| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
All – I wanted to update you on some offline questions and answers I’ve been having with Tingting on modeling of noise. I will remind you that we chose to use an additive white gaussian noise model for the noise, and we have had discussion about adding impulse noise to this. As I have said before, for building consensus, showing common models, or showing
common models as well as relationships to other models, is going to make it clearer for the group to compare results.
As I have explained to the group, there are a large number of ways to model noise – none are necessarily more ‘accurate’ than others – they represent different cases and characteristics. The model being used has been substantially simplified
for the purposes of repeatability. AWGN models for these types of channels are rarely, fully accurate, but are useful proxies in comparison. We could use various scalings of PSD level, changes in the spectral shape of the underlying noise, and changing of
the rolloff of the disturbing source, averaging bandwidth of the noise, linear averaging vs. geometric (dB) averaging of the noise to produce levels for simplified sources, and passing PSDs of the noise, and even actual signals through modelled or measured
coupling functions. All of these have merit – none are necessarily more “accurate” as the actual noise will vary from scenario to scenario. Simply put, noise in these physical systems is not terribly repeatable. In my experience, variations of SNR are not
uncommon up to 2 dB. Usually the crosstalk noise levels observed are better (less) than would be obtained from our ANEXT & AFEXT coupling limits – for several reasons – including that we generally pick our limits biased towards worst-case, not ‘typical’ case,
and the limit line approach would be touched at a single frequency, generally not met across the frequency band. In the modeling that I presented, I used the 100BASE-T1L AFEXT-only case– not the mixed crosstalk case which I also described, as we discussed. In this case you don’t add in 10BASE-T1L and you do not decrease the disturber PSD by 3 dB.
AND, the disturber is only attenuated by a short (100m) line. The total power is averaged linearly over a 60 MHz bandwidth. This result is approximated by a -113 dBm/Hz AWGN source for a 75 Mbaud signal with a scaled version of the 10BASE-T1L frequency mask.
To represent the higher in-band PSD seen with lower baud signals (and lower with higher baud), I have recommended a simple adjustment of 10log10(fbaud/75). Because this adjustment is a simple dB shift, and is small, it can be left out of simulations – it is, in my opinion, fairly easy for individual experts to apply the correction to the results themselves. I notice that the results presented
by Brian Murray at the last meeting used a constant level (7Vrms = -113 dBm/Hz) noise source. There are many ways to ‘improve’ this model – those that result in small shifts, can be treated similarly, those that use spectrally shaped noise are harder to compare and would need to be presented and justified. Simple shifts include
averaging over different bandwidths, varying the bandwidth that the noise is averaged, or varying the power level. Generally, varying the out of band rolloff (for 100BASE-T1L) shouldn’t be a substantial effect as this could (and probably should) be excluded
in an implementation by receiver filtering. For those looking at the mixed crosstalk models, Tingting had asked about the sharp attenuation at 20 MHz for 10BASE-T1L in the figure on slide 11 of
https://www.ieee802.org/3/dg/public/May_2022/zimmerman_3dgah_01b_01292024.pdf, This is because the PSD for 10BASE-T1L is not defined beyond 20 MHz. By that point it has already rolled off below where it would be a factor in the calculation, and, real
PSDs would be far below the -70dBm/Hz level of the upper PSD mask… hence it doesn’t make sense to consider it a wideband source at that point. Again, I would suggest you run results consistent with what others are doing/have done and anlso with what you think is right - that way we can calibrate results. If you anre going to another noise model, I would suggest posting it to
the reflector - that way we move to consensus. The actual transmit and disturber pads will be different yet, especially near the band edge and roll off region. We now have some presented results. I would suggest starting with those noise scenarios for comparison. I’ve been through the process of committees trying to reach consensus results and they are never perfect. Different simulation effects, different receiver models considered, different transmit psds both for disturbers and victims, as well
as external noise models tend to limit matching of results. From what I’ve seen if we get agreement and some understanding of differences within about 1db, that will be very good. Finer matching is usually not reflective of general principles, but more about
how the simulation is constructed. Matching worse than 2db (differences of more than 2db between individual presenters) is usually a suggestion that there is some factor they are not aligned on which needs to be resolved. Between 1 and 2 db is a judgement
call…. Again, these are my personal experiences having based on how this kind of work in standards group has gone in various projects both in ieee 802 and other groups (dsl from the 1990s). As a result, starting from a common basis before making changes is important, and often the more important answers are not the absolute margins but the relative results of each evaluator. I hope this helps you all as we move toward making some decisions in Denver. George Zimmerman, Ph.D. President & Principal CME Consulting, Inc. Experts in Advanced PHYsical Communications 310-920-3860 To unsubscribe from the STDS-802-3-SPEP2P list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-SPEP2P&A=1 |