DSP Xover project (part 2)

I used Najdą only so will recommend nothing. I know people who used minidsp or Behringer.
Najda is basically.digital crossover so i would look in that section . Flex Eight boxed, seems to fill the bill. But from minidsp you can also buy separate modules/PCBs and configure custom solution for your needs. I'm m not familiar with those.
 
Behringer, even when heavily modded (input and output stages) is meh, compared to a standard Najda.
Feed Najda linear power, change the op-amps and run 192KHz🙂 if you can, and it's excellent.

DEQx is expensive, can't comment on the Sound Quality.
Only works at 96KHz.

miniDSP I've always managed to avoid 😂
The 8 and 10 output ones only work at 96KHz.
Just not convinced it has the SQ and usability for me.

I'd be looking at CamillaDSP if there was no Najda, but don't know its details well enough to say if that can meet your requirements.
 
The higher clock frequency reduces low frequency resolution. I have not had my Najda powered up for a while, but I think I was playing with 96kHz. Noting the observations about 96kHz-related performance of Najda, I will think twice about choosing that frequency again. Back when Nick was around, I had questions about the accuracy of the graphic representation of the NUC interface (i.e., the displayed effect of various applied filters vs actual), but Nick dismissed my concerns (I was not convinced, though). So, I'll be trying 48kHz, but I wonder if anyone has positive feedback about Najda's performance at this clock frequency(?).
 
96 Khz is way more than enough unless you are a bat, and linear power supplies are a bad idea for digital circuits.
Feel and hear totally different.

It's not about the sampling frequency of the tracks here, it is the resolution the dsp is working at.
That greater resolution is right across the frequency range, of course.
192KHz is calmer, more composed, a more resolved sound.

Linear power bad for digital circuits 😂
It's about reducing noise injection.

Have a really good listen, compare analogue to your digital, hear that stuff that should be there?
Then eradicate it.

If you can't hear it, well it's not for you, that's all.
 
The higher clock frequency reduces low frequency resolution. I have not had my Najda powered up for a while, but I think I was playing with 96kHz. Noting the observations about 96kHz-related performance of Najda, I will think twice about choosing that frequency again. Back when Nick was around, I had questions about the accuracy of the graphic representation of the NUC interface (i.e., the displayed effect of various applied filters vs actual), but Nick dismissed my concerns (I was not convinced, though). So, I'll be trying 48kHz, but I wonder if anyone has positive feedback about Najda's performance at this clock frequency(?).

I tried 48KHz, was not good.
 
Feel and hear totally different.
Indeed. You often feel you hear stuff that actually isn't there.
It's not about the sampling frequency of the tracks here, it is the resolution the dsp is working at.
Indeed. That is why the precision (number of bits per sample) is more important than the sample rate.
That greater resolution is right across the frequency range, of course.
Whenever someone says "of course" it usually flags something as not based in fact.
192KHz is calmer, more composed, a more resolved sound.
If you say so...
Linear power bad for digital circuits 😂
It's about reducing noise injection.
Exactly. That is why linear power supplies are not a good idea for digital circuits.
Have a really good listen, compare analogue to your digital, hear that stuff that should be there?
Then eradicate it.

If you can't hear it, well it's not for you, that's all.
I suggest you have a really good listen too - but only with your ears. In a proper, level-matched double blind test.
 
Dear Juff,
I used the English 'feel' expression knowing it would light you up..
It's a perfectly valid expression.

All very good in theory.

What is your actual experience of Najda?
Have you built one, own one, listened to one extensively?

What implementation do you have?
What drivers do you use?
How many channels do you use with Najda?
Mouser PSU or other giving +5v and 12-0-12v?
Have you played SPdif vs I2S input, tried the analogue input?
What gain issues might you have come across?
What slopes do you use, what room correction?
What crossover points?
Have you listened to it at 48KHz / 96KHz / 192KHz?

Pictures of your build?

Do tell.
 
Last edited:
What do you mean by "low frequency resolution"? Word length/precision is much more important (for intermediate calculations).
I mean the smallest increment in frequency that is available to the user between filter centre frequencies at low frequency end of the audio band. At the highest sampling rate this figure is larger, but reduce the sampling rate and you can get a finer frequency selection on the low end.

Example: The setup is configured using biquads to set up your crossovers, including PEQ. Then FIR is applied to do room correction (typically done for below 200Hz, YMMV). You will quickly see that at higher sampling rates frequency selection is quite course and at lower sampling rates it gets finer (but with current Nadja level of development, still with room for improvement).

I was in Nick's ear to implement decimation - which would enable partitioning of the audio band with different, optimised, sampling rates applied to each sub-band. This he had committed to doing, but he never got that far.
 
  • Like
Reactions: Julf
Calling all Nadja knowledgeable..

Here are two plots of a standard Holmimpulse test signal sweep.
Something I've used since I first got Najda V1.0 all those years ago and I have found invaluable in shaping the output in various rooms, to achieve a sound I like.

The blue plot is with Najda set to upsample at 192KHz,

the red plot is with Najda set to upsample at 96KHz.

The crossover frequencies, slopes / room correction (tapped horn subs only) and time alignment settings to my 4 way Najda V1.3 (+ tweeter on passive X/O) set up, are identical for both 96 and 192KHz setups.
As are the input dB reductions I have.

Clearly something is going on between 96KHz and 192KHz!

This is measurable, repeatable and I assume accurate enough, as Holmimpulse has been a great tool for me.

Everything in the room was of course identical for both sweeps.

So! Implementation differences in the way Nick handled 96 vs 192KHz? Wittingly or unwittingly?

I'm at a loss to explain what I am heading and measuring, otherwise.
Oh yes, the differences are audible between the two when playing music. Esp the mid and upper mid frequencys - typically vocals etc.
I mean, look the dB difference!

This of course just measures spl. Quality or grain/less grain in the music presentation, is another thing..

Najda 192 vs 96KHz Upsampling.PNG