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

Solid State Talk all about solid state amplification.

How low can you go?
How low can you go?
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 9th February 2018, 08:58 PM   #51
MarcelvdG is online now MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
I forgot the main imperfection of practical interpolation filters: they need a transition band, for example from 0.45 fs to 0.5 fs. That makes the reconstruction imperfect and can cause overshoots, quite independent of the intersample overshoot issue due to peak normalization. Still, it has nothing to do with distortion in the sense of nonlinear distortion.
  Reply With Quote
Old 10th February 2018, 07:54 AM   #52
mchambin is offline mchambin  France
diyAudio Member
 
Join Date: Mar 2011
How can one expect a fine 20KHz sine signal at a 44 KHz sampling rate.
You expect a miracle from some interpolation technique to reconstruct a sine signal from two samples.
It seems there is a FORTRAN code to do that.
I am amazed to see it has not been recoded in C code since that 50 years old technology.
__________________
Transistor junction temperature is not transistor case temperature.
  Reply With Quote
Old 10th February 2018, 12:42 PM   #53
matze is offline matze  Europe
diyAudio Member
 
Join Date: Sep 2012
Quote:
Originally Posted by mchambin View Post
How can one expect a fine 20KHz sine signal at a 44 KHz sampling rate.
You expect a miracle from some interpolation technique to reconstruct a sine signal from two samples.
It seems there is a FORTRAN code to do that.
I am amazed to see it has not been recoded in C code since that 50 years old technology.
What abaut re-reading a good EE book? Distortion stems from quantification.
  Reply With Quote
Old 10th February 2018, 02:38 PM   #54
mchambin is offline mchambin  France
diyAudio Member
 
Join Date: Mar 2011
matze, how does your answer relate to my post ? Does it mean, I know nothing and should read books ?
__________________
Transistor junction temperature is not transistor case temperature.
  Reply With Quote
Old 10th February 2018, 02:42 PM   #55
MarcelvdG is online now MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
Quote:
Originally Posted by mchambin View Post
How can one expect a fine 20KHz sine signal at a 44 KHz sampling rate.
You expect a miracle from some interpolation technique to reconstruct a sine signal from two samples.
It seems there is a FORTRAN code to do that.
I am amazed to see it has not been recoded in C code since that 50 years old technology.
It's not a miracle, just a good low-pass filter. Sampling at 44.1 kHz is equivalent to multiplying the time signal with an impulse series with 44100 Hz repetition rate. The spectrum of such an impulse series consists of all multiples of 44.1 kHz. This means that the original signal gets mixed (in the RF sense of the word) around all multiples of 44.1 kHz, so 20 kHz becomes k*44.1 kHz +/- 20 kHz with integer k. The first image is at 44.1 kHz - 20 kHz = 24.1 kHz, so you need a filter that passes 20 kHz and very much attenuates 24.1 kHz. In fact I like it to pass 20 kHz and very much attenuate anything above 22.05 kHz, so aliases of signals just above 20 kHz are also suppressed. No big deal for a digital FIR filter.

The Parks-McClellan algorithm is quite famous, so I'm sure it has been recoded in any computer language you can think of. As I'm very interested in historical electronics, I prefer to use the original FORTRAN code (also because I could get it for free by just looking up the article in a library and typing in the code). It's probably more readable than C anyway, despite the spaghetti-style programming.

Last edited by MarcelvdG; 10th February 2018 at 02:46 PM.
  Reply With Quote
Old 10th February 2018, 03:06 PM   #56
MrMagic is offline MrMagic  Greece
diyAudio Member
 
MrMagic's Avatar
 
Join Date: Nov 2017
Location: Athens
Quote:
Originally Posted by MarcelvdG View Post
According to Benchmark Media Systems ( Intersample Overs in CD Recordings - Benchmark Media Systems, Inc. ), the main cause of overshoot is the use of peak normalization on a sampled waveform during mastering. They explain how that can amplify the waveform such that the peaks in between the samples exceed full scale.
Interesting to take into account when mastering -as I'm into this.

Quote:
Originally Posted by MarcelvdG View Post
I forgot the main imperfection of practical interpolation filters: they need a transition band, for example from 0.45 fs to 0.5 fs. That makes the reconstruction imperfect and can cause overshoots, quite independent of the intersample overshoot issue due to peak normalization. Still, it has nothing to do with distortion in the sense of nonlinear distortion.
If we can't have a perfect filter, we can't have a "perfect" reconstruction, which is one reason why the use of high sample rates is so more simple and straightforward, and far less dependent on unknown compromising parameters and complexities, and most likely, less costly too, both by eliminating the cost of filters, and because the storage media prices have dropped to ridiculous levels nowadays. That is, when we do have an option.

BTW, we are so many decades -36years(!) ahead of CD 44.1Khz it's rather embarrassing that we still use the same format! Think about that.

The physical media should have been replaced by high-resolution ones, decades ago. The only reason they are still produced, is because the technology we adopt (and research too) depends on greed-based profit and big-companies' domination, (profit-driven instead of benefit driven): if it costs a penny less to produce a compact disk, they will keep producing them instead of high-res ones, until human kind populates the galaxy by instant Star Trek-like transportation.

Quote:
Originally Posted by MarcelvdG View Post
the option to store the impulse response on punched cards is not supported anymore.
That was extremely funny, as you wrote it on a serious tone:
"sorry, computer running on coal is not supported anymore"

(I'm a little biased, as I know from a personal research that a computer is possible to be made and run literally on anything that carries any form of energy. Electrons and semiconductors is just one way).


Quote:
Originally Posted by MarcelvdG View Post
If you are interested in the subject, PM me and I will send you a list of articles related to interpolation filter design.
judging from the excellent article you suggested, I'm sure your list will be interesting as well, you've got a PM.

I'm not sure though, whether I'll end up using an FPGA, or a DSP or even a DSP microcontroller (for more flexibility/cost reduction), provided of course I won't find a quality DAC that does not clip at a reasonable cost.
  Reply With Quote
Old 10th February 2018, 04:45 PM   #57
matze is offline matze  Europe
diyAudio Member
 
Join Date: Sep 2012
Quote:
Originally Posted by mchambin View Post
matze, how does your answer relate to my post ? Does it mean, I know nothing and should read books ?
It is always a good idea to read books. Do not speculate how much you know; from information on this site I would conclude: a lot.

The question of signal reconstruction after discretisation in time is discussed in detail in many EE and more general books. As pointed out earlier in this thread, Nyquist did seminal work in this area, and results now are part of common canon in information technology.

Matthias
  Reply With Quote
Old 11th February 2018, 06:18 AM   #58
MarcelvdG is online now MarcelvdG  Netherlands
diyAudio Member
 
Join Date: Mar 2003
Location: Haarlem, the Netherlands
Quote:
Originally Posted by MrMagic View Post
I'm not sure though, whether I'll end up using an FPGA, or a DSP or even a DSP microcontroller (for more flexibility/cost reduction), provided of course I won't find a quality DAC that does not clip at a reasonable cost.
You could buy a Benchmark Media DAC for 2200 euros including VAT (at least that's what they cost down here), or build one of my DAC designs (cheaper, but definitely not cheap, and with lower dynamic range numbers), or take any I2S PCM DAC and attenuate the I2S signal before it goes into the interpolation filter. I still have to work out the details, but I think that can be done with a plain old 74HC158 and a couple of 74HC74's. It won't be perfect because it rounds the signal without any dither, but that's better than clipping.

See the attachment for the basic idea. It boils down to playing the MSB twice and delaying everything else by one bit clock period. This gives you a 1 bit (6.02 dB) headroom at the expense of 1 bit of signal level. I thought that by redrawing the logic a bit, the EXOR, inverter and multiplexer could all be made with a 74HC158, but they can't because the 74HC158 multiplexers have a common select input. I've got to think about it a bit more.
Attached Images
File Type: jpg I2Sattenuator.jpg (243.1 KB, 55 views)

Last edited by MarcelvdG; 11th February 2018 at 06:30 AM. Reason: Didn't remember common select input 74HC158
  Reply With Quote
Old 11th February 2018, 12:13 PM   #59
MrMagic is offline MrMagic  Greece
diyAudio Member
 
MrMagic's Avatar
 
Join Date: Nov 2017
Location: Athens
Quote:
Originally Posted by MarcelvdG View Post
You could buy a Benchmark Media DAC for 2200 euros including VAT (at least that's what they cost down here), or build one of my DAC designs (cheaper, but definitely not cheap, and with lower dynamic range numbers)
Ouch, 2200 euros/dollars just for a DAC is way beyond the cost I have in mind, as I don't want it for myself, I want to embedded it as an added bonus and convenience in the ES headphones amp, to ensure no clipping and also have excellent performance -while not necessarily the best-of-the-best on the market due to its low cost.
The amp itself though, should have incomparably better specs than anything has existed so far (did I mentioned incomparably?) while as a whole (+ the DAC and everything else) should have a very reasonable cost.

I didn't know you have designed a DAC, it reminds me a 10bit ADC I had designed in the 90's using cheap common ICs and discrete components (with a very good performance though), after a very frustrating experience trying for a week to make an ADC chip from NS to work properly, that was proved to be flawed (=completely trash)! (That was a hobby exercise, I couldn't pay a premium to buy an expensive ADC.)

Quote:
Originally Posted by MarcelvdG View Post
See the attachment for the basic idea. It boils down to playing the MSB twice and delaying everything else by one bit clock period. This gives you a 1 bit (6.02 dB) headroom at the expense of 1 bit of signal level. I thought that by redrawing the logic a bit, the EXOR, inverter and multiplexer could all be made with a 74HC158, but they can't because the 74HC158 multiplexers have a common select input. I've got to think about it a bit more.
Thanks, but dropping 1 bit would mean a significant digital-to-audio-conversion disadvantage that would cancel its advantage of no clipping. I prefer the opposite approach, that is, to expand the dynamic range instead of reduce it. Ideal would be a 17bit DAC, or a 24bit DAC (that can also be used for 24bit input), scaling up the input signal while leaving the same amount of headroom, eg to 23bits, thus retaining the whole input dynamic range, while avoiding clipping.

BTW, I appreciate all the help, but unfortunately I can't devote time for that right now, as I am completely overwhelmed by another project that requires a huge amount of research & development that I have to finish first (if everything goes well, it will fund this project too). I barely have time to post a comment once-a-day and I'm only doing it because I like chatting with great people here, about our common passion: electronics.

Last edited by MrMagic; 11th February 2018 at 12:24 PM.
  Reply With Quote
Old 11th February 2018, 01:54 PM   #60
MrMagic is offline MrMagic  Greece
diyAudio Member
 
MrMagic's Avatar
 
Join Date: Nov 2017
Location: Athens
Quote:
Originally Posted by MrMagic View Post
digital-to-audio-conversion
That was funny, I meant digital to analog
  Reply With Quote

Reply


How low can you go?Hide 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


New To Site? Need Help?

All times are GMT. The time now is 06:23 AM.


Search Engine Optimisation provided by DragonByte SEO (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Resources saved on this page: MySQL 15.00%
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2018 DragonByte Technologies Ltd.
Copyright ©1999-2018 diyAudio
Wiki