[SI-LIST] Re: Do you really ship products at BER 10e-xx ?
- From: <art_porter@xxxxxxxxxxx>
- To: <gtang@xxxxxxxx>, <Chris.Cheng@xxxxxxxxxxxx>, <si-list@xxxxxxxxxxxxx>
- Date: Wed, 20 Apr 2005 09:59:53 -0600
If you believe in statistical analysis, you also need to believe in =
calculating your confidence interval before making extrapolations. For =
small sample populations, extrapolating out to the fringes of the normal =
distribution can be very questionable, or to state it more precisely, =
the confidence interval can be unacceptable.=20
For a bit error rate of, say, 1E-12, you could have a large number of =
errors in the first few bits you examine; on the other hand, you also =
could see zero errors in 1E16 (or any arbitrarily large number) of =
consecutive bits. You could determine the bit error rate with near 100 =
percent confidence, but this could take an unnecessarily long time, =
especially if you never see any violations. Assuming you get no =
violations, you can calculate the confidence interval for predicting a =
given error rate from less data. For example, for a confidence level of =
95% of meeting a given BER, you would have to examine (3.84/BER) unit =
intervals. For a 99% confidence interval, you would have to examine =
(6.66/BER) unit intervals. =20
If you do see some errors, then these ratios (above) are invalid and you =
cannot extrapolate from them. In that case, you need to let the system =
run until you see a large number of errors to have good confidence that =
the number of violations divided by the number of unit intervals (UIs) =
tested is a good predictor of the long-term bit error rate.
For a further discussion of this topic, please refer to the application =
note "Memory-Depth Requirements for Serial Data Analysis in a Real-Time =
Oscilloscope," available at =
http://cp.literature.agilent.com/litweb/pdf/5989-1256EN.pdf.
The other complication that you have to deal with is deciding how much =
of the measured jitter is deterministic and bounded, and how much is =
random and unbounded. If you blindly follow the recommendation below =
(multiply the observed standard deviation by 14), you may come up with a =
way overly pessimistic answer. The deterministic, bounded portions of =
the jitter don't need to be multiplied by anything. That's where you =
need a tool to help you separate jitter into random and deterministic =
portions. There are a variety of commercially available tools to do that =
job. Since SI-LIST is not supposed to be for "commercials" I won't list =
any here.=20
Art Porter =20
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of George Tang
Sent: Tuesday, April 19, 2005 8:46 PM
To: Chris.Cheng@xxxxxxxxxxxx; si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Do you really ship products at BER 10e-xx ?
Chris,
If you believe in statistical analysis, you can use the following method =
to
determine BER of your system. You first measure the PLL output (for
simplicity, assume DJ=3D0) and record the time that each rising edge and
falling edge occurs. From these timed events, you can calculate jitter.
Plot this distribution as a function of jitter and determine the =
standard
deviation. About 68% of the sample data is within +/- 1 standard =
deviation.
This means if your eye opening inside the receiver is larger than +/- 7
standard deviation of the PLL RJ, your BER is 1e-12 or better.
Alfred, please email me the papers offline, too.
Thanks,
George
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx
[mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of Chris Cheng
Sent: Monday, April 18, 2005 7:36 PM
To: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Do you really ship products at BER 10e-xx ?
I thought that's the reason why some specs like Sonet spec out jitter
transfer Fc to roll off way before the loop bandwidth. How do you get =
there
is another problem.
I will be intrested to know what parasitics in the PLL can approach the
unity gain frequency to impact the stablity. Long long time ago there =
were
concerns about the PD output rc time constant to the loop filter or the
discrete Z sampling pole but I don't know how they look like with =
internal
filters and faster sampling rate. Still, I claim you can always observe =
the
problem by the cold start step response and thereby going back to the
drawing board. No suprise there.
Here's what really get me started on this thread. Customer service =
always
want us to give an MTBF estimate for everything so as to generate a =
support
cost structure. I think I understand the power supply, fan fail, bad =
chips,
bad discrets and bad disks problem. But I never quite "get" the 10e-12 =
link
error rates. Like Paul says, I've seen peta bytes of disk humming for =
weeks
(minus the disk or real components failure not related to the link =
itself).
I don't want to scare myself silly for a statistical number that no one
really care but everyone put it in the book. But if you tell me that's =
real,
I will get phone calls from CS engineers. How does one budget CS service
calls for the BER down the road at 8G FCAL for a peta byte installation =
?
Even if the BER is a million times better than 10e-12 ? My money is on
device (disk,fan,power supply etc) failures will dwarf the link failure
rate. But I can't proof that on a satistical basis.
Al, I will appreciate the papers off line also.
TIA,
Chris
-----Original Message-----
From: Alfred P. Neves [mailto:al.neves@xxxxxxxxxxx]
Sent: Monday, April 18, 2005 4:20 PM
To: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Do you really ship products at BER 10e-xx ?
When we had to optimize the loop characteristics for PLL's an overriding
consideration has been the conflicting requirements of low jitter
generation (GR-253, Sonet) and jitter transfer. If you make the loop BW
large, and the loop gain high you often have poor loop stability of
phase/frequency, such that you reject intrinsic VCO noise, but have poor
jitter transfer performance. Remember that there are parasitics in the
loop and they cause peaking in the loop response, along with higher
order poles in VCO, so when the loop BW increases loop stability may
suffer. Verify that the equations are correct (they are usually fairly
lousy linearized approximations of a charge-pump sampling system)and the
loop is set correctly, we can check the loop dynamics with Spectrum
Analyzer and autocorrelation analysis often using Wavecrest instruments.
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
On Behalf Of Chris Cheng
Sent: Monday, April 18, 2005 2:46 PM
Cc: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Do you really ship products at BER 10e-xx ?
My PLL design is getting rusty so if I am mumbling nonsense please
correct me. If you are dealing with classical PLL design, the loop
filter is always a trade-off between the high pass VCO phase noise and
the low pass input phase noise. My money is always on the VCO phase
noise and I optimize it for such. Afterall, if I read Geogre's response
in the same thread, "But if you solve all these problems to the extent
that there is sufficient eye-opening inside the receiver, then you are
dealing with errors caused by the second-order effects, mainly the RJ
from TX PLL, RX PLL / CDR circuits." That's sounds like a controlled
input phase noise (good eye opening) vs. a out of control VCO phase
noise problem (you have non-zero BER). And that's why our employers pay
us peanuts to design a power distribution that ensure no "large supply
glitches" or at least some good PLLVDD filters to avoid that to happen,
right ?
As to making sure the loop filter damping factor, one can either observe
directly the VCO control voltage or monitor the VCO output frequency in
the modulation domain during a cold start to deduce the stability
factor, there is no magic about it. You either get it right or you are
back to the drawing board. There is no 10e-xx probably you are either
right or wrong. And if you are talking about these bang bang PLL's.
AFAIK, if you are operating under the slew rate limit, your hunting
jitter is bounded and is related to the metastability limit of the FF
you use. And they have the added bonus of even if the input phase noise
is large, they are limited by Jwalk which is sqrt of input jitter. I
can't argue that a 1ps rms xtalk MAY have a 100ps jitter within the life
time of the universe but it is hard to extend that to say a properly
design system is one big 10-exx distribution.
------------------------------------------------------------------
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:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
------------------------------------------------------------------
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:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org
List archives are viewable at: =20
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
=20
------------------------------------------------------------------
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:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
