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

Re: [802.3_B400G_OPTX] [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

Hi Mike, Richard,

Focus: Explanation for the large TDECQ difference between Phase=0 and Phase= 24.

It was nice speaking with you during breakfast, and I looked into your concern regarding Phase=0 has TDECQ_fixed= 6.39 dB while Phase= 24 has TDECQ_fixed= 2. 70 dB which seems odd! 

Yes, I am not truncating the histograms.

The large reduction in TDECQ_fixed (Phase0= PhaseHist) from 6.39 dB at Phase 0 to.2.70 dB at Phase= 24 is mainly due to the "improved timing alignment" between the sampling phase and the channel response.
- This alignment enables the equalizer to significantly reduce residual ISI, increase effective eye opening (OMA_TDECQ) and require less DFE correction.
- Although the estimated noise is actually higher at Phase= 24, the reduction in ISI more than compensates, resulting in a much lower TDECQ.
- Comment: Phase= 24 provides a substantially cleaner equalizer eye than Phase= 0, and the TDECQ improvement is primarily ISI-driven rather than noise-driven.

How do you see?

Below are the key observations.
Parameter
Phase= 0
Phase= 24
TDECQ_fixed
6.39 dB
2.70 dB
TDECQ_search
4.06 dB
1.62 dB
Ceq
4.06 dB
1.62 dB
OMA_TDECQ
0.417
0.688
DFE
0.33
0.19
sigmaG
0.010
0.017

Below is the reasoning for Phase= 24 being better.
- At Phase= 24, the equalizer is able to concentrate much more signal energy into the target symbol and suppress residual ISI more effectively.
- Evidence of this includes:
* Equalizer gain (Ceq) increases from 1.576 to 3.615.
* Effective eye opening (OMA_TDECQ) improves from 0.417 to 0.688.
* Required DFE decrease from 0.33 to 0.19, indicating that less post-cursor cancellation is needed.
- Comment: Sampling phase at Phase= 24 is much better aligned with the dominant cursor of the channel response, resulting in a cleaner equalizer eye.

Below is the reasoning noise being not the main driver.
- Interestingly, the estimated noise actually increases, wherein Phase=0 has sigmaG= 0.010 and Phase= 24 has sigmaG= 0.017.
- If noise were the dominant factor, we would expect TDECQ to worsen but instead TDECQ improves dramatically.
- Comment: This indicates that the improvement comes primarily from:
* Reduced residual ISI.
* Better timing alignment.
* Improved equalizer solution.
* Greater effective eye opening.
rather than from any reduction in Gaussian noise.

Regards,
Hansel

From: Hansel DSilva <00004708d0869f3e-dmarc-request@xxxxxxxxxxxxxxxxx>
Sent: Tuesday, July 14, 2026 6:26 AM
To: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_B400G_OPTX] [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning
 

CAUTION: EXTERNAL EMAIL

Hi,

Apologizes for the lengthy email, I am trying my best to recognize the different positions of individuals.

Clause 180 in IEEE 802.3dj D3p1: "Two vertical histograms are measured through the eye diagram, nominally centered 0.05 UI before and after sampling phase phi0. Each of the histogram windows spans all of the modulation levels of the eye diagram, as illustrated in Figure 180-11. The precise time position of the pair of histogram windows is adjusted to minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart."

I was able to script it out at https://opensource.ieee.org/hansel.dsilva/com_code/-/tree/tdecq/Exploratory/tdecq?ref_type=heads. 

For today Tuesday, below are a few observations that may help frame the discussion between Phase0 equal to PhaseHist and Phase0 approximately equal to PhaseHist.

How do you see the below?

Below is a brief description of the setup.
1] samples per UI= 25
- Comment: Clause 180: "Each histogram window has a width of 0.04 UI."

2] Discrete phase indices: Phase= 0, 1, 2, ..., 24.

3] Equalizer optimization phase: Phase0= Phase/ samples per UI.

4] Histogram phase: PhaseHist= Phase0+ delta
Where delta is the histogram-center offset.
Under "local optimization" interpretation: -0.1UI<= delta<= +0.1UI.

5] Histogram windows are then positioned as:
Left= PhaseHist- 0.05 UI
Right= PhaseHist+ 0.05 UI
Therefore: Right- Left= 0.10 UI is always maintained.

6] Delta UI= PhaseHist- Phase0= delta
- Comment:
- It is a wrap-around (modulo-1 UI) phase difference.
- The reported phase offset should represent the minimum phase displacement between PhaseHist and Phase0 on a circular UI scale.
- Since the phase is periodic with a period of 1 UI, the offset is then wrapped into the interval: -0.5<= Delta UI<= +0.5
- Applied is the following:
Delta UI= Delta UI- 1...for Delta UI> +0.5
Delta UI= Delta UI+ 1...for Delta UI< -0.5
Delta UI= Delta UI...otherwise

7] TDECQ_fixed= TDECQ(Phase0)
- Comment. Histogram pair centered at Phase0.

8] TDECQ_search= TDECQ(Phase0, PhaseHist)
- Comment. Equalizer remains optimized at Phase0, but the histogram is allowed to move to PhaseHist.

9] Improvement in TDECQ: Delta TDECQ= TDECQ_search- TDECQ_fixed.
- Comment: Negative values indicate an improvement in TDECQ.

Below are the results.
Phase
Phase0 [UI]
PhaseHist [UI]
Delta UI [UI]
TDECQ_fixed [dB]
TDECQ_search [dB]
Delta TDECQ [dB]
0
0.000
0.925
-0.075
6.39
4.06
-2.33
1
0.040
0.080
0.040
5.91
5.07
-0.84
2
0.080
0.160
0.080
4.95
4.85
-0.10
3
0.120
0.140
0.020
4.36
4.26
-0.09
4
0.160
0.225
0.065
4.05
3.96
-0.09
5
0.200
0.220
0.020
3.90
3.82
-0.08
6
0.240
0.240
0.000
3.75
3.75
0.00
7
0.280
0.280
0.000
3.81
3.81
0.00
8
0.320
0.280
-0.040
3.84
3.84
0.00
9
0.360
0.320
-0.040
3.89
3.81
-0.08
10
0.400
0.300
-0.100
3.97
3.89
-0.08
11
0.440
0.340
-0.100
4.02
3.94
-0.09
12
0.480
0.400
-0.080
4.04
4.04
0.00
13
0.520
0.500
-0.020
3.99
3.99
0.00
14
0.560
0.640
0.080
4.08
4.02
-0.05
15
0.600
0.620
0.020
4.17
4.12
-0.06
16
0.640
0.640
0.000
4.66
4.66
0.00
17
0.680
0.780
0.100
4.87
4.80
-0.07
18
0.720
0.820
0.100
5.21
4.84
-0.37
19
0.760
0.860
0.100
5.24
4.27
-0.97
20
0.800
0.900
0.100
5.11
3.70
-1.41
21
0.840
0.925
0.085
4.77
3.37
-1.40
22
0.880
0.925
0.045
3.86
3.13
-0.73
23
0.920
0.925
0.005
3.08
2.85
-0.23
24
0.960
0.025
0.065
2.70
1.62
-1.08


Below are four observations.
1] Observation #1: Phase0 is approximately equal to PhaseHist
Phase
Phase0 [UI]
PhaseHist [UI]
Delta UI [UI]
TDECQ_fixed [dB]
TDECQ_search [dB]
Delta TDECQ [dB]
6
0.240
0.240
0.000
3.75
3.75
0.00
7
0.280
0.280
0.000
3.81
3.81
0.00
12
0.480
0.400
-0.080
4.04
4.04
0.00
16
0.640
0.640
0.000
4.66
4.66
0.00
23
0.920
0.925
0.005
3.08
2.85
-0.23
- The optimum histogram position is either equal to, or very close to Phase0.
- Moving the histogram pair provides little or no additional benefit.
- The eye is already being evaluated near its optimum sampling position.
- Comment. These cases suggest that the optimum histogram location can, in some circumstances, coincide with or remain very close to the equalizer optimization phase.

2] Observation #2: PhaseHist is offset from Phase0
Phase
Phase0 [UI]
PhaseHist [UI]
Delta UI [UI]
TDECQ_fixed [dB]
TDECQ_search [dB]
Delta TDECQ [dB]
16
0.640
0.640
0.000
4.66
4.66
0.00
17
0.680
0.780
0.100
4.87
4.80
-0.07
18
0.720
0.820
0.100
5.21
4.84
-0.37
19
0.760
0.860
0.100
5.24
4.27
-0.97
20
0.800
0.900
0.100
5.11
3.70
-1.41
21
0.840
0.925
0.085
4.77
3.37
-1.40
- Allowing the histogram center to move away from Phase0 produces a noticeable reduction in TDECQ (improves TDECQ).
- ***Important***: Optimum histogram phase does not seem to wander arbitrary over the entire UI.
=> Even for the cases showing the largest TDECQ improvements, the optimum PhaseHist remains within ***approximately*** +-0.1 UI of Phase0 after modulo-1 wrap-around is taken into account.
=> In this dataset, the maximum observed offset is approximately 0.10 UI.
- Several phases exhibit improvements approaching or exceeding 1 dB when the histogram center is shifted by approximately 0.08-0.10 UI.
- Comment: This suggests that, while PhaseHist may differ from Phase0, the optimum offset remains relatively localized around Phase0 rather than moving freely across the entire UI.

3] Observation #3: Phase0 Boundary/ Wrap-around Effect
Phase
Phase0 [UI]
PhaseHist [UI]
Delta UI [UI]
TDECQ_fixed [dB]
TDECQ_search [dB]
Delta TDECQ [dB]
0
0.000
0.925
-0.075
6.39
4.06
-2.33
- Phase 0 exhibits the largest TDECQ improvement observed in the dataset.
- Although the numerical value of PhaseHist appears significantly different from Phase0, the wrapped phase difference is only delta UI= -0.075 UI.
- Due to the periodic nature of phase, PhaseHist= 0.925 UI may be interpreted as a local phase shift ocurring 0.075 UI before Phase0= 0.000 UI.
- Comment: This observation provides a strong example that PhaseHist= Phase0 does not always yield the minimum TDECQ.
- Question. Is the observed TDECQ reduction attributable to a realistic receiver behavior near the phase boundary, or is it a consequence of introducing additional optimization freedom into the metric?

4] Observation #4: Largest improvement occurrence
Phase
Phase0 [UI]
PhaseHist [UI]
Delta UI [UI]
TDECQ_fixed [dB]
TDECQ_search [dB]
Delta TDECQ [dB]
19
0.760
0.860
0.100
5.24
4.27
-0.97
20
0.800
0.900
0.100
5.11
3.70
-1.41
21
0.840
0.925
0.085
4.77
3.37
-1.40
24
0.960
0.025
0.065
2.70
1.62
-1.08
- Interestingly, the largest TDECQ improvements occur near the later phases where the optimum histogram offset approaches the imposed search limit.
- This could be interpreted in two different ways:
A. The eye is asymmetric and a local histogram shift better captures receiver behavior
B. The TDECQ calculation is benefiting from an additional optimization degree of freedom.
- Comment: Distinguishing between these two interpretations may ultimately require agreement on the receiver's behavior being modeled.

Below are the final comments.
1] In support of Laurent's position: I am not seeing evidence from this dataset that PhaseHist= Phase0 must always hold
- Comment: Several cases show improved TDECQ when the histogram center is allowed to move away from Phase0.

2] In support of Adee/ Mike Dudek/ Norm: I am not seeing a need based on this dataset for PhaseHist to be completely independent of Phase0 over the entire unit interval.
- Comment: The optimum histogram phase appears to remain relatively localized around Phase0 rather than wandering arbitrarily across the full UI.

3] The results appear more consistent with: PhaseHist= Phase0 + delta
Where, delta is allowed to vary over a limited range while maintaining:
Left  = PhaseHist - 0.05 UI
Right = PhaseHist + 0.05 UI
Where, Right- Left= 0.10 UI is always preserved
- Comment: In the results above, the optimum offset was observed to remain withing approximately |delta|<= 0.10 UI after modulo-1 wrap-around is taken into account.

4] Question. Your simulation only examined +-0.1 UI, so how can you conclude anything about the full UI?
Answer. Fair point. The analysis only evaluated a +-0.1 UI histogram-center search range. Therefore, I am not claiming that the global optimum must lie within +-0.1 UI. Rather, I am observing that, within the search range evaluated, the optimum PhaseHist was consistently found within that local neighborhood of Phase0. Determining whether larger offsets provide additional improvement would require extending the histogram-center search beyond +-0.1 UI.

In support of the question raised by Richard, which I believe may be the key question underlying this discussion: What receiver behavior or impairment is the histogram phase adjustment intended to model? Clarifying that receiver model may ultimately determine which interpretation is most appropriate.

Below is some insight in helping start a discussion.
1] PhaseHist= approx. Phase0 with only small local adjustment
- Histogram pair represents the same timing reference used by the equalizer optimization and symbol decisions.
- The small local adjustment accounts for finite timing resolution, implementation uncertainty, interpolation error, clock recovery tolerance or other practical imperfections.
- The histogram search is not intended to represent an independent timing optimization, but rather a refinement around the receiver's selected sampling phase.
- Under this interpretation, moving the histogram center slightly allows the measurement procedure to avoid artifacts caused by discrete phase granularity while remaining representative of the actual receiver operation.
- This interpretation appears consistent with the Clause 180 wording: ""nominally centered 0.05 UI before and after sampling phase phi0" since the histogram pair remains associated with the sampling phase rather than becoming a separate optimization variable.

2] PhaseHist independent of Phase0
- The histogram pair represents a timing reference that is not necessarily tied to the equalizer optimization phase.
- The histogram search models a receiver in which monitoring, slicing, adaptation or quality-estimation functions may operate using different timing references.
- Under this interpretation, Phase0 determines where the equalizer is optimized, while PhaseHist independently determines where the noise distributions are evaluated.
- The histogram adjustment seeks the location that minimizes TDECQ regardless of where the equalizer optimization occurred.
- A sophisticated receiver architecture with independent monitoring circuitry could potentially justify this interpretation.
- However, it raises the question of whether the resulting improvement represents a realistic receiver capability or simply additional mathematical freedom withing compliance metric?

3] Adee's proposed Redacted comments L08 and L09: https://www.ieee802.org/3/dj/comments/D3p1/8023dj_D3p1_Non-BallotComments_Adee_Ran_proposed_Redacted.pdf
- The comments are stronger than supporting a relationship between PhaseHist and Phase0.
- These comments specially argue from a receiver-model perspective.

- A few interpretations of L09:
A. Real receivers optimize their equalizers based on an average sampling phase.
B. The concept of moving the histograms away from being symmetrically spaced around a sampling phase are not from the old TDECQ...
C. This additional degree of freedom increases the space on which TDECQ should be optimized...
D. ...expecting a real receiver to optimize at one phase but sample at another phase is not realistic..

Comment: One interpretation of L09 is that the Equalizer Optimization Phase, Receiver Sampling Phase, and Histogram Center Phase are fundamentally linked. Under this view, introducing an independently optimized histogram-center phase may create an additional degree of freedom beyond that expected of a minimally compliant receiver.

- A few interpretations of L08:
A. The purpose of using two separated histograms is to account for varying sampling phase in a real receiver.
B. The histogram does not model any real behavior and is only widening each histogram and increases variability of measurement results.
C. Ideally a zero width histogram could be used to capture the vertical opening at each specific phase.

Comment: L08 appears to emphasize measurement of signal quality at specific phase locations. In the limit of sufficient oversampling or interpolation, the histogram measurement approaches a specific phase point rather than relying on wider histogram windows. Under this interpretation, the role of the histogram is to characterize signal quality at a sampling phase rather than to identify a more favorable sampling phase.


4] Questions to ask ourselves and discuss among each other are the following.
- Is the histogram offset intended to model clock recovery uncertainty?
- Is it intended to model finite timing resolution of the measurement process?
- Is it intended to model phase-tracking error in the receiver?
- Is it intended to model implementation variation between the equalizer and decision circuitry?
- Is it intended to model an independently optimized timing path within a sophisticated receiver architecture?
- If PhaseHist differs significantly from Phase0 and yields a lower TDECQ, does that represent an actual receiver capability, or does it simply represent a larger optimization space within the compliance metric?
- Finally, is the purpose of TDECQ to report the mathematically lowest achievable value or to emulate a specific receiver behavior?


Regards,
Hansel D'Silva
Standards Development Engineer- Amphenol

From: Hansel DSilva <00004708d0869f3e-dmarc-request@xxxxxxxxxxxxxxxxx>
Sent: Monday, July 13, 2026 11:34 AM
To: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_B400G_OPTX] [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

CAUTION: EXTERNAL EMAIL

Hi,

I do not mean to put my hat in the ring as this is not my cup of tea.

Though I have started to code out an open-source script in calculating TDECQ under the guidance of Norm, Laurent, Roberto, Amit, Richard and a few others, which is posted at https://opensource.ieee.org/hansel.dsilva/com_code/-/tree/tdecq/Exploratory/tdecq?ref_type=heads and a final version release is pending. Reach out if interested.

Clause 180 in IEEE 802.3dj D3p1: "Two vertical histograms are measured through the eye diagram, nominally centered 0.05 UI before and after sampling phase phi0. Each of the histogram windows spans all of the modulation levels of the eye diagram, as illustrated in Figure 180-11. The precise time position of the pair of histogram windows is adjusted to
minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart."

The text clearly requires the histogram windows to remain 0.1 UI apart. The question is whether their center is constrained to Phase0 or may vary independently? Laurent's interpretation is consistent with preserving historical TDECQ behavior and receiver architectures with independent slicer timing. However, the wording "nominally centered" suggests the histogram-center search should remain local to Phase0 rather than becoming a completely independent optimization over the entire unit interval.

Therefore my interpretation is:
- It is not PhaseHist= Phase0 nor PhaseHist is free over entire UI.
- PhaseHist = approx. Phase0 with a constrained local optimization.
- For example, PhaseHist= Phase0+ delta where delta belongs to the interval from minus 0.1 UI to plus 0.1 UI.
    => Then Left = PhaseHist - 0.05 UI and Right = PhaseHist + 0.05 UI in selecting the minimum TDECQ.

How do you see it?

My reading of the discussion so far is recognizing three positions.
1] Laurent's interpretation
Phase0  = equalizer optimization phase
PhaseHist = histogram center phase
PhaseHist and Phase0 are independent
With only requirement: Left histogram  = PhaseHist - 0.05 UI and Right histogram = PhaseHist + 0.05 UI wherein Right- Left= 0.1 UI being always maintained.

Comment: 
A. In this interpretation:
1) Optimize equalizer at phase0.
2) Build eye.
3) Move histogram pair center.
4) Find minimum TDECQ.
B. This introduces another search dimension: TDECQ(phase0, histogram_center) rather than TDECQ(phase0) alone.

2] Norm/ Mike/ Adee interpretation
- The key sentence in Clause 180: "Two vertical histograms are measured through the eye diagram, nominally centered 0.05 UI before and after sampling phase phi0".
- The above implies: Histogram center= approx. Phase0 and not histogram center completely independent of phase0.
- The subsequent sentence: "The precise time position of the pair of histogram windows is adjusted to minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart"
- The above implies: small adjustment around Phase0 rather than free search over the entire UI...which is Norm's position.

3] Rich's question (most important)
- Rich asks: What receiver impairment or uncertainty is the histogram intended to model?
- This is a fundamental question.
- If phase0 already comes from the TDECQ optimization itself then: what additional physical receiver capability does PhaseHist not equal to Phase0 represent?
- Laurent's answer: independent slicer timing which can occur in a sophisticated receiver.

I will attempt to code something and share some insight, I do not have a preference whether position 1] or 2] though wish the ambiguity is cleared in helping me complete the coding of the script of TDECQ.

Regards,
Hansel D'Silva
Standards Development Engineer- Amphenol

From: Richard Mellitz <000014533bad0b9c-dmarc-request@xxxxxxxxxxxxxxxxx>
Sent: Sunday, July 12, 2026 3:44 PM
To: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_B400G_OPTX] [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning
 

CAUTION: EXTERNAL EMAIL

This is the first time you received an email from this sender (000014533bad0b9c-dmarc-request@xxxxxxxxxxxxxxxxx). Exercise caution when clicking links, opening attachments or taking further action, before validating its authenticity.

Since the measured eye already includes all transmitter jitter and phase 0 is optimized directly through the TDECQ search, what receiver impairment or uncertainty is the histogram intended to model? Without identifying the physical meaning of the histogram, it is difficult to determine whether the histogram center should be tied to phase 0 or treated as an independent optimization variable.

…Rich




Richard Mellitz, Signal Integrity (SI) Engineer

Samtec Southeast

Office: 803-908-4411

www.samtec.com
From: Mike Dudek <mtdudek@xxxxxxxxx>
Sent: Sunday, 12 July 2026 10:09:41
To: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>
Subject: Re: [802.3_B400G_OPTX] [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning
 
You don't often get email from mtdudek@xxxxxxxxx. Learn why this is important
⚠ CAUTION: This email originated from outside the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.

At lower speeds and D3.0 the reference phase was the average zero crossing.  These were the words that allowed a phase search from that zero crossing reference.   With D3.1 the phase search is explicit

Mike Dudek
C 949 201 0001
mtdudek@xxxxxxxxx
2005 Via Mango
San Clemente
CA 92673



On Sun, Jul 12, 2026, 5:18 AM Adee Ran (aran) <0000147b29386f6c-dmarc-request@xxxxxxxxxxxxxxxxx> wrote:
Hi all
Please note that two of my (late, non-ballot) comments address the histogram window and their positions.
L09 suggests removing the sentence that can be interpreted as allowing asymmetry. The text of the comment includes the rationale for using symmetrically placed histograms (matching real receiver behavior).
L08 suggests removing the nominal width of each histogram window, allowing instead some maximum width. Ideally the width should be 0 to capture a specific phase. This is possible if the signal is captured/interpolated to an integer oversampling ratio.



</Adee>

From: Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>
Date: Saturday, 11 July 2026 at 22:19
To: Mike Dudek <mdudek@xxxxxxxxxxx>; norm.swenson@xxxxxxxx <norm.swenson@xxxxxxxx>; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; Mark Nowell (mnowell) <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; Adee Ran (aran) <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx <miguangcan@xxxxxxxxxx>; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Cc: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>; STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx>
Subject: Re: [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

Hi Mike,

Can you explain then your comment #308 against draft 3.0, in which you proposed to eliminate reference to the 0.45 and 0.55 position of the zero crossing ? I interpreted this suggestion as an acknowledgement that the position of the histogram is not necessarily tied to the middle of the eye, as determined by the sampling phase selected to reconstruct the eye. 

Laurent


From: Mike Dudek <mdudek@xxxxxxxxxxx>
Sent: Saturday, July 11, 2026 2:10 PM
To: norm.swenson@xxxxxxxx <norm.swenson@xxxxxxxx>; Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx <miguangcan@xxxxxxxxxx>; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Cc: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>; STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx>
Subject: Re: [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning
 
I agree with Norm's interpretation and also have a comment to delete "nominally".
From: norm.swenson@xxxxxxxx <norm.swenson@xxxxxxxx>
Sent: Saturday, July 11, 2026 09:03
To: 'Alloin, Laurent' <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; Mike Dudek <mdudek@xxxxxxxxxxx>; 'Mahadevan, Amitkumar' <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx <miguangcan@xxxxxxxxxx>; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Cc: STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-OPTX@xxxxxxxxxxxxxxxxx>; STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx <STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx>
Subject: [EXTERNAL] RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning
 
Prioritize security for external emails:
Confirm sender and content safety before clicking links or opening attachments
 

Hi Laurent

 

Adding the reflectors to try to save time discussing this at the face-to-face meeting.

 

The sentences under discussion in D3.1 are:

“Two vertical histograms are measured through the eye diagram, nominally centered 0.05 UI before and after sampling phase \phi_0 …  The precise time position of the pair of histogram windows is adjusted to minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart."

I do agree that the second sentence was in D3.0 and I recall discussing the meaning of it with you in Munich.  I interpret that sentence a bit differently than you do.  You believe this means optimizing the taps at \phi_0 then sliding the two histogram windows across the entire \phi_0 -T/2 to \phi_0 + T/2 range to find their optimal location.  (Please restate if I have not stated your position accurately.)  I interpret those sentences to place the histogram windows, separated by 0.1UI, approximately centered at \phi_0 (from the word nominally) and make fine adjustments as necessary to optimize the symbol error rate.  I do not interpret that to mean sliding the center of the windows across the entire unit interval.

 

I am interested in how others interpret those sentences.  Again, since we are searching for the best \phi_0, I don’t think this adjustment of the histogram locations is necessary, which is why I have comments suggesting that the word “nominally” and the second sentence should be deleted.  But if others disagree, I think we should try to make the meaning of those sentences clear so that there is no ambiguity in how it should be interpreted.

 

Regards,

Norm

 

 

From: Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>
Sent: Friday, July 10, 2026 6:37 PM
To: norm.swenson@xxxxxxxx; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; 'Mike Dudek' <mdudek@xxxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Subject: Re: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

 

Yes, Norm,

 

I think that this degree of freedom has been part of the specification for a long time, and the zero-crossing reference in the figure was obsolete for some time.  I must admit it was before my time, but I see that this degree of freedom brings benefit when the transmitter has group delay distortion or when equalized eyes show some nonlinear skew, which a well-designed receiver can take advantage of. But foremost, I believe that at this point, we do not want to change the measurement results of the TDECQ specification as compared to what it yields prior to the effort of rewriting the specification for clarity, which I think was your goal in Draft D3.1. 

 

Laurent

 

 

PS: The exert of D3.0 spells out clearly that this degree of freedom exists in the prior specification.

 

 

 

 

 

From: norm.swenson@xxxxxxxx <norm.swenson@xxxxxxxx>
Sent: Friday, July 10, 2026 7:32 PM
To: Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; 'Mike Dudek' <mdudek@xxxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx <miguangcan@xxxxxxxxxx>; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Subject: RE: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

 

HI Laurent

 

Do you interpret that line to mean that you choose a \phi_0 (at which you optimize the taps), you place the windows at \phi_0 +/- .04T and measure TDECQ there, then you scan the center of those windows across the entire range of 0 to T, but keep the same \phi_0 for the tap optimization purposes?  That adds yet another parameter to search over and I am not sure that buys us anything.

 

I believe the original TDEC or TDECQ did not have that in there, as they assumed zero crossings defined 0UI and 1UI, and the histograms were placed around the midpoint between 0UI and 1UI.  So I am not so sure that that wisdom would apply here.

 

Regards,

Norm

 

 

From: Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>
Sent: Friday, July 10, 2026 3:03 PM
To: norm.swenson@xxxxxxxx; 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; 'Mike Dudek' <mdudek@xxxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Subject: Re: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

 

Hi Norm,

 

Yes, I think we need to retain this wording from 2 perspectives. First, the rewrite of the TDECQ procedure that you pursue and that I support to a certain extent should not change the outcome of the measurement at this point in the project. Secondly, a well-designed receiver may well have the capability to determine the optimum slicer time stamp independently of the CDR phase selection. I rely on the wisdom of those individuals who wrote the original TDECQ specification to know that it matters. 

 

Laurent

 

 

 

 

From: norm.swenson@xxxxxxxx <norm.swenson@xxxxxxxx>
Sent: Friday, July 10, 2026 5:08 PM
To: 'Hansel DSilva' <Hansel.D'Silva@xxxxxxxxxxxxxxxx>; Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>; 'Richard Mellitz' <richard.mellitz@xxxxxxxxxx>; 'Mike Dudek' <mdudek@xxxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; 'Roberto Rodes' <roberto.rodes@xxxxxxxxxxxx>; 'Mark Nowell (mnowell)' <mnowell@xxxxxxxxx>; 'Ali Ghiasi' <ali@xxxxxxxxxxxxxxxxx>; 'Adee Ran (aran)' <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx <miguangcan@xxxxxxxxxx>; 'Adam Gregory' <adam.gregory@xxxxxxxxxx>; 'Heck, Howard' <howard.heck@xxxxxx>; 'Samuel Kocsis' <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Subject: [**EXTERNAL**] RE: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

 

Laurent

 

Do you think we really need that line in the standard:

 

“The precise time position of the pair of histogram windows is adjusted to minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart."

 

We test at \phi_0 +/- .05T, then we scan T_0 to find the best TDECQ.  Do we need further optimization then that?  It seems that this would represent the operation of the clock recovery and then looking at the values .05 UI early and late would account for jitter.  Why would we need to do more than that?

 

Thanks,

Norm

 

 

From: Hansel DSilva <Hansel.D'Silva@xxxxxxxxxxxxxxxx>
Sent: Friday, July 10, 2026 11:05 AM
To: Alloin, Laurent <laurent.alloin@xxxxxxxxxxxxxxxxxxxxxxxx>; Norman Swenson <norm.swenson@xxxxxxxx>; Richard Mellitz <richard.mellitz@xxxxxxxxxx>; Mike Dudek <mdudek@xxxxxxxxxxx>; Mahadevan, Amitkumar <amahadev@xxxxxxxxx>; Roberto Rodes <roberto.rodes@xxxxxxxxxxxx>; Mark Nowell (mnowell) <mnowell@xxxxxxxxx>; Ali Ghiasi <ali@xxxxxxxxxxxxxxxxx>; Adee Ran (aran) <aran@xxxxxxxxx>; miguangcan@xxxxxxxxxx; Adam Gregory <adam.gregory@xxxxxxxxxx>; Heck, Howard <howard.heck@xxxxxx>; Samuel Kocsis <Samuel.Kocsis@xxxxxxxxxxxxxxxx>; 'Ahmad El-Chayeb' <ahmad.el-chayeb@xxxxxxxxxxxx>
Subject: IEEE 802.3 TDECQ: Clarification on Histogram Window positioning

 

Hi Laurent,

 

Thank you for the feedback you provided on the presentation shared in the email titled "IEEE 802.3dj: Script in calculating TDECQ".

 

I would like to clarify my understanding of the following comment by you:

   "You have to make sure that the histograms are 0.1 UI apart, and not necessarily tied to phase0UI = phase0 / sps."

 

I am trying to reconcile this comment with the draft text in Clause 180, which states

"Two vertical histograms are measured through the eye diagram, nominally centered 0.05 UI before and after sampling phase 0. Each of the histogram windows spans all of the modulation levels of the eye diagram, as illustrated in Figure 180-11. The precise time position of the pair of histogram windows is adjusted to minimize TDECQ while keeping the histogram windows spaced 0.1 UI apart."

 

When you mention that the histograms should not necessarily be tied to phase0/samples_per_ui, are you suggesting that the histogram-pair center should be treated as an additional optimization variable that is independent of the sampling phase phase0?

 

Wherein, instead of forcing: left= phase0- 0.05UI and right= phase0+ 0.05UI it should be left= phase_hist- 0.05 UI and right= phase_hist+ 0.05 UI where phase_hist may differ from phase0 and is adjusted to minimize TDECQ while maintaining the required 0.1 UI sepetation?

 

In my current implementation, the histogram locations are determined in extract_historgrams():

https://opensource.ieee.org/hansel.dsilva/com_code/-/blob/tdecq/Exploratory/tdecq/src/extract_histograms.m?ref_type=heads [opensource.ieee.org]

 

phase0UI= phase0/ samples_per_ui

leftCenter  = mod(phase0UI - 0.05,1);

rightCenter = mod(phase0UI + 0.05,1);

 

Background,

- For every phase: phase0= 0, 1, 2, ..., samples_per_ui-1

- Then phase0UI= phase0/ samples_per_ui

- Then leftCenter= mod(phase0UI - 0.05,1) and rightCenter= mod(phase0UI + 0.05,1)

- Hence, the histogram pair moves around the eye as phase0 changes.

- For example, samples per UI= 8

phase0= 0; phase0UI= 0.000; left= 0.950; right= 0.050.

phase0= 1; phase0UI= 0.125; left= 0.075; right= 0.175.

phase0= 2; phase0UI= 0.250; left= 0.200; right= 0.300.

phase0= 3; phase0UI= 0.375; left= 0.325; right= 0.425.

phase0= 4; phase0UI= 0.500; left= 0.450; right= 0.550.

phase0= 5; phase0UI= 0.625; left= 0.575; right= 0.675.

phase0= 6; phase0UI= 0.750; left= 0.700; right= 0.800.

phase0= 7; phase0UI= 0.875; left= 0.825; right= 0.925.

Comment. The histogram windows are therefore phase-dependent, but their sepeartion is always 0.1 UI.

 

Regards,

Hansel D'Silva

 

To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1

To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1


To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1

To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1

To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1

To unsubscribe from the STDS-802-3-B400G-OPTX list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-OPTX&A=1