Error found in FFT black box, trying to correct * Ryan has updated the BOF file status page with the results of many new tests
There are errors with the bandpass shape in the i0800x0064, i0800x0128, i1500x0128, i1500x0256, and i1500x8192 modes * Many of these seem related and may have a root cause
All 1500 MHz incoherent modes have phase drift that is not related to the AP signal itself
i0800x8192 has a repeated AP signal
Randy thinks this may be due to the way that the packets are being broken up and recombined, since only 2048 channels fit into one packet.
Software
TBIN parameter is being rounded with inadequate precision because 1500 MHz modes don't have a power-of-two bandwidth that divides nicely by number of channels.
Could simply increase precision of this parameter value
Could also rebuild BOFs to be "1600 MHz" modes (i.e., clock the FPGA at a faster rate) even though we can't get that much down the IF path
GBT Tests
Manager performed very well with VEGAS hardware
We saw significant packet loss in coherent modes. This manifested as incorrect values of TFOLD in the actual psrfits files because they were calculated based on missing integrations. Packet loss of 30% is common. We believe this is due to suboptimal process affinity. We tried changing the way processes were pinned to various CPU cores but still saw many dropped packets in certain modes. * Joe has a new version of guppi_daq_server with an optimized transpose that may help with packet loss in some modes. We will try this in the simulator and later with the real VEGAS hardware.