# Properties of Optimization Techniques in Filter Design

I have this design:

Can you tell me about the type of filter and the algorithm just viewing this design, or should there be other information?

• this is amplitude of what ? and after which operations you got this amplitude response ? – AlexTP Nov 3 '17 at 16:44
• This is amplitude of a single passband , what do you mean the operations ,do you mean the algorithm? – N.na91 Nov 3 '17 at 20:12
• Alex means: what on earth are we looking at? Is this a plot of how much high a sandwich splatters if excited at a given frequency, is this a characterization of a digital filter as measured, is this a theoretical result of a filter type analysis? So, no, we can't tell you much by just looking at the amplitude response. Sure, we can exclude a few classes of filters, but without knowing what filter design approach you used, the rest is just plain, bad guesswork. Your question, by the way, reeks a lot of the XY Problem,where you ask something strange in hopes of … – Marcus Müller Nov 3 '17 at 21:57
• … solving another problem, which you really should mention, because someone else might have a much better idea how to solve the original problem. – Marcus Müller Nov 3 '17 at 21:58
• @N.na91 Please read this: meta.stackexchange.com/help/how-to-ask for some advice on how to ask a good question. Your question is vague, does not show what you have researched so far to solve it yourself, and has no context or details. It seems you have posted this other question: dsp.stackexchange.com/questions/44925/… that seems to be related to the same problem. – Atul Ingle Nov 5 '17 at 2:50

## 1 Answer

It's a bandpass filter, which is fairly obvious. FIR or IIR is hard to tell without phase. A FIR design would likely to have assumed linear phase, and omitted the phase response. and a package probably wouldn't calculate it, like in an IIR design, so I would tend towards a FIR.

The band goes to .5 which implies whoever designed the filter was taught unit sampling. A current person would most likely have the band go to pi or 1. I’m guessing who ever designed this is retired or dead.

The stop bands don't have a constant level, so I would rule out remez or linear programing.

I would guess if a FIR filter, a window technique, probably a Kaiser window. Maybe a Fejer window. The stop bands are also symmetric around the pass band, consistent with a rect function in the frequency domain, followed by a window.

The stop band is only about 55 db down, so the filter order is definitely less than 100 taps, probably more than 30. I could count the nulls but I don’t have my glasses.

This is all a lot of speculation. Can’t tell if this fixed point or floating point arithmetic.

I'm reading:

Mastermind: How to Think like Sherlock Holmes, by Maria Konnikova

which inspired me to answer your odd question. I would put little trust in this answer. This looks like one of those situations where a legacy function is being migrated and someone thinks making it better would break something. Good luck. Windowing as a filter design method is more of a hueristic search than a direct optimization.

• Your answer to a question really helped me and gave me another look what I was thinking, This is the original article from which this design was taken: ''Complex eigenfilter design of arbitrary complex coefficient FIR digital filters‏'' – N.na91 Nov 9 '17 at 10:49