[SI-LIST] Re: Jitter measurement floor on different high bandwidth oscilloscopes
- From: "Alfred P. Neves" <al@xxxxxxxxxxxxx>
- To: "'Mark Randol'" <Mark.Randol@xxxxxxxxxx>, <si-list@xxxxxxxxxxxxx>
- Date: Fri, 11 Jun 2010 10:29:50 -0700
Mark,
Thanks for the excellent info on spectrum analysis.
We have used spectrum analysis for PLL design. You can examine the gain of
the PLL, see evidence of loop peaking and any other phase/frequency loop
dynamics that might be creating issues. This analysis completely undresses
the PLL. A major source of jitter for the PLL is VCO noise, converted to
jitter by linear addition, then shaped by the PLL loop dynamics. With
spectrum analysis you see everything. John McNeil and Mike Li both have
written a bit about this.
For designing transmitter PLL, RX CDR's, a spectrum analyzer is a wonderful
tool, especially when you have a discrete model of a bang-bang type PLL that
you can establish correspondence.
On the other topic: It isn't we don't trust the existing DJ-RJ extraction
tools, its that the systems today are messy from a stochastic view and their
job is impossible to do. Crosstalk aggressors, plane resonances,
resonances from two impedance mismatches (vias connetors, etc.,), really
small UI's for very fast data rates... it is really tough for the tools to
separate out periodic components of the jitter PSD from RJ and ISI. There
are spectral lines everywhere and we have found, for practical real life
cases, that you need to supplement your backplane analysis with both 3D
electromagnetic solvers, and time/frequency domain measurements. Examine
the Power spectral density of most practical compliance patterns and they
are very white like, they look like noise. So, in almost every case, RJ is
grossly overestimated when analyzing crosstalk for most of the RJ-DJ
extraction tools.
I always see Moore's Law as a rough guide to scaling computer technology.
Similarly, the Central Limit Theorem suggests that eventually none of the
stochastic estimator tools will work eventually as the data rates scream and
systems become more integrated. The systems jitter pdfs looks more
Gaussian as you add more deterministic jitter PDF's.
Its DesignCon abstract submission time, if anyone out there disagrees ping
me, we can provide the hardware for some interesting test cases.
I also agree with your comment regarding Wavecrest, which is now Gigamax
(new and what looks like very good management). The old Wavecrest (I
worked there years ago), IMHO, probably had some of the most useful jitter
technology and really good technologists. Their advanced clock tester
which is TIA based, along with a spectrum analyzer would be really slick for
PLL/clock design/verification.
Alfred P. Neves <*)))))><{
Hillsboro Office:
735 SE 16th Ave.
Hillsboro, OR, 97123
(503) 718 7172 Business
(503) 679 2429 Mobile
Main Corporate office:
Teraspeed Consulting Group LLC
121 North River Drive
Narragansett, RI 02882
(401) 284-1827 Business
(401) 284-1840 Fax
http://www.teraspeed.com
Teraspeed is the registered service mark
of Teraspeed Consulting Group LLC
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] On
Behalf Of Mark Randol
Sent: Friday, June 11, 2010 7:17 AM
To: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Jitter measurement floor on different high bandwidth
oscilloscopes
I've done a good amount of work with high performance sub-picosecond
clocks and oscillators, but not serial data. So I know this applies to
clocks, but I don't have the experience to know how or if it applies to
data streams.
When looking at discrete tonal noise components using a 'tuned' receiver
like a spectrum analyzer or phase noise system, the RBW (Resolution Band
Width) of the measurement system and the shaping factor of the RBW
filter do not figure into their contribution. That's because of the
different nature of broadband noise-like signals and spectrally discrete
or CW signals. That is, as the RBW shrinks towards 0Hz, the measured
power of the discrete component remains constant, but the noise power in
the bandwidth decreases as a function of the bandwidth.
I believe (no firm analysis) that a phase noise system or spectrum
analyzer with a sufficiently low noise floor are the best instruments
for correctly characterizing RJ. DJ can be estimated, but its actual
contributions aren't directly measurable since the phase relationships
between the different frequency components are unknown.
Both types of systems, frequency and time domain based, are useful
tools. I've found it useful to use them to cross check each other since
there are perturbations in both types of systems that can lead to much
head scratching without the sanity check.
Gigamax (formerly Wavecrest) offers some interesting alternatives. I've
found their analysis software to be more powerful and easier to use than
any of the oscilloscope manufacturer's packages. The oscilloscope tools
may have improved in the several years since I used them, but if you can
get a look at one, it's worth your while to see. No affiliation other
than using the products.
--
Mark Randol, Evaluation Engineer
ON Semiconductor
My opinions, not ON's.
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
On Behalf Of T.K. Jeon
Sent: Thursday, June 10, 2010 8:30 PM
To: Alfred P. Neves
Cc: si-list@xxxxxxxxxxxxx; 'prasad'
Subject: [SI-LIST] Re: Jitter measurement floor on different high
bandwidth oscilloscopes
I guess that using a spectrum analyzer is one of the good ideas to
benchmark jitter packages, especially, for PJ-RJ separation.
I noticed that some postprocessing algorithms show poor performance to
separate PJ and RJ effectively when the source is clock-like(1010)
signal. In other words, when there is a spectral peak from the signal
spectrum, which should correspond to PJ, a jitter package could fail to
take that for PJ calculation. The phase noise at a spectral peak can be
easily obtained without any special tools. In order to compute phase
noise from signal spectrum, you can read X dB on the peak in the
sideband using the marker, and then you will need to subtract noise
power bandwidth, which is usually 1.2*RBW(Resolution Bandwidth).
Furthermore, a correction factor (roughly 2.5dB for Agilent spectrum
analyzer) should be added for noise distribution.
Regards,
TK
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at:
//www.freelists.org/archives/si-list
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at:
//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:
