Re: [BP] how to evaluate signaling method
adam,
Thank you for the comments-questions.
1. Your description of my procedure is correct. Since my initial
posting, Mike Altmann asked about possible limits to using a
single frequency and i have suggested using a PRBS pattern to
get around that issue.
The reason for using a sine wave instead of the the result of (c)
is that the sine wave spends more time near its peak giving more
exposure to worst case conditions and reducing the number of bits
we need to simulate. See my exchange with Joe Abler and mike.
Yes, we will need to repeat steps (a)-(c|d) but fortunately that
is not very hard. The method is outlined in my Portland talk on
impulse modeling of channels.
2. Linear equalization in the receiver, like Tx equalization, will
change somewhat the worst case pattern. Of course if everyone
uses DFE the interference will not be affected by the Rx
equalization and everything will be fine.
charles
> Charles,
>
> Thank you for taking the time to pull this outline together so that we
> may discuss it prior to next Thursday's meeting. I have just now had
> time to review the thread and I had some questions related to the nature
> of the sinusoidal interferer. I'll start by quoting one your earlier
> responses.
>
> "The xtalk to use is the upper bound of the Xtalk which the system specs
> allow. The amount of Xtalk will depend on the channel, the Tx low pass
> characteristics, Tx equalization, and signaling type. All these need to
> be taken into account but some things like Tx equalization may be hard
> to know a-priori and worst case values need to be used."
>
> 1. I agree that the requirements for interference tolerance are system
> (and signaling) dependent. Correct me if I am wrong, but based on your
> statement above, the procedure appears to be:
>
> (a) assume a certain low-pass characteristic and worst-case equalization
> setting for the transmitter
> (b) synthesize a waveform that represents some worst-case pattern and
> reflects the assumptions in (a)
> (c) pass the resulting waveform through the crosstalk channel
> (d) measure the peak-peak value of this waveform and then substitute a
> sinusoid of a particular frequency with the same peak-peak amplitude.
>
> My confusion is that if we go through the trouble of (a)-(c), why not
> just sum that result of (c) with the forward response? What is the
> value of the substitution for these simulations?
>
> I think the original concept of sweeping the sinusoidal amplitude to
> derive the interference tolerance is interesting. However, I'm not sure
> how one would compare these results on an apples-apples basis since not
> all systems will see equal interference. Comparing margins to the
> required interference tolerance may be a way around this, but to do so
> you need to go back and do the calculations for (a)-(c).
>
> 2. Let's assume for the sake of argument that the receiver contains a
> equalization function (possibly adaptive) consisting of some zeros and
> poles or perhaps a feed-forward equalizer. This shapes the spectrum of
> the crosstalk (perhaps in a very complex way) and would yield a change
> in the peak-peak amplitude of the noise as seen at the slicer. It is
> not clear to me that gain/phase of this filter response at a the
> frequency of the injected sinusoid will yield a comparable change in the
> peak-peak value. Can you comment on this?
>
> Thank you,
> -Adam
>
>
>
> -----Original Message-----
> From: owner-stds-802-3-blade@listserv.ieee.org
> [mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of Charles
> Moore
> Sent: Wednesday, August 25, 2004 6:29 PM
> To: STDS-802-3-BLADE@listserv.ieee.org
> Subject: Re: [BP] how to evaluate signaling method
>
>
> mike,
>
> First to comment on your possible responses:
>
> 1) If the interference is a fairly large fraction of the baud rate
> the adaptation will not be able to track it directly and it
> tries
> to chop out that frequency it will miss any data in that range.
> 2) Slightly re-worded: it may be much more sensitive to some
> frequencies. So far at Agilent we have not seen this, but it
> might happen. We will have to remain alert.
> 3) If the effective Rx EYE is very sharp, with very small
> difference
> between the minimum EYE opening and the maximum amplitude this
> will happen. In the real world noise will prevent it from
> happening. In any event the amplitude just below where failure
> will be the interference tolerance.
>
> The reasons for using a constant frequency are:
>
> 1. It spends a lot of time near its peak therefore doing a good
> job
> providing worst case interference.
> 2. In actual measurements the interfering signal cannot be
> injected
> right at the Rx and will pass through some dispersive
> elements.
> A single frequency sine wave is the only signal which will not
> have its wave shape altered by the path.
>
> For simulation purposes, we could use some sort of PRBS signal since it
> need not suffer degradation. It will have many frequency components and
> remain at its peak value (+/-) essentially all the time.
>
> charles
>
> > Charles,
> >
> > It seems that if the interference is a single frequency interferer,
> then > we need to consider the possible responses of a PHY to the
> interference: > 1) it might, in the case of an adaptive PHY, track some
> of it out > 2) it might fail at one frequency, where at another it
> might pass > 3) If might pass until a defined amplitude, then fail
> totally (likely) > > In all of these cases, however, a single
> frequency test ignores the > effect of the frequency distribution of
> real interferers. Some > frequency components will be tracked out (by
> CDRs and adaptive > equalizers, etc), while others will not. The
> real-world performance of > the PHY will be a combination of these. I
> think, for this reason, that > we need to consider the interferer
> frequency distribution in this test. > Can this test be modified to
> account for this? Can we use measured > NEXT/FEXT response to > >
> .../Mike > > -----Original Message----- > From:
> owner-stds-802-3-blade@LISTSERV.IEEE.ORG
> > [mailto:owner-stds-802-3-blade@LISTSERV.IEEE.ORG] On Behalf Of
> Charles > Moore > Sent: Tuesday, August 24, 2004 2:16 PM > To:
> STDS-802-3-BLADE@LISTSERV.IEEE.ORG
> > Subject: Re: [BP] how to evaluate signaling method
> >
> > john,
> >
> > 1. a. In measurement we pick one frequency and sweep the
> amplitude
> > of interference. By plotting BER vs interference
> amplitude
> > at the we can find the Rx noise. Knowing Rx noise we can
> > compute an adder to the speced Xtalk tolerance to
> > extrapolate
> > from a measurable BER to the BER we want. This really
> > should
> > only need to be done at one frequency.
> >
> > b. The xtalk to use is the upper bound of the Xtalk which
> the
> > system specs allow. The amount of Xtalk will depend on
> the
> > channel, the Tx low pass characteristics, Tx
> equalization,
> > and signaling type. All these need to be taken into
> account
> > but some things like Tx equalization may be hard to know
> > a-priori and worst case values need to be used.
> >
> > c. We do not have much comparison.
> >
> > 2 Hspice allows elements which are defined in terms of
> > S-parameters
> > in touchstone (s4p) files. You measure them, we will use
> them
> > in
> > simulation, and that will be real data.
> >
> > 3. The number of bits used a trace off between simulation time
> and
> > confidence. Maybe we should hunt for the interference
> > tolerance
> > limit using a small number of bits, then use a larger number
> to
> > confirm it. Also, by worst-casing the interfering signal we
> > enormously reduce the amount of simulation needed.
> >
> > 4. The PWL will be complex but manageable i think. I should
> have
> > a
> > start at it to show by next meeting.
> >
> > charles
> >
> >> Charles,
> >> Some questions for you.
> >> 1. I have gone back and reviewed your presentation, and it looks to
> me > > for > >> crosstalk that you are proposing for a given
> frequency doing an > > amplitude > >> sweep then plotting the BER
> versus the Interference amplitude.
> >> a. I assume for multiple frequencies this process has to be
> >> repeated?
> >> b. This process seems to look at the xtalk performance of the
> >> channel
> >> only, rather than the product of the Tx spectrum and
> channel.
> >> Can
> >> you comment?
> >> c. Have you compared the summation of signal and proposed
> xtalk
> >> procedure against signal and broadband xtalk generation to
> >
> > see
> >
> >> how
> >> they compare? Can you share results?
> >>
> >> 2. You are proposing modeling the channel in HSPICE? Why would we
> do > > this >
> >> as opposed to use of real data?
> >>
> >> 3. The number of bits being proposed to run is 127 * 4 = 508 plus
> some >> offset bits. This seems like a very small number of bits to
> consider, >> especially in light of the BER we are trying to achieve.
>
>>> >> 4. Won't the PWL become complex pending the Tx solution being >
>
>> proposed? > >> >> John >> >> >> >> -----Original Message-----
>
>>> From: owner-stds-802-3-blade@listserv.ieee.org
>
> >> [mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of
> Charles > > Moore > >> Sent: Monday, August 23, 2004 6:11 PM >> To:
> STDS-802-3-BLADE@listserv.ieee.org
> >> Subject: [BP] how to evaluate signaling method
> >>
> >> guys,
> >>
> >> What i propose for signaling evaluation is an extension of
> >> ideas i presented at the July meeting in Portland in the talk >>
> "Receiver Testing Using Interference Tolerance Measurements". >>
> >> The basic idea is to do a time domain simulation of the Tx,
> >> channel, and Rx using a standard, generally available simulator. >>
> To provide a simple and reproducible model of cross talk, a >>
> sinusoidal interfering signal will be added at the input to the >>
> receiver. The amplitude of the interfering signal will be >> increased
> until the signal at the output of the Rx is deemed to >> be no longer
> usable. The highest level of interference at which >> the Rx provides
> a usable output will be called the interference >> tolerance. If the
> interference tolerance is below a specified >> (and perhaps signaling
> method dependent) value the simulated data >> path is non-compliant.
> If the interference tolerance is above >> the specified value, it is
> compliant and has a margin which is >> equal to the difference between
> the simulated tolerance and the >> spec. In general more margin is
> better. >>
> >> This is the basic idea. Details which i suggest be adopted as
> >> part of the method but which can be changed without altering the >>
> basic idea include: >> >> 1. Use hspice as the simulator. >> >> 2.
> Model the transmitter as:
> >> a. 1 or 2 piecewise linear (PWL) current sources:
> >> i. 1 current source with NRZ encoded data for NRZ
> >
> > signaling.
> >
> >> ii. 1 current source with precoded data for duo-binary
> >> signaling.
> >> iii. 2 current sources with 1/3 and 2/3 NRZ amplitude with
> >
> > LSB
> >
> >> and MSB data respectively, for PAM4 signaling.
> >> b. Rise times of PWL current sources set at about 30ps for NRX
> >
> > or
> >
> >> duo-binary or 60ps for PAM4
> >> c. R,L,C network between current source and Tx pins to provide
> >> reasonable return loss vs frequency
> >>
> >> 3. Include signaling method dependent Tx equalization in PWL
> current
> >> source model. Control equalization with hspice parameters.
> >>
> >> 4. Include some modeled Tx Jitter in PWL current sources. >> >>
> 5. Use hspice S-parameter network modeling capabilities to model
> >> channel.
> >>
> >> 6. Allow proprietary Rx models by Encryption.
> >>
> >> 7. Add interference signal at Rx with sinusoidal current source.
>
>>> >> 8. Allow for Rx input noise in the minimum interference
>
> tolerance > > spec. > >> >> 9. Determine that the Rx provides a
> usable output by:
> >> a. showing that data out of Rx gives the correct bits.
> >> b. If Rx does not include re-clocking, show that output eye is
> >> wide enough
> >> c. If Rx does not include a limiting stage, show that the
> output
> >> eye has enough amplitude.
> >>
> >> 10. Use 127 bit long PRBS pattern for data. Offset a few bits
> between
> >> pattern for MSB and LSB to give PAM4 pattern or use shorter
> PRBS
> >> pattern for LSB. Repeat PRBS pattern at least 4 times to allow
> >> interference to "walk through" the data.
> >>
> >> 11. Test with more than one interfering rate. Interesting rates
> >> should include:
> >> a. For NRZ or PAM4, (1+.2/127)*(baud rate)/2
> >> b. For duo-binary, (1+.2/127)*(baud rate)/4
> >>
> >> charles
> >>
> >> --
> >>
--
|--------------------------------------------------------------------|
| Charles Moore
| Agilent Technologies
| ASIC Products Division
| charles_moore@agilent.com
| (970) 288-4561
|--------------------------------------------------------------------|