Speaker
Description
The CERN Linac4 low-level RF relies on an adaptive feed-forward (AFF) to correct the beam-loading transient that the field feedback (PI or LQG) cannot remove within a single pulse. The AFF learns a correction from the field error of previous pulses and reduces the beam loading at the batch start to below 0.4%, a factor of two better than the specification. Over several years of operation we have identified conditions in which the AFF learns a wrong correction: intermittent beam transmission through the cavity, pulses taken with the beam stopper inserted, breakdowns or faults that interrupt the beam, and outdated corrections left after a long pause or a change of settings. In such cases the beam quality degrades or the RF trips. We present these conditions together with the measures taken to keep the AFF running reliably without operator intervention. We added a beam-present input, restricted learning to pulses with the beam stopper out and no active interlock, introduced an automatic reset after a long pause or a change of settings, refined the correction for longitudinal painting, and tuned the parameters for greater robustness.
| Abstract Classification | Control Architecture and Strategies |
|---|