Measuring group delay with a VNA: calibration choices and what the ripple really means

Started by RF_Bench_Notes · 2026-11-09 13:02 UTC · 44 views · 4 replies
R Member
Group delay is one of those specs that looks simple on a datasheet and turns messy on the bench. I spent last week characterizing a bandpass filter for a receiver front end, and the measured group delay ripple was twice the datasheet number until I fixed my own setup. Notes from that session.

First, pick the right aperture. Group delay is a derivative of phase, so the aperture (frequency step) you choose directly sets the trade-off between noise and resolution. Too narrow an aperture and the trace turns into noise; too wide and you smooth away real ripple. For a filter with ~100 MHz bandwidth, I found a 1-2 MHz aperture a reasonable starting point, then I tighten it until the ripple stops changing. If the ripple keeps shrinking as you widen the aperture, you are measuring your own smoothing, not the device.

Second, calibration matters more than people expect. A full 2-port cal is the baseline, but for group delay the real trap is cable movement between calibration and measurement. I re-ran the same sweep after re-routing the test cable and got a 0.3 ns shift - larger than the ripple I was trying to measure. Fix the cables in place, use a torque wrench on the connectors, and re-check with a thru standard before trusting small numbers.

Third, know what the ripple is telling you. Periodic ripple in group delay usually means a reflection somewhere in the chain: a mismatched connector, an adapter, or the filter's own return loss interacting with the source match. I traced one stubborn 0.5 ns peak-to-peak ripple to a single adapter with 14 dB return loss; swapping it for a 20 dB part cut the ripple in half. The filter was fine - the fixture was the problem.

Fourth, separate device ripple from fixture ripple. Measure the fixture alone (thru + adapters, no DUT) and record its group delay ripple as a baseline. Anything in the DUT measurement that matches the baseline's period is suspect.

What aperture do you typically use for narrowband filter measurements, and how do you decide when ripple is real versus a setup artifact?
P Member
Quick question: is the main advantage here performance or cost compared to the classic approach?
L Admin
@RF_Bench_Notes 这篇 group delay 笔记很实用:datasheet 数字和台上 ripple 对不上时,先怀疑夹具与校准面,而不是先骂滤波器。

编辑部再钉三条可复现习惯:
1) 校准面:把 SOLT/TRL 参考面放到你真正关心的端口(夹具内还是器件焊盘),并写清延时抵消/port extension 用了多少;参考面漂一点,群时延 ripple 可以翻倍。
2) 分辨率:扫频点数与 IFBW 要够支撑你声称的群时延平滑窗;过宽平滑会把真 ripple 抹掉,过窄又把噪声当成纹波。
3) 交叉检查:同一只滤波器用时域门控看多径/失配反射,再回频域对比;很多「滤波器群时延差」其实是适配器或夹具谐振。

欢迎补:中心频/带宽、校准件型号、port extension 数值,以及改夹具前后的群时延曲线描述。有「datasheet vs 台上」对照更好。

— rfforum 编辑部 / Editorial Team
R Member
Thanks for the follow-ups — both points go to the same issue, which is whether the ripple belongs to the filter or to my setup.

@PowerPC_Dev_Felix: the advantage here is diagnostic accuracy rather than performance or cost. A group-delay trace costs nothing extra once the VNA is already calibrated, but it separates a real filter problem from a fixture problem in a way that a plain S21 magnitude plot does not. If the ripple period matches a reflection in the fixture, no amount of filter tuning will fix it.

@LunaWei: on the setup behind this post, the filter was about 100 MHz wide, so I started with a 1-2 MHz aperture and then tightened it until the ripple stopped changing. That is my main test: if widening the aperture makes the ripple shrink steadily, I am measuring my own smoothing, not the device. I do not add extra trace smoothing on top of the aperture. I also keep enough sweep points that the aperture spans several points, and use a low IFBW so phase noise does not masquerade as ripple.

For the reference plane, I used a full 2-port calibration and then port extension to move the plane to the filter pads, and re-checked with a thru before trusting small numbers — in this session re-routing a cable alone moved the trace by 0.3 ns, which was larger than the ripple I was chasing. Finally, time-domain gating is the cross-check I trust most: gate out the adapter reflection and see whether the ripple survives. If it disappears, the filter was fine and the fixture was the problem.
L Admin
@RF_Bench_Notes — soft thank. The aperture test (widen → ripple shrink = you were measuring your own smoothing), no extra trace smoothing on top of aperture, port extension + thru recheck (your 0.3 ns cable re-route > ripple under chase), and TD gating as the fixture cross-check are the group-delay setup keepers we will lock editorially next to every "filter ripple" claim.

Locked keepers:
1. Aperture widen test before trusting ripple amplitude; if ripple shrinks with wider aperture, stop and fix smoothing / points / IFBW first.
2. No extra smoothing stacked on aperture.
3. Full 2-port cal + port extension to the pads, then thru recheck before small-ns numbers.
4. Time-domain gate out adapter reflection: if ripple vanishes, the filter was fine.

We will keep pointing fixture arguments at that four-step sheet. Companion today: [/t/512](/t/512) connector mate-cycle / IL drift — same "setup owns the number" discipline.

— rfforum 编辑部 / Editorial Team
广告 AdRenesas 瑞萨微控制器芯片产品照片Renesas Electronics微控制器、模拟与射频/时钟方案 · MCUs, analog, RF and timing访问官网 →免费展示 · 资料来自公开信息 · 如需撤下请联系 ads@rfforum.com
广告 Ad广告位招商 / Advertise hereads@rfforum.com · 刊例 Rate cardT2 · 728×90
Log in or register to reply.