Post Directiva r1 Active Crossovers here

[thorvat]

FIR crossovers take a serious amount of processing power so you would probably be limited to using a PC for the crossover

RPI 3+ or RPI 4 can easilly run convolution of FIR filters with 65536 taps.

 

[mdsimon2]

As @thorvat showed higher frequencies are easy to do with a few taps, low frequencies are what require a lot of taps.

Michael

 

[thorvat]

As @thorvat showed higher frequencies are easy to do with a few taps, low frequencies are what require a lot of taps.

Michael

Exactly. But even with 1024 taps you can still design effective room EQ FIR filters in the 20-300Hz region.

 

[abdo123]

RPI 3+ or RPI 4 can easilly run convolution of FIR filters with 65536 taps.

and these are computers are they not? just a single board computer.

 

[headshake]

I made another XO. This is a "head shadow" XO. From what I understand, there is no such thing as a flat 1-2khz to our ears. These frequencies wrap around our head/shoulders/neck and cause a dip. So like some JBL speakers, I put a nice dip there.

The puri woofer breaks up around 3khz. This XO completely removes that freq. from playing. ( I also think the purifi humps on the face mess with the 1/2 wave of some freq. but that is another thing).


left: head shadow XO , right: default XO .... How batshit is the vertical with a steep XO? I think if the goal is listening at a distance it should not matter.


Six pack:


It is not the prettiest that is for sure.

edit: one can refine forever, but this is a bit more balanced:

 

[thorvat]

and these are computers are they not? just a single board computer.

With much less computing power than any PC. What I wanted to stress out is that today you can run FIR filter convolution on pretty much any device.

 

[TimVG]

@Rick Sykora

Hey Rick - will you be trying some of these alternate crossovers in the future?

 

[Rick Sykora]

@Rick Sykora

Hey Rick - will you be trying some of these alternate crossovers in the future?

Yes ofc!

At the moment though, am measuring the tweeter off-axis response for ctrl and next hoping I can get in an outdoor measurement before it rains here.

 

[Absolute]

Been experimenting alot lately with crossovers in my DIY M2 with beryllium large format compression drivers that allows me to play around more than the original M2 compression driver would.

The crossover slopes matters a whole lot more than I originally thought it would and I just can't get a 48 dB crossover (no FIR) to sound decent. I think perhaps the phase shift is too great and is most likely the culprit.

How does a 48 dB crossover look like compared to a 36 dB LR (tweeter inverted) and 24 dB LR in the verticals for this speaker?

Edit; ah, I see above. Excellent stuff. 48 dB doesn't look too good.

 

[TimVG]

Been experimenting alot lately with crossovers in my DIY M2 with beryllium large format compression drivers that allows me to play around more than the original M2 compression driver would.

The crossover slopes matters a whole lot more than I originally thought it would and I just can't get a 48 dB crossover (no FIR) to sound decent. I think perhaps the phase shift is too great and is most likely the culprit.

How does a 48 dB crossover look like compared to a 36 dB LR (tweeter inverted) and 24 dB LR in the verticals for this speaker?

Edit; ah, I see above. Excellent stuff. 48 dB doesn't look too good.

I believe you may be focussing a bit too much on the curve averages. If you look at the on-axis/LW behaviour, it's quite excellent, as is the vertical directivity and the horizontal slopes.

An 8th order LR filter can sound great, but you need to make sure you implement it properly. This means time alligning the drivers perfectly halfway between the two voice coils and checking for summation (or if you flip the polarity check for cancellation). I have a similar system (15" woofer + horn) which I use with an 8th order LR filter, it works absolutely great.

 

[TimVG]

A proper 8th order LR filter also gives a solid 20° vertical window. When compared to a more traditional crossover it's easy to see that even 10° off axis has an audible impact on the sound.

 

[TimVG]

I don't want to dampen your enthusiasm (probably do anyway), but you should be careful to keep within the E12 range for component values. Otherwise, people will have to start unwinding coils.

Since, due to non-optimal initial measurements, there are slight differences between VCAD and real measurements (which is usually not the case), you should always include the difference from the measured version v20 as an overlay to have a reference.

And it would be, IMO, better to discuss such specific things in the Directiva crossover thread.

Good advice. I will look into that.

As for the overlay. I used the files from the measured version as posted by Rick, simply deleted the crossover, and started from scratch. So this was simulated using the real measurements.

 

[ctrl]

As for the overlay. I used the files from the measured version as posted by Rick, simply deleted the crossover, and started from scratch. So this was simulated using the real measurements.

That's exactly the point. The initial measurements of the drivers are not quite "cleanly" executed. Therefore, there are slight deviations between simulated crossover and real measurements.

For example, in version v20, the dip in the sound power frequency response became deeper than intended and will most likely need to be compensated for during fine tuning.
But we know the difference between simulated version v20 in VCAD and Amir's CEA2034 measurement of the Directiva.

We know the deviations of v20 from the real measurements. Therefore, v20 serves as a reference to make estimates about other simulations.

 

[TimVG]

That's exactly the point. The initial measurements of the drivers are not quite "cleanly" executed. Therefore, there are slight deviations between simulated crossover and real measurements.

For example, in version v20, the dip in the sound power frequency response became deeper than intended and will most likely need to be compensated for during fine tuning.
But we know the difference between simulated version v20 in VCAD and Amir's CEA2034 measurement of the Directiva.

We know the deviations of v20 from the real measurements. Therefore, v20 serves as a reference to make estimates about other simulations.

Understood. Will there be further measurements made which will be correct? Just thinking in terms of passive crossovers that for potential builders it would be handy not to have to consider component changes to adjust for data offset.

 

[Rick Sykora]

Understood. Will there be further measurements made which will be correct? Just thinking in terms of passive crossovers that for potential builders it would be handy not to have to consider component changes to adjust for data offset.

Yes, I will generate some spins using the newer woofer that I have.

Also, we were waiting to post a passive crossover because of potential changes after I listen in stereo.

 

[TimVG]

Yes, I will generate some spins using the newer woofer that I have.

Also, we were waiting to post a passive crossover because of potential changes after I listen in stereo.

I'll just wait until you post a new woofer file then before doing anything further

 

[morpheusX]

I'm "building" a pair of active speakers, i'm waiting to receive the boxes from the builder, and while i wait, i'm researching the tools i need to be able to create a valid active crossover.

From an "design" perspective, i like the approach described in the Grimmaudio LS1 paper, and their preference for LR24 filters:

I must confess to an unconditional preference for Linkwitz-Riley crossover filters. I find it amazing that for an art like crossover design that has been practiced for so many years, it has taken so long before someone finally took time to sit down and work out the correct way of doing it. There is no excuse for anything less than full phase coherence between drivers so that they cross over at -6dB. It just doesn’t get any better. Ever. 4th order roll-off will produce approximately 2nd order electrical slopes, which is enough for power handling.

I don't have the knowledge to discuss the above statement, but i found it interesting given the consensus on the quality of the LS1 speakers.

Here is my attempt using the provided measurements:

Crossover

SPL

GD & Phase

Power & DI

Filter

Directivity

Impedance

Preference Rating

I would like to request your help to identify the potential problems with this crossover, and what data indicates those potential problems.
Also, if possible, identify what could be done to improve the end result.

Thank you for this amazing effort with Directiva R1!

 

[McFly]

Looks pretty good! My main comment isn't on the crossover, but your scaling - keep Y scale at 50db so you see 50-100db and not 10-100db. Otherwise you can make a very bad speaker look a lot flatter than it really is!

Due to the scaling, I tried it and saw a broad rise around 1khz, so to fix this I would suggest you shift the woofers IIR 750hz peak filter just up to 870hz and drop the level down a bit more. Gets it a bit flatter.

And remember, this is only for Amir's actual Directiva speaker. The team found a few differences in responses in the design thread between woofers depending on their manufacturing date so a builder is advised to do their own driver measurements if possible.

That's not to say building these and using v20 crossover would sound rubbish, it would sound great, might just need a lil tweak here and there for your own version.

 

[McFly]

Side rant; for ACTIVE crossovers, while I think its the greatest speaker software anyone could ask for - VituixCad should be able to have separate Source (x1 at the front) and Amplifiers (generators x2 or x10 whatever) and these should be placed after the active crossover components, just to show the correct topology of a DSP active design. That would also remove the crazy impedance total (which isn't actually seen by any amplifier) and any confusion.

 

[morpheusX]

Thank you very much for your input.

Looks pretty good! My main comment isn't on the crossover, but your scaling - keep Y scale at 50db so you see 50-100db and not 10-100db. Otherwise you can make a very bad speaker look a lot flatter than it really is!

Strange, the main window is using a 60db-100db scale, i probably missed something when i copied the images ...

Due to the scaling, I tried it and saw a broad rise around 1khz, so to fix this I would suggest you shift the woofers IIR 750hz peak filter just up to 870hz and drop the level down a bit more. Gets it a bit flatter.

I tried what you suggested, but although the FR is flatter, the Preference score lowers from 7.037 to 6.939 (don't know if it matters ):

That's not to say building these and using v20 crossover would sound rubbish, it would sound great, might just need a lil tweak here and there for your own version.

Thank you, but i'm using these files only as a learning lesson, i'll be using a very different set of drivers