I have no further details and simply divided the 200nsec in two equal halves for slewing and settling.
Compare this to the 1.5usec for the TDA 1541.
But I still don’t see any reason in the Sim why the sound became worse.
So how the Dac is really behaving will probably remain a mystery, but it might explain why TI never propagated anything else but a fast active I/V converter, even in their recently revised version with the OPA1612.
Hans
Compare this to the 1.5usec for the TDA 1541.
But I still don’t see any reason in the Sim why the sound became worse.
So how the Dac is really behaving will probably remain a mystery, but it might explain why TI never propagated anything else but a fast active I/V converter, even in their recently revised version with the OPA1612.
Hans
What is notable in the left image relative to the right is that the output hasn't settled in at what is guessed a 196KHz sampling rate, meaning there isn't a kind of comparative "locked in" period. This would be unlike a 48KHz or NOS as continuing to settle in for 3 more periods under circumstances perhaps with 0.5mA current steps. It is suspected that non-settled waveforms contribute to auditory obscuration likened unto masking, veiling, noise, "carpet", etc. The waveform on the right has significantly improved settling, whereupon an improved lack of obscuration can reveal higher frequency detail as could also then reveal unwanted harmonic distortions as well.
Gerrit
Attachments
I'm planning to test that in the following weekend. It seems rather easy but spare time is restricted this period.
That is simple to explain.
The conventional I/V had it's -3dB point at 100Khz, whereas the MFB, being a second order filter has its -3dB point at 300Khz.
At 1Mhz both have roughly the same attenuation around -12dB.
And as discussed before, an LP filter has no effect on distortion whereas too low Slew Rate has.
So this difference in settling time can be disregarded.
As a matter of fact, in TI's most recent updated circuit diagram they also advise using 300Khz BW for the RC feedback network instead of the original 100Khz.
Hans
.
Attachments
MFB and Sallen Key are defining toplogies to realize multi order active filters.
So in this case a second order MFB variant was used where with the omission of an input resistor the Dac is directly connected to the first cap.
This causes the voltage excursion to the opamp as shown in the images to follow a much smoother path potentially preventing the opamp’s input to become overloaded.
So far the theory, but the result in this case is a step back in experienced sound reproduction.
Hans
So in this case a second order MFB variant was used where with the omission of an input resistor the Dac is directly connected to the first cap.
This causes the voltage excursion to the opamp as shown in the images to follow a much smoother path potentially preventing the opamp’s input to become overloaded.
So far the theory, but the result in this case is a step back in experienced sound reproduction.
Hans
I did a multi tone test with the MFB I/V converter with tones spaced 1Khz apart.
I don't see anything spectacular, so there is seemingly nothing malfunctioning.
Noise level of the scope is -129dBV, well below the measured noise in this recording with a noise level slightly below -100dBV.
Hans
.
I don't see anything spectacular, so there is seemingly nothing malfunctioning.
Noise level of the scope is -129dBV, well below the measured noise in this recording with a noise level slightly below -100dBV.
Hans
.
Attachments
The deeper one goes, the more complex it becomes.
I should have known better, but I assumed that the PCM1792 was switching its output current at 1/192Khz intervals.
However the ΔΣ converter operates at a much higher frequency.
I offered a dithered 16bit zero signal to the Dac, while measuring at the output of the LT1468 I/V converter with only 1K/1nF filter across, followed by a HP461 amplifier.
In the image below showing the result, the scale has to be divided by 10 because of the HP461 and the used probe, giving a 220uV rms output signal.
As can be seen, the signal hovers around zero volt, not even at fixed switching moments, but when looking at the spectrum, the centre is at ca 1.2Mhz.
More to come.
Hans
.
I should have known better, but I assumed that the PCM1792 was switching its output current at 1/192Khz intervals.
However the ΔΣ converter operates at a much higher frequency.
I offered a dithered 16bit zero signal to the Dac, while measuring at the output of the LT1468 I/V converter with only 1K/1nF filter across, followed by a HP461 amplifier.
In the image below showing the result, the scale has to be divided by 10 because of the HP461 and the used probe, giving a 220uV rms output signal.
As can be seen, the signal hovers around zero volt, not even at fixed switching moments, but when looking at the spectrum, the centre is at ca 1.2Mhz.
More to come.
Hans
.
Attachments
Here is the spectrum when offering a 1Khz -93.1dBFs 24bits dithered signal to the conventional active I/V converter in blue, versus the hybrid MFB version in red.
Obvious is that the noise production of the conventional version is superior.
At 200Khz the difference between the two is even 45dB.
It just seems that the current output does not like looking directly in the MFB's cap.
I also redid the multitone test, this time avoiding folding back of the very high noise from the MFB version.
And just like with the 1Khz test, noise beyond 20Khz stays on the same level instead of ramping down ??
Intermodulation products are now clearly visible.
Having added all of these experiences, I decided to go back to my original circuit diagram.
Hans
.
Obvious is that the noise production of the conventional version is superior.
At 200Khz the difference between the two is even 45dB.
It just seems that the current output does not like looking directly in the MFB's cap.
I also redid the multitone test, this time avoiding folding back of the very high noise from the MFB version.
And just like with the 1Khz test, noise beyond 20Khz stays on the same level instead of ramping down ??
Intermodulation products are now clearly visible.
Having added all of these experiences, I decided to go back to my original circuit diagram.
Hans
.
Attachments
Below, is an interesting e-mail exchange I had with the creator of the PGGB resampling software. That's the one based on a Windowed-SINC FIR filter selectable up to 8 billion taps. He shares some of his thoughts on implementations of OS versus NOS playback. Especially interesting, are his last two messages on why he believes that NOS and OS sound different when it seems they shouldn't. My messages to him are shown in italics, his to me are shown in plain text.
******************************************************************************
Hello,
I'm currently leading a diyAudio thread investigating high accuracy DAC up-sampling solutions. The PGGB appears to be the highest accuracy up-sampling solution which we've encountered. A major impediment to our members potentially utilizing a PGGB offline solution, however, is the cost of acquiring an PC/MAC for those whose systems do not already utilize a computer. In reviewing your website, I believe that it mentions your OEM support for 3rd party, streaming PGGB based products.
My question is, do you have a list of 3rd party vendors offering PGGB based streaming resampling within some form of dedicated/embedded computing module? With the objective of locating it inside the DAC box, directly feeding the DAC chip to serve as an oversampling digital interpolation-filter.
Regards,
Ken Newton
________________________________________________________________________________
Hello Ken,
It is great to hear the diy community is interested in PGGB. PGGB is in a way a DIY effort on creating my own upsampler and began as an experiment out of COVID boredom. Curious to know more about how you determined the quality of upsampling, appreciate a link to the relevant thread too.
Regarding requiring a PC/Mac, even with 16GB RAM you can get very good results and some are running PGGB on a Mac M1 now. PGGB scales with more resources. So one does not really need a 'super computer'.
Currently no one is offering an embedded computing module for this and I have not thought about it from that angle, I have SDKs for Mac/Windows and Arch Linux, with the idea of a Server or OS vendor or a music player would approach me with the interest to integrate.
Regards,
-ZB
-------------------------------------
Author PGGB
___________________________________________________________________________
Hi, ZB,
Good to get your response and interest. I launched the below linked thread at diyAudio to investigate the following question: Why don't NOS and OS DACs essentially have the same subjective character? It seems they should, once the zero-order-hold based -3.17 dB @ 20KHz response droop has been EQ'd to flat. Since all of the image-bands from 44.1KHz sample rate audio, and higher playback are ALL ultrasonic, the audio band should be fairly well reconstructed. So, why the characteristic difference? A difference which we basically subjectively judge for NOS as being much more relaxed, more 'natural' sounding and more 3D, but also, however, with a much narrower soundstage and diffuse bass and less dynamic sounding.
Our investigation does not just focus on image-band suppression, but every factor we could think of that technically differentiates NOS and OS DAC playback. The major categories we've identified for our investigation outline are: A) Frequency response error, B) Reconstruction/Image-Band Handling, C) Altered Jitter as a function of sample-rate, and D) Sample-Period related Quantization Errors. Prime among the suspects, but not yet proven, are Parks-McCllelan Equiripple FIR interpolation-filters and some other hardware based FIR filter artifacts.
Which, has led us to plan the testing of fully convolving Windowed-SINC FIR interpolation-filters in software, such as I believe are SoX and PGGB. Via feeding the output of those up-samplers to DACs with no other (internal) hardware digital FIR interpolation filters. Our ultimate hope is to subjectively marry the best of NOS sound with the best of OS sound. Any perspective or experience you have along those lines and on these issues would be most welcome for our thread to gain.
Best regards,
Ken Newton
________________________________________________________________________________
Oh, I forgot to answer your filter accuracy question.
I define audio interpolation-filter accuracy as one which most complies with Shannon-Nyquist. Most chip based hardware FIR interpolation-filters are, as I'm certain you know, compromised Equiripple, half-band implementations chosen to maximize processing efficiency, but not application performance. The only chip based filters, of which I'm aware, that are not Equiripple were the highly regarded Pacific Microsonics PMD-100 and PMD-200 FIR interpolation-filter ICs. Which, I suspect we're both (non-Equiripple) Windowed-SINC implementations.
1. Features a passband which is flat to within about +/- 0.5 dB @ 20KHz. However, does not exhibit the repeating FR ripples (and time-domain signal echoes) across the passband of Parks-McClellan Equiripple filters.
2. Features a Transistion-Band narrow enough to fully reach the filter's Stop-Band by the very start of the lowest CD image band frequency of 22KHz. Unlike Half-Band filters.
3. Features a Stop-Band attenuation of -120dB or better.
Best regards,
Ken Newton
_______________________________________________________________________________
Ha yes got it regarding accuracy.
Regarding marrying the best NOS DAC with the best OS software, I am in full agreement. Some of the newer NOS R2R DACs such as Denafrips T+ and Holo May operate at 1.5MHz and they scale very well with good software oversampling.
But compared to delta sigma DACs, which operate at much higher frequencies and have the advantage of noise shaping R2R DACs are subject to distortion due to quantization noise and nonlinearity due to resistor mismatch etc which is why PGGB uses noise shaping to push the quantization noise beyond audible spectrum.
Regarding NOS DACs that employ a simple zeroth order sample and hold or a first order interpolator, I agree that the images are ultrasonic and one would think it should not have an audible effect, but yet they do. I believe the answer lies in the time domain. If the original sample rate is F, zeroth order hold allows no variation between two successive samples at n/F, (n + 1)/F even in the analog domain (let us assume there is a very gentle or no analog filter). However, you can have infinitely many variations of true band limited analog signals in this interval (n/F, (n + 1)/F) recorder in the real world instead of the rather un realistic staircase even if these infinitely different analog signals have the same frequency domain response. These variations have both amplitude and time components that will help reproduce the sound more accurately and truer to the original band's limited source.
Another rather naive example is how a dirac impulse compares to white noise, both are broad band, yet in the time domain when all the tones arrive at the same instant you get a dirac pulse vs when they are rather chaotic resulting in white noise,
Regards,
-ZB
-------------------------------------
Author PGGB - remastro
_______________________________________________________________________________
Hi, ZB,
If I correctly understand what you are suggesting, we had a similar thought, which was captured as an item on our investigation outline. Our speculation was whether, even though unfiltered image-bands are ultrasonic, an insufficiently reconstructed signal baseband might still exhibit audible error. Except, that we couldn't reconcile why such an improper signal reconstruction should make NOS playback subjectively so more relaxed and natural sounding than OS. This seems unlikely to be simply due to serendipity, but who knows?
-Ken
_______________________________________________________________________________
There are two combinations possible
1. Higher reconstruction error but no aliasing artifacts in audible range (NOS DACs)
2. OS DACs with varying degrees of aliasing (because of limited on board processing) combined with varying degrees of reconstruction error (because Whitaker-Shannon coefficients are not used)
I believe NOS DACs have the advantage of not requiring anti aliasing filters used for OS and avoiding aliasing artifacts and tonally in my experience this translates to a denser/weighty presentation with a more relaxed treble. Reconstruction accuracy affects the quality of leading edges and transients in general, also affects resolution and depth. When reconstruction accuracy improves, one of the first things you will notice is the quality of bass. So it will not be entirely surprising that NOS DACs can sound a little off in the low end.
Poorly implemented OS DACs can fail on both reconstruction and suppressing aliased images and end up sounding less natural than NOS DACs but well implemented OS DACs can in my opinion surpass NOS DACs
Regards,
-ZB
-------------------------------------
Author PGGB - remastro
******************************************************************************
Hello,
I'm currently leading a diyAudio thread investigating high accuracy DAC up-sampling solutions. The PGGB appears to be the highest accuracy up-sampling solution which we've encountered. A major impediment to our members potentially utilizing a PGGB offline solution, however, is the cost of acquiring an PC/MAC for those whose systems do not already utilize a computer. In reviewing your website, I believe that it mentions your OEM support for 3rd party, streaming PGGB based products.
My question is, do you have a list of 3rd party vendors offering PGGB based streaming resampling within some form of dedicated/embedded computing module? With the objective of locating it inside the DAC box, directly feeding the DAC chip to serve as an oversampling digital interpolation-filter.
Regards,
Ken Newton
________________________________________________________________________________
Hello Ken,
It is great to hear the diy community is interested in PGGB. PGGB is in a way a DIY effort on creating my own upsampler and began as an experiment out of COVID boredom. Curious to know more about how you determined the quality of upsampling, appreciate a link to the relevant thread too.
Regarding requiring a PC/Mac, even with 16GB RAM you can get very good results and some are running PGGB on a Mac M1 now. PGGB scales with more resources. So one does not really need a 'super computer'.
Currently no one is offering an embedded computing module for this and I have not thought about it from that angle, I have SDKs for Mac/Windows and Arch Linux, with the idea of a Server or OS vendor or a music player would approach me with the interest to integrate.
Regards,
-ZB
-------------------------------------
Author PGGB
___________________________________________________________________________
Hi, ZB,
Good to get your response and interest. I launched the below linked thread at diyAudio to investigate the following question: Why don't NOS and OS DACs essentially have the same subjective character? It seems they should, once the zero-order-hold based -3.17 dB @ 20KHz response droop has been EQ'd to flat. Since all of the image-bands from 44.1KHz sample rate audio, and higher playback are ALL ultrasonic, the audio band should be fairly well reconstructed. So, why the characteristic difference? A difference which we basically subjectively judge for NOS as being much more relaxed, more 'natural' sounding and more 3D, but also, however, with a much narrower soundstage and diffuse bass and less dynamic sounding.
Our investigation does not just focus on image-band suppression, but every factor we could think of that technically differentiates NOS and OS DAC playback. The major categories we've identified for our investigation outline are: A) Frequency response error, B) Reconstruction/Image-Band Handling, C) Altered Jitter as a function of sample-rate, and D) Sample-Period related Quantization Errors. Prime among the suspects, but not yet proven, are Parks-McCllelan Equiripple FIR interpolation-filters and some other hardware based FIR filter artifacts.
Which, has led us to plan the testing of fully convolving Windowed-SINC FIR interpolation-filters in software, such as I believe are SoX and PGGB. Via feeding the output of those up-samplers to DACs with no other (internal) hardware digital FIR interpolation filters. Our ultimate hope is to subjectively marry the best of NOS sound with the best of OS sound. Any perspective or experience you have along those lines and on these issues would be most welcome for our thread to gain.
Best regards,
Ken Newton
________________________________________________________________________________
Oh, I forgot to answer your filter accuracy question.
I define audio interpolation-filter accuracy as one which most complies with Shannon-Nyquist. Most chip based hardware FIR interpolation-filters are, as I'm certain you know, compromised Equiripple, half-band implementations chosen to maximize processing efficiency, but not application performance. The only chip based filters, of which I'm aware, that are not Equiripple were the highly regarded Pacific Microsonics PMD-100 and PMD-200 FIR interpolation-filter ICs. Which, I suspect we're both (non-Equiripple) Windowed-SINC implementations.
1. Features a passband which is flat to within about +/- 0.5 dB @ 20KHz. However, does not exhibit the repeating FR ripples (and time-domain signal echoes) across the passband of Parks-McClellan Equiripple filters.
2. Features a Transistion-Band narrow enough to fully reach the filter's Stop-Band by the very start of the lowest CD image band frequency of 22KHz. Unlike Half-Band filters.
3. Features a Stop-Band attenuation of -120dB or better.
Best regards,
Ken Newton
_______________________________________________________________________________
Ha yes got it regarding accuracy.
Regarding marrying the best NOS DAC with the best OS software, I am in full agreement. Some of the newer NOS R2R DACs such as Denafrips T+ and Holo May operate at 1.5MHz and they scale very well with good software oversampling.
But compared to delta sigma DACs, which operate at much higher frequencies and have the advantage of noise shaping R2R DACs are subject to distortion due to quantization noise and nonlinearity due to resistor mismatch etc which is why PGGB uses noise shaping to push the quantization noise beyond audible spectrum.
Regarding NOS DACs that employ a simple zeroth order sample and hold or a first order interpolator, I agree that the images are ultrasonic and one would think it should not have an audible effect, but yet they do. I believe the answer lies in the time domain. If the original sample rate is F, zeroth order hold allows no variation between two successive samples at n/F, (n + 1)/F even in the analog domain (let us assume there is a very gentle or no analog filter). However, you can have infinitely many variations of true band limited analog signals in this interval (n/F, (n + 1)/F) recorder in the real world instead of the rather un realistic staircase even if these infinitely different analog signals have the same frequency domain response. These variations have both amplitude and time components that will help reproduce the sound more accurately and truer to the original band's limited source.
Another rather naive example is how a dirac impulse compares to white noise, both are broad band, yet in the time domain when all the tones arrive at the same instant you get a dirac pulse vs when they are rather chaotic resulting in white noise,
Regards,
-ZB
-------------------------------------
Author PGGB - remastro
_______________________________________________________________________________
Hi, ZB,
If I correctly understand what you are suggesting, we had a similar thought, which was captured as an item on our investigation outline. Our speculation was whether, even though unfiltered image-bands are ultrasonic, an insufficiently reconstructed signal baseband might still exhibit audible error. Except, that we couldn't reconcile why such an improper signal reconstruction should make NOS playback subjectively so more relaxed and natural sounding than OS. This seems unlikely to be simply due to serendipity, but who knows?
-Ken
_______________________________________________________________________________
There are two combinations possible
1. Higher reconstruction error but no aliasing artifacts in audible range (NOS DACs)
2. OS DACs with varying degrees of aliasing (because of limited on board processing) combined with varying degrees of reconstruction error (because Whitaker-Shannon coefficients are not used)
I believe NOS DACs have the advantage of not requiring anti aliasing filters used for OS and avoiding aliasing artifacts and tonally in my experience this translates to a denser/weighty presentation with a more relaxed treble. Reconstruction accuracy affects the quality of leading edges and transients in general, also affects resolution and depth. When reconstruction accuracy improves, one of the first things you will notice is the quality of bass. So it will not be entirely surprising that NOS DACs can sound a little off in the low end.
Poorly implemented OS DACs can fail on both reconstruction and suppressing aliased images and end up sounding less natural than NOS DACs but well implemented OS DACs can in my opinion surpass NOS DACs
Regards,
-ZB
-------------------------------------
Author PGGB - remastro
Last edited:
Hi Hans,
you cannot say this is ~ -100dBV -noise level. Only the 3 Hz BINs are ~ -100dBV. If you take the bandwidth of 20-20kHz into account your S/N ratio will be more like -65 dBV
Who is taking the challenge with his NOS Dac to do the up and downsample test and compare the original with the processed version ?
I have proposed 3 different .wav files.
Would it be of any help when these 3 files are also made available after being upsampled to 192/24 and back to 44.1/16 ?
I’m willing to do this to achieve some sort of reference for all NOS users to base their comments on the same files.
Maybe that can give this nice thread a push in the back.
Hans
I have proposed 3 different .wav files.
Would it be of any help when these 3 files are also made available after being upsampled to 192/24 and back to 44.1/16 ?
I’m willing to do this to achieve some sort of reference for all NOS users to base their comments on the same files.
Maybe that can give this nice thread a push in the back.
Hans