1. Klystron Cluster System (KCS)
Christopher Nantista
ILC 1st Baseline Assessment Workshop
September 7, 2010

2. Outline Concept and Layout Experimental Program Experimental Results
Future Experiments
ILC Operation With KCS (Power Requirements)
ILC Operation With KCS (Power Flow)

3. Concept and Layout

4. Klystron Cluster System Layout
surface rf power cluster building
surface
service tunnel eliminated
underground heat load greatly reduced
shaft
~1.06 km
upstream
downstream
~1.06 km
CTO
accelerator tunnel
TE01 waveguide

5. Shafts Each main linac is composed of 280 three-cryomodule rf units*
KCS would best be accommodated by moving the main shaft locations from the RDR upstream as indicated and adding two additional shafts, for a total of 5, each serving two klystron clusters, one feeding upstream and one downstream.
-- main shaft
-- additional KCS shaft
~1.064km
~684m
~380m
Each KCS:
28 rf units,
30 klystrons worth of power,
34? klystrons 28 28 28 28 28 28 28 28 28 28
beam
*additional CM’s in e- linac to compensate undulator loss will require extension of one system.
The RTML will use underground klystrons, not KCS.

6. Nominal Parameters # of shafts per main linac 5
# of KCS systems per main linac 10
# of rf units (tap-offs) per system
28 (1.064 km)
# of cryomodules per system 84
# of cavities per system
728
# of klystrons/modulators per system
~32-34*
peak rf power per system (MW)
× 2 systems/shaft
× 3 cryomodules/rf unit
× 8 2/3 cav.’s/cryomod.
½ beam power
18*
- 2
× 10 MW/klystron
* Two extra are included to allow for one failure. One or two more may be needed as overhead for LLRF control.

7. Combining and Distributing Power
Couplings ranging from ~1 to 1/28 to the TE01 (low loss, no surface E-field)mode are required.
CTO (Coaxial Tap-Off)
“3-port” coupler 1 3 2
determines coupling
For combining, the tap-offs are installed backwards.
Proper phase and relative amplitude needed for match (mismatched power goes to circulators).
First and last CTO’s are 3 dB units reversed relative to the others with port 1 shorted at the right phase.
Power Combining: 2 2 2 2 2 … 1 1 1 1 1 3 3 3 3 3
-3 dB
-3 dB
-4.8 dB
-6 dB
-7 dB l
Power Dividing:
-4.8 dB
-3 dB
-3 dB 1 1 3 1 3 3 2 2 2 l

8. Tunnel Cross-Section Layout (Preliminary )
Waveguide footprint will be reduced!
KCS main waveguide
and CTO
KCS Option American Version
(smaller along most of tunnel, as suggested by red circle)
Can be made more compact by:
not combining outputs
removing pump ports from between CTO and windows
redesigning CTO w/ one output (10 MW window?)
RDR Layout
courtesy of Tom Lackowski

9. Local RF Power Distribution Scheme
Power from each CTO is locally distributed along 3 cryomodules containing 26 cavities.
The cavities are mostly fed in pairs to eliminate the need for circulators.
Distribution is tailored to accommodate measured gradient limits of cavities.
CTO
customizable coupler
feed 1
feed 2
hybrid
load
38 m

10. Prototype Local Distribution Systems
(for Fermilab NML Facility)
For power division control, split and recombine through different phase lengths.
Version 1: VTO’s
For ILC, eliminate circulators.
Version 2: motorized U-bend phase shifters and folded magic-T’s
For ILC, eliminate circulators and replace T w/ magic T or hybrid.

11. New Local Distribution Idea
Version 2 (motorized U-bend phase shifters) is suitable where remote control is needed.
Version 1 (VTO’s) is suitable where occasional manual adjustment is foreseen.
For one time customization, the following version seems most economical and compact.
Changing widths of upper and lower spacers oppositely changes coupling while maintaining phases.
One phase shifter is eliminated, locking cavity phases for reflection cancellation.
PARTS LIST (2 cavities):
3 magic-T’s
2 H-plane U-bends
2 H-plane bends
1 E-plane bend
1 phase shifter
2 loads
2 bi-directional couplers
2 semi-flex guides
2 WR650 stubs (1/4 wavelength different)
4 customized spacers
1 waveguide run to next unit
25 gaskets + bolts

12. Experimental Program

13. Goals
Clearly, a full KCS demonstration is not feasible within the scope of pre-project commitments and funding. The KCS R&D program at SLAC is aimed at establishing confidence in the scheme as a viable single-tunnel option for powering an ILC. The current (hardware) program is focused on:
designing required components (rf & mechanical design)
prototyping components
testing functionality of components and assemblies
high power performance testing (evacuated and pressurized)

14. Goals
More specifically, our immediate goals are to efficiently transmit high-power rf through a ~10m run of the circular TE01 mode main waveguide using CTO’s and also to resonate the line to demonstrate robustness under high field levels such as it would see with full KCS power.
Transmission Line
4.5 MW
10 m of WC1890
±14.9 psig
0 dB
~4.5 MW
Resonant Line
<1 MW
back-shorted tap-in
10 m of WC1890
±14.9 psig
~-25 dB
~75 MW (300 MV equiv. peak fields)

15. Progress To date, we’ve made: Four 2.44m sections of WC1890*
Vacuum pumpout spool for WC1890
A ¼ wave WC1890 spacer
Two 3-dB CTO’s
Two tapers between CTO and WC1890
CTO shorting caps for transmission and resonant coupling
Four WR650 pumpout spool/ flange adaptors for connecting CTO ports to vacuum windows
* circular waveguide w/ 18.90” (0.48 m) diameter

16. Main Waveguide
2.438 m (8’)
0.480 m (18.9”)
Four 8’ sections (9.75 m total) of 0.48 m-diameter waveguide (WC1890).
Fabricated from formed aluminum sheets, welded and machined.
(1mm radius tolerance)
one-side double grooved flanges:
vacuum/pressure seal – Viton® O-rings
rf seal – Bal Seal® canted coil contact spring

17. Coaxial Tap-Offs (CTO’s)
The CTO (coaxial tap-off) is essentially a 3-port device which couples power from its central circular TE01 waveguide through a gap into a coaxial region. From here, power is coupled out into a wrap-around waveguide and split between two radial WR650 ports.
Variation of the gap allows different coupling values. In reverse, the CTO can be used for combining, given the correct power ratios and phases. By shorting one circular port, it can be made into a launcher.
We have two welded aluminum 3-dB CTO’s .

18. launching and resonant coupling.
CTO Auxiliary Parts
NOTE: not final depths
13.75” → 18.90”
Two circular step tapers to connect to WC1890 main waveguide.
Shorting end caps for
launching and resonant coupling.
(not final depths)
launcher
coupler

19. Vacuum Pumping SLAC L-band vacuum window WR650
pressure
WR650
crushed indium wire
vacuum
20 l/s ion pumps will be used on vacuum ports built into window adaptors on either side of each CTO.
Two 100 l/s ion pumps will pump through a special WC1890 pump-out insert in the middle of the large circular waveguide run.

20. Experimental Assemblies
CTO cold tests
input assembly
transmission tests m
9.990 m
resonant line tests
Location:
Roof of NLCTA bunker
Power source:
SNS modulator and
Thales “5 MW” klystron

21. Experimental Results

22. Program Installation RF from klystron CTO input ports WC1890
8-ft sections
Pump-out
CTO output ports with loads

23. Indium Seal Vacuum Test
In preparation for system assembly, we high-power tested, under vacuum, two WR650 vacuum window adaptor/pumpout spools s back-to-back between two windows.
The adaptors were connected with a grooved flange insert with crushed indium wire sealing. The dual purposes of the run were to test the indium seal and to test running aluminum WR650 under vacuum.
We ran into a problem running above kW which appears to be aluminum multipacting, unrelated to the indium. We will try repeating the experiment after cleaning and etching the parts, which may not have been properly prepared.
Power to Waveguide (kW)

24. Initial CTO Cold Test
This plot made by combining 4-port cold test measurements for a 2-port response of two CTO’s back-to-back, shorted with transmission “caps”. The latter were made deliberately too shallow with the intention, not yet realized, of tuning them by shims and remachining.
red – reflection
blue – transmission
yellow – loss
S Parameters (dB)
@1.300 GHz:
R = dB (0.906%)
T = dB (98.08%)
Missing: 1.02%
@ GHz:
R = dB (0.164%)
T = dB (99.04%)
Missing: %
These reasonable preliminary results demonstrate that the CTO basically works as expected.
Since these measurements represent two at a given spacing, the match of a single CTO may be better or worse.
Cap tuning and further characterization of the CTO’s was deferred due to (BAW driven) schedule reprioritization.

25. Resonant Test Setup
Due to the approaching workshop, we decided to proceed as expeditiously as possible directly to the pressurized resonant test.
We hoped to put power in and generate high fields in the 15 psig pressurized assembly before BAW. If KCS can work pressurized, there is no need to pursue the vacuum tests, and we can eliminate vacuum pumps from the envisioned ILC KCS system.
ceramic block pressure windows substituted for pillbox vacuum windows
end plate replacing taper and machinable tuning “cap”

26. Establishing a TE01n Resonance Near 1.3 GHz
spacer added here
Big pipe shorted w/ an end plate (no end taper or cap).
Analytic TE01n resonances for 11.73m of WC1890 (effective length for pipe, taper, CTO and shorting cap chosen to fit data):
1.2749
1.2851
1.2954
1.3058
1.3162
1.3266
Measured
Resonances:
1.3035
Measured
Resonances w/ lg/4 spacer:
Analytic TE01n resonances for 11.73m m (lg/4) of WC1890:
1.2699
1.2801
1.2903
1.3006
1.3109
1.3213
best
sharp
small
Outsmarting Murphy w/ a lg/4 spacer we’d made for cold testing: TE0,1,83 within a MHz of 1.3 GHz.
Murphy’s Law: TE01n resonances straddle 1.3 GHz as best they can.

27. Characterizing the Waveguide Resonator
S11 (dB)
S11 (dB)
Q circle fit
fr = GHz (→ psig)
QL = 33,787
b =
Q0 = 146,028
Imaginary
→ DfrFWHM = 38.5 kHz
tc = 2QL/w = 8.27 ms

28. Needed Input Power Calculation
(WC1890)
0.593 c (WC1375)
SW energy density: u(z) = 2Pt/vg(z), where Pt is the traveling wave power in either direction
To produce field levels equivalent to 300 MW TW, one only needs ¼, or 75 MW, in Pt.
Pi = 566 kW
U = 7.19 J
Pd = 402 kW
C = dB

29. Interlock Field Probe
As a safety interlock, a (non-directional, uncalibrated) probe of the WC1890 rf field was included, so that power could be shut off in the case of field collapse due to rf breakdown.
A loop antenna inserted about ½ way through a perforation hole in the WC1890 pumpout spool.
NWA showed ~-64 dB coupling from the input port. For 566 kW input, it should see 225 mW.

31. Future Experiments

32. Further Work ?
Further work can be done toward establishing the feasibility of the Klystron Cluster Scheme, including:
demonstrating the matched tap-off function with a third CTO
demonstrating power combining
designing a very high-power TE01 mode bend
demonstrating transmission efficiency of such a bend
demonstrating power (field) handling of such a bend
developing a TE01 mode diagnostic directional coupler
studying the combining scheme/efficiency
studying coupling errors, mode conversion, line resonances…

33. Overmoded Bends
To get from the surface cluster building to the linac, we must transport ~310 MW L-band rf pulses around 3(?) bends receiving and restoring the circularTE01 waveguide mode.
Designing, prototyping and testing an appropriate bend is a primary focus of our next phase of KCS R&D. Below are examples of approaches taken at X-band.
TE20
TE01
TE20
General Atomics profiled curvature bend in corrugated waveguide
SLAC compact bend in 40.6 mm circular waveguide tapering to and from rectangular TE20 mode
Finned waveguide, breaking TM11 degeneracy
(Nantista et al. PAC ‘93)

34. Resonant Ring (2012)
To test waveguide, bends, and CTO’s at full traveling wave power, a resonant ring could be built, incorporating a tap-off/tap-in section. This would require development of a CTO based directional coupler not required in ILC. It would also need a diagnostic coupler that could monitor power flow in either direction.
Such an undertaking would presumably be beyond the scope of the current Technical Design Phase.
phase shifter
directional coupler
tap-off
tap-in
160 m of WC1890
310 MW
diagnostic
bi-directional coupler

35. ILC Operation With KCS (Power Requirements)

36. Main Waveguide Power Transmission Loss
a(a=0.24m) = 5.67110-5 nepers/m, a(a=0.1746m) = 2.01410-4 nepers/m
WC1890
main waveguide
WC1375
CTO waveguide
For a 38 m rf unit that’s roughly 1.5 m WC1375 and 36.5 m WC1890, we have:
[1.5m*2.01410-4 nepers/m m*5.67110-5 nepers/m] = nepers
→ power transmission = (0.47% loss/rf unit in main waveguide)
For unit power at the 28th rf unit, we need units at the first and at the shaft. That’s 6.95% extra that gets attenuated in the main waveguide.
Peak Power Flow in Main Waveguide / 10 MW

37. Yamamoto/Ginsburg LCWS2010
Cavity Gradient Limit Distribution
While the design gradient for the ILC remains 31.5 MV/m, in order to increase the yield of usable cavities, the acceptable range of limiting gradient has been increased to within ±22.2% of this value.
From the following data plots, the drop-off of cavity yield with gradient would seem to be roughly linear over the range of interest, indicating a flat distribution of limits.
2007
2010
1st pass
2nd pass
Cavity Yield
Gradient (MV/m)
Yamamoto/Ginsburg LCWS2010

38. Cavity Pairing Sorting Recipe:
To eliminate the need for circulators at most of the cavities, they are powered in pairs through a hybrid or magic-T, with the reflected and discharged power combining into a load on the fourth port.
This requires running the cavities in each pair at the same gradient, the lower of the two limits, resulting in a loss in average gradient. To minimize this effect, some degree of sorting of cavities to closely match gradient limits is called for.
In our calculations, we assume an rf unit’s worth of cavities (26) is received and sorted between the 3 cryomodules as follows.
Sorting Recipe:
Order 26 random cavities by gradient limit.
Pair the first, last, or middle 24 in order, whichever minimizes gradient loss.
Assign pairs alternately to each 13-cavity feed, alternating between beginning and end of sort (i.e. of 12 ordered pairs, assign [ ] and [ ].)
Assign last 2 so as to minimize power requirement difference between feeds.

39. Effect on Average Gradient
Cavity Gradient Limit Distribution: flat from 24.5 MV/m to 38.5 MV/m
(31.5 MV/m ± 22.2%)
 raise cavity distribution to compensate.
Cavity Gradient Limit Distribution:
flat from MV/m to MV/m
( MV/m ± 22.2%)
Distribution
Distribution
0.62% gradient degradation due to pairing
Mean: MV/m
Std: kV/m
Mean: MV/m
Std: kV/m

40. Power Needed for Flat Acceleration
Achieving Flat Gradient
Flat acceleration across the beam pulse is important for maximizing gradient.
With control of power, Qe, and fill time, a steady state match can be achieved at the desired gradient in a cavity, giving constant acceleration with no power is reflected during the beam pulse.
With KCS, 748 cavities are locked to a common source timing. With power and Qe adjustable, a steady state constant gradient is still achievable, but there will generally be some level of power reflected.
Blue: timing set for zero reflection at center gradient
Red: timing set for zero reflection 5% below center gradient
Normalized to
31.5MV/m × m × 9mA = 294.2kW
Mean:
Power Needed for Flat Acceleration
For a given distribution of gradients, the timing can be set so as to minimize the average reflected power, not necessarily at the point where the center gradient is matched.
Mean:
20% spread example

41. Power Needed for Flat Acceleration
Achieving Flat Gradient (2)
Below are the power needed and coaxial coupler setting for optimal common timing across the gradient range we are considering.
Gradient Distribution: flat MV/m
( MV/m ± 22.2%)
timing set for match at 29.8 MV/m
power into beam
reflected power
Qe,max/Qe,min = 3.359
Power Needed for Flat Acceleration
QL / 106
Lowering the match point also reduces the range of QL adjustment required.
Average ~6.0% above the nominal kW.
For upstream(downstream) KCS systems, the effective spread in fill time is ~7.64(0.40) s. This is roughly 1.28%(.07%) of the nominal fill time and has little effect on efficiency.

42. Needed Power at Cavities
With the given distribution of cavities and sorting prescription, there will still be some statistical spread in how much power each rf unit needs. Since the CTO’s will not be adjustable, excess power will have to be provided to accommodate most (97.5%?) units.
With no adjustment to the split between halves of an rf unit, the statistics are over 13 cavity feeds, or 1/2 rf units. The difference is marginal.
mean + 2std (97.5%) = MW
Mean: MW
Std: kW
2(mean+2std) = MW
(0.307% more)
Mean: MW
Std: kW
Mean: 5.89% over nominal (294.3 kW)
Mean+2std (97.5%): % over (12.15% w/ feed division adjustment)
 5.9% reflected, 6.6% into end loads

43. RF Unit Layout w/o feed adjustment 38 m w/ feed adjustment
end load for excess power
w/o feed adjustment
CTO
adjustable coupler
feed 1
feed 2
hybrid
load
38 m
w/ feed adjustment
CTO
adjustable coupler
feed 1
feed 2
load
hybrid

44. Loss Tally
294.3 kW (nominal to beam per cavity = 31.5 MV/m 1.038m  9 mA)
× (for flat gradient w/ cavity gradient spread and common timing)
× (for statistical spread in feed/rf unit requirements w/ fixed couplings)
× 26 (cavities/rf unit)
÷ (~5% local distribution losses) = 9.06 MW/rf CTO
× 28 (rf units)
÷ (6.5% main waveguide losses) = beginning of linac run
÷ (shaft and bends)
÷ (combining CTO circular waveguide losses)
÷ (input circulator and WR650 losses)
÷ (CTO coupling/klystron amplitude mismatches)
MW from klystrons
~8%
klystron-to-tunnel

45. Klystrons Needed per KCS
The calculation/estimate suggests we need MW worth of klystron power.
At 10 MW each, 30 klystrons would give us 300 MW (1.7% to spare).
However, we want to be robust against a single klystron failure per system. With N sources combined in a passive network, failure of one source leaves combined the equivalent of (N-1)2/N sources.
With 32 klystrons, we have 320 MW (8% to spare).
With one failure, and 31 on of 32, we have MW available (1.8% to spare).
However, we also need 5% overhead for LLRF compensation of such effects as beam current (loading) variation and Lorentz force detuning. This overhead could be harnessed via phase control of the rf drives, oppositely dephasing the klystrons in pairs, as illustrated. Out of phase components will deflect power to loads, while the in phase components contribute to combined power reduced as P = Pmax cos2 f. f -f
V2n
V2n-1
Vtot
To avoid the asymptotic loss of control as f→0, we might allow f to vary from ˚ (93.3%) to 8.13˚ (98%).
The real power overhead needed for LLRF is then ~7% and the maximum power requirement rises to MW ÷ = MW, (of which up to 310 MW would be used)
With 33 klystrons and one failed, we have MW (1.8% low).
With 34 klystrons and one failed, we have MW (1.4% to spare).

46. Summary
There are multiple reasons for the increase in required klystrons per 28 rf units:
klystrons:
28 equivalent to the RDR requirement, 1 per rf unit.
+2 7% more for long range distribution for eliminating service tunnel.
+2 for redundancy (allowing one failure). In the RDR, such failures had to be covered by including additional rf units.
+2 to recover enough for 5(7)% LLRF overhead after a 12.5% hit due to cavity gradient spread (flat gradient w/ common timing and feed statistics). Most of this hit would exist in the RDR scheme. 34
NOTE: The preceding calculations could well be off by ~2-3 percent, depending on actual cavity distribution and error margins in loss estimates.

47. ILC Operation With KCS (Power Flow)

48. KCS Average Power Diagram*
4.395 MW
flow (MW)
heat loads
4.395
4.043
3.841
2.305
2.224
2.173
2.157
2.121
1.983
1.884
1.774
1.038
power supplies (.92)
351.7 kW
202.1 kW MW
80.9 kW
51.0 kW
15.3 kW
36.5 kW
138.0 kW
98.9 kW
109.8 kW
736.2 kW
“wall plug”
modulator (.95)
klystron collector (.60?)
klystron waveguides
combining loads
combining waveguide (CTO’s)
shaft & bends
tunnel
main KCS tunnel waveguide
local distribution waveguide
distribution end loads
cavity reflection loads
* 28 rf units, full power
beam
1.038 MW
(31.5MV/m1.038m9mA0.969ms5Hz26cav/unit28units)

49. Tunnel Heat Load Notes
Power dissipation along tunnel in the KCS main waveguide is non-uniform, declining roughly linearly from its peak at the shaft to ~1/30 this vale at the end of the KCS.
Main Waveguide Heat Load (W/m)
average: ~130 W/m
fill
Because ILC is pulsed, not CW, there is nominal power reflected from cavity into load (during fill and discharge) of ~38% of cavity input power. This is in addition to all attenuations and mismatch reflections.
discharge
Cavity Reflected Power (normalized)
matched case
Additional heat load associated with the rf comes from the electronics crates in the tunnel. Preliminary estimates set this at ~ kW per KCS.