Noise-source ENR vs physical temperature: the Y-factor that walked after the 346 sat on a warm bench

Started by LunaWei · 2026-11-18 01:57 UTC · 22 views · 5 replies
L Admin
[Technical article / FAQ, rfforum Editorial Team]

A noise-figure number that moved 0.2-0.4 dB after the ENR source sat on the bench, or after you moved it from a cart, is often not a DUT change. The ENR table is only valid at the temperature and connector state the cal lab assumed. A warm body, a hot attenuator in front, or a stale table invents a Y-factor walk that follows the 346 more than the amplifier.

1. What the source setup owns
- ENR is a function of the physical temperature of the noise diode and of any pad between the diode and the reference plane. A source left next to a PA, under a lamp, or held in a hand can move several tenths of a dB before the meter looks "settled."
- A table taken at 23 C and used at 30 C, or a table that still includes a pad you removed, is a systematic offset, not a noisy DUT.
- Excess noise from a poorly torqued or dirty 3.5 mm / SMA interface changes the delivered ENR. One remate that "fixes" NF is a connector clue, not an LNA clue.

2. Freeze list before you retune the LNA
- Source model, serial, ENR-table date, and whether the table includes the pad actually in the path
- Case or diode temperature, or at least "time since power-on / time since moved"
- A controlled remate under marked torque, then a second NF on a known attenuator or a second source
- Receiver / meter warm-up and the Y-factor method settings (averaging, SB, cold-source vs hot/cold)

3. Habit
Never freeze "NF drifted 0.3 dB" from a single scan after the noise source was moved or left on a warm bench. Archive table date, pad state, and a temperature or settle note next to every release NF. If a re-zero of the table, a remate, or a 15-minute settle kills the step, the DUT was never the patient.

4. Practical gate
- Leave the source and the first pad dressed in the final position for the settle time in the source manual (or >=15 min if unknown)
- Confirm the ENR table matches the pad stack
- Remate once; if NF jumps, stop the schematic change

Discussion: on your last NF walk that followed the 346 rather than the LNA, did the fix land in a stale ENR table, a warm source, or a remate - and do you require a dressed, settled source before a release number?

— rfforum 编辑部 / Editorial Team
L Admin
Addendum — freeze table date, pad state, and settle before an LNA ECO. A second source, or a known attenuator at the same plane, after a dressed 15-minute (or datasheet) wait belongs on the same sheet as the NF claim. If the walk dies after a remate or a re-load of the ENR table, the DUT was never the patient.

A source left next to a PA or held in a hand can move several tenths of a dB of delivered ENR before the meter looks settled. Archive case/diode temperature or "time since moved," and whether the table still includes a pad that is no longer in the path.

Open: on your last NF walk that followed the 346 rather than the LNA, did the fix land in a stale ENR table, a warm source, or a remate — and do you require a dressed, settled source before a release number?

— rfforum 编辑部 / Editorial Team
A Member
Agree that a 0.2–0.4 dB NF walk after the 346 sat on the bench is often the source, the pad list, or a remate, not the LNA. Three extras I freeze before touching the DUT.

First, the ENR table’s frequency grid. If the cal lab tabulated every 1 GHz and you are on an in-between tone, the meter’s interpolation is a systematic offset. A stale table is one failure mode; a perfectly current table used off-grid is another. Archive the exact frequency, the two neighbouring ENR points, and whether a user correction was applied.

Second, “the meter looks settled” is not the same as the diode having settled. After a move, a 28 V supply sag, or a source that was sitting next to a PA, the case may still be drifting after the displayed NF has stopped twitching. I want diode/case temperature or a datasheet wait, and I want the 28 V rail logged; a soft supply invents ENR error that no remate will fix.

Third, separate receiver calibration (K-factor / second-stage) from delivered ENR. If both moved, retuning the LNA will absorb a setup error into a new “gain.” A second source or a known pad at the same plane is useful, but only after both the ENR table and the receiver cal are frozen and the connector is dressed. If the walk dies when you reload the table or reseat, the DUT was not the patient.
L Admin
Thank you, ARM_Linux_Lee — treating the 0.2–0.4 dB NF walk after the 346 sat on the bench as often the source, the pad list, or a remate, then adding the ENR-table frequency grid, a real diode/case settle plus the 28 V rail, and a freeze of receiver cal versus delivered ENR, is the traveler we want next to any "the Y-factor walked after a warm sit" claim.

Editorial keepers:

1. Archive the exact tone, the two neighbouring ENR points, and whether a user correction was applied. A stale table is one failure mode; a current table used off-grid, with the meter interpolating, is another.
2. A settled display is not a settled diode. After a move, a 28 V sag, or a source that sat next to a PA, log diode/case temperature or a datasheet wait, and log the 28 V rail; a soft supply invents ENR error that no remate will fix.
3. Freeze receiver calibration (K-factor / second-stage) and delivered ENR before touching the DUT. If the walk dies when you reload the table or reseat, the DUT was not the patient; retuning the LNA will only absorb a setup error into a new "gain."

Open question: on the last walk that survived a frozen ENR table and a dressed remate, did the fix land in off-grid interpolation, a still-drifting diode or soft 28 V rail, or a receiver-cal change — and do you require both the table and the receiver cal to be frozen before opening an LNA ticket?

— rfforum 编辑部 / Editorial Team
N Member
The ENR-versus-physical-temperature correction is the one that catches people because the source datasheet says 5 dB and the instrument asks for 15 dB, and the 15 dB is only valid at 290 K. If the source body sits at 310 K in a rack, the excess noise ratio drops by roughly 0.1 dB per 10 K, which is exactly the order of the 0.2-0.4 dB walk described. We now log the source body temperature with every noise figure run and apply the correction afterwards, and the repeatability went from 0.3 dB to under 0.1 dB. How long do you let the source settle before the first Y-factor point?
R Member
We've tested a similar approach - decent stability, but BOM cost goes up a bit.
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.