FIR filtering

FIR filtering theory is described in Lars presentation but other documents and presentation from Doris and Kazu are available at Indico search: FIR

Which FIR is applied with VeloTELL1CableFIRFilter

The filter is designed to be applied on pedestal and common mode subtracted non zero suppressed data.

It is applied in a way as close as possible to a TELL1 realistic scenario (thanks to Guido, which designed me this algorithm):
You want to apply a filter of order say 5. For every channel, you have to compute:

newADC[i] = g[-2] * ADC[i-2] + g[-1] * ADC[i-1] + g[0] * ADC[i] + g[1] * ADC[i+1] + g[2] * ADC[i+2]
this can be trivially written
newADC[i] = 1/128 * ( g[-2]*512 * (ADC[i-2]/4) + g[-1]*512 * (ADC[i-1]/4) + g[1]*512 * (ADC[i+1]/4) + g[2]*512 * (ADC[i+2]/4) ) +  ADC[i]
g[0] is assumed to be one.
FIR coef inside the TELL1s must have a binary representation. The real values of the g[k] are of the order of the x-talk coefficients, that is O(0.05). Storing g[k]*512 on a 8 bit signed integer allows to explore the range +127/512 = 24% to -24% by step of 0.2%. Dividing the ADC value by 4 allows to drop the least significant bits which we do not want to enter the computation.

A VeloTELL1CableFIRFilter instance has a given N (x-talk order, defined by the amount of Kmi vector you give in options (see below)) and M (FIR order, given by the M option). Both have to be 3,5,7 or 9 (means you act on 1,2,3 or 4 chip channel neighbors).

The FIR filter acts only on a per-analog-cable basis, it cannot correct for a pulse that spreads over the boundary of two cables.

The VeloTELL1CableFIRFilter options

  • InputDataLoc and OutputDataLoc tells the algorithm where to read and write the data.
  • M is the FIR order, the default value is 9, you acts on -4, +4 neighbors.
  • DebugCable = x; prints for the given cable the ADC values before and after the FIR filter.
The Tell1List and coefficients options
  • Tell1List = { t1,t2,t3,...tn }; is the list of TELL1s to analyze the order matter. All the other TELL1s see their data duplicated to the OutputDataLoc.
  • Km4, Km3, Km2, Km1, Kp1, Kp2, Kp3, Kp4 are the x-talk (not FIR) real value coefficients per TELL1s and analog cable. For example Km2 defines all the h[-2] coefficients.
Each of the Kxy is a vector of double written with the following syntax:
Kxy = { t1_cable0_coef_xy, ..., t1_cable63_coef_xy, t2_cable0_coef_xy, ..., t2_cable63_coef_xy, ..., tn_cable0_coef_xy, ..., tn_cable63_coef_xy };
the t1, t2, ..., tn being ordered like in the Tell1List vector. This is of course awful to write by yourself, but VeloXTalkComputer spits you a file with these lines out.

How to inverse the x-talk into FIR coefficients

The algebraic theory to inverse the x-talk values into FIR coeficients is summarized here (thanks Lars):
Basic equation to determine the transfer function of the
filter to cancel the x-talk :


d[n] = sum_-inf^+inf { g[k]*h[k-n] }

where 
  d[n] is the kronecker delta
  g[n] is the transfer function of the FIR,
       i.e the coefficients for the filter.
  h[n] is the transfer function of the link,
       i.e the measured x-talk 

if the g[n] is non-zero for only M terms, which means FIR
filter of order M, then the sum is reduced to:

d[n] = sum_(-M-1)/2}^{(M-1)/2 { g[k]*h[k-n] }

This can be considered as an infinite series of equations,
one for each value of n, that will give the relation between
g[n] and h[n]. These equations will have non-trivial
solution only for values of n where h[k-n] are not zero for
all terms in the sum. Assuming that the x-talk only extends
(N-1)/2 bins forwards and backwards, the requirements on n
for non trivial equations are:

-(M-1)/2 - n <=  (N-1)/2   (only first term in sum non-zero)
 (M-1)/2 - n >= -(N-1)/2   (only last term in sum non-zero)

This gives N+M-1 equations, -(M+N)/2+1 <= n <= (M+N)/2-1,
that defines the relation between g[n] and h[n].

Writing out the equations explicitly for M=3 and N=3:

                    k=-1          k=0          k=1

n=-2    d[-2] = g[-1]*h[ 1] +      0      +     0
n=-1    d[-1] = g[-1]*h[ 0] + g[ 0]*h[ 1] +     0
n= 0    d[ 0] = g[-1]*h[-1] + g[ 0]*h[ 0] + g[1]*h[ 1]
n= 1    d[ 1] =      0      + g[ 0]*h[-1] + g[1]*h[ 0]
n= 2    d[ 2] =      0      +      0      + g[1]*h[-1]

This system of equations can be put in matrix form

Y = (d[-2], d[-1], d[0], d[1], d[2])^T  
  = (  0,     0,    1,    0,    0  )^T

G = (g[-1], g[0], g[1])^T 


A = { (h[ 1],   0,     0  ), 
      (h[ 0], h[ 1],   0  ),
      (h[-1], h[ 0], h[ 1]),
      (  0,   h[-1], h[ 0]),
      (  0,     0,   h[-1]) }

this give the equation AG = Y with M unknowns (the g[n]) and
N+M-1 equations that can be solved by the least square method.
 

-- Main.jborel - 30 Jun 2007

Edit | Attach | Watch | Print version | History: r2 < r1 | Backlinks | Raw View | Raw edit | More topic actions...
Topic revision: r1 - 2007-06-30 - unknown
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LHCb All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback