Hello Raimonds,
No mean to disrespect your work or disparage your product, as it looks really good and professional, and is one of the rare hardware convolution engines out there
I just feel float arithmetic would have been a better choice for filtering (which, again, is not the intended purpose of your box if I understand correctly).
The curves in the post above are what happens with a 1kHz filter and 0dBFS passband.
The stopband level will be higher with a lower crossover point and/or with attenuation.
No mean to disrespect your work or disparage your product, as it looks really good and professional, and is one of the rare hardware convolution engines out there
I just feel float arithmetic would have been a better choice for filtering (which, again, is not the intended purpose of your box if I understand correctly).
The curves in the post above are what happens with a 1kHz filter and 0dBFS passband.
The stopband level will be higher with a lower crossover point and/or with attenuation.
Hi POS,
Thank you! I am offering detailed EQ (available only on FIR) for 15 years.
The last 12 - on hardware.
Your mentioned stopband attenuation does not depend on bit number and will be the same for any filter tuning frequency but with "but" if you have unlimited number of fir filter coefficients.
We have limited number in the real life. Your mentioned example of filter on 1 kHz will be much higher affected by real length of fir filter instead of the bit number of its coefficients.
Please run your simulation for real length and you will see that it affects the stopband much higher than the bit number. And it highly depends on your skills to cut the filter to the length you need/have.
You can use any of APL`s FIR products for filtering / crossover functions with very high success.
But you should catch what is the most serious advantage of APL`s offers.
That advantage makes the job of sound producer / director / recording engineer two times more efficient ...
The first step should be to implement that for everyone who has serious passion in audio quality.
Good luck,
Thank you! I am offering detailed EQ (available only on FIR) for 15 years.
The last 12 - on hardware.
Your mentioned stopband attenuation does not depend on bit number and will be the same for any filter tuning frequency but with "but" if you have unlimited number of fir filter coefficients.
We have limited number in the real life. Your mentioned example of filter on 1 kHz will be much higher affected by real length of fir filter instead of the bit number of its coefficients.
Please run your simulation for real length and you will see that it affects the stopband much higher than the bit number. And it highly depends on your skills to cut the filter to the length you need/have.
You can use any of APL`s FIR products for filtering / crossover functions with very high success.
But you should catch what is the most serious advantage of APL`s offers.
That advantage makes the job of sound producer / director / recording engineer two times more efficient ...
The first step should be to implement that for everyone who has serious passion in audio quality.
Good luck,
Very high stopband rejection can be obtained with a good choice of windowing algorithm, independently of the number of taps :
4096 taps @ 48kHz, Albrecht 8-term window, 32 bits float:
4096 taps @ 48kHz, Albrecht 8-term window, 24 bits fixed:
4096 taps @ 48kHz, Albrecht 8-term window, 32 bits float:
4096 taps @ 48kHz, Albrecht 8-term window, 24 bits fixed:
Attachments
I am not sure I follow your argument here: are you implying that your unit does not have enough taps to do filtering with more than -90dB rejection, and that an analog device is a better choice?...
That 50Hz low-pass filter shown here is around 36dB/oct up to -20dB, then 48dB/oct up to -40db, and keeps increasing in slope after that. Quite usable I would say, and with linear phase
But that is beyond the point: 24bit fixed is enough for EQs, and that is what you do, apparently with great success.
32bit float is IMO a better choice when filtering is involved.
That 50Hz low-pass filter shown here is around 36dB/oct up to -20dB, then 48dB/oct up to -40db, and keeps increasing in slope after that. Quite usable I would say, and with linear phase
But that is beyond the point: 24bit fixed is enough for EQs, and that is what you do, apparently with great success.
32bit float is IMO a better choice when filtering is involved.
Yes, the analog crossover is better choice even we are "dancing" around the liner phase.
Your mentioned example is very close to 4th order analog filter.
It is extremely expensive to try to build a 4 way system using 4 FIR filters for crossover and only for the goal to have the linear phase crossover.
Much elegant solution is to use the analog crossover and then just one FIR filter to make its phase linear . FIR filter can serve not only mentioned time correction`s function but any other EQ functions you like in such case.
And it is extremely easy to make a phase correction for your crossover.
Measure or calculate its IR. Inverse it in time. Upload to your FIR. This filter is completely bit insensitive. Even 16 bits will be ok.
Your mentioned example is very close to 4th order analog filter.
It is extremely expensive to try to build a 4 way system using 4 FIR filters for crossover and only for the goal to have the linear phase crossover.
Much elegant solution is to use the analog crossover and then just one FIR filter to make its phase linear . FIR filter can serve not only mentioned time correction`s function but any other EQ functions you like in such case.
And it is extremely easy to make a phase correction for your crossover.
Measure or calculate its IR. Inverse it in time. Upload to your FIR. This filter is completely bit insensitive. Even 16 bits will be ok.
Last edited:
I have a 4 way FiR crossover and it wasn't that much expensive.
3 minidsp 2x4HDs ( 2 of them used as a 2x2)
It actually works quite well in fact I think I paid less for all three than a single APL would have cost. IIRC or very close to . And I can make that 50hz horbach crossover or whatever it is no problem.
3 minidsp 2x4HDs ( 2 of them used as a 2x2)
It actually works quite well in fact I think I paid less for all three than a single APL would have cost. IIRC or very close to . And I can make that 50hz horbach crossover or whatever it is no problem.
Last edited:
Hi,
What was your reason to use FIRs for crossover?
The most serious reason to have FIRs as crossover filters is the goal to have detailed EQ on each of bands. Are you able to arrange that?
How do you implement the master EQ for your system?
APL has lot of different APLs ... : )
There are no competitors for APL1s unit in terms of quality.
What was your reason to use FIRs for crossover?
The most serious reason to have FIRs as crossover filters is the goal to have detailed EQ on each of bands. Are you able to arrange that?
How do you implement the master EQ for your system?
APL has lot of different APLs ... : )
There are no competitors for APL1s unit in terms of quality.
Last edited:
Yes, that was the initially the sole purpose of rephase
But FIR filtering can bring many other things beside simple phase linearity, like the H-K crossover Oabeieo mentioned above.
Now if we are talking about optimization and opportunities to mix FIR and IIR, why not implement Balázs Bank's automated method with biquads in your solution to make better use of the coefficients your hardware can provide (ie better frequency resolution), and possibly keep a short FIR just for phase linearization?...
There are many ways to achieve good results. I personally like the idea of doing everything with one time domain convolution with a (floating point
) FIR: very simple and brutally elegant And the good news is that it can already be done today, and things will only get better in the coming years thanks to companies offering opened solutions like yours, miniDSP, Marani, Powersoft, etc.
Unfortunately, or rather ironically, its the I/O and not the compute engine that is expensive. At the heart of most of these DSP solutions is a $20 chip (in sufficient volume) surrounded by a myriad of input/output circuity, including AtoD and DtoA. What we pay for such a thing has more to do with the volume in which the solution is sold than its manufacturing cost.
So yes, you can do a brutally elegant FIR solution that runs on a $400 PC but then, depending on how many channels you have, you may spend several times that on the I/O to support it. As Raimonds said, the way to minimize that I/O cost is to use an analog crossover for each speaker so the number of DSP and AMP channels needed isn't multiplied by the number of ways in each speaker, H-K XO being an exception.
So yes, you can do a brutally elegant FIR solution that runs on a $400 PC but then, depending on how many channels you have, you may spend several times that on the I/O to support it. As Raimonds said, the way to minimize that I/O cost is to use an analog crossover for each speaker so the number of DSP and AMP channels needed isn't multiplied by the number of ways in each speaker, H-K XO being an exception.
Yes actually, it has 10peq on input and 10peq on output.
Output eqs are driver specific and input eqs are copied to all units to have same eq function.
That method worked good ..
Lately I have a Dirac box as a master eq ... it works too .
More than likely I'll own one of your boxes eventually because I just love to play with stuff.
The Tda software you have looks very promising and appealing.
So my remark in no way was aimed at diminishing your product, I was just stating that someone could build a 4 way with plenty of taps with multiple minis for less than the cost of one APL .
It's a little crude jumping from platform to platform, but the flexibility to me made it worth it .
And being able to export all my iirs and firs from rephase and work the correction from one platform actually works pretty good.
Last edited:
Hi Jack,
I'd rather not use a PC if I can avoid it
And the good news is I actually *can*
I use one openDRC-DI + SMSL M8 DAC per speaker (JBL M2 clones), and I get all the filtering capabilities, bandwidth and dynamic I need.
What else to ask for?... Well, I'd love to have an integrated box and maybe better ASRCs...
As you say the DSP in not the expensive part, but it is 2017 and the now ten-years-old HD2 is still state of the art.
I'd love to see a box with one say Sharc DSP per output for serious filtering capabilities without having to ressort to FFT, decimation, or helping biquads tricks.
Hi Pos:
Sounds like your speakers may employ analog crossovers. How else could you get by with just 2 DACs, nice as those two you have are? If so, that makes Raimonds' point, which was a good one. Its the total system cost that matters.
And I think you may have more $$$ in the DACs than in the OpenDRC.
As you said, you have all the DSP you need. Why wish for more? Our problems are not all minimum phase.
Jack
Sounds like your speakers may employ analog crossovers. How else could you get by with just 2 DACs, nice as those two you have are? If so, that makes Raimonds' point, which was a good one. Its the total system cost that matters.
And I think you may have more $$$ in the DACs than in the OpenDRC.
As you said, you have all the DSP you need. Why wish for more? Our problems are not all minimum phase.
Jack
I think Audirvana - > Soundflower -> AU Lab
with LAconvolver Audio Unit plugin should work.
If I have a two channel high end USB DAC, and i want to apply an active filter.
Is there a way to plug in another 2 channel DAC for the woofers?
So one computer for streaming out two channels on one USB port and two other channels on a second USB port?
Or am i forced to get a 4 channel USB DAC?
Is there a way to plug in another 2 channel DAC for the woofers?
So one computer for streaming out two channels on one USB port and two other channels on a second USB port?
Or am i forced to get a 4 channel USB DAC?
I completely agree on a box with 4 sharks. Heck I could elaborate a bit and
Want 4 sharks and 5th on the input
Just to make things perfectly clear. The 4096-6144 taps in these boxes out barley cover it and get things done, and still runout of taps for a GOOD LF manipulation.
Yeah iirs have there place and are very resourceful when dealing with arguably low amount of coefficients. OToH ; if someone could make a purely FIR mixed phase filter for input and one on each output that would be legit.
I'm convinced at this point it's the output delay that kills the idea . So make a HDMI delay that goes with it . But for us , I really don't care if there's 10 full seconds of delay. If I have complete control over every speaker I can do whatever.
Ajusting phase globally is great and all, however in crossover regions it takes the summed phase of two mixing drivers and makes a rotation for both (to whatever degree that may be based on crossover chosen and speaker behavior) drivers to sum them to linearity.
If you had a separate capable FIR bank on each driver you can make the driver linear to include the crossover area. the total sum would be two drivers inphase rather than two drivers that may be completely different in phase , also unually different in brand and specs and size have effectively the same phase at crossover by whatever has modulated into whatever that may be at that mic location.
Using close proximity measurements for separate channel linearizing and listening position mic averaging for gloabal changes...
That's what I was trying to make by buying separate 2x4HDs and running them as 2x2s. It works
And still don't have enough FIR power to do all of the correction in FIR , so that's why I chose to use a Dirac at the input. Free up the mini sharks to do more with each drivers behavior.
So yes, anyone that wants to make a million tap box you have at least one buyer
Want 4 sharks and 5th on the input
Just to make things perfectly clear. The 4096-6144 taps in these boxes out barley cover it and get things done, and still runout of taps for a GOOD LF manipulation.
Yeah iirs have there place and are very resourceful when dealing with arguably low amount of coefficients. OToH ; if someone could make a purely FIR mixed phase filter for input and one on each output that would be legit.
I'm convinced at this point it's the output delay that kills the idea . So make a HDMI delay that goes with it . But for us , I really don't care if there's 10 full seconds of delay. If I have complete control over every speaker I can do whatever.
Ajusting phase globally is great and all, however in crossover regions it takes the summed phase of two mixing drivers and makes a rotation for both (to whatever degree that may be based on crossover chosen and speaker behavior) drivers to sum them to linearity.
If you had a separate capable FIR bank on each driver you can make the driver linear to include the crossover area. the total sum would be two drivers inphase rather than two drivers that may be completely different in phase , also unually different in brand and specs and size have effectively the same phase at crossover by whatever has modulated into whatever that may be at that mic location.
Using close proximity measurements for separate channel linearizing and listening position mic averaging for gloabal changes...
That's what I was trying to make by buying separate 2x4HDs and running them as 2x2s. It works
And still don't have enough FIR power to do all of the correction in FIR , so that's why I chose to use a Dirac at the input. Free up the mini sharks to do more with each drivers behavior.
So yes, anyone that wants to make a million tap box you have at least one buyer
Last edited:
Oabeieo,
Very much agree with all your points.
6144 taps just plain falls short for LF work, or at least it falls short for my current FIR building skill level....
And I agree that the reason products we want don't already exist is most likely due to output delay...and all the forms of havoc delay provides for various applications....
Very much agree with all your points.
6144 taps just plain falls short for LF work, or at least it falls short for my current FIR building skill level....
And I agree that the reason products we want don't already exist is most likely due to output delay...and all the forms of havoc delay provides for various applications....