1. SCT Links Operation Performance from QA and operation
Setting up links during operation for barrel and EC. TX RX
Operational problems
Fatalities
Difficulties
How can we track performance?
Acknowledgements
thanks to Peter Phillips, Bilge Demirkoz, Caroline Magrath and Nigel Hessey for discussions of SCT procedures
Tony Weidberg
Pixel Opto November '06

2. Performance from QA On-detector VCSEL power
Two VCSELs in one opto-package have single VCSEL current setting Spread of power between two VCSELs important.
Tony Weidberg
Pixel Opto November '06

3. PIN Arrays & DRX-12 Slow tail to signal limits dynamic range
Count bit errors in 32k bit stream
Blue region is error free.
Slow tail to signal limits dynamic range
Saturation of DRX-12 at very high input current ~ 800 mA
Tony Weidberg
Pixel Opto November '06

4. RX links dynamic range (1)
RX link designed to work at power > 40 mW
Only tested on one channel but could operate with negligible BER at 35 mW (no error in 1 hour)
Bit Error Rate Scans
Different Optical power
BER
RX DAC

5. RX links dynamic range (2)
Look at range of RXmax/RXmin
Can’t set RXmax > 255  Need RXmin < 100  reasonable operating window  RXmax ~ 500  Power ~ 1000 mW
Would be better to have a larger range for DRX-12 threshold but can only easily increase this by 20%. Does this help?
However system should work over dynamic range > ~ 10.
Is this true in practice?
Tony Weidberg
Pixel Opto November '06

6. Off-Detector VCSELs High mean power but large spread.
Change 10 mA  15 mA increases power by ~50%
Tony Weidberg
Pixel Opto November '06

7. Setting Up Links TX links
Initially set TX DAC = 165  nominal VCSEL current of 10 mA. Newer TXs have higher fibre coupled power  new default TX DAX = 140
Check IPIN. Adjust if <100 (200) or > 700 (600) mA for barrel (EC).
Usually ok after this but if there are problems perform TX DAC scan (see later).
Tony Weidberg
Pixel Opto November '06

8. RX Links DC coupled links are more difficult to set-up.
Old procedure for barrel
Perform RX threshold scan.
Send configuration data and repeat for 10 triggers (8000 bits). Change RX DAC value and repeat  coarse scan of BER vs RX DAC. Optimal setting RX_DAC=0.3*RXmin+0.7*RXmax  figure. Works but is slow.
Set timing to global value (barrel) or perform RXdelay scan (Endcaps).
New method: superfast scan (7s)
2d scan of delay and Rxthreshold in clock/2 mode (ABCD returns 20 MHz clock). Works for most modules but use slower Rxthreshold based on configuration data for problematic channels.
Tony Weidberg
Pixel Opto November '06

9. RX threshold Scan Good links  wide operating window.
Not typical! Most VCSEL too bright so don’t see top edge of scan.
RX_Threshold
Errors at low RX DAC are not due to noise but are due to the slow tail in the p-i-n signals (off-detector array).
RXmin scales with power
Tony Weidberg
Pixel Opto November '06
Time Bin

10. MSR Adjustment (1)
Need accurate 50:50 MSR for 20 MHz clock  low jitter BC clock recovered by DORIC  don’t try to design perfect clock but allow for adjustments with MSR register.
Adjustment will be required on all channels to optimise timing but if it is very wrong can cause problems even in calibration runs.
i/p clock with non 50:50 MSR
o/p clock has large jitter
Tony Weidberg
Pixel Opto November '06

11. MSR Adjustment (2)
MSR adjustment can work  produce low jitter BC clock. Obtain jitter (full width) < 0.5 ns for optimal MSR but jitter could be very big if MSR wrong
Warning: we found a few VCSEL arrays that had very low MSR and could not be tuned to give 50:50
MSR Register

12. MSR Adjustment (3) How to set MSR values in full system? MSR scan
Idea is to tune system so that duty cycle of clock/2 is independent of BPM phase.
MSR scan
(1) Set clock/2 mode
(2) SetMSR value
(3) Measure width of on period with scan of RXdelay.
(4) send odd number of bits (SR) of TTC data to flip phase of BPM signal and repeat (3)
(5) Change MSR register  3
(6) Plot duty cycle for the two phases as a function of MSR register  figure
Plan to develop faster 2D scan in future
Tony Weidberg
Pixel Opto November '06

13. MSR Adjustment (4)
Scan does work but is very time consuming. Only use it now if we have problems with calibration runs

14. Operational Problems Fatalities
ESD deaths. ~ 0.4 % of on-detector VCSELs died after burn-in (72 10 mA and 50C and full QA (~ 30 minutes operation)  rely on redundancy. Maybe lower level damage on more VCSELs will compromise the lifetimes?
Slow turn on VCSELs (figure). QA used DC or PSBR data. Real operation involves long gaps between bursts of data. ~ 0.25% of on-detector VCSELs are not usable.
Tony Weidberg
Pixel Opto November '06

15. Slow Turn on VCSEL Link 1 has no good value for RX DAC use redundancy
Link 0 illustrates intermediate case: can find good RX DAC but lose dynamic range.
Tony Weidberg
Pixel Opto November '06

16. Operational Problems Difficulties
RX links: Intermediate cases of slow turn on VCSELs can be used but much more care required to set Rxthreshold  slow scan.
TX links: some require larger IPIN than normal, may be linked to slow turn on off-detector VCSELs?
Tony Weidberg
Pixel Opto November '06

17. Problem Solving Recipes
Runs do crash. Often after many minutes in noise occupancy scans because long events suffer more from slow turn on VCSEL problems.
Fixes
Hard reset ( module CLK/2) and repeat
If RXmin>100 reduce VCSEL current
If RXmin<100 increase VCSEL current
Redo RX threshold scan and RX delay scan.
Look at error codes: if they are L1 or BC ID errors  error at start of run  probably slow turn on VCSEL.
perform TXscan: scans TX DAC value and sends and returns configuration data (cf RX threshold scan). Increase or decrease IPIN if necessary.
(ABCD errors  increase VDD to 4.2V)
Have succeeded in simultaneous operation of large number of modules: 672 Barrel 6 in Oxford and 462 barrel modules in SR1 Cosmic run but required a lot of tweaking of parameters.
Tony Weidberg
Pixel Opto November '06

18. Tracking Performance TX links RX links.
Measure IPIN and check if we see expected decrease in responsivity with radiation damage
RX links.
Can track performance roughly with RXmin  fig.
Better if we could measure I_PIN in array for each channel
Tony Weidberg
Pixel Opto November '06

19. Tracking I(p-i-n) Difference I(p-i-n) in a run cf reference run.
Mean change in I(p-i-n) versus run number
10% spread due to off-detector VCSEL variation but means show stability over ~ 2 months
Tony Weidberg
Pixel Opto November '06

20. RX min correlation with Power
Weak correlation
Better way: use current measurement in off-detector p-i-n array
12 bits with LSB = 6 mA  ~ O(2%) precision.

21. Outlook Should take time to debug anomalous channels.
Look at signals with optical probe.
Easier to debug in SR1 than in pit
Can we make DAQ more robust so that single bit errors don’t cause system to crash in calibration runs?
Tony Weidberg
Pixel Opto November '06