data mining s-parameters slide -1 time before start...

Bogatin Enterprises, LLC, a LeCroy Company 2011

Slide -1

www.beTheSignal.com

Data Mining S-Parameters

How to Read S-Parameters

Like a Bookor

Tapping Into Some Of The Information Buried Inside S-

Parameter Black Box Models

Dr. Eric Bogatin, Signal Integrity

Evangelist, Bogatin Enterprises a LeCroy

Company

www.beTheSignal.com

Fall 2011

Bogatin Enterprises and

LeCroy Corp

No Myths Allowed Webinar

Time before start:

Downloaded handouts from www.beTheSignal.com, recent publications: PPT-NMA-855


Slide -2

www.beTheSignal.com


Overview

• S-parameters as a “black box” behavioral model

• Root cause analysis: a powerful methodology

• Four common patterns and their possible root cause

• A Pop Quiz


Slide -3

www.beTheSignal.com


For More Information

www.beTheSignal.com

Recent Publications

My Blog: What I learned this month

Class schedule

Published by Prentice Hall, 2009

Boston: Sept 28-29

Hyderabad: Oct 10-11

Bangalore: Oct 13-14

Singapore: Oct 18

Penang: Oct 20

Shah Alam: Oct 21

San Jose: Nov 8-10


Slide -4

www.beTheSignal.com


Hands on Labs in All Bogatin

Enterprises Classes(simulation software and lab exercises provided)

• Quite Universal Circuit Simulator (QUCS)

Circuit Simulator: transient, frequency domain, S-parameter

• LeCroy’s SI Studio

S-parameter viewer

Simulate eye diagrams from channel S-parameters

Serial data analysis

S21

S11


Slide -5

www.beTheSignal.com


A Special Thank you for all

Webinar Attendees

• 10% discount to any Boston or San Jose, Bogatin Enterprises class:

SPSI, Sept 28, 2011, Boston

DPD, Sept 29, 2011, Boston

EPSI, Nov 8-9, 2011, San Jose

SPSI, Nov 10, 2011, San Jose

• Use discount code NMA1109A when registering online at www.beTheSignal.com


Slide -6

www.beTheSignal.com


The Real World:

Interconnects are Not Transparent

Driver BoardPackage Backplane Board Package Receiver

TX RX

1. We use S-parameters to describe the behavior of the interconnects

2. We can answer some “why” questions from the S-parameters


Slide -7

www.beTheSignal.com


Characterize an Interconnect by How “Precision” Reference Signals Scatter Off a DUT

S11 S21

Incident Wave

Transmitted wave

TDR TDT

tTime Domain

tFrequency Domain

waveform out from a port

waveform in to a portparameter =


Slide -8

www.beTheSignal.com


What are S-Parameters?

• A collection of scattered responses (at every frequency):

Reflected sine waves

Transmitted sine waves

• Stored in a special format: Touchstone file

.s1p: scattered data from 1 port

.s2p: scattered data from 2 ports

E

1 port into wavesine

1 port from out wavesineS11 =

1 port into wavesine

2 port from out wavesineS21 =

Freq units are MHz

S-parameters

dB and degrees

Reference Z 50 ohm

Magnitude

and Phase or

Real, and

Imaginary


Slide -9

www.beTheSignal.com


System Simulation with S-Parameter

Behavioral “Black Box” Models

1 2 3 4 5 6 7 8 90 10

-30

-20

-10

-40

0

freq, GHz

Diffe

ren

tial R

esp

on

se

5 Gbps @ RX

U1

CMOS,3.3V,FAST

2

1

V1

TOP

BO...

TL1

83.5 ohms447.547 ps3.000 inStackupTL2

83.5 ohms447.547 ps3.000 inStackup

J2

m1_1234.s4p

Port1 Port2

Port3 Port4

J3

m1_1234.s4p

Port1 Port2

Port3 Port4

U2

CMOS,3.3V,FAST

2

1

TL3

83.5 ohms447.547 ps3.000 inCoupled StackupTL4


TL5

83.5 ohms447.547 ps3.000 inStackupTL6


J4

m1_1234.s4p

Port1 Port2

Port3 Port4

1 2 3 4 5 6 7 8 90 10

-30

-20

-10

-40

0

freq, GHz

Diffe

ren

tia

l R

esp

on

se

, d

B

1 2 3 4 5 6 7 8 90 10

-30

-20

-10

-40

0

freq, GHz

Diffe

ren

tia

l R

esp

on

se

, d

B

Turn S-parameter Behavioral Model into a SPICE

compatible model using pole-zero model of S-

parameters which any SPICE can use:

“broad band SPICE”: Simbeor, HyperLynx, ADS,

Sigrity, SiSoft E


Slide -10

www.beTheSignal.com


Opening the Lid to the Black Box

What treats lay within?


Slide -11

www.beTheSignal.com


Fastest Way to Solve a Problem is

to Identify its Root Cause

If you have the wrong

root cause, you will only

fix the problem by luck


Slide -12

www.beTheSignal.com


Don’t Think Frequency or Time Domain,

Think Frequency AND Time Domain

Single ended S-parameters Differential S-parameters

Differential T-parametersSingle ended T-parameters

Frequency Domain

Time Domain

Circuit simulation

Electromagnetic simulation

Measurement


Slide -13

www.beTheSignal.com


2 Port S-Parameters

• S11: Return loss

It is the reflected signal

Impedance mismatch from 50 ohms throughout the interconnect

A little about losses

• S21: Insertion loss

It is the transmitted signal

Impedance mismatches throughout the interconnect

Losses

~50Ω Z0 = 50Ω

Vsource

Magnitudeand phaseDetector

DUT

magnitude/phase

detector50Ω

Z0= 50Ω

Transparent interconnect:

S11: large, negative dB

S21: small, negative dB

Applies to single-ended and differential S-parameters


Slide -14

www.beTheSignal.com


Four Important Patterns in S11, S21

• Ripples in S11, sometimes in S21: reflections

• Monotonic drop in S21: losses

• Broad dips: ¼ wave stub resonances

• Sharp dips (hi Q), coupling to resonances

Insertion loss

return lossInsertion loss


Slide -15

www.beTheSignal.com


How Reflections Result in Return,

Insertion Loss Ripples

• When Len << ¼ λ

Reflections from front and back, 180 deg out of phase

No net reflection, all transmitted waves in phase and add

S11 large negative dB, S21 nearly 0 db

Z0 < 50 Ohms

When Len << ¼ λ

min S11 max S21

½

0

00

0 0

50 Ω 50 Ω S21

S11

At low frequency

ALL interconnects

are transparent


Slide -16

www.beTheSignal.com


When Len = ¼ λλλλ, S21 Lowest, S11

Highest

• When Len = (½ x n + ¼ ) λ

Max S11, min S21

Z0 < 50 Ohms

min S21

When Len << ¼ λ

When Len = 1/4 λ

½ ¼ ¼ + 0

½ max S11

½ + 0 ¾

min S11 max S21

½

0

00

0 0

S21

S11


Slide -17

www.beTheSignal.com


When Z0 ≠ 50 Ohms, Multiple Reflections

From The Terminations Cause Ripples

• When Len = n x ½ λ Reflected waves from front and back subtract

Minimum reflected signal

Transmitted waves all in phase, S21 max

• As frequency increases, insertion, return loss increase, decrease

Longer distance between reflections, shorter the frequency between high and low

Larger the impedance difference, the larger the modulation

Z0 < 50 Ohms

min S21

When Len << ¼ λ

When Len = 1/4 λ

½ ¼

¼ + 0 ½

max S11

½ + 0 ¾

min S11 max S21

½

0

00

0 0

When Len = ½ λ

min S11 max S21

½

1

½ ½ + 0

1 + 0 1½

S21

S11


Slide -18

www.beTheSignal.com


Attenuation and Insertion Loss

In real interconnects,

amplitude drops off

exponentially with distance

( )dB/len

nepers/lendd 20

out in inV d V e V 10α

−−α= =

d

S21 is attenuation when terminations are matched and there is no

coupling out of the transmission line

out

in

VS21

V=

[ ] [ ]out

in

VS21 in dB 20 x log dB / in x d attenuation

V

= = −α =


Slide -19

www.beTheSignal.com


Estimating Attenuation per Length

in 50 Ohm Interconnects

dB/inch

w = line width in mils

f in GHz

Dk = dielectric constant

Df = dissipation factor

Len

1atten f 2.3 x f x Df x Dk

w

= − +

w = 5 mils

Dk = 4

Df = 0.02Measured attenuation

from two different

transmission lines

Simple estimate

matches measured

behavior ok

Never sign off on a design

based on a rough estimate

( )Len GHzatten ~ 2.3 x Df x Dk ~ 0.1dB / inch / GHz− −


Slide -20

www.beTheSignal.com


The ¼ Wave Stub Resonance

TD, Lenstub 2 x TD

When 2 x TD = ½ cycle, minimum received signal

TD = ¼ cycle: the quarter wave resonance

res

1 1TD

4 f=

innsec

LenTD

6=

1.5

Len=res

1 1f

4 TD=

f in GHz Len in inches

Example: Len = 0.5 inch, fres = 3 GHz


Slide -21

www.beTheSignal.com


¼ Wave Resonance in the

Frequency Domain

Lenstub

1.5

Len=

res

1 1f

4 TD=

Stub 0.5 inches long

fres = 3 GHz

If Nyquist = 3 GHz,

no eye at BR = 6 Gbps

Stub length = 0.5 inches

S21 S11

Where does the energy go?


Slide -22

www.beTheSignal.com


Measured Insertion Loss with and without

Via Stub in a ¼ inch Thick Backplane

No via stub in the channel

0.2 inch via stub

res

1.5 1.5f 7.5 GHz

Len 0.20= = =

7.5 GHz


Slide -23

www.beTheSignal.com


Coupling to Hi-Q Resonances

• What are hi-Q resonators?

Any floating metal (copper fill)

Plane cavities


Slide -24

www.beTheSignal.com


Via to Cavity Coupling in a 4 Layer Board

Plane resonances expected:

Len = 3.25 in fres = 0.92 GHz



Len

GHz3

Lenx2Dk

GHz12fres ==

S21 wo return vias

3.25 inches

1.187 inches

0.8 inches

0.8 inches


Slide -25

www.beTheSignal.com


Four Important Patterns in S11, S21

• Ripples in S11, sometimes in S21: reflections

• Monotonic drop in S21: losses

• Broad dips: ¼ wave stub resonances

• Sharp dips (hi Q), coupling to resonances

Insertion loss

return lossInsertion loss


Slide -26

www.beTheSignal.com


Pop Quiz:

What Causes the Higher Loss?

Looks like bottom (blue) line has more SDD21 than top (red) line.

What causes the higher attenuation?

-Conductor loss from surface roughness?

- Poor copper plating

- High Df in one layer

- ???

Same measurement, but up to 20 GHzMeasured SDD21 for 2 differential pairs, up to 10 GHz

What causes the large dip?

- Stub resonance?

- Mode conversion?

- Resonant coupling to other structures?

- Bloch waves and glass weave?

- Dielectric absorption resonance?

Dip at 14 GHz

Data courtesy of Bob Haller, Enterasys

Which S-parameters might have the answers?


Slide -27

www.beTheSignal.com


Could it Be Mode Conversion?

SDD21 SDD11

SCD11

SCD21

Mode conversion terms:

SCD11, SCD21

SDD11 shows reflected energy

SCD11 shows mode conversion reflected

SCD21 shows mode conversion transmitted


Slide -28

www.beTheSignal.com


Summary

• Get in the habit of looking inside S-

parameters

“You can observe a lot by watching”, Yogi

Berra

• Look for the four patterns

Ripples

Monotonic drop

Broad dips

Sharp dips

• This has been just a “tip of the

iceberg” look at data mining S-

parameters


Slide -29

www.beTheSignal.com



Slide -30

www.beTheSignal.com


For More Information

www.beTheSignal.com

Recent Publications

My Blog: What I learned this month

Class schedule

Published by Prentice Hall, 2009

Boston: Sept 28-29

Hyderabad: Oct 10-11

Bangalore: Oct 13-14

Singapore: Oct 18

Penang: Oct 20

Shah Alam: Oct 21

San Jose: Nov 8-10

• Classes with Hands on Labs:

Quite Universal Circuit Simulator (QUCS)

LeCroy’s SI Studio

S21

S11

data mining s-parameters slide -1 time before start...

Documents