DSP Xover project (part 2)

Hi all,
Got my board yesterday and connected it up with LCD, LED's and buttons and it works great.
It would be nice with a possibility to load the biquad coefficients from e.g. "ACD".
I tried loading a FIR filter that I made in "Acourate". I had to generate a 512 tap filter (1024 is one tap too much) and save it as TDS *.dat file and it loads fine, but the low freq resolution is not good enough at 512 taps:
see attachment...65535vs512 taps
How many taps is possible for each channel?

I see that HOLM HOLM Acoustics is using something they call "16x decimation" that equals 16368 taps @ 44khz.

Anyway, this seems very promising for the first release. Thanks a lot Nick, you have done a great job!

best,
Paal

Hi Paal,

Great to hear from you and congratulations for your build. You guys are brilliant, yourself, Shaun and JCGA.

OK let's stop with compliments and get to your point.

The documentation is on the way, but basically you can load up to 1023 coefficients on Ch 1 and Ch5 and up to 511 coefs on the remaining channels.

Inputs take all 1023 coefs except Diff only 511.

The reason why it's 1023 instead of 1024 (or similarly 511 instead of 512) is connected to the way circular buffers are handled on a DSP56300 core.

Note that this is likely to change with future releases when I get feedback of what sort of filters users want to implement.

Regarding biquad coefs, it would be helpful if you could send me a ACD file so that I can make Najda compatible.

Is the Nadja sensitive in which order power comes up, and what restriction on rise time do we have to consider?

Otherwise I would prefer to apply power to digital section and ADACS first, and then when this is out of "boot" then apply to the analogue output section to minimize any experiences with digital noise output on speakers...

Still risks of thumping?

I'd recommend that you supply all power at once to Najda. There's an isolated relay command onboard for the issue you're mentioning, please see if that could fit the bill.
 
The documentation is on the way, but basically you can load up to 1023 coefficients on Ch 1 and Ch5 and up to 511 coefs on the remaining channels.

Inputs take all 1023 coefs except Diff only 511.

The reason why it's 1023 instead of 1024 (or similarly 511 instead of 512) is connected to the way circular buffers are handled on a DSP56300 core.

Note that this is likely to change with future releases when I get feedback of what sort of filters users want to implement.

It seems that some of the advanced DSP schemes like Bodzio's Ultimate Equaliser (UE)* uses something of the region of 16K to 32K coefficients. That is more than an order of magnitude greater than what is currently available (low frequency resolution is 46.92Hz for 48kHz Fs and 1023 coefficients). How big do you think you can go with this platform?

*UE is an integrated PC-based DSP crossover/digital room correction application that is able to deliver flat frequency and phase response I am hopeful that, once a capable stand-alone DSP platform becomes available, that these will also be supported by the likes of UE.
UE Open baffle
Bodzio Software
 
Hi Shaun

Thanks for the links

The HW recommended for Bodzio UE is
New Intel Core i7-960 3.2 GHz Quad-Core (BX80601960) SLBEU, Factory Sealed is selling on ebay for 450 $, CPU only ... about 20 times more powerful than the choosen DSP. Plus you have to buy a multi-channel PCIe sound card which is worth $$$ plus the noise and burden of the PC plus give your money to microsoft :D

I guess solutions for the Najda may be SW only to get the most of the board in the low range in focusing resources ?

BR
 
Last edited:
As Said psu with good values and that low min load few and far between here.
To get me started I Will go with a good One and
put 0.4 or 0.5A dummy load on the 5V, 0.2A on the 12+. 12- had no load demand according to the spec sheet.
I'll get my Scope back and give it a measure;)

Hi,

Why reasoning with SMPS with such low consumption (less than 5 W on idle and less than 1.5W in standby). I will probably go myself with a linear power supply using LDO regulators like the LT1963 and LT3015
JC
 
I see that HOLM HOLM Acoustics is using something they call "16x decimation" that equals 16368 taps @ 44khz.

I overlooked this yesterday and it's actually something interesting (and it's pretty honest of HOLM Acoustics to publish such detailed specs).

So what they do is that they are down-sampling the signal priory to applying the filter. The down-sampling ratio is 16, this is why they say that 1023 is equivalent to 16*1023 = 16368.

That could be one way to go. Thanks for posting the info Paal.
 
I overlooked this yesterday and it's actually something interesting (and it's pretty honest of HOLM Acoustics to publish such detailed specs).

So what they do is that they are down-sampling the signal priory to applying the filter. The down-sampling ratio is 16, this is why they say that 1023 is equivalent to 16*1023 = 16368.

That could be one way to go. Thanks for posting the info Paal.

Hi Nicolas

yes, decimation is a clever trick used also by some loudspeaker management units to improve resolution in the lows. It would be very very nice if you could implement it in Najda :)

JC
 
Cos Nick and Paal said so:)
A very quick search for linear PSU's throws up even less options for me - no all one options came up - Am I looking the wrong place? - Farnell, RS, Ebay...

My switching PSU arrived and it's tiny... I'll dummy load it and measure it + current in.

I fixed the Najda board down to a bread broad yesterday evening and got prepared for the PSU. Got a suitable metal box for the PSU. This evening I will wire up the push-to-make key pad.

Hi,

Why reasoning with SMPS with such low consumption (less than 5 W on idle and less than 1.5W in standby). I will probably go myself with a linear power supply using LDO regulators like the LT1963 and LT3015
JC
 
It seems that some of the advanced DSP schemes like Bodzio's Ultimate Equaliser (UE)* uses something of the region of 16K to 32K coefficients. That is more than an order of magnitude greater than what is currently available (low frequency resolution is 46.92Hz for 48kHz Fs and 1023 coefficients). How big do you think you can go with this platform?

So with the current direct convolution strategy, we're at the max, which is around 9000 coefs at 48kHz shared between all filters.
We can expand this by going FFT. How many coefs exactly I don't know, I would need to look at the details but don't expect 16k coefs per channel.
Fact is we're going to hit memory limits: ram memory for computing the filter and even eeprom for storing coefficients.

I'm surprised with the number of coefs that you are throwing. I'm convinced that 16k coefs at 48kHz sampling rate is way overkill and that there are more elegant solutions to achieve similar result with shorter filters. We'll find out how.

Cos Nick and Paal said so:)
Snitching is not nice ;)
My switching PSU arrived and it's tiny... I'll dummy load it and measure it + current in.
Yes. Check what are the output voltages with no load.
We don't really know what they mean when they say 'min. load'. I guess the PSU is not going to blow, what do you think? :)
Maybe it's just that the other specs are just outside their limits in the datasheet, but still OK for normal use. (maybe not)
 
Cos Nick and Paal said so:)
A very quick search for linear PSU's throws up even less options for me - no all one options came up - Am I looking the wrong place? - Farnell, RS, Ebay...

Sorry, I was not clear enough. I will build my own power supply using probably those LDO regulators.

Also I received today from Mouser some nice That 1200 and That 1606 drivers that I will use later to expand the Najda board with symetrical input-outputs

JC
 
Hi Shaun

Thanks for the links

The HW recommended for Bodzio UE is
New Intel Core i7-960 3.2 GHz Quad-Core (BX80601960) SLBEU, Factory Sealed is selling on ebay for 450 $, CPU only ... about 20 times more powerful than the choosen DSP. Plus you have to buy a multi-channel PCIe sound card which is worth $$$ plus the noise and burden of the PC plus give your money to microsoft :D

I guess solutions for the Najda may be SW only to get the most of the board in the low range in focusing resources ?

BR


Hi Jlop,

Thank you for your comments. I have the following response.


Regarding PC noise.

Contemporary PCs have excellent SNR. I have measured and compiled the results in http://www.bodziosoftware.com.au/Computers_SNR.pdf paper. You will find, that PCs are equal or better than some of the CD players.


Regarding PC burden.

PC is actually used as an audio server and the same PC runs UE4. You can even run “full digital” solution described in http://www.bodziosoftware.com.au/Ultimate Equalizer V4 in SPDIF.pdf paper. This is really uneatable combination for convenience and the cost of music.

The PC solution is simply unmatched for user friendliness, audio server function, and multi-channel HT application. You may have noticed, that iTunes has been offering ".m4a" and ".m4p files now, instead of ".mp3" files. The ".m4a" files are much less compressed, thus offering better sound quality.

I stopped buying CDs about a year ago. I buy my music exclusively as single files of very selectively chosen songs, and this necessitates a PC for a music server. But this approach also offers huge cost savings in purchasing the music - this is the way to do it. The same PC runs the UE4 - so it all matches for me to perfection.


Regarding sound cards.

In it’s basic configuration, you do not need any sound cards – you can use 2in/8out motherboard audio codec, such as ALC889 or ALC898. However, sound cards (delta1010LT for $220) expand capabilities of the UE. With two sound cards, you can have 16in/16out signal routing capabilities – example in http://www.bodziosoftware.com.au/Signal_Routing.JPG paper.

So, you can have 8 stereo sources, or 2 HT +2stereo sources connected to the UE4 system, and freely switch between inputs with single keyboard button press.



Best Regards,
Bohdan
 
Hi Bohdan

Thank you for your fast answer and very interesting attached documents, especially about PC electrical noise. These infomation is unfortunately not often available to audio PC builders. I was in fact speaking about acoustic noise from fans, which is difficult and costly to get rid of for powerful platforms.Same PC may not be used as server for many reasons including noise of HDD, backup disks, screens, etc

Let me stress that I did not intend to dismiss your software solution, which seems very comprehensive to me as far as I understand.

I just wanted to stress that a on a one board DSP you (may) have to restrain your ambitions compared to a powerfull PC based solution. Both have a public of course, depending on the requirements.

There are quite nice soundcard, but, as far as I know and I am definetly not a great specialist, their SQ cannot compete with some very good DACs, especially for output stages, so I have a feeling that going digital out of the PC and have DA conversion in a specific HW might be better. Just a thought.

Please feel free to comment !

BR
jean-Louis
 
thats a myth, there are some truly superb soundcards. apogee, the metric Halo LIO-8 and ULN-8 (both have external DC supply inputs and a pretty impressive DSP onboard) Lynx make some excellent hardware too. much more expensive than this board here, but both DSP and analogue are considerably more powerful.

its possible to construct a dedicated multichannel dac that outperforms them, but it would require considerable effort and funds and their drivers are rock solid. External wordclock high quality preamps.

Hi Bohdan

Thank you for your fast answer and very interesting attached documents, especially about PC electrical noise. These infomation is unfortunately not often available to audio PC builders. I was in fact speaking about acoustic noise from fans, which is difficult and costly to get rid of for powerful platforms.Same PC may not be used as server for many reasons including noise of HDD, backup disks, screens, etc

Let me stress that I did not intend to dismiss your software solution, which seems very comprehensive to me as far as I understand.

I just wanted to stress that a on a one board DSP you (may) have to restrain your ambitions compared to a powerfull PC based solution. Both have a public of course, depending on the requirements.

There are quite nice soundcard, but, as far as I know and I am definetly not a great specialist, their SQ cannot compete with some very good DACs, especially for output stages, so I have a feeling that going digital out of the PC and have DA conversion in a specific HW might be better. Just a thought.

Please feel free to comment !

BR
jean-Louis
 
Shaun has reported a bug that prevents from loading IIR files into the processor. This happens with files that involve lots of filters.
I have identified the issue and I have a fix for that. We'll do some testing then I'll post the new release asap.
Sorry for the inconvenience.
Thanks Shaun for reporting the bug!