next up previous
Next: Feedback Characterization Up: FOFB IMPLEMENTATION Previous: General Layout

Integration Issues

Basis for the FOFB is the real time operation mode of the digital BPM (DBPM) system where each BPM electronics continuously delivers data at a rate of 4 kHz. This rate is the result of down conversion and decimation in the digital receivers with their 31.2 MHz ADC clocks locked to the 500 MHz ring RF frequency. Presently, the 4 kHz data streams of the 72 digital BPMs are not synchronous to each other. Thus, the FOFB has to wait for the latest data acquisition before a new orbit correction can be calculated although the transfer time for BPM data between the different sectors takes only 8 $\mu $s. The asynchronous data rate introduces an additional delay of maximal one feedback cycle which amounts to 250 $\mu $s. This delay will be eliminated by a DBPM firmware upgrade. The passband width of each BPM is set to 2 kHz by means of programmable filters on the digital receiver. It results in a resolution of 1.2 $\mu $m and typical group delays of about 300-600 $\mu $s. The numerical controlled oscillator (NCO) frequencies on the DDCs are adjusted to the main RF frequency. An automatic loop on the BPM low level control system tracks the ring RF frequency and reprograms the NCO frequencies to keep the BPM signal in the passband width of the DDCs. This becomes important when even smaller passband widths of the BPMs are envisaged in order to reduce noise and group delays in the DDCs and hence to increase the bandwidth of the feedback loop. Frequency changes are necessary since horizontal path length effects are taken into account by the FOFB as well. Off-energy orbits are not corrected by the steerer magnets but by adjustments of the RF frequency. It is therefore necessary to subtract the dispersion orbit from the measured orbit before a correction is applied. The dispersion orbit is extracted on the DSP level by a one dimensional SVD fit on the position data from three sectors, which corresponds to 18 BPM readings. Whenever the fitted RF frequency error exceeds 5 Hz (equivalent to dP/P $\sim 2\cdot10^{-5}$) a high level application on the beam dynamics model server [4] corrects the RF frequency.

The start of the fast orbit feedback is performed in a sequence where the central high level orbit correction application (former 'slow orbit feedback, SOFB') corrects the electron beam to the required reference orbit within 5 $\mu $m and adjusts the RF frequency. Subsequently, all necessary feedback parameters including the inverted response sub-matrices and PID control parameters are downloaded to the DSPs at the twelve BPM stations. A global trigger from the timing system starts the FOFB on all BPM stations synchronously. Since the same number of correctors and BPMs are used to constrain the orbit to the ``Golden Orbit'' at each of the 72 BPM locations, it is indispensable to rely on each single position reading. BPM pickup cross checks of the four RF buttons have therefore been implemented on the DSP in order to detect BPMs with spurious bad readings. If the sums of the two diagonal BPM button raw values do not agree within a predefined level (default 20%) the reading is considered to be faulty. In such a case, the DSP disables the BPM and stops the feedback in this particular sector because the structure of the ``inverted'' response sub-matrices are not appropriate anymore. The halt of the feedback loop on one BPM station also directly affects the two adjacent sectors which do not get position readings over the fiber links anymore and consequently skip the correction cycles with 'data timeouts'. The localized structure of the fast orbit feedback allows this type of asymmetric operation mode. Nevertheless the feedback is then automatically stopped by the EPICS control system which permanently monitors the status of all sectors. The high level orbit control application reloads a new set of sub-matrices where the particular faulty BPM is disabled and finally restarts the feedback. This scenario has already been successfully tested during machine development shifts.


next up previous
Next: Feedback Characterization Up: FOFB IMPLEMENTATION Previous: General Layout
Michael Boege
2003-11-23