Hi John,
How sure are you in the Er value? Have you measured it with sufficient
confidence or is it a data sheet estimate? Also, if you have any
capacitive discontinuity at the boundaries (may come from connectors,
vias), it will electrically lengthen the DUT. Those two items can easily
account for the 4.45" vs 5" difference.
Regards,
Istvan Novak
Oracle
On 5/24/2016 12:01 AM, John Lin wrote:
Dear Istvan, Alfred, and ohers,
Thank you for your explanations. It helps.
I briefly calculate the period of the spikes. It is about 1.44Ghz
between two spikes. With PCB's Er=3.4 , the Half_length is about
4.45 inch.
I used 5 inch trace as fixture to de-embed 16 inch trace. How does the
half-length(4.45inch) relate to 5 inch or 16 inch traces.
Thanks again,
John Lin
2016-05-24 9:30 GMT+08:00 Istvan Novak <istvan.novak@xxxxxxxxxxx
<mailto:istvan.novak@xxxxxxxxxxx>>:
Al,
Yes, I agree. Though I was also thinking about the situation when
the minimum points of the main diagonal elements (S11 and S22 in
this case) become noisy, making the matrix inversion more
difficult. That is when lowering the trace noise and checking
data quality as you said becomes crucial.
Best regards,
Istvan Novak
Oracle
On 5/23/2016 10:13 AM, Alfred P. Neves wrote:
Istvan,
If the trace impedance is off by much more than 5% and the launch
signal integrity is bad (peak TDR impedances of >10ohms), you
would have a long resonator essentially, that could also cause
the half wave resonances also I believe. Would you agree?
- Al
/
/
/Products for the Signal Integrity Practitioner/
/
/
*
*
*
Alfred P. Neves*
Chief Technologist
Office: 503-679-2429 <tel:503-679-2429>
*/www.wildrivertech.com <http://www.wildrivertech.com>/**//*
/2015 Best In Design&Test Finalist/
On May 23, 2016, at 4:40 AM, Istvan Novak
<istvan.novak@xxxxxxxxxxx <mailto:istvan.novak@xxxxxxxxxxx>> wrote:
John,
Becasuse of the matrix inversions involved, deembedding can be very
sensitive to minor measurement errors and inaccuracies. You may see
those spikes near multiples of half wavelength frequencies of the
structure where the main diagonal elements of the S matrices go
through
zeros.
Regards,
Istvan Novak
Oracle
On 5/23/2016 12:00 AM, John Lin wrote:
Dear all SI gurus,
While doing de-embedding by removing 5 inch trace as
fixture from 16
inch trace,
I find there are many spikes on my de-embedded S21 plot.
Here are my approaches.
1. Transfer both 5 inch and 16 inch S2P file to T paramter (matrix)
2. cut the 5inch into half by T_2p5inch *T_2p5inch = T_5inch
==>
T_2p5inch=T_5inch^ 0.5; assuming symmetry.
3. do inverse(T_2p5inch)
4. De-embedded result = inv(T_2p5inch)*T_16inch*inv(T_2p5inch)
5. Convert De-embedded T parameters back to S parameters.
I also use Delta-L for de-embedding. Both S21 plot can match
well,
except the spikes.
I don't know what happen for the spikes and seek any idea from you.
Thanks,
John Lin