Professional Documents
Culture Documents
Adam Healey
Broadcom Inc.
January 2020 (r1)
Introduction
• Effective return loss (ERL) has been a topic of recent discussions and a
number of initial Task Force review comments
• This presentation takes a closer look at the underlying concepts of ERL
• It clarifies the role of ERL in the compliance methodology for backplane
PHYS and chip-to-chip interfaces
• The impact of test fixtures is also considered
TP0a TP5a
Rx-Tx re-reflection
Tx re-reflection
Rx re-reflection
(𝑟) (𝑡)
• Constraints on 𝑠11 , 𝑠22 , 𝑠11 , and 𝑠22 , e.g., ERL, are imposed to limit the re-reflection terms
𝑠11 𝑠12
(𝑡) (𝑡) (𝑟) (𝑟)
𝑠21 𝑠22 𝑠21 𝑠22 𝑠11 𝑠21
FFE Tx Channel Rx CTLE DFE
Stimulate the channel with a pulse with height 𝑨𝒗 and width 𝑻𝒃 (unit interval)
𝐻21
𝐻𝑓𝑓𝑒 𝐻𝑡 𝐻𝑟 𝐻𝑐𝑡𝑓
FFE Tx Channel Rx CTLE DFE
-5
Thru Channel information
Rx re-refl.
-10
Tx re-refl.
• “Heck2” as defined in heck_3ck_01b_0719.pdf
-15 Rx-Tx re-refl. • Loss is approximately 15.2 dB
Combined terms
-20 • Effective return loss is approximately 13.6 dB
Assembled link
-25 for port 1, 12.7 dB for port 2
-30 • Case 1 package model (𝑧𝑝 = 12 mm for Tx, Rx)
Magnitude, dB
-35
• Linear equalization determined using COM
-40
-45
-50
-55
-60
-65
-70
-75
-80
0 5 10 15 20 25 30 35 40 45 50
Frequency, GHz
0% Observations
-20% • Math works! Combination of terms agrees with
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Thru Time, UI Tx
Rx re-refl. at 53.125
re-refl. GBd Rx-Tx re-refl.
the transfer function of the assembled link
5.0% • Only Rx and Tx re-reflection terms need to be
4.0% considered (except, perhaps, for very low-loss
3.0%
channels)
Normalized amplitude
2.0%
Re-reflections add to 𝒔𝟐𝟏 product
1.0% • Re-reflection terms constructively/destructively
(𝑡) (𝑟)
0.0% combine with 𝑠21 product (𝑠21 𝑠21 𝑠21 ) and each
-1.0%
other
-2.0%
-3.0%
Tx and Rx re-reflections cancel • The 𝑠21 product includes a significant portion of
-4.0% the “tail” information
-5.0%
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time, UI at 53.125 GBd
ERL calculation begins with calculation of the pulse time-domain reflectometry (PTDR) response
• Response of the filtered reflection coefficient to a pulse with height 1 and width 𝑇𝑏
(𝑟) (𝑟)
• 𝑝𝑡𝑑𝑟 = ℱ −1 𝑇𝑏 𝑠𝑖𝑛𝑐 𝑓𝑇𝑏 𝐻𝑡 𝑠11 𝐻𝑟 where ℱ −1 𝑥 is the inverse Fourier transform of 𝑥
“Weighting”
• 𝐺𝑟𝑟 (re-reflection) is a function of the minimum effective return loss allowed for connected channel, Tx, or Rx
• 𝐺𝑙𝑜𝑠𝑠 is not a significant factor (its value is close to 1)
• Significant discontinuity at 𝑁𝑏𝑥 + 1 unit intervals from 𝑇𝑓𝑥 (related to the expected Rx equalization capability)
1.1
1.0
0.9
0.8
0.7
Weight
0.6
0.5
0.4 G_rr
0.3
0.2 G_loss
0.1 G_rr * G_loss
0.0
-4 -2 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
t − T_fx, UI at 53.125 GBd
0.150
0.100
0.050 • 𝑡 = 0 is the round-trip test fixture delay 𝑇𝑓𝑥
0.000
-0.050 • Weighted PTDR (a.k.a. 𝑅𝑒𝑓𝑓 ) is also shown
-0.100
-0.150
-0.200
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time, UI at 53.125 GBd
Rx re-refl.
0.25%
Re-reflection impulse response
Impulse response
0.20%
0.15%
0.10% • Normalized to pulse amplitude at 𝑡𝑠
0.05%
0.00%
-0.05%
-0.10%
-0.15%
-0.20%
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time,RxUIinterf.
at 53.125 GBd
1.50%
1.00% • Convolution of PTDR response (not weighted)
0.50% (𝑟)
0.00%
and re-reflection impulse response ℎ𝑟𝑟
-0.50%
-1.00%
• This is what the receiver would actually “see”
-1.50%
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time, UI at 53.125 GBd
0.150 𝑵𝒃𝒙 + 𝟏
0.100
0.050
• Weighting implies the PTDR response for time
0.000 [0, 𝑁𝑏𝑥 + 1) UI is not as impactful
-0.050
-0.100 Convolves with… • Weighting implies the impact gets progressively
-0.150
-0.200 smaller as 𝑡 approaches 0
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time, UI at 53.125 GBd
Rx re-refl.
• But the receiver “sees” the convolution of the
0.25%
PTDR response and the re-reflection impulse
Impulse response
0.20%
0.15%
response
0.10%
0.05% • This “low impact” region of the PTDR response
0.00%
-0.05% can generate interference terms later in time
-0.10%
-0.15% • Some of these terms can be beyond the reach
-0.20%
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 of68the 72
assumed
76 80 DFE
Time,RxUIinterf.
at 53.125 GBd
To yield… • The interaction between the Rx (or Tx) and the
2.00%
channel is not fully described by the weighting
Normalized amplitude
1.50%
1.00% function (and the 𝜌𝑥 parameter)
0.50%
0.00%
-0.50%
-1.00%
-1.50%
-4 0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80
Time, UI at 53.125 GBd
18.0
~4 dB drop in ERL and calculate ERL
16.0
14.0 • Multiple values of 𝑁𝑏𝑥
12.0
10.0
10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40
Package trace length (z_p), mm
T2 H2 T1 M1 K1 K3 K5 H3
7.00 COM
6.00 • Computed using parameter
COM, dB
1.00
0.80
zero
0.60
0.40 • Measure improvement in
0.20 COM (the equalizers are
0.00
10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 re-optimized)
Package trace length (z_p), mm
20.0
There is no clear justification for sharp decrease in ERL
when Tx or Rx reflections are later than 𝑵𝒃𝒙 + 𝟏 UI
ERL, dB
18.0
12.0
15 20 25 29 35 40 Rx pkg. length, mm
• May be some relationship between re-reflection impulse
1.6
response and channel PTDR response 1.4
• But channel ERL does not appear to be sensitive to 𝑁𝑏𝑥 1.2
1.0
ΔCOM, dB
0.8
K3
K5
0.6
H3
0.4 T2, H2, T1
M1
0.2
0.0 K1
-0.2
10.0 12.0 14.0 16.0 18.0 20.0 22.0 24.0
Channel ERL, dB
IEEE P802.3ck Task Force, January 2020 (r1) 13
Things to consider for ERL, part 1
• ERL limits the additional re-reflection interference that the Rx needs to
tolerate
• It captures the difference between component behavior measured with
reference terminations and the expected behavior in an assembled link
• It ignores features that will not matter considering the capabilities of the
reference equalizer
• But is it correctly discriminating between what matters from does not?
Channel
𝑠11 𝑠12
𝑠21 𝑠22
TP5a
Receiver effective return loss
(𝑟)
• We want to know 𝑠11 but we can only measure Γ5𝑎
Fixture Rx CTLE DFE
(𝑓) (𝑟) (𝑓) (𝑟) (𝑟)
(𝑓) 𝑠21 𝑠11 𝑠12 𝑠11
(𝑓) (𝑓)
𝑠12 𝑠11 𝑠21
Γ5𝑎 = 𝑠11 + (𝑓) (𝑟) Γ5𝑎
1 − 𝑠22 𝑠11 𝑠21
(𝑓) (𝑓)
𝑠22
(𝑓) (𝑟) (𝑓) (𝑓) (𝑓) (𝑟)
Γ5𝑎 ≅ 𝑠11 + 𝑠11 𝑠21 𝑠21 1 + 𝑠22 𝑠11