Go Back   Home > Forums > >
Home Forums Rules Articles diyAudio Store Blogs Gallery Wiki Register Donations FAQ Calendar Search Today's Posts Mark Forums Read

PC Based Computer music servers, crossovers, and equalization

Ideas wanted -- USB DAC product
Ideas wanted -- USB DAC product
Please consider donating to help us continue to serve you.

Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
Reply
 
Thread Tools Search this Thread
Old 10th December 2017, 09:57 PM   #31
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
Quote:
Originally Posted by MarcelvdG View Post
Features that could be useful when you use digital filtering:

1. Selectable filter characteristics:
1A. Apodizing filter for people worried about pre-ringing
1B. Steep filter with a stop band starting at fs/2 for people who like anti-imaging filters to really prevent imaging
1C. Compromise filter with low latency for professional and video applications

2. Headroom
Most filters with 0 dB passband gain for stationary sine waves have overshoots on signals that are not stationary or not sinusoidal, such as music. Many digital recordings go all the way up to 0 dBFS, so you need to keep some headroom yourself if you want to prevent clipping or overflowing.

Suggestion 2 goes at the expense of the SINAD and DR numbers you get when you measure with stationary sine waves, so please ignore suggestion 2 if you are really only after getting impressive measured performance without caring what it sounds like with music.
Thanks for your advice. I will include user selectable FIR filters (apodizing coefficients, steep coefficients and low latency coefficients). I may not do IIR since my design depends on polyphase identity of FIR to use DSP resource effectively.

I will include good headroom in my design. It has a 72-bit accumulator inside the DSP, and I will anyway scale signal down a little bit to prevent sigma delta modulator oscillation.

It won't affect SINAD and DR since my target SINAD and DR are 24-bit class, and all my DSP engines are at least 32-bit, so I have some wiggling room. Even if user inputs 32-bit, it will be dithered first.
  Reply With Quote
Old 10th December 2017, 10:07 PM   #32
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
Quote:
Originally Posted by MarcelvdG View Post
140 dB? Reed relay shorting the outputs when you detect an all zeros input signal?
The reason why everyone rates DR instead of SNR is because this number can be cleverly manipulated.

For instance, DR doesn't care about signal-related noise (multiplicative noise), it only cares about additive noise.

In my case, my OPAMP output stage generates <=0.5uV RMS noise, and my device can output 10Vpeak (7V RMS) single ended, and 14V RMS differential.

Now, by doing differential output, output noise gets boosted by sqrt(2), so it becomes 0.7uV RMS, so I get 20M:1 DR, which is 146dB.

As for SNR, my design has digital SNR more than 150dB in audible range, mainly contributed by aliasing folded back into audible range.

My clock system guarantees at least 150dB SNR at 10kHz, and unlike most other people's design, which uses a low jitter XO, I use a little different design which is immune to mechanical shock (TCXO feeding into PLL1, PLL1 uses a MEMS VCXO as VCO, then PLL1 feeds to LC VCO PLL2, PLL1 filter designed to suppress mechanical vibration frequency), so acoustic resonance on chassis won't degrade jitter performance.

My analog subsystem is yet to be characterized, so I don't know what's the overall SNR, but I would say at least 135dB.
  Reply With Quote
Old 10th December 2017, 10:14 PM   #33
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
Quote:
Originally Posted by pcan View Post
It would also be useful to support USB-C charging mode on the USB port, when using the Android smartphone as FLAC source with a USB-C cable connection. On my current setup with a conventional USB DAC, I am forced to use a powered USB-C hub to avoid the battery drain and to get a standard USB2 connector, and it is unpractical.
Believe or not, I initially designed for up to 60W USBPD charging (device powered by a 90W USBPD charger, up to 60W can be diverted to load, my device provided DC/DC conversion and USB PD CC communication to both DFP and UFP).

It ended up being scrapped due to the extra complexity, and anyway such features should not be provided to a prosumer grade gear. Also, it imposes some global EMI and safety certification issues. I decided to just remove USBPD feature and replace its input power connector to a plain old barrel jack.

My device is a self powered device, with type C interface. It only draws less than 0.2mA from USB port for VBUS detection.
  Reply With Quote
Old 10th December 2017, 10:55 PM   #34
MarcelvdG is offline MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
Quote:
Originally Posted by blueskull View Post
I will include good headroom in my design. It has a 72-bit accumulator inside the DSP, and I will anyway scale signal down a little bit to prevent sigma delta modulator oscillation.

It won't affect SINAD and DR since my target SINAD and DR are 24-bit class, and all my DSP engines are at least 32-bit, so I have some wiggling room. Even if user inputs 32-bit, it will be dithered first.
I don't understand that. Of course you can make the SINAD of the digital part as high as you like just as long as it fits in the DSP or FPGA, but keeping headroom also means that the mixed-signal and analogue parts have to be driven a little softer than when you had no headroom, which reduces the SNR.

Anyway, good that you plan to include a decent headroom. It is far more likely to give an audible advantage than a few dB of difference in an already quite soft noise floor.
  Reply With Quote
Old 11th December 2017, 06:01 AM   #35
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
Quote:
Originally Posted by MarcelvdG View Post
... and analogue parts have to be driven a little softer than when you had no headroom, which reduces the SNR.
Of course! Analog part is designed to output +-10V peak from a +-12.5V rail, leaving 2.5V headroom, which is close to maximum "minimum distortion point" with a safe margin according to OPA1612 datasheet.
  Reply With Quote
Old 11th December 2017, 10:45 PM   #36
MarcelvdG is offline MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
So with 0 dBFS you aim for a level where the SINAD of the op-amps is maximal, with intersample overshoots you go a bit beyond that point, but still low enough not to cause hard clipping. That sounds very reasonable.
  Reply With Quote
Old 12th December 2017, 12:00 AM   #37
Markw4 is offline Markw4  United States
diyAudio Member
 
Join Date: Jul 2016
Location: California
Intersample overs occur in the reconstruction filter inside the DAC. For that reason Benchmark Media reduces the digital input to the DACs by 3.5dB. That's enough to prevent distortion inside the DAC.
  Reply With Quote
Old 12th December 2017, 03:36 AM   #38
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
Quote:
Originally Posted by Markw4 View Post
Intersample overs occur in the reconstruction filter inside the DAC. For that reason Benchmark Media reduces the digital input to the DACs by 3.5dB. That's enough to prevent distortion inside the DAC.
My analog reconstruction filter has a pretty low Q thanks to huge digital interpolation ratio (isn't that why we do oversampling?). As a result, transfer function never exceeds its DC gain.

The SDM, however, does tend to crap out when input level is too high, so I gave it quite some margin in digital domain.

When designing an SDM DAC from scratch, there's another form of challenge, namely inter-symbolic interference, which basically means the current symbol's output level is dependent on the previous symbol, which causes auto-correlation, hence multiplicative noise, which results in THD degradation. For this reason, our SDM DAC has a blank pulse insertion mechanism to isolate output DAC bitstream symbols from each other. This decreases DC gain by ~1.9dB. IMHO, greatly reducing THD by a lot at the cost of 1.9dB DR loss is a good deal.
  Reply With Quote
Old 12th December 2017, 05:33 AM   #39
MarcelvdG is offline MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
What I understand from the articles Mark posted is that it can either be the digital interpolation filter or the analogue reconstruction filter that clips. Usually it will be the digital filter, as these are often designed without any headroom. Analogue filters will typically have some headroom to cover spreads and the increase of distortion that you get close to clipping.

By the way, the absolute worst-case input signal is full scale times the sign function of the impulse response reversed in time, that is, a signal that switches between positive and negative full scale in such a way that the absolute values of the filter coefficients all get added to each other. With a waveform like that you can easily need 11 dB of headroom. Fortunately, it is extremely unlikely to find something like that in a normal music recording.
  Reply With Quote
Old 12th December 2017, 05:47 AM   #40
blueskull is offline blueskull  China
diyAudio Member
 
Join Date: Jun 2013
Location: RTP Metropolitan, NC, USA
You are absolutely correct. Analog part don't usually clip due to the low Q of filter, but digital parts, both interpolation filter and SDM, tend to clip easily.

The worst thing is that FIR filters don't have a feedback, so even for the worst case (but still within audible frequency) I can guarantee no clipping happens, but since SDM is essentially an IIR with a twist, if it decides to oscillate and clip, nothing can stop it but careful design with saturation prevention and damping.

This is another reason why some ridiculously high specced DACs use very high interpolation rate -- ES9018 goes to 1.536Msps after OSF, and though I have no data, I think internally in its Hyperstream SDM core, it uses higher sample rate.

With higher sample rate, SDM filters don't have to be so aggressive at shaping noise from lower frequency to higher frequency -- because its input noise frequency is already very high. That reduces the risk of having it going haywire.

Also, soft clipping and active damping are used in my modulator to minimize the impact of clipping should it occur.

And of course, my digital part has more than enough headroom, this is also to prevent SDM from going crazy. When it clips, it tends to recover really slowly, so it's best not to let it clip in the first place.
  Reply With Quote

Reply


Ideas wanted -- USB DAC productHide this!Advertise here!
Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
3-way design ideas wanted! karha096 Multi-Way 34 10th October 2014 04:56 PM
Ideas wanted. Speakers in fire surround. jimbo1968 Full Range 1 11th November 2013 07:38 PM
Squeezebox help and ideas wanted zane Digital Source 8 22nd July 2013 07:21 AM
Input wanted on Speaker build ideas Moonfly Multi-Way 11 19th January 2013 01:45 PM
Compact sub ideas wanted. type Subwoofers 2 12th June 2007 09:57 AM


New To Site? Need Help?

All times are GMT. The time now is 07:33 AM.


Search Engine Optimisation provided by DragonByte SEO (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Resources saved on this page: MySQL 14.29%
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Copyright ©1999-2018 diyAudio
Wiki