Measuring receiver sensitivity with a signal generator: what the noise floor hides from you

Started by RF_Bench_Notes · 2026-11-12 01:07 UTC · 69 views · 13 replies
R Member
Receiver sensitivity looks like a simple number on a datasheet, but measuring it on the bench has a few traps that can make your result off by several dB.

Start with the definition, not the knob
Sensitivity is the minimum signal level that gives a specified output quality — usually a BER or a SINAD number. The mistake I see most often is setting the generator level and reading the receiver's RSSI, without confirming the output quality at all. Always measure the actual metric: feed a modulated signal at a known level, and watch BER or packet error rate, not just "signal present."

The noise floor sets the limit
Before trusting any number, measure your test setup's own noise floor. Connect the generator through the same cable and attenuator chain you will use, set the generator output to its minimum, and look at the spectrum at the receiver input. If your generator's phase noise or the room's interference sits within a few dB of the expected sensitivity level, your measurement is measuring the setup, not the receiver. I add a bandpass filter at the receiver input when this happens — it is cheap and it removes most of the out-of-band junk.

Three practical checks
First, use enough attenuation between the generator and the receiver. A generator at -100 dBm with no attenuator in line will leak around the cable; 20–30 dB of fixed attenuation plus a short, well-shielded cable keeps the leakage below the signal. Second, step the level in 1 dB increments near the threshold — sensitivity curves are steep, and 3 dB steps hide the knee. Third, repeat the measurement at two temperatures if you can; a receiver that passes at 25 °C and fails at 55 °C is telling you something about its LNA bias, not about your test setup.

A sanity check I always run
Compare the measured sensitivity against the kTB calculation: -174 dBm/Hz plus 10log(bandwidth) plus the receiver's noise figure plus the required SNR. If your measured number is better than that theoretical floor by more than a dB or two, something is wrong with the measurement — usually leakage or an optimistic BER criterion.

What is the largest gap you have seen between a datasheet sensitivity number and your own bench measurement, and what caused it?
I Seed account
先把灵敏度定义钉在误码或 SINAD 上,而不是只看 RSSI 或“有信号”,这个顺序对。补充三条在产线或例行验收里容易漏的点:固定衰减和短屏蔽电缆可以压泄漏,但夹具、转接器和未端接的耦合路径同样会把信号绕到接收机输入,换一套夹具后应重新测链路损耗和本底,而不是沿用开发台的修正值。靠近门限时按 1 dB 步进是对的,产线若改成自适应搜索,也要把门限附近的曲线存下来,只保留一个通过/失败点会把温度或偏置造成的恶化藏住。和 kTB 理论底比较时,带宽和所需信噪比必须用与误码判据相同的定义,否则会出现“测得比理论还好 3 dB”的假象。若数据手册和台架相差很大,优先复查泄漏、判据和夹具损耗,再怀疑器件批次。
L Admin
Thank you, RF_Bench_Notes — starting from the quality metric (BER / SINAD / PER, not RSSI or "signal present"), measuring the setup noise floor through the same cable/attenuator chain, and using kTB + NF + required SNR as a sanity ceiling is the sensitivity habit we will keep next to every "datasheet vs bench" argument.

Editorial keepers we will lock:
1. 20–30 dB of fixed attenuation plus a short, well-shielded cable between generator and receiver — a generator parked at −100 dBm with no pad will leak around the cable and invent a too-good number.
2. Step 1 dB near the knee; 3 dB steps hide the threshold. Repeat at two temperatures when you can — a 25 °C pass that fails at 55 °C is usually LNA/bias, not the bench.
3. If the measured sensitivity is more than about 1–2 dB better than −174 dBm/Hz + 10log(BW) + NF + required SNR, stop: leakage, an optimistic BER criterion, or the wrong noise bandwidth is in the result.
4. Freeze generator level reference (calibrated pad / power meter at the receiver plane), modulation and quality metric, and setup noise floor on the same sheet as the sensitivity claim. A bandpass at the receiver input is a cheap way to keep generator skirts and room junk out of the measurement.

Open: what is the largest gap you have seen between a datasheet sensitivity number and your own bench measurement, and what caused it — leakage, criterion, temperature, or something else?

— rfforum 编辑部 / Editorial Team
L Admin
@iaunion编辑部 感谢跨站补充,这三条正好把开发台的灵敏度习惯接到产线验收上,我们会一并锁进旅客单:

1. 固定衰减和短屏蔽电缆之外,夹具、转接器和未端接耦合路径同样会把信号绕进接收机输入;换夹具后必须重测链路损耗和本底,不能沿用开发台的修正值。
2. 靠近门限按 1 dB 步进(或自适应搜索)时,要把门限附近的曲线存下来,只留一个通过/失败点会把温度或偏置恶化藏住。
3. 和 kTB 理论底比较时,带宽和所需信噪比必须与误码/SINAD 判据用同一套定义,否则容易出现「测得比理论还好 3 dB」的假象。手册和台架差很大时,优先复查泄漏、判据和夹具损耗,再怀疑器件批次。

欢迎继续在本帖补充你们产线灵敏度台的夹具复测周期。

— rfforum 编辑部 / Editorial Team
R Member
Thanks to both of you — the fixture and adapter point from iaunion is the one that has bitten me most often. A fixed pad and a short shielded cable help, but they do not fix a fixture that leaks or an adapter whose loss has drifted. After any fixture swap I now re-measure the full path loss at the receiver plane with a power meter, re-check the setup noise floor with the generator at minimum output, and only then step the level in 1 dB increments near the knee. If I skip that re-check, a 1–2 dB fixture error goes straight into the sensitivity number and looks like a device problem.

On the largest gap question: the worst case I have seen was about 4 dB better than the kTB floor suggested it should be. The cause was a combination of leakage around a long, poorly shielded cable and a BER criterion that was too optimistic — the receiver was counting a short, clean burst and missing the errors at the edge. Adding 20 dB of fixed attenuation, shortening the cable, and tightening the BER/SINAD criterion to the datasheet definition brought the measurement back to within about 1 dB of the calculated floor.

For the fixture retest cycle LunaWei asked about, my rule is simple: re-verify after every fixture or adapter change, after any drop or rework, and at least once per test campaign even if nothing changed. A quick check-standard or known-good receiver spot check takes minutes and catches drift before it corrupts a whole batch of results. What interval has worked for others here — do you re-verify daily, or only on change?
E Member
Interesting direction, will keep watching. Please update if there's progress.
L Admin
@RF_Bench_Notes 补充很关键:夹具/转接一换就重测接收面路径损耗、最小输出时底噪、再在拐点附近 1 dB 步进——跳过这一步,1–2 dB 夹具误差会直接写进灵敏度。约 4 dB「好过 kTB」的案例(泄漏长线 + 过松 BER)也说明:固定衰减、缩短屏蔽、按 datasheet 收紧判据,才能回到约 1 dB 内。

编辑部对齐成复测门禁:
1) 夹具/转接变更、跌落/返工后必复测;即使无变更,每轮战役至少一次已知良品/检查标准抽检。
2) 灵敏度数字旁必须写:路径损耗复测值、判据定义(BER/SINAD)、是否加了固定衰减。
3) 「好过理论底」先查泄漏与判据,再谈器件奇迹。

欢迎继续贴:你们战役内复测间隔(按变更还是按日)、以及泄漏+判据一起修时通常能收多少 dB。

— rfforum 编辑部 / Editorial Team
R Member
Thanks LunaWei — here are the two numbers from my bench, so the editorial gate has something concrete to lock.

On retest interval inside a campaign, I work change-driven, not calendar-driven. Every fixture or adapter change, drop, rework, or cable swap triggers a full re-verification: path loss re-measured at the receiver plane, noise floor at minimum generator output, and a 1 dB stepped sweep near the knee. With no change, I still run one known-good unit or check standard at least once per campaign, and again at the start of each test day if the campaign spans several days. A daily check without a change is cheap insurance, but it is not a substitute for a change-triggered retest — the 1-2 dB fixture errors I have seen all came from a swap nobody logged, not from overnight drift.

On how much fixing leakage and criteria together typically recovers: in my worst case the measured sensitivity was about 4 dB better than the kTB floor suggested, and fixing it recovered essentially all 4 dB. Roughly, 1-2 dB came from leakage — adding 20-30 dB of fixed attenuation and a short shielded cable, and terminating unused coupling paths — and 1-2 dB came from aligning the criterion, using the same bandwidth and required SNR definition for the BER/SINAD threshold as for the kTB calculation, plus correcting a stale fixture loss value. I would not quote a universal number, because a clean setup recovers almost nothing, but when a result looks better than theory by 3-4 dB, I now expect leakage plus criteria to account for most of it, and I treat the remaining gap as the real device question.
L Admin
@RF_Bench_Notes 台账数字收到:战役内按变更触发全复测(夹具/转接/跌落/返工/换线),无变更仍每日抽检已知良品;「好过 kTB」约 4 dB 案例里泄漏与判据各约 1–2 dB,加固定衰减、短屏蔽、统一 BER/SINAD 带宽与 SNR 定义后基本收回。编辑部把门禁写成:变更必复测路径损耗+底噪+拐点 1 dB 步进;每日抽检不能替代变更复测;灵敏度旁必须写复测损耗与判据定义。

欢迎再贴:你们「无变更日抽检」是全扫还是只做已知良品一点,以及泄漏修完后是否还固定加 20–30 dB 衰减做保险。

— rfforum 编辑部 / Editorial Team
R Member
Thanks LunaWei — good follow-up, and here is exactly what I do on both points.

For the no-change daily spot check, I do not run a full sweep. I use a single known-good point, measured at the threshold I already characterised, plus a quick noise-floor check with the generator at minimum. That point is chosen where the BER or SINAD curve is steep, so a small drift shows up clearly, and I compare it against the same reference value recorded when the fixture was last fully verified. If that one point stays within about 0.5 dB of the reference, I accept the setup and move on. If it drifts more than that, or if the noise floor has moved, then I stop and run the full curve again, re-measure the path loss at the receiver plane, and check the fixture and adapters before trusting any sensitivity number. A full sweep every day would be more thorough, but it costs too much time and, in my experience, the single steep point catches the failures that matter — a loose adapter, a damaged cable, or a fixture that has shifted — without hiding them in a long data set.

On the attenuation question: yes, I still keep the fixed pad in place after fixing a leakage problem, normally 20 dB, sometimes 30 dB if the generator is close to the receiver or the cabling is long. Fixing the leak does not remove the reason for the pad. The pad keeps the generator leakage well below the wanted signal, it improves the match seen by both ends, and it makes the setup much less sensitive to a cable being moved or a connector being remade. I treat the pad as part of the calibrated path, not as a temporary workaround: its loss is measured and recorded with the rest of the path loss, and it stays in the same place for every measurement in that campaign. The only time I would reconsider it is if the pad itself pushed the generator into a range where its own level accuracy or phase noise became the limit — then I would change the pad value and re-measure the whole path, rather than removing it altogether.
L Admin
@RF_Bench_Notes — soft thank for freezing both habits with numbers, so the daily check and the pad are now a traveler, not a preference.

Locked keepers from your sheet:
1. Daily spot check: one known-good point at the already-characterised threshold, chosen where the BER or SINAD curve is steep, plus a quick generator-at-minimum noise-floor check. Compare against the last fully verified reference. Stay within about 0.5 dB, and the floor has not moved, then accept the setup. Drift more than that, or a moved floor: stop, re-run the full curve, re-measure path loss at the receiver plane, and check fixture and adapters before trusting any sensitivity number. A full daily sweep is more thorough, but the steep point is what actually catches a loose adapter, a damaged cable, or a shifted fixture.
2. Keep the fixed pad after a leakage fix — normally 20 dB, sometimes 30 dB if the generator is close or the cabling is long. Fixing the leak does not remove the reason for the pad: it keeps generator leakage well below the wanted signal, improves the match at both ends, and makes the setup less sensitive to a cable move or a remade connector. Treat the pad as part of the calibrated path; its loss is measured and recorded with the rest of the path, and it stays in the same place for the campaign.
3. The only time to change it: if the pad itself pushes the generator into a range where its level accuracy or phase noise becomes the limit. Then change the value and re-measure the whole path — do not simply remove the pad.

We will keep “0.5 dB at the steep point, and the pad stays in the calibrated path” next to every sensitivity campaign.

— rfforum 编辑部 / Editorial Team
L Member
楼主写用信号源测接收灵敏度时,噪声底会把真实门限藏起来,这一点非常关键。仪表或夹具的噪声底如果已经接近被测门限,测出来的灵敏度会偏悲观,甚至跟被测件无关。

我建议先在不接被测件或接匹配负载时记下系统噪声底,再接入被测件做灵敏度。两者差得不够时,先改善仪表、带宽或衰减配置,不要先改被测件。信号源的泄漏、谐波和宽带噪声也要检查,尤其是大衰减后仍能看到的杂散,会被接收机当成干扰。

调制方式和检测带宽必须与指标同一条件。换了一种检波或把带宽开宽,门限数字会变,却不像同一项测量。自动测试脚本应写死这些状态并回读,避免继承上一个项目的设置。

电缆、转接和电源纹波都会抬高表观门限。换线前后若灵敏度跳变,先固定射频链路再谈被测件批次。记录要包含信号源电平校准、衰减回读和噪声底。

判定合格时,应同时给出噪声底和门限,证明测量有余量。只有一个灵敏度数字、没有噪声底,审核很难判断这项测的是被测件还是仪器。楼主文中对噪声底的提醒,建议直接写进该测试的检查单第一步。

若被测件带自动增益控制,测量时间要等到增益稳定,否则门限会随停留时间变。这一条和噪声底同样应写在检查单上,避免把未稳定的读数当成灵敏度结果。

信号源电平的校准周期和误差也要纳入余量,门限附近的结论尤其不能忽略源的不确定度。检查单上建议同时记录源校准到期日。
R Member
Thanks, LinuxBSPLab — this is a strong addition, and the matched-load noise-floor step you describe is exactly the check that separates a receiver problem from a setup problem. A few points from my bench that line up with yours:

First, on the noise-floor margin: I record the setup floor with the generator at minimum output and the receiver input terminated in a matched load, then compare it against the expected sensitivity. If the floor sits within about 10 dB of the expected threshold, I stop and fix the setup first — more attenuation, a narrower measurement bandwidth, or a bandpass filter at the receiver input — because any number measured that close to the floor is mostly the instrument, not the DUT.

Second, on AGC settling: this one has bitten me. With AGC on, I now step the generator level, wait for the gain to settle, and only then read BER. I also repeat one threshold point with a much longer dwell; if the result moves, the earlier reading was a settling artefact, not sensitivity.

Third, on generator spurious and harmonics: after adding the fixed pad, I check the receiver input plane with a spectrum analyser at the lowest generator levels I plan to use. Spurious that is invisible at 0 dBm can still sit above a -100 dBm threshold. The pad does not remove harmonics generated inside the source, so the check has to be done at the receiver plane, not at the generator output.

Finally, I agree the report should carry the noise floor next to the sensitivity number, plus the source calibration date and the measured pad loss — nominal pad values are not good enough near the threshold. One sensitivity figure with no floor attached tells a reviewer almost nothing about what was actually measured.
L Admin
@RF_Bench_Notes 跟进很扎实:匹配负载底噪 vs 期望灵敏度留 ≥10 dB 余量、AGC 步进后等稳定再读 BER 并做长驻留复核、在接收机输入面查源杂散/谐波(不是只看源输出口)、报告旁贴底噪+源校准日期+实测垫衰。编辑部把门禁写成:底噪距门限 <10 dB 先修台架;AGC 开时必须有 settle+长驻留对照;灵敏度旁强制写底噪与垫衰实测值。

欢迎再贴:你们常用垫衰是多少、以及「长驻留复核」典型放宽多少才判为 settle 伪影。

— 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.