tdr using vna+agilent

时间：04-07 整理：3721RD 点击：

Hi All,

Can someone please recommand a practical de-embedding techniques for PCB networks in high frequency range (upto 10GHz)?

For example, how to take out the SMA and microstrip transmission line effects so that I have accurate s-parameter measurements on a packaged device?

Many thanks in advance

Make your own VNA calibration tool.
Put back-to-back two SMA connectors (identical to the ones in your design), and calibrate the VNA using this as a “Short”, instead using the original Cal-kit short.
In this way you de-embedded the SMA connectors.
To take accurate S-parameters in a package I would recommend using a Probe Station.
For frequencies up to 2GHz you can use small FSC connectors, connected direct to IC pin.

You have to prepare the SMA connectors + microstrip line calibration tools, that is open, short, through, 50ohm.
Calibrate the network analyzer with these tool before your measurement

I think this application notes can help you.

CMOSBJT

may i suggest an alternative method with respect de-embedding?
If your VNA has time domain use it rather de-embedding. It's more easy fast and practical. Don't need any post processing, and it's recognized to be one of the most accurate method.
Also, SMA connectors, formally, cannot be de-embedded because the discontinuity of the gap dimension (at reference plane) is unknown. It's known only the gap uncertainty.
Only precision connectors can be formally de-embedded.

Do you mean TDR? Thanks.

First of all, thank you all for your suggestions/comments.

I had tried calibration with a custom built calibration board (SMA+microstrip lines) using THRU-REFLECT-LINE (TRL) method. However, I only get good results at lower frequency range (i.e. less than 4-5 GHz). Of course, the accuracy depends on the precisions of the fabrication process. That's why I was wondering whether there are more accurate method for higher frequency applications.

I like Sergio's suggestion thought I have not tried it. Do you mean time gating out the first reflected pulse (from the SMA) for de-embedding? But what about the mulitple reflection assuming SMA and microstrip lines are not completely matched at all frequencies? Do you think up to what frequency can we get good results?

Many thanks !

? Why TRL?. Your DUT is the sma connectorized unit!
Do SOLT calibration, set start freq. as low as possible, it is important for the algoritm used into TimeDomain.
Collect the measured data relative to SMA DUT, Then, helped by TimeDomain, cut away the connectors discontinuity.

I've used hp8510. The TimeDomain into 8510 is friendly (perhaps more friendly that other menues) , and allow you to see on screen the one-way reflections, then whit markers you may cut away the reflections you want.
Try! it's easy

You might want to look at our product, Spectro VNA. Using Time Domain Substitution, you can move your reference plane to anywhere that you can temporarily create a short.

There is an Agilent VNA fixture cal wizard that is handy in removing two end fixture's S-par responses.

It assumes the two end blocks are nearly identical (not so true at high frequency).

It just needs a center thru block.

One measure is END block+thru+END block; another measure is: END Block+DUT+END block. The wizard will then automatically de-embed the end blocks effect from 2nd measurement