GaN PA gate bias feed resonance: a supply that looks quiet on a DMM and sings on a spectrum analyzer

Started by LunaWei · 2026-11-12 01:54 UTC · 48 views · 3 replies
L Admin
[Technical article / FAQ, rfforum Editorial Team]

A GaN PA gate bias that reads rock-steady on a DMM can still be a resonant network that sings under RF drive. The feed inductor, bypass set, and cable inductance form a tank; the spectrum analyzer sees sidebands or bursts the voltmeter never will.

1. What the feed actually does
- Gate current is small, so people undersize decoupling and leave long bias wires. Under RF, envelope and video products modulate the gate node through that inductance.
- A ferrite + capacitor combination that "looks fine" at DC can resonate in the low-MHz to tens-of-MHz range and up-convert into close-in skirts or pulsed ACLR mess.
- Cold start and warm bias can land on different sides of a stability boundary; a CW bias-point check after warm-up misses the start-up sing.

2. Freeze list before you blame the GaN die
- Bias network schematic at the gate: series R, ferrites, cap values, and wire/cable length to the supply
- Spectrum at the gate node (high-Z / resistive probe) under the same pulsed or modulated drive as the ACLR test
- Whether the skirt or burst moves when you shorten the bias lead or add a local bypass at the pin
- Cold-start versus warm spectrum on the same board, same load

3. Habit
Never close a "phase noise / ACLR die issue" without a gate-node spectrum under drive. A quiet DMM is not a quiet bias feed. Archive the bias-network one-liner next to every close-in skirt claim.

Discussion: on your last mysterious GaN skirt, did the fix land in local bypass, series gate R, or a shorter bias lead — and do you require a gate-node spectrum on every pulsed ACLR traveler?

— rfforum 编辑部 / Editorial Team
L Admin
Addendum — GaN gate-bias traveler: series R / ferrite / bypass values, bias-lead length, and a high-Z gate-node spectrum under the same pulsed or modulated drive as the ACLR test — next to every "phase noise / close-in skirt is the die" claim. A quiet DMM is not a quiet bias feed. Cold-start versus warm spectrum on the same board belongs on that sheet.

Open: on your last mysterious GaN skirt, did the fix land in local bypass, series gate R, or a shorter bias lead — and do you require a gate-node spectrum on every pulsed ACLR traveler?

— rfforum 编辑部 / Editorial Team
R Member
Do you have measured data? Our lab results followed the same trend.
L Admin
Complement to the open question: the last skirt we chased landed in a local bypass + a few ohms of series gate R, not in the die, and a quiet DMM had nothing to say about it. We now require a gate-node spectrum under the same drive as the ACLR shot — with one caveat about the probe.

1. Where the fix actually landed

- The tank was the bias-lead inductance (a few tens of nH on a ~8–10 cm wire plus a ferrite) against the local bypass. On the SA it showed up as sidebands at f_c ± f_tank, with f_tank in the ~5–20 MHz region — exactly the “looks like close-in phase noise / ACLR skirt” signature, invisible on a DMM.
- Shortening the lead helped; what closed it was a 100 pF–1 nF right at the gate pin (not centimetres away) and a series 5–22 Ω at the gate. That R kills the Q of the bias tank without eating much RF gain if it sits in the bias arm, not in the RF path. A ferrite hanging out on a long wire, with no local C, was the configuration that sang.
- Cold-start: the same board was quiet after warm-up and sang for the first 1–2 s after bias came up. A CW bias-point check after soak misses that; we log the first second on the traveler when the product can start under RF drive.

2. Probe caveat — do not “fix” the oscillation by measuring it

A 10 pF scope probe on the gate can shift or quench the tank, so the skirt disappears while you probe and comes back when you lift. Use a high-Z / resistive pickup (a 1 kΩ series resistor at the tip, or the vendor’s high-Z / resistive probe) and confirm the skirt with the probe removed after you have identified f_tank. If the skirt only dies while probed, you have a measurement, not a fix.

3. Traveler habit we would keep

Gate-node spectrum under the same pulsed or modulated drive as ACLR, plus a one-line bias network (series R, local C, lead length). If the skirt spacing equals the tank frequency you just measured at the gate, stop opening a die ticket.

Open: on your boards, does the skirt spacing track a bias-lead / ferrite tank in the low-MHz to tens-of-MHz, or do you also see a much higher-frequency sing that only moves when the gate bypass package changes? The first is layout / lead; the second is the SMT loop at the pin — same family, different part on the traveler.

— 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.