RF system for e#elinear collider of 0.5 to 1.0 TeV C.M. energy, Proceedings of the 18th International Linear Accelerator Conference, Geneva, Switzerland, August 26}30, 1996. [2] J.S. Oh et al., The "rst 111-MW C-band Klystron-modulator for linear collider, Proceedings of the Third Modulator-Klystron Workshop, SLAC, California, USA, June 29}July 2, 1998. [3] Z.D. Farkas et al., SLED: a method of doubling SLAC's energy, Proceedings of the Ninth International Conference on High Energy Accelerator, SLAC, 1974, p576. [4] P.B. Wilson, SLED: a method for doubling SLAC's energy, SLAC-TN-73-15, 1973. [5] T. Shintake, N. Akasaka, A new RF pulse-compressor using multi-cell coupled-cavity system, Proceedings of the "fth European Particle Accelerator Conference, Sitges, Spain, June 10}14, 1996. [6] G.N. Glasoe, J.V. Lebacqz, Pulse Generator, McGrawHill, New York, 1948. [7] E.C. Snelling, Soft Ferrites, Properties and Applications, Butterworths, London, 1988. [8] N.R. Grossner, Tranformers for Electronic Circuits, McGraw-Hill, New York, 1967. [9] T.F. Turner, An Improved Pulse Transformer for HighVoltage Applications, Rept. No. MIL609, Microwave Laboratory, Stanford University, Stanford, California, May, 1966. [10] H.W. Load, IEEE Trans. Magn. MAG-7 (1) (1971) 17. [11] T. Shintake, FCI "eld charge interaction program for high power Klystron simulator, Proceedings of the 1989 Particle Accelerator Science and Technology, Chicago, USA, March 20}23, 1989. [12] J.S. Oh et al., E$ciency issue in C-band Klystron-modulator system for linear collider, Proceedings of the 1997 Particle Accelerator Conference Vancouver, Canada, May 12}16, 1997.