Mini-DSP only at 48 or 96 Khz ?

[ofrappier]

Hi,

I've made a Digital FIR Crossover on my MAC OSX for my SUB , WITH REPHASE, and FIR filters at 110, 2048 taps, 32 bits FLOAT / 44100.

My audio files are flac : 16/44100

When i analyse after this CROSSOVER, it's 100% linear phase and 33% bit pecfect , volume perfect:
https://deltaw.org/

if i upsample to 48 : no more bitperfect change in Volume.

And it's a problem : if i want to do the same crossover with DSPproducts (AES/SPDIF product), the plugins upsamples at 48/96 Khz.

i don't want to upsample 44100 to 48000, it's a bad operation.

so all mini-DSP products plug-ins, upsample 44100 to 48000 Khz : Bad

Is there a DSP hardware product that accept 2048 taps at 44100 for 6 Channels (HIGH - MID - low stereo) ? for a good brickwall linear Crossover, like my DIGITAL CROSSOVER with my MOTU and my Compouter that use convolution FIR.....

Where ?

And why mini-DSP don't use 44100 ? ( i know latency.....)

With 2048 taps at 44100 on my computer i've only 23ms latency for multiple crossover.

I want a hardware DSP at 44100

Thanks,

Olivier

My OSX FIR CROSSOVER use this :

for SUB :

For Main SPeaker :

 

[LeftCoastTim]

i don't want to upsample 44100 to 48000, it's a bad operation.

False.

Instead of obsessing about bit perfection, why not measure your room and fix your room modes? It will change the sound more than what the upsampling is doing.

 

[DownUnderGazza]

False.

Instead of obsessing about bit perfection, why not measure your room and fix your room modes? It will change the sound more than what the upsampling is doing.

What he said, and for the better!

 

[scott wurcer]

When i analyse after this CROSSOVER, it's 100% linear phase and 33% bit pecfect , volume perfect:
https://deltaw.org/

What does "bit perfect" mean in the context here. The concept of 33% bit perfect needs a definition.

 

[pos]

@ofrappier, most hardware DSP solutions will resample to their own internal sampling frequency, even if this is supposed to be the same as the input one.
That means that even if you feed the minidsp with native 48kHz signal it will still resample it to his 48kHz.
That is called an ASRC, and is one of the most reliable and predicable ways of dealing with this type of situation.

If you want to avoid that the only really practical solution is to play and filter within the same device, so you only have one clock.
A PC/Mac paired with a filtering software and a multichannel DAC would do the trick.

You might still have an ASRC process taking place within the DAC itself though

 

[ofrappier]

Thanks to all.

But ASRC change Volume RMS ?

If i compare a 44100 file at 0dBFS upsampled with audio driver SRC at 48000, it clips with volume CHANGE...

(https://deltaw.org/)

upsampling 48 clips audio files between -2dBFS/0dBFS and 96 even more....

I will stay with my OSX COMPUTER DSP at 44100/32 bits FLOAT, without ASRC, with MOTU, better than all MINI-DSP.

 

[pos]

upsampling can generate (reveal...) intersample clipping, but you were going to have them anyway in the DAC (the ESS in your motu might have some margin to handle them in that case though).

Better keep a margin when filtering anyway, especially when doing phase manipulations.

 

[ofrappier]

upsampling can generate (reveal...) intersample clipping, but you where going to have them anyway in the DAC (the ESS in your motu might have some margin to handle them in that case though).

Better keep a margin when filtering anyway, especially when doing phase manipulations.

Hi,

REPHASE is a great great freeware tool.

Thanks for explanations. Can you advice me about a good DIGITAL DSP (without internal ADC or DAC) with :

AES IN
and X3/AES out for HIGH / MID / SUB

With FIR PLUG-INS that accept each 2048 TAPS created with REPHASE with a .wav at 32 bits. (3X2)

Thanks

Olivier

 

[pos]

Thanks for your kind words

There are still not a lot of FIR-capable hardware crossovers unfortunately, not with a usable number of taps and a good flexibility at least.

The BSS blu 160 (and blu 8xx derivates for network audio protocols) is an interesting solution, with a lot of taps at hand at 48kHz (I guess it uses FFT convolution), impressive routing capabilities, and multiple in/out options (digital, analog, network).

The minidsp minisharc offering is also still a good option, and you can combine several of them using waveflex' boards.

 

[Fledermaus]

A Solaro DSP unit (Xilica) with only AES I/O cards might do the job too - to which you might add analog I/O, USB, GPIO cards as needed.
The last software version allows it to run 48 or 96 kHz and the FIR filters are 4096 taps max :



https://xilica.com/products/qr1/

 

[pkane]

What does "bit perfect" mean in the context here. The concept of 33% bit perfect needs a definition.

Bit perfect in DeltaWave is defined as the percent of all the samples in the two files that are a perfect match at the selected sample size. Obviously, at 100% the files match perfectly, all the samples are the same. 33% is a very good match, since even a 1/2 LSB random dither will reduce the bit match result to about 50%. DeltaWave also computes the number of bits the two files must be reduced to in order to get a 50% of all samples to be a bit-perfect match.

 

[scott wurcer]

33% is a very good match, since even a 1/2 LSB random dither will reduce the bit match result to about 50%. DeltaWave also computes the number of bits the two files must be reduced to in order to get a 50% of all samples to be a bit-perfect match.

Each type of noise shaped dither will have a different statistical distribution so the probability distribution of exact matches of any given sample in two files can probably be computed. Not sure this level of detail gives any useful insight.

 

[pkane]

Each type of noise shaped dither will have a different statistical distribution so the probability distribution of exact matches of any given sample in two files can probably be computed. Not sure this level of detail gives any useful insight.

Random (white) dither, including TPDF will result in a 50% bit perfect metric. Different shaped dither may produce different results.

The point of the metric is not to measure how similar or dissimilar the files are but to determine how molested the bits were in processing when a bit-perfect playback is desired. For better similarity measures, DeltaWave includes many other metrics that can be used, from amplitude and spectra differences, to correlated null depth to measurements of phase differences and jitter.