Predicting the In-System Performance of the CMS Tracker Analog Readout Optical Links

Predicting the In-System Performance of Predicting the In-System Performance of the CMS Tracker Analog Readout Optical the CMS Tracker Analog Readout Optical

LinksLinks

Stefanos DrisStefanos Dris

CERN & Imperial College, CERN & Imperial College, LondonLondon

CERN & Imperial College Stefanos Dris 15/9/2004

IntroductionIntroduction• 10 million channels in the CMS

Tracker read out by ~40 000 analog optical links.

• Electronic, optoelectronic, and optical components making up the readout links will vary in gain.

• What spread in gain and dynamic range from link to link can we expect in the final system as a result?

• Will system specifications be met?

• We are now in a position to find out by simulation based on real production test data.

IntroductionIntroduction

The Tracker Optical Links

Simulation

Specifications

Results

Conclusions


Analog Readout Optical Analog Readout Optical LinksLinks

Introduction

The Tracker The Tracker Optical LinksOptical Links

Simulation

Specifications

Results

Conclusions


SimulationSimulationSimulation: LLD DataSimulation: LLD Data

0.20

0.15

0.10

0.05

0.00

Pro

babi

lity

16141210864

Transconductance (mS)

Gain Setting 0Gain Setting 1Gain Setting 2Gain Setting 3

LLD Transconductance

• LLD has four gain settings, allowing a certain amount of gain equalization of the optical links in the final system.

Introduction


SimulationSimulation

SpecificationSpecificationss

Results

Conclusions


SimulationSimulationSimulation: Laser Tx DataSimulation: Laser Tx Data

0.020

0.015

0.010

0.005

0.000

Pro

babi

lity

0.0500.0450.0400.0350.030Slope Efficiency (mW/mA)

Laser Transmitter Slope Efficiency

Introduction




Results

Conclusions


SimulationSimulationSimulation: AOH DataSimulation: AOH DataIntroduction




Results

Conclusions

0.14

0.12

0.10

0.08

0.06

0.04

0.02

0.00

Pro

babi

lity

0.70.60.50.40.30.20.1

Efficiency (mW/mV)

Gain Setting 0Gain Setting 1Gain Setting 2Gain Setting 3

AOH Efficiency

• Calculated AOH efficiency, using LLD and Laser Tx data


SimulationSimulationSimulation: Patch PanelsSimulation: Patch Panels

0.040

0.030

0.020

0.010

0.000P

roba

bilit

y1.41.21.00.80.60.40.20.0

Insertion Loss (dB)

MFS Connector Insertion Loss0.08

0.07

0.06

0.05

0.04

0.03

0.02

0.01

0.00

Pro

babi

lity

1.41.21.00.80.60.40.20.0Insertion Loss (dB)

MU Connector Insertion Loss

Introduction




Results

Conclusions


SimulationSimulationSimulation: Receiver DataSimulation: Receiver Data

0.020

0.015

0.010

0.005

0.000

Pro

babi

lity

4540353025Responsivity (mA/mW)

ARx12 Responsivity

Introduction




Results

Conclusions


SimulationSimulationSimulation: Load ResistorSimulation: Load Resistor

• Resistors used have 1% tolerance.• Assumed a 4σ Gaussian distribution.

• The load resistor offers another handle in tuning the gain of the optical link.

• The ARx12 contains a current mode amplifier that converts the optical power into an electrical current. This is changed into a voltage by the load resistor.

• The current value is 100, but is not frozen.

Introduction




Results

Conclusions


Simulation: MethodSimulation: MethodSample PDF of each component

Multiply samples together to obtainoverall link gain for each LLD setting

Gain = LLD * Laser Tx * Connector Insertion Losses* ARx Responsivity * Load Resistor

Result: Four gains for each LLDsetting

Choose LLD setting whichresults to overall gain

closest to 0.8V/V

With LLDSwitching?

YES

Choose same LLD setting(0,1,2, or 3) every time.

NO

Repeat until 1 millionlink gains have been

produced

0.040

0.030

0.020

0.010

0.000

Pro

babi

lity


MFS Connector (Distributed Patch Panel) Insertion Loss0.080.070.060.050.040.030.020.010.00

Pro

babi

lity


MU Connector (In-Line Patch Panel) Insertion Loss

0.020

0.015

0.010

0.005

0.000

Pro

babi

lity

0.0500.0450.0400.0350.030Slope Efficiency (mW/mA)

Laser Transmitter Slope Efficiency

0.020

0.015

0.010

0.005

0.000

Pro

babi

lity

4540353025Responsivity (mA/mW)

ARx12 Responsivity

0.35

0.30

0.25

0.20

0.15

0.10

0.05

0.00

Pro

babi

lity

16141210864

Transconductance (mS)

Gain Setting 3

LLD Transconductance

Gain Setting 0

Gain Setting 1

Gain Setting 2

µ = 5.38mS

µ = 8.06mS

µ = 13.41mS

µ = 10.74mS • Use Inverse Transform Method to obtain random sample from each component PDF.

• Multiply each sample together to get overall optical link gain. 4 LLD settings 4 link gains.

• Choose one of the four link gains according to switching algorithm.

• Repeat 1 million times.

Introduction




Results

Conclusions


SpecificationsSpecifications

Single-ended gain

Min Typical Max

0.25 0.8 2.5 V/VAt nominal LLD setting 1, no switching

0.3 0.8 1.3 V/V With LLD switching

• Target link gain = 0.8V/V• This corresponds to 80 ADC counts per 25 000 electrons from the

detectors.• Data is digitized, and there are 8bits of information for the signal. At the

target gain, 80 000 electrons can be transmitted with 8bits. • Assumption:

– For 320μm detectors, 1MIP = 25 000 electrons– For 500μm detectors, 1MIP = 39 000 electrons

• Therefore, for thin detectors, 3.2MIP signals can be transmitted without clipping using 8bits (2MIPs for thick detectors).

Introduction


Simulation


Results

Conclusions


ResultsResults30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

Gain 0

Gain 1

Gain 2Gain 3

Gain Spreads Without Switching

Introduction


Simulation

Specifications

ResultsResults

Conclusions

• ‘Single gain’ (no switching) distributions are roughly Gaussian.

• Gain 0 distribution tail exceeds 0.8V/V target. These links cannot be set to a lower setting.

• ‘Switching algorithm simply selects LLD gain setting which results in overall link gain closest to 0.8V/V.

• Switching is like cutting through ‘single gain’ distributions and selecting the slices centered on the target 0.8V/V.


30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

Gain Spreads Without Switching

Gain 0

Gain 1

Gain 2Gain 3

ResultsResults

30000

25000

20000

15000

10000

5000

0

Cou

nt

1.21.11.00.90.80.70.60.5

Gain (V/V)

LLD Gain Setting Gain 0: 65.46% Gain 1: 34.45%

Gain 2: 0.09%Gain 3: 0.00%

Gain Spread with Switching

Introduction


Simulation

Specifications

ResultsResults

Conclusions

• ‘Single gain’ (no switching) distributions are roughly Gaussian.

• Gain 0 distribution tail exceeds 0.8V/V target. These links cannot be set to a lower setting.

• ‘Switching algorithm simply selects LLD gain setting which results in overall link gain closest to 0.8V/V.

• Switching is like cutting through ‘single gain’ distributions and selecting the slices centered on the target 0.8V/V.

• Possible to estimate ‘hard limits’ of the switched distribution by calculation.

• Average gain ~0.775V/V.• 99.9% of the links will have gains

from 0.64 to 0.96V/V.


ResultsResults

Dynamic Range• Analog data is digitized by 10-bit ADC, but 2 MSBs are discarded later.• Dynamic Range: Maximum signal size in electrons that can be transmitted

without clipping using 8bits.• Can translate this value into MIPs for both thin (320μm) and thick (500μm)

detectors.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000

Optical Link InputElectrons


ResultsResults

• Optical link output signal size in ADC bits, as a function of the input from the detectors in electrons.

• Solid line corresponds to the typical (target) gain value of 0.8V/V.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000



ResultsResults


• Solid line corresponds to the typical (target) gain value of 0.8V/V.• 2 MSBs of the 10-bit ADC are discarded, therefore 8bits available to

accommodate the signals.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000


0.80

V/V


ResultsResults


• Solid line corresponds to the typical (target) gain value of 0.8V/V.• 2 MSBs of the 10-bit ADC are discarded, therefore 8bits available to accommodate the

signals.• At link gain = 0.8V/V, signals up to 80 000 electrons can be transmitted without clipping (80

000 e-/8bits).

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000


0.80

V/V


ResultsResults

• Shaded area corresponds to minimum and maximum link gains determined by the simulation.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000


0.8V

/V


ResultsResults


• Minimum dynamic range = ~63 200 e-/8bits• Maximum dynamic range = ~100 500 e-/8bits

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000


0.80

V/V

1.01

V/V

0.64

V/V


ResultsResults


• Minimum dynamic range = ~63 200 e-/8bits• Maximum dynamic range = ~100 500 e-/8bits• Knowing the distribution of the gain, we can calculate the distribution of the

dynamic range.• 99.9% of the links will have a dynamic range from ~66 500 to 100 000 e-/8bits.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000

Optical Link Input

15000

10000

5000

Count

Electrons

0.80

V/V

1.01

V/V

0.64

V/V

0


ResultsResults

• For thin detectors (320μm), 1MIP=~25 000e-

• Hence at 0.8V/V gain, the system can transmit ~3.2MIPs/8bits without clipping.• 99.9% of the links will have a dynamic range from ~2.65 to 4 MIPs/8bits

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000

Optical Link Input

15000

10000

5000

Count

6543210MIPs (320µm detectors)

Electrons

0

1.01

V/V

0.80

V/V

0.64

V/V


ResultsResults

• For thick detectors (500μm), 1MIP=~39 000e-

• Hence at 0.8V/V gain, the system can transmit ~2.1 MIPs/8bits without clipping.• 99.9% of the links will have a dynamic range from ~1.7 to 2.6 MIPs/8bits

Introduction


Simulation

Specifications

ResultsResults

Conclusions

10

8

6

4

2

0

Opt

ical

Lin

k O

utpu

t (A

DC

bits

)

150000100000500000

Optical Link Input

15000

10000

5000

Count



Electrons

1.01

V/V

0.80

V/V

0.64

V/V


ResultsResults

Real System Data• 123 Tracker End Cap system optical links in test beam. • Relative agreement between simulation and real data.• ‘Single gain’ distributions of the real system are a few percent higher than

in simulation, with larger spread. This is most likely due to electronic components on either side of the optical link which were not simulated.

• Upper and lower limits of both switched distributions are almost identical – not surprising, given same switching algorithm is used and the dominant effect of gain settings 0 and 1 in both cases.

15

10

5

0

Cou

nt

1.21.11.00.90.80.70.60.5Gain (V/V)

30000

25000

20000

15000

10000

5000

0

LLD Gain SettingGain 0: 72.4%Gain 1: 21.1%Gain 2: 1.6%Gain 3: 4.9%

Test Beam Simulated

Introduction


Simulation

Specifications

ResultsResults

Conclusions


ResultsResults

Previous Study• Uniform component distributions assumed.• Typical LLD gain setting was thought to be 1, now it is 0.• Larger range of gains than those predicted with real production data.

– Due to low-end tail of Gain 3 distribution.– Hence, low-gain links could not be compensated as well as high-gain links (the

opposite of the current situation).

• Ignoring the Gain 3 low-end tail, the extents of the spread are the same as in current simulation.

Introduction


Simulation

Specifications

ResultsResults

Conclusions

50000

40000

30000

20000

10000

0

Cou

nt

1.41.21.00.80.60.40.2Gain (V/V)

LLD Gain SettingGain 0: 11.00%Gain 1: 36.01%Gain 2: 30.18%Gain 3: 22.80%


ResultsResults40000

35000

30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

100 Ohms

30000

25000

20000

15000

10000

5000

0

Cou

nt

1.21.11.00.90.80.70.60.5Gain (V/V)


Gain 2: 0.09%Gain 3: 0.00%

Load Resistor• In addition to the

switchable LLD, there is a second handle on the gain of the full optical link.

• The Load Resistor value can still be changed to shift the ‘single gain’ distributions.

100Ω

Introduction


Simulation

Specifications

ResultsResults

Conclusions


ResultsResultsLoad Resistor

• In addition to the switchable LLD, there is a second handle on the gain of the full optical link.


90.9Ω

Introduction


Simulation

Specifications

ResultsResults

Conclusions

30000

25000

20000

15000

10000

5000

0

Cou

nt

1.21.11.00.90.80.70.60.5Gain (V/V)


Gain 2: 0.96%Gain 3: 0.00%

40000

35000

30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

90.9 Ohms


ResultsResultsLoad Resistor

• In addition to the switchable LLD, there is a second handle on the gain of the full optical link.


80.6Ω

Introduction


Simulation

Specifications

ResultsResults

Conclusions

40000

35000

30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

80.6 Ohms

30000

25000

20000

15000

10000

5000

0

Cou

nt

1.21.11.00.90.80.70.60.5Gain (V/V)

LLD Gain SettingGain 0: 6.29%

Gain 1: 85.30%Gain 2: 8.38%Gain 3: 0.03%


ResultsResults40000

35000

30000

25000

20000

15000

10000

5000

0

Cou

nt

2.52.01.51.00.50.0

Gain (V/V)

75 Ohms

30000

25000

20000

15000

10000

5000

0

Cou

nt

1.21.11.00.90.80.70.60.5Gain (V/V)

LLD Gain SettingGain 0: 1.16%

Gain 1: 77.50% Gain 2: 21.10%

Gain 3: 0.24%

Load Resistor• In addition to the

switchable LLD, there is a second handle on the gain of the full optical link.


75Ω

Introduction


Simulation

Specifications

ResultsResults

Conclusions


ConclusionsConclusionsIntroduction


Simulation

Specifications

Results

ConclusionsConclusions

• A model has been developed for the CMS Tracker analog optical link and used in a Monte Carlo simulation to assess the performance that can be expected in the final system.

• The specifications will be met for every one of the 40 000 links.

• The gains of the links will lie between 0.64 and 0.96V/V, i.e. 32% of the specified switched gain spread.

• The ‘single gain’ (non switched) distributions are slightly higher in mean than expected, due to higher gains in the laser transmitter and LLD, as well as low insertion loss of the connectors. This means the typical setting for the LLD is gain 0.

• The model can be used to demonstrate the effect of changing the values of readout components (e.g. the receiver’s load resistor). We could tailor the positions of the single-gain distributions to achieve a typical LLD setting of 1, or change the shape of the distribution to something more desirable.

• Different switching algorithms can be tested.• Components of the readout system (not part of the optical link) can be

added to achieve even more realistic results. Temperature effects on the AOH can also be included.




Simulation

Specifications

Results



• The specifications will be met for every one of the 40 000 links.• The gains of the links will lie between 0.64 and 0.96V/V, i.e.

32% of the specified switched gain spread. • The ‘single gain’ (non switched) distributions are slightly higher in

mean than expected, due to higher gains in the laser transmitter and LLD, as well as low insertion loss of the connectors. This means the typical setting for the LLD is gain 0.


• Different switching algorithms can be tested.• Components of the readout system (not part of the optical link) can

be added to achieve even more realistic results. Temperature effects on the AOH can also be included.




Simulation

Specifications

Results



• The specifications will be met for every one of the 40 000 links.• The gains of the links will lie between 0.64 and 0.96V/V, i.e. 32% of

the specified switched gain spread. • The ‘single gain’ (non switched) distributions are slightly

higher in mean than expected, due to higher gains in the laser transmitter and LLD, as well as low insertion loss of the connectors. This means the typical setting for the LLD is gain 0.


• Different switching algorithms can be tested.• Components of the readout system (not part of the optical link) can be

added to achieve even more realistic results. Temperature effects on the AOH can also be included.




Simulation

Specifications

Results




the specified switched gain spread. • The ‘single gain’ (non switched) distributions are slightly higher in



• Different switching algorithms can be tested.• Components of the readout system (not part of the optical link) can

be added to achieve even more realistic results. Temperature effects on the AOH can also be included.




Simulation

Specifications

Results




the specified switched gain spread. • The ‘single gain’ (non switched) distributions are slightly higher in



• Different switching algorithms can be tested.• Components of the readout system (not part of the optical

link) can be added to achieve even more realistic results. Temperature effects on the AOH can also be included.

Documents

Predicting the In-System Performance of the CMS Tracker Analog Readout Optical Links