Yes, primarily aimed at multiple measurements of the same speaker. I don't really follow the second part of your comment, since cross correlation alignment is an alignment process, not an averaging process. It is aligning to the first measurement, so it's phase response will depend on the placement of t=0 for that first response. Vector average is a complex average. Your "AverageResponse" looks to be a little acausal given the way the phase seems to rise above a few kHz.
I know... probably the time alignment for my left tweeter is wrong on that side. The right side does not show that behavior, but is still different from the vector average in REW. Staging in the car is really good though with that setting, so i kept away from changing this tbh.
And Averager does a time alignment of the different measurements first.
I think the most important point is, that both Smaart and Averager can do power averages where dips in the response are weighted less than in REW with the vector averages. Just like the RMS average in REW is doing, but without averaging the phase information.
RMS averaging in REW corresponds way better with power averages from Averager and Smaart while both Averager and Smaart retain the averaged phase information and REW does not.
If REW was able to average phase when doing RMS averages like it's doing with vector averages it would be great. But of course i don't know if this is doable or planned for futher REW development.
And Averager does a time alignment of the different measurements first.
I think the most important point is, that both Smaart and Averager can do power averages where dips in the response are weighted less than in REW with the vector averages. Just like the RMS average in REW is doing, but without averaging the phase information.
RMS averaging in REW corresponds way better with power averages from Averager and Smaart while both Averager and Smaart retain the averaged phase information and REW does not.
If REW was able to average phase when doing RMS averages like it's doing with vector averages it would be great. But of course i don't know if this is doable or planned for futher REW development.
I looked at a few hybrid averaging schemes. The difficulty with them is the phase response no longer corresponds to the magnitude response, so the impulse response of the average ends up looking pretty odd with a lot of acausal content. That has held me back from putting any of those schemes in a release.
Fair enough, although for the task that Oabeio is doing absolute phase response is not even necessary. I myself trim down the phase response from my car down to an FDW of 1 to just correct the trend of the phase response, not every minor detail.
Question would be if the phase average tied to the vector average is so much more correct than if compared to a phase average based on RMS averages.
Apparently with Smaart and their way of doing power averages it is working.
Question would be if the phase average tied to the vector average is so much more correct than if compared to a phase average based on RMS averages.
Apparently with Smaart and their way of doing power averages it is working.
@Cathul and @emailtim
I can’t start a 3 way PM
Tim this is Peter “Cathul”… he’s the one I told you would benefit from your expertise… super awesome to be able to share different experience’s
Would you mind sending him a private message (or a new thread) and showing him what you’ve done I think this is right up his alley as he also does a convolution like yours, and I think he would really like to see what you’ve done with DRC-FIR
He has programmed a raspberrie pi to do a convolution in his car and he has the brains to pull off what you’ve done
……..
Peter , this guy blew me away wow is all I could say I was completely taken aback by what he’s done…
Way above my head most definitely but I understand what he is doing and it is super cool… if you can make sense of it then you can teach it to me in a more for dummies kind of way lol
He re wrote the software to do driver level convolution and can be used with a usb DAC
But more then that his method for getting the sub and tweet step perfect… (which doesn’t seem really physically possible right as the sub has no peak) …. You’ll see what I’m talking about it’s super super cool…
And then we can catch up privately about it..
I can’t start a 3 way PM
Tim this is Peter “Cathul”… he’s the one I told you would benefit from your expertise… super awesome to be able to share different experience’s
Would you mind sending him a private message (or a new thread) and showing him what you’ve done I think this is right up his alley as he also does a convolution like yours, and I think he would really like to see what you’ve done with DRC-FIR
He has programmed a raspberrie pi to do a convolution in his car and he has the brains to pull off what you’ve done
……..
Peter , this guy blew me away wow is all I could say I was completely taken aback by what he’s done…
Way above my head most definitely but I understand what he is doing and it is super cool… if you can make sense of it then you can teach it to me in a more for dummies kind of way lol
He re wrote the software to do driver level convolution and can be used with a usb DAC
But more then that his method for getting the sub and tweet step perfect… (which doesn’t seem really physically possible right as the sub has no peak) …. You’ll see what I’m talking about it’s super super cool…
And then we can catch up privately about it..
Fair enough, although for the task that Oabeio is doing absolute phase response is not even necessary. I myself trim down the phase response from my car down to an FDW of 1 to just correct the trend of the phase response, not every minor detail.
Question would be if the phase average tied to the vector average is so much more correct than if compared to a phase average based on RMS averages.
Apparently with Smaart and their way of doing power averages it is working.
Yes smaart does sound very good….
The REW method also sounds good, the data isn’t what is happening however which is frustrating…
I’m sure everybody has a preferred technique, and I did a study on a bunch of different cars and realized the excess phase and measured phase are very very similar. (Makes sense in such a reflective environment)
So using the measured phase in smart may not be optimal but in a car it works quite nice…
I would be absolutely thrilled if I could get the data presented the same way that smart does it…
Even if they ignore the dips that the vector makes, and I make a correction in rephase, without moving the microphones or changing anything still literally in the same tuning session, once I send my filter to the DSP, what shows on the live measurement isn’t exact to what I was working with like it does work in smaart
I thought it was because before I had multi mic capabilities, that it was me not holding the microphone in the exact same positions unfortunately that’s not the case even with the microphones in the exact same position the outcome is inconsistent..
You need to set "hide result phase below" to a lower value, but you might face precision issues.
In any case, the phase at -100dB will not have any impact on your crossover.
Thanks @pos .
I thought the options in the rightmost panel were for display purposes only, not affecting the filter generation. They do not appear to make any change on the binary data.
I am using the "extended" phase width for some computational purposes. I noticed if I convolve a linear phase all pass filter against a linear phase band pass XO in FFTW, I get some linear phase extension on the low side of the band pass XO, but that maybe an artifact in REW's display.
I thought the options in the rightmost panel were for display purposes only, not affecting the filter generation. They do not appear to make any change on the binary data.
I am using the "extended" phase width for some computational purposes. I noticed if I convolve a linear phase all pass filter against a linear phase band pass XO in FFTW, I get some linear phase extension on the low side of the band pass XO, but that maybe an artifact in REW's display.
The "hide result phase below" option gives a better picture of what is going on.
It looks like the phase is defined down 300dB within the band pass.
Is there any reason why phase on the low side jumps 180 degrees and then bounces around 180 degrees instead of bouncing around 0 degrees ?
This explains the extra low side phase extension when convolving the LP All Pass against the LP band pass.
It looks like the phase is defined down 300dB within the band pass.
Is there any reason why phase on the low side jumps 180 degrees and then bounces around 180 degrees instead of bouncing around 0 degrees ?
This explains the extra low side phase extension when convolving the LP All Pass against the LP band pass.
Last edited:
@pos , I have a question on boost eq
If the max amplitude is above 0db , (like using eq boost) and you don’t use the master volume to lower it and use the fir with a positive value …. What happens?
I ask because sometimes I have dips that I would like to add boost to …. And would be easier to do it in fir and keep all channels the same relative volume…
Does it drive it into clipping immediately?
Because of my upstream source has a volume control and I never go to max wouldn’t leave the head room that way?
Thanks in advance
Andrew
If the max amplitude is above 0db , (like using eq boost) and you don’t use the master volume to lower it and use the fir with a positive value …. What happens?
I ask because sometimes I have dips that I would like to add boost to …. And would be easier to do it in fir and keep all channels the same relative volume…
Does it drive it into clipping immediately?
Because of my upstream source has a volume control and I never go to max wouldn’t leave the head room that way?
Thanks in advance
Andrew
Controlling the volume at the source is indeed a good solution.
long answer: https://www.audiosciencereview.com/...he-convolution-step.38654/page-2#post-1507357
long answer: https://www.audiosciencereview.com/...he-convolution-step.38654/page-2#post-1507357
If the DSP is floating point you can exceed 0dB and bring the volume down later, if it is fixed point you cannot as it will clip straight away and then bringing the volume down later will just make the clipping quieter
Easy to test as any digital DSP clipping sounds like crunch crunch crunch and is truly horrible.
Easy to test as any digital DSP clipping sounds like crunch crunch crunch and is truly horrible.
I have 1 more question if I can…
After a correction rough cut I have been zooming in and doing fine corrections
What is raised cosine mean?
This one make very flat small bumps high Q fraction of a db changes
Is that ok ? Like Q 5-10 .2db changes or so
Attachments
Less spread of frequency correction than typical PEQ
https://www.twaudio.de/wp-content/u...Grath_et_al-Raised_Cosine_Equalization_EN.pdf
https://www.felusch.de/?p=445
https://www.twaudio.de/wp-content/u...Grath_et_al-Raised_Cosine_Equalization_EN.pdf
https://www.felusch.de/?p=445
Okay yeah just tried it … disaster and for the first time on minidsp I could hear pre ringing
It sounded weird …. Like short echoes…
Okay it’s cool , would be really handy if that bell shape could be done in minimum phase
I added some boost filters in and a whole .84db above the 0db line
Turned it on waiting for the nasties and it sounds good still … whooo hooo … this is going to make things much easier now
It sounded weird …. Like short echoes…
Okay it’s cool , would be really handy if that bell shape could be done in minimum phase
I added some boost filters in and a whole .84db above the 0db line
Turned it on waiting for the nasties and it sounds good still … whooo hooo … this is going to make things much easier now
The main selling point of raised cosine EQs is that you get a flat response when adjusting adjacent frequencies to the same gain, when Q and frequency spacing are chosen accordingly, eg Q=2.0 for 2/3 octave (the default configuration) and Q=4.0 for 1/3 octave.
I must confess I have never used them in any project myself
I must confess I have never used them in any project myself