[SI-LIST] Re: Lower epsilon achieves lower crosstalk

Thank you Ivan for the detailed explanation and others for participating in the 
dialog.
 
From a 30,000 foot view, it sounds like a lot of the cancellation effects 
depend on the symmetry and homogeneity of the stripline environment.
 
I take it then, that the asymmetric stripline case is not as quiet in terms of 
the Xtalk as the symmetric stripline?
 
Also, what piqued my interest in the first place was whether there is any 
significant advantage to keeping high speed traces on internal layers, for us 
these are usually asymmetric  strip lines, over microstrip routing. 
 
I will have to dig up some of those papers and presentations that have been 
referenced.
 
Best Regards,
 
Stefan Milnor
Kontron America
Fremont, CA
 
________________________________

From: Istvan Novak [mailto:istvan.novak@xxxxxxx]
Sent: Fri 11/12/2010 5:41 AM
To: Stefan Milnor
Cc: ZHENGGANG CHENG; Young; Liu Xike; si-list@xxxxxxxxxxxxx
Subject: Re: [SI-LIST] Re: Lower epsilon achieves lower crosstalk



Stefan,

Yes, it was correctly stated by Zhenggang that there is no (inherent)
far end
crosstalk in homogeneous interconnects.  The two minor clarifications are
useful to understand, but in the end result they do not make a big
difference.
The first clarification is that what is NOT present in a homogeneous
interconnect is the INHERENT far end crosstalk.  This means you get no
far end crosstalk if all ports are matched terminated.  With multiple
coupled
interconnects matching with single-ended resistors is just approximate,
so we need to assume the coupling is not very tight.  If you have major
mismatches, part of the near end crosstalk gets reflected back and you
will get a signal at the far end even in homogeneous coupled interconnects;
this is however due to reflections.  The second clarification is the need
for homogeneity: it means the entire cross section carrying the coupled
waves is electrically the same.  When we think about how stripline is
built up,
we immediately see ways how we can be in minor violation of this
requirement.
The dielectric layer below and above the trace come from different sources,
so even if they are nominally the same, practically there could be small
differences in Dk (and also in Df).  There are situations when even the
nominal parameters will be slightly different, for instance when the layer
below and above the trace have different glass-resin ratios: it changes
Dk and Df slightly.  Take also the fact that resin from the prepreg has
to flow
out from the dielectric to fill the voids where you etched away copper,
throw in glass-weave effects, and we can identify multiple reasons why
a stripline in practice may not be perfectly homogeneous.  On bare fabs
you can measure a small far end inherent crosstalk, but as said above,
these are minor second-order effects.  On the big picture there is no
far end crosstalk in stripline in matched case.

Zhenggand also correctly stated the reason expressed mathematically:
in striplines relative capacitive and inductive couplings are identical. A
physical explanation for it is that in homogeneous materials there is
only a single speed for any electromagnetic wave (it can be frequency
dependent though).  In non-homogeneous interconnects there will
be different propagation speeds and the racing condition between
the faster and slower waves will create the crosstalk waveform at the
far end.  In a bandlimited fashion the far end crosstalk in microstrips
can be compensated for, either on purpose, or sometimes we may
get it as a welcome side effect.  There have been many publications on
this subject: you can find one of our early publications
"Reduction of Far-End Crosstalk on Coupled Microstrip PCB Interconnect"
from 1994 posted at http://www.electrical-integrity.com/ >> Paper
download.

All in all, stripline has not only no (or very low) inherent far end
crosstalk,
but also in stripline near end crosstalk drops off much faster with
distance,
and therefore for the same overall crosstalk you can pack traces more
tightly.

Regards,

Istvan Novak
Oracle






On 11/12/2010 1:31 AM, Stefan Milnor wrote:
> Other Experts -
>
> Is this correct (that there is no FEXT for the stripline case)?
>
> Does this make stripline routing "better" than microstrip?
> ________________________________
>
> From: si-list-bounce@xxxxxxxxxxxxx on behalf of ZHENGGANG CHENG
> Sent: Thu 11/11/2010 5:18 PM
> To: Young
> Cc: Liu Xike; si-list@xxxxxxxxxxxxx
> Subject: [SI-LIST] Re: Lower epsilon achieves lower crosstalk
>
>
>
> Due to Cm/C-Lm/L =0, meaning negative inductive coupling canceled with
> positive coupling at far end.
> Zhenggang
>
> On Thu, Nov 11, 2010 at 5:10 PM, Young<long.0.yang@xxxxxxxxx>  wrote:
>
>> Zhenggang
>>         Could you tell us why there is no FEXT in the case of stripline?
>>
>>
>>
>> On Fri, Nov 12, 2010 at 03:25, ZHENGGANG 
>> CHENG<zhenggang.cheng@xxxxxxxxx>wrote:
>>
>>> Hi Liu,
>>> Xtalk analysis is very different for microstrip and stripline cases, and
>>> it
>>> also depends what xtalk you are talking about such as FEXT or NEXT.
>>>
>>> For stripline, there is no FEXT, and NEXT proportional to Cm/C. So in this
>>> case, lower Er, to keep the same impedance, H will be decreased, so Cm/C
>>> will be smaller.
>>>
>>> For microstrip case, FEXT is proportional to Cm/C-Lm/L, and NEXT is
>>> proportional to Cm/C+Lm/L. In your case, Lm/L is becoming larger (Lm is
>>> not
>>> changing, L is decreased due to smaller H), Cm/C may be smaller, then FEXT
>>> will become larger due to lower Er.
>>>
>>> Zhenggang CHENG
>>>
>>>
>>> On Wed, Nov 10, 2010 at 10:46 PM, Liu Xike<xike.silist@xxxxxxxxx>  wrote:
>>>
>>>> We could achieve lower crosstalk by using dielectric with lower
>>>> permittivity, while keeping *Z0* as a constant.
>>>>
>>>>
>>>> I'm new in SI, my understanding about this question is:
>>>>
>>>>
>>>>
>>>> If we use dielectric with lower permittivity while keeping *Z0* as a
>>>> constant.
>>>>
>>>> We need to reduce *h *to maintain a constant *C*, because *C*>
>>> *epsilon***w*
>>>> /*h*, and *Z0*=1/*vp*/*C*.
>>>>
>>>>
>>>>
>>>> Since *Z0* = sqrt(*L/C*), to keep* Z0* as a constant, the stripe line
>>> need
>>>> to adjust it's current distribution to maintain a constant *L*.
>>>>
>>>>
>>>>
>>>> Considering the loop area decrease because *h* is reduced, the current
>>>> distribution should be more tighter to maintain a constant *L*.
>>>>
>>>>
>>>>
>>>> Finally the tighter current distribution on the returning pass results
>>> in
>>>> the reduction of the cross talk.
>>>>
>>>>
>>>>
>>>> Are these correct? I will appreciate your help
>>>>
>>>>
>>>>
>>>> Best Regards
>>>>




------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
http://www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field


List technical documents are available at:
                http://www.si-list.net

List archives are viewable at:     
                http://www.freelists.org/archives/si-list
 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  

Other related posts: