FIR filter with minimum phase - page 6

New comment
 
keekkenen:
the conclusion is simple - not Juan's sombrero!
I don't know.
Or maybe... simpler signals.
Like a synthetic 100 hertz + 1,000 hertz.
Let it draw, isolate, summarize.
But this time with a visual phase control.

P.S. Don't just stick whatever the hell it is
whatever you need to build a model
signal = LF + HF + noise
Generate a simpler one - play with the model.
Then make it more complicated.
 
How about this? The delay is right, but it is only critical in some cases. It is wrong to think that it should be in principle minutes or even hours and there is no way to reduce it without a fatal loss of signal quality. Yes, reducing the delay is due to the fact that the filter becomes less and less like a perfect bandpass. But nobody forbids us to slightly increase sampling frequency of a signal to allow selecting a "spare" boundary frequency of a filter, i.e. above the boundary of a signal spectrum, but below half of sampling frequency. In this case, the non-ideal stepped amplitude-frequency response of the filter will not matter much. Finally, some people seem to confuse non-linear distortion with distortion of the amplitude-frequency response of the filter.
 
In the light of this, it is problematic to make a system with these filters in that the impulse response length is repeatedly increased. And if we take into account the non-uniform sampling rate - different number of ticks per minute. Then the weighting function will have a dynamic length. As a consequence of this it is necessary to either adapt the kinematic - completely regenerate its characteristics and adapt to the spectrum permanently on each sample, or use IIR filters.
 

Why a FIR filter? Wouldn't it be better to get the spectrum first? Then pick up the filter and see the result?

DSP is also possible...

.... Thanks for the topic, I've been wanting to do it myself for a long time, but haven't got around to it.

 
Zhunko:

The FIR filter can be made any way you like. Payback for this calculation time.

Correction.

Payback is not for FIR filter itself, but for a desire to implement it on processor.

Only I do not quite understand where this wish comes from.

Hardware special calculator can calculate value of any FIR filter in 2 clock cycles.

 
http://www.metolit.by/ru/dir/index.php/2512 Neurocomputer with extensible architecture for special tasks
 
Judging by the example of wipers, for a 1024 bar depth, the number of wipers needed increases to tens, hundreds, or in the best case, thousands or more. To calculate such a number of filters using digital filters instead of wipers is all the more difficult.
 
http://physics-animations.com/rusboard/themes/22453.html Found some interesting discussions not afraid to move away from theories. Discussed everything from quantum mechanics to kotelnikov. Taki in the highlighted post is a bit similar to what I wrote about here about intermediate values. There's not much information about filter delay and its reduction. But here's the gist. I quote: "About the delay is correct, but it is critical only in some cases. It's not right to think that it should be minutes or even hours, and you can't possibly reduce it without a fatal loss of signal quality. Yes, reducing the delay is due to the fact that the filter becomes less and less like a perfect bandpass. But nobody forbids us to slightly increase sampling frequency of a signal to allow selecting a "spare" boundary frequency of a filter, i.e. above the boundary of a signal spectrum, but below half of sampling frequency. In this case the non-ideal stepped amplitude-frequency response of the filter will not be of much importance. Finally, the author seems to confuse non-linear distortion with distortion of the filter's amplitude-frequency response." I'll highlight this very point in particular:"... Yes, the decrease in delay is due to the fact that the filter becomes less and less like a perfect bandpass. But no one prohibits us from slightly increasing the sampling frequency of the signal, which would allow us to choose a "spare" boundary frequency of the filter, i.e. above the boundary of the signal spectrum, but below half the sampling frequency. In this case, the non-ideal stepped amplitude-frequency response of the filter will not be of special importance....".
 

The delay may or may not be important. It all depends on what purpose the filters are used for. In my case, filters are used to decompose a complex curve into simple sine-like components. More precisely, for visual representation of the curve as a sum of components on the screen, because I perceive such components better, and I don't need these components for any calculations.

So, a simple experiment (decomposition of a sine wave) shows that this decomposition is useful only in one case - if the phase shift of the filter is zero. Otherwise, the picture becomes not easier to understand, but more complicated.

After reading the topic diagonally, I still could not find an answer to the question from the title: what is the minimum phase shift of the FIR filter? Although I have not finished my work yet, I have grounds to believe that the minimum possible phase shift of FIR filter is zero. In books, such filters are called physically unrealizable, and that is usually the end of the discussion. Nevertheless it is obvious, that such filters can be used on history and under some conditions they will work in real time as well.

 
I haven't seen any indicators that analyse dynamic phase shifting. That is, the filters of the kicks shift the phase in different ways. If, for example, the average between samples is plotted, then in some cases the optimal shift is not required by half a period, but by +- another fractional part. That is, if instead of the dummy smoothing by the method of inscribed complexities the tangents to the edges connecting the adjacent samples will give additional points which will have complementary samples along the price axis, and at the same time they will have non-uniform length samples, something will be shifted a bit more, something less. Thus, we will get a function not only along the price axis but also along the "time" axis. For example, many people build the scales starting from period 1,2,3,.... and so on, but there are wands with periods of 1/2, 1/4, 1/64.... and so on, and the intersection points of these shapes also have their own information. And then, adding, say, an interpolation straight line containing 1000 additional discrete points (or for example, a dynamically varying function as a range width, or the same tick volume as a function can be attached to these 1000 intermediate points) between samples, we will have dummies with fractional weights. And since the additional points between samples will have a non-uniform phase step, the weights of the dips or other ticks, will also vary.
12345678910111213...20
New comment