Hai all, Which FIR filter design method is used to design a non-symmetric LP FIR filter as the following method will give linear phase i.e.,symmetric coefficients 1.window 2.frequency sampling 3.equiripple Which is the best method to design a non-symmetric LP FIR filter? regards, faz
FIR filter design methods
Started by ●May 28, 2008
Reply by ●May 28, 20082008-05-28
faz wrote:> Hai all, > > Which FIR filter design method is used to design a non-symmetric LP > FIR filter as the following method will give linear phase > i.e.,symmetric coefficients > 1.window > 2.frequency sampling > 3.equiripple > > Which is the best method to design a non-symmetric LP FIR filter?Hello again. You still don't write why you don't like symmetric LP filters. Regardless, you can use FDLS to design any type of FIR filter - most easily if you have a prototype frequency repsonse. You can download a Matlab implementation of FDLS here: http://apollo.ee.columbia.edu/spm/external/tipsandtricks/files/TandT_Jan2007.zip Regards, Andor> > regards, > faz
Reply by ●May 28, 20082008-05-28
Hai, I am not an enemy to symmetric filter coefficients ...My filter realization structure is not according to symmetric structure ... (i.e.,half the number of coefficients of non-symmetric filter).Though non-symmetric structure can perform symmetric coefficient computation but will miss the advantage for less multiplier at the cost of more adders... For example 256-tap non-symmetric==512 symmetric tap.. By the way how to use that MATLAB file to generate filter coefficients and filter response calculation?? which is the best filter design method ?? regards, faz Andor wrote:> faz wrote: > > Hai all, > > > > Which FIR filter design method is used to design a non-symmetric LP > > FIR filter as the following method will give linear phase > > i.e.,symmetric coefficients > > 1.window > > 2.frequency sampling > > 3.equiripple > > > > Which is the best method to design a non-symmetric LP FIR filter? > > Hello again. You still don't write why you don't like symmetric LP > filters. Regardless, you can use FDLS to design any type of FIR filter > - most easily if you have a prototype frequency repsonse. You can > download a Matlab implementation of FDLS here: > > http://apollo.ee.columbia.edu/spm/external/tipsandtricks/files/TandT_Jan2007.zip > > Regards, > Andor > > > > > regards, > > faz
Reply by ●May 28, 20082008-05-28
Hai, I am not an enemy to symmetric filter coefficients ...My filter realization structure is not according to symmetric structure ... (i.e.,half the number of coefficients of non-symmetric filter).Though non-symmetric structure can perform symmetric coefficient computation but will miss the advantage for less multiplier at the cost of more adders... For example 256-tap non-symmetric==512 symmetric tap.. By the way how to use that MATLAB file to generate filter coefficients and filter response calculation?? which is the best filter design method ?? regards, faz On May 28, 6:12�pm, Andor <andor.bari...@gmail.com> wrote:> faz wrote: > > Hai all, > > > Which FIR filter design method is used to design a non-symmetric LP > > FIR filter as the following method will give linear phase > > i.e.,symmetric coefficients > > 1.window > > 2.frequency sampling > > 3.equiripple > > > Which is the best method to design a non-symmetric LP FIR filter? > > Hello again. You still don't write why you don't like symmetric LP > filters. Regardless, you can use FDLS to design any type of FIR filter > - most easily if you have a prototype frequency repsonse. You can > download a Matlab implementation of FDLS here: > > http://apollo.ee.columbia.edu/spm/external/tipsandtricks/files/TandT_... > > Regards, > Andor > > > > >
Reply by ●May 28, 20082008-05-28
On May 28, 8:41�am, faz <fazulu.v...@gmail.com> wrote:> Which is the best method to design a non-symmetric LP FIR filter?The "best" method depends upon many factors, but described here is an outline of how to convert a zero-phase (symmetric) FIR filter to an equivalent minimum-phase filter: http://www.geo.uni-potsdam.de/forschung/Software/FIR/fir_corr.htm . I haven't attempted to do this, but I had saved the link for analysis at a later time. Greg
Reply by ●May 28, 20082008-05-28
faz wrote:> Hai, > > I am not an enemy to symmetric filter coefficients ...My filter > realization structure is not according to symmetric structure ... > (i.e.,half the number of coefficients of non-symmetric filter).Though > non-symmetric structure can perform symmetric coefficient computation > but will miss the advantage for less multiplier at the cost of more > adders... > > For example 256-tap non-symmetric==512 symmetric tap..... No higher authority demands that you exploit the potential efficiency that symmetry allows. Just pretend that the filter lacks symmetry. Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
Reply by ●May 28, 20082008-05-28
Another method per Schussler does this: 1) Find a symmetric minimax filter with the same "structure" of the response you need. Because the filter is even in time, it will be real in frequency. 2) Add a small constant to the filter response that's equal to the peak stopband ripple. This makes the filter response completely positive and with double zeros in the stopband. Or, in place of 1 and 2, find a symmetric minimax filter with the same "structure" of the response you need but with all-positive response in the stopbands (suggesting double zeros in the stopband). You can do this by changing the filter design criterion at each step in the Remez exchange algorithm by adding the peak error to the stopband objective function / desired response. 3) Once you have the all-positive, symmetric FIR filter, you will have a filter with conjugate zeros around the unit circle. Remove all the zeros outside the unit circle and make all the double zeros on the unit circle into single zeros. In effect, this takes the square root of the frequency response. The resulting filter is minimum phase. Note that the stopband ripple will change from R*10^-K to sqrt(R)*10^-(K/2) so 10^-6 will be 10^-3. This is quite a change in dB attenuation. So, to fix this, shoot for 2x the stopband attentuation you need (in dB). Having done this, there are also a number of other nonsymmetric filters possible: - a maximum phase filter where you select all the zeros outside the unit circle. - some in-between phase filter where you select zero pairs from both the inside and outside the unit circle - eliminating their images on the other side of the circle. If the original filter is length 2N and the desired filter is length N then there are N/2 ways to select the zeros .. some may have characteristics (such as coefficient values or phase) that are more or less desirable. Fred
Reply by ●May 29, 20082008-05-29
hai, For the same symmetric coefficients,can anyone tell the difference between FIR filter response computed in direct form symmetric(optimized) and non-symmetric(conventional) structure?? regards, faz Fred Marshall wrote:> Another method per Schussler does this: > > 1) Find a symmetric minimax filter with the same "structure" of the response > you need. Because the filter is even in time, it will be real in frequency. > > 2) Add a small constant to the filter response that's equal to the peak > stopband ripple. This makes the filter response completely positive and > with double zeros in the stopband. > > Or, in place of 1 and 2, find a symmetric minimax filter with the same > "structure" of the response you need but with all-positive response in the > stopbands (suggesting double zeros in the stopband). You can do this by > changing the filter design criterion at each step in the Remez exchange > algorithm by adding the peak error to the stopband objective function / > desired response. > > 3) Once you have the all-positive, symmetric FIR filter, you will have a > filter with conjugate zeros around the unit circle. Remove all the zeros > outside the unit circle and make all the double zeros on the unit circle > into single zeros. In effect, this takes the square root of the frequency > response. > > The resulting filter is minimum phase. > > Note that the stopband ripple will change from R*10^-K to sqrt(R)*10^-(K/2) > so 10^-6 will be 10^-3. This is quite a change in dB attenuation. So, to > fix this, shoot for 2x the stopband attentuation you need (in dB). > > Having done this, there are also a number of other nonsymmetric filters > possible: > - a maximum phase filter where you select all the zeros outside the unit > circle. > - some in-between phase filter where you select zero pairs from both the > inside and outside the unit circle - eliminating their images on the other > side of the circle. > If the original filter is length 2N and the desired filter is length N then > there are N/2 ways to select the zeros .. some may have characteristics > (such as coefficient values or phase) that are more or less desirable. > > Fred
Reply by ●May 29, 20082008-05-29
"faz" <fazulu.vlsi@gmail.com> wrote in message news:719debdb-28b8-4b47-a030-b516f2af3b0a@w5g2000prd.googlegroups.com...> hai, > > For the same symmetric coefficients,can anyone tell the difference > between FIR filter response computed in direct form > symmetric(optimized) and non-symmetric(conventional) structure?? > > regards, > faz >No doubt.
