• These commercial threads are for private transactions. diyAudio.com provides these forums for the convenience of our members, but makes no warranty nor assumes any responsibility. We do not vet any members, use of this facility is at your own risk. Customers can post any issues in those threads as long as it is done in a civil manner. All diyAudio rules about conduct apply and will be enforced.

Hi-end DSP based multi-channel integrated Preamp/Crossover/DAC project

Jump to Latest
D

DSP_Geek

Member
Joined 2003
DSP_Geek said:
That is... sub-optimal.

Off to faff about with hideously expensive HDMI to SPDIF extractors, then, and weed through the morass of 2-channel analog out plus TOSLink devices clogging the search results. Feh.
Click to expand...

The only thing which comes close is an HDMI to 4 AES/EBU extractor -- only $2400! If you can find the bloody thing, that is. Yay.

HDMI-2A - ARVUS

Everything else has TosLink for multichannel -- compressed of course -- and RCA analog outs for downmixed two channel.

If anyone else has a better idea I'm all ears.
 
K

Kal Rubinson

Member
Joined 2003
DSP_Geek said:
The only thing which comes close is an HDMI to 4 AES/EBU extractor -- only $2400! If you can find the bloody thing, that is. Yay.

HDMI-2A - ARVUS

Everything else has TosLink for multichannel -- compressed of course -- and RCA analog outs for downmixed two channel.

If anyone else has a better idea I'm all ears.
Click to expand...

I do not follow this thread but, if it is relevant, there is a nice 8 channel USB-AES3 IO from miniDSP, the U-DIO8.
 
D

DSP_Geek

Member
Joined 2003
Kal Rubinson said:
I do not follow this thread but, if it is relevant, there is a nice 8 channel USB-AES3 IO from miniDSP, the U-DIO8.
Click to expand...

I could cobble together an HDMI to 8 AES output kluge from various silicon vendors' development boards, but every one of them cautions that'd work only for non-HDCP streams. Near as I can figure USB isn't crippled with copy protection / user harassing -- yet -- so USB interfaces are a relatively clean design.

In other news Arvus tells me the HDMI-2A does work with HDCP streams, and it's $1750 plus shipping from New Zealand
 
N

NATDBERG

Member
Joined 2006
DSP_Geek said:
That is... sub-optimal.
Click to expand...

I think what you mean is

" That doesn't match my requirements "

Consider that it has no HMDI or video related input at all.. you do seem to be hung up on the fact that its I2S input uses the same connector as that used in video connections... however it is an I2S input for I2S. A similar situation would be if someone saw lots of XLR sockets on the rear of their analogue pre-amp and was complaining that none of them were AES/EBU digital inputs.
 
Last edited:
D

DSP_Geek

Member
Joined 2003
NATDBERG said:
I think what you mean is

" That doesn't match my requirements "

Consider that it has no HMDI or video related input at all.. you do seem to be hung up on the fact that its I2S input uses the same connector as that used in video connections... however it is an I2S input for I2S. A similar situation would be if someone saw lots of XLR sockets on the rear of their analogue pre-amp and was complaining that none of them were AES/EBU digital inputs.
Click to expand...

I mean what I said. With all due respect, the unit is 95% of the way to filling my requirements, so if the HDMI connector doesn't extract audio streams from video then the specs are suboptimal for my use, not a show-stopper since I've already found an alternate solution.

In any event, T Bass looks to have something rather nice. Ta!
 
D

DSP_Geek

Member
Joined 2003
orangeart said:
Many thanks for the enthusiasm. I'm glad you like it.
Click to expand...

It's gotten better, even though my engineering skepticism is yelling burn-in isn't real. Perhaps I'm hearing stuff not obvious from the git-go since my ears weren't used to the unit.

Anyway, the system's grown a newfound sense of s-c-a-l-e. Cool Struttin' by Sonny Clark lives quite comfortably in the room, as do Joni Mitchell, Joe Jackson, Bittersweet, and lots of friends.

Now remember the speakers have one RS-150 and one 27TDFC each. A demo to a fellow working on the house provoked the question "Where's the woofer?" "You're looking at it" He was gobsmacked, a testament to nCore's ability at plumbing the depths with aplomb.

Once again, well done. Even the chassis is brilliant, with the front panel and power button setting the right understated tone.
 
O

orangeart

Member
Joined 2010
kjfaudio.co.uk
DSP_Geek said:
It's gotten better, even though my engineering skepticism is yelling burn-in isn't real. Perhaps I'm hearing stuff not obvious from the git-go since my ears weren't used to the unit.

Anyway, the system's grown a newfound sense of s-c-a-l-e. Cool Struttin' by Sonny Clark lives quite comfortably in the room, as do Joni Mitchell, Joe Jackson, Bittersweet, and lots of friends.

Now remember the speakers have one RS-150 and one 27TDFC each. A demo to a fellow working on the house provoked the question "Where's the woofer?" "You're looking at it" He was gobsmacked, a testament to nCore's ability at plumbing the depths with aplomb.

Once again, well done. Even the chassis is brilliant, with the front panel and power button setting the right understated tone.
Click to expand...

Thanks for the extra enthusiasm :)

I've got these amps up on my website now for those who are interested. I'd rather not start a discussion about the amps on someone else's thread though. If anyone is interested I had a thread here or I can be PM'd

Thank you should also go to David for bringing the attention of you guys, his potential customers, to my amp. Thanks.
 
D

DSP_Geek

Member
Joined 2003
Code:
The DSP processing is specified at 192 kHz. Is that hard & fast or could I run at 96 or 48 kHz? I suppose a sample rate converter block for lows and mids would be reaching for the moon...

Here's the deal: FIR filters need the square of DSP clocks as the processing rate goes up. Say for example you have a 100 point FIR at 48 kHz; it must be 200 points to keep the same frequency response at 96 kHz as at 48. However, you're executing the longer filter twice as many times, so you need 2 x 2 = 4 times the clocks to run it at double the sample rate. 192 kHz, well, the filter is 4 x as long and runs 4x as often, so it wants 16 times the clocks. Suddenly a middling frequency resolution 100 tap FIR looks like it'll chew up a fair bit of DSP power, and never low to mid crossover FIRs wanting 960 taps at 48 kHz! You go from 46 million multiply-accumulates (MMACs) at 48 K to 737 MMACs at 192 kHz. For one filter. Each channel takes one for the woofer, one for the midrange, two channels: now we're talking 2949 MMACs, or about three billion operations a second. That'll warm up the room some.

There's another gotcha: the poles of IIR filters approach -1 as their corner frequency decreases. For example, a 100 Hz Butterworth lowpass filter sampled at 48 kHz has poles at 0.9907442546 + j 0.0091708587, 0.9907442546 - j 0.0091708587, leading to a recurrence relation of
Code: 
y[n] = (  1 * x[n- 2])
     + (  2 * x[n- 1])
     + (  1 * x[n- 0])

     + ( -0.9816582826 * y[n- 2])
     + (  1.9814885091 * y[n- 1])
which features a gain of 2.356080688e+04 (about 14.5 bits) around the denominator feedback loop. Given a 32 bit floating point mantissa is 24 bits, scaling the input down by that much before pushing it through the filter leaves about 9.5 bits of resolution, or about 57 dB signal to noise ratio (SNR) through this filter. Arguably you can get away with that filter -- maybe -- but the situation gets quadratically worse as the sample rate increases.

This Ultimate Preamp 2 says math can happen in 40 bit floats (helpful) or 80 bit fixed point (really useful), which would alleviate the IIR problem. Can we specify which mode is used by each filter, or do all filters use the same operations?
 
Last edited:
Tranquility Bass

Tranquility Bass

Member
Joined 2013
analog-precision.com/
DSP_Geek said:
Code:
The DSP processing is specified at 192 kHz. Is that hard & fast or could I run at 96 or 48 kHz? I suppose a sample rate converter block for lows and mids would be reaching for the moon...

Here's the deal: FIR filters need the square of DSP clocks as the processing rate goes up. Say for example you have a 100 point FIR at 48 kHz; it must be 200 points to keep the same frequency response at 96 kHz as at 48. However, you're executing the longer filter twice as many times, so you need 2 x 2 = 4 times the clocks to run it at double the sample rate. 192 kHz, well, the filter is 4 x as long and runs 4x as often, so it wants 16 times the clocks. Suddenly a middling frequency resolution 100 tap FIR looks like it'll chew up a fair bit of DSP power, and never low to mid crossover FIRs wanting 960 taps at 48 kHz! You go from 46 million multiply-accumulates (MMACs) at 48 K to 737 MMACs at 192 kHz. For one filter. Each channel takes one for the woofer, one for the midrange, two channels: now we're talking 2949 MMACs, or about three billion operations a second. That'll warm up the room some.

There's another gotcha: the poles of IIR filters approach -1 as their corner frequency decreases. For example, a 100 Hz Butterworth lowpass filter sampled at 48 kHz has poles at 0.9907442546 + j 0.0091708587, 0.9907442546 - j 0.0091708587, leading to a recurrence relation of
Code: 
y[n] = (  1 * x[n- 2])
     + (  2 * x[n- 1])
     + (  1 * x[n- 0])

     + ( -0.9816582826 * y[n- 2])
     + (  1.9814885091 * y[n- 1])
which features a gain of 2.356080688e+04 (about 14.5 bits) around the denominator feedback loop. Given a 32 bit floating point mantissa is 24 bits, scaling the input down by that much before pushing it through the filter leaves about 9.5 bits of resolution, or about 57 dB signal to noise ratio (SNR) through this filter. Arguably you can get away with that filter -- maybe -- but the situation gets quadratically worse as the sample rate increases.

This Ultimate Preamp 2 says math can happen in 40 bit floats (helpful) or 80 bit fixed point (really useful), which would alleviate the IIR problem. Can we specify which mode is used by each filter, or do all filters use the same operations?
Click to expand...

There are dipswitch settings on the board to run the native sample rate at 48KHz or 96KHz but of course you need to change your filter coefficients to match. You can also use the down convert and up-convert modules in Audioweaver to reduce the sampling rate for better use of the DSP resources.

Floating point numbers are handled internally with 40 bit precision and fixed point at 80 bits. There are fixed point and floating point filters in the Audioweaver library so you get to choose what best suits your application ;)

cheers
david
 
D

DSP_Geek

Member
Joined 2003
Tranquility Bass said:
There are dipswitch settings on the board to run the native sample rate at 48KHz or 96KHz but of course you need to change your filter coefficients to match. You can also use the down convert and up-convert modules in Audioweaver to reduce the sampling rate for better use of the DSP resources.

Floating point numbers are handled internally with 40 bit precision and fixed point at 80 bits. There are fixed point and floating point filters in the Audioweaver library so you get to choose what best suits your application ;)

cheers
david
Click to expand...

I imagine 80 bit math takes more time than 40 bits, but anyone doing FIRs in 80 bits wants a gentle talking-to.

<look outside> Funny, there's no snow on the ground but it sure feels like Christmas. I couldn't ask for any more. Well, maybe for delay specs on the up & down converters :tongue: Happy dance!
 
You must log in or register to reply here.
Top Bottom