I agree but I think the most important thing is how you plan to implement it. Microcontroller, DSP, PC, FPGA? For a microcontroller, for example, it is easier to use IIR because it doesn't need so much memory and the number of operations is relatively small. And in any case, I normally use any software package to calculate the filter coefficients (Analog Devices has a free package that does this).
There is a less often used method called eigen filters which helps to consider time
an frequency domain given in "Multirate Systems and Filter Banks" by P.P.Vaidyanathan.
p50 -56 which would seem to be attractive.
Cheers
Dear Hugh,
digital filter design is a multifold topic. There are many design methods available. This depends on you application topic in genreal. However, if you are not familiar with filter design, there are two "classical" methods for designing digital filters.
a) FIR-filters: I would recommend the classical frequency response / windowing technique
b) IIR-filters: Here you design a classical analogue filter first, then you have to transform the analogue frequency response / Transfer function (in s, Laplace-Transform) with the help of the Bi-Linear-Transform and z-Transform in the z-domain.
A good reference is: http://web.mit.edu/~cjoseph/Public/Discrete-Time_Signal_Processing_-_Oppenheim_-_2nd_Edition.pdf
In this reference you will find many hints regarding digital filter design. Futhermore, I recommend the Matllab Filter Toolbox (http://www.mathworks.de/discovery/filter-design.html).
Volker
Hi Narasim
I have read a bit about eigen-filters previously (like Slepian windows, with maximal concentration). Is it possible to create non-linear phase versions with the desired group delay as a design parameter?
Hugh
It depends on what the exact filter requirements are. If the passband ripple and the transition band should be rather small, I would suggest using a wave digital filter, either as a ladder filter or as lattice filter.
It depends on the order of the filter based on the transition band allowed. If it a higher order filter ,better to go for IIR implementation because it is easy to implement.
Hi Hugh
i am of the opinion that it should be possible but it might involve complex
coefficients . Basically the prolate sequence / eigenfilter relies on
the properties of Hermitian matrices using Rayleigh's principle
which states that max of v' A v (for A Hermitian and all v unit norm ) occurs only
iff v is eigenvector corresponding to max eigenvalue. So what is needed
to consider is if the form of H(z) =sigma h(n)*z^-n for n=0 to N can be split up into
a quadratic form v' A v with a Hermitian matrix or alternatively impose such conditions while allowing for nonlinear phase. Hope this helps. If it is of interest
we could explore further.
Cheers
Narsim
I think it is depend on what is ur bottle neck design or what is your design criteria.u can easiky design any kind of digital filters with any kind criteria in matlab fda tool!
Depending upon the application it will vary. For subband coding, I can apply wavelet filter method.
Depending on the transition band and attenuation rate in stop band decides the order of the filter. If the order is high better go for IIR Chebychev Filter configuration.
For low order IIR Chebychev filter is preferred, which requires less memory. But when stability issue is considered FIR is preferred but it requires more memory .
Regarding delay, transposed structure will give less delay
Concerning the need for configurable delay I would prefer FIR designs (stability, phase response) unless it must be a very high selective filter (= high order FIR).
Perhaps the attached pdf on type 1 eigen filters may be of interest. The coding in Scilab could be improved and other features added with some effort. It should be possible to translate into Matlab fairly easily .
Cheers
Your new favourite low-pass filter :) ...
Article Improved IIR Low-Pass Smoothers and Differentiators with Tunable Delay
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing