How to Accurately Measure Insertion Loss and Return Loss of RF Filters with a VNA

You’ve just finished laying out a new band‑pass filter on your bench, and the simulation says it should lose only 0.8 dB. The real question is: does it really perform that well? In today’s crowded spectrum, a few tenths of a decibel can be the difference between a clean link and a noisy mess. That’s why getting reliable insertion loss (IL) and return loss (RL) numbers from your vector network analyzer (VNA) matters more than ever.

Why the Numbers Matter

When we talk about RF filters, insertion loss tells us how much signal power makes it through the filter, while return loss tells us how much of that signal is reflected back toward the source. High insertion loss eats away at link budget, and poor return loss can cause standing waves, spurious oscillations, and even damage to sensitive front‑ends. In short, accurate measurements let you validate designs, troubleshoot problems, and give your customers confidence that the filter will behave in the field.

Insertion Loss vs Return Loss

Insertion Loss (IL) – Usually expressed as S21 (or S12) in dB. It is the ratio of output power to input power when the filter is in the signal path.
Return Loss (RL) – Expressed as S11 (input port) or S22 (output port) in dB. It is the ratio of reflected power to incident power at a given port. Higher RL (more dB) means less reflection.

Both are measured with the same VNA, but the setup and data‑processing steps differ enough that a careless approach can give you numbers that look good on paper but hide real issues.

Preparing Your VNA

Calibration – The First Step

A good calibration is the foundation of any trustworthy measurement. I still remember my first solo lab session, where I skipped the full 2‑port SOLT (Short‑Open‑Load‑Through) and ended up with a mysterious 2 dB ripple across the band. The lesson? Never trust a VNA that hasn’t been calibrated for the exact frequency range and connector type you’re using.

Choose the right kit – Use the calibration standards that match your test ports (e.g., SMA, N‑type).
Perform a full 2‑port calibration – This removes systematic errors from mismatch, tracking, and isolation.
Save the calibration – Most modern VNAs let you store the state, so you can reload it later without re‑doing the whole routine.

Choosing the Right Port Extensions

If you need to measure a filter that sits inside a fixture or a housing, use precision coaxial extensions or a calibrated test fixture. The key is to keep the electrical length short and the loss low. Longer cables add their own insertion loss and can mask the filter’s true performance. When I first measured a cavity filter inside a metal box, I used a 2‑inch SMA extension and a calibrated thru. The result? A clean 0.9 dB IL that matched simulation within 0.1 dB.

Measuring Insertion Loss

S21 Sweep Basics

Set the VNA to a linear sweep covering the filter’s passband plus a margin of at least 10 % on each side. Use a sufficient number of points (e.g., 801 or 1601) to capture any ripple. For most filters, a resolution bandwidth (RBW) of 10 kHz is a good compromise between speed and noise.

Select the magnitude (dB) format – This makes it easy to read IL directly.
Enable averaging – A factor of 4–8 reduces random noise without sacrificing measurement time.
Check the power level – Keep the source power low enough to avoid compression but high enough for a good signal‑to‑noise ratio. Around –10 dBm is a safe starting point for most passive filters.

De‑embedding the Test Fixture

If you used any adapters, cables, or a test fixture, you must remove their contribution. Most VNAs have a “de‑embed” function where you can load the S‑parameters of the fixture (measured separately) and let the instrument subtract them automatically. Alternatively, you can manually subtract the fixture’s loss in dB:

IL_filter = IL_measured - IL_fixture

Make sure the fixture’s S‑parameters are measured with the same calibration kit and over the same frequency range.

Measuring Return Loss

S11 and S22

Switch the VNA to the reflection measurement mode. For input return loss, display S11; for output return loss, display S22. The same sweep settings used for S21 work here, but you may want a finer frequency step if you’re hunting for narrow resonances.

Enable the “Port Extension” correction if you have a known length of coax before the filter. This aligns the reference plane to the filter’s face.
Set the display to dB – A higher RL (e.g., >20 dB) indicates good matching.

Time Domain Gating for Clean Data

Sometimes the measurement includes unwanted reflections from connectors or the test fixture. The VNA’s time‑domain gating feature lets you isolate the filter’s response. Here’s a quick workflow:

Transform to time domain – Use the built‑in FFT.
Place a gate around the time window that corresponds to the filter’s physical location.
Inverse transform back to frequency domain.

The result is a smoother RL curve that reflects only the filter, not the surrounding hardware. I’ve used this trick on a high‑Q SAW filter where the raw S11 showed a puzzling dip at 2.45 GHz that vanished after gating.

Common Pitfalls and Quick Fixes

Mismatched reference planes – Forgetting to set port extensions can shift the measured IL by up to 1 dB.
Cable drift – Coax cables can change length with temperature. Re‑calibrate if the lab temperature swings more than a few degrees.
Source power too high – Over‑driving a passive filter can cause heating and a false rise in IL. Drop the power and re‑measure.
Insufficient averaging – A noisy trace can look like ripple. Increase the averaging factor or lower the RBW.

Putting It All Together

Calibrate the VNA with a full 2‑port SOLT kit that matches your connectors.
Attach the filter using the shortest, lowest‑loss adapters you have. Record the fixture’s S‑parameters if you plan to de‑embed.
Set up a linear sweep that covers the filter’s band plus margin, with enough points and a modest RBW.
Measure S21, apply de‑embedding, and note the insertion loss at the center frequency and at the band edges.
Switch to S11/S22, enable port extensions, and optionally use time‑domain gating to clean up reflections. Record the return loss across the band.
Document the settings (cal kit, cable lengths, power level, averaging) so you can repeat the test later or hand it off to a colleague.

When you follow these steps, the numbers you write down will be the same ones you see in the simulation, and you’ll have confidence that the filter will behave as expected in the field. In my own work, this disciplined approach has saved countless hours of redesign and helped me convince skeptical customers that the filter really does meet its spec.

Happy measuring, and may your dB curves stay flat!