LNA stability circles on the bench: when K-factor alone misleads you

K > 1 and |Δ| < 1 is the textbook unconditional-stability check, and I used to stop there. Then a prototype LNA oscillated at 3.8 GHz, well outside its 2.4 GHz operating band, while happily passing the K-factor test at the design frequency. Here is what I do differently now.
1. K-factor is a single-frequency verdict
K and Δ are computed per frequency. A device can be unconditionally stable at 2.4 GHz and potentially unstable at 1 GHz or 6 GHz, where your matching network presents a very different impedance. Always sweep the full S-parameter range the device is specified for, not just the band you care about.
2. Plot the stability circles
Input and output stability circles on the Smith chart show exactly which source and load impedances are dangerous. In my case, the output circle at 3.8 GHz carved through the region where my matching network sat — the network was fine at 2.4 GHz but wandered into the unstable region out of band.
3. Check stability with the actual matching network attached
Measuring S-parameters of the bare device is not enough. Once the matching network is in place, re-measure stability with the network's impedance presented, or better, probe the finished board with a spectrum analyzer while sweeping the bias — oscillation often shows up as a spur that moves with supply voltage.
4. Add margin, not just compliance
A design that is "just barely" unconditionally stable will drift into instability with temperature, bias variation, or component tolerance. I now aim for a stability margin (μ-factor comfortably above 1 across the band) rather than a bare pass, and I verify it at temperature extremes.
5. Watch the low-frequency end
Bias networks are notorious here: a poorly decoupled bias line can create a low-frequency resonance that the K-factor at RF never sees. A quick check with the bias tee and a low-frequency sweep has saved me more than once.
Has anyone else been bitten by out-of-band instability that the datasheet K-factor did not catch? What is your go-to stability verification routine?

Thank you, RF_Bench_Notes — the 3.8 GHz out-of-band oscillation that sailed past a 2.4 GHz K-factor pass is exactly the failure mode LNA reviews keep rediscovering.

What we will keep quoting from your checklist:

1. Sweep the full device S-parameter range, not the band you care about. K and Δ are per-frequency; a matching network that is fine in-band can sit inside an output stability circle a few hundred megahertz away.

2. Plot input/output stability circles with the intended matching network attached. Bare-device circles alone miss the impedance the finished board actually presents.

3. Prefer μ-factor margin over a bare unconditional pass, and re-check at temperature extremes and bias corners — “just barely K>1” rarely survives production scatter.

4. Do not skip the low-frequency bias-network sweep. A poorly decoupled bias line can create a resonance the RF-band K-factor never sees; a spectrum-analyser bias sweep on the finished board is still the cheapest catch.

Open question for the room: when you freeze a stability gate for a 2–6 GHz LNA, do you set a minimum μ across the full S-parameter span, or a circle-clearance rule (e.g. output circle stays outside |Γ|≤0.8 of the matching-network locus) — and do you re-verify that gate after a thermal cycle?

— rfforum 编辑部 / Editorial Team