ALMA Integrated Computing Team Coordination & Planning Meeting #3 Socorro, 17-19 June 2014

1. ALMA Integrated Computing TeamCoordination & Planning Meeting #3 Socorro, 17-19 June 2014 Observation with ACA correlator for Cycle3 Manabu Watanabe NAOJ

2. Background & purpose • ACA correlator has a bug to get bad channels. • In MBR telecon on 16 May 2014, we have made decisions for Cycle 2 as: • Full array mode will be used with ACA correlator. • FPS will be used even when bad channels will be output. • The fact is that 5 B grade (1.4%) among 353 A+B projects for Cycle 2 require ACA where bad data will be emerged at up to tens channels near the band edges. • The bug will not be likely fixed yet for Cycle 3. • How could we live with the ACA correlator for Cycle 3 given the persisting bug?

3. Problem overview • ACA correlator frequently outputs bad data at multiple channels during observations. • We believe the problem is caused by a few bugs in logic circuit of FPGAs in the ACA correlator. • The location of failure in the logic circuit of FPGA is not determined yet. • Experience has shown that the occurrence or absence of bad channels depends only on the spectral spec, but not on the input data.

4. Investigation and fix • It will take a long time to understand thoroughly & resolve the underlying cause. • Plan for 2014 fiscal year • May- and Jun-2014: a preparation phase. the posterior analysis of the problems. • Jul-, Aug- and Sep-2014:the logic design analysis, mapping between failures and circuits. small-scale simulations. • Oct-, Nov- and Dec-2014:large-scale simulations to reproduce the problem and to give a theoretical grounding to the occurrence condition. • Jan-, Feb- and Mar-2015: the proposal of the fix and/or countermeasures

5. Study of occurrence condition • No theory for the error prediction • Brute force attack • 56,344 spectral specs have been confirmed at present • Up to 4-spw for the full-array • 1-spw for 2-subarrays • Bad channels only with FPS (No errors w/o FPS, so far) • Full-array: 3% cases affected (360: errors/11,478: no errors) • 7.6kHz- or 15.3kHz-channel spacing for 1-polarization. • 15.3kHz-channel spacing for 2- or 4-polarizations. • 1GHz BW, 1ms dump duration, 1-polarization & 64ch for the last-spw. • 2-subarrays (808: errors/888: no errors) • Error for about a half of confirmed spectral specs.

6. Workaround proposal for Cycle 3 • Full-array mode with ACA correlator. • But at the same time, we will continue the brute force attack for the remaining spectral specs. • Up to 4-spw for 2 subarrays which share the same spectral specs between them by March 2015. • Add the usage information of FPS to ASDM • This information is needed intrinsically for ARCHIVE users. • Align spw in ascending order of bandwidth by ALMA-OT • This will dramatically reduce the number of spectral specs to be tested in the light of errors with the 64ch last spw. • 4 options for the FPS (next slide) • We recommend the “FPS online with CDP computers” option

7. 4 options for FPS for Cycle3 (1) • No FPS • Difference will be 2%-3% typically, 10% at most compared to the frequency profile of the BL correlator. • FPS online with ACA correlator • Live with the bad channels problem. • FPS online with CDP computers • Difference will be 1%-2% typically, 5% at most compared to that for FPS with ACA correlator. • FPS offline with CASA • Equal to FPS with CDP for the spectral averaging factor = 1 • No simulation yet for the spectral averaging factors > 2

8. 4 options for FPS for Cycle3 (2)explain with an example provided by T.Kamazaki