Re: [10GBASE-CX4] Comments on initial draft of 10GBASE-CX4 PMD
Mike,
This is not quite "compliance channel", but does it address your
issues?
+-----------+ +----------------+
| TX PMD | v(t) |Standard range |
| device +------------>| of test load |
+-----------+ \ +----------------+
\
\ +--------+ +-------------+
--->| G(s) +---->| time-domain |
+--------+ | template |
+-------------+
The principles:
(1) TX output v(t) is being constrained by the spec. This is
defined at the physical connector (interop point)
(2) The constraint on v(t) is expressed as a temporal template
applied *after* a linear filter
(3) As a design spec, the constraint should be met over a
full range of legal test loads and legal test signals
(4) As a conformance test, perhaps one test signal (the
low freq pulse already proposed) and two limit conditions
on test load are satisfactory.
G(s) is a complex frequency response function which we need to make
up in committee. It need not exactly model a feasible (or nominal)
channel. In fact, if G(s) is optimized for the purpose of ensuring
interoperability without overspecifying, it may deviate quite a
bit from real channel characteristics.
In any case, let's make sure that we recognize a design spec and
a conformance test are ideally based on the same model but are
*two different things*.
Cheers,
Chuck Harrison
Far Field Associates, LLC
+1 360 863 8340 (voice) PST = GMT-0800
Mike Jenkins wrote:
>
> Howard and all,
>
> First of all, my apologies for jumping in somewhat late on this
> discussion. I appreciate all the work that has gone into
> 10GBASE-CX4 to date, and I hope my comments will be received as they
> are intended -- as constructive.
>
> I will list several specific comments below, but my overall
> not-so-hidden agenda is to persuade the group not to abandon the
> XAUI compliance interconnect concept and to avoid required TX
> near-end measurements. I offer three reasons for this:
>
> 1. (The practical reason) Near-end TX waveforms have maximum high
> frequency content, exciting any and all fixture resonances.
> Hence near-end measurements -- especially with restrictive
> templates -- can be a nightmare of small excursions outside the
> template. Yet these high frequency aberrations are attenuated
> by the transmission path and affect the far-end signal quality
> very little if at all. Far end measurements are much more
> benign and to the point.
> 2. (The theoretical reason) A jitter budget, with jitter
> increasing down the transmission path made sense for
> trapezoidal waveforms, but not for pre-emphasized TX
> waveforms. With pre-emphasis, the jitter can be better at the
> far end of the path than at the TX output. Specifying such TX
> waveforms to guarantee an adequate RX waveform is complex and
> sensitive to assumptions.
> 3. (The time-to-market reason) It appears inevitable that the RX
> will require additional equalization compared with XAUI.
> Hence, keeping the TX specifications as close as possible to
> XAUI(at least in format) would seem to offer the lowest risk,
> fastest route to market for 10GBASE-CX4.
>
> For what it's worth, a compliance channel approach to TX specs would
> also render the issue of minimum TX amplitude irrelevant.
>
> Here are some specific comments on sections of the draft:
>
> * 54.7.3.2 Load For accurate measurements of 3.125 Gb/s
> signals, especially at the TX, 2.5 GHz is probably not adequate
> bandwidth. (At the far end, after 12 or 20 dB of attenuation
> at these frequencies, this is much less of an issue.)
> * 54.7.3.4 Output Impedance and 54.7.4.5 Input Impedance I
> realize these sections are out of XAUI, but I would like to
> point out that the values specified (if my math is correct) are
> equivalent to 1.04 pF of dif'l load capacitance on the TX and
> 0.424 pF of dif'l load capacitance on the RX. These would be
> problematic limits.
> * 54.7.3.6 Differential Output Template (I gather there has
> already been discussion on this item, but I'll add my two
> cents.) This template is unworkably tight. The definition of
> normalized amplitude guarantees that the waveform will never
> exceed +/-1, so template values outside that range are
> meaningless. This leaves a mere 7 percent of peak-peak
> amplitude -- 3.5% at each extreme -- as a target. With
> resonances and +/-5% load tolerances, this won't work. The
> implied risetimes are also unrealistic. The slowest risetime
> of a trapezoidal signal that would fit within the template
> would have a risetime of 101.4 ps. Realistic, curved waveforms
> would need to be even faster. Of course, all these problems go
> away with the compliance interconnect approach.
> * 54.7.3.7 Transmitter Jitter The added requirement on the mean
> of jitter distributions will probably invalidate a number of
> existing jitter measurement approaches. A lot of capital
> equipment might be obsoleted with this additional requirement.
> * 54.8.2 Cable Assembly Insertion Loss Just some questions
> here. I do not understand the relevance of the 1/sqrt(f)
> term. This is large at low frequency and smaller at higher
> frequencies. Also, it should be stated that "f" in the
> expression is in units of Hz. Also, I believe the inequality
> is in the wrong direction. And, lastly, the sentence ending
> "...deviate by more than 10% from equation 54.3." might need
> some explaining. 54.3 is, after all, an inequality. Does this
> allow +/-10% deviation from the right hand side expression? Or
> only -10%?
>
> Again, apologies for the last minute comments.
>
> Regards,
> Mike
>
> --
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> Mike Jenkins Phone: 408.433.7901 _____
> LSI Logic Corp, ms/AH260 Fax: 408.433.2840 LSI|LOGIC| (R)
> 1873 Barber Lane mailto:Jenkins@LSIL.com | |
> Milpitas, CA 95035 http://www.lsilogic.com |_____|
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
>