Problems setting-up the ALBA FOFB DEELS 14 12

Problems setting-up the ALBA FOFB DEELS 14 12 -13 May - ESRF Angel Olmos

OUTLINE ALBA FOFB Overview Latest Results Current Problems of Data Transfer on c. PCI Setting the PI loop (No) Correction of Kickers signal Summary A. Olmos DEELS’ 14

ALBA FOFB Overview DEELS’ 14 We aim to achieve orbit stability on the sub-micron level up to frequencies in the 100 Hz range This is a first stage “low-cost” system to learn about beam stabilization and better define the future system 88 BPMs out of the 120 available will be used on FOFB 88 Horizontal / 88 Vertical corrector magnets Libera Brilliance running Diamond Communication Controller protocol for data transfer “Recycled” timing boards used as position reading nodes (sniffers) Distributed correction calculation on 16 dedicated CPUs No Slow Orbit Feedback RF frequency control by standalone process A. Olmos Not only Desy recycles

DEELS’ 14 ALBA FOFB Overview Main equipments interconnection One sector out of 16 Next sector Libera c. PCI crate PMC board (sniffer) Libera Previous sector Libera CPU IP modules Libera Tx Board Corrector PS A. Olmos Tx Board Corrector … Corrector PS PS Corrector PS

ALBA FOFB Overview DEELS’ 14 e. BPM electronics Libera Brilliance - software Release 2. 09 Diamond Communication Controller to handle the position data transfer between units Optical links from 2 Liberas on each sector are laid to a central patch panel Routing of each link can be done from-to any sector A ring-type topology is used for the time being Only one optical link is used to send BPMs data to the PMC FPGA board A. Olmos

ALBA FOFB Overview DEELS’ 14 PMC FPGA boards Decision to re-use some Micro-Research EVR-230 boards that we had in-house These boards were meant for timing purposes on Beamlines but never installed Do only have one single optical link for position data transfer No redundancy and so low FOFB reliability Xilinx Virtex-II FPGA is an already obsolete device Integration of CC has been done by Diamond The boards are already known by ALBA controls staff Overall cost reduction of the FOFB becomes significant A. Olmos

ALBA FOFB Overview DEELS’ 14 Correction Calculation CPU Retrieves the position data from the PMC FPGA board and performs the calculation of the needed correction setpoints Adlink single, dual and 4 -Cores c. PCI CPUs have been analyzed Also different Kernel and Linux OS versions were tested because the handling of the interruptions forced CPU dead-times Adlink 4 -Cores c. PCI-3970 CPU running soft real time Linux 2. 6. 27 Processes distributed to different Cores (Read BPM, Calculation, CPU-c. PCI stuff) A. Olmos

DEELS’ 14 ALBA FOFB Overview Power Converters Power converter Diamond Provided by OCEM company (was in bankrupt but I heard they’re back) I/V transducer Provide ± 1 mrad of DC deflection and ± 40µrad @ 100 Hz ADC PSI Controller RS 232 service port 1 k. Hz Bandwidth / 18 bits Resolution Correctors PCs rack Controls rack Optical Correction Calculation CPU Tx Board IP modules Electrical to Optical protocol management Carrier c. PCI Bus A. Olmos

DEELS’ 14 ALBA FOFB Overview Correctors Magnets Horizontal Steering Vertical Steering ALBA sextupoles have extra wiring to provide H/V beam steering Eddy currents on the vacuum chamber reduce the effect of the magnetic field at high frequencies To have a more effective penetration field chamber thickness reduction to 2 mm in the correctors A. Olmos f. H=235 Hz f. V=1550 Hz

ALBA FOFB Overview DEELS’ 14 No Slow Orbit Feedback Integration of x. BPM Integration of the photon monitor (x. BPM) of MISTRAL beamline is already implemented in the SOFB Libera Photon + Communication Controller to be used Control of RF frequency Handling of Interruptions / ACK does not allow FOFB to readback the correctors setting. FOFB just assumes that setpoint is OK No possibility to set correctors AC and DC by FOFB External process that will monitor dispersive pattern on correctors and change RF frequency A. Olmos

First Results (before Easter shutdown) DEELS’ 14 HORIZONTAL 88 BPMs FOFB OFF (ID stopped) FOFB ON (ID closing) FOFB OFF (ID opening) 5 um 50 sec 2 um A. Olmos VERTICAL 88 BPMs

DEELS’ 14 First Results (before Easter shutdown) IDs SOURCEPOINT HORIZONTAL POSITION 10% Beamsize 1 um 100 nm 1 Hz Frequency 100 Hz 1 k. Hz IDs SOURCEPOINT VERTICAL POSITION 1 um 10% Beamsize 100 nm 1 nm A. Olmos 1 Hz Frequency 100 Hz 1 k. Hz

DEELS’ 14 First Results (before Easter shutdown) XALOC HORIZONTAL ANGLE 1 urad 10% Beam Divergence 1 nrad 1 Hz Frequency 100 Hz 1 k. Hz XALOC VERTICAL ANGLE 1 urad 10% Beam Divergence 1 nrad 1 Hz A. Olmos Frequency

Problems of Data Transfer on c. PCI DEELS’ 14 Brust Mode should ideally allow BPMs reading within 20 us c. PCI crate PMC board (sniffer) But brust is stopped after 2 cycles and restarted again Like that, reading 88 BPMs takes >100 us It can only be warratied reading at 5 k. Hz 2 c. PCI Bridges And we can also have problems due to CPU interruptions … When reading 88 BPMs 1 cycle lost every 2, 4 or 8 seconds CPU 15 h tests 0. 007% correction cycles lost If reading 104 BPMs 0, 009% If reading 120 BPMs 0, 32% A. Olmos

DEELS’ 14 Setting the PI loop Kp + E(s) U(s) Ki/s U(n) = U(n-1) + A 0 * E(n) + A 1 * E(n-1) A 0 = (Ki * Ts /2) + Kp A 1 = (Ki * Ts /2) - Kp Ts = 1/5 k. Hz No previous experience setting PI loops It has been done experimentally Trim Coil that introduces noise at predefined frequencies Best results found for Kp =0, Ki=1000 A. Olmos

DEELS’ 14 (No) Correction of Kickers signal FOFB OFF Kickers OFF A. Olmos

DEELS’ 14 (No) Correction of Kickers signal FOFB ON Kickers OFF A. Olmos

DEELS’ 14 (No) Correction of Kickers signal 3 Hz injection rate FOFB ON Kickers ON A. Olmos

DEELS’ 14 (No) Correction of Kickers signal FOFB ON Kickers Pulse A. Olmos FOFB effect

DEELS’ 14 (No) Correction of Kickers signal 10% H-Beamsize HORIZONTAL 1 um 100 nm 1 Hz A. Olmos Frequency 100 Hz 1 k. Hz

DEELS’ 14 (No) Correction of Kickers signal VERTICAL 1 um 10% V-Beamsize 100 nm 1 Hz A. Olmos Frequency 100 Hz 1 k. Hz

DEELS’ 14 Summary Limitation to 5 k. Hz BPMs data due to c. PCI architecture Spurious CPU interruptions We believe we’ve to live with that Don’t really know how to proceed setting the PI loop Integration of x. BPM in the loop (possible problems? ) No Slow Orbit Feedback. RF control by standalone process Effect of Kicker pulses 1 st try was to discard position values higher than some levels No success 50 Hz signal Did anyone try to reduce it by external means? Dedicated filtering of 50 Hz on FOFB is enough? Any comment / suggestion / critic is really welcomed A. Olmos