The CLIC study focuses on high-gradient, high-frequency (30 GHz) acceleration for multi-TeV linear colliders. Short RF pulses of high peak power are typically required in high-frequency linear colliders. In the case of CLIC, 130 ns long pulses at about 230 MW per accelerating structure are needed, but no conventional RF source at 30 GHz can provide such pulses. This leads naturally to the exploration of the two-beam acceleration technique [3.1] , in which an electron beam (the drive beam) is accelerated using standard, low-frequency RF sources and then used to produce RF power at high frequency.

In linear collider projects based on conventional RF sources (klystrons), pulse compression or delayed distribution techniques are used in order to obtain the needed high peak power and short pulse length. Similar techniques can be used in two-beam accelerators. In the CLIC case, however, the compression and distribution are done with electron beams [3.2]. The main advantage of electron beam manipulation, with respect to manipulation of RF pulses, consists in the very low losses that can be obtained while transporting the beam pulses over long distances and compressing them to very high ratios. A further advantage is the possibility of frequency multiplication, achieved by interleaving bunched beams by means of transverse RF deflectors [3.3] . In the following we will describe the CLIC RF power source complex used to generate all the RF power needed for one of the two main linacs (electron or positron). Possibilities to combine some elements of both the e ⁺ and e ^- complexes are under study. A schematic layout of one complex is shown in Fig. 3.1.

Fig. 3.1 : Schematic layout of the CLIC RF power source. Two such complexes (one for each of the main linacs) will be needed to provide the power for
3 TeV c.m. CLIC operation. Only two of the 22 decelerator/accelerator units composing a linac are shown.

The CLIC RF power source can be thought of as a `black box' that combines and transforms several long, low-frequency RF pulses into many short, high-power pulses at high frequency. During the process, the energy is stored in a relativistic electron beam, which is manipulated in order to obtain the desired time structure and then transported to the place where the energy is needed. The energy is finally extracted from the electron beam in resonant decelerating structures, which run parallel to the main accelerator and are called Power Extraction and Transfer Structures (PETS). The key points of the system are an efficient acceleration of the drive beam in conventional structures, the introduction of transverse RF deflectors to manipulate the drive beam, and the use of several drive-beam pulses in a counter-flow distribution system, each one powering a different section of the main linac. The primary characteristic of this scheme consists of using the energy stored in different time bins of a long electron-beam pulse to create the RF necessary for different sections of a long linac. Thus, the same accelerator and beam manipulation system is used to create all the beam pulses needed for powering one of the two main linacs. The method discussed here seems relatively inexpensive, very flexible and can be applied to the beam-acceleration in various linear colliders over the entire frequency and energy range applicable.

The drive-beam generation complex is located at the centre of the linear collider complex, near the final-focus system. The energy for the RF production is initially stored in a 92 ms long electron beam pulse (corresponding to twice the length of the high-gradient, main linac) which is accelerated to about 1.2 GeV by a normal-conducting, low-frequency (937 MHz) travelling wave linac. The linac is powered by conventional long-pulse klystrons. A high-energy transfer efficiency is paramount in this stage. The drive beam is accelerated in relatively short structures (3.4 m long), such that the RF losses in the copper are minimized. Furthermore, the structures are fully beam-loaded, i.e., the accelerating gradient is zero at the downstream end of each structure and no RF power flows out to a load. In this way, about 98% of the RF energy can be transferred to the beam.

The beam pulse is composed of 32¥ 22 subpulses, each one 130 ns long. In each subpulse the electron bunches occupy alternately even and odd buckets of the drive-beam accelerator fundamental frequency (937 MHz). Such a time structure is produced after the thermionic gun in a subharmonic buncher, whose phase is rapidly switched by 180° every 130 ns. This provides us with a means to separate the subpulses after acceleration, while keeping a constant current in the accelerator and avoiding transient beam-loading.

With nominal phase-switching times, the resulting pulse of the acceleration voltage is rectangular. By delaying the phase-switching time, it is also possible to obtain subpulses of different lengths. When the different subpulses are superimposed at the end of the combination process, one can thus obtain a current ramp of about 22 ns at the leading edge of the pulse. This in turn produces a ramp in the PETS power output, which is used for beam-loading compensation in the main linac. An illustration of this technique [2.11] is given in Fig. 3.2 under the assumption that only one combiner ring is used, folding the beam by a factor two only. If the phase-switching is delayed after the first two trains, the shape of the voltage pulse is not flat anymore. The trailing bunches which are located after the nominal switching time add to the train considered and consequently append to the tail of the final pulse. In the train which follows the delayed switching, the first few bunches are missing and this generates gaps at the head of the final pulse. This results in a variation of the density of bunches and therefore in a ramp of the current. The unwanted tail is of no concern since it goes through the drive-beam decelerator after the passage of the main beam.

Fig.3.2 : Illustration of the delayed switching scheme. In the upper case, the phase is switched at nominal times, creating a rectangular pulse.
In the lower case, the phase shift is delayed in order to create a ramp of beam current.

As the long pulse leaves the drive-beam accelerator, it passes through a delay-line combiner [3.2] where `odd' and `even' subpulses are separated by a transverse RF deflector at the frequency of 468.5 MHz. Each `even' bunch train is delayed with respect to the following `odd' one by 130 ns. The subpulses are recombined two-by-two by interleaving the electron bunches in a second RF deflector at the same frequency. The net effect is to convert the long pulse to a periodic sequence of drive-beam pulses with gaps in between. After recombination, the pulse is composed of 16  ¥  22 subpulses (or trains) whose spacing is equal to the train length. The peak power and the bunch frequency are doubled.

The same principle of electron-bunch pulse combination is then used to combine the trains four-by-four in a first combiner ring, 78 m long. Two 937 MHz RF deflectors create a time-dependent local deformation of the equilibrium orbit in the ring. This bump is used for injection of a first train in the ring (all its bunches being deflected by the second RF deflector onto the equilibrium orbit). The ring length is equal to the spacing between trains plus l /4, where l is the spacing between bunches, equal to the wavelength of the RF deflectors. Thus, for each revolution period, the RF phase seen by the bunches circulating in the ring increases by 90°, and when the second train is injected, the first one does not see any deflection and its bunches are interleaved with the ones which are injected (at a l /4 distance). This is repeated twice, then the four interleaved trains are extracted from the ring by an ejection kicker half a turn later, and the same cycle starts again. After the first combiner ring the whole pulse is composed of 4¥ 22 trains.

The trains are combined again, using the same mechanism, in a second combiner ring, 312 m long, yielding another factor four in frequency multiplication, and obtaining the final 22 trains required for the main linac. At this point, each final train is 39 m long and consists of 1952 bunches with a charge of 16 nC/bunch and an energy of 1.18 GeV.

Such drive-beam pulses are distributed down the main linac via a common transport line, in a direction opposite to the direction of the main beam. The distance between trains is now 1248 m, corresponding to twice the length of the linac section which they will power, so that they will arrive at the appropriate time to accelerate a high-energy beam travelling in the opposite direction.

Pulsed magnets deflect each beam at the appropriate time into a turn-around. After the turn-around each pulse is decelerated in a 624 m long sequence of low-impedance Power Extraction and Transfer Structures (PETS) down to a minimum energy close to 0.12 GeV (Fig. 3.22), and the resulting output power is transferred to accelerate the high-energy beam in the main linac. As the main beam travels along, a new drive-beam train periodically joins it and runs in parallel but ahead of it to produce the necessary power for a 624 m long linac unit. At the end of a unit the remaining energy in the drive beam is dumped while a new one takes over the job of accelerating the main beam. The main characteristics of one drive-beam unit are given in Table 3.1.