Characterization of PCB
Insertion Loss with a New
Calibration Method
Jia Gongxian
Zhu Wenxue Huawei Technologies
Long Faming
Cheng Ning
Keysight Technologies
Mike Resso
Agenda
• Background - The Importance of characterizing PCB
insertion loss
• Traditional methods for characterizing PCB insertion loss
• New Characterization method – 1xAFR
• Comparisons and validations of different methods – With simulation
– With measurements (different layers, reference impedance, stub or backdrill)
• Considerations on the gating range of the 1xAFR method
• Conclusion
Determination of PCB unit length insertion loss is critical for:
• Material electrical parameters extraction
• PCB electrical performance estimation
• Evaluation of the passive channel design
• PCB material selection
• Evaluation of the PCB fabrication quality, etc
Traditional Characterization Methods
Direct Loss Subtraction TRL Calibration Automatic Fixture Removal
With 2xThru
Direct Loss Subtraction
Two differential transmission
lines with different lengths on
the PCB
Subtracting the insertion loss of the two
differential lines, resulting in the insertion
loss of PCB trace with length of ΔL.
Limitations:
• The launches of the two lines must be
consistent
• The launches must have good match
Any mismatch in the launches will become residual error after
subtraction and will be transferred to the insertion loss measurement of
the PCB trace, causing ripples on the insertion loss characterization
Removing launch effects with TRL Calibration
TRL Calibration Error Model
A typical TRL calibration kit on PCB
Concept:
• TRL cal kit is fabricated with the same
characteristics of the launches
• Total 10 measurements are made to calculate the 8
error terms
• Error correction is applied to remove the launch
parts, resulting only the PCB trace S-parameters
Limitations:
• It needs much experience in TRL cal kit design,
fabrication and verification
• The phase difference between Thru and lines must be
within (20 ~ 160) + N * 180 degrees in the frequency
range used
• The launches of the cal standards must be consistent
• The Thru and lines should have small impedance
variations
• Cal kit model needs to be characterized for calibration
Automatic Fixture Removal with 2xThru
Two differential transmission
lines with different lengths on
the PCB
• The launches are fully extracted from a shorter
2xThru, then de-embedded from the longer
differential line, resulting in insertion loss of PCB
trace with length of ΔL.
• Based on time domain gating and signal flow
diagram calculations.
• Same accuracy with TRL calibration, more accurate
than Direct Subtraction.
Limitations:
• The launches of the 2xThru must be consistent with
that of the longer line.
• The 2xThru and longer line need to be symmetrical
top to bottom
• The return loss and insertion loss of the 2xThru
cannot cross each other in the measurement
frequency range, often at least 5 dB separation is
required
Separation between RL and IL of the 2xThru
Proposed New PCB Loss Characterization
Method using 1xAFR Technique
PCB Loss Characterization with 1xAFR
Two differential Open
standards with different
lengths on the PCB
• Extracting the 4-port S-parameters of the two Opens
with 1xAFR, which is also based on Time Domain
Gating and Signal Flow Diagram calculations
• De-embed the 4-port S-parameters of the shorter
Open from the longer Open, resulting in the S-
parameters of the PCB trace with length of ΔL.
• Same accuracy with TRL calibration and 2xThru AFR,
but save more PCB area and measurement time.
Limitations:
• The launches of the two Opens must be consistent.
• The impedance variations of the PCB trace must be
as small as possible, especially when getting close to
the Open response
• The gating range needs to be selected carefully to
achieve good accuracy (will be discussed later)
Comparisons and Validations with
ADS Simulation
Simulation Procedure
• Create structures of PCB trace
without launch, launches,
differential lines, differential Opens
with ADS
• Simulate the S-parameters of all
these structures separately
• Characterize the PCB trace loss
with the above methods and
compare to the actual data
• 100 Ohm and 90 Ohm Z0 cases
are studied independently
Structure Line length (mil)
PCB trace length without
launches
11000
Launch 150
2xThrough AFR Longer line
(Direct Subtraction longer line)
11300
2xThrough AFR Shorter line
(Direct Subtraction shorter line)
1300
1xAFR Longer Open 11150
1xAFR Shorter Open 1150
Dielectric Constant: 3.85
Simulation Schematic of the Differential Open
• The S-parameters of the
Differential Open standard can
be simulated with the schematic
on the left
• S-parameters of other structures
can be simulated similarly
Overcome the Gating Effects
The intended measurement
bandwidth is 20 GHz, but
AFR may have error at high
frequency due to gating
effects.
Simulate/measure to higher
frequencies to overcome the
gating effects, in this case
study, the simulation is up to
25 GHz.
Direct Subtraction Result (100 Ohm Z0)
The maximum difference
between actual data and
Direct Loss Subtraction is
0.2216 dB at 19.99 GHz (ΔL
= 11 inches)
Ripples at high frequency
due to residual mismatch
error
2xThru AFR Result (100 Ohm Z0)
The maximum difference
between actual data and
2xThrough AFR is 0.0012 dB
at 810 MHz (ΔL = 11 inches)
1xAFR AFR Result (100 Ohm Z0)
The maximum difference
between actual data and
1xAFR is 0.0848 dB at 10
MHz (ΔL = 11 inches)
Direct Subtraction Result (90 Ohm Z0)
The maximum difference
between actual data and
Direct Loss Subtraction is
0.193 dB at 17.55 GHz (ΔL =
11 inches)
Ripples at high frequency
due to residual mismatch
error
2xThru AFR Result (90 Ohm Z0)
The maximum difference
between actual data and
2xThrough AFR is 0.0561 dB
at 19.98 GHz (ΔL = 11
inches)
1xAFR AFR Result (90 Ohm Z0)
The maximum difference
between actual data and
1xAFR is 0.08 dB at 10 MHz
(ΔL = 11 inches)
Summary of Simulation Results
• The 2xThru AFR and 1xAFR
are both very close to the
actual data, either with 100
Ohm or 90 Ohm Z0
• The Direct Subtraction
shows more error at high
frequency due to residual
mismatch
Z0 Characterization
Method
Error (dB/inch)
100 Ohm Direct Subtraction 0.022
2xThru AFR 0.0001
1x AFR 0.0008
90 Ohm Direct Subtraction 0.02
2xThru AFR 0.0056
1x AFR 0.008
Comparisons and Validations with
Measurement
Comparison outlines
For 100 Ohm and 90 Ohm Z0, compare the
performance of above methods in following
cases:
Comparisons
Layer 4
Stub
Backdrill
Layer 6
Cross-
comparis
on of
1xAFR
8 layer PCB
Measurement Structures
Dielectric Constant: 3.9 at 5 GHz
Structure Line length (mil)
PCB trace length without
launches
10000
TRL Open 500
TRL thru 1000
TRL Line1 3480
TRL Line2 1496
TRL Line3 1099
TRL DUT 11000
2xAFR Longer line (Direct
Subtraction longer line)
12327
2xAFR Shorter line (Direct
Subtraction shorter line)
2327
1xAFR Longer Open 11400
1xAFR Shorter Open 1400
8-layer PCB layout and fabricated PCB
(coaxial connectors are not mounted yet)
Measurement results after TRL
calibration is treated as reference for
comparisons
100 Ohm, Layer 4, stub
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.055
2xThru AFR 0.024
1xAFR 0.023
Even some deviation in TRL here
The big deviation in TRL
should be due to cal kit
fabrication error
100 Ohm, Layer 4, backdrill
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.06
2xThru AFR 0.016
1xAFR 0.02
100 Ohm, Layer 6
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.044
2xThru AFR 0.027
1xAFR 0.026
Cross comparisons
Very repeatable results with
1xAFR on different layers,
either Stub or backdrill
90 Ohm, Layer 4, stub
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.068
2xThru AFR 0.029
1xAFR 0.039
90 Ohm, Layer 4, backdrill
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.057
2xThru AFR 0.034
1xAFR 0.037
TRL has big deviation in this range
90 Ohm, Layer 6
Error compared to TRL:
Char
method
Error
(dB/inch)
Direct
Subtraction
0.039
2xThru AFR 0.026
1xAFR 0.021
Cross comparisons
Difference < 0.015 dB/inch
with 1xAFR on different
layers, either Stub or
backdrill
Measurement Comparison Summary
• 2xThru AFR and 1xAFR are both very close to TRL calibration result
• Direct Loss Subtraction result shows ripples at high frequency due to
mismatch effects
• TRL calibration result may have big deviation, too, due to TRL cal kit
fabrication quality
• The new 1xAFR works for both 100 Ohm and 90 Ohm Z0
• The 1xAFR results are very repeatable on different PCB layers, either
with Stub or backdrill
Considerations on the Gating Range of
1xAFR Method
If the gating range is too narrow
• In the 1xAFR method,
bandpass time domain
gating is used for the
extraction of the fixture
insertion loss from the Open
standard.
• If the gating range is too
close to the Open response,
part of the Open response
may be gated off and causes
the extracted insertion loss
to have some ripples
If the gating range is too wide
• If the gating range is too
wide, although the complete
Open response will be
maintained, some mismatch
effects caused by the
impedance variations of the
PCB trace will also be
included after gating and
introduce some ripples to the
extracted insertion loss
Optimum gating range
• The optimum gating range
needs to be adjusted as a
compromise of the two
aspects above
Extracted insertion loss with different gating range
In real applications
• Optimize the fabrication
quality to make sure the
impedance variations are as
small as possible
• Select the gating range
carefully to optimize the
extracted insertion loss if
there are still some
impedance variations
Conclusion
Direct Loss
Subtraction
TRL
calibration
2xThrough
AFR
1xAFR
Complexity Easy Complicated Easy Easy
Accuracy Low High High High
Cost Low High Low Lowest
Measurement
Bandwidth
Low High High High
Thank You!
Q & A