Configurable electronics with low noise and 14-bit dynamic range for photodiode-based photon detectors

ARTICLE IN PRESS

0168-9002/$ - se

doi:10.1016/j.ni

�CorrespondE-mail addr

Nuclear Instruments and Methods in Physics Research A 565 (2006) 768–783

www.elsevier.com/locate/nima

Configurable electronics with low noise and 14-bit dynamic range forphotodiode-based photon detectors

Hans Mullera,�, Rui Pimentaa, Zhongbao Yinb, Daicui Zhoub, Xi Caoc, Qingxia Lic,Yingzhuang Liuc, FeiFei Zouc, Bernhard Skaalid, Terry C. Awese

aCERN, PH Department, 1211 Geneva 23, SwitzerlandbInstitute of Particle Physics, Huazhong Normal University,Wuhan 430079, China

cElectronics and Information Department, Huazhong University of Science and Technology, Wuhan 430074, ChinadDepartment of Physics, University of Oslo, Blindern 0316, Norway

eOak Ridge National Laboratory, Oak Ridge, TN 37831, USA

Received 20 February 2006; received in revised form 12 May 2006; accepted 13 May 2006

Available online 5 July 2006

Abstract

We describe the principles and measured performance of custom configurable 32-channel shaper/digitizer Front End Electronics

(FEE) cards with 14-bit dynamic range for gain-adjustable photon detectors. The electronics has been designed for the PHOS calorimeter

of ALICE with avalanche photodiode (APD) readout operated at �25 1C ambient temperature and a signal shaping time of 1 ms.The electronics has also been adopted by the EMCal detector of ALICE with the same APD readout, but operated at an ambient

temperature of +20 1C and with a shaping time of 100 ns. The CR-RC2 signal shapers on the FEE cards are implemented in discrete

logic on a 10-layer board with two shaper sections for each input channel. The two shaper sections with gain ratio of 16:1 are digitized by

10-bit ADCs and provide an effective dynamic range of 14 bits. Gain adjustment for each individual APD is available through 32 bias

voltage control registers of 10-bit range. The fixed gains and shaping times of the pole-zero compensated shapers are defined prior to

FEE production by the values of a few R and C components. For trigger purposes, ‘‘fast OR’’ outputs with 12-bit dynamic range are

available. FPGA based slave logic, combined with a USB processor supports a variety of remote control and monitoring features,

including APD gain calibration. The measurements presented here for APDs at �25 1C ambient temperature and 1ms shaping time

achieve an average RMS noise level of 0.25 ADC counts or 290 electrons.The linearity over the dynamic range is better than 1%, as is the

uniformity of shaping time and gain over 32 channels. Due to the excellent correspondence of the output pulse shape with offline fit,

a differential timing resolution of less than 1.5 ns between channels has been achieved at ca. 2GeV, i.e. at 1.5% of the dynamic range

of PHOS.

r 2006 Elsevier B.V. All rights reserved.

PACS: 29.40.Vj; 85.60.Dw; 85.60.Gz

Keywords: Photon detector readout; Photodiode; APD; Configurable front end electronics; Discrete low-noise shaper; Dynamic range; Timing resolution;

Altro ADC

1. Photoelectron signal of PHOS and EMCal detectors

The PHOS detector [1] of ALICE is an electromagneticcalorimeter consisting of 17920 PbWO4 crystals of equalsize 180� 22� 22mm3. The detector is operated at an

e front matter r 2006 Elsevier B.V. All rights reserved.

ma.2006.05.246

ing author. Tel.: +4122 767 3533; fax: +41 22 767 0170.

ess: [email protected] (H. Muller).

ambient temperature of �25 1C to enhance the scintillationlight yield. Each crystal is attached to an Avalanche Photo-Diode (APD) of type S8148, produced by HammamatsuCo., Japan. The APDs have an active area of 5� 5mm2

and are mounted in a ceramic case of 9� 11mm2.The EMCal detector [2] of ALICE is a large acceptance

lead/scintillator sampling calorimeter with ‘‘shashlik’’-typereadout via wavelength shifting fibers and consists of

www.elsevier.com/locate/nima

dx.doi.org/10.1016/j.nima.2006.05.246

mailto:[email protected]

ARTICLE IN PRESS

Fig. 1. Digital resolution over PHOS dynamic range.

Fig. 2. Various contributions to the EMCal energy resolution.

H. Muller et al. / Nuclear Instruments and Methods in Physics Research A 565 (2006) 768–783 769

12 672 readout towers. The fiber bundle of each tower isread out with a S8148 APD operated at the ambienttemperature of the ALICE experiment (+20 1C).

Tests of the FEE electronics described here have beenperformed with S8148 APDs, excited with photon pulseswhich imitate the characteristics of PbWO4 scintillationlight via LEDs of type Kingbright L7104PCB with 470 nmpeak wavelength. The number of APD photoelectrons perMeV at gain M ¼ 1 can be estimated from the PbWO4 lightyield measurements [1] taken with a photomultiplier XP2262b at room temperature. By taking detector surfacesand quantum efficiencies of photomultiplier and APD inrelation for a nominal PbWO4 light-yield of 10 photoelec-trons/MeV at +25 1C one obtains

NAPD ¼ NPMT � ðSAPD=SPMTÞ � ðEAPD=EPMTÞ

¼ 10� ð25=484Þ � ð0:7=0:25Þ ¼ 1:45 e=MeV:

At the PHOS operating temperature of �25o C the light-yield is about 3 times higher, hence

NPHOSAPDð�25 �C;M¼1Þ�4:4 e=MeV:

With a nominal APD gain for PHOS of M ¼ 50, the APDcharge signal is

NPHOSAPDð�25 �C;M¼50Þ�220 e=MeV:

Preliminary results from test beam measurements for theEMCal gave corresponding numbers for the EMCal withAPD gain M ¼ 1 and 30 of

NEMCAPDð25 �C;M¼1Þ�4:4 e=MeV and

NEMCAPDð25 �C;M¼30Þ�132 e=MeV:

It is accidental that PHOS and EMCal obtain the samelight yield figure. With a 1.2 pF charge capacitor of thepreamplifier, the charge conversion gain is 0.83V/pC andthe preamplifier output step voltages for PHOS andEMCal at APD gain M ¼ 50 and 30 are:

UPHOSstepð�25 �C;M¼50Þ�29:2 mV=MeV and

UEMCstepðþ20 �C;M¼30Þ�17:5 mV=MeV

respectively.

2. Dynamic range

The dynamic range of PHOS has been defined by therequirement of 80GeV maximum single channel energyand 250GeV maximum single channel energy for EMCal.Given the 16:1 ¼ 24 gain ratio of the high and low gainchannels, the two overlapping 10-bit ADC ranges corre-spond to an effective 14-bit dynamic range.

High gain range : 4:9MeV to 5GeVðPHOSÞ and

15:3MeV to 15:6GeV ðEMCalÞ

Low gain range : 78MeV to 80GeV ðPHOSÞ and

248MeV to 250GeV ðEMCalÞ.

The minimum value corresponds to the Least SignificantBit (LSB) of the high-gain ADC. The digital resolution isgiven by the LSB value relative to a signal value of energyE. For the above PHOS ranges, the digital resolutionsLSB/E ¼ 4.9/E (MeV) for high gain and 78/E (MeV) forlow gain are shown in Fig. 1, compared to first testbeamdata taken by PHOS with old electronics [3]. Apart from asmall region above 5GeV, the digital resolution lies alwayssignificantly below the targeted PHOS energy resolution athigh energies of 3%/OE � 0.8%. The combined noise from3� 3 crystal areas of 7.5MeV/ E (2.5MeV single channelnoise) lies closely above the digital resolution of the highgain shaper.The intrinsic energy resolution of the EMCal is expected

from simulations to be about 6.9%/OE � 1.4% [2]. Therelative contribution of the digital resolution to the totalEMCal resolution is shown as bottom dual-range curve inFig. 2. A 1% constant contribution due to relativecalibration is assumed. The electronics noise has been

ARTICLE IN PRESSH. Muller et al. / Nuclear Instruments and Methods in Physics Research A 565 (2006) 768–783770

assumed to be 2000 electrons per APD for 100 ns shapingtime. The dotted curve represents the corresponding 3� 3electronics noise contribution term of 160MeV/E which isnegligible to the intrinsic resolution (top solid line).

3. Shaper gain

The shaper gain is defined with the above choices sincethe maximum energies must correspond to the maximumanalogue input ADC Voltage of 1.0V. With a nominalpreamplifier output of 29.2 mV/MeV for PHOS with APDgain M ¼ 50, the two gain ranges are defined as

High gain: LSB ¼ 143 mV (5.0MeV), Umax ¼ 0.146V(5GeV) �4shaper gain ¼ 6.85

Low gain: LSB ¼ 2.28mV(78MeV), Umax ¼ 2.34V(80GeV) �4shaper gain ¼ 0.427.

And in the case of EMCal with the above nominalpreamplifier output of 17.5 mV/MeV and APD gainM ¼ 30, the two gain ranges are defined as

High gain: LSB ¼ 267 mV (15.3MeV), Umax ¼ 0.273V(15.6GeV) �4 shaper gain ¼ 3.66

Low gain: LSB ¼ 4.34mV(248MeV), Umax ¼ 4.37V(250GeV) �4 shaper gain ¼ 0.229.

The real implementation does not need to reproducethese exact gain values since the assumptions about light-yield are only nominal and may be adjusted via the APDgains.

4. Energy resolution

The energy resolution of the calorimeter is parameterizedas [1]

sE¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffia2

E2þ

b2

Eþ c2

s

with the following parameters:

�
noise term a: parallel and serial noise from detectordiode and preamplifier; � stochastic term b: shower fluctuations and APD excess
noise;
�
1www.mathworks.com.

constant term c: detection losses, inter-calibration ofchannels, non-uniformity, non-linearity, instabilities(temperature and bias).

In the case of PHOS, a fit to PHOS test beam data [3],gives the following preliminary values: a ¼ 13MeV (3�3crystals), b ¼ 3.58%GeV0.5 and c ¼ 1.12%. For the highenergy region E 4 1GeV, the fit parameters arecompatible with

sE¼

3%ffiffiffiffiEp � 0:8%.

The interest of PHOS in a high signal/noise ratio is tominimize the noise term a in order to achieve a good energyresolution at low energies. In particular, the two-photon

invariant mass peaks of the p0 (134.97MeV) and Z(547.75MeV) are important for yield measurements andfor energy calibration of the calorimeter [4].

5. Channel readout

Fig. 3 depicts the readout principles for one FEEchannel. An APD is biased via a resistor Rp to a voltagewhich corresponds to the required gain. The chargeproduced by the photodiode is the product of number ofphotoelectrons and APD gain. As noted above for PHOSwith gain M ¼ 50, the nominal detector charge isQ ¼ 220 e/MeV and 132 e/MeV EMCal with M ¼ 30. Forfurther signal analysis, the charge dynamics is approxi-mated as a delta function, although in reality, thescintillating light of the PbWO4 crystals has a finite riseand decay time (of order 1 and 10 ns, respectively). Ideally,the charge sensitive preamplifier (CSA) converts the deltacharge into a voltage step of amplitude U ¼ Q/Cf. Inreality the step function has a finite risetime of order 20 nswhich is also limited by the finite risetime of the CSA. Forsimplicity, the capacitor Cf is auto-discharged via a resistorRdis with a time-constant which is very large (of order100 ms) compared to the time of measurement. The CSAand photodiode are physically close to the calorimeter,hence operated at the detector’s ambient temperature.The CSA step voltage signal is connected via a short

cable to the shaper and digitizer of the FEE electronicswhere it is amplified by a low noise buffer, followed by twoparallel CR-RC2 shapers of equal time constant and gainratio of 16:1. The products of the buffer gain (g) and theshaper gains (HG, LG) correspond to the gains requiredfor the given dynamic range. For each shaper section ofhigh and low gain, there is one 10-bit ADC to digitize thedata.

6. Shaper bandpass

The FEE shaper is a second-order bandpass filter for acentral frequency of fc ¼ 1/(2p�t) where t is the shapingconstant. It consists of a simple RC high-pass of �3 dB/octave attenuation below fc, followed by two low-passfilters with �6 dB/octave above fc. The combined bandpassenhances the Signal-over-Noise ratio by separating the‘‘noise sea’’ from the signal’s main Fourier components.The choice of the cutoff frequency fc (or shaping constant)defines the Signal-over-Noise ratio. The shaper’s Bodediagram of Fig. 4 represents the frequency domain transfercharacteristics, which is strictly linked to the shaper’s timedomain behaviour or output pulse shape discussed later.The Bode diagram of Fig. 4 was obtained by simulatingthe circuit diagram of the shaper with Matlab.1 It depictsthe high pass (dotted) and lowpass (solid red) transfercharacteristics as well as the combined bandpass transfercurve (interrupted line). The peak frequency at

http://www.mathworks.com

ARTICLE IN PRESS

Fig. 3. Photodiode and preamplifier, followed by dual gain shaper and ADCs.

Fig. 4. Bode diagram of the PHOS shaper bandpass for t ¼ 1ms.


fc ¼ 160 kHz corresponds with fc ¼ 1/(2pt) to a shapingtime of t ¼ 1 ms.

7. Signal-over-noise

With the aim to achieve a good energy resolution also atlow energies, the shaper was designed with a low-noiseinput buffer stage. The noise of cascaded amplifiers withN1 and N2 noise levels is calculated as

Nnoise ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiN12 þ

N2

K1

� �2s.

K1 is the amplification of the first buffer stage. By choosingK141 the contribution of the shaper noise N2 can bereduced by choosing a buffer with a low-noise figure N1.The standard noise analysis [5] is depicted in Fig. 5.

The noise components are:

�
Current noise (parallel): Shottky 2qIdiode+shunt resistorthermal current 4kT/Rb + equivalent input current. � Voltage noise (serial) 4kTRs thermal Johnson noise at
the amplifier input.
� 1/f noise.
The Equivalent Noise Charge (ENC) at the output of theshaper is a function of the shaping time constant t, of theabsolute temperature T, of the parallel, serial, and constantnoise sources, and of the detector capacitance. The currentnoise is proportional to the shaping time t, the voltagenoise is inversely proportional to t, and the 1/f noise isindependent of t. With an estimate for Rs ¼ 2/3� (1/gm)where gm is the forward transconductance of the JFETtransistor of the CSA, the ENC as function of shaping timeis given by

ENC2¼

4 kT

q2 � Rb

� Fp � tþ4 kT

q2� 2=3 �

1

gm

� F s �C2

d

t

þ C2d � const:

For the PHOS electronics the following values apply:Rb ¼ 20MO, gm ¼ 30mS, and Cd ¼ 100 pF (APD ¼ 90 pFat �251C, JFET ¼ 10 pF). The noise figures [6] for a CR-RC2 shaper are: Fs ¼ 0.84 and Fp ¼ 0.63.The ENC noise in dependence on the shaping time t

(without the negligible constant 1/f noise term) is shown inFig. 6 for EMCal (+20 1C) and for PHOS (�25 1C ). Thenoise minimum lies at 2–3 ms. At shorter shaping times, thevoltage noise dominates while at higher shaping times,the current noise dominates. PHOS has nevertheless chosena shaping time constant of 1 ms as a compromise forimproved offline timing resolution (see Section 17.4). Theachieved average RMS noise level of 0.25 ADC counts (seeSection 16.3) corresponds to 39 mV at the output of thepreamplifier, or with Nq, noise ¼ 1/q (U�Cf ) this corre-sponds to 292 noise electrons. Corrected for APD gainM ¼ 1, this corresponds to a noise level of ca. 6 primaryphotoelectrons.In the case of the EMCal, the intrinsic resolution is

worse than PHOS and so the electronics noise contri-bution is less important. As shown in Fig. 2, an electro-nics noise contribution of as much as 2000 electronsper channel is not significant relative to the intrinsicresolution. Therefore EMCal has chosen a much shortershaping time of 100 ns to suppress energy deposit fromlate event-related neutron background, and to reducethe data volume per channel. The calculated noise forEMCal would correspond to the upper curve (+20 1C) ofFig. 6.

ARTICLE IN PRESS

Fig. 5. Equivalent noise of detector and preamplifier, followed by the shaper

Fig. 6. Calculated ENC noise as function of shaping time for PHOS at �25 1C and EmCal +20 1C.

Fig. 7. Full diagram of the analogue stages of a CR-2RC shaper.

H. Muller et al. / Nuclear Instruments and Methods in Physics Research A 565 (2006) 768–783772

8. Shaper analysis

The Laplace representation of an nth order shaper oftype CR-RCn with step function input from a chargesensitive amplifier is a product of equivalent operators.With the Laplace equivalent ‘‘s’’ for the step voltage, thepreamplifier-shaper chain of Fig. 7 can be transcribed withshaping constant t0 and shaper gain A into an Laplaceoperator H(s) which represents the full chain of the FEEelectronics analogue part

HCSAþshaperðSÞ

¼ s�1

1þ ð1=RdisCf Þ

� ��

t01þ st0

� �

�A

1þ st0

� �n

.

The first bracket term represents the ‘‘auto-discharge’’combination Rdis�Cf of the CSA. As shown below, thisterm can be eliminated by including the ‘‘pole-zero-cancellation’’ RC network with the inverse operator in

the input of the shaper. The second bracket is the Cp�Rd

highpass (differentiator) and the third term are n equallowpass (integrator) stages. The shaping time constants t0of differentiator and integrators are equal. The generalsolution [6] of this Laplace operator in the time domain is aGamma function of order n

VnðtÞ ¼nnQ � An

Cf � n!

� ��

t

tp

� �n

� e�nt

tp .

ARTICLE IN PRESS

Table 1

Values of differentiator and pole-zero filter for 1 and 2 ms shaping time

Shaper constant t0 Rz Cz R0

t0 ¼ 2 ms 143 k 470pF 9.1 k

t0 ¼ 1 ms 143 k 470pF 4.22 k

Fig. 9. Oscilloscope output of uncompensated step pulser time response

for two shaping times: 1 and 2ms.

Fig. 10. Dual integrator circuit with multi-feedback architecture.


The zero-crossing of the first derivative shows that the peakamplitude is proportional to the APD charge Q, hence tothe scintillation light

Vmax ¼Q � An � nn

Cf � n! � en.

For the n ¼ 2 second-order CR-RC2 shaper, the pulseshape Vn(t) in the time domain is described with a peakingtime tp at Vmax. The relation between peaking time andshaping time is tp ¼ n�t0, hence for the FEE electronicswith n ¼ 2, the peaking time is twice the shaping time andthe pulse shape is

VoutðtÞ ¼2Q � A2

Cf

� ��

t� t0

tp

� �2� e�2�ðt�t0=tpÞ.

Please note that this solution is the time domain-equivalentrepresentation of the frequency diagram shown previouslyin Fig. 4.

9. Shaper implementation

The shaper highpass (differentiator) stage can simply beimplemented as an RC filter with R0�Cz ¼ 1/(2p�fc) atthe input of the low noise buffer (Fig. 8). In order to cancelthe first bracket operator of the Laplace operator above, itis sufficient to add a resistor Rz in parallel to Cz and choosethe combined time constant equal to Rdis�Cf. The finalvalue of Rz is deviating slightly from the theoretical value;it was determined by measurement, i.e. by adjusting thesignal Vout(t) to zero undershoot.

The combined resistor values of the differentiation andpole-zero cancellation are listed in Table 1 for two shapingtimes.

The effect of the CSA auto-discharge term is demon-strated in Fig. 9 by replacing the CSA output with a steppulse generator: the pole-zero cancellation filter whichwould normally compensate the signal overshoot producesthe opposite undershoot. The fully compensated FEE pulsewith CSA is shown later in Fig. 12.

The shaper lowpass was implemented as a dualintegrator stage using a single operational amplifier with

Fig. 8. Differentiator and pole-zero cancellation stage of the shaper input

stage.

multiple feedback architecture (MFB) as shown in Fig. 10.The values of the components are obtained as solution ofthe transfer equations, described in the following sections.The dual integrator transfer function can be expressed [7]

with gain K as a second order polynomial of the Laplace s

operator

HðsÞ ¼K

1þ asþ bs2.

For a flat Bessel filter transfer characteristics of second-order, the following conditions for the coefficients of thepolynomial apply:

K ¼ R2/R3 (gain of dual integrator for f 5fc)a ¼ ocC1 (R2+R3+R2R3/R1) ¼ 1.3617b ¼ oc

2 C1C2R2R3 ¼ 0.618.


In order to avoid complex solutions the capacitor valuesmust fulfil the condition

C2XC1 �4b � ð1� KÞ

a2.

The solutions for the resistances, given as a set of capacitorvalues and a shaper gain K for a cutoff frequency fc ¼ 1/(2p* t0) with shaping time constant t0, are shown below

R2 ¼aC2 �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffia2 � C2

2 � 4bC1C2 � ð1� KÞ

q4pf c � C1C2

R1 ¼R2

�K

R3 ¼b

4p2f 2c � C1C2R2

f c ¼1

2p �ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiR2 � R3 � C1 � C2

p �1

FSF.

The frequency scaling factor is FSF ¼ 1.2736. The valuesobtained for 100 ns and 1 and 2 ms shaping time for highand low gain are shown in Table 2.

The full implementation of one shaper channel, includ-ing pole-zero-compensation, is depicted in Fig. 11 for the

Table 2

CR-RC2 shaper component values for three shaping time constants

t0 R1 R2

2ms (high gain) 649O 4.42 kO2ms (low gain) 10.5 kO 4.22 kO1ms (high gain) 681O 4.87 kO1ms (low gain) 4.02 kO 1.69 kO100 ns (high gain) 78.7O 316O100 ns (low gain) 590O 147O

Fig. 11. Detailed schematics of one dual-

2 ms shaping time ( ¼ 4 ms peaking time). A low-noisebuffer of gain 2 with RC highpass as shown in Fig. 8 isfollowed by two parallel, 2nd-order lowpass filters of gainratio 16:1, each implemented with the same Bessel filtertransfer characteristics and with the same shaping time.Their differential output, required for the ADCs, isimplemented by inverting and non-inverting drivers.Analogue anti-aliasing filters of ca. 4MHz cutoff frequencyare inserted in the input lines of the ADC in order toremove out-of-band components, due to digital ADCswitching noise.

10. Digitization and triggered readout

As shown in Fig. 4, the �10 dB attenuation of the PHOSshaper bandpass is around 500 kHz, i.e. the Nyquistfrequency (2� signal bandwidth) is ca. 1MHz, hence asampling frequency of fs ¼ 2MHz would be sufficient. Forthe measurements presented here, over-sampling at10MHz was applied in order to study timing resolutiondependencies. The ADC is an ALTRO chip [8] containing10 bit ADC’s which are followed by multi-event samplingbuffers of up to 512 samples. The loading with ADCsamples requires a timing strobe to start transfer of samplesfrom a maximal 15-deep pre-sampling pipeline to the

R3 C1 (pF) C2 K

1.78 kO 150 2.2 nF 3.35

5.90 kO 220 470 pF 0.21

1.96 kO 68 1000 pF 3.35

5.36 kO 150 470 pF 0.21

205O 100 1000 pF 1.95

442O 220 470 pF 0.12

shaper channel for 2ms shaping time.

ARTICLE IN PRESSH. Muller et al. / Nuclear Instruments and Methods in Physics Research A 565 (2006) 768–783 775

buffer. The timing strobe (derived from the trigger of theexperiment) must not arrive later than 5/fs(MHz) ms after thesignal in order to store at least 10 pre-samples formeasuring pedestals and RMS noise of the signal. Thedigitized pulse-shapes are read out from the data buffersvia the protocol of the custom, 40-bit ALTRO bus whichwas adopted from the ALICE TPC. One readout partitionconsists of 2� 14 FEE and 2 TRU cards and is masteredby a Readout Control Unit (RCU) [9] with a networkedDCS processor card [10]. The memory mapped configura-tion and readout protocols provide remote access to allFEE register resources, defined by programmable slavelogic (PCM) of the FEE cards. The data transfer from databuffers to the offline computers is established via thecustom, optical DDL link [11] of the ALICE experiment.The software framework for the ALICE Data Acquisitionis DATE [22] which can be conveniently used also in testenvironments for reading raw data files from front endelectronics onto disk.

11. Offline analysis

For offline data analysis, the ROOT offline analysispackage [12] developed at CERN was used. Fig. 12 depictsthe ROOT reconstruction of a typical, 2GeV equivalentLED pulse. This example shows the high-gain shaperoutput with 1 ms shaping time, sampled at 10MHz. TheGamma-2 fit is in all cases in very good correspondencewith the data, from the first pre-samples up to sample 60.Fits beyond this level of the down-slope of the signal startto deviate from the data due to the pole-zero compensationeffect. In the test environment, the ADC strobe latency wasset to 15 pre-samples, allowing determination of pedestalsand RMS noise on an event-by-event basis.

After pedestal correction, the signal peak amplitude ofthis event is 410 ADC counts and the peaking timecorresponds to 2 ms. Since the dynamic range definition ofPHOS implies that 5MeV corresponds to 1 ADC count,the peak amplitude corresponds to 2.05GeV.

Fig. 12. Offline high gain ADC counts versus samples [100 ns]. The red

line is a Gamma-2 fit.

12. FEE prototype response to electrons and protons

The first FEE card prototypes were tested with atriggered electron beam, directed to a 16� 16 PbWO4

crystal matrix [4] in August 2004, and with mixed beams ofprotons, pions, and electrons with momenta of 4–120GeV/cdirected to an 8� 8 array of EMCal towers in November2005. Fig. 13 shows the digitized pulse shape output with10MHz sampling (100 ns/channel) for an electron eventin PHOS (right) and a proton event in EMCal (left).The different peaking times of 2 ms and 200 ns, respec-tively, corresponding to the different shaping times areapparent.The test beam results confirm the FEE gains and shaping

times and allow extracting the light yields from thecalibrations in terms of MeV/ADC count and the overallenergy resolution including the FEE noise contribution.Gain adjustments within a factor 2 are easily accomplishedvia APD bias adjustment.

13. FEE card

The architecture of FEE electronics within the PHOS read-out and trigger system has been described in Refs. [13,14].The photograph of Fig. 14 depicts the 10 layer FEE card withall its details. The total power consumption is ca. 5W.Special attention was given to a hierarchical power

filtering and careful layout of ground planes for analogue,digital, and high voltage sections, avoiding coupling ofdigital noise into the analogue shaper sections. Withtemperature, voltage and current monitoring, individualFEE cards within a closed detector volume can be disabledin case of power or temperature problems. The cards aredesigned for operation inside water-cooled copper cas-settes. For this purpose, the FEE component layout leavesroom for inlet and outlet water pipes and contains holes formounting screws of the surrounding cassette.

14. Test environment

A test environment with APD diodes requires ideally thesame conditions as exist in the calorimeter, i.e. a stableambient temperature and a pulsed photon light source withsimilar wavelength and timing characteristics as thescintillating light produced by electromagnetic showers inPbWO4 crystals. The corresponding test setup of the PHOSlaboratory in Wuhan is shown as a simplified diagram inFig. 15.The inputs of the FEE card are connected to an array of

APD/CSA in a testbox which is kept inside a refrigerator ata temperature of �25 1C in order to mimic the PHOSoperating conditions. A pulse generator is used to triggeran avalanche LED pulser, containing a LED with 470 nmpeak spectral emission. The emitted light-pulse is distrib-uted via optical fibers to the APD array. All APDs havebeen set via bias voltage registers to a gain value whichgenerates the same amplitude for the common light input.

ARTICLE IN PRESS

Fig. 13. FEE electronic pulse shapes (100 ns/channel) with testbeam: Left-EMCal, Right-PHOS.

Fig. 14. Photograph of the 32 channel FEE card

Fig. 15. Test environment in the Wuhan PHOS laboratory at CCNU.



Upon reception of the trigger output from the trigger pulsegenerator, the Readout Control Unit (RCU) sends a timingstrobe to all ALTRO chips to store ADC samples of allselected channels into their sample buffers. Thereafter,these buffers are transmitted, via RCU and the optical fiberDDL link to a Data Acquisition Computer running theDATE software. The ROOT analysis tools are used to sortthe sampled ADC data into histograms and to performuser-defined data analysis. The results are shown anddiscussed in the following chapters.

14.1. LED pulser

An avalanche LED pulser [15] was built for the test setupto generate up to 109 photons of 470 nm for distribution viaoptical fibers with a 1–2 ns light risetime. Fig 16 shows a�20V LED pulse amplitude of o1 ns risetime and theAPD testbox with optical fibers, fanned out from thepulsed LED.

The charge/voltage step response of the CSA, whichcorresponds to an (attenuated) photon flash from a pulselike Fig 16, transmitted over ca. 50 cm of optical fiber tothe APDs, is depicted in Fig. 17. The LED (KingbrightL7104PCB) is specified for typically 350mcd at 20mA.When measured over a series resistor, a pulse correspond-ing to 10A peak current can be converted over lumen andhuman eye correction factors into ca. 8� 109 photons. Thestep function of the attenuated light flash arriving overoptical fibers at the APDs, shows a slight overshoot which

Fig. 16. Left: LED voltage pulse at the pins of a Kingbright L7104PCB/

Right: APD test box with optical fibers facing each APD.

Fig. 17. CSA response to a LED pulse.

is due to the test-setup. This effect is not observed with realparticles in the testbeam.

15. Fast analogue OR

A threshold trigger can be based on analogue sumsgenerated on the FEE cards from 2� 2 input channels. Atrigger region unit (TRU) has been designed [16] to receiveall fast OR signals from 14 FEE cards and to generatecharge sums over 4� 4 sliding windows over 448 channels.In order to generate an analogue sum from 4 inputchannels, the input signals on the FEE cards are split(compare Fig. 11) into semi-gaussian energy channels andinto fast-OR trigger channels with 12 bit dynamic range.The fast ORs are generated by a simple RC shaper with100 ns pulse width.

16. Measured FEE characteristics (PHOS 1 ls shapingtime)

The measurements presented here for the PHOS detectorwith the choice of 1 ms shaping time are results of theROOT offline analysis with data generated via thedescribed test system and read out via a computer foroffline analysis.

16.1. Linearity and gain over dynamic range

The theoretically linear relation between the peakamplitude of the Gamma-2 function and the APD charge(input voltage step amplitude) was confirmed by themeasurement of Fig 18.The lower curve represents the full low gain range

(2.3mV–2.39V input step) the upper line shows the highgain range up to 0.149V, extrapolated to input stepvoltages below 10mV. The voltage levels below 10mVcould not be measured precisely with our oscilloscopes.

Fig. 18. PHOS linearity measurement over dynamic range.


The two shaper gains can be derived from the slopes ( i.e.p1 coefficient) after converting ADC counts in Voltageaccording to 1ADC bit ¼ 1V/1024 ¼ 0.97656V:

�
High gain, measured ¼ 6.770.051, � Low gain, measured ¼ 0.39770.001.
These values correspond with the design values definedin Section 2. The gain ratio of 16.87 is close to the valueobtained later in Section 16.5 by direct measurement.

16.2. APD gain calibration for PHOS

The gain spread of the APDs for a common bias voltageof 340V was measured in a PHOS testbeam October 2004with the first FEE prototypes with 2GeV electrons.Individual gain control via 10-bit bias voltage registers[17] on the FEE cards allows for controlled gain calibrationof 0.20V per bit in the range 210–400V. The measurementof Fig. 19 shows the relation between APD bias Voltageand remote controlled DAC settings

Vbias ¼ 209:9þ 0:2022 �DAC:

Fig. 19. Bias register counts versus APD bias voltage.

Fig. 20. APD gain spread at �25 1C with 2GeV partic

With a gain dependence on the voltage of 1/M � (dM/dV) ¼ 3.3%/V for gain M ¼ 50 [18] and a bias voltageincrement 0.20V/bit, the gain resolution (T ¼ constant )is 0.66%. Fig 20 shows the gain distribution before andafter FEE gain calibration, i.e. after individual biasadjustment via bias control registers. The un-calibratedvariations amount to more than 300%. After (non-exhaustive) hardware pre-calibration the variation wasreduced to ca. 5%.A direct determination of the APD gain versus bias

voltage correspondence of individual APDs requiresknowledge of their signal amplitude for gain M ¼ 1. Bymeasuring the signal amplitude at gain M ¼ 1, andmultiplication by the desired gain, one obtains thecorresponding bias voltage. Such a measurement is shownin Fig. 21 over the full bias voltage range for two APD’s.This determination of the gain M ¼ 1 is not very preciseand depends also on the light wavelength. As described inRef. [18], APD gain M ¼ 1 is should be measured at theinitial plateau of the photocurrent between 10 and 50 Volt.With this, the nominal gain of M ¼ 50 is reached at ca.350V for the two APDs.In view however that an inter-calibration of better than

1% is aimed for in the constant term of the energyresolution term of the PHOS and EMCaL detectors, thismethod of gain calibration is only useful for pre-calibra-tion, while inter-calibration to the o1% level requirescalibration with electromagnetic showers.

16.3. Pedestal and RMS measurement

The pedestal and RMS measurements are taken from N

pre-samples of the digitized signal as shown in Fig. 12,measured over a sample of events. The RMS for a singleevent is calculated as

RMSðNÞ ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1

N

Xiðyi �MeanÞ2

r

where Mean is the average of the N samples. With N ¼ 10pre-samples like shown in Fig. 12 the measurement in thelaboratory with the described PHOS test setup gave the

les (left) and after hardware pre-calibration (right).


following results: For an unconnected FEE card, Fig 22(left) shows the pedestal distribution over all 32 high gainFEE shaper channels (bottom) and their derived RMSnoise (top) for 2050 events. The constant spread of pedestallevels is normal, due to component tolerances. The uniformRMS noise level is around 0.22 ADC counts. Afterconnecting FEE channels 24–31 to 8 APD detectors in an

Fig. 22. Left: Pedestal and RMS noise measurement for a FEE card without d

with 8 APDs connected and 24 channels unconnected. Note that the pedestal

Fig. 21. APD gain measurement for two different S8148 APD’s relative to

their M ¼ 1 gain plateau below 50V.

absolute dark environment with gain setting M ¼ 50 at�25 1C the noise distribution taken from pre-samples ofLED pulses is a little more disturbed as shown in Fig 22(right). There is a clear RMS noise enhancement in theAPD channels over the unconnected channels to anaverage of 0.23 ADC counts.With the dynamic range definition of PHOS, 1 ADC

counts corresponds to 5MeV, hence the single channelRMS noise for PHOS is 1.16MeV. This figure issignificantly better than measured with first FEE proto-types in the 2004 testbeam. For the final noise situationinside the PHOS module we assume 2.5MeV as a moreconservative noise figure, hence the PHOS noise term for a3� 3 crystal area can be assumed with new FEE electronicsas a ¼ 7.5MeV (compare Section 4).

16.4. APD noise as function of bias voltage

As shown in Section 7 the FET voltage noise dominatesat shaping times below the minimum noise, hence the APDcurrent has little influence at shaping times below 2 ms.However the APD capacitance is a strong function of theAPD bias voltage with a strong increase of capacitance andnoise towards low voltages. Towards high bias voltages,the excess noise, due to charge avalanche, increases theeffective noise level.The excess noise is a statistical process which broadens

the signal resolution, hence belongs to the stochastic term‘‘b’’ of the calorimeter energy resolution. Given an excess

etector connected; top: RMS noise averaged over 2050 events; right: same

s are unchanged.

ARTICLE IN PRESS

Fig. 24. Uniformity of gain ratio High/Low.


noise factor of F ¼ 2.27 for gain M ¼ 50 it can be shownthat [18]

bffiffiffiffiEp ¼

ffiffiffiffiffiffiffiffiF

Npe

s.

With Npe ¼ 4 e�/MeV and F ¼ 2.27 at (M ¼ 50), the lowerlimit of the single channel stochastic term at 1GeV can beapproximated as b42.3%GeV1/2.

Fig. 23 depicts an all inclusive noise measurement versusAPD bias voltage for 8 APDs at the PHOS ambienttemperature of �25 1C. The effect of excess noise can beseen for some APDs already above 380V. The averagenoise at 350V (M ¼ 50) corresponds here to 0.3 ADCcounts (1.5MeV).

16.5. Gain ratio of high /low gain

The design gain ratio for FEE cards (high gain/lowgain ¼ 16/1) of the two shapers sections cannot be imple-mented exactly as 16/1. The 32-channel gain ratio measure-ment of Fig. 24 reports a value of 16.83 which effectivelyincreases the dynamic range to 14.075 bits. This measurementalso demonstrates that excellent uniformity has been achievedwith the discrete component shapers of the FEE cards.

16.6. Peaking time measurement

The design peaking time of the second order PHOSshapers with 1 ms shaping time is 2� t ¼ 2 ms. Fig. 25

Fig. 23. Measurement of effective noise (ADC counts) versus APD bias

voltage for 8 APDs, read out and digitized by FEE electronics.

Fig. 25. Peaking time measurement of the PHOS FEE shaper with 1 msshaping time.

shows the measurement taken with 8 APDs pulsed by asingle LED. The measured mean value is 2.17570.01 ms.

17. Timing resolution

Great attention was given to the PHOS requirement fora timing resolution of the order 1–2 ns, for time-of-flightcuts that will be used to discriminate against anti-neutronbackground, predominantly around 2GeV. The timingresolution achievable with semi-gaussian shapers is a weakfunction of noise and shaping time but a strong function ofthe signal amplitude (ADC bits) and of the samplingfrequency. Considerable improvement was gained by usingthe proper offline method and by using a LED pulser withlight risetime of the same order or better than the targetedresolution.The intrinsic timing resolution of PbWO4 crystals is

reported as 0.13 ns [19] and timing measurements intestbeam [4] report a constant timing resolution of 0.5 nsabove 1.5GeV. The only affordable way of measuringtime-of-flight in each individual FEE channel is by offline


determination of the time reference obtained from theGamma-2 fit. In order to achieve the 1 ns level of timingresolution, it is important to first determine the peakingtime constant of each individual channel and then use thesevalues for the Gamma-2 fit.

Fig. 28. Timing resolution dependence on equivalent energy (ADC

counts).

17.1. Timing resolution of FEE card

The method applied for measuring the differential timingresolution of the FEE alone is depicted in Fig. 26. A stepvoltage of equivalent energy E is connected to the inputs oftwo or more shapers. The differences of the time referencest1 and t2 of the offline fits are compared and the differencesare accumulated in a histogram. With sufficient statistics,the variance of the difference plot sðt2 � t1Þ is a measure ofthe differential timing resolution. Assuming that bothchannels have the same resolution, the timing resolution iss/O2.

The measurement of the differential time resolution of 2FEE channels without detector, using a step pulser of2GeV equivalent energy is shown in Fig 27. With 2050events and a sigma of 0.712, the differential time resolutionof PHOS FEE electronics at 2GeV is 0.5 ns

The energy dependence of offline timing resolutionmeasurement is shown in Fig. 28 for pulsed FEE channels.

Fig. 26. Principle of measuring differential time resolution measurement.

Fig. 27. Offline measurement timing resolution of pulsed FEE channels.

It is due to the digital resolution limit of the dynamic range,as shown in Fig. 1. In the limit of 10-bit signals (1024counts), the FEE electronics can resolve 300 ps at a shapingtime of 1 ms. This result hints that for future upgrades, thedigital resolution should be increased to a dynamic rangeof 16-bits.

17.2. Timing resolution with APD and LED source

The timing resolution with particles and APD diodes canbe measured in good approximation by operating theAPD’s in the same bias, temperature, and connectivityenvironment as in the experiment and by pulsing a LED ofsimilar wavelength and timing characteristics as thescintillation light of a PbWO4 detector. The test environ-ment described previously provides this possibility. Thetiming resolution measurement is depicted in Fig 29 (highgain) and Fig 30 (low gain). The equivalent LED lightenergy was ca. 2GeV (compare Fig. 12 for the amplitude /energy equivalence of Run 157).The differential time resolution of the high gain shaper is

1.5 ns and of the low gain shaper is 6.22 ns. The combinedhigh/low timing resolution for 1 ms shaping time at 2GeVequivalent energy is

tresð1msÞ ¼ 1:455 0:026 ns:

The difference in resolution with and without APD/preamplifier of ca. 1 ns. This difference is due to thelimited timing resolution of the APD (order 400 ps) and thefinite risetime of the LED light.

17.3. Timing resolution dependence on sampling frequency

The timing resolution obtained with a LED source isshown in Fig 31 as a function of ADC sampling frequency.The plateau above 1MHz sampling frequency complieswith the Nyquist theorem that a sampled signal is fullydefined by sampling it at twice the signal bandwidth. Asshown in Fig. 4, the �10 db signal bandwidth above fc is ca600 kHz for 1 ms shaping time. With 10MHz sampling rate,

ARTICLE IN PRESS

Fig. 29. Histogram of high gain timing resolution at ca. 2GeV with LED

pulser.

Fig. 30. Histogram of timing resolution measured with the low-gain

shaper channel. The resolution is worse compared to the high gain channel

due to the low digital resolution at 2GeV.

Fig. 31. Measurement of timing resolution dependence on the ADC

sampling frequency.

Fig. 32. Comparative measurement of Fig. 21 but with 2 ms shaping time

(4ms peaking time).


semi-gaussian signals for 1 ms shaping time are by far over-sampled. Unless over-sampling [20] methods can be usedoffline to improve timing resolution (by gaining moreeffective bits), a sampling frequency of 2MHz is sufficientfor 1 ms shaping time.

17.4. Comparative measurement with 2 ms shaping time

As depicted in Fig. 6, a shaping time of 2 ms should resultin even lower RMS noise. Fig. 32 confirms this for 2 msshaping time by a measurement under exactly the sameconditions as Fig. 22. Obviously the ENC noise from theshaper reaches the noise limit of the ADC at ca 0.2 ADCcounts.

The reduced noise with 2 ms shaping time entails howevera worse timing resolution of 1.76 ns as measured in Fig. 33

with a LED source under the same conditions as Fig. 29.The conclusion of the two comparative measurements isthat the RMS noise reduction has less importance forPHOS than the ca. 17% improved timing resolution, andtherefore a 1 ms shaping time has been adopted.

ARTICLE IN PRESS

Fig. 33. Comparative measurement of Fig. 28 with 2ms shaping time.


18. Conclusions and outlook

The 32-channel low noise FEE electronics presented herehas been designed for the high resolution PHOS electro-magnetic calorimeter of ALICE with a dynamic range of14-bits. The design performance for the PHOS calorimeterhas been verified by measurements in the dedicated testfacilities at CCNU and HUST in Wuhan, China. Withcooled APDs and a shaping time of 1 ms, the 0.25 ADCcount RMS noise level corresponds to 290 electrons at gainM ¼ 50 and the offline timing resolution is better than1.5 ns. The configurable shaper properties and the hard-ware gain calibration feature allow for adaptation to otherphoton detectors. The EMCal project of ALICE [2] hasadopted this electronics with 100 ns shaping time. Anupgrade towards a 16-bit dynamic range for better timingand noise performance and a direct, optical readoutconcept [21] via 1Gbit/s network protocols is underconsideration for new applications.

Acknowledgements

This work is supported by the ALICE LHC project atCERN, by the Norwegian Research Council and INTASProject 03-5747, by the National Natural Science Founda-tion of China (NSFC) under Grant reference 10575044,10547118 and the China Ministry of Education (CMOE)under Grant of the MoU between CMoE and CERN/ALICE and sponsored by SRF for ROCS, SEM, byCentral China Normal University (CCNU) under Grantreference (2005)80 and by the US Department of Energyunder Contract no. DE-AC05-00OR22725.

References

[1] Photon Spectrometer PHOS, ALICE Technical Design Report,

CERN /LHCC 99-4, 5 1999.

[2] The Electromagnetic Calorimeter, Addendum to the ALICE Tech-

nical Proposal, CERN-LHCC-2006-014, CERN/LHCC 96-32-Add3,

14 April 2006.

[3] D.V. Aleksandrov, et al., A high resolution electromagnetic

calorimeter based on lead–tungstate crystals, Nucl. Instr. and Meth.

A 550 (1–2) (2005) 169.

[4] M. Ippolitov, et al., Energy, timing and two photon invariant mass

resolution of a 256 channel PWO calorimeter, Proceedings CALOR

04 Conference, April 2004, Perugia, Italy.

[5] Particle detectors, low noise electronics, Rev. Part. Phys. 15 (1–4)

(2000) (Chapter 24.9).

[6] W.M.C. Sansen, et al., Limits of low noise performance of detector

readout front ends in CMOS technology, IEEE Trans. Circuits and

Syst. 37(11) (1990).

[7] Active Filter design techniques, Texas Instruments, Literature

Number SLOD006A (Excerpt from ‘‘O-Amps for Everyone’’)

(Chapter 16).

[8] The ALTRO chip: A 16-channel A/D converter and digital processor

for gas detectors, Proceedings of IEEE NSS/MIC, Norfolk,

November 2002.

[9] J.A. Lien, The readout control unit of the ALICE TPC, Thesis,

CERN-THESIS-2005-013.

[10] S. Bablok, et al., Front-end-electronics communication software for

multiple detectors in the ALICE experiment, NIM-A 557 (2) (2006)

631.

[11] G. Rubin, et al., The ALICE Detector Data Link, Article p493 in

CERN Report 99-9.

[12] ROOT, An Object-Oriented Data Analysis Framework, http://

root.cern.ch.

[13] H. Muller, et al., Front-end electronics for PWO-based PHOS

calorimeter of ALICE, NIM-A (2005).

[14] H. Muller, et al., Trigger electronics for the ALICE PHOS detector,

Nucl. Instr. and Meth. A 518 (2004) 525.

[15] B.K. Lubsandorzhiev, et al., LED calibration system of the Lake

Baikal Neutrino Telescope NT-200+, 29th International Cosmic Ray

Conference, Pune, 2005, pp. 101–104.

[16] Trigger Region Unit for ALICE PHOS Calorimeter, Proceedings

LECC05,11th Workshop on Electronics for LHC and future

Experiments, Heidelberg, September 2005, CERN-LHCC-2005-038,

pp 384–387.

[17] I. Sibiriak, A. Tsvettkov, A. Vinogradov, APD Power Control for the

ALICE PHOS Prototype, Alice Internal Note, ALICE-INT-2005-

013.

[18] A. Karar, et al., Characterization of avalanche photodiodes for

calorimetry applications, Nucl. Instr. and Meth. A 428 (2–3) (1999)

413.

[19] R. Novotny, et al., Response of a PbWO4 scintillator array to

electrons in the energy regime below 1GeV, IEEE Trans. 44(3) (Part

1) (1997) 477.

[20] J.C. Candy, G.C. Temes, Oversampling methods for data conversion,

IEEE Pacific Rim Conference on Communications, Computers and

Signal Processing, May 9–10, 1991.

[21] H. Muller, et al., Common gigabit interfaces for HLT and L1T

trigger links of LHCb, Proceedings of the 9th Workshop on

Electronics for LHC, Amsterdam, September 29–October 3, 2003,

CERN LHCC-2003-055, pp 139–143.

[22] Alice DAQ and ECS User’s Guide, Alice Internal Note/DAQ ALICE

–INT-2005-015, January 2006, Alice DAQ Project.

http://root.cern.ch

http://root.cern.ch

Documents

Configurable electronics with low noise and 14-bit dynamic range for photodiode-based photon detectors