Why do some audiophiles spend thousands of dollars on a DAC? Are they searching for a "sound signature" they like, or just greater "accuracy"?

[MLXXX]

14 minutes ago, Grant Slack said:

Just want to point out that the “attenuation” doesn’t get “amplified”. The attenuation stays exactly the same from one end of the amplifier to the other, in dB.

Yes of course.  It is almost too obvious to say, but if the treble has been rolled off by a DAC, that roll-off will remain after the DAC signal is amplified, assuming of course a standard amplifier with a frequency response that is flat.

[Grant Slack]

Yes I agree it is obvious, but the wording used was open to misinterpretation.

 

Hi @Ittaku, a slow roll off filter can still avoid a HF droop, if it is an oversampling DAC.

[MLXXX]

32 minutes ago, Grant Slack said:

This shouldn’t be happening with an over-sampling DAC. It is certainly not a necessary consequence of using 44.1k sampled data for storage.

 

Assuming your measurements are OK, and the unit is performing to spec, and the designers were competent, then this is deliberate tailoring, IMO.

 

23 minutes ago, Ittaku said:

Does the DAC concerned use a "slow roll-off filter"? By design they roll off at the top of the audible range.

 

The FCA1616 audio interface provides 8 balanced DAC outputs. I chose 2 of these on my unit at random for my tests. (I never use this interface for listening to music per se, only for making and quickly reviewing recordings of live music.) I've never looked into the DAC filter design, e.g. whether it uses a slow roll-off filter.  (I've seen largely favourable comments on the net about the DAC performance.)

 

I might try to get an answer by practical means, e.g. using an ADC operating with a 96kHz sample rate to see how the reconstruction filters of the FCA1616 DACs seem to apply for a sample rate of 44.1kHz.

 

It's interesting to review the results I found with kukynas's test recordings. No "slow roll-off" filtering apparent there! :-

	RME	RMS level at 1000 Hz	RMS level at 11025Hz	dB down at 11025Hz relative to level at 1000Hz
	Left	-9.08159	-9.08713	-0.00554
	Right	-9.07955	-9.08327	-0.00372
	Stereo	-9.07955	-9.0852	-0.00565
	TOPPING	RMS level at 1000 Hz	RMS level at 11025Hz	dB down at 11025Hz relative to level at 1000Hz
	Left	-9.23363	-9.28215	-0.04852
	Right	-9.23142	-9.28033	-0.04891
	Stereo	-9.23252	-9.28124	-0.04872

 

Only a 20th of a dB attenuation at 11.025kHz with the Topping DAC [incorporating the effect if any of the RME ADC].  Even less apparent attenuation with the RME DAC.

[Stereophilus]

Interesting that this 20th of a dB is nonetheless audible, albeit in a minor way.

[aussievintage]

1 hour ago, Stereophilus said:

Interesting that this 20th of a dB is nonetheless audible, albeit in a minor way.

 

You cannot conclude this is the slight difference we are hearing.  In fact, I'll bet it isn't

[Stereophilus]

40 minutes ago, aussievintage said:

 

You cannot conclude this is the slight difference we are hearing.  In fact, I'll bet it isn't

Just play the 11kHz test tones (not the music) back to back a few times. There is an audible difference.

[MLXXX]

1 hour ago, Stereophilus said:

Just play the 11kHz test tones (not the music) back to back a few times. There is an audible difference.

Kukynas's files as supplied by him have not been normalised to the same reference level at 1kHz. 

 

Below are measurements I've done using the RMS analysis tool available in Audacity.  (To load this tool, use the add plug-in option under the Analysis tab.)  The RMS measurements can vary in the last two decimal places depending on exactly how much of the waveform you select.

 

The Topping file is 0.153dB quieter at 1kHz, and 0.196dB quieter at 11.025kHz than the RME file.  0.196dB can be rounded to 0.2dB or a fifth of a decibel.

 

(For my personal listening tests I normalised kukynas's files in their Left and Right channels to match the 1kHz level in file 1. This can be achieved by using the split stereo file option in Audacity at the start of an Audacity  track, selecting the track content for the relevant channel, and then using the effect "amplify", set for the appropriate amount of amplification.  An easier and automatic method would be to use foobar 2000 with ABX plug-in and allow it to normalise the files, though foobar might not give exactly the right gain correction for these particular files. ) 

 

	RME	RMS level at 1000 Hz	RMS level at 11025Hz	RMS level discrepancy at 11025Hz relative to 1000Hz, in dB
	Left	-9.08159	-9.08713	-0.00554
	Right	-9.07955	-9.08327	-0.00372
	Stereo	-9.07955	-9.0852	-0.00565
	TOPPING	RMS level at 1000 Hz	RMS level at 11025Hz	RMS level discrepancy at 11025Hz relative to 1000Hz, in dB
	Left	-9.23363	-9.28215	-0.04852
	Right	-9.23142	-9.28033	-0.04891
	Stereo	-9.23252	-9.28124	-0.04872
	FILE 1	RMS level at 1000 Hz	RMS level at 11025Hz	RMS level discrepancy at 11025Hz relative to 1000Hz, in dB
	Left	-9.03091	-9.03106	-0.00015
	Right	-9.03091	-9.03106	-0.00015
	Stereo	-9.03091	-9.03106	-0.00015
		MISMATCH	RMS stereo level at 1k Hz	RMS level discrepancy at 1000Hz compared with file 1, in dB
		FILE 1	-9.03091	0
		RME	-9.07955	-0.04864
		TOPPING	-9.23252	-0.20161

[aussievintage]

11 hours ago, Stereophilus said:

Just play the 11kHz test tones (not the music) back to back a few times. There is an audible difference.

 

Then something else is changing.  You cannot hear 1/20th of a db

[Stereophilus]

10 hours ago, aussievintage said:

 

Then something else is changing.  You cannot hear 1/20th of a db

Fair enough.  Where to from here then? The pub?

[aussievintage]

2 hours ago, Stereophilus said:

Fair enough.  Where to from here then? The pub?

might as well 

[andyr]

12 hours ago, aussievintage said:

 

Then something else is changing.  You cannot hear 1/20th of a db

 

 

Just like you can't hear a 0.1 mm raise at the arm pivot?  

 

Andy

 

[aussievintage]

10 hours ago, andyr said:

 

Just like you can't hear a 0.1 mm raise at the arm pivot?  

 

Andy

 

Don't be a twerp

[MLXXX]

11 hours ago, andyr said:

Just like you can't hear a 0.1 mm raise at the arm pivot? 

Recording the cartridge preamp output of a track on an LP for multiple playings would deliver variations that might swamp other variables such as arm pivot height. A minor amount of record warp, a little bit of of bounce of the stylus in the groove, and no two playings would deliver exactly the same waveform captures by an ADC.  (People sometimes refer to groove wear but I understand that should be minimal with a non-worn stylus and correct stylus weight and anti-skating.)

 

And then there is the question of audibility of a very small arm pivot height adjustment. More likely to be possible if able to do an A B comparison between two time-aligned ADC recordings of the cartridge preamp output. 

[andyr]

4 hours ago, aussievintage said:

Don't be a twerp

 

Have no problem being called a 'twerp', av.  

 

But I would've been extremely annoyed if you'd called me a 'woke twerp'!  

 

Andy

 

[muon*]

level matching isn't such a big deal then.

[MLXXX]

6 hours ago, muon* said:

level matching isn't such a big deal then.

 

There may have been some confusion.  Let me try to clarify.

 

1. The discrepancy in level of the lead-in tone  between the raw files kukynas supplied for the RME and Topping was just under 0.2dB and under very good conditions such a difference in level could, of itself, be just audible for some listeners. (A lot of people though wouldn't be able to hear such a small difference in amplitude even with a continuous tone and even with immediate A B switching.   And older people might not be able to hear the 11.025kHz tone at all.)

 

2. However, if kukynas's files are adjusted in gain so as to play at exactly the same level at 1kHz, the discrepancy in amplitude at 11.025kHz falls to less than a 20th of a dB, and that of itself would not be audible.

 

3. Listening only to the lead-in tone at 11.025kHz is no substitute for listening to the whole file. There could be audible differences somewhere other than at the start of each "round trip file".

 

Edited November 27, 2020 by MLXXX

[muon*]

27 minutes ago, MLXXX said:

There may have been some confusion.  Let me try to clarify.

I forgot to add this

[Guest DrSK]

On 22/11/2020 at 6:12 PM, MLXXX said:

I'd note that In the case where the DAC is the centre of interest, one would be more likely to measure the electrical signal emerging from it directly, rather than look later in the reproduction chain after a power amplifier, speakers, room acoustics and a microphone, all of which would affect the sound and could make it harder to identify performance issues of the DAC itself.

This is what I used to do 20 years ago in military spec telecommunications for analogue and digital devices. 

 

Set up automated test beds of client and competitor product. 

 

Pass real audio signals and compare outputs to the input. Start by stripping off acoustic transducers and just use electrical signals in and out 

 

You can then add in each stage  so you get the effect of each component in the system.

 

Real signals were more useful than frequency sweeps and tone tests beyond the basics and you could relate things better to customer feedback and whether what they heard was real and conversely whether measurable differences could be heard. 

 

We also did this to the front end, eg remove the mic, feed an electrical signal in and test all steps of the chain. 

And with a good system for play back you can obviously test the results with a group of listeners. The focus for the testing I did with a group of listeners was focused on intelligibility. 

 

 

 

[Guest DrSK]

On 21/11/2020 at 6:07 PM, LHC said:

 

If you narrow down the key thing to being the reverberant fields from the two amps, that would make comparative measurement a lot easier. 

Sorry, I'm not following. 

 

The reverberant field is a function of the room and energy output, and by definition so muddled in time response I'm not understanding how this would relate to perception of image depth between the two amps?

 

Eg my early Metaxas Soliloquy and and my Electrocompaniet AW180 monos are measuring very very close in low frequency energy level, but the Metaxas has crisper base. I don't see how measuring the reverberant field, which by definition is time smeared, would show any differences? 

 

 

[Guest DrSK]

On 21/11/2020 at 6:07 PM, LHC said:

 

If you narrow down the key thing to being the reverberant fields from the two amps, that would make comparative measurement a lot easier. 

Sorry, I'm not following. 

 

The reverberant field is a function of the room and energy output, and by definition so muddled in time response I'm not understanding how this would relate to perception of image depth between the two amps?

 

Eg my early Metaxas Soliloquy and and my Electrocompaniet AW180 monos are measuring very very close in low frequency energy level, but the Metaxas has crisper base. I don't see how measuring the reverberant field, which by definition is time smeared, would show any differences? 

 

PLEASE REMOVE. REFRESH STUFF UP. 

 

 

Edited November 30, 2020 by DrSK

[Guest DrSK]

On 21/11/2020 at 6:07 PM, LHC said:

 

If you narrow down the key thing to being the reverberant fields from the two amps, that would make comparative measurement a lot easier. 

Sorry, I'm not following. 

 

The reverberant field is a function of the room and energy output, and by definition so muddled in time response I'm not understanding how this would relate to perception of image depth between the two amps?

 

Eg my early Metaxas Soliloquy and and my Electrocompaniet AW180 monos are measuring very very close in low frequency energy level, but the Metaxas has crisper base. I don't see how measuring the reverberant field, which by definition is time smeared, would show any differences? 

PLEASE REMOVE. REFRESH STUFF UP. 

 

Edited November 30, 2020 by DrSK

[Guest DrSK]

On 21/11/2020 at 6:07 PM, LHC said:

 

 

On 22/11/2020 at 3:25 AM, MLXXX said:

I think further explanation is needed here as to why higher sample rates are considered necessary in some circumstances.

 

A continuous 30Hz signal that is a pure sinewave unvarying in amplitude and with no jitter, that is present at the output terminals of an amplifier, could be successfully captured with an ADC set for an extremely low sample rate such as, for example, 100Hz. On playback of the capture,  a DAC would reconstruct the waveform as a smooth analogue sine wave at 30Hz. Perfection!

 

Reasons for needing to use a much higher sample rate l(such as 48kHz) for a "30Hz signal" might include:

 

• the waveform has a repetition rate of 30Hz but is a sawtooth, or other non-sinusoidal shape, with harmonics that extend to very high frequencies
• the nominal 30Hz waveform is creating a standing wave pattern, in a room, that is highly unstable because of factors such as flexing of the walls of the room, or rapid changes in the composition of the air in the room, and a microphone is being used to capture the nominal 30Hz waveform
• the waveform is being launched from the cone of a bass driver in a speaker enclosure, leading to so-called doppler effects and other distortions
• the waveform is subject to small rapid irregular phase shifts because the DAC clock is continually hunting for a lock rather than smoothly reaching a stable equilibrium average clock rate to match the clock rate of the incoming stream.

 

All of the above are contaminations or distortions of some kind, relative to a pure 30Hz sine wave.

The latter is closer to the measurement I was referring to. 

 

Had 30Hz single cycles with small uneven delays between them. The delay was a function of amplitude and there was non-linear feedback affecting the next 30Hz cycle. The feedback also switched between positive and negative gain. 

 

Edited November 30, 2020 by DrSK

[Guest DrSK]

On 21/11/2020 at 2:26 PM, Grant Slack said:

 

 

I also get frustrated by completely wrong descriptions about digital audio like the above.  “From experience”?

 

(To anyone else reading the above, I just want to assure you that the 22 kHz audio generated from a 44.100 kHz digital process, is perfect. There is in no sense a need to ‘plot’ the shape of the analog audio waveform with sample points, as the above paragraph suggests. The wording, and the 10x factor mentioned, implies that CD audio can only reproduce audio well up to about 4 kHz, and above that it is less and less accurate, relying on gross interpolation. Nothing could be further from the truth. But don’t wait for DrSK and I to win you over with debating skills (which I am disinclined to do, for the umpteenth time over the years): just pick up a DA textbook. It should be somewhere near to the beginning.)

 

The only valid reason for higher bitrates [edit: as a result of sampling at a higher frequency], from the listener perspective, is to avoid having a fraction of a dB of attenuation in the 18-20 kHz region, caused by the anti-aliasing filter at 22 kHz. And it would be highly debatable whether that fraction of a dB is even audible, and compared to how one’s tweeter probably behaves in that region, irrelevant. But, just to get out of debating it, let’s use 48 kHz, and the question becomes moot.

 

From a non-audio perspective, it is also much cheaper to make high-bitrate DACs with sigma-delta technology, than low-bitrate DACs with multi-bit ladder technology.

 

Modern DACs are so good that the idea we can hear their imperfections is quite unhelpful. The only explanations for imperfections reaching audible levels are bad design, bad construction, or a misguided attempt to tailor the frequency response, in either digital or analog domains.

 

cheers

Grant

 

You could well be right in terms of audible differences. My expertise is less so on digital sampling. More on acoustic aspects. 

 

I found it simpler to just oversample, as I wouldn't miss anything.  Nothing worse than analysing data then finding you've missed something. 

 

And it made energy calcs more robust as you can get true RMS across the frequency range without having to assume sine waves. Ie, for what I was doing it matched analogue measurements, at Nyquist limit things don't. 

 

 

[LHC]

1 hour ago, DrSK said:

Sorry, I'm not following. 

 

The reverberant field is a function of the room and energy output, and by definition so muddled in time response I'm not understanding how this would relate to perception of image depth between the two amps?

 

Eg my early Metaxas Soliloquy and and my Electrocompaniet AW180 monos are measuring very very close in low frequency energy level, but the Metaxas has crisper base. I don't see how measuring the reverberant field, which by definition is time smeared, would show any differences? 

 

 

Since you are replying to some post from a few weeks ago, that may have caused the confusion. I recalled it was you who first mentioned reverberant field back then, while was going on about measuring the entire sound field, not just the reverberant component. 

Edited November 30, 2020 by LHC

[Guest DrSK]

1 hour ago, LHC said:

Since you are replying to some post from a few weeks ago, that may have caused the confusion. I recalled it was you who first mentioned reverberant field back then, while was going on about measuring the entire sound field, not just the reverberant component. 

May have both confused each other. 

 

In terms of depth perception I was referring to reverb and reflections/delays in the recording.