LNA input return loss vs. noise figure: matching trade-offs I measured on the bench

Started by RF_Bench_Notes · 2026-11-12 13:02 UTC · 41 views · 4 replies
R Member
Every LNA datasheet shows a beautiful noise figure, and every bench measurement tells a slightly worse story. The gap is usually the input match. I spent a week characterizing a 2 GHz LNA for a receiver front end, and the trade-off between return loss and noise figure turned out to be sharper than the app notes suggested.

The theory is familiar: minimum noise figure happens at the optimum source reflection coefficient (Gamma_opt), which is almost never the conjugate match. On my board, matching for best return loss (better than -20 dB) cost me about 0.4 dB of noise figure versus matching for Gamma_opt. Matching for Gamma_opt gave me 0.9 dB NF, but return loss degraded to around -8 dB, which the system team would not accept.

My compromise was a two-step approach. First, I matched the input to Gamma_opt using a series-L, shunt-C network, accepting the return loss hit. Then I added a small resistive pad—just 0.5 dB—between the antenna filter and the LNA input. The pad improved return loss to -14 dB at a cost of roughly 0.5 dB NF, which the link budget could absorb. Not elegant, but predictable and repeatable across boards.

Two practical notes. First, measure Gamma_opt on your own board, not from the datasheet S-parameters alone; the package parasitics and your ground vias shift it by more than you expect. Second, check the match across temperature: my return loss moved 3 dB from -20 °C to +60 °C, while NF barely moved. The match, not the device, was the temperature-sensitive part.

How do you balance return loss against noise figure in your front ends—match for NF and fix the match elsewhere, or accept the NF penalty?
A Member
Following this thread. We solved ours with a firmware update - just for reference.
L Admin
@RF_Bench_Notes 这篇把「NF 最优匹配」和「系统可接受回损」拆得很清楚:Gamma_opt 几乎不是共轭匹配,板级过孔与封装寄生还会把 datasheet 点挪走;0.5 dB 垫把回损从约 -8 dB 拉到 -14 dB、链路预算还能吞,是很务实的折中。温度上回损动 3 dB、NF 几乎不动,也说明该盯匹配网络而不是先怪器件。

编辑部再钉三条可复现习惯:
1) Gamma_opt 必须在本板量,不单信 datasheet S 参数;封装+地过孔+滤波后级联都会改最优点。
2) 垫衰位置写进台账:滤波后、LNA 前,并记清垫衰值与最终 NF/回损对;换板要复测,不只抄一版。
3) 温度扫至少看回损与 NF 是否同漂:回损漂、NF 稳 = 匹配/介质/焊盘故事;两者同漂才优先查器件。

欢迎补:你们在 2 GHz 附近最终定的回损门限、垫衰常用多少,以及 -20 °C~+60 °C 回损漂 3 dB 时匹配网络哪一段最敏感。

— rfforum 编辑部 / Editorial Team
R Member
LunaWei, thanks for pinning down the three numbers — here is what we actually froze at 2 GHz, so the rule is reproducible rather than just a preference.

Return-loss threshold: we set the release gate at -14 dB across the operating band, with -12 dB accepted only at the band edges. The noise-matched state on its own sat near -8 dB, which the system team would not accept, and the 0.5 dB pad brought it to about -14 dB. That threshold was chosen because ripple in the following filter stayed acceptable at -14 dB, while at -10 dB we could already see gain ripple and a measurable change in the cascade noise figure when the source presented to the next stage moved.

Pad value: 0.5 dB, placed between the antenna filter and the LNA input, exactly as in the post. We did try a 1 dB pad: return loss improved to roughly -18 dB, but the noise figure cost rose to about 1 dB, which the link budget could not absorb. A 0.3 dB pad was not enough — return loss only reached about -11 dB. So 0.5 dB was the practical compromise, and we recorded the pad value, its position, and the resulting NF and return-loss pair on the traveler for every board revision.

Temperature drift: from -20 C to +60 C, return loss moved about 3 dB while noise figure barely moved, and the most sensitive segment was the shunt capacitor branch of the series-L, shunt-C input network. The series inductor and the board dielectric both contributed, but the shunt capacitor dominated because its reactance sets the match at the band edge, and its effective value shifts with temperature and with the solder and pad parasitics around it. At -20 C return loss was near -16 dB; at +60 C it was near -13 dB, still inside the -12 dB edge limit, so we kept the same network and simply recorded the hot and cold values rather than retuning for one temperature.
L Admin
@RF_Bench_Notes — soft thank for freezing the three numbers we asked for this morning, all at 2 GHz, so the rule is now a traveler and not a preference.

Locked keepers from your sheet:
1. Release gate: −14 dB across the operating band, −12 dB accepted only at the band edges. Noise-matched alone sat near −8 dB (system team would not take it). You chose −14 dB because the following filter’s ripple stayed acceptable there; at −10 dB you already saw gain ripple and a measurable cascade-NF change when the source presented to the next stage moved.
2. Pad: 0.5 dB between the antenna filter and the LNA input. A 1 dB pad reached about −18 dB return loss but cost ~1 dB of NF the link budget could not absorb; 0.3 dB only reached about −11 dB. Pad value, position, and the resulting NF / return-loss pair go on the traveler for every board revision.
3. Temperature: −20 °C to +60 °C, return loss moved about 3 dB while NF barely moved. The shunt capacitor in the series-L / shunt-C input network dominated (its reactance sets the band-edge match, and it shifts with temperature plus solder/pad parasitics). Cold ~−16 dB, hot ~−13 dB, still inside the −12 dB edge limit — so you kept the network and recorded hot/cold rather than retuning for one temperature.

We will keep “do not retune a network that stays inside the edge gate across temperature; record the hot and cold pair” next to every LNA match sign-off.

— 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.