DSP Xover project (part 2)

Didn't do the display swap.
I moved the upper mid and tweeters forward. Physically time aligned the tweeters and upper mids and did the rest inc the tapped horns in DSP.
Horns in free space sounds a bit different and has more presence which is good.

I probably have about 15 EQs in total on the subs, bass and mids. I prefer the sound non EQ'd on the upper mid Vitavox S2's - they sound more alive and natural. Perhaps that's cos' I've heard them so much without. I don't know?

I got the Mouser PSU
MTW30-51212 TDK-Lambda | Mouser
PSU. Quite a bit bigger physically. Still just fits in my metal case. Connected it up - same great sound. No dummy load needed on this one. Draws 4.5w of mains on Standby and 8.7w with Najda working.
Thanks for the tip off kazam.

Display can wait;) Listening time now...

Looking forward to the updated SW.
 
OK I'll do that.

For example, you had this FIR LR approximation with one 511-tap LF and one 255-tap HF.
We found out that it's better to take the delay into account when computing the phase otherwise the Electrical Sum looks ugly.
So I changed this and now the Sum of your FIR looks beautiful, that's great.

But that will maybe introduce problems for other users who keep to IIR filters because they use delays for another purpose. So, if you design a IIR LR4 crossover, you expect that the bands sum flat, right? But if you have added delay on one channel, then it's not going to sum flat anymore.

The above is right because we make an electrical sum. It's like if you were taking the output channels and summing them in the analogue domain - meanwhile when you add delay in a IIR setup, you do that in order to time-align your drivers.

(Anyway, that's a long talk I've been having with myself and it's not really clear to me :) )


/QUOTE]

Hi Nick,
Yeah, it was plain luck with the woodwork (angled baffle) that I had to use 128sample delay between low and high (exact acoustic fit between drivers). I guess you have to tell the soft the acoustic delays.
I won't interupt your discussion with yourself in that area, but here is something to sleep on... have a look on the LP of this LR4 (attached). Is it correct?

best,
Paal
 

Attachments

  • LR4 LP dev.pdf
    91.7 KB · Views: 95
I prefer the sound non EQ'd on the upper mid Vitavox S2's - they sound more alive and natural. Perhaps that's cos' I've heard them so much without. I don't know?

Yeah, a good audio guy I know used to tell me: "the best processing is no processing". I suppose he meant "no useless processing".

I got the Mouser PSU
MTW30-51212 TDK-Lambda | Mouser
PSU. Quite a bit bigger physically. Still just fits in my metal case. Connected it up - same great sound. No dummy load needed on this one. Draws 4.5w of mains on Standby and 8.7w with Najda working.
Thanks for the tip off kazam.

That's great to hear, thanks.

Yeah, it was plain luck with the woodwork (angled baffle) that I had to use 128-sample delay between low and high (exact acoustic fit between drivers). I guess you have to tell the soft the acoustic delays.

Well, if we're talking about the same configuration (511 taps/255 taps), then the 128-sample delay you have added to the tweeter was to compensate for the extra delay on the woofer due to more taps. So it has nothing to do with acoustic path. Not sure though we talk about the same setup.

I won't interupt your discussion with yourself in that area, but here is something to sleep on... have a look on the LP of this LR4 (attached). Is it correct?

Hmm you don't like it because it's not perfectly symmetrical right? :)
 
Well, if we're talking about the same configuration (511 taps/255 taps), then the 128-sample delay you have added to the tweeter was to compensate for the extra delay on the woofer due to more taps. So it has nothing to do with acoustic path. Not sure though we talk about the same setup.



Hmm you don't like it because it's not perfectly symmetrical right? :)

Hi Nick,
I think it is the same thing, but the way I discovered the delay was when I measured the speaker. The pulse response showed about 128 samples delay (well actually 130) between the drivers. This could not be merely physical placement. I then understood that most of it was "difference in prosessing" delay.:)

You are right on the symmetrical issue;) I had to use a lot of eq's to get it back to LR4:)
HP seems perfect so why is this?

best,
Paal
 
This is getting very interesting. What kind of Dac's?

I am fed up with waiting for the Hypex DLCP.......

This is out, apparently
Hypex Electronics BV - DLCP
and costs 550€ + VAT.

Anybody tried it? Every other mini digital crossover i find (Behringer DCX2496, etc) seem to be good for car audio but not for studio monitoring.

I'm looking for a digital crossover i can use in a pair of DIY studio monitors without having to design an analogue crossover.

Thanks
 
Indeed I'll make a soft+firm revision available tomorrow, thanks Steve. The release is now under test.

In the program:

1. Independent analogue output gains.
2. More flexibility with FIR taps.
3. EQ filters now downto 10 Hz instead of 20 Hz previously.
4. Switch on/off and preset switching revisited in order to minimize pop in loudspeakers.
5. Output graphs now take delays into account - optionally - when computing phase.
6. Two more steps of LCD contrast adjustment.
7. Led 7 will light on when one of the DSPs (or both) are overloaded.
8. Bugfix for this silence issue that Steve reported.

Hope you'll like it.

Best,

Nick
 
I was already asked this question.

You will have some elements of answer if you watch this
video.

It's getting really into the topic from min 12 on - so be patient :)

That was an interresting lesson. Even if most of it went over my head, I think I got the basic picture.:)

Indeed I'll make a soft+firm revision available tomorrow, thanks Steve. The release is now under test.

In the program:

1. Independent analogue output gains.
2. More flexibility with FIR taps.
3. EQ filters now downto 10 Hz instead of 20 Hz previously.
4. Switch on/off and preset switching revisited in order to minimize pop in loudspeakers.
5. Output graphs now take delays into account - optionally - when computing phase.
6. Two more steps of LCD contrast adjustment.
7. Led 7 will light on when one of the DSPs (or both) are overloaded.
8. Bugfix for this silence issue that Steve reported.

Hope you'll like it.

Best,

Nick

Great news! Looking forward to try it.

best,
Paal
 
That was an interresting lesson. Even if most of it went over my head, I think I got the basic picture.:)

Yes these are great DSP classes!! The professor talking there is Oppenheim himself. He co-authored with Schafer what became the Number 1 textbook on digital signal processing during many years. It's dating a bit now, but it's still absolutely relevant. There are other videos that I recommend watching if you're interested in DSP.

Coming back to your question, let's take a simple example of an analogue low-pass LR4 cutting at 10 kHz.
At 10 kHz, the filter is attenuating the input signal by 6 dB. You're OK with this, right?

Let's agree that the slope of the filter is 24 dB/oct straight after the corner frequency (it's not exactly the case, but it doesn't matter here).
This would mean that at 20 kHz, your analogue filter is attenuating the signal by 6+24 = 30 dB.
At 40 kHz, attenuation equals 6+24+24 = 54 dB.
At 80 kHz, attenuation equals 6+24+24+24 = 78 dB.
Etc.

You notice that the theoretical analogue model is not limited in frequency.

When you translate an analogue filter for the digital domain, sampling comes in with the problem that the system is limited in frequency. If you sample say at 48 kHz, then any frequency above Nyquist = 48 kHz/2 has no meaning.
With a system sampling at 48 kHz, we can't specify an attenuation by 54 dB at 40 kHz - it's just impossible.

This gives you an intuitive hint that digital realization of analogue filters are only approximations of the theoretical analogue model.
 
Yes these are great DSP classes!! The professor talking there is Oppenheim himself. He co-authored with Schafer what became the Number 1 textbook on digital signal processing during many years. It's dating a bit now, but it's still absolutely relevant. There are other videos that I recommend watching if you're interested in DSP.

Coming back to your question, let's take a simple example of an analogue low-pass LR4 cutting at 10 kHz.
At 10 kHz, the filter is attenuating the input signal by 6 dB. You're OK with this, right?

Let's agree that the slope of the filter is 24 dB/oct straight after the corner frequency (it's not exactly the case, but it doesn't matter here).
This would mean that at 20 kHz, your analogue filter is attenuating the signal by 6+24 = 30 dB.
At 40 kHz, attenuation equals 6+24+24 = 54 dB.
At 80 kHz, attenuation equals 6+24+24+24 = 78 dB.
Etc.

You notice that the theoretical analogue model is not limited in frequency.

When you translate an analogue filter for the digital domain, sampling comes in with the problem that the system is limited in frequency. If you sample say at 48 kHz, then any frequency above Nyquist = 48 kHz/2 has no meaning.
With a system sampling at 48 kHz, we can't specify an attenuation by 54 dB at 40 kHz - it's just impossible.

This gives you an intuitive hint that digital realization of analogue filters are only approximations of the theoretical analogue model.

Hi Nick,
Thanks,
So basically changing my samplerate to 96kHz solved my problems :)

The new release is available for download at the usual place.

It includes a FW update as well, so this will reset your board to the factory settings. Make sure not to pull off the USB cable or the power chord while the new FW is being transferred :)

Great!
The update went well. Can you tell a bit about the flexibility of the FIR taps?

best,
Paal
 
It's getting more and more appealing :)

Just for my understanding, doesnt the max samplerate 192kHz give the best sound quality ? Is there a reason to stay on 96 kHz (save processor resources or whatsoever)?

BR
Jean-Louis

I preferred the sound / presentation of 96KHz. I did those tests before I applied my crossovers and some correction.
192 sounded glassy and artificial. 48 sounded rather lowfi. 96 was sweet and pretty much transparent comparing A/B to my passives.
Slightly different presentation but pleasing.

I am at 52% and 47% load currently.
 
Last edited: