RE: [SI-LIST] : Guard trace question

Chris Cheng (hycheng@3pardata.com)
Mon, 27 Sep 1999 16:31:44 -0700

This is a multi-part message in MIME format.

------=_NextPart_000_0065_01BF0905.C9118BC0
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: 7bit

good point. guard trace is only meaningful if the shield has a low impedance
path
to the signal return current. if the signal jump layer, the shield trace has
to jump
with via and if the signals you try to shield switch layers in different
location, you
are out of luck. worst, if you have multiple signal coming in and out of
your
guardtrace, you might have an return SSO problem. all these guard trace
ideas
looks good on paper but very hard to do in real life.
chris
-----Original Message-----
From: owner-si-list@silab.eng.sun.com
[mailto:owner-si-list@silab.eng.sun.com]On Behalf Of Shimon Amir
Sent: Monday, September 27, 1999 6:36 AM
To: si-list@silab.eng.sun.com; si-list@silab.eng.sun.com;
gsanders@csidaq.com
Cc: eric
Subject: RE: [SI-LIST] : Guard trace question

Eric,

what was the guard trace connected to?

thanks,

Shimon

At 02:34 PM 9/25/99 -0500, Eric Bogatin wrote:
>>>>

Gary-

I think your original problem you are trying to solve is related to
achieving high isolation between two signal lines in a stripline
configuration, with the use of a guard trace. Your question was related to
how to estimate the isolation using some approximations and rules of thumb,
based on a reasonable starting place. I am a strong advocate of using both
of these types of analysis, and applaud your attempts. In addition, there is
also a third type of analysis you can use that, in this case, will save you
a lot of time and which should be a tool in your portfolio of skills:
numerical analysis.

There are a slew of 2D field solvers on the market that are accurate and
easy to use, from which you can get an answer to your problem, accurate to
probably better than 5%, in minutes. In one of the classes I teach, I review
the use of 2D field solvers for calculating the cross talk coefficients in
various structures and how to use these tools to help establish design rules
that minimize coupling. For mixed signal applications, there are often cases
where you need 100 dB isolation or more, as when you have a sensitive
receiver line near a digital line. I have appended a copy of an example I
use that illustrates the extra isolation gained by a guard trace between two
stripline signals. This particular example was done with the Ansoft Maxwell
2D field solver. In addition, the Hyperlynx tool can also analyze these
types of problems.

The specifics for this case are: 2, 50 Ohms lines, 5 mil line width,
about 12 mil thick dielectric between the planes, FR4 dielectric. In the
blue curve, we look at the near end cross talk coefficient of just the two
traces, as we grow the separation (edge to edge) between them. You can get
100 dB of isolation without the guard trace if you increase the separation
to about 50 mils. (These results are a little different from Ron's below. He
offered 40 dB at 30 mil separation and 100 dB at 75 mil separation. I
calculated 40 dB at 15 mils and 100 dB at 50 mils. However, I'm not sure the
geometry in my example is the same as Ron's)

The purple trace is the near end cross talk coefficient when a guard
trace is added to the space between the lines, with the condition that it is
wide enough to fill the gap, leaving a 5 mil space to the nearest adjacent
trace, and it starts out 5 mil wide. With this condition, typical of what a
fab house would require, the spacing between the signal lines has to be at
least 15 mils before a guard trace will fit. You can see that the presence
of a guard trace does have a dramatic effect on lowering the coupling. At 15
mil spacing, there is about -42 dB coupling in the stripline, all by itself,
but -60 dB with the guard trace.

This illustrates that guard traces may not have a common role in
straight digital applications (as distinct from mix signal), as when do you
need more than 40 dB isolation, which you can get without a guard trace?
But, when you do need more than 40 dB isolation, the guard trace can add
almost 20 dB isolation. By 30 mil separation, the stripline is at -70 dB and
the guard trace is better than -100 dB.

I can already hear folks saying- why does the isolation of the guard
trace start to flatten out at about 30 mil separation- why doesn't it get
any better? I'm going to leave this an open question to the forum to see
what folks post as possible explanations, then I'll comment.

--eric

Eric Bogatin
BOGATIN ENTERPRISES
Training for Signal Integrity and Interconnect Design
26235 W. 110th Terr.
Olathe, KS 66061
v: 913-393-1305
f: 913-393-1306
pager: 888-775-1138
e: eric@bogent.com
web: www.bogatinenterprises.com

-----Original Message-----
From: owner-si-list@silab.eng.sun.com
[mailto:owner-si-list@silab.eng.sun.com]On Behalf Of S. Weir
Sent: Saturday, September 25, 1999 2:01 AM
To: si-list@silab.eng.sun.com
Subject: Re: [SI-LIST] : Guard trace question

Ron,

Either you have a different formula, or I believe that you slipped a
decimal point:

The isolation required: -100db watts/watt = 1*10^-5 volts/volt

The high frequency coupling limit = 1/(1 + D/H^2)

Solving, I get:

D/H = sqrt( 1/99,999 ) = 1/316.

Which is a bit much to ask for.

I do not have a formula which is both simple and accurate for the case
where a via fence is used but for this high of an isolation, it is likely a
practical solution. There was a paper referred to about a week ago which I
have not got around to reading yet which might provide a decent guestimate.

Regards,

Steve.
At 04:01 PM 9/24/1999 -0700, you wrote:

Gary

I approach this from a slightly different angle but it may be useful
to
you.

1. For 40 db isolation = 1 % voltage coupling at the worst coupling
frequency quarter wave coupled length the spacing should be 3 X the
total height or 2 X 5mil material = 30 mil spacing. for 100 db it
should
be about 2.5 X 30 mils = 75 mils with no guard trace at all.

2. Another rule of thumb you should consider is that you can get
coaxial, near complete isolation if the spacing between the
stitching
is much shorter than a quarter wavelength at the highest frequency
of concern. Figure about 170 ps per inch or about 6 mils per ps in
FR4.

These are rules of thumb I use and I am willing to listen to other
rules if anyone wishes. However, I will not defend the accuracy.
They are quick and dirty.

Ron Miller

Gary Sanders wrote:

I want to use guard traces on a multilayer PCB to provide almost
complete isolation, simulating the results as if coax were used
instead.
I want to bury two signal layers between ground plane layers.
It seems to me that superposition must apply to EM fields in this
situation. I want to find the spacing and trace width of the guard
traces to provide about 100 dB of total isolation. I know that the
additional isolation provided by a guard trace is 6-12dB
(depending on
whether ground stitching is provided). This assumes that the guard
trace
is the same width as the signal trace, and is centered between the
aggressor and the victim traces, spaced by the width of the signal
trace.

Here's the specific question- is the isolation provided by a very
wide
guard trace the superposition of the isolation of several normal
width
guard traces placed side-by-side? Example--- assume 5 mil signal
traces,
a 5 mil layer thickness of FR4, and a spacing between them of 20
mils
(between centers). The isolation calculated by formula without any
guard
trace is 24.6dB. A stitched-ground guard trace would increase this
isolation to 36.6dB.

NOW- if we instead used a 25 mil wide guard trace, centered such
that it
is 25 mils from the other traces, what is the isolation? I
calculate
48dB without the guard trace. If we assume superposition, then the
25
mil guard trace is like having five 5 mil guard traces, each
providing
12dB of isolation; so we have a total of 60dB isolation. This
added to
that of the total spacing is 48dB + 60dB = 108dB of isolation. SO-
is
this correct? Can I assume that the wide guard trace is the
superposition of several narrow guard traces? Note that this is a
big
improvement on using a normal width guard trace, since the spacing
between the aggressor and victim would have to be 1" to get 104dB
of
isolation (with a 5mil guard trace); and the wide guard trace
approach
gives the same isolation with a spacing of 80 mils. I appreciate
anyone's thoughts.
--
Regards,
Gary L. Sanders, Staff Analog Engineer, gsanders@csidaq.com
L3 Communications, Inc. Celerity Systems
<http://www.csidaq.com/>www.csidaq.com
Cupertino, CA dir 408-861-7325 fax 408-873-1397
Ultra Fast Acquisition & Data Generation Systems
"The fog is wine. The sun is my gold."

**** To unsubscribe from si-list: send e-mail to
majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
si-list, for more help, put HELP. si-list archives are accessible at
<http://www.qsl.net/wb6tpu/si-list>http://www.qsl.net/wb6tpu/si-list ****

--
Ronald B. Miller&nbsp; _\\|//_&nbsp; Signal Integrity Engineer
(408)487-8017&nbsp;&nbsp;&nbsp; (' 0-0 ')
fax(408)487-8017&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;
&nbsp;&nbsp;&nbsp;&nbsp; ==========0000-(_)0000===========
Brocade Communications Systems, 1901 Guadalupe Parkway, San Jose,
CA&nbsp; 95131
rmiller@brocade.com,&nbsp; rbmiller@sjm.infi.net

Attachment Converted: "c:\eudora\attach\guard.pdf"

<<<<

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Shimon G. Amir Avici Systems, Inc.
samir@avici.com 101 Billerica Ave
Signal Integrity N. Billerica, MA 01862-1256
(978)964-2269 http://www.avici.com
=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-

**** To unsubscribe from si-list: send e-mail to
majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE
si-list, for more help, put HELP. si-list archives are accessible at
http://www.qsl.net/wb6tpu/si-list ****

------=_NextPart_000_0065_01BF0905.C9118BC0
Content-Type: text/html;
charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable

<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">

good=20 point. guard trace is only meaningful if the shield has a low impedance=20 path
to the=20 signal return current. if the signal jump layer, the shield trace has to = jump
with=20 via and if the signals you try to shield switch layers in different = location,=20 you
are=20 out of luck. worst, if you have multiple signal coming in and out of = your=20
guardtrace, you might have an return SSO = problem. all=20 these guard trace ideas
looks=20 good on paper but very hard to do in real=20 life.
chris
-----Original Message-----
From:=20 owner-si-list@silab.eng.sun.com = [mailto:owner-si-list@silab.eng.sun.com]On=20 Behalf Of Shimon Amir
Sent: Monday, September 27, 1999 = 6:36=20 AM
To: si-list@silab.eng.sun.com; si-list@silab.eng.sun.com; = gsanders@csidaq.com
Cc: eric
Subject: RE: = [SI-LIST] :=20 Guard trace question

Eric,

what was the = guard trace=20 connected to?

thanks,

Shimon

At 02:34 PM 9/25/99 = -0500,=20 Eric Bogatin wrote:
>>>>
Gary-

I=20 think your original problem you are trying to solve is related to = achieving=20 high isolation between two signal lines in a stripline = configuration, with=20 the use of a guard trace. Your question was related to how to = estimate the=20 isolation using some approximations and rules of thumb, based on a=20 reasonable starting place. I am a strong advocate of using both of = these=20 types of analysis, and applaud your attempts. In addition, there is = also a=20 third type of analysis you can use that, in this case, will save you = a lot=20 of time and which should be a tool in your portfolio of skills: = numerical=20 analysis.

There=20 are a slew of 2D field solvers on the market that are accurate and = easy to=20 use, from which you can get an answer to your problem, accurate to = probably=20 better than 5%, in minutes. In one of the classes I teach, I review = the use=20 of 2D field solvers for calculating the cross talk coefficients in = various=20 structures and how to use these tools to help establish design rules = that=20 minimize coupling. For mixed signal applications, there are often = cases=20 where you need 100 dB isolation or more, as when you have a = sensitive=20 receiver line near a digital line. I have appended a copy of an = example I=20 use that illustrates the extra isolation gained by a guard trace = between two=20 stripline signals. This particular example was done with the Ansoft = Maxwell=20 2D field solver. In addition, the Hyperlynx tool can also analyze = these=20 types of = problems.

The=20 specifics for this case are: 2, 50 Ohms lines, 5 mil line width, = about 12=20 mil thick dielectric between the planes, FR4 dielectric. In the blue = curve,=20 we look at the near end cross talk coefficient of just the two = traces, as we=20 grow the separation (edge to edge) between them. You can get 100 dB = of=20 isolation without the guard trace if you increase the separation to = about 50=20 mils. (These results are a little different from Ron's below. He = offered 40=20 dB at 30 mil separation and 100 dB at 75 mil separation. I = calculated 40 dB=20 at 15 mils and 100 dB at 50 mils. However, I'm not sure the geometry = in my=20 example is the same as = Ron's)

The=20 purple trace is the near end cross talk coefficient when a guard = trace is=20 added to the space between the lines, with the condition that it is = wide=20 enough to fill the gap, leaving a 5 mil space to the nearest = adjacent trace,=20 and it starts out 5 mil wide. With this condition, typical of what a = fab=20 house would require, the spacing between the signal lines has to be = at least=20 15 mils before a guard trace will fit. You can see that the presence = of a=20 guard trace does have a dramatic effect on lowering the coupling. At = 15 mil=20 spacing, there is about -42 dB coupling in the stripline, all by = itself, but=20 -60 dB with the guard trace. =

This=20 illustrates that guard traces may not have a common role in straight = digital=20 applications (as distinct from mix signal), as when do you need more = than 40=20 dB isolation, which you can get without a guard trace? But, when you = do need=20 more than 40 dB isolation, the guard trace can add almost 20 dB = isolation.=20 By 30 mil separation, the stripline is at -70 dB and the guard trace = is=20 better than -100 = dB.

I=20 can already hear folks saying- why does the isolation of the guard = trace=20 start to flatten out at about 30 mil separation- why doesn't it get = any=20 better? I'm going to leave this an open question to the forum to see = what=20 folks post as possible explanations, then I'll = comment.

--eric



Eric=20 Bogatin
BOGATIN=20 ENTERPRISES =
Training for Signal Integrity = and=20 Interconnect Design =
26235 W. 110th = Terr.=20
Olathe, KS = 66061
v:=20 913-393-1305
f: 913-393-1306 =
pager: = 888-775-1138
e: eric@bogent.com=20
web: www.bogatinenterprises.com=20


-----Original=20 Message-----
From: owner-si-list@silab.eng.sun.com=20 [mailto:owner-si-list@silab.eng.sun.com]On Behalf Of S.=20 Weir
Sent: Saturday, September 25, 1999 2:01 = AM
To:=20 si-list@silab.eng.sun.com
Subject: Re: [SI-LIST] : Guard = trace=20 question

Ron,

Either you = have a=20 different formula, or I believe that you slipped a decimal=20 point:

The isolation required: -100db watts/watt =3D = 1*10^-5=20 volts/volt

The high frequency coupling limit =3D 1/(1 +=20 D/H^2)

Solving, I get:

D/H =3D sqrt( 1/99,999 ) =3D=20 1/316.

Which is a bit much to ask for.

I do not = have a=20 formula which is both simple and accurate for the case where a via = fence=20 is used but for this high of an isolation, it is likely a = practical=20 solution. There was a paper referred to about a week ago which I = have not=20 got around to reading yet which might provide a decent=20 guestimate.

Regards,


Steve.
At 04:01 PM = 9/24/1999=20 -0700, you wrote:
Gary

I approach this from a slightly different = angle=20 but it may be useful to
you.

1. For 40 db isolation = =3D 1 %=20 voltage coupling at the worst coupling
frequency quarter = wave=20 coupled length the spacing should be 3 X the
total height or = 2 X=20 5mil material =3D 30 mil spacing. for 100 db it should
be = about 2.5 X=20 30 mils =3D 75 mils with no guard trace at all.

2. = Another rule of=20 thumb you should consider is that you can get
coaxial, near = complete=20 isolation if the spacing between the stitching
is much = shorter than=20 a quarter wavelength at the highest frequency
of concern. = Figure=20 about 170 ps per inch or about 6 mils per ps in FR4. =

These are=20 rules of thumb I use and I am willing to listen to other =
rules if=20 anyone wishes. However, I will not defend the accuracy.
They = are=20 quick and dirty.

Ron Miller

Gary Sanders wrote: =
I want to use guard traces on a multilayer PCB to = provide=20 almost
complete isolation, simulating the results as if = coax were=20 used instead.
I want to bury two signal layers between = ground=20 plane layers.
It seems to me that superposition must apply = to EM=20 fields in this
situation. I want to find the spacing and = trace=20 width of the guard
traces to provide about 100 dB of total = isolation. I know that the
additional isolation provided = by a=20 guard trace is 6-12dB (depending on
whether ground = stitching is=20 provided). This assumes that the guard trace
is the same = width as=20 the signal trace, and is centered between the
aggressor = and the=20 victim traces, spaced by the width of the signal
trace.=20

Here's the specific question- is the isolation = provided by a=20 very wide
guard trace the superposition of the isolation = of=20 several normal width
guard traces placed side-by-side? = Example---=20 assume 5 mil signal traces,
a 5 mil layer thickness of = FR4, and a=20 spacing between them of 20 mils
(between centers). The = isolation=20 calculated by formula without any guard
trace is 24.6dB. A = stitched-ground guard trace would increase this
isolation = to=20 36.6dB.

NOW- if we instead used a 25 mil wide guard = trace,=20 centered such that it
is 25 mils from the other traces, = what is=20 the isolation? I calculate
48dB without the guard trace. = If we=20 assume superposition, then the 25
mil guard trace is like = having=20 five 5 mil guard traces, each providing
12dB of isolation; = so we=20 have a total of 60dB isolation. This added to
that of the = total=20 spacing is 48dB + 60dB =3D 108dB of isolation. SO- is
this = correct?=20 Can I assume that the wide guard trace is the =
superposition of=20 several narrow guard traces? Note that this is a big =
improvement=20 on using a normal width guard trace, since the spacing =
between the=20 aggressor and victim would have to be 1" to get 104dB of =
isolation=20 (with a 5mil guard trace); and the wide guard trace approach =
gives=20 the same isolation with a spacing of 80 mils. I appreciate=20
anyone's thoughts.
--
Regards,
Gary L. = Sanders, Staff=20 Analog Engineer, gsanders@csidaq.com
L3 Communications, = Inc.=20 Celerity Systems <http://www.csidaq.com/>www.csidaq.com=20
Cupertino, CA dir 408-861-7325 fax 408-873-1397
Ultra = Fast=20 Acquisition & Data Generation Systems
"The fog is = wine. The=20 sun is my gold."

**** To unsubscribe from si-list: = send e-mail=20 to majordomo@silab.eng.sun.com. In the BODY of message put:=20 UNSUBSCRIBE si-list, for more help, put HELP. si-list archives = are=20 accessible at=20 = <http://www.qsl.net/wb6tpu/si-list>http://www.qsl.net/wb6tpu/si-lis= t=20 ****

--
Ronald B. Miller&nbsp;=20 _\\|//_&nbsp; Signal Integrity=20 Engineer
(408)487-8017&nbsp;&nbsp;&nbsp; (' 0-0 = ')=20 = fax(408)487-8017&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nb= sp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;= &nbsp;&nbsp;&nbsp;=20
&nbsp;&nbsp;&nbsp;&nbsp;=20 = =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D0000-(_)0000=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
Brocade Communications Systems,=20 1901 Guadalupe Parkway, San Jose, CA&nbsp;=20 95131
rmiller@brocade.com,&nbsp; rbmiller@sjm.infi.net=20


Attachment Converted:=20 "c:\eudora\attach\guard.pdf"=20 =

<<<<



-=3D-=3D-=3D-=3D-=3D= -=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-= =3D-
Shimon=20 G. Amir Avici Systems, Inc.
samir@avici.com 101 Billerica Ave =
Signal=20 Integrity N. Billerica, MA 01862-1256
(978)964-2269=20 = http://www.avici.com
=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D--=3D-= =3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-=3D-


****=20 To unsubscribe from si-list: send e-mail to = majordomo@silab.eng.sun.com. In=20 the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. = si-list=20 archives are accessible at http://www.qsl.net/wb6tpu/si-list=20 ****
------=_NextPart_000_0065_01BF0905.C9118BC0-- **** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****