It is necessary to solve a following problem. Frequency of sampling – 120kH. It is necessary to calculate and realize the low-frequency filter with a passband 29 kH, frequency of stopband 30 kH, ripple in a passband no more 0,1dB, suppression not less 96dB, the phase should be linear. After a filtration a signal decimated twice, up to frequency of sampling 60 kH. What approaches can be for the decision of the given problem? I tried to count FIR. The order of the filter approximately from 500 up to 1000 taps. Whether it is possible with the purpose of reduction of the order of the filter and accordingly reduction of charges of the equipment to solve a problem by means of two IIR filters - the first IIR forms necessary magnitude-frequency characteristic, and the second IIR – all-pass filter -corrects a phase of the first IIR the filter? In other words, it is necessary to solve a problem with the minimal hardware expenses.
Effective realization of the filter with an abrupt transitive strip?
Started by ●November 5, 2007
Reply by ●November 5, 20072007-11-05
alex65111 wrote:> It is necessary to solve a following problem. Frequency of sampling =F2==C0=D3> 120kH. It is necessary to calculate and realize the low-frequency filte=r> with a passband 29 kH, frequency of stopband 30 kH, ripple in a passban=d> no more 0,1dB, suppression not less 96dB, the phase should be linear. > After a filtration a signal decimated twice, up to frequency of samplin=g> 60 kH.=20 >=20 > What approaches can be for the decision of the given problem?The best thing to do is to restate the problem. Something tells me that the task can be approached in much better way.> I tried to count FIR. The order of the filter approximately from 500 up=to> 1000 taps.=20First, decimate and then correct the frequency response. However the=20 requirement of 0.1dB accuracy at 97% of Nyquist is tough to riducule.> Whether it is possible with the purpose of reduction of the order of th=e> filter and accordingly reduction of charges of the equipment to solve a=> problem by means of two IIR filters - the first IIR forms necessary > magnitude-frequency characteristic, and the second IIR =F2=C0=D3 all-pa=ss filter> -corrects a phase of the first IIR the filter? >=20 > In other words, it is necessary to solve a problem with the minimal > hardware expenses.Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●November 5, 20072007-11-05
==The best thing to do is to restate the problem.== What do You consider unsuccessful in conditions of a problem? ==requirement of 0.1dB accuracy at 97% of Nyquist is tough to riducule== Such requirement is caused by a problem with ISI. And how many in Your opinion it is normal? Basically are supposed and more ripple, but they should be low-frequency that on an any interval 1-2kH magnitude-frequency characteristic it was possible to consider as a constant. And as your opinion on a combination of two IIR filters?
Reply by ●November 5, 20072007-11-05
alex65111 wrote:> It is necessary to solve a following problem. Frequency of sampling – > 120kH. It is necessary to calculate and realize the low-frequency filter > with a passband 29 kH, frequency of stopband 30 kH, ripple in a passband > no more 0,1dB, suppression not less 96dB, the phase should be linear. > After a filtration a signal decimated twice, up to frequency of sampling > 60 kH. > > What approaches can be for the decision of the given problem? > > I tried to count FIR. The order of the filter approximately from 500 up to > 1000 taps. > > Whether it is possible with the purpose of reduction of the order of the > filter and accordingly reduction of charges of the equipment to solve a > problem by means of two IIR filters - the first IIR forms necessary > magnitude-frequency characteristic, and the second IIR – all-pass filter > -corrects a phase of the first IIR the filter? > > In other words, it is necessary to solve a problem with the minimal > hardware expenses. >If you can spec the passband ripple to some tolerance that's greater than "none" and if you can tolerate ripple in the stop band (which you probably can) then you may indeed be able to reduce the number of taps. I suspect that you'll want a combination of IIR and FIR filters, though. Sounds like a fun problem to tackle. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com Do you need to implement control loops in software? "Applied Control Theory for Embedded Systems" gives you just what it says. See details at http://www.wescottdesign.com/actfes/actfes.html
Reply by ●November 5, 20072007-11-05
alex65111 wrote:> ==The best thing to do is to restate the problem.== > What do You consider unsuccessful in conditions of a problem?Give us the full picture of the project.> ==requirement of 0.1dB accuracy at 97% of Nyquist is tough to riducule== > Such requirement is caused by a problem with ISI.The ISI should be handled separately.> And how many in Your opinion it is normal?0.8 of Nyquist is OK, 0.9 is probably the maximum reasonable value.> Basically are supposed and more ripple, but they should be low-frequency > that on an any interval 1-2kH magnitude-frequency characteristic it was > possible to consider as a constant. > > And as your opinion on a combination of two IIR filters? >It depends. Describe the full project in the details. Vladimir Vassilevsky DSP and Mixed Signal Design Consultant http://www.abvolt.com
Reply by ●November 5, 20072007-11-05
Without nuances which do not influence statement of a question the project consists in the following. There is an analog low-frequency filter at which the useful working band is 29kH, and a transitive band such, that on frequency 60kH suppression is 96dB. The signal from an output of the filter acts on ADC with frequency of digitization 120kH. The problem - to result frequency of digitization according to a useful working band, that is 60kH. Accordingly a signal it is necessary in the digital way to filter a signal before new sampling. One of the main features of a problem it is necessary to solve a problem with the minimal hardware expenses. The working band is set of independent digital signals (PSK8) a band from 1kH up to 3kH. Problem ISI in my opinion is necessary for considering by development of "our" filter that it has not brought additional ISI.
Reply by ●November 5, 20072007-11-05
Why do you need the 120kHz sampling and all of the foregoing shamanism if your goal is 8-PSK in the 3kHz band? VLV alex65111 wrote:> Without nuances which do not influence statement of a question the project > consists in the following. > > There is an analog low-frequency filter at which the useful working band > is 29kH, and a transitive band such, that on frequency 60kH suppression is > 96dB. The signal from an output of the filter acts on ADC with frequency of > digitization 120kH. > The problem - to result frequency of digitization according to a useful > working band, that is 60kH. Accordingly a signal it is necessary in the > digital way to filter a signal before new sampling. > > One of the main features of a problem it is necessary to solve a problem > with the minimal hardware expenses. > > The working band is set of independent digital signals (PSK8) a band from > 1kH up to 3kH. > Problem ISI in my opinion is necessary for considering by development of > "our" filter that it has not brought additional ISI. >
Reply by ●November 5, 20072007-11-05