The LHCb VELO analog transmission line

Transcription

1 Version 2. The LHCb VELO analog transmission line A. Bay,a, J. Borel b R. Frei b, G. Haefeli a, L. Locatelli b, F. Zehr a, J. Buytaert c a LPHE, Ecole Polytechnique Fédérale de Lausanne (EPFL), Switzerland b LPHE, present address... c CERN, Switzerland 1 Abstract The Vertex Locator is one of the most important sub-detectors of the LHCb experiment which is devoted to B physics. The signals from the silicon strip sensors are multiplexed at 4 MHz and transported by 6 m copper lines to the counting stations placed in a radiation safe area. This paper describes the development of this long transmission line and its performances Key words: LHCb, Electronics, Data acquisition, Silicon strip detectors, Common mode PACS: 1 1 Introduction The LHCb experiment at CERN s Large Hadron Collider (LHC) is dedicated to the studies of CP violation and rare decays in the beauty sector (i. e. This project is funded in part by the Swiss National Science Foundation (SNSF) Corresponding author tel: ; fax: Aurelio.Bay@epfl.ch URL: (A. Bay). Preprint submitted to Elsevier Science 16 October 29

2 Fig. 1. The LHCb detector. The Vertex Locator (VELO) is at the collision point, left of the figure the set of particles containing a b quark). The physics of interest in LHCb being carried by particles emitted mostly at small angles with respect to the beam, the detector covers only a region from about 15 to 3 mrad (Fig. 1). The proton beams will collide with s = 14 TeV inside the VErtex LOcator (VELO), shown on the left-hand side of Fig. 1. The produced particles will cross two ring imaging Cherenkov counters, the Tracking system consisting of a 4 Tm dipole magnet, and four stations (the Silicon Tracker for the inner region, and the Outer Tracker with straw chambers for the outer part), a calorimetric system, and five muon chambers. An efficient and flexible multilevel trigger is foreseen based on custom electronics and on a PC cluster [2]. The first level of trigger, L, reduces the rate from the LHC bunch crossing rate of 4.78 MHz down to 1 MHz, based on calorimetry and muon data. L is fully synchronous and has a latency of 4 µs. In the end the High Level Trigger (HLT) has access to the full physics information and will be able to reconstruct a set of channels of interest reducing the event rate down to 1 khz. 28 The VELO is a very central element for LHCb physics: placed at only 8 mm 2

3 from the beam axis can locate B mesons decay vertices with 5µm precision. In section 2 we give a general overview of the VELO electronics. The aim of this paper is to present the development of the transmission lines bringing the data from the VELO to the counting house. The particularity of the VELO data transmission resides in the fact that we decided not to use optical links as is the case for the other sub-detectors, but we have adopted copper cables instead. A careful choice of the cable was a key point for the success of the project, as discussed in section 3. In section 4 we will present the electronics circuits and the system performances. The present situation in LHCb will be briefly discussed in section Overview of the VELO electronics The VELO sensors are readout by the radiation hard Beetle front-end chip [3]. The Beetle handles 128 channels, it has a pipeline 187 columns wide to cope with L latency, and 4 output lines. Each output serializes the amplitudes of 32 sensor channels. In total 5376 lines are needed to readout the full VELO. In addition to the 32 data samples a pseudo-digital header of 4 bits is present on each output. The header carries the Beetle internal status information and the Pipeline Column Number (PCN) used in the event The interface with the HLT is made by the TELL1 boards [4]. The boards are placed in a radiation safe area quite far from the detector: in the case of the VELO 6 m of cable are needed. Unacceptable signal distortions are present when the Beetle is driving more than 1 m of cable. This requires the insertion of a repeater station. For comparison in the case of the Silicon Tracker (also using the Beetle) it was possible to place digitization and optical 3

4 drivers within 1 m of the detector, in a region with quite low irradiation. In the case of the VELO we could not find a location close enough to place such electronics. Hence a first design was made with a line driver placed just outside the VELO tank ( 1 m distance) connected by 2 m of cable to an intermediate station performing digitization and data transmission to the TELL1 via optical fibers. Subsequently the uncertainty on the precise location of the intermediate station triggered the study presented in this paper, with the goal to transport the analog data from the detector to the TELL1 boards on copper wires, without intermediate processing. A key point for this choice is that the fact that TELL1 accepts analog signals from copper lines as well as digital data from optical fibers. 64 The final design (now implemented in LHCb) consists of the repeater boards carrying the line drivers. They are placed close to the VELO tank and connected to the Beetles by short lines 6 m of twisted pair cables to the counting house the receiver boards carrying 1 bits ADCs. They are placed on the TELL The ADC must be precisely clocked in phase with the Beetles: a digital delay with 1 ns resolution is available on the TELL1 for the fine tuning of the phase of each channel. The 6 m line induces a loss of amplitude with modifications in the frequency spectrum. The electronics must amplify the analog signal and also correct the frequency response of the system. The required bandwidth is defined by the analog signals from the Beetle which are rectangular pulses at the frequency of 4 MHz The development of the system went through several configurations. A poste- riori we can say that significant time saving was made by an extensive use of 4

5 electronic simulations (based on SPICE [1] program), validated by measurements. The two main topologies considered were circuits with frequency precompensation, with frequency compensation done at the line driver level, and post-compensation, with the compensation done at the receiver just before digitization. In the end the pre-compensation circuit was preferred because it has a better immunity to high frequency pick-up from the line The choice of the cable To ensure the best immunity to electromagnetic pick-up a differential transmission is used, hence our choice is restricted to twisted pair cables. Tests were made on two types of cables available at CERN: ND26P, and ND36P, and two standard Ethernet cables, CAT5, and CAT6. Some technical features of the selected cables are summarized in Table 1. ND26P ND36P CAT5 CAT6 Number of pairs Shield global Al global Al global Al individual Al External diameter [mm] Impedance [Ω] Capacitance/m [pf/m] Insertion loss at 4 MHz [db] Propagation delay [ns] Propagation delay skew [ns] ACR at 4 MHz [db] ELFEXT at 4 MHz [db] Table 1 Upper part: technical features of the tested cables. Bottom part: physical performances measured on 6 m long cables 5

6 9 91 For 6 m of cable of each kind we have measured the insertion loss from 1 to 8 MHz (Fig. 2, left). We find that CAT5 has the lowest attenuation at 4 MHz ( 4 db). On the other hand the lower propagation delay is given by Fig. 2. Left: Insertion loss, and, right: ELFEXT. Black points: ND36, blue squares: ND26, red triangles: CAT5, purple stars: CAT6 the CAT6 cable (25 ns) and also the very lowest propagation delay skew, i. e. the maximal difference in the propagation delays between couples of twisted pairs ( 1.7 ns) We also measured the Near End CROSSTalk (NEXT) and the Far End CROSS- Talk (FEXT), the inter-cable coupling measured at the input and the output of a neighbor pair respectively. These quantities must be related to the insertion loss: the Attenuation to Crosstalk Ratio (ACR) is the difference between the insertion loss and the NEXT, which can be interpreted as a signal-to-noise ratio considering only the noise induced by a neighbor pair. In a similar way the Equal Level Far End CROSSTalk (ELFEXT) is defined as the difference between the insertion loss and the FEXT. The curves obtained with the ND36P, the CAT5, and the CAT6 are quite similar (NEXT 5 db at 4 MHz) while the attenuation is less important with the ND26P ( 3 db at 4 MHz). The ACR curves are also equivalent ( 4 db at 4 MHz) except for the ND26P where it is only of the order of 2 db. The better results for FEXT are obtained 6

7 with the CAT family and specially with the CAT6 ( -65 db at 4 MHz). This also appears in the ELFEXT values (Fig. 2, right). In particular the CAT6 has 54 db at 4 MHz while the ND cables have only 3 db. The results of these measurements are summarized in Table 1. The best high frequency behaviour was found with CAT5. A reduced high frequency loss factor, low dielectric constant, and high propagation velocity are indicators of low distortion cable. In this respect CAT5 had the best performances. Nevertheless CAT6 was selected in the end: decisive parameters were the high values for the ACR and ELFEXT and also the lowest propagation delay skew. But notice that by the end of development it was decided to introduce an individual timing for each channel with 1 ns precision, relaxing the need of a minimal delay skew. Another serious advantage of the CAT6 with respect to the others is that the four twisted pairs are shielded individually, providing a better protection against RF pick-up and crosstalk Fig. 3 shows the measured response of a CAT6 cable superposed to the results of our simulation. In order to reproduce the cable frequency behavior with 1 db precision we implemented a two stage RC network as shown in Fig. 4. The setup reproduces the module of the transfer function with sufficient precision. We use voltage controlled sources to decouple each stage. Incidentally this assumption is not far from the real situation: we have measured that the cable attenuation is enough to avoid that reflected signals at the receiver can reach the driver (the reflected fraction is few %). 7

8 Gain (db) Frequency (MHz) Fig. 3. Transfer function of the CAT6 cable. Crosses are the measured points, the line is the result of the the Spice model of Fig.4 First low pass filter Second low pass filter Spice decoupling Fig. 4. Spice model for the cable 13 4 Conception of the circuit The next step was to conceive a circuit to drive the line and compensate for the attenuation above 1 MHz. As previously said we have chosen to introduce the frequency compensation at the level of the line driver installed on the repeater boards and to keep the receiver response flat. Other constraints in the design were: the amount of radiation close to the VELO, 7 krad for 1 years of operation. Hence the amount of sensitive electronics must be minimized the power consumption in the VELO region must be as low as possible to avoid cooling the dynamics must be larger than the Beetle range of [ 5, +5] MIPs to allow 8

9 Common Mode (CM), corresponding to about ±.5 V with 1 Ω load. 1 MIP is equivalent to 22 electron-holes in the VELO silicon The line driver The line driver schematics is shown in Fig. 5. It consists of a two pole-zero network, hence the two slopes of 2 db per decade we see on the Bode diagram of Fig. 6. Because the cable transfer function is well reproduced by a double RC filter circuit and considering that the driver has exactly the same kind of behavior (two poles, two zeros) one could in principle just reverse the cable function and put resistors and capacitors into the feedback network such that to cancel the line effect. Unfortunately the optimal values are incompatible with the operational amplifier requirements which demand 5Ω feedback resistors to set the bandwidth [11]. The two other resistors are then determined to set the low and high frequency gain in order not to saturate the amplifier while keeping enough amplification. Thus the only free parameters of the network are the two capacitors. Fig. 5. Schematics of the line driver 9

10 Explicitly the transfer function of this driver is: G(s) = G (1 + sτ za )(1 + sτ zb ) (1 + sτ pa )(1 + sτ pb ) 156 where s=jω, (j 2 = 1) R L G = R 4 + R 6 R 6 R 5 + R L τ za = C 3 R 5 τ pa = C 3 (R 5 /R L ) τ zb = C 1 (R 2 + R 4 /R 6 ) τ pb = C 1 R (the / operator is R 1 /R 2 = 1 1/R 1 +1/R 2 ) and we assume that the layout of the driver is symmetric i.e. R 2 = R 3, R 5 = R 9, etc.. With the values indicated on the circuit the poles are at 12.6 and 53 MHz, and the zeros at 8.36 and 26.5 MHz. One sees on Fig. 6 that the module of the transfer function is accurate before the high-frequency cut-off, which is due to the operational amplifier characteristics and parasitic effects. To improve the precision at high Gain (db) Frequency (MHz) Fig. 6. Line driver transfer function. The crosses are the measured points. The continuous line is the calculated module of the transfer function. The dotted line is the asymptotic Bode diagram 162 1

11 frequency a Spice simulation was set up using the model of the operational amplifier provided by the manufacturer [11]. Fig. 7 shows the results with and without taking into account the effect of stray capacitances. Parasitic effects are included by the insertion of a 1 pf capacitor in parallel to the feedback resistors. Gain (db) Frequency (MHz) Fig. 7. Line driver transfer function as in Fig. 6. The dotted and continuous lines are the Spice predictions without and with stray capacitances The receiver Fig. 8. Schematics of the receiver The receiver shown in Fig. 8 is designed to process the symmetric signal from the twisted pair cable and to amplify it by a factor two before digitization. The receiver is built with a differential amplifier [11] with a flat response until about 1 MHz, according to the manufacturer. The results of our measurements and 11

12 173 simulation are given in Fig. 9: The amplification remains constant within 2 db up to about 9 MHz, followed by a fast drop. Gain (db) Frequency (MHz) 174 Fig. 9. Transfer function of the receiver. The crosses are the measured points. The dotted and continuous lines are the Spice results without and with stray capacitances Simulation of the whole system The global behavior of the system was studied by cascading the circuits of the three elements: driver, cable, and receiver. Parasitic capacitors have been included in the simulations to improve the accuracy in the frequency response. A comparison between the measurements and the simulation of the analog link system is shown in Fig. 1. Over 6 MHz the predicted gain is too high compared to the measured response (+2 db at 7 MHz and +6 db at 9 MHz). On the other hand the region up to 6 MHz, which is most relevant to us, is well reproduced including the 2 db dip between 3 and 2 MHz. This last feature is not without consequences as can be seen in Fig. 11 showing measured and predicted responses to the injection of rectangular pulses 25, 5, and 75 ns wide. The figures show a signal integration tail extending over 1 ns. On the figures we also indicate the amplitudes of the overshoot measured on the tail in a window 25 ns after the end of the pulse. The predicted results are 12

13 Gain (db) Frequency (MHz) 189 Fig. 1. Transfer function of the whole system. The dots are the measurements and the line is the response given by the model very close to the measured ones showing the reliability of our simulation It is important to keep control on the overshoot and minimize it because the remainder after the 25 ns pulse will appear as a inter-symbol crosstalk, i. e. a signal transfer between adjacent channels during the serial data transmission from the Beetle. The overshoot can be significantly modified by a trial re- amplitude (mv) % 1 5% 1 7.5% time (ns) amplitude %.2 6.8%.2 1.9% time (ns) Fig. 11. Response of the analog link to rectangular pulses of 25, 5 and 75 ns width. The result of measurements are the 3 upper plots, the mode predictions at the bottom. We have indicated the fraction of the height of the pulse after a delay of 25 ns tuning of the driver compensation network. For instance taking C 1 = 22 pf 13

14 195 and C 3 = 18 pf introduces a +1 db amplification in the 1 MHz region with respect to low frequencies, and makes the crosstalk to vanish, Fig. 12. Amplitude time (ns) 196 Fig. 12. Pulses of 25, 5 and 75 ns from Spice simulation setup with the values C 1 = 22 pf and C 3 = 18 pf So far we have assumed rectangular pulses at the input of our simulation and the measurements have been performed with a fast rise-time and fall-time signal generator. In real life the pulse is produced by the Beetle chip with its own particular spectral distribution as will be discussed in the following section Tests with (and without) the Beetle Our aim was to study the different sources of noise, random and correlated, the amount of inter-symbol crosstalk, etc.. Several iterations were done mainly trying to reduce the level of crosstalk and to adjust the global gain of the system to achieve a dynamics of about 1 MIPs. We used a setup with a Beetle chip mounted on a small PCB (two Beetles 1.3 were available in our lab, the two giving very similar results). The Beetle allows internal pulse injection but for our tests we also had 1 inputs bonded to the PCB. A given amount of charge could then be injected by a voltage step on a 3 pf capacitors placed at each bonded input. The test circuit was equipped with a line driver, the cable, and a receiver board with 8 bit ADCs (instead of the 1 bits on the final version) mounted on a special version of the TELL1. The 6 m of cable 14

15 was uncoiled, running around the lab, in a region well known for its heavy electromagnetic background Fig. 13 depicts the analog signal observed directly at the output of the Beetle. The measurement is done with a 2 GHz oscilloscope equipped with.7 pf active probes. A 2 MHz frequency cut is applied and a set of 4 events is recorded. The average amplitude is set to zero by a pedestal subtraction (pedestals are evaluated over 1 events in the absence of pulse injection). At the left we see the pseudo-digital header: the last two bits (time ) correspond to the PCN pseudo-bits and are different for each event; the first two bits (before 215) being constant they disappear after pedestal subtraction. A charge equivalent to 1 MIP was injected in the third channel, visible as an almost rectangular pulse 25 ns wide. We already notice that the pulses at the Beetle output have a tail which penetrates the next 25 ns sample. This will result in channel crosstalk. For each measurement a delay Amplitude (mv) Time (ns) Fig. 13. Analog signal measured at the Beetle output scan is performed to adjust the phase of the ADC so as to get a maximal value 15

16 for the signal, keeping the crosstalk at a reasonably low level: our target was to keep it under 5 %. Quantitatively the crosstalk from a channel (pulsed with amplitude A ) to the next (j=1), to the next-to-next (j=2), and so on, is defined as the ratio of amplitudes C j = A j /A, with j=1,2, The Beetle internal crosstalk seen above is of about 4 % for j=1. We have found that the frequency compensation circuit can be tuned to have a small undershoot for j=1 to cancel the Beetle crosstalk. By doing this we have obtained the C j values of.1 %, 2.8 %, 2.1%, and 2. %, for j=1 to 4 respectively, well within our target. Notice that with this tuning the crosstalk measured at the generator gives 5 % for j=1, i. e. an undershoot, precisely what is needed to compensate for the Beetle internal crosstalk After the delay adjusted to get the ADC clock on phase the data acquisition is started. 128 pedestals are calculated as the average over 1 events taken without signal injection. The first step in the analysis of data consists of the subtraction of such pedestal values from each channel. Subsequently a Common Mode (CM ) correction can also be applied. CM appears as a correlated (noise) amplitude of two channels far apart. This is illustrated in Fig. 14 showing an example of such correlation before and after CM correction. In the present study we have used a simple CM correction which consists in subtracting the average of all the channels with exception of the ones with charge injection (see also section 4.5). The procedure is done independently for the four output links of the Beetle The calibration in amplitude is obtained by the injection of a 1 MIP equivalent charge in one of the bonded inputs. Fig. 15 shows the associated peak at

17 Amplitude Ch 2 (ADC counts) Amplitude Ch 12 (ADC counts) Fig. 14. Amplitude correlation for two channels far apart but on the same Beetle link. The large scatter plot shows the CM correlation, the insert is the result after correction ADC counts. The other peak centered at zero comes from events with zero charge injected. The r.m.s. width of 1.14 ADC counts is interpreted as the uncorrelated part of the noise. When the Beetle is disconnected the r.m.s. decreases to.4 or less, showing that the Beetle contribution dominates the random noise. By quadratic subtraction we estimate such contribution to be 1.7 ADC counts Amplitude (ADC counts) Fig. 15. Amplitudes registered by a Beetle channel, pedestal subtracted and CM corrected. The peak at mean position of 32.5 ADC counts was obtained with the injection of a charge equivalent to 1 MIP. The peak centered at zero was obtained with no charge injected 17

18 26 We can identify the following sources of electronic noise in the VELO: (1) RF pickup by the silicon sensors (2) the Beetle input amplifier noise (3) the Beetle pipeline non-uniformity (4) the line driver noise (5) RF pickup by the 6 m of cable (including inter-line crosstalk) (6) the line receiver noise (7) the effect of digitization ) and 7) are not addressed here. 5) has been reduced to a negligible amount by the usage of CAT6 shielded pairs, and by having the equalization in the driver. 6) has been found negligible. The origin of 3) comes from the fact that the pipeline capacitors are not identical in principle requiring a PCN dependent calibration. For channels not too close to the header the PCN contribution to noise was found.97 ADC counts. On the other hand it was found that this noise is strongly correlated and that it almost disappears after CM correction: the residual (random) noise is <.26 ADC counts. The more complicated situation close to the header will not be discussed here [7,?] In the setup of our lab the total CM component was 1.86 ADC counts. Subtracting the PCN contribution seen before gives 1.59 ADC counts for the CM produced by items 2), 4), and 5). Finally a measurement directly at the output of the Beetle shows that the chip is the main source of CM, as it is for the random component In conclusion these measurements indicate a negligible noise contribution from the line. The results are summarized in Table 4.4 with the noise figures also given in MIP units. The 1 MIP signal-to-noise ratio of the system with the 18

19 source ADC MIP 1 3 Random Beetle Random other Random tot CM tot Beetle PCN CM + PCN Table 2 Sources of random and CM noise. The r.m.s. values are given in ADC counts for our setup, and converted in MIP units (1 MIP = 22 electron-holes) Beetle not connected to the sensors is expected to be of 3, after CM correction Processing of the VELO signals in LHCb The disadvantage of the CAT6 cable was the small number of pairs per cable. At the time of our tests a new cable became available [12] with 16 pairs individually shielded and specs close to CAT6, or better. This is the cable presently installed in LHCb. Only a minor re-tuning of the frequency compensation was needed. Receivers with 1 bit ADCs are now used in LHCb with a gain of 5 ADC counts per 1 MIP. For a more flexible crosstalk correction a Finite Impulse Response (FIR) filter of 3rd order acting on the 1-bit data can be invoked from the TELL1 firmware. Several schemes of CM correction have been studied [5], and other strategies are being developed taking into account the very complex topology of the VELO sensors. More details can be found in [7,?] Recent and preliminary measurements done in the pit on the VELO give a noise inferior to MIPs, after CM correction. Considering the additional 19

20 noise from the Silicon sensors this result is not far from the lab results showing that the long line performs well and does not contribute in a significant way to the noise Conclusion We have described the method adopted for the transmission of the LHCb Vertex Locator signals from the detector to the counting house. Twisted pair lines (similar to CAT6) 6 m long are used each requiring a line driver with frequency compensation, placed just after the front-end chip, and a line receiver with flat gain, mounted on TELL1 boards The contribution from the line to the electric noise was measured in our lab and found much lower than other sources. The signal to noise ratio of 3 was obtained for the system (Beetle was not connected to the sensors) after CM correction When data is being transmitted, serialized with 32 samples on each Beetle output, the finite bandwidth of the system can introduce crosstalk. We have shown that this can be made negligible (under 5 %) by a careful tuning of the frequency compensation network placed on the line driver. 318 References [1] L. Nagel et al., SPICE (Simulation Program with Integrated Circuit Emphasis), Memorandum No. ERL-M382, University of California, Berkeley, (1973) 321 [2] LHCb Collaboration, R. Antunes Nobrega et al., Trigger Technical Design 2

21 322 Report, CERN LHCC/23 31 (23) 323 [3] The Beetle 1.3, 1.4, 1.5 Reference Manual, CERN-LHCb (25) [4] Guido Haefeli et al., The LHCb Off-Detector ElectronicsDAQ interface board TELL1, Nucl. Instrum. Meth. A56, , (26); Guido Haefeli et al., FPGA-based signal processing for the LHCb silicon strip detectors, Nucl. Instrum. Meth. A569, , (26) [5] Patrik Koppenburg, Contribution to the Development of the LHCb Vertex Locator and Study of Rare Semileptonic Decays, Thesis of the Université de Lausanne, (22) [6] Guido Haefeli, Contribution to the development of the acquisition electronics for the LHCb experiment, CERN-THESIS-24-36, (24) [7] Laurent Locatelli, Direct search for Higgs boson in LHCb and contribution to the development of thevertex detector, Thesis of the Ecole Polytechnique Fédérale de Lausanne (27); Laurent Locatelli et al., Tests on the VELO analogue transmission line with the TELL1 prototype RB3, LHCb Note (24) [8] Jérémie Borel, The analog readout of the LHCb vertex detector and study of the measurement of the Bs oscillation frequency Thesis of the Ecole Polytechnique Fédérale de Lausanne (28) [9] T. Szumlak, C. Parkes, Application of the Beetle Header cross-talk correction algorithm for the VELO Detector, CERN-LHCb-PUB-29-6 (29) [1] Jérémie Borel et al., Measurements on the VELO analog link system, LHCb Note 25-3 (25) 345 [11] [12] 21