FIR filter with minimum phase - page 7

New comment
 
If we select cocks for each timeframe for each pair, for example for the sample of 1024 minute bars, then the length of the impulse characteristic will vary from 1024 points to 3 points and if we include intermediate lines connecting one-minute bars then the number of cocks will increase exponentially, But there is one more thing about timeframes, they will become "bigger" and when adjusting the "length" of all TFs to the width of the smallest discreteness TF, the shift of points will occur and the bars' slopes will fall to intermediate values.
 
It is possible to solve through geometry. But there is no way to fit it into the multicurrency, and it also has a useful component for analysis. Of course we can build indices and use them for calculating target levels, and choose the pair according to more dissimilar ones, but it's also an ambiguous picture, and we cannot build the index correctly without the entire spectrum of all pairs, because all components are important there. So to say that there is no noise, if you have the whole spectrum of frequencies, they are all needed in the calculation, but it is impossible to calculate, so you have to sacrifice something and extrapolate the slowest components, but the highest frequency components will remain unpredictable, so it seems that I separate the signal and noise, in fact this "noise" is also a useful component in the signal, which participates equally in calculations.
 
About shifting filters and stuff. Has anyone tried building a Pascal triangle out of these filters?
 
setting progressions for a pascal triangle is generally good, i.e. a pascal triangle can be made "stretched/compressed" as it were by changing the progression coefficient. In essence, we get a hierarchy of filters with a set of weight coefficients. But their ach isn't smooth with such coefficients. If you make a triangle with smooth decaying edges instead of a clipped one, it would be better. Now, it would be nice to set this parameter. That way in every filter hierarchy we can shift them without major repainting and then for building another filter with a set of smooth coefficients we can take the values of the previous one. I will try to describe it in the evening.
 
A pascal triangle can be thought of as a set of ki filters with weight functions more like trapezoids at even levels in the triangle, and triangular ones at odd levels in the pascal triangle. So, how will the kinds of these functions change if we build a pascal triangle from a pascal triangle and so on. For example, we have got a Pascal triangle for a 100th bar depth, we take the extreme values on the last bar from all levels of the triangle (that is, the values on the last bar from filters whose coefficients are the values of rows of levels in the Pascal triangle multiplied by corresponding values of bars, then from these hundred values, and so on, setting the number of times we recalculate the triangle from the results of the previous triangle. Or maybe the coefficients here will have some variable function stretching/shrinking the pascal triangle initially, i.e. maybe there are formulas for variations of the pascal triangle so as not to do these triangle-to-triangle calculations.
 
Nik1972:
Has anyone tried building a Pascal triangle out of these filters.
I don't get it... Pascal's triangle is constructed from certain numbers. And what is a Pascal triangle from filters? And most importantly, what is it for, what do we want to get from it, what is the physical meaning of it?
 
AlexeyFX:
I don't get it... A Pascal triangle is constructed from certain numbers. And what is a Pascal triangle made of filters? And most importantly, what is it for, what do we want to get out of it, what is the physical meaning of it?
The meaning is unimportant. What matters is the Pascal's triangle.
 
Correct, the Pascal triangle is constructed from numbers and the filters have fractional coefficients like a linearly weighted waving machine. By constructing a fan of wizards (simple) and then constructing the average between the averages and so on, we obtain a Pascal triangle of fractional coefficients. Where in the numerator is the Pascal triangle itself-numbers that surround it, and in the denominator is the number increasing by 2 bases. In essence, the levels in the Pascal triangle will change from integers to fractional numbers, which will become weight series (functions) in filters of different depths. We can see why the shifting filters should be of odd order, they will have a tending to parabolic (base up) shape.) Even-order filters will be like a trapezoid, with decreasing upper base. It can be seen that in order to have an overlap in phase, it is necessary to take (using the example of wipers) wipers1-3,3-5,5-7.... and so on. Therefore the Pascal triangle can also be seen as a system of nested (if you take sets of filter weights that are not even separately) triangle/parabolic. It is necessary to connect these weighting functions to get a triangle not as an inverted parabola with cut off ends, but to get the ends smoothly passing into a decaying wave. But actually it is probably already close to the calculation of Kikh filters.
 
This construction will be needed when getting the following, so that the difference between the price and the filter For example, we build a large period LF filter, for example 2000 bars, from it we take the remainder, i.e. the LF cloze. Then we filter the remainder and so on. The system of filters should be such, that the remainder is approximately equal, being directional in the sign of increment. Then, when we shift the filter system, we will substitute the missing data by the least moduli method so that their sum is minimal in co-direction.
 

In the limit, this construction will produce a Gaussian filter (as the limit of the binomial coefficients). Its advantage is that it also produces a Gaussian bell in the frequency domain. In other words, by rapidly decreasing the Gaussian curve, effectively limiting the time window, we at the same time limit the frequency domain just as effectively. (Those who know DSP theory will remember that this is a big advantage for DSP, because spectrum clipping from the high frequencies tends to creep into the low frequencies, causing lots of problems.)

Another thing is that it's much easier not to screw around and calculate the coefficients of the Gaussian impulse response curve beforehand.

1234567891011121314...20
New comment
Reason: