When to use the Isolation Calibration in a two-port cal.

In a two-port calibration, you have the option to omit the isolation cal, so how do you decide when this is important? The isolation cal removes the errors due to crosstalk between test ports in a transmission measurement (S21 or S12). On most network analyzers, this crosstalk is in the range of -100 to -120dB. So you can see that the error would only be significant if your device under test has a very high insertion loss, such as the stopband of a real good filter. If you are measuring an amplifier or a filter with -60dB stopband, you can omit the isolation cal. If you need the maximum dynamic range of the network analyzer, use the isolation cal.

When you perform the isolation cal, it is important that you use averaging to reduce the noise in the calibration measurement. Without it, you will have a noisy measurement of the crosstalk, and when the network analyzer uses this in its error correction, it could end up raising the noise floor by 3dB (not what you wanted!). Set the averaging factor at least four times higher than what you will use for measuring your DUT, so you can get a clean measurement of the crosstalk for your cal.

In some network analyzers, the isolation cal doesn't actually remove all of the crosstalk in the system. This is because the system error model only accounts for crosstalk between the test ports, and there are other crosstalk paths inside the box. In particular, there may be internal crosstalk from the port1 reflected signal to the port2 transmitted path, known as "A to B" crosstalk because of the internal receiver channel names. When you perform the isolation cal, you will put loads on both ports, so the reflected signals are very small and this crosstalk is not present. But when you measure your filter, it probably has a high reflection in its stopband, so you get the A to B crosstalk error that was not present during the isolation cal, and therefore is not corrected for.

Hence, the ultimate isolation cal would go like this. Turn on a lot of averaging. Connect the port1 of the filter you will be measuring to port1 of the network analyzer and loads on port2 of both the filter and network analyzer, and make the "Forward Isolation" measurement. Then connect filter port2 to network analyzer port2 and loads on both port1's, and make the "Reverse Isolation" measurement. Turn off averaging, and finish the rest of the two-port cal measurements. Unfortunately, this isolation cal would only be good for that particular filter! The next one may have a different phase shift in S11 and S22, so the correction of the A to B crosstalk is no longer working.

Check for A to B crosstalk on your network analyzer by doing the usual isolation cal with loads on both ports (remember the averaging). Then measure S21 and S12 with shorts or opens on the ports, and see if you see any crosstalk sticking up out of the noise floor. If you do, it is a potential limitation in your measurements that need high dynamic range.