VNA error caused by sweeping too fast through a device with long electrical delay.

Sometimes when measuring S21 of a device like a long cable or SAW filter, the data definitely looks wrong. The magnitude of the response looks too low, and there may even be small jumps in it. Then when you slow down the VNA's sweep time, the data looks correct. Apparently, the VNA sweep rate is too fast for this measurement, but why? What is causing the bad data at fast sweep rates?

The delay of the DUT (device under test) causes a shift in the frequency of its output signal relative to that of its input signal, because the frequency is sweeping with time. This frequency shift is determined by the VNA's sweep rate and the DUT's electrical delay:

The DUT's output signal goes into the VNA receiver, which is tuned to the frequency of the DUT's input signal, and therefore the output signal frequency is "mis-tuned" from the receiver by Fshift. Consequently, the receiver IF signal is not centered in the IF filter passband, and it has some attenuation due to the filter's rolloff. The faster the sweep rate, the greater the frequency shift, and the more the S21 measurement drops off. There can also be jumps in the data if the frequency sweeps across more than one VNA frequency band, and the VNA has to change its sweep rate in the different bands.

Now what can be done if the data looks bad? Well, obviously you can slow down the VNA's sweep rate. In fact, it might be a good idea to forego sweeping and select a "stepped frequency" mode if it is available, and eliminate the frequency shift problem all together. An alternative may be to add a compensating delay in the VNA's reference channel. If the VNA offers access to an R channel jumper, then it may be possible to add enough cable to the R channel path to approximately match the delay of the DUT. This cancels the effect of the frequency shift when sweeping, and allows you to maintain the fast speed of the swept measurement.