Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: GOOD non-oversampling DAC? (Read 50136 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

GOOD non-oversampling DAC?

Reply #50 –
Quote from: oatmeal769 on
Okay then, back on topic, 'member I mentioned about op-amps versus a mo'betta output stage?  (not just with DAC's but anything with gain)  That I can A/B/X, even now, but no one called me out on that.  Am I generally considered to be correct in that regard?


The oy reason why I didn't call you out before is that I never heard of this.

Yes, TOS 8 is still there, mind you!

The idea that one can readily wine taste good (and even shabby) op amps is yet another audiophile myth.

Fact is that even the much-maligned 709 op amp can be compared to a straight wire and come up clean if you use it within its obvious limits. In case you don't know, the 709 was one of the very first op amps ever sold in volume and is eay to abuse so that it  slew rate limits, is too noisy, and even rolls off the high end. But, the audio output buffers for CD/DVD/digital players is really pretty easy duty, so with some careful  mouth holding 709s can even be sucessfully used for that.

I love my 5532s as well if not better than the next guy, and have used them myself in a number of custom designed equipment. I have a ton of gear that uses them and their more recent betters.

However, as output buffers for CD players, 5532s are both gross overkill, and also apparently swapped out en masse by ignorant audiophools, no doubt based on the overheated non-sensiing, non-thinking that passes for audio journalism these days. 

The current generation of ultra low distortion op amps from National and TI primarily exist so that people can bench test the ultra low converters that are now becoming readily available.  IOW, they are mostly about chasing impressive numbers, and are not requirements for good sound. Not that they sound bad. They are just gross overkill for DAC output buffers for reproducing music as it can be possibly be recorded and played.

GOOD non-oversampling DAC?

Reply #51 –
Oh, without a doubt, I doubt I can tell a difference between OpAmps themselves, at least not when swapping them out in the ZERO, but I can tell the diff. between two pieces of hardware. The PSA DLIII  I have uses no OpAmps anywhere.

Here's my first little experiment, between my Zero DAC (op-amp) and My PS Audio DLIII  I don't know how many times I need to do it,  I got 6 out of 7 so far - (???)  or how to post results (haven't read that far yet). But I volume leveled and timed the two clips exactly, etc. (I still use an ancient copy of SoundForge, hope that's okay.)  I did notice though that the difference between the two graphics (in SoundForge, the PSA DLIII looks  'fuller' to me in the graphic) were much more apparent than the actual ear test - It was harder than I thought by listening - I thought I'd nail it instantly, but I did have to listen a while.  I want to have a go again tomorrow, when I'm fresh. 

It does bring up an interesting point though, I don't know that it's op-amps PER SE vs. Dedicated circuits, or whether circuit design itself is at fault...  I can only A/B/X the actual units. 

GOOD non-oversampling DAC?

Reply #53 –
Quote from: greynol on
1) These clearly aren't level-matched.

2) You should make sure that your recording device isn't doing any limiting.
Oops, I think you're right.  It's been a while since I've used that software, I think I used the wrong tool.  I'll RTFM and re - do them...

GOOD non-oversampling DAC?

Reply #54 –
If you want accurate results you should level-match before recording and not in software.

GOOD non-oversampling DAC?

Reply #55 –
Quote from: oatmeal769 on
Okay then, back on topic, 'member I mentioned about op-amps versus a mo'betta output stage?  (not just with DAC's but anything with gain)  That I can A/B/X, even now, but no one called me out on that.


I did, at the end of my previous post on this thread, but I guess I was too subtle about it.

Quote
Am I generally considered to be correct in that regard?


Depends on how much the forum you make that claim in values good evidence.  Here, as you've seen, the answer would be no.

GOOD non-oversampling DAC?

Reply #56 –
Quote from: krabapple on
Quote from: oatmeal769 on
Okay then, back on topic, 'member I mentioned about op-amps versus a mo'betta output stage?  (not just with DAC's but anything with gain)  That I can A/B/X, even now, but no one called me out on that.


I did, at the end of my previous post on this thread, but I guess I was too subtle about it.

Quote from: krabapple on
Quote from: oatmeal769 on
Ultimately for DAC's you want to get away from Op-Amps and use those which have a discrete output section.  Then comes better internals, clocking and synching, etc.

    It's a shame you mix good advice, e.g., diminishing returns, with so much that's dubious. There may be no audiophile advice more insidious than that given by people who get stuff half right .

Anyway, I think you're back to "mumbo-jumbo" land with that notion, oatmeal769.  It's far easier to get matching differential pairs when working with a single piece of silicon then you can with discrete components.

GOOD non-oversampling DAC?

Reply #57 –
Quote from: rpp3po on
If you want accurate results you should level-match before recording and not in software.
That sounds reasonable.  Searched around, couldn't find instructions so I'm going to:
Play a 1KHz at 0dB track through each, and simply use a multi-meter to measure voltage at the line input back at the PC (?)  That would be about the only way I can think of to be certain.  Or can I just do it in whatever recording software I'm using?  I'm sure there are instructions somewhere, but I searched...

GOOD non-oversampling DAC?

Reply #58 –
Quote from: greynol on
I think you're back to "mumbo-jumbo" land with that notion, oatmeal769.
We soon shall see, that is if I can get some good samples recorded.  If I am wrong I'll own it.  I'm not above that, learning can be painful at times.  Besides, think of the money I can save not investing in more wizardry without proof.  I do still have the idea of a sum of components possibly  contributing to a greater whole...

GOOD non-oversampling DAC?

Reply #59 –
Quote from: oatmeal769 on
Play a 1KHz at 0dB track through each, and simply use a multi-meter to measure voltage at the line input back at the PC (?)  That would be about the only way I can think of to be certain.  Or can I just do it in whatever recording software I'm using?  I'm sure there are instructions somewhere, but I searched...


This is correct. You multimeter must be capable of measuring alternating voltage, though. You measure at the output of your two sources. Alternatively, record with your favorite software at 24 bit resolution and watch the amplitude of your recording of a 1 kHz tone. Adjust the output gain of your two sources until you get identical decibel readings with at least 0.1db accuracy. You don't need to record at 0db full scale, better is synchronizing both sources to -3db or -6db to avoid clipping. Many ADCs use digital volume controls, which reduce resolution. Leave your input gain at its default setting and only adjust the output gain of your two sources.

To time-sync your records, you can use a software, [a href='index.php?act=findpost&pid=639719']Synchrotron[/a], I have written some time ago. Its intended purpose was fixing encoder gaps, so you cannot feed it with large time differences; but aligning both tracks visually to lets say <10000 sample difference and then let Synchrotron do the rest for perfect sample accuracy should do the job. I might update Synchrotron in the future to allow arbitrary differences, which would make the preceding visual aligning obsolete. But I haven't found the time, yet.

GOOD non-oversampling DAC?

Reply #60 –
As most multi-meters don't have a dB scale, the voltage equivalent for a 0.1dB match is:

With a 1.0000 volt reference the 0.1dB range is 0.9886V to 1.012V

Frequency response and voltage accuracy of the multi-meter really don't matter.
Kevin Graf :: aka Speedskater

GOOD non-oversampling DAC?

Reply #61 –
Quote from: rpp3po on
If you want accurate results you should level-match before recording and not in software.


I've done it both ways literally 100s if not 1,000s of times.

It is easier in software. Potentially more accurate, too.

Also, I wouldn't match levels with a FS digital signal, but rather one that is 3. 6, or even 10 dB down. Some digital audio equipment clips before 0 dB. If you level match two different pieces of equipment, one that clips before FS and one that does not, you may cause a mismatch for most other levels.

Finally, its easy to estimate dB from a meter that does not have dB markings, especially over a 1 dB range.  A 10% mismatch is pretty close to 1 dB, a 1% mismatch is about 0.1 dB, and so on for smaller changes.

Level matching should be done with a DVM that has at least 4 digits, and the reference level should use at least 3 of those digits. 4 digit DVMs can be found for under $40 at Home Depot and Lowes. I posted the results of testing one such device here a few months back. Themeter is now at a job site, so I can easily post the make and model.

GOOD non-oversampling DAC?

Reply #62 –
Quote from: Arnold B. Krueger on
It is easier in software. Potentially more accurate, too.
I'm pretty sure SoundForge can do it too, I've used it recording 100's of times. - It's just been quite a while, I need to DL the manual and read it again.  I do know that you can monitor directly what's coming in, and I'm pretty sure you can adjust resolution of the record/play meter to fractions of single dB's.
Quote from: Arnold B. Krueger on
I wouldn't match levels with a FS digital signal, but rather one that is 3. 6, or even 10 dB down.
Good point.
Quote from: rpp3po on
Leave your input gain at its default setting and only adjust the output gain of your two sources.
No adjustment available, but I can use the recording software, rather than the input gain of the Windows recorder, That should be much more precise.
Quote from: rpp3po on
To time-sync your records, you can use a software, [a href='index.php?act=findpost&pid=639719']Synchrotron[/a], I have written some time ago.
This is really cool!  I'll give it a look.  I'm busy for the weekend, but I will do all of this.  I intend to use this system from now on, I've got loads of applications for it already.
 

GOOD non-oversampling DAC?

Reply #63 –
I read what I could find, and did some playing around, Here's what I came up with.

This time I set the Windows recorder level at it's default, and used S.F. to fine tune the input.  Without gain control over each DAC's output individually, I felt that voltage based adjustments would be extremely difficult if not impossible.  I also found that my input gain could be controlled much more finely using SF, where I adjusted the meter for highest resolution. I also did as suggested and made it so the peak loudness of each recording was 3dB down, in order to ensure against clipping.

Once I'd captured, I did some reading and playing around with volume leveling and normalization.  There are two means of doing it.  You can use an average measurement of the whole track - Root/Mean Square.  This takes the overall volume average, compares it to a set point, and then changes the file to appear 'louder'.  I believe it's the same thing as is done in the loudness wars.  The file itself is changed, and dynamics are compressed or expanded to make the overall volume more consistent.

The othe method is simple peak leveling.  The file is not changed as far as dynamics are concerned, rather the peak point of volume amplitude is taken, then the entire file is made to fit within a given set point.  In this instance I made each fit in at -3dB before clipping.  In this manner the file itself remains unchanged.  It is simply made so that it's loudest point is placed within the given set point.

If you look at each graphic, they do appear different, but looking closely, one can tell that each has the same point of maximum loudness. (-3dB)  I found these readily distinguishable, as the results show.

Subjectively, I couldn't hear what I'd call a significant difference in volume.  One simply sounded more 'full' than the other.  It had a richness in harmonic overtones the other didn't.  While I couldn't see which unit was which in the ABX, I'm pretty sure I know, and listening to the files by name afterword confirmed it. 

Then I tried something else - I applied the RMS standard to each file.  This made each one seem like one big splat on the graph, but I listened anyway.  I found the difference in sound MORE pronounced, rather than less.

I submit that any differences in loudness in one file vs. the other are due to the unique signature and timbre of the unit itself.  They are in fact 'as loud as one another'.  I even noticed that the ZERo had more dynamic range than did the PSA-DLIII.

I'd be happy to upload FLAC's of the files (a truncated version under 30 sec. the originals are timed at 42 sec's.)
I don't know of any set in stone standard as to how to achieve equal volume, so with that in mind....,
Fire away!

GOOD non-oversampling DAC?

Reply #64 –
Quote from: oatmeal769 on
I'd be happy to upload FLAC's of the files (a truncated version under 30 sec. the originals are timed at 42 sec's.)
I don't know of any set in stone standard as to how to achieve equal volume, so with that in mind....,


Please upload your FLACs ASAP.

GOOD non-oversampling DAC?

Reply #65 –
Posted  HERE

GOOD non-oversampling DAC?

Reply #66 –
Arny notes a significant (1.8dB) RMS level difference between them in that thread (I think such observations) are best kept on this thread, btw).  Could be one of those DACs is *broken*.

GOOD non-oversampling DAC?

Reply #67 –
Why the tedious sloppiness after you have been given so much good advice? First level match (e. g. 1kHz sine) then record. What's so hard about that?

GOOD non-oversampling DAC?

Reply #68 –
Quote from: rpp3po on
Why the tedious sloppiness after you have been given so much good advice? First level match (e. g. 1kHz sine) then record. What's so hard about that?
I DID level match and record, as I wrote in my post, using the recording software as Mr. Krueger had suggested.  Peak levels were identical, (at 3dB under 0) through each unit prior to capture.  I simply did not use a multimeter.
Quote from: krabapple on
Arny notes a significant (1.8dB) RMS level difference between them in that thread (I think such observations) are best kept on this thread, btw).  Could be one of those DACs is *broken*.
I have no doubt that the RMS rating is lower - as the graph plainly shows, and as I described in my post. 
As I also said, matching them by compressing / expanding the file to make them both of the same 'average' (RMS) volume will change the file itself.  Squashing one sample, or expanding the other will change it's sound.  These samples are 'volume level matched'.  The loudest point of each is at precisely -3.000 dB.  Remember, it is the SAME file, just played through different equipment.

GOOD non-oversampling DAC?

Reply #69 –
Quote from: oatmeal769 on
Once I'd captured, I did some reading and playing around with volume leveling and normalization.  There are two means of doing it.  You can use an average measurement of the whole track - Root/Mean Square.  This takes the overall volume average, compares it to a set point, and then changes the file to appear 'louder'.  I believe it's the same thing as is done in the loudness wars.  The file itself is changed, and dynamics are compressed or expanded to make the overall volume more consistent.

The othe method is simple peak leveling.  The file is not changed as far as dynamics are concerned, rather the peak point of volume amplitude is taken, then the entire file is made to fit within a given set point.  In this instance I made each fit in at -3dB before clipping.  In this manner the file itself remains unchanged.  It is simply made so that it's loudest point is placed within the given set point.

If you look at each graphic, they do appear different, but looking closely, one can tell that each has the same point of maximum loudness. (-3dB)  I found these readily distinguishable, as the results show.

Are you hear to learn, or to show everyone how clever you think you are?

I ask, because if it's the latter, you really need to leave right now.

If it's the former, someone with immense patience might be able to make you understand how very wrong your procedure is. But before they spend this time, you need to say "yes, I want to understand what I did wrong, and where I am mistaken in what I think I know".

...starting with this bizarre fictional "understanding" of RMS normalisation, which in reality has nothing to do with dynamic range compression!

Cheers,
David.

GOOD non-oversampling DAC?

Reply #70 –
Quote from: oatmeal769 on
I DID level match and record, as I wrote in my post, using the recording software as Mr. Krueger had suggested.


Sorry, it rather looks to me as if you DID two recordings and then normalized them to peaks (maybe fucking up even more using your flawed understanding of normalization). Adjusting the input gain in a second round with a digital gain control by the offset found in your first step leads to identical results BTW. The procedure is flawed.

It also rather looks as if you think you have understood everything and can skip what you consider needless HA pedantry. You obviously skipped volume matching using a sine, because you thought you know better. Stuff like this makes communicating with you very tiring. Using two recordings with musical signals isn't suited for volume matching! Even little sub-sample offsets, that you cannot avoid with your equipment, lead to different digital peaks for the same position in an analog sample. That's just the nature of PCM. Using sines for level matching pretty much eliminates this.

You need to provide the following:
  • Recording of DAC1, time-synched, level-matched.
  • Recording of DAC2, time-synched, level-matched.
  • Original sample.
  • Documentation about the procedures used and exact input gain values used for each recording. The recorded samples should not be touched, except for time-synching. If you don't understand how to do the latter, upload completely untouched recordings and I'll provide time-synched versions.

Else we can finish this conversation.

GOOD non-oversampling DAC?

Reply #71 –
Quote from: krabapple on
Arny notes a significant (1.8dB) RMS level difference between them in that thread (I think such observations) are best kept on this thread, btw).  Could be one of those DACs is *broken*.


This is what I posted on the sample thread:

There is also an approximate 1.8 dB difference in average RMS level.

I noticed that the average RMS levels differed by amounts that were significantly different from the difference in peak values. This may indicate that one of the DACs had considerable nonlinear distorion and thus failed the criteria of being "a good DAC".

The supplied samples pretty much *have* to sound different because of the apparent RMS level mismatch.

Once I corrected for most of the level mismatch, the spectral responses were nearly identical over the range from 40 to 16 KHz, and varied only by a fraction of a dB outside that.

GOOD non-oversampling DAC?

Reply #72 –
Quote from: 2Bdecided on
Quote from: oatmeal769 on
Once I'd captured, I did some reading and playing around with volume leveling and normalization.  There are two means of doing it.  You can use an average measurement of the whole track - Root/Mean Square.  This takes the overall volume average, compares it to a set point, and then changes the file to appear 'louder'.  I believe it's the same thing as is done in the loudness wars.  The file itself is changed, and dynamics are compressed or expanded to make the overall volume more consistent.


No.

Quote
The othe method is simple peak leveling.  The file is not changed as far as dynamics are concerned, rather the peak point of volume amplitude is taken, then the entire file is made to fit within a given set point.  In this instance I made each fit in at -3dB before clipping.  In this manner the file itself remains unchanged.  It is simply made so that it's loudest point is placed within the given set point.



Whether one makes a linear gain adjustment so that the average levels match, or so that the peak levels match, there is no change to the dynamic range. Matching average levels is more likely to yield two files that sound the same, because the ear tends to base its perceptions of loudness on the averages, rather than the peaks.

But let's be clear about this, it is very rare for two files to be different in the way that these two are. Usually, if you match the peaks, you usually match the averages pretty closely, and vice-versa. I've been matching levels for ABX tests for over 30 years and have not ever seen two files that differed this way, this strongly.

I'm going to presume that these samples were processed linerally with good faith. While their gains may have been altered either electronically or numerically, the alterations were presumably linear and were done in good faith. That is the simplist explanation.

That leaves us with one or two DACs that are for all intents and purposes very nonlinear and therefore broken. The difference between them has nothing to do with oversampling or non-oversampling, because DACs made either way can be very linear. I've found that to be true over the past 20 or more years and tests of literally 100s of DACs.

If light is going to be shed on any possible differences between oversampling and non-oversampling DACs, either these DACs must be repaired so that we don't get weirdness like this, or one or both DACs has to be replaced.


GOOD non-oversampling DAC?

Reply #73 –
Quote from: 2Bdecided on
"... you need to say "yes, I want to understand what I did wrong, and where I am mistaken in what I think I know".and where I am mistaken in what I think I know."
All right.  I want to understand what I did wrong, and where I'm mistaken in what I think I know.
Quote from: rpp3po on
You obviously skipped volume matching using a sine, because you thought you know better.
I'm sorry, I did not explain that I believe I had set the levels using a 1KHz wave recorded at 0dB.  (Apparently, I did it incorrectly.) It is a file which exists on an audio testing disc I have.  I did this with each DAC prior to recording the music sample.  The only change made was the DAC, I even used the same power cord, as I didn't have an extra outlet.
Quote from: rpp3po on
  • Recording of DAC1, time-synched, level-matched.
  • Recording of DAC2, time-synched, level-matched.
  • Original sample.
  • Documentation about the procedures used and exact input gain values used for each recording. The recorded samples should not be touched, except for time-synching. If you don't understand how to do the latter, upload completely untouched recordings and I'll provide time-synched versions.
Okay, I will do this, and upload the files to be time synched.  But before I make said recordings, I need to be sure that I'm capturing the samples correctly. 
As I said, I do not think I can do so with a multi-meter, unless I hold the leads to the contacts on my motherboard which is connected to the input jack.  The other way would be to take it from the plug directly (?)
I'm also hearing Mr. Krueger say that to level match via recording software is acceptible, which is what I think I did... 


There is also a request for me to post the original samples before any processing, which I can do if still wanted. (I can roughly clip them at 29 seconds)
Quote from: Arnold B. Krueger on
Matching average levels is more likely to yield two files that sound the same, because the ear tends to base its perceptions of loudness on the averages, rather than the peaks.
So then can I assume that 'peak' leveling would be the preferred method?
Quote from: Arnold B. Krueger on
I'm going to presume that these samples were processed linerally with good faith. While their gains may have been altered either electronically or numerically, the alterations were presumably linear and were done in good faith. That is the simplist explanation.
Yes, I believe this is what I did.  Everything has been done in good faith, I have not and will not intentionally manipulate(d) data to serve my points.
Quote from: Arnold B. Krueger on
If light is going to be shed on any possible differences between oversampling and non-oversampling DACs, either these DACs must be repaired so that we don't get weirdness like this, or one or both DACs has to be replaced.
I think I need to get the sampling process down first, then we'll know for sure.  Sampling vs. non-sampling I don't know, at first my curiosity is to see if there's an objectively audible difference between my ZERO and my PSA to match my subjective opinion.

First, shall I do this using software, or shall I try to do this using a Multi-Meter?
Second, I'm guessing at that point I just hit record and capture about 30 seconds or so worth.
Thirdly,  clip the file to less than 30 seconds, convert to FLAC, and upload it, correct?

Incidentally, these were made 'direct' from the DAC's, which is why I had no means of level control other than the software, and the Windows recorder.
I can use my Pre-amp / Headphone amp to control output that way.  would that be better?

GOOD non-oversampling DAC?

Reply #74 –
Would it be useful to play the 1 kHz signal through both DACs and record the output? That would allow them to be tested for linearity.