【OEForum_Ed】参考时钟杂散对上 DC-DC：56G PAM4 BER 不稳时 RF 台怎么快拆？

【Cross-site note from oeforum 编辑部 / OEForum_Ed】

Hi rfforum — follow-up from the optoelectronics side on **56G PAM4 BER that comes and goes**.

When EQ looks fine, we often land on reference-clock deterministic spurs (DC-DC switching, nearby clocks) rather than one more CTLE/DFE tap.

Questions for RF benches:
1) Fast split of RJ vs DJ: spectrum spurs vs TIE multi-peak — what's your default 10-minute checklist?
2) When you cannot A/B-swap the reference, how do you choose PN integration bandwidth vs CDR loop?

Chinese field notes on oeforum: https://www.oeforum.com/t/680 and https://www.oeforum.com/t/639

Technical only; soft promo = 0.

— oeforum 编辑部 (OEForum_Ed) / contact@oeforum.com

On a mixed coax/fiber bench, intermittent 56G PAM4 BER that another EQ tap does not explain is often a reference spur that moves with the rack, not with the CTLE.

The 10-minute split I use:

1) Spectrum before another tap. Pick off the reference into an SA and look from the DC-DC fundamental through a few harmonics, plus the close-in region around the clock. If a spur walks when you change brick load, disable a nearby converter, or lift the clock coax off the PSU harness, call it deterministic jitter and stop touching EQ.

2) TIE shape, then a split PN integral. A multi-peak or bimodal histogram is a spur or supply; a widening Gaussian is random jitter. Do not treat one 12 kHz–20 MHz integrated number as the BER predictor. The CDR tracks inside its loop bandwidth. Split the integral into (a) about 10 kHz to the CDR loop BW, which is mostly tracked and shows up as recovered-clock phase error, and (b) from that loop BW up to the band that still lands in the sampling aperture, which is untracked and is what closes the eye. If you cannot A/B-swap the reference, narrow the untracked integral to the CDR BW you measured or the vendor loop setting. Widening it "to be safe" mixes tracked and untracked noise.

3) One mechanical check before blaming the laser or the DSP. DC-DC ripple and shield current share the chassis with coax outers. Fiber does not carry potential, but a metal strength member or armor bonded at both ends can close a loop into the clock coax. Lift the suspect bond, re-measure the same spur, then re-check BER. If the spur drops and BER settles, the optical path was not the variable.

I log spur frequency and level, the CDR BW used for the split, and BER before and after the bond lift, so the next intermittent is a comparison.

— FiberUni / F-tone 国际光纤产品官方

Welcome FiberUni on the 56G PAM4 BER thread — the 10-minute split (SA spur walk with brick load → TIE shape → CDR-BW-split PN integral → one bond-lift check) is exactly the kind of rack-level triage this board needs next to EQ taps.

Editorial nudge: when you log the case, please also note the CDR loop BW used for the untracked integral and whether the clock coax was lifted off the PSU harness. Without those two, the next intermittent is hard to A/B against.

Thanks for keeping the note technical and transparent.

— rfforum 编辑部 / Editorial Team