Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: [802.3ae] XAUI jitter tolerance test




Tom- good point, getting my English confused, I was thinking you meant the noise from the dividers in the PLL. A complete PLL will definately introduce RJ, most probably the dominant for a CMOS design. Sorry for adding confusing (not jitter to the thread). 

I put the email distribution list on again, as I want to make sure I didn't confuse anyone out there..... Don't want people only reading half the thread.


-Anthony



-----Original Message-----
From: Lindsay, Tom [mailto:tlindsay@stratoslightwave.com] 
Sent: Dienstag, 21. Mai 2002 18:18
To: Michae de Bie; anthony.sanders@infineon.com
Subject: RE: [802.3ae] XAUI jitter tolerance test


Hi guys - I'm trimming the list down.

Michael - some of your statements are not clear to me. Do you understand that the XAUI spec requires that quantification of RJ be done above 20 MHz, not below? That is, a broadband spectrum is required.

Anthony - doesn't it depend on the multiplier properties and what it is multiplying? If the initial clock is Gaussian (but not white), then it will still be after multiplication. Also, if the multiplier has a lot of VCO noise at its input, that could also be Gaussian. So can you explain what will necessarily be bounded and deterministic? A lot of the stuff I've seen is pretty random.

Tom

-----Original Message-----
From: Michae de Bie [mailto:mdebie@wavecrest.com]
Sent: Tuesday, May 21, 2002 8:35 AM
To: anthony.sanders@infineon.com; Lindsay, Tom; qiyu@silabs.com; stds-802-3-hssg@ieee.org
Subject: Re: [802.3ae] XAUI jitter tolerance test


Greetings,

I would concur with Anthony's amendments and Tom's main points.  It is important to add the Gaussian noise with sufficient frequency so as to emulate the high frequency noise that will be present of a real network as a result of the system components.  The noise spectrum of a live network does extend to the carrier frequency and as such should be represented in the tolerance test by applying full spectrum Gaussian jitter to at least 20MHz. This reference to 20MHz is more a reflection on the challenge of creating Gaussian jitter at high frequency while maintaining a reasonable peak to RMS ratio.

I would highly recommend using some sort of jitter analysis equipment when setting up a tolerance test to ensure that the jitter components you are attempting to create meet specification.  Some examples are stated in Annex 48B and include a TIA (See www.wavecrest.com for a complete listing of automated XAUI analysis equipment) and a BERT.  You might be hard pressed to get PJ amplitudes from the BERT, so, I would recommend either the TIA or some combination of sampling oscilloscope and spectrum analyzer for the PJ amplitude and frequency calculation.

Regards,
Michael de Bie

----- Original Message -----
From: <anthony.sanders@infineon.com>
To: <tlindsay@stratoslightwave.com>; <qiyu@silabs.com>; <stds-802-3-hssg@ieee.org>
Sent: Tuesday, May 21, 2002 2:21 AM
Subject: RE: [802.3ae] XAUI jitter tolerance test


>
> Would almost agree with the first statement, except that the noise
that is
generated by the multipliers would be bounded and therefore according to our definitions deterministic. Could be they are still coloured in the spectrum, but they are not Gaussian.
>
> Definately agree with the second statement, the DJ has a very
complicated
spectrum which at least theoritically can go to very low frequency, such that the CDR starts tracking low frequency components from "killer patterns". This is exactly the reason why one shouldn't have a CDR BW that is too high, otherwise you will start tracking wander that is not actually there. (i.e. it's down converted HF jitter).
>
> -Anthony.
>
>
> -----Original Message-----
> From: Lindsay, Tom [mailto:tlindsay@stratoslightwave.com]
> Sent: Freitag, 17. Mai 2002 22:07
> To: Qicheng Yu; stds-802-3-hssg@ieee.org
> Subject: RE: [802.3ae] XAUI jitter tolerance test
>
>
>
> I'll try this.
>
> 1. The purpose of the statement was to help ensure that someone would
not
create RJ with mostly low frequency content that would be tracked out by the tested CDR. So, in that sense, the spec is intending to allow high frequency RJ and to ensure that a CDR can appropriately dissipate it.
>
> In reality, RJ will probably not have a lot of high frequency content,
and
in fact may have frequency content dominated below a few MHz. Sources include clock synthesis, multipliers, etc. The spec assumed that if a CDR can tolerate high frequency RJ, it can also tolerate low frequency RJ.
>
> 2. I am still putting some thought into this, but I believe it is too
simple to say that deterministic jitter is mostly high frequency. I agree that common mechanisms that create DJ may be high frequency (such as high frequency rolloff in the channel, DCD, etc.), but the jitter spectrum generally tracks the spectrum of the data itself. Indeed, CJPAT creates significant low frequency jitter when passed through a lossy channel.
>
> Also, mechanisms such as inadequate low frequency response also map
into
low frequency DJ. But again, the jitter spectrum will tend to match that of the data.
>
> (Of course, a benefit of 8B10B is that its spectral content rolls off
at
the low end when compared to scrambled codes).
>
>
> So, in real systems, you will have to tolerate other jitter sources -
both
RJ and DJ.
>
>
> Thoughts?
>
> Tom Lindsay
> Stratos
> (425) 672-8035 x105
>
>
> -----Original Message-----
> From: Qicheng Yu [mailto:qiyu@silabs.com]
> Sent: Friday, May 17, 2002 5:05 AM
> To: stds-802-3-hssg@ieee.org
> Subject: [802.3ae] XAUI jitter tolerance test
>
>
>
>
>
>
>
> I have a question on the jitter tolerance test signal for XAUI
receiver.
In P802.3ae/D4.3, section 47.4.3.2 Jitter Tolerance, on page 305, it is said "Random jitter is calibrated using a high pass filter with a low frequency corner of 20MHz and 20dB/decade roll off below this." Does this mean the random jitter component of the jitter tolerance test signal is mostly high frequency, above 20MHz? If so, since deterministic jitter is also by nature high frequency, the amount of jitter a receiver has to tolerate at, say, 2MHz, would be mostly the 0.1UIpp sinusoidal jitter and not much more. Is this correct? Thanks for your attention.
>
> Qicheng Yu