I keep seeing the same two oscillator selection mistakes in design reviews

Peter Hu — Thu, 08 Oct 2026 15:10:00 +0000

After a few too many design reviews, I've noticed the exact same two mistakes come up over and over again:

The "just in case" over-spec Someone will put a $150+ OCXO into a design that only needs ±2ppm stability. Then they wonder why their battery life dropped from 5 years to 6 months, and why the board needs a 5-minute warm-up before it's stable. The thing is: TCXOs are good enough for most consumer and industrial use cases. ±0.5 to ±2.5ppm over temperature covers 90% of designs. You don't need an oven unless you actually need sub-50ppb stability.
The "we'll just fix it in software" under-spec On the flip side: someone uses a basic XO or cheap TCXO as the reference for a 1GSPS ADC, then spends weeks debugging why the SNR is 10dB off. Clock phase noise at 10kHz offset doesn't care about your software filters. Or the classic: picking a TCXO for a 5G small cell that actually needs Stratum 3E stability (±50ppb), because someone looked at the ±2ppm spec and thought "that's fine, right?" My lazy decision tree after 10 years of this: Need to pull the frequency with a control voltage for a PLL? → VCXO. Need better than ±50ppb over -40 to +85°C? → OCXO. Need to hold time without GNSS for more than a few hours? → OCXO. Literally everything else? → TCXO. Stop overthinking it. The other thing nobody talks about: warm-up time. A lot of engineers forget that OCXOs can take 20 seconds to 10 minutes to reach stable operating temperature. If your device has to boot and transmit in <1 second, an OCXO is going to give you problems. What's the most common oscillator mistake you see in designs? Is it the over-spec "just in case" crowd, or the under-spec "software will fix it" crowd?

What's the lowest phase noise OCXO you've actually put on a bench?

Peter Hu — Tue, 29 Sep 2026 15:17:11 +0000

I've been working with precision timing components for longer than I'd like to admit, and I've tested a lot of oscillators over the years. Most high-performance OCXOs land somewhere around -155 to -160 dBc/Hz at 1kHz offset, which is already pretty impressive if you've ever had to debug a noisy clock.
But the best one I've personally worked with hit -170 dBc/Hz at 1kHz offset.
For context, that's another factor of 10 quieter than a -160 dBc/Hz part — the kind of performance you'd normally associate with a rubidium atomic reference, except it's in a normal DIP package you can solder right on your PCB, at a fraction of the cost and size. When we first measured it on the bench I had to double-check the setup because it felt like the numbers shouldn't be possible with a crystal.
I'm curious — what's the lowest phase noise OCXO you've actually used in a real design? What were you building, and how much phase noise did your system actually need?
Drop it in the comments, I'm always interested to see what other engineers are working with.

DEV Community: Peter Hu

I keep seeing the same two oscillator selection mistakes in design reviews

What's the lowest phase noise OCXO you've actually put on a bench?