Re: [802.3_EPOC] Early comments on dai_01_1112.pdf
Dear Christian,
It would probably help further discussion, if the use of "we" in the
statements made by an individual could be avoided. I tend to read that as a
royal "we" and it sets me back instantly. We are also all individuals here
and express individual opinions, rather than represent group wisdom .
Regarding the first point you're making, which I assume picks on the part
that I directly contributed to this deck. To simplify implementations, in
the past projects that I participated in, people were trying to come up with
the ways to limit the number of different clock rates required to implement
the given PHY. If all the data rates are multiples of a single base clock,
implementation becomes simpler. This was the case for 10G-EPON, and drove
the selection of the specific FEC code, among the others.
The point that I was trying to make is that with MMP approach (at least the
way I understand this could work), data rates at some of the interfaces will
be varying widely from packet to packet due to different capacity of the
channel over which they will be transmitted. I am not talking about some PHY
clock (which I do not know what it is, really . I would expect to have
312.5MHz+/-100ppm clock as basis for our system, given that this is the rate
at which XGMII makes data transfers; whether that is the PHY clock, I do not
know - you seem to create new terms as you go) and I am not sure what
buffering and demupltiplexing has to do in here. You have to be able to push
out frames across PMA into PMD at some speed for one profile and another
speed for another profile, to make sure that they can be transmitted
correctly within the PMD (perhaps at the fixed PHY clock you referred to).
Different data rates at PMA = different base clocks, hence the clock
variability I am referring to within the slides.
Also, I am not arguing for any additional clock signals (not sure from where
and to where such signals would go). All I am saying is that with the
increase in different data rates within the PHY, an implementer will have to
synthesize more clock rates from the base clock. Effectively, I am arguing
in favour of making the implementation simpler for any chip designer,
something you should appreciate I believe.
Regards
Marek
From: Pietsch, Christian [mailto:cpietsch@xxxxxxxxxxxxxxxx]
Sent: Tuesday, November 13, 2012 18:55
To: STDS-802-3-EPOC@xxxxxxxxxxxxxxxxx
Subject: [802.3_EPOC] Early comments on dai_01_1112.pdf
Hi Eugene and Marek,
Concerning your joint presentation, we would like to point out a few things
in advanced. We would appreciate if you could clearly explain or justify
your assumptions during the presentation. We believe that your conclusions
are based on technical incorrect assumptions:
- Multiple modulation profiles (MMP) do not require additional
clock signals. The PHY layer clock rate is related to the sampling clock of
the analog-digital converter and the digital-analog converter. This clock
does not change with the order of the modulation scheme. MMPs would be
implemented by (de)multiplexing and buffering, which is also required for a
single modulation profile (SMP).
- Your proposal of stronger FEC for symbols with higher modulation
order seems to be ill-posed. If bit loading is applied, you would encode all
bits with the same rate. Actually, you use bit loading in order to use the
same FEC for all bits and still be efficient.
- SNR measurements for a single carrier transmission scheme (6/8
MHz channels) actually provide a good indication of what gains could be seen
for MMP. Such measurements average the SNR over 6/8 MHz. With a finer
bandwidth resolution for SNR measurements, the benefits of MMP will
increase. Meanwhile, there have been several presentations pointing out that
there are gains.
- The different in-home configurations that you describe actually
motivate the use of MMP because SNR will vary much more for different users
in the same plant. Hence, this is the best example where you need MMP to
make EPoC work. It will be difficult to mandate the use of only 1x2
splitters in-home. There will be always users disobeying. Also, replacing
all AM fiber would be very costly and not desired by many MSOs.
- Your proposal in slide 16 is not a solution of the problem you
posed. You propose to do MMP for different frequencies, but the problem you
described motivates MMP for different users.
- End users do benefit from MMP in one way or the other. RF
bandwidth is limited and the subscribers (end users) will have to share a
certain amount of RF bandwidth. With MMP, the aggregated data rate of all
end users is higher. Hence, each user gets a larger share of data rate
(unless the system is not fully loaded and all the end users are limited by
the maximal rate of their subscription). Alternatively, an MSO could
increase the number of users in a plant.
- The choice of MMP is automated based on SNR and the allowed
profiles. The PHY will automatically make sure that it operated reliably at
the highest rate supported. We believe it is desired by (most/all?) MSOs to
avoid manual monitoring and configuration, which would add to the
operational costs.
- Multicast and broadcast could be done with a profile common to
the worst user. There is no need to duplicate data or rescaling of profiles
in addition to what needs to be done for MMPs anyways.
- It seems that only your universal modulation profile refers to
what is commonly referred to as single modulation profile for a plant. A CLT
broadcasts into multiple segments of the cable plant, i.e. you have the same
DS signal on all segments. Your single modulation profile requires all the
features your MMP scheme needs.
Thanks,
Christian
_____
________________________________________________________________________
To unsubscribe from the STDS-802-3-EPOC list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-EPOC&A=1