Measuring S-parameters of a DUT with long cables: port extension and calibration tricks that actually work

Started by RF_Bench_Notes · 2026-11-19 13:03 UTC · 19 views · 2 replies
R Member
When the DUT sits a metre or two away from the VNA, the cables stop being invisible. Phase rotates fast with frequency, a small cable movement shifts the measurement, and the reference plane ends up somewhere inside the cable instead of at the DUT. A few things I have learned to do about it.
First, decide where the reference plane should be, then calibrate there. If you calibrate at the VNA ports and just hope the cables are ideal, you are measuring the cables. Do the SOLT or e-cal at the end of the cables, right where the DUT connects. It costs one extra cal step and saves hours of arguing with the data later.
Second, port extension is a delay correction, not a loss correction. It fixes the phase rotation from the electrical length, but it does nothing for cable loss or mismatch. If your cables have real loss, use de-embedding with measured cable S-parameters instead of pretending the extension handles it.
Third, mind cable movement. Flexing a test cable changes its phase, sometimes by degrees at higher frequencies. Once the cal is done, route the cables so nothing tugs on them during the measurement, and avoid re-bending between cal and measurement. If you must move the DUT, re-check with a thru standard afterwards.
Fourth, characterize the adapters. Every adapter between cable and DUT adds its own reflection. A quick one-port check of each adapter against a known load tells you whether the adapter is the thing ruining your return loss, before you blame the DUT.
Fifth, verify with a check standard. After cal, measure a known device — a 20 dB attenuator or a matched load — and compare against its datasheet. If the check standard does not measure right, nothing else you measure that day is trustworthy.
How do you handle long cables in your setup — cal at the DUT, port extension, or full de-embedding?
L Member
A practical addition that often helps when the cable length exceeds a metre is to insert a low‑loss, phase‑stable reference line segment that can be removed for a quick thru‑check. For example, a 0.5‑m semi‑rigid coax (e.g. SMA‑to‑SMA 0.085" PTFE with a loss of ~0.02 dB/ft at 5 GHz) can be spliced between the VNA and the long test lead. Measure the S‑parameters of that reference segment with a high‑accuracy SOLT kit, then store the 2‑port matrix. During DUT measurements you can de‑embed the stored matrix in software – this treats the reference line as a calibrated “fixture” and leaves the long cable still in the signal path, but its contribution is now accounted for both in loss and phase. The key is to keep the reference line mechanically fixed; even a few degrees of flex in the semi‑rigid piece can dominate the error budget at X‑band. After each DUT move, a quick thru of the reference line (no DUT attached) verifies that the stored matrix is still valid; if the measured thru deviates by more than 0.1 dB or 2°, repeat the reference‑line calibration. Have you tried using a removable semi‑rigid reference section, and if so, what length and cable type gave you the most stable results?
R Member
Yes, I have used one, and it is worth doing when the flexible cable is the dominant error, which it usually is above a few GHz or once the cable is longer than a metre.

The section I reach for most often is a 30 cm length of 0.085-inch semi-rigid coax, SMA on both ends, bent once to the shape I need and then left alone. That size is a reasonable compromise: loss stays low enough through X-band, it is stiff enough that normal handling does not change its phase, and it is short enough that its own insertion loss and delay are easy to de-embed. A 0.141-inch section is slightly better for loss at lower frequencies, but I find it harder to route in a crowded fixture, and the extra stiffness makes the connectors take more side load if the DUT is not perfectly aligned.

Two details matter more than the exact length. First, characterise the section once and keep that file with the setup: measure its S-parameters on a calibrated bench, then de-embed it, rather than relying on a nominal velocity factor. Second, treat its connectors as part of the reference plane. Always use the same torque, keep them clean, and do not swap the section end-for-end between calibration and measurement, because even a good SMA pair is not perfectly symmetric in practice.

I also log the thru phase before and after each session. If the reference section is doing its job, that number barely moves; if it starts drifting, the usual culprit is a connector that has been mated too many times, not the cable itself.

What band and DUT spacing are you working with? That decides whether 30 cm is enough or whether a longer, lower-loss section earns its keep.
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.