GoKawiilTCXO failure analysis
2026-03-06 14:00
Backstory
Back in January, the ThunderScope team sent me a PCIe card version of their prototype open hardware oscilloscope.
I was very excited about this, because I was already testing ngscopeclient with a Thunderbolt version of the ThunderScope but the only machines I had with Thunderbolt were laptops. The PCIe card version would be usable with my (much more powerful) desktops, allowing me to really push ngscopeclient and the ThunderScope to its limit. It took a bit of experimenting with external GPU enclosures and such to figure out how to get the ThunderScope to sit out on my lab bench so I wouldn’t need to crawl under the bench and run cables up from the workstation on the floor up to probes on my DUT, but I figured that out easily enough.
After getting the drivers installed I fired up ngscopeclient, cabled channel 1 of the ThunderScope to my Siglent SSG5060X-V vector signal generator, and gave it a 100 MHz unmodulated sinewave to check everything out. Everything looked good until I fired up the FFT in ngscopeclient and saw a peak… at around 106 MHz, and unstable - it was moving slightly up and down.
Just to rule out any issues with the Siglent, even though it had been recently checked against cal standards and was locked to my lab 10 MHz distribution system, I directly measured the 10 MHz outputs from my SRS FS752 GPSDO and my Symmetricom rubidium standard with the ThunderScope. Both showed 10.665 MHz, while in reality they were both 10.0000000000 MHz give or take about 2e-11. So the Siglent was fine, and the ThunderScope’s timebase was 6.6% slower than it should have been.
After a bit of back and forth, I was able to read the PLL lock bit for the ADC clock generator via a debug interface - and it wasn’t locked.
A bit of hardware debug later, I confirmed that the output of the 10 MHz TCXO (ECS-TXO-3225MV-100), which provides the primary timebase for the oscilloscope, was flatlined. The PLL VCO was running wild with no edges to lock to, with the nominally 1 GHz ADC clock hovering around 938 MHz but unstable.
The lack of a PLL reference clock certainly explained why the ThunderScope wasn’t behaving as it should. I reworked the solder joints in case of an assembly defect, but that didn’t solve the problem. So I ordered a new TCXO, swapped the bad one out, and the scope was happy.
... continue reading