|
Hi firiends Please excuse me if I have posted this query on irrelevant discussion group. I have a basic question on Lowpass Filter Design. I am talking in the context of Upsampling and Down sampling. What is the criteria to choose the ripples in passband and stop band? We all know that the larger the ripple values we allow, the Filter Order required will be lower for the same transition band. Hence by allowing larger ripples we get advantage on the computational speed. But what is the system parameter which will decide that the ripple can't be more than a prescribed value. Thanks in advance. regards arun |
|
|
Choosing ripples
Started by ●September 21, 2000
Reply by ●September 21, 20002000-09-21
|
Hi, I were innteresting this problem for real microprocessor application. The solution depend on start and finish necessary conditions. You can use Butterworth filter of 10th or more order. Thus enable work in plain segment of AFR. Also You must to remember about phase of filter. Phase is more important that amplitude!!! Best regards Please write to me if you have any question- i'm interesting this problem. Nick Packo ----- Original Message ----- From: <> To: <> Sent: Thursday, September 21, 2000 7:51 AM Subject: [matlab] Choosing ripples > > Hi firiends > Please excuse me if I have posted this query on irrelevant discussion > group. > > I have a basic question on Lowpass Filter Design. > I am talking in the context of Upsampling and Down sampling. > What is the criteria to choose the ripples in passband and stop band? > > We all know that the larger the ripple values we allow, the Filter > Order required will be lower for the same transition band. Hence by > allowing larger ripples we get advantage on the computational speed. > > But what is the system parameter which will decide that the ripple > can't be more than a prescribed value. > > Thanks in advance. > > regards > arun > To Join: > > To Post: > > To Leave: > > Archives: http://www.egroups.com/group/matlab > > More DSP-Related Groups: http://www.dsprelated.com > |
Reply by ●September 22, 20002000-09-22
|
Hi Emer, I am thankful for your response. I do agree with your statement thet it depends on the application. But, the question is what value we have to choose judiciously for FIR filters in decimation and interpolation applications. Since the application I was interested in was Decimation and Interpolation, what I felt in case of (1.) Interpolation: was that, the stopband ripple should be such that the total energy in stop band after filtering can be decided based on no of bits used for filtering. ie. for example if 16 bits are used for arithmetic in 2s complement format, the last bit will have weight equal to 2**(-15). One thing that can be done is that the ripples designed be such that its total energy equal to (2**(-16))*(2**(-16))/12(assuming unifrom probability for out of band signals). So, we have a basis on which we can calulate the total ripple energy and then if we choose equiripple then it can be equally distributed over stop band. (2) Decimation: was that, since the frequency response folds at new fs/2 after decimation, (and considering that aliasing components which folds into transition band and pass band, as the undesirable signal) the contribution due to aliased frequency spectrum should not visible in the output samples of 16 bits(of 2s complement arithmetic). This means the variance of aliased signal after filtering in the pass band and transition band should be equal to (2**(-16))*(2**(-16))/12. So here also we have a basis on which we can choose ripples. These are my own ideas. Not from any books. I have been trying to get info regarding this, but not successful. Is there any book which you suggest? More over, I don't know the basis on which you can choose the passband ripple. I may be wrong.Please let me know any flaw on the idea have. regards arun > -----Original Message----- > From: Emerson Tan 95-28046 [mailto:] > Sent: Friday, September 22, 2000 10:28 PM > To: > Subject: Re: [matlab] Choosing ripples > > > Hi! It really depends upon your application. There is no 'best' filter. > Sometimes it is ok for some ripple in the passband, sometimes you'd > want the passband to be maximally flat. There are also filters that > put equal ripple in the passband and the stopband; they are called > equiripple filters. However, they are IIR filters and produce nonlinear > phase response. > > --emer > > On Thu, 21 Sep 2000 wrote: > > > I have a basic question on Lowpass Filter Design. > > I am talking in the context of Upsampling and Down sampling. > > What is the criteria to choose the ripples in passband and stop band? > > > > We all know that the larger the ripple values we allow, the Filter > > Order required will be lower for the same transition band. Hence by > > allowing larger ripples we get advantage on the computational speed. > > > > But what is the system parameter which will decide that the ripple > > can't be more than a prescribed value. > > > > Thanks in advance. > > > > regards > > arun > > To Join: > > To Post: > > To Leave: > > Archives: http://www.egroups.com/group/matlab > > More DSP-Related Groups: http://www.dsprelated.com |
Reply by ●September 22, 20002000-09-22
|
Ah, I did not realize that you wanted to know about stop-band ripples as well. I believe you are correct, except that you are assuming evenly distributed noise when most noise sources are gaussian. There is also one other thing to watch out for. Noise is tenacious. For two uncorrelated noise sources which add together, SigmaTotal^2 = SigmaA^2 + SigmaB^2. So let's say that SigmaA is your converter noise, uniformly distributed over +/- 1/2 LSB. SigmaB is the noise you have allowed to alias in. If you allow aliasing at the same level as your converter noise, the result is 3 dB higher than the converter noise! You can suppress SigmaB further and further, but it will always add at least a little bit to the converter noise. The question is, how much additional noise can you stand? If you allow aliasing at 1/5th of converter noise, you have added about 0.17 dB. So just bear in mind that if you want noise performance near that of the A/D, you must add some extra aliasing protection. Sincerely, Glen Ragan |
Reply by ●September 22, 20002000-09-22
|
Hi! It really depends upon your application. There is no 'best' filter. Sometimes it is ok for some ripple in the passband, sometimes you'd want the passband to be maximally flat. There are also filters that put equal ripple in the passband and the stopband; they are called equiripple filters. However, they are IIR filters and produce nonlinear phase response. --emer On Thu, 21 Sep 2000 wrote: > I have a basic question on Lowpass Filter Design. > I am talking in the context of Upsampling and Down sampling. > What is the criteria to choose the ripples in passband and stop band? > > We all know that the larger the ripple values we allow, the Filter > Order required will be lower for the same transition band. Hence by > allowing larger ripples we get advantage on the computational speed. > > But what is the system parameter which will decide that the ripple > can't be more than a prescribed value. > > Thanks in advance. > > regards > arun > |
|
|
