1
$\begingroup$

I'm investigating the propeties of Cascaded Integrator Comb (CIC) filter. (for sigma-delta modulator). I have 2 quesitons and below is what I have done in Wolfram Mathmatica:

a). What's the correct method to design a fast settling decimation filter, is the combination of SINC3+SINC1 in the right direciton? or any other topology? (narow bandwidth LPF, decimation ratio osr, settling within one cycle osr/Fs, also with good anti-aliasing? better than SINC1?)

b). Any alternative method to improve my calculation speed? is my method too clumsy?

1. SINC1

The simplist one, SINC1's transfer function, it's impulse response length is osr

osr=128
normalSinc1 = (1 - z^(-osr))/ (1 - z^-1);
normalSinc1Freq = (normalSinc1/osr) /. z -> Exp[I*\[Omega]];
normalSinc1FreqPlot = 
  Plot[20*Log10[Abs[normalSinc1Freq]], {\[Omega], 0, \[Pi]}, 
   PlotStyle -> Blue];
normalSinc1Time := If[InverseZTransform[normalSinc1, z, n] > 0, 1, 0]
For[n = 0, normalSinc1Time > 0, n++]
normalSinc1ImpluseLength = n

after decimation by osr, alising effect of power

(*ideal brickwall lowpass filter, bandwith=1/osr *)
idealOSRpower = Integrate[Sqrt[0.5]^2, {f, -1/osr, 1/osr}];
(*Normal SINC1 power after decimation, Ratio=1 *)
normalSinc1FreqPower = Conjugate[normalSinc1Freq]*normalSinc1Freq;
normalSinc1FreqPowerSample = 
  Integrate[normalSinc1FreqPower, {\[Omega], -\[Pi], \[Pi]}]/(2*\[Pi]);
normalSinc1Ratio = normalSinc1FreqPowerSample/idealOSRpower

2. SINC3

Then I do same thing for SINC3 filter, I found the caculation efficiency is very low, is my method is too clumsy? Any method to improve the calculation speed? Thank you!

normalSinc3 = (normalSinc1)^3;
normalSinc3Freq = (normalSinc1^3)/osr^3 /. z -> Exp[I*\[Omega]];
normalSinc3FreqPlot = 
  Plot[20*Log10[Abs[normalSinc3Freq]], {\[Omega], 0, \[Pi]}, 
   PlotStyle -> Black];
normalSinc3Time := If[InverseZTransform[normalSinc3, z, n] > 0, 1, 0]
For[n = 0, normalSinc3Time > 0, n++]
normalSinc3ImpluseLength = n

the power after decimation

(*Normal SINC3 power after decimation Ratio=0.55 *)
normalSinc3FreqPower = Conjugate[normalSinc3Freq]*normalSinc3Freq;
normalSinc3FreqPowerSample = 
  Integrate[
    normalSinc3FreqPower, {\[Omega], -\[Pi], \[Pi]}]/(2*\[Pi]) ;
normalSinc3NBW = normalSinc3FreqPowerSample/idealOSRpower

2. Fast Settling SINC3+SINC1?

Now I understand, if my Data Ouput Rate is Fs/OSRsettling time is:OSR/Fs for SINC1;(3*OSR-2)/Fs for SINC3;

Is it possible to design a filter (FIR/CIC) to achieve both fast settling (with one cycle osr) and good anti-aliasing? Then I try the combination of SINC3+SINC1. Notice! This calculation is very time-consuming!

(* sinc3(osr=r1) + sinc1(osr1=r2) *)
osr=128;
r1 = 4;
r2 = osr/r1;
fastSincN = ((1 - z^-r1)/ (1 - z^-1))^3;
fastSinc1 = (1 - z^(-r2 *r1))/ (1 - z^(-r1));
fastSinc = fastSincN*fastSinc1;
fastSincFreq = fastSinc/(r1^3)/r2 /. z -> Exp[I*\[Omega]];
fastSincFreqPlot = 
  Plot[20*Log10[Abs[fastSincFreq]], {\[Omega], 0, \[Pi]}, 
   PlotStyle -> Red];
fastSincTime := If[InverseZTransform[fastSinc, z, n] > 0, 1, 0]
For[n = 0, fastSincTime > 0, n++]
fastSincImpluseLength = n
fastSincImpluseLength2 = 3*r1 - (3 - 1) + r1*(2^Log2[r2] - 1)/(2 - 1)

The Impluse Response Length fastSincImpluseLength is 134, which is close to osr. The fastSincImpluseLength2 is the relationship I summed up. Unfortunately, I found the anti-alasing capacibilty is very very poor, which is close to SINC1.This calculation is very time-consuming!

(*Fast SINC power after decimation,Ratio=0.98 *)
fastSincFreqPower = Conjugate[fastSincFreq]*fastSincFreq;
fastSincFreqPowerSample = 
  Integrate[fastSincFreqPower, {\[Omega], -\[Pi], \[Pi]}]/(2*\[Pi]);
fastSincRatio = fastSincFreqPowerSample/idealOSRpower
$\endgroup$

1 Answer 1

0
$\begingroup$

A first order CIC filter for decimate by $N$ is mathematically identical to the cascade of a moving average filter over $N$ samples, followed by the down-sampler. This is a linear phase filter whose group delay will be $(N+1)/2$. Increasing the order of the CIC is equivalent to cascading multiple first order CIC filters, with the commensurate increase in delay.

The CIC filter is very simple in implementation so is an attractive solution in many cases. As in the moving average filter, it will also suffer from "passband droop" which itself is easily compensated for. If settling time is of upmost concern, and a non-linear phase (group delay distortion) is acceptable, then a minimum phase anti-alias filter would be an alternate approach to consider.

CIC filter

CIC for decimation

The above graphics show how a (first order) CIC is equivalent to a moving average filter followed by a down-sampler. Thus if we want to predict or model the CIC filter, we can do so simply from the frequency response of a moving average filter. A higher order CIC is equivalently the cascade of multiple moving average filters, such as the construction of a 2nd order CIC for decimate by 4 shown in the graphic below, with the commensurate increase in overall delay associated with each moving average filter.

higher order CIC

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.