Hi all, I seek your insight into an issue which my collaborators and I have been troubleshooting for some time. It has to do with a 1024 channel Roach2 spectrometer which uses the MUSIC DAC/ADC board. The Roach is clocked at 256 MHz, which it derives from a 512 MHz input to the MUSIC board (taken from one of the ADC channels). After making several duplicates of the system (hardware, firmware and software), we noticed that a few of them (something like one out of six) exhibit spurious voltage spikes in the ADC time streams, which are visible via snap block readout. The effect of the glitches is to increase the overall amplitude noise level at the channel outputs, so we'd like to minimize or eliminate them if at all possible.
Here is a rundown of observations and some attempts at troubleshooting so far: Signs and symptoms: - Voltage spikes in the digitized time streams of multitone frequency combs. The glitches appear to be non-periodic, and vary in amplitude over the full scale of the ADCs. They are present in both the I and Q channels, but can be worse in a single channel. They are also observed, to a lesser extent, when the I/Q inputs are terminated in 50 ohms. No glitches are observed in the DAC output, which is looped directly back to the ADCs during these measurements. The glitches only present in waveforms containing more than a single frequency. - On an afflicted system, the glitches disappear when the clock is either slowed to < ~430 MHz, or increased above ~530 MHz. On a system which is not afflicted, the glitches can be induced by slowing the clock to below ~500 MHz. - They have not been remedied by swapping MUSIC boards, although their characteristics differ somewhat between different board combinations. When a MUSIC board is taken from an afflicted system and used with a reliable Roach, no glitches are observed. - Glitches are not observed in five systems which have had their FPGA fans removed and replaced with heat pipe assemblies. However, one of these systems exhibited glitches when the mounting bolts on the FPGA heat sink were overtightened by a small amount. When they were loosened, the glitches subsided immediately. Attempts at troubleshooting: - Firmware: Identical firmware used on each of the systems. Given the correlation with clock rate, timing is a likely culprit. Using MSSGE, the firmware compiles with a timing score of zero. However, the best case positive slack is quite small (a few picoseconds). I've attempted to increase this by using pblocks in PlanAhead, as well as the get lucky approach with SmartXplorer, without much success. The ADC MMCM is configured for high bandwidth, in accordance with a previous post on MUSIC board glitches. - Hardware: As mentioned above, the glitches haven't been observed when running with heat pipes instead of fans. While watching a live stream of the ADC snap readout on one of the afflicted systems, I unplugged the power leads to the FPGA fan. The spikes gradually subsided over a few seconds. When the fan was powered on again, they returned on the same time scale. This effect was not reproducible on an identical system. - We've had no luck using different synthesizers as clock sources, filtering, attenuating or otherwise isolating the clock line to the MUSIC boards. Much appreciated, Sam Gordon Astrophysics PhD candidate Arizona State University -- You received this message because you are subscribed to the Google Groups "casper@lists.berkeley.edu" group. To unsubscribe from this group and stop receiving emails from it, send an email to casper+unsubscr...@lists.berkeley.edu. To post to this group, send email to casper@lists.berkeley.edu.
