A 3 way design study

Crossover updated 😀
1656265000075_2wayhorn_fullpolar3 XO-schema-1.png


1656265000301_2wayhorn_fullpolar3 Six-pack.png

Probably, i think i will get a bit more energy and reduce cancellation and nulls around crossover in the vertical response, if i make this entire configuration 2.5 way and roll off one woofer earlier around 350Hz. The 2 woofer responses itself causes some lobing around crossover in vertical direction.

The two woofers themselves are causing some lobing around 1kHz
Fluid had shown something similar to this long back in his sims for twin woofer configuration as shown below if i understood it correctly
DW V Polar (2).png
 
Last edited:
One probably wants some kind of high pass for the tweeter, perhaps even a capacitor despite DSP to protect it for fault conditions and accidents. For the rest of it it doesn't matter mucho which filters to use as the acoustic response is what matters in the end, with resulting group delay and phase stuff no matter what filters were used to get the response.

Very high low pass is neat trick, Geddes used one years ago and perhaps others before him and is quite easy to come up with in VituixCAD as well. Small cheap capacitor, good response and protection. Works very well with ~constant directivity waveguides and compression drivers, not too well with direct radiating tweeters as they are quite flat already, hence you don't see this implemented too often just because the constant directivity speakers are still a rare thing.
 
Is there any particular advantage if a high-pass is used in place of a low-shelf ? Wouldn't that also reduce the number of biquads required to implement the filter ?

As far as i know, a single digital biquad can implement upto any second order transfer function. In this case I have used a 1st order high pass filter which can take up a single biquad. Does a low shelf filter not require a biquad?

Anyway in the current system configuration/DSP capabilities I have, the number of Biquads available is endless.. 😀 Because Equalizer apo running on my laptop can implement any digital transfer function with any kinds of filtering i will ever need. Currently I run 8192 tap FIR filters per driver. Each FIR filter transfer function can contain all the frequency shaping features needed for Driver passband linearization, crossover filters and phase linearization if required. This will result in about 185ms of delay as per EQApo but at the moment my music collection is in harddrives and laptop and hence delay is a non issue.. 🙂

Also the current response and crossover that i showed above is just a prototype that i cooked up in 15minutes. I will refine it further. As you pointed out, in a real hardware dsp implementation, biquads are limited, FIR capabilities are scarce. Hence we need to implement most of the things with IIR filters. In such a scenario every biquad should be carefully used.

In fact, the type od crossover blocks i used are to be used with IIR filter based DSP implementation. With FIR capability, to utilize it better, i should have done it smarter.. 🙂 But it doesn't matter given that i can change crossover design any time and Vituixcad can get me the impulse response files for convolution of any arbitrary transfer functions that i want.FIR or IIR. 🙂
 
One probably wants some kind of high pass for the tweeter, perhaps even a capacitor despite DSP to protect it for fault conditions and accidents. For the rest of it it doesn't matter mucho which filters to use as the acoustic response is what matters in the end. This is neat trick, Geddes used one years ago, perhaps others before him and is quite easy to come up with in VituixCAD as well. Small cheap capacitor, good response and protection. Works very well with ~constant directivity waveguides and compression drivers, not too well with direct radiating tweeters as they are quite flat already, hence you don't see this implemented too often just because the constant directivity speakers are still rare.
In fact i already have a 150uF capacitor in series with the compression driver. I can try a smaller value if required.. 🙂 All measurements of the compression driver on horn that i posted above include this series capacitor also.
 
And number of biquads shouldn't matter at all, if I remember correctly the ubiquitous ADAU1701 DSP chip for example uses all the processing capacity on every cycle, meaning it doesn't matter how many biquads there is as long as there is not too many, which I suppose is plenty. Its just math that is calculated for every sample thats it ( well, I'm not too familiar with how DSP works so, but its not black magic, just math and computers are good at math ). Only thing that matters is the acoustic response and one has to just implement it what ever the means. If DSP affects sound quality its in the AD and DA stages and associated analog circuitry but I bet the filters one puts in to make resulting acoustic response (for the whole system) is far more important and audible thing than the ADDA stages.
 
Last edited:
  • Like
Reactions: vineethkumar01
Currently I run 8192 tap FIR filters per driver. Each FIR filter transfer function can contain all the frequency shaping features needed for Driver passband linearization, crossover filters and phase linearization if required. This will result in about 185ms of delay as per EQApo but at the moment my music collection is in harddrives and laptop and hence delay is a non issue.. 🙂
There doesn’t have to be any delay in a minimum phase FIR filter. If the peak of the impulse response is at the front instead of centered. A 65,536 tap filter can be 1 or 2 ms latency.

Often the high pass is more like 20K for a waveguide to compensate the mass rolloff. You might be able to undo the effect of the 150uF cap in Vituix to see what you could do with just a small cap for the high pass.
 
  • Thank You
Reactions: vineethkumar01
As far as i know, a single digital biquad can implement upto any second order transfer function. In this case I have used a 1st order high pass filter which can take up a single biquad. Does a low shelf filter not require a biquad?
So I guess there's no particular advantage. Yes, a low-shelf needs a biquad as much as the high-pass.

Anyway in the current system configuration/DSP capabilities I have, the number of Biquads available is endless.. 😀 ...... but at the moment my music collection is in harddrives and laptop and hence delay is a non issue.. 🙂
There's no problem as long as it is played back by a computer.

As you pointed out, in a real hardware dsp implementation, biquads are limited, FIR capabilities are scarce. Hence we need to implement most of the things with IIR filters. In such a scenario every biquad should be carefully used.
Yes, I'm mostly hardware-headed but I think that also ensures judicious use of resources, as you said. I'm currently making a 16ch processor for active multi-amplifier playback (no ADCs like you), with plenty of biquads per band (and therefore per channel), but still like to keep things simplified so as to avoid clutter. Two channel shouldn't be very complicated, even with so many filters, especially if a computer is used.
 
  • Thank You
Reactions: vineethkumar01
There doesn’t have to be any delay in a minimum phase FIR filter. If the peak of the impulse response is at the front instead of centered. A 65,536 tap filter can be 1 or 2 ms latency.

Often the high pass is more like 20K for a waveguide to compensate the mass rolloff. You might be able to undo the effect of the 150uF cap in Vituix to see what you could do with just a small cap for the high pass.
Thanks fluid. 🙂
I will try out both the suggestions.
 
Yes, I'm mostly hardware-headed but I think that also ensures judicious use of resources, as you said. I'm currently making a 16ch processor for active multi-amplifier playback (no ADCs like you), with plenty of biquads per band (and therefore per channel), but still like to keep things simplified so as to avoid clutter. Two channel shouldn't be very complicated, even with so many filters, especially if a computer is used.
Awesome.. 🙂
Is there a thread or posts in this forum or others where I can read up more about about your work..?
I am very interested in DSP based implementations on hardware.
 
...
with plenty of biquads per band (and therefore per channel), but still like to keep things simplified so as to avoid clutter. ...

Don't let this affect you in a way that it keeps you not achieving best possible sound! You can add 100 filters and if it makes the response better it would be better than just using 10, if it was limit from the own head for no good reason. I mean there is no other restriction than our own heads, especially with computer as the hardware. Just make proper measurements so one can see what can be tuned with the EQ, there is no point EQ diffraction for example as its different to every direction and not correctable with EQ other than for one position making everywhere else worse. These are limitations from time before good simulation software and knowledge for us hobbyists existed but now with VituixCAD there is no problem to make "perfect" crossover with DSP as long as the measured data is reliable. Perfect in the sense that only thing that limits the system performance is how it measures, coming up with filters is just matter of fiddling around. Basically only limit for performance is the construct you have built, crossover can be thought as solved nowadays, trivial. Although there is differences what kind of crossover one makes it is no mystery anymore, just matter of playing around and coming up with the best sounding one (needs some experience like it always has, to translate visuals to sound at our own listening environments). This is again matter of the construct, how the system measures and from this insight I've reasoned that to make better sound one needs to build better construct, what ever that is for the application you are using it at!🙂 This is the limit of performance, the construct, as filters and XO can be thought of as ideal because that we are capable of today.

Clutter can lead to mistakes and problems and is worthy thing to keep in mind you are very right with that one. I don't know what kind of DSP software the Equalizer APO or others are what you are using, but to avoid clutter one could be clever with the UI and how things are grouped, if nothing else, to keep the DSP implementation UI clutter free as much as possible while still using as many filters as needed (thinkinf of the Sigmastudio, which is kind of drag'n drop DSP arrangement UI). Anyway, everyone can do what ever they like, just be aware about the constrains if they have reason behind them or not. All we want is best possible sound, right?

ps. checked out ADAU1701, about the cheapest DSP chip, and it has 1024 instructions per sample, what ever that means. I suppose it could be 1024 filters, divide for 4 outputs and one could have 256 per channel. Even if it was 10 instructions per filter it would be still 25 filters per channel, plenty to knock out any peaks and dips from any reasonable system without any extra cost in sound quality as long as the EQ was on things that don't make things worse, IOW for good reason.
 

Attachments

  • adau1701-instructions.png
    adau1701-instructions.png
    46.4 KB · Views: 74
Last edited:
@tmuikku : With Equalizer Apo and VituixCAD, one advantage that we can make use of is we can just design all filters in VituixCAD and see its effects on the driver responses first (including driver passband linearization and crossover filter).
Then VituixCAD can create the impulse response of a "super filter" for each driver that includes the effects of all filters in the driver's signal path and dump it into a file. Now this file is the only thing that we need to use in Equalizer Apo along with any Pre amp blocks/shelving filters, channel mapping, if required. So there is no need to go and again specify all the filters as we designed in VituixCAD in the DSP software. So the interface looks clean.. 😀
This is how my DSP interface looks like on PC:
1656320655707.png

each of the '.wav' filters in above snippet is an FIR "super filter" for each driver 🙂
So if 2 way system, we just need to mention the 2 impulse response files per speaker (let's say left channel). Copy the entire thing for right channel as well. DSP done 🙂

No biquads or anything need to be used in the DSP implementing software if designed this way. EQApo can also be used in the traditional way where we design filters in VituixCAD and then copy all the filters and create it again in EQApo with appropriate plugin like traditional DSP. And there are other convolution softwares as well which might have more flexibilities.
For example, we can have Camilla DSP running on a Raspberry Pi and a multichannel DAC or a compatible audio interface can do all this and more if we don't want to get tied to the PC doing the DSP. In fact I had tried out part of it in the past on streaming music from spotify and TIDAL etc and it worked really well. I will try it out again sometime later. 🙂 Again no ADCs in the chain and plenty capable DSP processor.

Anyway, this is just one easy way of doing things. There are other smarter approaches also as fluid mentioned. I am slowly learning more.. 🙂

@fluid : Can you please point me to a resource where there is some discussion on FIR filter design with impulse peak in the front instead of centered?
 
Last edited:
  • Like
Reactions: tmuikku
Is there a thread or posts in this forum or others where I can read up more about about your work..?
I am very interested in DSP based implementations on hardware.
Don't let this affect you in a way that it keeps you not achieving best possible sound! You can add 100 filters and if it makes the response better it would be better than just using 10.....

I don't know what kind of DSP software the Equalizer APO or others are what you are using, but to avoid clutter one could be clever with the UI and how things are grouped, if nothing else, to keep the DSP implementation UI clutter free as much as possible...

ADAU1701 .......has 1024 instructions per sample .... could be 1024 filters ... Even if it was 10 instructions per filter it would be still 25 filters per channel, plenty to knock out any peaks and dips from any reasonable system.....

Though I'm doing a hardware implementation, I'm not using a DSP in the way you people do, as my application also involves decoding proprietary formats. Here is the link to the rather uninteresting thread where I haven't posted in a while, as progress has been slow. Nevertheless, it helped me realise that most people these days favour PC-based processing, for more or less the same reasons you guys mention above.

https://www.diyaudio.com/community/...quirements-preferences-opinions-ideas.381423/

It is worth considering that some of the peaks/valleys in a response may not be consistent in frequency and /or amplitude and could shift under varying operating conditions, making narrow-band EQs less effective when compared to the broader ones. So, unfortunately, there could be a point beyond which further EQ might not make a big difference, even when resources are available. I believe that this number should not more than 10 biquads, across the entire audio range.
 
  • Like
Reactions: vineethkumar01