A recurring RF front-end failure: the LNA datasheet (or a single-frequency K-factor check) says “unconditionally stable” at the operating band, yet the finished board oscillates a few hundred megahertz away.
Practical checklist we use on the bench (open for critique):
1. K and Δ are per-frequency. Sweep the full S-parameter span the device is specified for, not only the band you care about. Matching networks present very different impedances out of band.
2. Plot input/output stability circles with the intended matching network attached. Bare-device circles miss the impedance the finished board actually presents; circles that carve through the match locus out of band are a classic warning.
3. Prefer μ-factor margin over a bare unconditional pass, and re-check at temperature and bias corners. “Just barely K>1” rarely survives production scatter.
4. Sweep the bias network at low frequency. 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.
5. Probe the finished board while sweeping bias — oscillation often shows up as a spur that moves with supply before it shows up in a tidy Smith-chart argument.
Question for engineers here: when you freeze a stability gate for a broadband LNA, do you set a minimum μ across the full S-parameter span, or a circle-clearance rule against the matching-network locus — and do you re-verify after a thermal cycle?
— rfforum editorial ·
https://www.rfforum.com
Cosmos