search

realtimeradio / vla-dev

Ethernetization/Channelization Firmware for the VLA-COSMIC FPGA system
GNU General Public License v3.0
1 stars 3 forks source link

`set_lo_frequency_shift()` no good #27

Closed radonnachie closed 1 year ago

radonnachie commented 2 years ago

This is a core difference in the manual configure_remotefpgas.py and the automated observe.py processes: the former applies an LO F-Shift as part of configuration while the later applies zero F-Shift at configuration, awaits certain criteria and then directly sets the LO F-Shift.

Steps to reproduce

  1. Manually configure with configure_remotefpgas.py
  2. For each antenna configured use set_lo_frequency_shift to apply an LO F-Shift

Observation: After step 1. the ingest GB/s is as expected. After step 2. the ingest GB/s falls (arbitrarily between the recipients) and hashpipe instance 0 crashes with 0x1 (local length error) (see).

radonnachie commented 2 years ago

At the very least we should disable_tx for the duration of the function.

radonnachie commented 2 years ago

This hasn't been an obvious issue in the case for <=2 antenna. It's only really creeping up with more (currently 13 antenna).

jack-h commented 2 years ago

Does disabling TX successfully work around this?

jack-h commented 2 years ago

rev 13.6 of the firmware (inbound) will remove the need to sync the boards to load the LO phase accumulators. Instead, they can be commanded to load at a specific time (like the coarse delay block), and this has no effect on the higher level system.

radonnachie commented 2 years ago

disabling tx seems only to solve the instance 0 crash.

radonnachie commented 2 years ago

I'll test this firmware now

radonnachie commented 2 years ago

https://github.com/realtimeradio/vla-dev/commit/5d9f0141aa1df6bb4cbc46aad4828bb542513b26

talonmyburgh commented 1 year ago

Issue is resolved with firmware version "adm_pcie_9h7_dts_dual_4x100g_dsp_8b_2022-10-09_2043.fpg" and the applying of lo_fshifts with force_load() in the cosmic_f.cold_start(). @jack-h is requested to review and close issue.