I need to measure a differential lines.
At first i got low causality about 50% so i played with the connector screwed unscrewed tried wipe of some spots i saw on the ground pad.
Then i saw that the causality got to almost zero and i cant improve it back to the values before.
I tried to changes cables to shorter ones.
I dont know if my 1mm to 1.85mm screwed proprly to the PNA.
I dont know how to place propectry the SMA and how much presure to put on the screws so it will work properly.
Any help?
photos attached.
Thanks.

The resolution of TDR was limited by the rise time of the incident step function, and the internal multiple reflections within the network.

Given a frequency sweep of a cable, done with a network analyser, it is possible to compute from S11, the impedance of the network, at different ranges, or times, into the network. The result is limited by the frequency band swept by the analyser, and the resolution of the frequency steps taken by that analyser, within that band.

That computation starts from the known impedance of the adapter cable between the analyser and the network under test. That adapter cable can be computationally removed. When the TDR is computed, an estimate of "causality" shows how reliable the computed TDR of the network can be.
https://www.signalintegrityjournal....ling-and-bandwidth-shape-time-domain-fidelity

I imagine that the problem is there because of practical problems generating and sampling a direct time domain response graph. Those problems are mirrored in the limited bandwidth of the 'sample' frequency steps / samples. So the bottom line will be the resulting SNR and Error Rate in the data.

I haven't understood why, because causality must always apply in reality, an imperfect simulation cannot predict the overall delay of the first sample through the system. If multiple signal paths are involved then the delays must be equalised - to an 'acceptable' degree. Perhaps this 'acceptable' criterion will depend on the actual shape of data pulses and even the pattern of the data. I looked at the reference and it seems that they use the term 'causality' to calculate the errors - hence their comparison between the causal and non-causal treatment.

All this is reminiscent of analogue colour TV circuit design where the image on the screen is basically a time varying electron beam current which (for a stationary picture) appears as distance variation of phosphor brightness which poses demands in both frequency response and time response (edges and patterns). Simulating that in the old days was demanding and test signals had to be pretty ingeniously designed.

There are many versions of RF connectors which are cheaper than other versions and they have inferior specs, even though they have the same mechanical dimensions and will fit high quality cables.

@yefj Your Analyser can give you S parameters for frequencies which may be outside of the range you are using. If the solderless connection continues to give trouble then consider a soldered connector and also pay attention to the circuit board tracks to ensure a good continuous impedance from cable to the board. Good Engineering requires identify the relevant problems and solve them, one at a time, in order of importance. This often needs experience.

If you use the wrong connector you may deform the mate, so it will not contact correctly again without correction or replacement. As an example: If you use a 50Ω BNC male to connect to a 75Ω BNC socket, it will seem to work with only a 50:75 impedance mismatch, but the socket may be physically spread, sufficiently to make unreliable contact with a 75Ω BNC male in the future.

The connector image you show in post #1, looks like a "vertical launch" connector, that has an angle of 90° between the trace and the connector axis. It is not an "end launch" connector, one that mounts on the edge of the PCB keeping the connector axis in line with the trace on the PCB. The 90° bend may restrict the bandwidth. Or maybe that is the SMA you refer to, that does not mate with a 1.85mm connector?

A thorough internet search (AI) for work on the problem of connection to a circuit board could be good value. You need to find someone who's actually had similar problems and solved them.