VEGAS Pulsar Project Meeting - Impromptu Meeting: 2017 January 27th 2:00 - 3:00pm ET
Ryan, Joe, Steve, Jason, Randy, Natalia
During acceptance testing of the coherent modes Ryan recognized artifacts in scans of the noise diode at L-Band in all coherent modes except c0800x0064. The artifacts manifest as noise in certain frequency channels. Initial inspection shows a repetition in frequency space of about 12 or 13 channels in the c0800x0512 mode. The artifacts are not visible in scans of B1937+21, but this could be because of dispersion and faster sampling. For example see the report on acceptance testing from the previous sprint review
Randy had previously noticed that the PFB filter bin width used in the coherent modes was too large (1.0 vs 0.95 compared to GUPPI), which could lead to a few dB of extra spectral leakage. A new BOF file was built with a PFB core similar to that used in GUPPI and tested by Ryan, but there was no impact on the artifacts (see log
Ryan had previously noticed that, in incoherent modes, the 950 MHz baseband IF module filter seemed to leading to large increases in noise and spurious bandpass features. Switching to the 1400 MHz IF module filter seemed to produce better results in incoherent modes. The 1400 MHz IF module filter should now be used in coherent modes, though this has not been confirmed via inspection of ROACH register values.
The first instance of the artifacts in historical data are from early July 2016, but they do not always appear in test data after that. There is no obvious change to hardware or software around this time that would explain their emergence.
Ideas for other things to check:
- Does this show up in all banks? This should be fairly easy to test by changing the data source in the vegas.conf file.
- Could this be caused by a problem with LO2/LO3? Steve thinks this is unlikely but we can check it "for free" by routing to different banks.
- Check that IF modules are actually be set correctly via inspection of ROACH registers.
- Double check that valon is locked and using external reference.
- Test for ADC spurs using a test tone and/or the wideband artificial pulsar. The WBAP is needed because the spurs do not manifest if there is no signal above 200 MHz baseband.
- Check per-bank data to make sure that this is not a product of data combination software.
- Try sending GUPPI packets to VEGAS HPCs and vice versa, to isolate to FPGA or HPC hardware. This might be difficult and will be left to a last resort.
- Ryan will do more detailed characterization of artifacts in all modes, look more closely at historical trends.
- Acquire data with different ROACHs and banks
- Try to replicate artifacts using test-tone injection
- Test with WBAP in simulator