Measuring Filter Response: Cutoff, Bandwidth, Q, Phase, and Loading

What This Measurement Does and Does Not Design

The Practical Filter Design Guide chooses a topology and nominal targets; the Passive Filter Nonidealities Guide models ESR, DCR, Q, and source/load effects. This page begins after a prototype exists. Its deliverable is a reproducible measurement of the assembled transfer function at specified source, load, supply, signal level, and temperature. It does not derive new component values or claim that a measured room-temperature sweep proves production tolerance or stability.

Measure the Actual Transfer Ratio

A 50 Ω generator driving a non-50 Ω filter may not deliver the voltage shown on its front panel. Measure the voltage at the input terminals. Likewise, a 10× passive probe can add enough capacitance to shift a high-impedance pole. Use a buffer or high-impedance probe when justified, but that buffer has its own bandwidth and output impedance. Keep the final test load connected during measurement. If the application uses a different load, the measurement describes a different filter.

Use a dual-channel scope with frequency-response capability for audio to moderate frequencies or a calibrated network analyzer where its frequency range and port impedances fit the DUT. A VNA normally expects defined port impedances; forcing a high-impedance active stage into a 50 Ω environment can change its gain or overload it. Baseline cables and fixtures before attributing a slope to the filter. For active filters, reduce stimulus until the input and output remain sinusoidal and proportional to level.

Extract Cutoff and Band-Pass Q Consistently

For a band-pass with fL = 9.5 kHz and fH = 10.5 kHz, the measured bandwidth is 1.0 kHz, geometric center is about 9.987 kHz, and loaded Q is about 9.987. The Filter Q Factor & Bandwidth Calculator converts these measured edges into the same declared convention. It does not calculate the component's intrinsic, unloaded Q. If peak gain changes when the source or load changes, measure and report Q for each configuration separately.

For a Butterworth low-pass, the complete response is -3.0103 dB at its specified cutoff relative to the passband level. Do not demand that every individual second-order stage is -3 dB at the cascade cutoff. For a peaking or high-Q network, the -3 dB point relative to the peak is not necessarily the same as a passband-edge definition. State which reference you used before comparing measurements to a model.

Choose Sweep Resolution From the Expected Bandwidth

A log-spaced overview is good at finding unexpected response regions but can miss a narrow peak or notch. After locating the band, run a fine sweep with multiple points across each expected -3 dB edge and near the extremum. Record the raw points; do not present an interpolated peak as an exact measurement when it lies between sparse samples. Check instrument settling time at each point, especially for a high-Q filter. Repeat the sweep in both directions when drift, auto-ranging, or active-stage heating may be significant.

At deep attenuation, the output approaches the analyzer noise floor and feedthrough between cables may set the measured minimum. A 60 dB notch relative to a 1 V input implies roughly 1 mV output, but 60 dB relative to a 10 mV input implies only 10 µV. Check stimulus dynamic range, probe pickup, and channel isolation. If the minimum is below the dependable measurement floor, report “at least” the demonstrated rejection, not a fictional exact depth. This caveat is especially important for the Twin-T Notch Measurement Guide.

Read Phase Without Inventing a Stability Claim

Magnitude and phase are measured together because loading, resonances, and active stages can change both. Compare phase at the same physical input/output reference planes. A phase trace wrapped at ±180° should be unwrapped before estimating its derivative. Group delay is -dφ/dω for a consistently unwrapped phase response, but differentiation amplifies phase noise and requires adequate sweep density. Do not call a standalone filter's phase response a control-loop phase margin: that requires a defined loop gain and measurement injection point. An active filter can have acceptable small-signal transfer yet clip, slew-limit, or oscillate at another amplitude or load.

A Reproducible Bench Sequence

Save the sweep trace, marker frequencies, raw Vin/Vout magnitudes, phase, instrument settings, DUT revision, fitted component values, and environmental conditions. Use the Multiple-Feedback Band-Pass Calculator or Butterworth Filter Calculator to compare measured response against the intended topology, not to replace the measurement. A discrepancy can indicate loading, component tolerance, op-amp limits, layout coupling, or a mistaken reference level.

Common Mistakes

• Using generator set voltage as Vin without measuring at the filter input.
• Comparing a 50 Ω analyzer result with a high-impedance application load.
• Taking -3 dB relative to 0 dB when the passband has insertion loss.
• Calling geometric center the universal peak frequency of every loaded topology.
• Reporting loaded Q as an individual inductor or capacitor Q.
• Using a coarse sweep that misses a narrow peak or notch.
• Claiming exact notch depth below the noise and feedthrough floor.
• Increasing stimulus until an active filter clips or slews.
• Treating wrapped phase jumps as physical group-delay spikes.
• Comparing measurements at different temperature, supply, or load conditions.

Measurement Sources

Tektronix frequency-response analysis guidance describes swept input/output magnitude and phase measurements. Keysight network-analyzer filter measurement guidance covers swept RF filter responses. Use instrument-specific calibration and safety instructions for your setup.