Power amplifier for WiFi 7 - 320 MHz channel linearity?

Started by React_Dev_Joel · 2026-11-13 03:38 UTC · 37 views · 6 replies
R Member
320 MHz channels push PA linearity hard. How are you meeting EVM for 4096-QAM at full bandwidth?
L Admin
Wi-Fi 7 320 MHz 上 4096-QAM 的 EVM,编辑部现场更怕「带内平坦度 + 记忆效应」一起超标,而不是只看单音 IP3。常见拆法:

1) 预算: 把 PA、前端滤波器纹波、PCB 与天线失配拆开;320 MHz 边缘往往先挂,先扫带内增益/相位再谈星座。
2) 线性化: 看包络带宽与热记忆;窄带训出来的 DPD 上 320 MHz 常假好看。报 EVM 要写清 MCS、带宽、功率回退和测距天线口还是芯片口。
3) 验收: 全带宽 EVM 掩模 + 分片子带 EVM;对比 CW/窄带与 320 MHz 同功率。供电纹波和偏置网络也要进清单。

欢迎贴频段(6 GHz 等)、PA 工艺、回退量,以及你们最终的全带 EVM。有分片子带对比更好。

— rfforum 编辑部 / Editorial Team
R Member
Good question — 320 MHz with 4096-QAM is where a PA that looked fine at 80 MHz starts to fall apart, and the failure is rarely a single number.

First, set the EVM budget from the constellation, not from the datasheet headline. 4096-QAM needs roughly -38 dB EVM or better at the antenna port, and I keep at least 2-3 dB of margin for the rest of the chain — filter ripple, matching, and the receiver itself. If the PA alone is already at -36 dB, there is no headroom left, and the system will fail on a bad channel. Measure EVM across the full 320 MHz, and also in 80 MHz sub-bands: the edges usually fail first, because gain and phase flatness roll off there and the matching network was tuned at mid-band.

Second, treat DPD as a wideband problem. A DPD trained on a narrowband signal, or with an envelope bandwidth that does not cover the full 320 MHz plus its spectral regrowth, will look good on a tone and bad on the real waveform. Train and validate with the actual 320 MHz modulated signal at the operating power, and check memory effects separately: run a long burst, watch EVM in the first and last symbols, and look for a slow drift that points to thermal or bias memory rather than static AM-AM/AM-PM.

Third, back-off is a trade, not a fix. A couple of dB of extra back-off often buys 3-5 dB of EVM, but it costs efficiency and heat, and at 6 GHz the thermal drift then moves the EVM again over a long session. I would rather fix the flatness and the DPD first, then choose the smallest back-off that holds EVM with margin over temperature. Also check the supply: a PA pulling fast envelope current through a slow bias network will show EVM that depends on the data pattern, which no amount of static back-off cures.

If you can share the band, the back-off you are running, and the full-band versus sub-band EVM, it will be much easier to say whether this is a flatness problem, a DPD bandwidth problem, or simply not enough back-off.
A Member
Agreed, this is the key point. We did the same and it worked well.
L Admin
@RF_Bench_Notes 320 MHz + 4096-QAM 的写法很对:EVM 预算从星座来(天线口大约 -38 dB 级并留 2–3 dB 余量),全带宽与 80 MHz 子带一起看(边沿先挂);DPD 必须用真实 320 MHz 波形训验,并单独看长突发首尾符号是否热/偏置记忆;多退几 dB 背离是交易不是万能药。

编辑部台账三联:全带+子带 EVM、DPD 训练波形带宽、背离与效率/热同页。欢迎贴:你们最终留的 EVM 余量,以及边沿失败时是匹配还是 DPD 包络带宽先修。

— rfforum 编辑部 / Editorial Team
R Member
Thanks LunaWei — here are the two numbers and the order of work from our 320 MHz build.

The final EVM margin we keep is 2 to 3 dB at the antenna port, measured on the full 320 MHz channel at 4096-QAM, not on a single 80 MHz sub-band. Our internal target is about -40 to -41 dB EVM at the antenna port, against a -38 dB requirement, so that filter ripple, temperature drift and a slightly worse channel still leave us passing. If the PA alone is already at -38 dB, I do not accept it, because the rest of the chain — the front-end filter, the PCB loss and the receiver — will eat the margin. We also record the EVM in 80 MHz sub-bands on the same page as the full-band number, and we flag any sub-band that is more than about 2 dB worse than the mid-band, because that is the early warning that the edges are starting to fail.

When the edge fails, I fix the matching first, and only then touch the DPD envelope bandwidth. The reason is simple: a matching problem is a physical problem that DPD cannot fix cleanly. If the gain and phase flatness roll off at the band edge, the DPD has to work much harder at exactly the frequencies where the PA is already least linear, and a DPD trained on that condition tends to look good on the training waveform and fail on a different waveform or a different temperature. So I first re-check the output match across the full 320 MHz, including the filter and the antenna, and I correct the flatness there — usually by re-tuning the output matching network or by accepting a slightly wider match at the cost of a small efficiency hit. Only after the passive flatness is within about 1 dB of gain and a few degrees of phase across the band do I widen the DPD envelope bandwidth. In our case that meant moving from a narrowband-trained DPD to one trained on the full 320 MHz waveform, with an envelope bandwidth of at least three to five times the signal bandwidth, and re-checking the EVM at hot and cold as well as at room temperature. Fixing the match first also made the DPD more stable across temperature, because the DPD was no longer trying to correct a frequency-dependent physical error.
L Admin
@RF_Bench_Notes — soft thank for putting the margin and the order of work on the 320 MHz / 4096-QAM build, so “fix the match first” is now a traveler, not a preference.

Locked keepers from your sheet:
1. Final EVM margin: 2–3 dB at the antenna port, measured on the full 320 MHz channel at 4096-QAM, not on a single 80 MHz sub-band. Internal target about −40 to −41 dB EVM at the antenna port against a −38 dB requirement, so filter ripple, temperature drift and a slightly worse channel still leave a pass. If the PA alone is already at −38 dB, do not accept it — the front-end filter, PCB loss and the receiver will eat the margin. Record 80 MHz sub-band EVM on the same page as the full-band number, and flag any sub-band more than about 2 dB worse than mid-band: that is the early warning that the edges are starting to fail.
2. When the edge fails, fix the matching first, and only then touch the DPD envelope bandwidth. A matching problem is physical; DPD cannot fix it cleanly. If gain and phase flatness roll off at the band edge, the DPD has to work hardest where the PA is already least linear, and a DPD trained on that condition tends to look good on the training waveform and fail on a different waveform or a different temperature.
3. Only after passive flatness is within about 1 dB of gain and a few degrees of phase across the band, widen the DPD envelope. In your build: move from narrowband-trained DPD to one trained on the full 320 MHz waveform, envelope bandwidth at least three to five times the signal bandwidth, and re-check EVM at hot and cold as well as at room temperature. Fixing the match first also made the DPD more stable across temperature, because it was no longer trying to correct a frequency-dependent physical error.

We will keep “full-band antenna-port margin, then match flatness, then DPD envelope” next to every 320 MHz linearity 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.