Strengths and weaknesses of digital filtering Example of ATLAS LAr calorimeter C. de La Taille 11 dec 2009.

Présentation au sujet: "Strengths and weaknesses of digital filtering Example of ATLAS LAr calorimeter C. de La Taille 11 dec 2009."— Transcription de la présentation:

1 Strengths and weaknesses of digital filtering Example of ATLAS LAr calorimeter
C. de La Taille 11 dec 2009

2 ATLAS ready for data taking (2007)
11 dec 2009 C. de La Taille Journées thématiques IPNO

3 Signal and noise in LAr calorimetry
Ionization signal, triangular shape Large dynamic range : 50 MeV-2 TeV High accuracy : < 1% 11 dec 2009 C. de La Taille Journées thématiques IPNO

4 Analog shaping in ATLAS LAr
Minimize overall noise Electronics noise : ENI =A/tp3/2 + Ap/√tp, varies with capacitance Pileup noise : ENE = C √ tp, scales with machine luminosity Optimum shaping time varies with luminosity and detector position 11 dec 2009 C. de La Taille Journées thématiques IPNO

5 C. de La Taille Journées thématiques IPNO
Digital filtering Multiple Sampling technique [Cleland NIM A338 (1996)] Linear combination of N samples on waveform to optimize noise Finite Impulse Filter Signal : s(t)=Ag(t)+b(t) A : amplitude g(t) : normalised signal shape b(t) : noise Sampled signal : si=Agi+bi Filter : weighted sum Σ aisi Normalization : <S> = A Σ ai gi = 1 Sampled signal shape 11 dec 2009 C. de La Taille Journées thématiques IPNO

6 Digital filtering Σ aiajRij – λ( Σaigi -1)
Searching Amplitude A, minimizing Variance S σ2 = Σ aiajRij Rij = autocorrelation matrix = <bibj> Finite Impulse Filter Lagrange multiplier : λ on constraint Σ aigi =1 Minimize Σ aiajRij – λ( Σaigi -1) ai = Σ R-1ij gi gi = signal shape Autocorrelation function 11 dec 2009 C. de La Taille Journées thématiques IPNO

7 C. de La Taille Journées thématiques IPNO
Example gi = (0, 0.63, 1, 0.8, 0.47) gives ai = (0.17, 0.34, 0.40, 0.31, 0.28) Effect on signal : Low pass filter Reduction of electronics Noise Tp : 50 ns -> 100 ns Expect noise / 3 Get 1.8 !… 11 dec 2009 C. de La Taille Journées thématiques IPNO

8 Noise improvement Measured values, for various peaking times
noise before and after digital filtering 11 dec 2009 C. de La Taille Journées thématiques IPNO

9 C. de La Taille Journées thématiques IPNO
Various questions What is the noise improvement ? How does-ti compare to the optimum analog filter ? How many samples are needed on the waveform ? How do they vary with the phase ? What accuracy is needed on Rij and gi What relation should be verified between sampling time and analog shaping time ? Is the analog filter really necessary ? Can the signal also be accelerated Can it be used to measure the arrival time ? 11 dec 2009 C. de La Taille Journées thématiques IPNO

10 Finding the transfer function
Using Z-transform : H(Z) = Σ aiZ-i with Z = exp(jωTech) Low pass filter, f0 = 4 MHz Aliasing at multiples of sampling frequency Fech = 40 MHz 11 dec 2009 C. de La Taille Journées thématiques IPNO

11 C. de La Taille Journées thématiques IPNO
Noise improvements 11 dec 2009 C. de La Taille Journées thématiques IPNO

12 C. de La Taille Journées thématiques IPNO
Noise improvement With higher sampling frequency the aliasing effect is reduced With more samples, the signal can be further slowed down 11 dec 2009 C. de La Taille Journées thématiques IPNO

13 C. de La Taille Journées thématiques IPNO
Variation with phase Optimum phase with one sample at signal peak 11 dec 2009 C. de La Taille Journées thématiques IPNO

14 C. de La Taille Journées thématiques IPNO
Time measurement Sign al becomes Si=Ag(ti-T)+ni Expand to Si=Ag(ti) + ATg’(ti)+ni Optimization of A and AT terms with 2 Lagrange multipliers New set of coefficients bi : bi proportionnal to signal derivative g’i 11 dec 2009 C. de La Taille Journées thématiques IPNO

15 C. de La Taille Journées thématiques IPNO
Time measurement Time resolution, measured on calibration and physics data 11 dec 2009 C. de La Taille Journées thématiques IPNO

16 Accelerating the signal
Change of autocorrelation function with pilepup noise New coefficients : A = (-0.75, 0.47, 0.75, 0.07, -0.19) Accelerates (and symetrizes) the signal A = (-0.75, 0.47, 0.75, 0.07, -0.19) 11 dec 2009 C. de La Taille Journées thématiques IPNO

17 Overall noise performance
Black curve : constant analog filtering (tp = 50 ns) Green curve : digital filtering Dotted curve : optimum analog filter 11 dec 2009 C. de La Taille Journées thématiques IPNO

18 C. de La Taille Journées thématiques IPNO
Conclusion Digital filtering is a powerful method to reach optimum filter, it has rendered complex analog filters nearly obsolete It allows non causal filters It is adaptative and simple to implement in DSPs It allows to perform several different optimizations (amplitude, time…) References B. Cleland, W. Stern NIM A338 (1996) C. de La Taille, L. Serin ATLAS Internal note LArG 80 (1998) M. Abarrouche et al. NIM A597 (2008) 11 dec 2009 C. de La Taille Journées thématiques IPNO