Hello,

I applied de-embedding function in PNA to different traces to see its effects.

For low loss and well matched structres, the de-embedding worked pretty well. However, the de-embedding had issues with high loss (>10dB) or poor match strutures.

Can you explain why this is happening?

Is this due to measurement unceratinties or de-embedding function?

Thanks.

I applied de-embedding function in PNA to different traces to see its effects.

For low loss and well matched structres, the de-embedding worked pretty well. However, the de-embedding had issues with high loss (>10dB) or poor match strutures.

Can you explain why this is happening?

Is this due to measurement unceratinties or de-embedding function?

Thanks.

When you are de-embedding a high loss device, the error correction arrays are modified by the values of the depembedding network. If the loss is high, the directivity error term is degraded by 2 x loss. A very common problem among users is using test port cables that are not stable in return loss.

The situation is often this: VNA -> cable -> loss -> DUT

If the cable has a return loss of stability of 33 dB, for example, and the loss is 10 dB, then the effective stabiliyt of the RL of the cable at the DUT interface is only 13 dB. So, measuring any load of less than 13 dB return loss would be completely masked by the cable instability. This has nothing to do with the de-embedding math or the VNA, just the cable connection.

Another common problem is that the S-parameters of the loss element are not carefully measured, and if it has a high reflection and there is an error in the reflection measurement (due to instable cables, not careful cal, or any other reason) then the de-embedd function removes the wrong (very large) number, so the residual is very high. It is the problem of subtracting two large numbers, so you can have an issue of very small errors (such as phase errors) causing large errors in the difference.

My experience is that if your loss is greater than 10 dB, you must have very stable cables, more stable than 50 dB. I don't think it is possible to de-embedd more than 20 dB, as the RL signal (raw) is degraded by 2x the loss, so there is nothing but noise in the reflection measurement.

If you are serious about de-embedding high loss you must use narrow IFBW (100) and very careful techniques