2. BE-RF-FB THE LINAC4 LOW LEVEL RF 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF2 P. Baudrenghien, J. Galindo, G. Hagmann, J. Noirjean, D. Stellfeld, D.Valuch

3. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF3 LINAC4 @ CERN Linac4 is a new 86-m long normal-conducting linear accelerator that will provide 160 MeV H- to the CERN PS Booster

4. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF4 RF STRUCTURES 1 LLRF Crate 3 LLRF Crate 7 LLRF Crate 6 LLRF Crate 3 LLRF Crate + 1 Debuncher in the transfer line to PSB Main machine and beam parameters Freq RF: 352.2Mhz Type of particle accelerated: H- ions Output energy: 160 MeV Maximum repetition rate: 2Hz Beam pulse length: 400 µs Mean pulse current: 40 mA

5. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF5 LINAC 4 LLRF

6. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF6 CRATE ARCHITECTURE VME STANDARD CPU VME P1; Standard VME VME P2; CERN L4 LLRF VME Clock Distribution, Power distribution, Timing Distribution, Ref Phase Distribution, JTAG Chain, Interlocks, Alarms… RF SIGNAL DISTRIBUTION Forward and reflected power between klystron and circulator Forward and reflected power upstream of main coupler Up to four antennas per cavity One reference line signal

7. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF7 LLRF L4 HARDWARE SWITCH AND LIMIT prevents klystron saturation during loop transients switches off RF drive in few tens of ns when interlock occurs (vacuum, HV, RF high power...) CLOCK DISTRIBUTOR using the reference line signal, it generates harmonically related clocks for the I/Q demodulation-modulation acquisition (ADC 88.05Mhz) demodulation/modulation of RF signals (LO 330.1Mhz)

8. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF8 LLRF L4 HARDWARE CAVITY LOOPS Regulates the klystron drive. It includes three systems: Feedback loops for regulation of the cavity field. It compensates for the various perturbations (klystron noise, vibrations, beam loading) Adaptive Feedforward for the compensation of the transient at the head of the beam batch Polar loop to compensate for the klystron gain and phase shift variations due to the High Voltage (HV) supply fluctuations (ripples up to 10 kHz), the HV droop during the 400 microsec long pulse and the long-term drifts It includes functionalities for conditioning the cavity o Xilinx Virtex5, XC5VSX95T o 4 x RF channels inputs o 4 x Single ADI ADC, 125 MSPS, 14 bit o 2 x fast link optical tranceivers (2 GBps) o 2 x 72 Mbit SRAM o 1 x RF output(Quadrature modulator/BW 300-1000 Mhz)

9. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF9 LLRF L4 HARDWARE TUNER LOOP keeps the structure on resonance Computes the phase difference between the cavity antenna and the signal from the forward coupler Modifies the tuning via PLC/step motor acting on plungers (or temperature of the cooling water - RFQ). o Xilinx Virtex-5, XC5VLX110 o ADI SHARC DSP, ADSP-21369, 400 MHz core clock o 4 x Dual ADI ADC, 125 MSPS, 14 bit o 8 x RF Front-end channels, ENOB 11.04 bits o 2 x 72 Mbit SRAM o 2 x SerDes transceivers, 1.5 GBps

10. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF10 CAVITY LOOPS FEEDBACK PI CONTROLLER REDUCTION OF KLYSTRON NOISE Loop Delay 1.6 us Cavity Filling Ramp 50 us No beam No Polar Loop 10 kHz ripple  switching frequency of the HV power converters Feed-Forward will reduce this further as it is reproducible DTL1 @ 10MV Open Loop; 60 kV amplitude 0.2 deg phase drifts Closed Loop; 6 kV pk-pk Ripple (0.06%) 0.05 deg pk-pk Ripple

11. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF11 PI, TRACKING PI CONTROLLER REFERENCE VOLTAGE TRACKING Voltage modulation of the last two PIMS will provide Longitudinal Painting for optimum filling of the PSB bucket With the PI controller, the Cavity field Step Response show a rise time ~5 us that is consistent with the 1.6us loop delay But this is too slow for a good tracking of the voltage in the painting cavities: we will apply a triangular amplitude modulation with 40 us period Upgrade underway: Kalman predictor + LQR regulator, expected to increase the regulation bandwidth

12. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF12 PI, BEAM LODADING PI CONTROLLER BEAM LOADING RESPONSE Transient at the sharp end of beam batch, sudden drop of beam loading. Open Loop; induced voltage around 300 KV Close Loop; induced voltage around 70Kv compensated in about 10 us Planned upgrades (AFF + Kalman estimator) will further reduce this transient

13. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF13 SPACE STATE ESTIMATOR LOOP DELAY Negative implication for feedback Discrete model for delay SPACE STATE CONTROL Discrete model of the cavity / plant Precise estimation of loop delay Apply same drive to the plant and a Kalman Estimator in the LLRF controller; Virtual access to the cavity voltage (states) at times {n, n-1,.., n-N} State Feedback using these “measurements”

14. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF14 LQG REGULATOR LINEAR QUADRATIC GAUSSIAN REGULATOR includes; NON-DELAYED CAVITY STATES ESTIMATION for FEEDBACK Kalman estimator; estimate the cavity voltage Linear Quadratic Regulator; using the Kalman estimator states in feedback Integrator to track the voltage set-point The (delayed) cavity voltage measurements are used to update the predictor’s estimates, and correct for the inexact model and the unknown noise sources.

15. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF15 LQG REGULATOR DTL1 SIMULATIONS PI controller vs LQG regulator (amplitude only). Top: response to a step in reference voltage, risetime improved 5us  1us Bottom: response to the step beam loading, risetime improved 10us  3us Structure; DTL1 On tune Proc. Noise 10db SNR Sens. Noise 50db SNR Loop delay 1,6us

16. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF16 LQG IMPLEMENTATION Simulations with Simulink/Matlab; Full State Feedback FPGA Implementation with System Generator; 1.6us @ 88Mhz > 100 taps (x2 I,Q) Huge combinatorial path for LQR adder Full State Feedback not feasible CURRENTLY ONGOING WORK Multi-rate system; Kalman at slower sampling rate (decimation) LQR feedback of cavity states only

17. 02/11/2015 LLRF15, THE LINAC4 LOW LEVEL RF17 Thankyou!