Progress on signal processing techniques for Lumical ... · sources – laser, 90Sr source,...
Transcript of Progress on signal processing techniques for Lumical ... · sources – laser, 90Sr source,...
Progress on signal processing techniques for
Lumical testbeam and CLIC detectors
Szymon Kulis, Marek Idzik
Faculty of Physics and Applied Computer Science
AGH University of Science and Technology
Predeal, Romania | FCAL Collaboration Workshop 30.V - 01.VI.2011
Agenda
• Motivation & Goals
• Deconvolution principle
● Experimental & MC setup
● Preliminary Measurement & MC results
• Gated integrator concept
• Summary & Future plans
Motivation
• During next FCAL testbeam we will have to use technique which allows us to reconstruct amplitude (and maybe time) based on relatively rare samples (asynchronous sampling)
• In CLIC– Bunch crossing separation ~ 0.5 ns– No external event trigger will be available
• Detector subsystems will need triggerless readout electronics fulfilling a number of requirements:– Precise time information
– Good amplitude resolution
– Good pileup rejection
– Low power consumption
Readout electronics diagram - deconvolution
• Pulse at output of shaper v(t) is convolution of input signal (current from sensor – s(t) ) and impulse response of readout chain h(t):
• Using data from continuously running ADC and taking advantage of known pulse shape one can perform invert procedure – deconvolution – to get information about event time and amplitude
v t =∫−∞
∞
h t−x s x dx
Deconvolution for CR-RC shaping
• Only two multiplications and three additions (very fast and light !)
• Deconvolution produces non-zero data only when one or two first samples are on baseline,and second/third is on pulse
• Initial time of pulse is found from ratio of those samples
• Amplitude is found from sum of those samples, multiplied by time dependent correction factor
• Deconvolution reduces (infinite number) of CR-RC pulse samples to 1 or 2 non zero samples !
Look Up Tables used
Can be done off-line
CR-RC, Tsmp
=Tpeak
=1, amp =1
}
d i=siw1 si−1w2 s i−2
CR-RC Deconvolution properties | Pileup
• Two events can be separated and precisely measured if they are distant 2-3 T
smp
• For shorter distance between pulses additional signature of not resolvable pileup may be used (more than 2 non zero samples)
CR-RC, Tsmp
=Tpeak
=1, amp =1
Measurement Setup configuration
Signal source:
• Pulse generator (AGILENT 81150A)
– Signal : Current pulse
– Trigger : second channel (channel to channel jitter < 100ps)
• Radioactive source 90Sr
– Signal : β- ~546 keV
– Trigger : Scintillator + PMT (poor time resolution >~ 1ns)
• Laser (PDL 800-D)
– Signal : Infrared photons 1060nm
– Trigger : provided by device(excellent time resolution, jitter < 100ps)
• MSO7104B 4Gsps (250ps) Digital sampling oscilloscope used as ADC
New prototype with Multichannel ADC SoC
ASIC is almost ready
Photograph of Measurement Setup
MSO7104Bsampling oscilloscope
E3612Apower supply
81150Apulse generator
Sensor & FE BOXScintillator & photomultiplier
Keithley 237HV Supply
Laser Head
PicoQuantPDL 800D
Radioactive source
Monte Carlo Software architecture
object oriented library written in Python
A dedicated software was developed to reproduce all measurements in order to compare deconvoluted experimental results with MC simulations results
Linearity check – measurements with laser
• Deconvolution algorithm is linear (as expected)
• For small SNR (<10) reconstructed amplitude is not yet optimised
• NF is slightly above 0, almost no degradation of SNR
• MC fits well to measurements
Tpeak
= Tsmp
= 60ns
NF=20 log10 SNRin
SNRout
Time reconstruction – measurements with laser
• Reconstructed time doesn't depend on amplitude (as expected)
• For Tpeak
= Tsmp
= 60ns time resolution in range 2-7 ns is obtained
• MC fits well to measurements
Constant Trigger delay
Tpeak
= Tsmp
= 60ns
Deconvolution as a function of Sampling Period Laser pulses, SNR ~20, T
peak = 60ns
• For Tsmp
between 20-40 ns time
resolution between 1-2 ns can be obtain for SNR 20 and T
peak = 60 ns
• Tsmp
should be a compromise between
power consumption (ADC) and requested timing resolution
• Wide plateau (30-70ns) with optimum amplitude resolution (NF)
Very good agreement between measurements and MC simulations
Deconvolution studied with various signal sources – laser, 90Sr source, generator
• Good qualitative agreement for results obtained with different sources
• Some quantitative differences are connected to :
– For generator relatively long rise time > 2.5ns
– For radioactive source (90Sr) poor timing resolution of reference photomultiplier signal
– Laser gives the most precise measurement
Tpeak
= Tsmp
= 60ns
Deconvolution results vs sensor bias voltages – pulse shape degradation
• Weak electric field (low HV) increases charge collection time and lengthen current pulse
• Non delta like pulse causes deviation from ideal CR-RC shaping
• Deteriorated pulse shape causes quantitative decrease of timing resolution
Tpeak
= 60ns
Altogether quantitative differences are not very big, the method is moderately sensitive to pulse shape
Deconvolution Summary
• Extraction of time and amplitude information using the deconvolution principle has been studied with MC simulations and verified experimentally
• Deconvolution performed on measured (with laser) data with the setup comprising of: sensor + Front-End + ADC shows very good agreement with MC simulations
• Precise timing information down to 1-2ns, good amplitude reconstruction and pileup rejection was obtained using a simple CR-RC shaper and T
peak=60ns
• Readout comprising of Front-End + ADC + DSP is a very good candidate for use in triggerless systems (CLIC / testbeam), replacing traditional dual chain readouts
• Subnanosecond resolution may be possibly obtained using shorter shaping/sampling times or multilayer detector systems, but have not been studied yet...
Deconvolution Future plans
• Short term : Verify fully custom triggerless system on testbeam (July 2011, see Adrian Matoga & Jonathan Aguilar talks)
• Long term : work in progress on extending the analyses (and measurements if possible) of deconvolution performance on:– optimization of power consumption, sampling time, shaping time for
given specifications
– precise calculation of correlated noise for given shaping/sampling
– implementation of higher order shaping like CR-RCn
– study ADC quantization effects
– study pileup effects
LumiCal Front-end ASIC Multichannel ADC ASIC Xilinx FPGA
Occupancy estimation for LumiCal@CLIC
(by Tel Aviv group)
• CLIC train structure :
– BX separation : 0.5ns
– BX per train 320 (whole train 160ns)
1% occupancy : → 3 hits per train → event separation ~ 50 ns
Technique for high occupancy CLIC
detectors – gated integrator & CDS
Gated Integrator - properties
• Simplest possible shape (step) will allow amplitude information reconstruction even in case of high occupancy
• Time resolution of single event tagging limited by rise time of preamplifier and ADC sampling frequency
• Feasible sampling frequencies :
– Only (fast) ADC : ~ 50 - 100 Msps (10 - 20ns)
– Analog memory followed by (slow) ADC : ~ 100 - 200 Msps (5 - 10ns)
• Dynamic range ~ 10 (12) bits
• SNR – to be estimated
Gated Integrator - Future plans ...
•
• We have already some prototypes with can be used to verify this concept
• Singal to Noise study has to be performed
• More information about time structure of incoming events is needed
Marek Idzik talkFCAL Collaboration Meeting
LAL-Orsay, France, 5-6 Oct. 2007
Summary
•
• Two methods for signal processing for CLIC detectors were presented
– Deconvolution gives very good results (both in amplitude and time domain) for relatively rare pulses (time interval > ~ 50ns)
– Gated integrator may be solution capable to process very high occupancy channels (signal to noise performance and time resolution are worst than for other methods)
• More information about time structure of incoming events / occupancy is needed
Backup slides
Very soon scope digitizer will be replaced by our new Multichannel ADC SoC ASIC
ADC ASIC 3rd prototype• 8 channels of 10 bit pipeline ADC
(verified to work up to 50Msps)
• Power scales linearly with fsmp
• Power switching on/off mechanism
• Cross talk < -70dB
• Digital multiplexer/serializer:
– Serial mode (max fsmp
~ 4 MSps)
– Parallel mode (max fsmp
~ 30MSps)
– Test mode (single channel readout)
• High speed LVDS drivers (~1GHz)
• Low power DAC control references
• Precise BandGap reference source
• Temperature sensor
• Die size ~ 2 x 3 mm ADC ASIC 2rd prototype
~ 0.8mW/MSps
Front-end Electronics & Sensor(designed for LumiCal detector@ILC)
Front-end spec:
– Cdet
≈ 0 ÷ 100pF
– 1st order CR-RC shaper (Tpeak
≈ 60 ns)
– variable gain
– SNR ~ 20 for MIP
Standard Silicon sensor :– Thickness 300um
– Capacitance ~ 5 – 25 pF / channel
– Leakage current ~ 5nA / channelAMS 0.35 μm
ASIC contains 8 channels
See more : M. Idzik, Sz. Kulis, D. Przyborowski "Development of front-end electronics for the luminoisty detector at ILC" Nucl. Instr. and Meth. A 608 (2009) pp.169-174
Few definitions
(for many events)
NF=20 log10 ( SNR%in
SNRout)
•Noise Figure (definition)
• (definition)
Recovered time = mean valueTime recovery error = RMS of such distribution
• NF = 0 → SNR doesn't change
• NF < 0 → SNR increases
• NF > 0 → SNR decreases
SNRin
= 20
• NF = 1db → SNRout
= 17.8
• NF = 2db → SNRout
= 15.9
• NF = 3db → SNRout
= 14.1
Deconvolution related issues
• How sampling time / peaking time / SNR affects timing resolution?
• What is optimum relation between sampling time, peaking time and power consumption?
• How deconvolution impacts on SNR ?• How pulse non-ideality affects deconvolution ?• How ADC resolution impacts on deconvolution ?
A dedicated experimental setup comprising of the whole readout chain (sensor + Front-End + ADC) was build to answer quantitatively some of these questions
Deconvolution – history, goals
• Deconvolution idea was proposed for use in pulse shaping in HEP experiments at the beginning of 90's
• It was then implemented in different versions of APV's (analog pipeline voltage) ASICs designed for synchronous with beam experiments like CMS at LHC, where deconvolution was performed by analog pulse shape processor (APSP)
• The main goal was amplitude measurement with good pileup rejection plus a rough estimation of time (to identify beam crossing, repeating every 25ns)
• Our goals are to:
– apply deconvolution principle in asynchronous systems
– obtain precise timing information (~1-10 ns rms) in addition to amplitude and good pileup rejection
– add ADC in each readout channel and use digital signal processing (much more robust than analog)
Deconvolution – principle and choice of adequate shaping
Requirements :– Simple hardware shaper realization
– Simple deconvolution formula
V sh s =1
s1/2
D s =1
V sh s=s1 /2
I sen t = t I sen s =1
z=esT
- peaking time
D z =z2−2 e−T /e−2T /
z-1 is a unit time delayT – sampling interval
d t i=z−1 D z =V sh t i−2e−T /V sh t i−1e−2T /V sh t i−2
CR-RC shaper
Sensor pulse: CRRC response
Deconvolution formula (s domain)
Deconvolution formula (time domain)
Deconvolution formula (z domain)
d t i=V sh t i−0.74V sh t i−10.14V sh t i−2
=TFor :
CLIC machine time structure – Power savings
Average power can be reduced 20ms/200ns ~= 100.000 times ! if switching on/off mechanism is implemented
Deconvolution alghoritm
• Look Up Tables used• Time and amplitude finding can be done off-line or in
external DAQ
Event processing
• Interpolation is used to find trigger time (< 250ps)
• Acquired (oversampled 4 Gsps) waveform is down-sampled to number of events for further analyses (but trigger information is left precise)
Testing all possibles phases between sampling clock and incoming event