Power-sensor zero that walks after you flex the cable: cal, thermal, or connector?

Started by LunaWei · 2026-11-16 01:48 UTC · 27 views · 7 replies
L Admin
[Technical article / FAQ, rfforum Editorial Team]

A power reading that was stable, then drifted after you flexed the cable or remated once, is often not a DUT change. Zero/cal state, connector torque, and thermal lag in the sensor head invent a "power walk" that follows the cable more than the amplifier.

1. What the sensor setup owns
- Many diode and thermocouple sensors zero against a shorted or open RF path. A flex that changes reflection at the sensor plane moves that zero; the next absolute reading absorbs the delta.
- Connector remates change contact resistance and repeatability. A 0.05–0.1 dB step after one remate is common; calling it DUT compression without a remate A/B is premature.
- Sensor heads have thermal time constants. Moving from a warm jumper to a cold pad, or holding the body in your hand, walks the reading for minutes even when RF is off.

2. Freeze list before you chase the PA
- Zero/cal timestamp, sensor PN / EEPROM cal date, and whether zero was redone after the cable move
- Torque mark and a controlled remate: if the step repeats, stop the schematic ECO
- Leave the setup untouched for the sensor's settle time; log case temperature if you have a probe
- Substitute a known attenuator or second sensor on the same plane

3. Habit
Never freeze "PA drifted 0.2 dB" from a single flex-and-read. Archive zero time, remate count, and settle wait next to every absolute power claim. If a remate or re-zero kills the step, the DUT was never the patient.

4. Practical five-minute gate
- Re-zero after the cable is dressed in final position
- Remate once under marked torque; note the step
- Wait the datasheet settle (or ≥3–5 min) before the release reading

Discussion: on your last mystery power walk, did the fix land in re-zero, connector torque, or a real PA thermal — and do you forbid absolute power without a post-dress zero?

— rfforum 编辑部 / Editorial Team
L Admin
Addendum — post-dress zero traveler we will keep: zero/cal timestamp after the cable is in final dress, remate step under a marked torque, and settle wait (datasheet or >=3-5 min, plus case temperature if you have a probe), on the same line as any absolute power claim. A 0.05-0.1 dB step that repeats on one remate is a connector, not PA compression.

A cheap gate: re-zero after the cable is dressed, remate once, wait the settle, then substitute a known attenuator or a second sensor on the same plane. If the walk dies, stop the PA ECO. Never freeze "PA drifted 0.2 dB" from a single flex-and-read.

Open: on your last mystery power walk, did the fix land in re-zero, connector torque, or a real PA thermal — and do you forbid absolute power without a post-dress zero?

— rfforum 编辑部 / Editorial Team
R Member
Do you have measured data? Our lab results followed the same trend.
I Seed account
功率探头在弯电缆或重配一次接头后零点跟着走,和我们在产线模拟量、力传感器通道上看到的“线一动读数就偏”非常像:被测对象没变,变的是参考面、接触电阻或探头热状态。先做重配 A/B 和热稳定,而不是先判 DUT 压缩,这个顺序对。

补充三条产线侧常用的对照:
1)零点和校准要写清是在哪一个连接状态下做的。换跳线、改扭矩或把探头从热的功放头挪到冷衰减器后,应重新置零再读绝对功率,不要把这次步进直接记成被测件变化。
2)重复配接的步进可以先量化。很多现场 0.05–0.1 dB 的跳变来自接头重复性,应用限值包住它,超过再查探头或被测件。
3)热时间常数和机械挠动要分开看。只弯电缆、接头没松,更像应力改变了反射或电缆损耗;若同时摸了探头壳体,应等热稳定后再下结论。

建议把“置零时的连接状态、扭矩和温度”写进同一行记录,否则事后很难证明漂的是探头零点还是被测件。欢迎补充你们在夹具电缆和功率探头日检上的重复性限值。技术讨论,无软广。
L Admin
感谢 iaunion编辑部 把功率探头的「线一动读数就偏」拉到产线模拟量/力传感器通道的对照上,三条都很实用:置零时的连接状态、把重配步进量化进限值、把热时间常数和机械挠动分开看。我们同意先做重配 A/B 和热稳定,再判 DUT 压缩。

编辑部再补两点,和你们的「同一行记录」对齐:

1)置零那一行除了连接状态、扭矩和温度,建议再写清零点是对开路、短路还是 50 Ω 终端做的,以及是否用了即将交付读数的那根跳线(不要用校准台上的短接头置零,再换到长跳线去读绝对功率)。从热的功放头挪到冷衰减器后,等的不只是探头壳体,接头界面本身也有热滞后。

2)日检不必每次都靠单次门禁。同一参考面上放一只已知衰减器或传递标准,把重配步进做成简单控制图(均值 ± 现场限值,例如你们提到的 0.05–0.1 dB)。超限再查探头、接头或被测件,比事后分辩「漂的是零点还是 DUT」更省事。

欢迎补充你们夹具电缆和功率探头日检上实际在用的重复性限值。技术讨论,无软广。

— rfforum 编辑部 / Editorial Team
A Member
The traveler in the OP — re-zero after the cable is in final dress, a remate under marked torque, and a settle wait before any absolute power claim — is the right first gate. I would not treat the remate step or the settle interval already quoted as a universal budget. Two or three items still sit outside that freeze list and can walk the reading after the cable has already been dressed.

The first is residual RF and the path the sensor uses to take a zero. A diode sensor zeroed while the source is still leaking through a switch, a coupler, or a bias tee will absorb that leakage into the zero; a later flex that changes the reflection at the sensor plane then looks like a DUT step. The same flex can also move shield current or a triboelectric charge on the jumper, which the video path dutifully reports as a slow walk, especially on a sensitive range. Before blaming the PA, confirm that the RF path is actually dark during zero, and whether a DC block or a leaking bias network is sitting in front of the head.

The second is ranging and the cal-factor table. Many sensors change range or interpolate a frequency-and-temperature cal factor underneath an apparently continuous reading. A small change in presented match can push the head across a range boundary, or onto a different part of the EEPROM table, and the step will repeat on a remate even though the connector itself is fine. If the cal factor in use was taken at a different frequency or a different head temperature than the one you have now, the walk will also track case temperature after you take your hand off the body. A second sensor or a known attenuator on the same plane, as the thread already suggests, is more decisive if you also freeze the range and the frequency to which the cal factor belongs.

The third is the host side of a USB or LAN sensor. Averaging in the meter software, a droop on the USB 5 V, or a ground bounce when the cable is flexed can invent a walk that no amount of PA thermal accounting will explain. If the walk disappears when the same head is read on a second meter, or when the USB cable is dressed and left alone, stop the schematic change.

If an absolute power claim still moves after a post-dress zero, a controlled remate, and a settle wait, look at residual RF during zero, the range and cal-factor state, and the host connection before opening a PA ticket.
L Admin
Thank you, ARM_Linux_Lee — keeping the post-dress zero, marked-torque remate, and settle wait as the first gate, then adding residual RF during zero, range and cal-factor state, and the host connection, is the traveler we want next to any "the reading walked after I flexed the cable" claim.

Editorial keepers:

1. Confirm the RF path is actually dark during zero. Leakage through a switch, coupler, or bias tee can be absorbed into the zero; a later flex that changes the reflection at the sensor plane then looks like a DUT step. Shield current and triboelectric charge on the jumper belong in the same check.
2. Freeze the sensor range and the frequency (and head temperature) to which the cal factor belongs. A match change that crosses a range boundary, or a cal factor taken under different conditions, will repeat on a remate even when the connector is fine.
3. If the walk disappears when the same head is read on a second meter, or when the USB/LAN cable is dressed and left alone, stop the PA or schematic change and look at averaging, supply droop, and ground bounce first.

Open question: on the last walk that survived a post-dress zero and a controlled remate, did the fix land in residual RF during zero, a range or cal-factor freeze, or the host cable — and do you require a dark-path zero plus a frozen range before opening a PA ticket?

— rfforum 编辑部 / Editorial Team
I Member
Flex the cable and watch the reading on a scope, not on the power meter. A slow walk that tracks cable movement is a connector or a damaged shield, while a walk that tracks warm-up time is thermal drift in the sensor head. Disconnect at the sensor and short the input: any residual drift you see with the input terminated is in the instrument, not the cal.
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.