RE: [10GBASE-CX4] SIGNAL_DETECT
As an aid in studying the high vs. low frequency content of IPGs, I built a spread sheet that contains about all the possible IPG patterns that meet XAUI specs. I have attached this both as an Excel spread sheet and a PDF file. The columns headed 'RD' reflect the running disparity at that point in a sequence. Almost all the patterns shown are three symbol patterns (the exceptions being the Start of Frame patterns). Among Ordered_Sets I have only included ||LF|| and ||RF||. The Start of Frame and Sequence Ordered_Set patterns are lane-specific, (Lanes1 & 2 are identical), the others are common to all lanes (except the unexpanded ||Tn|| bits, where 0 to 3 lanes are random data for this table).
The conclusion is that all the patterns contain a lot of 3- & 4- bit runs, and very few 1-bit runs, so that although the data-detecting circuit has to look at am equalized eye, a signal detect circuit can expect to see transitions close to the maximum value possible in an IPG. Hence a circuit that accepts near-LF signal amplitudes within a few bit times as 'signal present', but requires more than a full packet length absence before declaring 'signal not present', should have no trouble generating a reliable 'SIGNAL_DETECT' function within the terms of a XAUI-like requirement for CX4.
Peter Bradshaw
> -----Original Message-----
> From: Howard A. Baumer [mailto:hbaumer@broadcom.com]
> Sent: Monday, January 27, 2003 1:35 PM
> To: DOVE,DANIEL J (HP-Roseville,ex1)
> Cc: Rogers, Shawn; 10GBCX4
> Subject: Re: [10GBASE-CX4] SIGNAL_DETECT
>
>
>
> Dan,
> If we use the low freq. content of the IPG then the loss
> at those freq is
> going to be around 12-15db (for 312.5MHz los components are:
> ~6db from Tx eq.,
> ~6-9db from the channel). The Hf & Lf levels are within ~5dB
> (Rx eq takes up
> this slack) so the sig det thresholds would then be withn a
> factor of 2 (or
> less) of each other. The IPG will guarantee us that we will
> have some low
> frequency content so there will be something to check. If we
> look at the max
> packet size of 1600 Bytes (rounding to make it easy) then
> there is 400 bytes
> per lane which equates to 4000 symbols. Since thse could
> conceivably be the
> "1010.." pattern (valid 8B10B coded output) we would need to
> wait that time.
> 4000*0.32ns =1280ns.
> It becomes a trade off of the time used to get hysterisus
> to indicate
> signal loss versus the level we use to get signal on. The
> level to detect Hf
> will more than likely end up close to a noise threshold.
> Waiting a longer time
> (more like 4us to allow non-standard jumbo packets) is not that bad a
> compromise to get a bit more robust signal detect threshold.
>
> Howard
>
>
> "DOVE,DANIEL J (HP-Roseville,ex1)" wrote:
>
> > Howard,
> >
> > Your approach is similar to mine except you assume a worst
> > case "010101" pattern and I was willing to take advantage of
> > the fact that the IDLE pattern will contain more low-frequency
> > content and thus we could increase the amplitude of the
> > squelch. Your approach allows a shorter time-frame for
> > detection of the OFF state though.
> >
> > Dan
> >
> > -----Original Message-----
> > From: Howard A. Baumer [mailto:hbaumer@broadcom.com]
> > Sent: Monday, January 27, 2003 11:37 AM
> > To: Rogers, Shawn
> > Cc: 10GBCX4
> > Subject: Re: [10GBASE-CX4] SIGNAL_DETECT
> >
> > My $0.02,
> > I think Shawn has a good point here. The resultant receivers'
> > sensitivity
> > could end up quite low and therefore random noise could trigure a
> > SIGNAL_DETECT. If we were to use the CL 48 state machine
> fault sections
> > then
> > we presuppose there will always be a PMD and a PCS in the same Si.
> > Another way to look at the problem would be to look at what the
> > amplitude
> > of the high frequency signal would be (the "1010...", the transition
> > ..,0,1,...). Allowing for a loss of 16-20dB from transmit
> mdi to receive
> > mdi
> > then we could specify if the receive signal goes above -20dB of the
> > transmitted
> > signal SIGNAL_DETECT goes on and if it doesn't for 32ns
> after that it goes
> > off.
> >
> > Howard
> >
> > "Rogers, Shawn" wrote:
> >
> > > My $0.02:
> > >
> > > I do not see how SIGNAL_DETECT can be specified in terms
> of a receiver
> > > sensitivity threshold when, based on data presented, the
> receive eye will
> > > likely be completely closed.
> > >
> > > Even if not completely closed, there is a high likelihood that the
> > receiver
> > > min threshold will be so low that, even after EQ, you
> will not be able to
> > > accurately apply a level test to determine whether there
> is adequate
> > signal
> > > to make a decision.
> > >
> > > I believe the right approach lies in using the state
> machines provided in
> > > Clause 48 to declare a Local Fault condition.
> > >
> > > Shawn
> > >
> > > ____________________________________________
> > > Shawn Rogers, PMP s-rogers@ti.com
> > > High Speed Serial Link Marketing
> > > Texas Instruments
> > > 12500 TI Boulevard / M/S 8732/ Dallas, Texas 75243
> > > Office: 214.480.2678 Cell: 214.549.4868
> > > ______________________________________
> > >
> > >
> > > -----Original Message-----
> > > From: Petre Popescu [mailto:popescu@mail.quaketech.com]
> > > Sent: Monday, January 27, 2003 12:16 PM
> > > To: 10GBCX4
> > > Subject: [10GBASE-CX4] SIGNAL_DETECT
> > >
> > > Howard,
> > > Please find attached the text proposed for SIGNAL_DETECT.
> > > I found the most appropriate text for our applications in clause
> > > 39 (1000BASE-CX).
> > > Regards, Petre
> > >
> > > --
> > > Petre Popescu
> > > Quake Technologies
> > > 613.270.8113.x2229
> > > 613.220.8982 (cell)
>
>
>
XAUI_IPG.pdf
XAUI_IPG.xls