L
Admin
Group delay is the negative derivative of phase with respect to frequency. On a VNA it answers a practical question: how much does the envelope of a narrowband signal get delayed through the DUT, and how flat is that delay across your channel.
Why engineers look at it
- Filters and cables with ripply phase create envelope distortion even when |S21| looks fine.
- In pulsed radar and digital radios, group-delay variation across the occupied bandwidth shows up as EVM or range sidelobes.
- Comparing two paths (for example main and diversity) needs matched delay, not only matched magnitude.
How to set the measurement
1. Calibrate with enough points that the phase unwrap is trustworthy. Sparse frequency grids make the derivative noisy.
2. Use a smoothing aperture that matches your channel bandwidth. Too narrow and the trace is noise; too wide and you average away real ripples.
3. Prefer a thru or TRL cal when absolute delay matters. SOLT with a poor definition of the thru standard shifts the absolute number even if relative ripples look right.
Three mistakes I see often
1. Reading absolute group delay from an uncorrected fixture. The coax and adapters dominate. Report relative variation across band, or de-embed first.
2. Ignoring wrap artifacts. If phase unwrap fails at a resonance, group delay spikes that are not physical. Zoom the span and raise the point count before trusting spikes.
3. Comparing DUTs at different temperatures or cable dress. A few degrees or a re-seated SMA can move absolute delay more than the DUT difference you care about.
Quick check before you trust the plot: terminate the DUT output (for a two-port) or replace the DUT with a known airline / delay line and confirm the VNA reports a delay close to the electrical length. If that fails, fix cal and fixtures before debugging the DUT.
What aperture / point density do you use for FR1 channel filters versus narrow crystal filters?
Why engineers look at it
- Filters and cables with ripply phase create envelope distortion even when |S21| looks fine.
- In pulsed radar and digital radios, group-delay variation across the occupied bandwidth shows up as EVM or range sidelobes.
- Comparing two paths (for example main and diversity) needs matched delay, not only matched magnitude.
How to set the measurement
1. Calibrate with enough points that the phase unwrap is trustworthy. Sparse frequency grids make the derivative noisy.
2. Use a smoothing aperture that matches your channel bandwidth. Too narrow and the trace is noise; too wide and you average away real ripples.
3. Prefer a thru or TRL cal when absolute delay matters. SOLT with a poor definition of the thru standard shifts the absolute number even if relative ripples look right.
Three mistakes I see often
1. Reading absolute group delay from an uncorrected fixture. The coax and adapters dominate. Report relative variation across band, or de-embed first.
2. Ignoring wrap artifacts. If phase unwrap fails at a resonance, group delay spikes that are not physical. Zoom the span and raise the point count before trusting spikes.
3. Comparing DUTs at different temperatures or cable dress. A few degrees or a re-seated SMA can move absolute delay more than the DUT difference you care about.
Quick check before you trust the plot: terminate the DUT output (for a two-port) or replace the DUT with a known airline / delay line and confirm the VNA reports a delay close to the electrical length. If that fails, fix cal and fixtures before debugging the DUT.
What aperture / point density do you use for FR1 channel filters versus narrow crystal filters?