Just a note, you will probably not find any "NOS" ADCs today. As a matter of fact I don't event know if there are any ADC chips out there that don't use sigma delta technology... that means sampling at a rate way higher than the target sampling rate and then decimating down (using a digital filter to remove the OOB contents).
If there are some, please enlighten me.
EDIT: I did some reading and there's the "successive approximation" type of ADC which has a DAC as integral part of it's structure. If you would use a NOS DAC in this position I guess you could build a true NOS ADC. Are there any? I don't know...
If there are some, please enlighten me.
EDIT: I did some reading and there's the "successive approximation" type of ADC which has a DAC as integral part of it's structure. If you would use a NOS DAC in this position I guess you could build a true NOS ADC. Are there any? I don't know...
Last edited:
According to a theory of R. Lagadec and T. G. Stockham, 0.03 dB peak-peak ripple is equivalent to a pre- and postecho of -61.27 dB. Considering the filter length, at 340 us before and after the main signal.
It's straightforward to make such pre- and postechoes in GoldWave. Using an ABX testing program and a NOS DAC, you can then check whether this makes the NOS DAC sound different (like a DAC with an SAA7220?)
There's this thread, but they are oversampling the SAR ADC...
SAR ADC for high performance audio ADC project [LTC2380-24]
If you would run a SAR ADC at the target sample rate, it would indeed be a non-oversampling ADC, and you'd need a pretty selective analogue filter to get rid of aliases.
Looks very impressive.
Many years ago I was intending to design a device around the parallel load 50 Msps LTC1668 16-bit device, although I never got around to trying it. The idea was to remove the serial stream of data/clock signals from the internals of the ADC to reduce internal noise, that along with high speed settling was thought would produce a clean pair of differential current step outputs...
Many years ago I was intending to design a device around the parallel load 50 Msps LTC1668 16-bit device, although I never got around to trying it. The idea was to remove the serial stream of data/clock signals from the internals of the ADC to reduce internal noise, that along with high speed settling was thought would produce a clean pair of differential current step outputs...
Member
Joined 2009
Paid Member
Maybe that’s why I have given up on digital lately, just can’t get a good sound. If NOS is the way to go there’s apparently none available. The most bearable digital source I have is a Marantz CD player but the network streamers I’ve tried had to go. I don’t have the patience to keep trying them until I find something decent (and that includes an acceptable user interface) so apart from a few CD’s I’m back to buying records and CD’s. This thread could yet point me towards a better solution.
Last edited:
Don't want to bust your expectations, but how should this thread point you towards a better digital solution if (almost) any digital recording did involve a non-NOS technology? Or are you talking about a new (yet unheard of) analog format? Compact Cassette?
Sorry for being so sarcastic and pointing out non-sensical (at least to me) arguments.
This thread is still very viable and from everything that I read the oversamlping/decimation algorithm and filter very much matter. At least for my terms that's not non-sensical, why should quality not matter if trying to achieve quality re-preoduction. Just don't be fooled to think that the re-production chain can fix everything for you - There's always a recording chain involved, which you cannot get rid of. It's flaws are always baked into the digital recording, there's no way around that.
To alleviate most of the concerns revolving around digital recordings - Here's my POV: I still very much enjoy even lossy 320k MP3 (and even lower, down to 256k and sometimes 192k) formats even on a pretty revealing system. In the end I enjoy the MUSIC, not the transport format and its shortcomings.
EDIT: If I ever had golden ears, they just broke off after posting this and vanished into fairy dust. Sorry guys, I'm not qualified for any audiophile listening tests anymore I fear
Sorry for being so sarcastic and pointing out non-sensical (at least to me) arguments.
This thread is still very viable and from everything that I read the oversamlping/decimation algorithm and filter very much matter. At least for my terms that's not non-sensical, why should quality not matter if trying to achieve quality re-preoduction. Just don't be fooled to think that the re-production chain can fix everything for you - There's always a recording chain involved, which you cannot get rid of. It's flaws are always baked into the digital recording, there's no way around that.
To alleviate most of the concerns revolving around digital recordings - Here's my POV: I still very much enjoy even lossy 320k MP3 (and even lower, down to 256k and sometimes 192k) formats even on a pretty revealing system. In the end I enjoy the MUSIC, not the transport format and its shortcomings.
EDIT: If I ever had golden ears, they just broke off after posting this and vanished into fairy dust. Sorry guys, I'm not qualified for any audiophile listening tests anymore I fear
Last edited:
Marcel, that seems like an interesting tool. While I'm not familiar with it, I'm intrigued. Coincidently, I was recently lamenting to Hans that, while we seem to have identified pedestrian OS interpolators as being at fault, we did not identify exactly which functional aspect is directly responsible. With the help of the GoldWave tool, perhaps, we can answer the potential guilt of what I've always felt was the prime suspect. The Lagadec echo. Could you discuss and coordinate with Hans by PM, regarding providing a few songs files made with GoldWave for us to potentially offer a listening test of the echo hypothesis?
While I would like for us to determine the exact technical reason why typical interpolators are introducing audible artifacts, I'm also cautious about pressing the patience of our voluntary group to participate in too many experiments. Perhaps, by utilizing some new songs for an echo test it might enhance participation interest.
Feel free to PM me with your thoughts, should you prefer.
Last edited:
IMHO, you are not necessarily wrong about high bitrate MP3s. Sometimes the perceptual encoders are able to remove some small imperfections of the recording process. In some cases it can end up sounding just as 'good' as the original, even though it sounds somewhat different. AC3 is probably a little better than MP3 at the same bitrate though.
Studies suggest a steady diet of velvet beans (Mucuna pruriens). Apparently they contain high levels of L-dopa, the precursor molecule to dopamine (the pleasure drug). I like to imagine a hypodermic filled with dopamine being pushed down my MUSIC. So whatever pushes the contents of your plunger... 192k MP3 or otherwise...
Just as a side note, this app note contains some useful information related to the sinc function that occurs at first order sample and hold in digital-to-analog conversion. Most notably, it mentions two equalization technique:
- using an inverse of the sinc function in the digital domain (pre-equalization)
- or using an analog boost where the sinc droop occurs (post-equalization)
- using an inverse of the sinc function in the digital domain (pre-equalization)
- or using an analog boost where the sinc droop occurs (post-equalization)
OTOH, in my setup, I find the rolloff to be excessive. My best subjective results so far are with a passive filter that is fairly flat out to 20 kHz.
A high quality 88.2k/24 or 96K/24 soundcard in your PC will do the job.
It has many advantages to record at a frequency beyond 44.1K and to downsample afterwards because it’s almost impossible to do the anti alias filtering in the analogue domain. But this then needs digital processing before downsampling to 44.1k which as we have seen may corrupt the sound to some degree.
Recording at 88.2K can probably be done with an analoque anti alias filter.
When playing this file at 88.2K through a NOS Dac, will give you a digitally unproceesed file, which seems like a very attractive option.
Hans
I highly doubt that, because most likely the 88.2k will stem from a oversampled SD ADC and the downsampling process to 88.2k will have involved a digital filter of some sort.
Digital filter and upsampling 2
I did some experiments and will do a more solid one tomorrow and post again with images and listening experience…
So what I did,
I installed the trial version (30 minutes before restart, which is practical enough) of HQ Player.
In Roon, I embedded the HQ Player so it is real time upsampling and filtering.
I upgraded my RPI3 with Ropieee to RopieeeXL which has an option to be an endpoint for HQPlayer (!)
In Roon you can select HQPlayer as endpoint and just play any track. HQP will than upsample and filter to your choice through the settings menu.
There is a crazy amount of filters and all have some kind of purpose according to the manual. If you can really distinguish, I will leave for now.
Ok, so I did put all in my workshop and connected the DDDAC output to my scope.
Played a 12 kHz square wave with FS48 and watched the result from all filters
I found out that Polynomial-1 was the closest to the original square wave
ALL other filters gave the typical pre and after ringing or only after ringing.
I ran a wide band FFT of the 48 and 192 version and as expected the noise of the latter was significant lower
So what I will do next is
Create a -6dB square wave soundtrack with 1kHz 44.1 kHz (to avoid any overload of filter and enough harmonics in typical audio spectrum) also make a 1,2kHz version with 48kHz. You know why
Play both with and without some different filters and upsample with 2x and 4x
Measure again to make 100% sure the difference in filter is still what I thought it was in this long session yesterday and make screenshots just to give you an idea what is happening in the time domain
Of course run the FFT again of the original and upsample signals to show out of audio band noise
Last but NOT least, listen to the 4 original tracks from Hans with and without HQPlayer upsampling with the polynomial-1, FIR and IIR filter settings and see if I can hear any significant difference and if so what I like best
As very last, compare the one setting and track I (maybe) liked best and compare with the built in filter and upsampling function in Roon….
Stay tuned….
PS, VERY unfortunate, also in HQPlayer there is no zero-order sample and hold filter which just up samples…. But as said, the polynomial-1 comes very close
I did some experiments and will do a more solid one tomorrow and post again with images and listening experience…
So what I did,
I installed the trial version (30 minutes before restart, which is practical enough) of HQ Player.
In Roon, I embedded the HQ Player so it is real time upsampling and filtering.
I upgraded my RPI3 with Ropieee to RopieeeXL which has an option to be an endpoint for HQPlayer (!)
In Roon you can select HQPlayer as endpoint and just play any track. HQP will than upsample and filter to your choice through the settings menu.
There is a crazy amount of filters and all have some kind of purpose according to the manual. If you can really distinguish, I will leave for now.
Ok, so I did put all in my workshop and connected the DDDAC output to my scope.
Played a 12 kHz square wave with FS48 and watched the result from all filters
I found out that Polynomial-1 was the closest to the original square wave
ALL other filters gave the typical pre and after ringing or only after ringing.
I ran a wide band FFT of the 48 and 192 version and as expected the noise of the latter was significant lower
So what I will do next is
Create a -6dB square wave soundtrack with 1kHz 44.1 kHz (to avoid any overload of filter and enough harmonics in typical audio spectrum) also make a 1,2kHz version with 48kHz. You know why
Play both with and without some different filters and upsample with 2x and 4x
Measure again to make 100% sure the difference in filter is still what I thought it was in this long session yesterday and make screenshots just to give you an idea what is happening in the time domain
Of course run the FFT again of the original and upsample signals to show out of audio band noise
Last but NOT least, listen to the 4 original tracks from Hans with and without HQPlayer upsampling with the polynomial-1, FIR and IIR filter settings and see if I can hear any significant difference and if so what I like best
As very last, compare the one setting and track I (maybe) liked best and compare with the built in filter and upsampling function in Roon….
Stay tuned….
PS, VERY unfortunate, also in HQPlayer there is no zero-order sample and hold filter which just up samples…. But as said, the polynomial-1 comes very close
Last edited:
My point was, that also the BCK needs to go up to reduce the noise of the DSM, as The DDDAC1794 uses the PCM1794 in no digital filter mode with SCK = BCK and that leads to in audio band noise for 44.1 and 48 tracks. I wanted to find out if I speed up the DSM with the same track and sample content, if sound reproduction perception would be better