Oscilloscope probe loading on high-speed signals: passive vs. active probes, and where each lies to you

Started by RF_Bench_Notes · 2026-11-10 13:03 UTC · 40 views · 4 replies
R Member
A probe is part of the circuit the moment it touches the board, and on fast edges it lies to you in ways that look like real signal problems. I have chased "ringing" that turned out to be my probe, and missed real ringing because the probe filtered it out. Here is how I choose and how I sanity-check what I see.

First, know what a passive probe does to the signal. A typical 10x passive probe presents about 10 MΩ at DC but only a few pF of capacitance, and that capacitance is what matters: at 500 MHz, 10 pF is roughly 30 Ω to ground. On a high-impedance node that is a heavy load. It rounds off edges, kills overshoot, and can make a marginal driver look clean. If the edge looks suspiciously perfect on a fast node, suspect the probe before you trust the waveform.

Second, active probes solve the loading problem but bring their own traps. Input capacitance drops below 1 pF, which is why they are the right tool above a few hundred MHz. The catch is dynamic range: most active probes clip at a few volts, and the clipping is not always obvious on screen — it can look like a flat top on a signal that is actually still rising. Check the probe's linear range against your signal amplitude before you believe the top of the pulse.

Third, the ground lead is usually the biggest error source. A 6-inch ground clip adds roughly 100 nH, which resonates with the probe capacitance right in the 100–300 MHz band — exactly where probe-induced ringing lives. When I see ringing in that band, my first move is to shorten the ground to a spring tip or a soldered ground and re-measure. If the ringing moves or disappears, it was the probe.

Fourth, compare two probes on the same node. If a passive probe and an active probe disagree on rise time or overshoot, the truth is usually closer to the active probe — but only after you have checked the active probe's own bandwidth and calibration. A quick sanity check: measure a known-good fast edge, like the scope's own calibration output or a reference pulser, with each probe and confirm the numbers make sense.

Fifth, remember the scope itself. Probe bandwidth, scope bandwidth, and the channel's own front end all stack. A 1 GHz probe on a 500 MHz scope gives you a 500 MHz system, not a 1 GHz one. Match the chain to the signal: for edges faster than about 1 ns, I want the whole chain at 1 GHz or better, and I verify it with a reference edge rather than trusting datasheet numbers.

My rule of thumb now: passive probes for power rails and slow control signals, active probes for anything with edges under a few nanoseconds, and a short ground always. What is your go-to probe check when a waveform looks too good — or too bad — to be true?
R Member
We've tested a similar approach - decent stability, but BOM cost goes up a bit.
L Admin
@RF_Bench_Notes 这篇把「探头在骗你」拆得很清楚:无源容性负载、有源动态范围、地线电感三段,基本覆盖了高速沿上最常见的假象。

编辑部再钉三条台上习惯,方便后来者对照:
1) 先缩短地回路再谈换探头:弹簧地 / 焊地 vs 长鳄鱼夹,很多 100–300 MHz「振铃」一改地就搬家或消失。
2) 同一节点双探头交叉:无源 vs 有源对上升沿/过冲不一致时,优先信有源的上升时间,但要先核对有源线性区,避免削顶被当成平顶。
3) 记录三联:探头型号与衰减、地线长度/方式、示波器带宽与采样;没有这三条,别人很难复现你的「假干净」或「假振铃」。

欢迎补:你常用的节点阻抗量级、无源/有源具体型号,以及「改地前后」或「双探头对照」的波形描述。有带宽受限故意掩盖真振铃的反例也特别有价值。

— rfforum 编辑部 / Editorial Team
R Member
Good follow-up — here are the specifics from my bench, so the comparison is reproducible.

Node impedance matters more than probe bandwidth alone. The nodes where I see the biggest probe lies are high-impedance ones: a CMOS clock input or crystal oscillator node at roughly 1–5 kΩ, and a 50 Ω RF trace driven through a series resistor. On the 50 Ω trace a 10 pF passive probe is a minor load; on the high-Z node the same 10 pF is about 30 Ω to ground at 500 MHz, so it rounds the edge and kills real overshoot.

Probes I actually use: a 10x passive probe, 500 MHz, about 10 pF, for power rails and slow control; and a 1.5 GHz active FET probe, under 1 pF, for fast edges. I always record probe model, attenuation, ground length and method, and scope bandwidth, as you suggested.

Ground change, same node, same 800 ps edge: with a 6-inch ground clip I measured about 320 mV of overshoot ringing near 210 MHz and a 1.4 ns rise time. With a spring ground tip soldered close to the node, the ringing fell to about 90 mV and the rise time came to 850 ps. The ringing moved with the ground length, which confirmed it was probe-induced.

Dual-probe check on the same node: the passive probe reported 1.1 ns rise time, the active probe 620 ps. I trust the active probe here, after confirming it was inside its linear range — the top was flat because the signal had settled, not because the probe was clipping.

One masking counterexample: a 200 MHz passive probe on a fast driver hid a real 450 MHz ringing completely; the waveform looked clean until I repeated it with the active probe and saw about 180 mV of ringing that the limited bandwidth had averaged away. That is why I never call a fast edge clean on a bandwidth-limited probe alone.
L Admin
@RF_Bench_Notes — thank you for the reproducible numbers. They turn this morning's "probe can lie" caution into a traveler we can actually freeze.

Locked keepers from your bench:
1. Node impedance owns the lie more than probe bandwidth alone. On a 50 Ω trace a 10 pF passive probe is a minor load; on a 1–5 kΩ CMOS clock or crystal node the same 10 pF is about 30 Ω to ground at 500 MHz, so it rounds the edge and kills real overshoot.
2. Ground length is a first-order experiment: same 800 ps edge, 6-inch clip → ~320 mV ringing near 210 MHz and 1.4 ns rise; spring tip close to the node → ~90 mV and 850 ps. If the ringing moves with ground length, it is probe-induced.
3. Dual-probe check: passive 1.1 ns versus active 620 ps on the same node, after confirming the active probe was in its linear range (flat top because the signal settled, not because the probe clipped).
4. Bandwidth can hide a real problem: a 200 MHz passive probe erased a real 450 MHz / ~180 mV ringing that the active probe showed. Never call a fast edge clean on a bandwidth-limited probe alone.

Traveler: probe model, attenuation, ground length and method, scope bandwidth, node-impedance class, and a dual-probe or bandwidth-adequate check — on the same sheet as any "edge is clean / overshoot is X mV" claim.

— 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.