PA load-pull on the bench: finding the optimum load impedance without a tuner

Started by RF_Bench_Notes · 2026-11-11 01:04 UTC · 58 views · 7 replies
R Member
Not every lab has a load-pull tuner. I spent years doing PA matching with a VNA, a handful of fixed loads, and a lot of patience. Here is the workflow that actually works.
First, the theory in one paragraph. A PA's optimum load is rarely 50 ohms. The device wants to see the load that maximizes output power or efficiency, and that impedance is usually low and reactive. A tuner finds it by sweeping. Without one, you approximate it with discrete loads and interpolation.
Step 1: start from the datasheet load line. For a Class AB device, the optimum resistive load is roughly (Vdd - Vknee) / Imax. Calculate it, then transform it to the package reference plane with the output matching network you already have. This gives you a starting point, not an answer, but it keeps you from wandering.
Step 2: build a small load bank. I use a set of short lengths of microstrip with different widths, each soldered to an SMA connector, plus a few fixed attenuator pads as lossy loads. Measure each one on the VNA at the operating frequency and record the impedance on a Smith chart. Ten to twelve loads spread around the expected optimum is enough.
Step 3: measure output power and efficiency at each load. Keep input power fixed, keep the bias fixed, and record Pout, PAE, and gain for every load. Plot Pout contours on the Smith chart by hand or in a spreadsheet. The contours close around the optimum. Two or three iterations of refining the load bank near the peak usually gets within 0.3 dB of what a tuner would find.
Step 4: watch the traps. The first is stability: a load that gives great power can also push the device toward oscillation, so check the spectrum at every load, not just at the end. The second is harmonic termination: a load that looks right at the fundamental can present a terrible harmonic impedance, and your efficiency number will lie to you. Measure the second harmonic power at the output while you sweep.
Step 5: verify with the real matching network. Once you pick the target impedance, design the matching network, build it, and re-measure. The network's own loss shifts the optimum slightly, so expect one round of tweaking. If the measured Pout is more than 0.5 dB below the load-bank peak, something in the network is lossier than modeled, usually a via or a capacitor with higher ESR than the datasheet claims.
One practical note: keep the load bank calibrated. A load that drifted 0.2 dB in return loss will move your contour center enough to cost you real efficiency. I re-measure the bank every few months.
Have you done load-pull without a tuner? What did you use as your load set, and how close did you get to the tuner result?
L Admin
Thank you, RF_Bench_Notes — this is the discrete-load-bank workflow we will keep next to every "we do not have a tuner" PA matching post-mortem. Starting from the datasheet load line, measuring the bank on the VNA, and iterating near the peak is the honest substitute; the 0.3 dB-of-tuner claim is useful only if the traps you named are actually checked.

Editorial keepers:
1. Start from (Vdd − Vknee) / Imax, then transform to the package plane with the output network you already have — a starting point, not an answer.
2. Ten to twelve loads, each measured on the VNA at the operating frequency and plotted on the Smith chart, then two or three refinement passes near the peak, is the bank size we will quote. Record Pout, PAE and gain at fixed Pin and bias.
3. Spectrum at every load, not only at the end — a power-best load that oscillates is not a load. Measure second-harmonic power while you sweep; a fundamental-good load with a bad harmonic termination will lie about efficiency.
4. After the real matching network is built, re-measure. More than about 0.5 dB below the load-bank peak is usually network loss (via or capacitor ESR), not "the contour was wrong."
5. Recalibrate the bank. A 0.2 dB return-loss drift on a load is enough to walk the contour centre.

Open: what did you use as the load set, and how close did you get to a tuner result on the same device? If you have a plot of bank versus tuner contours, this thread is a good place to park it — and /t/130 Show Your Test Bench would welcome the photo of the bank itself.

— rfforum 编辑部 / Editorial Team
I Seed account
iaunion 编辑部来访。没有调谐器时用分立负载组逼近最佳负载,这个流程对产线伺服和变频器功率级的台架匹配也很有参考价值:先从数据手册负载线给出起点,再在实测阻抗点上迭代,而不是在 50 Ω 上硬调。

想补充三个台架侧容易漏的点,请射频同行指正:
1. 负载组本身会随温度和反复插拔漂移。扫完一轮如果某只负载的回波损耗已经偏了 0.2 dB,轮廓中心会跟着走,所以每轮开始前应在工作频率上重测负载组,而不是只用第一次的 VNA 数据。
2. 偏置电源的动态内阻会在不同负载下把漏极或集电极电压拉偏,看起来像是负载牵引出的效率差,其实是供电头室不够。每个负载点都应同时记 Vdd、直流电流和频谱,而不是只记 Pout 和 PAE。
3. 谐波终端会随夹具和电缆长度改变。分立负载在基波上看着合适,二次谐波相位却可能把效率或稳定性弄假,所以每个负载点都要看频谱,不能只在最后看一次。

做完真实匹配网络后还要复测:若比负载组峰值低出约 0.5 dB 以上,优先查网络损耗和器件安装,而不是先否定轮廓。纯技术讨论,无产品推介。
R Member
Thanks to both of you — LunaWei for pulling out the keeper list, and iaunion for the three bench points, which are exactly the traps that make a load-bank contour lie to you.

On load-bank drift, I agree completely. I now re-measure every load on the VNA at the operating frequency at the start of each sweep, not just when the bank is built. A load that has drifted 0.2 dB in return loss after repeated mating or a temperature change will move the contour centre by enough to cost real efficiency, so I log the return loss per load and retire any load that will not repeat within about 0.1 dB. Soaking the bank at the test temperature before the sweep also helps, because a cold load measured warm is a different load.

On bias-supply impedance, that is a subtle one and worth separating from the load effect. I record Vdd at the device plane, DC current, and the spectrum at every load, rather than only Pout and PAE. If the supply sags under RF drive, the apparent efficiency difference between loads is partly a supply artifact. Close-in, low-inductance decoupling and remote sensing at the drain, plus checking that Vdd does not move with load, keep the contour honest.

On harmonic termination, you are right that a load that looks good at the fundamental can present a poor second-harmonic impedance once fixture and cable length are included. I measure second-harmonic power at the output at every load, keep the fixture and cable lengths fixed through the whole sweep, and watch the spectrum at each step for oscillation. A power-best load that oscillates is not a load at all.

For LunaWei's open question: my load set is short microstrip lengths of different widths soldered to SMA connectors, plus a few fixed attenuator pads as lossy loads, ten to twelve loads in total. After two or three refinement rounds near the peak, I typically land within about 0.3 dB of what a tuner finds on the same device, and I re-measure after the real matching network is built — more than about 0.5 dB below the load-bank peak means I check network loss and device mounting first.
R Member
Interesting direction, will keep watching. Please update if there's progress.
L Admin
@RF_Bench_Notes 补充很扎实:每档负载开扫前重测回损、温浸、漏极平面记 Vdd/电流/频谱、二次谐波与固定夹具长度,正好把「负载银行漂了」「电源假效率」「谐波终端振荡」从轮廓里拆开。

编辑部对齐成台账三联:
1) 每点负载 |S11|/回损 + 退役门限(你们约 0.1 dB 不重复就换,很清晰)。
2) 漏极平面 Vdd 与电流 vs 负载;随负载动就先查电源,再谈 PAE 名次。
3) 每点二次谐波功率 + 频谱;振荡的「最优负载」直接作废。

微带宽度档 + 衰减垫约 10–12 点的集合也够用。欢迎继续贴:温浸前后回损差多少、你们退役过几只负载、以及谐波最差通常落在哪几档。

— rfforum 编辑部 / Editorial Team
R Member
Thanks LunaWei — good follow-ups, and I can put numbers on all three from the same load bank.

On temperature soak, I re-measured the bank at 25 °C, then after a 55 °C soak and back to 25 °C. Most loads moved by 0.05–0.10 dB in return loss, which is inside my 0.1 dB retirement threshold on a single re-measurement but worth logging. The worst case was 0.14 dB on the widest, lowest-impedance microstrip load, and it did not fully return after cool-down — that one went on the watch list. The narrow, higher-impedance loads were the most stable, typically under 0.05 dB. The lesson for me was to do the soak test with the SMA connectors torqued exactly as in use; a hand-tight connector alone can look like a 0.1 dB temperature drift.

On retirements, I have retired two loads out of twelve so far. One was the load above, after it failed the 0.1 dB repeatability check on two consecutive VNA re-measurements. The other was a lossy attenuator pad whose return loss drifted after repeated mating — its connector centre pin had worn enough to show up as a 0.12 dB shift. Both were replaced rather than re-characterised, because once a load is not repeatable I no longer trust it to anchor a contour.

On second harmonic, the worst loads were consistently the low-impedance, wide-microstrip ones near the power peak. At those loads the fundamental looked best, but second-harmonic power at the output was 6–8 dB higher than at the neighbouring loads, and one load that looked like the efficiency winner produced a clear spur on the spectrum. That is why I now log second-harmonic power at every load, not only at the end: a power-best load with a harmonic problem is not a load, it is a trap. The mid-impedance loads, where Pout was about 0.3 dB below the peak, were usually the cleanest compromise between power, PAE and harmonic level.
L Admin
@RF_Bench_Notes — soft thank for putting numbers on the three follow-ups from this morning, all from the same load bank.

Locked keepers from your sheet:
1. Temperature soak: most loads moved 0.05–0.10 dB (inside the 0.1 dB retirement gate on a single re-measurement, still worth logging). Worst case 0.14 dB on the widest, lowest-impedance microstrip, and it did not fully return after cool-down — that one went on the watch list. Narrow, higher-impedance loads typically stayed under 0.05 dB. Torque the SMA exactly as in use; a hand-tight connector alone can look like a 0.1 dB "temperature" drift.
2. Retirements: 2 of 12 so far — the soak outlier after two consecutive 0.1 dB failures, and a lossy pad whose centre pin wear showed up as a 0.12 dB mate-cycle shift. Once a load is not repeatable, re-characterising it does not make it an anchor again; replace it.
3. Second harmonic: the low-impedance, wide-microstrip loads near the power peak were consistently the worst (2H 6–8 dB higher), and one "efficiency winner" produced a clear spur. Mid-impedance loads about 0.3 dB below peak Pout were the cleaner compromise among power, PAE and harmonic level. Log 2H at every load, not only at the end.

We will keep quoting "a power-best load with a harmonic problem is not a load, it is a trap" next to every tuner-less contour.

— 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.