1. Performances of the HV system out of Drawers Roméo Bonnefoy, Robert Chadelas, Christian Fayard, Marie-Lise Mercier, Eric Sahuc and François Vazeille (LPC Clermont-Ferrand) Tilecal upgrade meeting (2014 February 7 th ) ● Reminder of the basic principles ● Set ups ● Definitions of the measures ● Tested combinations ● Results ● Next developments ● Conclusions ● Appendix: Component references 1

2. ● Reminder of the basic principles ◊ General constraints - HV distribution towards Mini-Drawers (Maximum of 12 HV channels). - HV performances similar to the present ATLAS conditions: they are better than the initial specifications of HV stability ≤ 0.5 V. - Possibility of individual HV channel switching. - Improved HV safety (Dust, metallic pieces, humidity, hazards…). - Increased radiation levels. ◊ Specific principles of the studied option - Regulation system in USA15 using the same electronics schemes. - Distribution of the individual HVs via Multiconductor cables. - Passive HV bus cards.  Permanent access to the electronics and reliability.  Full insensitivity to radiations.  Cheap. ◊ R&D requested, - In particular the regulation and noise aspects. - If validated  HV supply for the Demonstrator in Building 175. 2

3. ● Set-ups - The set-up is the same in the tests made at LPC Clermont-Ferrand and in the building 175 at CERN. - The same components are used also (HV crate, cables, HV bus cards, FE parts). CERN 3

4. HV Bus boards (no active components) 12 channels. - A 4-layer printed circuit: External layers as shielding, Internal layers to distribute the HVs. - Noise killers (1K  ) everywhere. HV regulation crate: ¼ of the total space (4 Modules later) HV outside LV’s inside CANBUS Rear Front DCS HV Source LV HV cards on special supports  The LV/HV supplies.  Channel ordering. recovery. 4

5. ● Definitions of the measures ◊ Noise measures i 48 1 T1T2T3T4 FE - Digital scope on 1 M  (20 ms/scale unit, cut frequency at 20 MHz). -Measure = “Peak to Peak“ variations =   RMS on the total applied HV ≤ (  /2) x (1/k) /sqrt(12) for a uniform distribution which is the most pessimistic case. -Noise level very low: around 1 mV  The estimate is not easy: accuracy of about 10%, bigger for larger noises > 5 mV. -Made through Dividers. -At the last stage level: a fraction k of the applied HV (See the schemes in the Back up slides). 5

6. ◊ HV regulation measures -Made as usually on the HV Opto cards, from the DCS recorded data. (Frequency of one applied HV measure every 10 seconds on the individual 48 channels.) -Other information recorded: applied HV source, LVs, temperatures … enable to qualify the stability of the environment.  Results on stability of each channel given by its distribution RMS.  Comparison with the present Tilecal HV stability of channels. ◊ HV drop measures -Made to see the possible effects of the multiconductor cable lengths. ◊ PMT/LED/full HV set-up signals Divider + PMT + LED (Scope 50  ) 6 -Made to see the possible noise effects.

7. ▪LV inside HV crate ▪LV outside HV crate ▪HV LPC1 ▪HV LPC2 ▪HV Tilecal ▪HV Opto Tilecal ▪HV Opto Prototype ▪1.52 m ▪20 m (4) ▪100 m ▪Load ▪Passive Divider ▪Active Divider 2 sites: all combinations ▪ LPC Clermont-Ferrand ▪ Building 175 CERN without/with load ● Tested combinations ▪HV Bus inside Bench ▪HV Bus outside Bench BUS A big number of combinations ! 7

8. ◊ FE part: several set-ups are compared - Pure load : Resistor of 3 M  (Close to 3in1 card). Comment: Not fully equivalent to the reality where there are also capacitance effects, not necessarily harmful at the noise level. - Passive/Active Dividers, without or with the 2 groundings (Readout and HV) interconnected. Comment: As the Active Dividers have active components (Transistors, Diodes), some additional noise should be visible, by wishing it stays very low. ◊ Multiconductor cable lengths -These cables connect the HV regulation crate to the Super-Drawer Patch Panel. - Three lengths were compared, each one corresponding to specific conditions: ▪ Short cable (1.52 m) : to be not far away the present Tilecal conditions. ▪ Medium cable (20 m): to fit the Building 175 conditions. ▪ Long cable (100 m): to reproduce the distance USA 15-Detector. - Short additional lengths must be added to supply inside, respectively, the Mini-Drawers 1 to 4: 0.145, 0.835, 1.525, 2.225 m (with an uncertainty of 5 mm). 8

9. ◊ Low Voltage Power Supplies -Two locations and types of LV Power Supplies are compared: ▪ Inside the HV regulation crate: specially designed Switching Power Supplies. ▪ Outside the HV regulation crate: Linear Power Supplies used in workshop. - Comment: The inside implementation offers a self working crate system. ◊ High Voltage Source Power Supplies - Three types of HV Power Supplies are compared: ▪Commercial Power Supplies used in workshop studies (Siemel Cie and FUG). ▪Tilecal Power Supply (Tesla Cie). - Comment: The HV source is adjusted at – 830 V and must deliver up to 12 mA. ◊ Electromagnetic shielding -The HV Bus cards have grounded their 2 external layers. -Tests made in an open space then inside a closed box. 9

10. ◊ HV Opto cards -Some story! ▪ Historically, the first design (Scheme 1) of the HV Opto card had no transistors outside the Optocouplers of the regulation loop. ▪ The first campaign of radiation tests showed the sensitivity of Optocouplers:  gain losses at the 2 Opto sides: Opto and Transistor  regulation failures at a certain radiation level. 1 2 5 46 Opto side Transistor and HV side ▪ The solution (Scheme 2) was to “boost” the 2 sides by putting Transistors  New scheme.  New radiation/qualification tests.  Satisfying working of the HV system (well demonstrated by Tilecal data) but regulation loop more sensitive to the environment (Cable lengths, etc.). MOC8204 10

11. Scheme 1Scheme 2 See the full schemes in the Back up slides - In this R&D study, the 2 schemes are compared, because there are no longer radiation concerns. - The best one will be chosen to measure the whole performances. 11

12. ● Several campaigns of tests -Because of the low noises, it is difficult to reproduce everywhere/every time the same environment. Example: Changing the sector mains of the various elements (HV and LV supplies, scope) can affect the results. -After a time consuming learning phase of several months,  Tests made in 5 periods: LPC 2013 November 19-21 CERN 2013 November 25-27 LPC 2013 December 18-19 LPC 2014 January 24-25 LPC 2014 February 3-4 12

13. ● Results ◊ Noise studies Step 1: Search of the most significant effects. Step 2: Performances of the best solution. Step 1: Search of significant effects □ What could be expected, even though the noise could be very low ? - Every time Transistors (  gains) are concerned in the comparisons  noise is expected to be increased. A priori there are 2 concerns: - The HV Opto Scheme 2/Scheme 1 - The Active Dividers/Passive Dividers - The noise would increase when the cables are longer. □ Other effects, likely smaller, cannot be easily foreseen before testing.  The major role of the experiments. [Everywhere: HV in = - 830 V, all individual HV out = -700 V.] 13

14. □ Effects of Transistors and cable lengths: Search of the most significant effects (LPC, Linear LV, HV LPC2) Opto Type Divider1.5 m100 m Scheme 1Passive1.40±0.10 (0.35)1.43±0.10 (0.34) Active1.70±0.10 (0.34)1.78±0.17 (0.40) Scheme 2Passive1.16±0.11 (0.37)4.54±0.35 (1.22) Active1.54±0.10 (0.35)8.80±0.82 (2.83) Table 1: Mean (RMS) noise in mV from 12 channels (Drawer T1) using the same Passive/Active Dividers. ▪ With HV Opto Scheme 1 (Within the uncertainties): - The noise is independent from the cable length: 1.40  1.43 mV 1.70  1.78 mV - The noise is slightly higher using Active Dividers: + 0.3 mV for 1.5 m, + 0.35 mV for 100 m. ▪ With HV Opto Scheme 2: it is no longer true - The noise is very dependent from the cable length: 1.16  4.5 mV 1.54  8.8 mV - The noise is higher using Active Dividers: + 0.28 mV for 1.5 m, + 4.3 mV for 100 m. 14

15. Comment: The scheme 2 would be slightly better for a short cable of 1.5 m - 0.24 mV for Passive and – 0.16 mV for Active Dividers. Is it true for the present shorter cables (0.165 m) in Scheme 2 Tilecal ? □ Effects of Low Voltage Power Supplies (LPC, HV LPC2) DividerLV Power Supply1.5 m100 m PassiveLinear1.40±0.10 (0.35)1.43±0.10 (0.34) Switching1.45±0.10 (0.33)1.51±0.13 (0.45) ActiveLinear1.70±0.10 (0.34)1.78±0.17 (0.40) Switching1.80±0.14 (0.47)1.84±0.09 (0.31) Table 2: Mean (RMS) noise in mV from 12 channels (Drawer T1) for the Scheme 1, using the same Passive/Active Dividers. ▪ The Switching PS increases the noise of about 0.05-0.10 mV … within the uncertainties. ▪ All the conclusions of Table 1 are still valid. Conclusion: The Scheme 1 is taken as the best solution for long cables. Conclusion: The HV Regulation crate can be used with its Switching PS. 15

16. □ Effects of parasitic noises in an open space (LPC, Switching LV, HV LPC2) DividerSpace100 m PassiveOpen1.43±0.10 (0.33) Closed1.45±0.11 (0.37) ActiveOpen1.78±0.17 (0.40) Closed1.76±0.08 (0.27) Table 3: Mean (RMS) noise in mV from 12 channels (T1) for the Scheme 1 and Switching PS, using the same Passive/Active Dividers. -The measures are made in the case: Switching PS and 100 m long cable. -The closed space is the Drawer Test Bench  Some electromagnetic shielding. ▪ No difference within the uncertainties. ▪ Comment: We cannot guarantee that this box gave a perfect shielding. Conclusion: The HV Bus boards would be well shielded themselves. 16

17. Site HV Source Divider100 m LPC LPC1 HV Passive1.57±0.14 (0.48) Active2.05±0.09 (0.30) LPC LPC2 HV Passive1.43±0.10 (0.34) Active1.78±0.17 (0.40) CERN LPC1 HV Passive2.09±0.08 (0.29) Active2.40±0.07 (0.25) CERN Prague HV Passive2.16±0.08 (0.27) Active2.49±0.10 (0.34) □ Effects of HV Source and sites (LPC and CERN, linear LV) - CERN measures in the Building 175. - Open space at LPC/CERN. - At CERN, after a while, LPC1 HV failed - Back to LPC  LPC2 HV. 175 ▪ Effect of site (same LPC1 HV)  noise slightly higher: Passive + 0.52 mV Active + 0.35 mV  Growing compatible within uncertainties. ▪ Effect of HV source types: CERN  LPC1 HV/Prague HV very close (0.07, 0.09) LPC  LPC2 HV better than LPC1 HV (-0.14, - 0.27) within uncertainties or … environment ? Conclusion: The environment can play a role, and at a lower level the HV sources. Table 4: … always for 12 channels. 17

18. □ Extreme conditions - Obtained at CERN, in the following conditions : Linear LV, Prague HV, Scheme 2, 100 m long cable, T1, Active Divider  with large instantaneous variations : Mean of 135± 22 (78.0) mV Minimum channels (#9 and 11): 57.5 mV Maximum channel (#10): 322 mV Channel # 1 used later for LED test: 92 mV - Obviously not recommended ! 18

19. Step 2: Performances Data recording of tests Four 20 m long cables - Using the set-up with 4 times 20 m long cables  Goal to instrument the Building 175. - Whole Super-Drawer (4 Mini-Drawers) and whole HV set (1 HV Micro + 2 HV Opto + 4 HV Bus + internal cables + 4 long cables). Example of the CERN set-up 19

20. DividerLPCCERN Passive1.26±0.04 (0.26)1.82±0.03 (0.23) Active1.65±0.04 (0.28)2.31±0.05 (0.33) Table 5: … for 48 channels and 20 m long cables. ▪The results are slightly better than those ones as shown till now, but compatible within the uncertainties: The noise is always < 2.5 mV. ▪ The site effect is very clear: Passive  + 0.56±0.05 mV Active  + 0.66±0.06 mV but it is difficult to reproduce exactly the same conditions. ▪ The noise is higher using active Dividers … but stays very low in absolute value. LPC  + 0.39 ±0.06 mV CERN  + 0.49 ±0.06 mV 20

21. DrawerDividerLPCCERN T1Passive1.24±0.08 (0.29)1.79±0.07 (0.25) Active1.62±0.06 (0.20)2.28±0.10 (0.35) T2Passive1.25±0.08 (0.26)1.80±0.08 (0.29) Active1.63±0.07 (0.23)2.34±0.12 (0.42) T3Passive1.25±0.06 (0.21)1.81±0.04 (0.15) Active1.61±0.06 (0.22)2.21±0.06 (0.22) T4Passive1.28±0.09 (0.31)1.88±0.06 (0.22) Active1.71±0.12 (0.43)2.40±0.09 (0.31) Table 6: Comparison of the 4 Drawer sets and 20 m long cables. ▪ Made of different components: Cables (internal, long), HV Opto cards HV Bus cards the 4 sets give very close results within the uncertainties. Conclusion: The noise levels are the same within a Super-Drawer. independently from components. 21

22. ◊ Regulation studies □ From the Drawer T1 (only way to compare with the 100 m long cable) Cable length (m) Run duration (H:Min) Mean Variation (µV) Mean±  Mean (  Distr. ) Minimum variation Maximum variation 2014:4327.4±5.1 (17.6)0.51.7 10016:0840.0±7.6 (26.4)0.76.9 Table 7: Stability of 12 channels in µV (Linear LV, HV LPC1), at LPC [Scheme 1].  The cable lengths have no or a very low impact on the HV regulations. Site (100 m) Run duration (H:Min) Mean Variation (µV) Mean±  Mean (  Distr. ) Minimum variation Maximum variation LPC16:0840.0±7.6 (26.4)0.76.9 CERN1:0035.4±6.8 (23.5)0.77.2 Table 8: Stability of 12 channels in µV (Linear LV, HV LPC1 at LPC, HV Prague at CERN) [Scheme 1].  The results are very close. 22

23. Loads Run duration (H:Min) Mean Variation (µV) Mean±  Mean (  Distr. ) Minimum variation Maximum variation Resistors1:35107±16 (54.3)45.8205.1 Dividers0:3431.8±5.2 (18.1)0.50.2 Table 9: Stability of 12 channels in µV for Scheme 2 and a 100 m long cable (Linear LV, HV LPC1 at LPC, HV Prague at CERN). ▪ Comment: In the second case, Passive and Active Dividers (Grounds open) equip the odd and even channels. ▪ In this worst set-up: - The regulation stays within the Tilecal specifications (≤ 500 µV). - Loads having resistors + capacitors (Such as the Dividers) are more efficient than Pure resistive loads. 23 What stability in the worst case ? (See previous noise studies): Scheme 2 and 100 m long cable. These results show the efficiency of the regulation loop … but nevertheless this set-up is never recommended.

24. □ From the whole set-up using the four 20 m long cables at LPC Site Run duration (H:Min) Mean Variation (µV) Mean±  Mean (  Distr. ) Minimum variation Maximum variation LPC14:4329.8±6.2 (21.6)0.89.8 Table 10: Stability of 48 channels in µV (Linear LV, HV LPC1) [Scheme 1]. ▪ The regulation stays within the Tilecal specifications (≤ 500 µV). ▪ It is compatible with the ATLAS recorded data in 2012-2013: RMS = 100 µV [Reference: Loic Valéry] Conclusions: - The stability of the regulation fits fully the Tilecal specifications. - It is guaranteed independently from the cable lengths and from the other equipment's (Power supplies, for example). - It is still working even though the system is very noisy. 24

25. ◊ Noise/regulation correlation studies (Study made only in the optimum set-up [Scheme 1]) Noise (mV) HV Stability (µV) Stability/Noise of 48 channels (Linear LV, HV LPC1) (Table 10). Conclusion: There is no correlation between the stability and the noise. ◊ HV drop studies ▪ Direct voltage measures  Drop of 1 mV for 20 m.  Drop of 5 mV for 100 m. 25 Conclusion: The voltage drop is negligible with respect to the applied voltages.

26. ◊ LED/Divider/PMT studies Set-up with a PMT lighted by a LED and supplied by a 20/100 m long cable  PMT signal of 1 V (equivalent to 100 GeV) ▪ Reference: output of the HV source (HV LPC1). The LED system gives a rather large signal of RMS about 14 mV for a short cable. ▪ Measures made at LPC/CERN. ▪ At CERN: large temperature fluctuations  RMS below 14 mV at low temperature ▪ Connected at the output of the HV regulation crate:  Difficult to see the noise effects, except in the extreme case (Scheme 2, 100 m long cable, etc.)  Values up to 17.7 mV for channel #1 (noise of 92 mV). Conclusion: It is difficult to see the effect of low noises of some mV with respect to the applied 700 V  A whole readout would give more information. 26

27. ● Next developments ◊ Multiconductor cable aspects -CERN certification: more information on the HV certification must be made. (Halogen free already guaranteed). Comment: That we made previously for the present Tilecal HV cables. -ATLAS routing : First contacts (Sergei Malyukof) promising, but we must go on. ◊ Space aspects in USA15 : estimate to do. 27 Comments: ▪ In ATLAS today, Tilecal is the less consuming of service space. ▪ Despite the increased needs: - Tilecal has perfectly the right to make this request. - It is very likely that will stay the less demanding. ◊ Missing measures to make: noise versus applied HV Till now: at 700 V. Next: range 600-800 V.

28. ◊ Optimization of the scheme -Individual channel switching : As the individual channel failures are rare  Individual straps could be a simple, safe and cheap solution since the access to the crate is permanent. -Decision of the “re-use” of the HV cards (long sizes, special PCB as supports and channel organization) or a new making by optimizing the sizes and without special supports. -Optimization of the LV Power supplies inside the regulation crate. -Improvement of the HV bus cards at the Noise Killer level ▪ Noise Killers in the opposite side conditioned by existing connectors going though the printed circuits. ▪ If not available: Stiff coat or small plastic covers. 28

29. ● Conclusions ◊ A lot of combinations have been tested  The simplest scheme is the best solution: Scheme 1 (No Transistors in the regulation loop). 29

30. ◊ It is satisfying fully the Tilecal specifications: - Very low noise level: below 2.5 mV in any set-ups  Relative value < 0.0025/700 = 3.6 10 6. ▪ Minimum noise pick up at the HV Bus board  Design. ▪ Negligible increase with respect to the cable length. - Regulation performances: Stability < 0.1 mV. - Insensitivity to radiations, humidity, dust and mechanical damages (Falling pieces, handling) [See talk in Tilecal maintenance weekly meeting about present HV Bus cards]. - Suppression of 3 Finger LVPS Bricks or 3 Voltage Regulators. - Possibility of cheap and safe individual channel switching, using straps. - Permanent access to the electronics  100% working of all channels - Likely a "cheap" solution … but a realistic estimate must be made. 30.

31. ◊ The present prototype can be used (as it is) to supply a Demonstrator in the Building 175 - The four 20 m long cables are in place. - A whole HV Bus set is at CERN. - Minor changes must be made on the regulation crate (Fans, front Patch Panel…). - Standard DCS is operational and can be connected to the existing DCS. 31 The 4 cables are routed HV LPC1 HV Prague HV crate

32. DrawerT1 External T2T3T4 Internal HV Bus #1324 HV Opto # Scheme 1 External 004 Internal 005 HV Opto # Scheme 2 External 543 Internal 034 HV Micro #219 Appendix: Component references Noise and LED measures Passive Divider #108458 Active Divider #21 32 LED measures PMT #AA0491

33. SideChannels/Divider numbers Odd1357911 233133343536 Even24681012 108382108387108390108405108448108469 Power Supplies HV HV LPC1C ie SIEMEL HV LPC2C ie FUG MCN 140-1250 HV PragueC ie TESLA LVLV LPC1 (Linear)C ie GW LV LPC2 (Switching)LPC assembly When resistive loads are not used 33

34. Back up 34

35. 35

36. 36

37. HV noise measure from Passive Divider 37

38. HV noise measure from Active Divider 38