• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Building a 2-way small active speaker with software crossover

OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,602
Likes
10,769
Location
Prague
These are the electrical responses I assume. As KTSR mentions, the electronic filter response is in addition to the natural acoustic responses of the drivers (I'm pretty sure you know this, tell me to stop if I am too elementary here...).
Yes of course, I know this. The result is a combination of electrical and acoustical transfer functions. You are right, the comment has been quite elementary :D.
 
  • Like
Reactions: MAB

Sokel

Master Contributor
Joined
Sep 8, 2021
Messages
6,060
Likes
6,081
Normally and "active" speaker has the amplifier(s) and crossover built-in.

It looks like you're building a bi-amplified speaker a with customized active crossover.

And it looks like a lot of work for speaker with a tiny 4-inch "woofer", but that's just my personal bias/prejudice. :p
Isn't 8381A active?
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,602
Likes
10,769
Location
Prague
And it looks like a lot of work for speaker with a tiny 4-inch "woofer", but that's just my personal bias/prejudice. :p
In fact it is a little bit of work, as the crossover is in SW and I use my current setup. No need for any additional HW, no need to build a bulky RLC crossover.
 

dualazmak

Major Contributor
Forum Donor
Joined
Feb 29, 2020
Messages
2,850
Likes
3,047
Location
Ichihara City, Chiba Prefecture, Japan
Hhm, the filter section of EKIO looks very limited to me. No variable Qs for highpass/lowpass is pretty restrictive, you'll have to work around that with additional PEQs etc.
BTW I do not understand your remark on Qs of LP and HP filters. The choice is Butterworth, Linkwitz-Riley and Bessel, which I find appropriate.

Neither do I.
@KSTR, would you please kindly further clarify/explain what would be your concern(s) on EKIO?

Guillaume of LUPISOFT kindly responded to me informing (ref. here);
"EKIO uses IIR filters. The processing is done using a cascade of second order transposed direct form II biquad sections. Every calculation is done using 64 bit floating point numbers."

and EKIO also has really flexible EQ capabilities which can be controlled by mouse operation on nicely designed GUI screen, and even ABX comparator. You can export/import your XO/EQ configuration by its detailed text configuration file;
スクリーンショット 2023-09-21 062243.png
 
Last edited:

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,796
Location
Sweden
Building a 2-way small active speaker with software crossover

1. Goal

To try software crossover method for a 2-way, active speaker

2. Software and instrumentation used

Per @dualazmak hint, I will use EKIO software crossover. It is very easy to use and it is free up to 2 IN/4 OUT channels.
For acoustical SW measurements, I am using REW software in combination with VituixCad , VituixCad is also used for crossover and acoustical simulations.
For acoustical HW interface, I use Behringer ECM8000 microphone with the calibration curve, iConnectAudio4+ soundcard and Roland Duo-Capture Ex soundcard.
As an audio amplifier, I use my A250W4R or AIYIMA A07.

3. Speaker cabinet and drivers

The speaker cabinet used is made from 18mm MDF, dimensions 180 x 280 x 170 mm (w x h x d)

View attachment 313317

View attachment 313318

Woofer is a 4" Magnat MW 111 CP 754-I (no link unfortunately)
View attachment 313319

Tweeter is a Monacor HT-88 horn loaded PA driver
View attachment 313320

4. Connectors

I need 4 wires to bring to the speaker, so the 4-way Speakon-like Cliff connector is used.

View attachment 313322

5. Crossover

I plan to use 48db/oct LR filters for the woofer and the tweeter. This will be completed with baffle step compensation shelf filter and necessary slope filter for the tweeter, with gain compensation between the low frequency and high frequency crossover sections. The goal is to get as flat on-axis response as will be possible. With the tweeter used, the useful HF range will be restricted to some 12kHz-15kHz, but I do not care, it fits to my hearing limit, being near to 70 years old :).

As already stated, the main goal is to try software crossover method. The passive R-L-C crossover with baffle compensation and air-core inductors would be that big that it would not fit inside the cabinet. And, the SW crossover is much more flexible for changes, and, is able to provide necessary filter slope to cut parasitic driver behaviour.

I am not posting the preliminary simulations, I will wait for first real measurements and post them then.
Looks like a nice project. With my 3 years of work with the HYBRID, an active dsp 3 way speaker with very good directivity, I came to the conclusion that I most often prefered a linkwitz Riley acoustical crossover 24 dB LP/HP for the mid and tweeter and a third or fifth acoustical order crossover at 75 Hz . We have made hundreds of comparisons with different topology .
This is also the conclusion GRIMM audio has made regarding the use of a 24 dB linkwitz riley crossover .

A 48 dB slope or higher order crossover has audible limitations in certain ways even if the dsp crossover is a very good one . Ofcourse the measurements will look great ;)

If you want a really good result, you must play with two dsp speakers in a stereo setup and do the necessary stereo system compensations + 1,5 dB q3 at 1,8 KHz , - 1,5 dB at 3,5 KHz and + 1,5 dB at 6,5 KHz . Very easy to implement with a dsp. Some, but not all recordings in stereo will sound better this way .
This is documented on the Swedish forum Faktiskt.io , some of it by Ingvar Oehman.

This is if you want the next level of performance in sweetspot , beyond a result for mono optimised speakers .

If you need to do notch filtering for the midbass, a better result can be had using a passive filter , as MAB has shown.

The art of making good stereospeakers are very complicated and one must read a lot and do own experiments about what matters, we hear very different compared to a microphone , the mic takes up all the sound , the brain/ear starts selecting sounds at a certain distance .
 
Last edited:

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,377
Likes
18,295
Location
Netherlands
This is also the conclusion GRIMM audio has made regarding the use of a 24 dB crossover .
They also conclude that cables and clocks matter, I’d hardly call that a benchmark.
If you want a really good result, you must play with two dsp speakers in a stereo setup and do the necessary stereo system compensations + 1,5 dB q3 at 1,8 KHz , - 1,5 dB at 3,5 KHz and + 1,5 dB at 6,5 KHz .
Where does this come from? What’s the science behind this?
 

Geert

Major Contributor
Joined
Mar 20, 2020
Messages
1,945
Likes
3,551
Where does this come from? What’s the science behind this?

This I guess, if he corrects the highest frequency he mentioned:
index.php

However, fixing this with basic eq doesn't work as a fix should only affect the stereo phantom image an not the frequency response of L and R signals.
 

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,796
Location
Sweden
They also conclude that cables and clocks matter, I’d hardly call that a benchmark.

Where does this come from? What’s the science behind this?

Start reading - Google translate is your friend

At the beginning of the thread :

”For the simplest case, we can imagine that we are playing a mono signal (something that is panned in the middle, eg a singer). If we think that she is standing and singing for real in front of listeners, we will have a tone curve deviation right in front of the outer ear, which is primarily due to the shape of the head. We can have this deviation as a reference. That's how it should be when we listen to mono sound. If you let the singer sing through two speakers, as the picture shows, we will not have the same tone curve deviation in front of the outer ears. The situation is symmetrical so we only need to check one ear. Let's select the right ear. What happens? The sound from the right speaker will contain more high-frequency sounds in that it plays more 'straight into' the ear. The sound from the left speaker, however, will lack a bit of treble. If you combine the sound from the left and right speakers and measure at the ear, due to the above differences, the tone curve will not be the same as if the sound had come straight through. Then you can simply make sure that you adjust the direct tone curve from the speaker so that the sound is included as if it had come straight from the front. Then you will get a more 'living' sound image.

Now this is not entirely without complications... When you play sounds that are completely panned either to the right or to the left, you get a deviation that is infected by the compensation to make it sound right on mono sound. It's probably less of a problem because there you have quite a lot of agreement with where the sound should give the illusion of coming from and where it really comes from :) It will still be right s a s. Although it is still a problem because we often use relative tone curve differences to determine directions in the soundscape. In addition, it is not so easy to just check the difference between center-panned sounds and real straight-ahead sounds. You have to see how it should behave from the whole sound image from the front (perhaps at least +/-30 degrees) and make a balance. There will be a lot of trade-offs! 8)

Another stereo error artifact that IÖ fixes in the pi60(s) is the height projection. It also contributes to a more convincing released sound image. Although it does not affect the sound on axis on the speakers.”

IMG_4272.gif
 
Last edited:

Tangband

Major Contributor
Joined
Sep 3, 2019
Messages
2,994
Likes
2,796
Location
Sweden
This I guess, if he corrects the highest frequency he mentioned:
index.php

However, fixing this with basic eq doesn't work as a fix should only affect the stereo phantom image an not the frequency response of L and R signals.
Correct - to do it right one must also play with the baffle shape , as done in the speaker Ino pi60 and Guru qm60.


But one can get a hint of it with dsp manipulation using a flat baffle .

Now, - back to PMA :s active dsp speaker and sorry for OT
 
Last edited:

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,752
Likes
6,147
Location
Berlin, Germany
BTW I do not understand your remark on Qs of LP and HP filters. The choice is Butterworth, Linkwitz-Riley and Bessel, which I find appropriate.
@KSTR, would you please kindly further clarify/explain what would be your concern(s) on EKIO?
I've designed numerous analog active and passive speaker crossover as day job and never, absolutely never, any of the highpass or lowpass filters was one of the textbook filters. Textbook filters are only useful, in very rare cases, for final acoustical target... in practice you need variable Q. You know, the textbook filter XO phase response falls apart the moment you have another slope in the vicinity, be it a natural roll-off (LF system corner frequency) or be it the XO to an adjacent way. In a two-way, these additional slopes are sufficiently far away from the XO point, but from three ways up (notably including any subwoofer XO) you cannot use any textbook XO filter function anymore as it stops summing correctly because phases are are not matching anymore. And that's just for the final target functions, mind you.

For example, if you settle for a target for low-to-mid 3rd-order XO with a constant 60 degree phase offset (because 90 degree standard 3rd order Butterworth would give more lobing issues and less efficient summing etc), how you are going to do that without free access to filter Q? As mentioned, you then must use additional PEQ and shelves to bend your fixed filter's magnitude and, much more importantly, phase to the required target. Again, that's for the target function alone, not the actual required electrical filter which is pretty much arbitrary and thus needs arbitrary control over parameters.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,377
Likes
18,295
Location
Netherlands
This I guess, if he corrects the highest frequency he mentioned:
index.php

However, fixing this with basic eq doesn't work as a fix should only affect the stereo phantom image an not the frequency response of L and R signals.
That's almost like the Harman target curve for headphones, just missing the final high-frequency correction. Some others seem to add that back in though. If this is a thing, why don't we see it anywhere?
 

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,241
Likes
1,389
Location
Budapest
However, fixing this with basic eq doesn't work as a fix should only affect the stereo phantom image an not the frequency response of L and R signals.

You can do that with M/S processing (applying the desired EQ filters on the Mid channel only)
 

3ll3d00d

Active Member
Joined
Aug 31, 2019
Messages
212
Likes
176
would you please kindly further clarify/explain what would be your concern(s) on EKIO?
various filter topologies can be constructed using variable Q 2nd order filters so if you have access to that directly then you can implement some other filter types if the software doesn't support what you want directly

for example, jriver has relatively weak direct support for filter types but does support variable q filters. From this I could implement (in https://yabb.jriver.com/interact/index.php/topic,129609.0.html) a couple of different types of bessel filters and also matched delay subtractive filters as well being able to use any specific individual filter I want. In the absence of this, you have to approximate using peaking filters which is not ideal. If ekio supports biquad coefficient import then it's not an issue as you can just do that design externally and then import.
 

Geert

Major Contributor
Joined
Mar 20, 2020
Messages
1,945
Likes
3,551
That's almost like the Harman target curve for headphones, just missing the final high-frequency correction. Some others seem to add that back in though. If this is a thing, why don't we see it anywhere?

Like I said, because it's not straightforward to implement. To do it right you need an advanced solution like https://www.bacch.com (their initial basic stereo solution, not the new spatial stuff), or some DIY DSP algorithm as suggested above (which is not at the level BACCH is).

And not all records need this fix, the issue might already be accounted for in the recording process. (Although I have different great sounding albums which missed this problem).
 

Sokel

Master Contributor
Joined
Sep 8, 2021
Messages
6,060
Likes
6,081
This is one channel built as I have understood,right?
So no need for a multichannel DAC yet.
I'm looking forward for the finished built measurements.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,377
Likes
18,295
Location
Netherlands

Start reading - Google translate is your friend

At the beginning of the thread :

”For the simplest case, we can imagine that we are playing a mono signal (something that is panned in the middle, eg a singer). If we think that she is standing and singing for real in front of listeners, we will have a tone curve deviation right in front of the outer ear, which is primarily due to the shape of the head. We can have this deviation as a reference. That's how it should be when we listen to mono sound. If you let the singer sing through two speakers, as the picture shows, we will not have the same tone curve deviation in front of the outer ears. The situation is symmetrical so we only need to check one ear. Let's select the right ear. What happens? The sound from the right speaker will contain more high-frequency sounds in that it plays more 'straight into' the ear. The sound from the left speaker, however, will lack a bit of treble. If you combine the sound from the left and right speakers and measure at the ear, due to the above differences, the tone curve will not be the same as if the sound had come straight through. Then you can simply make sure that you adjust the direct tone curve from the speaker so that the sound is included as if it had come straight from the front. Then you will get a more 'living' sound image.

Now this is not entirely without complications... When you play sounds that are completely panned either to the right or to the left, you get a deviation that is infected by the compensation to make it sound right on mono sound. It's probably less of a problem because there you have quite a lot of agreement with where the sound should give the illusion of coming from and where it really comes from :) It will still be right s a s. Although it is still a problem because we often use relative tone curve differences to determine directions in the soundscape. In addition, it is not so easy to just check the difference between center-panned sounds and real straight-ahead sounds. You have to see how it should behave from the whole sound image from the front (perhaps at least +/-30 degrees) and make a balance. There will be a lot of trade-offs! 8)

Another stereo error artifact that IÖ fixes in the pi60(s) is the height projection. It also contributes to a more convincing released sound image. Although it does not affect the sound on axis on the speakers.”

View attachment 313554
So this is not just something you can do with a simple EQ on every speaker, because the EQ you should apply is only valid for part of the sound: the bit that is common in both channels (sound coming from the front).
 

ppataki

Major Contributor
Joined
Aug 7, 2019
Messages
1,241
Likes
1,389
Location
Budapest

maty

Major Contributor
Joined
Dec 12, 2017
Messages
4,596
Likes
3,168
Location
Tarragona (Spain)
@pma
EKIO looks great!
Alternatively you can use Jriver with whatever VST plugin you want for the EQ, crossover, delay, etc.
With that you can even have linear phase crossover.... (probably not 48dB/octave due to pre-ringing but 24dB shall work fine with something like the T-Racks Linear Phase EQ)
In Jriver I recommend using DDMF Metaplugin to manage your plugins for each channels

I agree. JRiver MC 64 bits and its DSP have allowed me very fine tuning to the sound I get from my internal modified KEF Q100 coaxials (plugged front bass-reflex). And the JRMC convolution of the filter created with rePhase.
 
Top Bottom