| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
For the averaging method, based on discussions this does not necessarily address the variation due to the CRU tracking of the test pattern. It may result in the measured EOJ value being lower than EOJ. This should not be an option.
[[Adee]] I don’t have a strong opinion on this one, but I don’t see why it should reduce measurement accuracy. Averaging is a good way to reduce measurement noise. I’ll check again where the consensus is.
Hi Matt
Thanks for the feedback. Please see some comments inline.
</Adee>
From: Matt Brown <mbrown8023@xxxxxxxxx>
Sent: Thursday, October 29, 2020 18:02
To: Ran, Adee <adee.ran@xxxxxxxxx>
Cc: STDS-802-3-100GEL@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_100GEL] Resolving Even-odd jitter comments
Hi Adee,
Thanks for putting this proposal together.
Here are my thoughts regarding a portion of the proposal I've repeated here:
Even-odd jitter is calculated using the measurement method specified in120D.3.1.8.2 with the following exceptions:
- The PRBS13Q pattern may be substituted with a shorter odd-length pattern that includes the 12 possible transitions between two different PAM4 symbols if it improves the measurement precision.
- The corner frequency of the clock recovery unit (CRU) may be set lower than 4 MHz if it improves the measurement precision.
- Even-odd jitter EOJ is the average of the 12 measurements defined in 120D.3.1.8.2 instead of the maximum.
Given the choices and non-specific allowances, some opening text should be provided to indicate why this list of the options is being proposed.
[[Adee]] Good idea, but I don’t have good text at this time. Since we are trying to build consensus in a short period (actually meeting today) we may want to defer that to a later draft. Until then, there is record of our recent discussions.
For the pattern method, I think that at least one particular pattern should be specified. The reason being is that this would ensure that both device and test equipment are compatible with at least one pattern. There could be a statement that any other pattern with particular characteristics could be used. In addition, there has been discussion that the pattern should cover all transitions, so that might have to be specified for the "other patterns". [[Adee]] This is included in the proposal above: “shorter odd-length pattern that includes the 12 possible transitions…” This alternate pattern would presumably be chosen such that the CRU does not track the test pattern.
[[Adee]] Practically any shorter pattern would satisfy that. I think the opening text can cover the reasoning.
For the corner frequency method, more guidance should be provided as to what the lower bandwidth should achieve. So either specify a particular frequency or provide guidance such as "lower bandwidth so that the CRU does not track the test pattern". Perhaps a particular bandwidth can be recommended for the PRBS13Q pattern.
[[Adee]] Good idea. I will try to build consensus over that.
For the averaging method, based on discussions this does not necessarily address the variation due to the CRU tracking of the test pattern. It may result in the measured EOJ value being lower than EOJ. This should not be an option.
[[Adee]] I don’t have a strong opinion on this one, but I don’t see why it should reduce measurement accuracy. Averaging is a good way to reduce measurement noise. I’ll check again where the consensus is.
Cheers!
Matt Brown
Huawei Technologies Canada
Chief Editor, IEEE 802.3ck Task Force
On Sun, Oct 25, 2020 at 5:26 AM Ran, Adee <adee.ran@xxxxxxxxx> wrote:
Hi all
I’m sending this over the reflector to include people who have not been part of the discussion so far but may be interested.
We should be resolving EOJ related comments (48, 186, 189, 52, 187, 188, 127, 190) hopefully this week. We have seen presentations calvin_3ck_01_1020 and ran_3ck_01_1020 and had a discussion and some straw polls.
The straw polls results indicate support to the following ideas
- Increasing the EOJ limit to 0.025 UI
- Specifying or allowing a lower CRU bandwidth
- Use the average of the 12 transition measurements rather than the worst case
The suggestion to use PRBS9Q or another pattern suitable for EOJ measurement (which would be the best technique to overcome the interaction between CRU bandwidth, pattern length, and other sources of jitter) had low support in the straw polls.
The combination of the 3 items above does not seem to provide a good enough solution:
- Taking the average of 12 measurements instead of worst one may be a good idea by itself, but it does not solve the main issue – the interaction above can affect all measurements in a similar way and results can still have the same repeatability issues even with this change.
- Using lower CRU bandwidth may work in some cases, but may close the eye (due to untracked jitter) and prevent accurate measurement in other cases.
- Increasing the limit to 0.025 UI may not be enough if the variation between measurements is large, as shown in some cases.
I think using a shorter pattern is the key component required for solving this issue. Results in calvin_3ck_01_1020 (backup slides) show that measurements with “PRBS9Q” (although not currently defined in the standard) are more consistent than with any other method. PRBS13Q does not provide this consistency (for reasons that were explained) and there is nothing special about this pattern that requires us to keep using it.
Allowing shorter patterns (with adequate criteria) without specifying a single pattern, as in the text below, would be a relatively easy change and a good solution. It may be seem that allowing more freedom in measurements opens the door to inconsistency in results, but all measurements have a mathematical basis that should work regardless of the pattern – if the specific pattern affects the measurement it is usually a bad thing (as in this case). We may choose to add a specific pattern in the future if there is a compelling reason.
I would suggest the following as a response to the comments above:
- Change text in 162.9.3.3 as follows
162.9.3.3 Output jitter
Output jitter is characterized by three parameters, J3u, JRMS, and even-odd jitter. These parameters are calculated from measurements with a single transmit equalizer setting to compensate for the loss of the transmitter package and host channel. The equalizer setting is chosen to minimize any or all of the jitter parameters.
J3u and JRMS are calculated using the measurement method specified in 120D.3.1.8.1. J3u is defined as the time interval that includes all but 10–3 of fJ(t), from the 0.05th to the 99.95th percentile of fJ(t).
Even-odd jitter is calculated using the measurement method specified in 120D.3.1.8.2 with the following exceptions:
- The PRBS13Q pattern may be substituted with a shorter odd-length pattern that includes the 12 possible transitions between two different PAM4 symbols if it improves the measurement precision.
- The corner frequency of the clock recovery unit (CRU) may be set lower than 4 MHz if it improves the measurement precision.
- Even-odd jitter EOJ is the average of the 12 measurements defined in 120D.3.1.8.2 instead of the maximum.
- In 120F.3.1.3, change the cross-reference for EOJ measurement from 120D.3.1.8.2 to 162.9.3.3 to incorporate the exceptions above.
- In Table 162–10, Table 163–5, and Table 120F–1, change the value for “Even-odd jitter, pk-pk” from 0.019 to 0.025.
Discussion prior to the teleconference would be welcome.
Best regards
Adee
To unsubscribe from the STDS-802-3-100GEL list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100GEL&A=1
To unsubscribe from the STDS-802-3-100GEL list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-100GEL&A=1