Page 1 of 2

Linear phase filters

Posted: 17 Jun 2020 13:03
by sadhill
Hi Uli, hi All,

I want to try using linear phase filters instead of minimum phase. A few questions are arising :

1. When generating the crossover, is there a filter type more recommended for linear phase ? I do usually use Linkwitz XO for minphase.

2. If I understand properly, the normal way to add filters or/and to apply driver linearization to a linear phase XO is convolution followed by cut'n window. Is it better to cut'n window after each convolution, or can it be done once at the end after several convolutions ? Which position option should be chosen in cut'n window ?

3. Linear phase filters are said to be more processor power-consuming. But my resulting Cor files seem to have the same size as a minphase Cor files (512k for 48kHz files). Is it normal ?

Of course these questions will look trivial to many forum readers...

Francois

Re: Linear phase filters

Posted: 17 Jun 2020 15:00
by UliBru
Hello Francois,

1. of course you may still use Linkwitz XOs. The NT and UB filters offer quite steep filter slopes This allows to remove an undesired driver behaviour, e.g. cone break-ups, in a better way. Of course there is no lunch for free, the stepper filters have a bit more ringing. Simply try them and decide which XO you refer by listening.

2. you can apply CutNWindow after several convolutions. For convolution of linphase XO and minphase linearization simply select the original XO length and start position 0, no other option ticked.

3. the required power is dependent on filter length. A filter file size 512k = 65536 filter taps is ok, 512k filter taps would be very long. It is better to talk about filter size in taps.
Linearphase filters have a delay of half filtersize (in taps) divided by samplerate. This is by principle and not belonging to CPU power.

Re: Linear phase filters

Posted: 17 Jun 2020 17:40
by sadhill
Dear Uli,

Thank you for the answer !

The filter steepness is of concern for me. I do usually mainly use 24dB LR filters. This time I am using 48dB slopes, precisely for the purpose of removing unwanted cone movements. Is it best using NT or UB instead of a steeper LR filter ? Is a bigger delay with a steep LR the only drawback ?

You say the required processing power comes on the length of the filter, which made sense for me as I stressed the file size as the comparison point. So this means the required processing power depends on the sample rate and not on the filter type, linphase or minphase ? Or is it better using a bigger filter length with linphase filters ?

Re: Linear phase filters

Posted: 17 Jun 2020 17:51
by UliBru
Francois,

feel like a child and play around with Acourate ;) . So e.g. create a UB jPol7 filter 2nd order with your XO data and compare it to your Linkwitz filters. You should quickly grab the difference. There is no additional delay.

The filter length (n) defines the resolution of the filter. You get n/2 +1 frequency bins with a frequency resolution of samplerate/n.
Thus of course longer filters procude a better resolution on the cost of more required CPU power and more delay in case of linearphase filters.
Example: 65536 taps @ 48 kHz result in 32769 frequency bins, the resolution is 0.732 Hz. The linphase filter delay is 0.682 seconds.
A filter of 262144 taps would result in 131073 frequency bins with resolution 0.183 Hz and a linphase filter delay of 2.73 seconds

Re: Linear phase filters

Posted: 18 Jun 2020 09:00
by sadhill
Hello Uli,

Interesting, these UB filters. I have to give them a try, I hadn't payed attention to them up to now.

I went through the thread where you explain them. A post drew my attention. You write that linearizing drivers nearfield often gives bad results with dipoles. In my experience with LX 521 speakers, the result is far better with 40cm nearfield linearization than at the listening position...

Francois

Re: Linear phase filters

Posted: 18 Jun 2020 09:25
by UliBru
Francois,

it is great if the nearfield linearization works for you. I have made a different experience, at the end we have even omitted the linearization and just applied an overall correction.

Re: Linear phase filters

Posted: 18 Jun 2020 11:37
by sadhill
works well too, I'm presently listening without linearization...

Re: Linear phase filters

Posted: 18 Jun 2020 23:29
by sadhill
I tried UB Jpol7 2nd order linphase modified filters with a very interesting result. Wide and accurate soundstage already with the first trial, without any linearization or delay/phase checkups. Really promising.

By the way, the shape of these filters is quite close to the LR 8th order, but with a better result...

Thank you Uli !

Re: Linear phase filters

Posted: 19 Jun 2020 12:41
by UliBru
That's why the UB filters are invented :D

Re: Linear phase filters

Posted: 21 Jun 2020 19:45
by sadhill
Hello Uli,

For comparison purposes, I tried to set up a linphase filter with asymetric slopes (LR 2nd and 4th order + several adjustments with convolutions). Everything goes fine until I use Cut'N window. The curve shape is completely changed with this last operation ! (I checked the result is ok with a symetrical filter after the same convolution processes)

On the picture, the black curve is the cut'nwindow of the red one. (length 65536, start position, position = 0, everything else unchecked).

Normal behaviour ? (the raw asymetric filter itself was built through convolution of to half-filters with different slopes, no cut'n window was performed at this stage)
CutNw24-12.png
CutNw24-12.png (17.4 KiB) Viewed 4835 times