Audibility thresholds of amp and DAC measurements

[xr100]

Quite the opposite - the essence. When the output of a DUT is corrupted - it is corrupted, or you need to insert some DSP between amplifier and headphones/speakers. For a listener this corruption is final/irreversible.

(1) Spectrum of triangle wave:

(2) Spectrum of triangle wave after modification by an all-pass filter:

(3) Spectrum of triangle wave after modification by an all-pass filter with an inverse phase response to the previous:

(4) Invert (3), add to (1):

The eagle-eyed might have noticed a tell-tale sign that the images for (1), (2), and (3) are actually the exact same image file/screen capture. Indeed they are, and that's because all three spectral plots were, of course, identical.

 

[pkane]

(1) Spectrum of triangle wave:

View attachment 46595

(2) Spectrum of triangle wave after modification by an all-pass filter:

View attachment 46595

(3) Spectrum of triangle wave after modification by an all-pass filter with an inverse phase response to the previous:

View attachment 46595

(4) Invert (3), add to (1):

View attachment 46596



The eagle-eyed might have noticed a tell-tale sign that the images for (1), (2), and (3) are actually the exact same image file/screen capture. Indeed they are, and that's because all three spectral plots were, of course, identical.

I’m not sure I understand the point of this illustration. You’re obviously doing some DSP to match the delayed version. Isn’t that what Serge was saying you could do if you do use DSP?

 

[Serge Smirnoff]

(1) Spectrum of triangle wave:

View attachment 46595

(2) Spectrum of triangle wave after modification by an all-pass filter:

View attachment 46595

(3) Spectrum of triangle wave after modification by an all-pass filter with an inverse phase response to the previous:

View attachment 46595

(4) Invert (3), add to (1):

View attachment 46596



The eagle-eyed might have noticed a tell-tale sign that the images for (1), (2), and (3) are actually the exact same image file/screen capture. Indeed they are, and that's because all three spectral plots were, of course, identical.

All-pass filter has flat FR, so magnitudes of spectral components are not changed. Phases of these components are changed. This results in change of the form of the signal in time domain. If you compute the same spectra in complex values (with phases) the distortion of the signal becomes visible in frequency domain too. Df level for the signal accounts all types of distortion, so, in this case it shows the amount of waveform distortion caused by phase modification/inaccuracy. Df level computed in time domain is math. equal to the one computed in frequency domain in complex values.

And now a simple question. What plot one should use for research of distortion in this case:
(1) in time domain, where distortion of waveform is clearly visible?
(2) in freq. domain magnitudes only, where such distortion is not visible?
(3) in freq. domain complex spectra, where such distortion is visible?

 

[solderdude]

all of the above.
One should look in all measurements.
One could also look at the null and disect what from the null comes from what aspect and then weigh depending on audibility tests what is objectionable or pereived sound degrading.
That basically is just a different road leading to the same results but using a difference signal + source signal to further analyze.

I agree that using real music may shed some lights that may remain 'hidden' when looking at pure tones that are not testing the outer boundaries of the audible range.


Question... is there are relation between 'regular test tone' performance and music performance when excluding the lower and higher frequencies ?
Are there things (band limited in the same manner) that pop out as being substantially different between test signals and music (in good performing gear) ?

 

[xr100]

I’m not sure I understand the point of this illustration. You’re obviously doing some DSP to match the delayed version. Isn’t that what Serge was saying you could do if you do use DSP?

DSP was used (FIR filters.) The point is the absence of new frequency components and the reversibility of the operation, being a linear system.

 

[j_j]

You can choose on this page the devices of your interest, I will compute df levels for them from supplied recordingd. Also you can do this yourself but I'm ready to help.

I am sorely tempted to send you a couple of samples. They will all be synthetic, never passing through analog means. But they ARE synthetic. Of course I'd send both original and corrupted.

 

[Serge Smirnoff]

Question... is there are relation between 'regular test tone' performance and music performance when excluding the lower and higher frequencies ?
Are there things (band limited in the same manner) that pop out as being substantially different between test signals and music (in good performing gear) ?

Not sure I got you correctly, is your question about correlation of df measurements with various t-signals to the measurements with m-signal?

 

[Serge Smirnoff]

I am sorely tempted to send you a couple of samples. They will all be synthetic, never passing through analog means. But they ARE synthetic. Of course I'd send both original and corrupted.

You are welcome. Don't forget to formulate the question in regards to your test vectors.

 

[solderdude]

when you mean t-signals = test signals and m-signals = music then yes.
And if the correlation is better when lowest and highest frequency differences of the audibe range are not included in weighing of the music signals.

Just curious.. I could probably look myself but am lazy this way and rather ask to someone that will have looked for this.

 

[Serge Smirnoff]

Sorry, t-signal means technical signal, m-signal - music signal.

And if the correlation is better when lowest and highest frequency differences of the audibe range are not included in weighing of the music signals.

I don't know. I think the distortion in output music signal is already weighted to some extent according to particular spectra of that m-signal. Usually high freq. components of m-signal have lower energy. So, their distortion is less accounted by null naturally.

 

[xr100]

And now a simple question. What plot one should use for research of distortion in this case:

(1) in time domain, where distortion of waveform is clearly visible?
(2) in freq. domain magnitudes only, where such distortion is not visible?
(3) in freq. domain complex spectra, where such distortion is visible?

For distortion, (2).

To quote from Douglas Self's "Audio Power Amplifier Design Handbook":

"Phase and group delay have been an area of dispute for a long time. As Stanley Lipshitz et al. have pointed out, these effects are obviously perceptible if they are gross enough; if an amplifier was so heroically misconceived as to produce the top half of the audio spectrum 3 hours after the bottom, there would be no room for argument. In more practical terms, concern about phase problems has centered on loudspeakers and their crossovers, as this would seem to be the only place where a phase shift might exist without an accompanying frequency-response change to make it obvious. […] This controversy is of limited importance to amplifier designers, as it would take spectacular incompetence to produce a circuit that included an accidental all-pass filter. Without such, the phase response of an amplifier is completely defined by its frequency response, and vice versa; in Control Theory this is Bode’s Second Law, and it should be much more widely known in the hi-fi world than it is. A properly designed amplifier has its response roll-off points not too far outside the audio band, and these will have accompanying phase shifts; there is no evidence that these are perceptible."

(Slight caveat--Self's book is not a definitive treatise on questions of audibility.)

 

[Serge Smirnoff]

For distortion, (2).

To quote from Douglas Self's "Audio Power Amplifier Design Handbook":

"Phase and group delay have been an area of dispute for a long time. As Stanley Lipshitz et al. have pointed out, these effects are obviously perceptible if they are gross enough; if an amplifier was so heroically misconceived as to produce the top half of the audio spectrum 3 hours after the bottom, there would be no room for argument. In more practical terms, concern about phase problems has centered on loudspeakers and their crossovers, as this would seem to be the only place where a phase shift might exist without an accompanying frequency-response change to make it obvious. […] This controversy is of limited importance to amplifier designers, as it would take spectacular incompetence to produce a circuit that included an accidental all-pass filter. Without such, the phase response of an amplifier is completely defined by its frequency response, and vice versa; in Control Theory this is Bode’s Second Law, and it should be much more widely known in the hi-fi world than it is. A properly designed amplifier has its response roll-off points not too far outside the audio band, and these will have accompanying phase shifts; there is no evidence that these are perceptible."

(Slight caveat--Self's book is not a definitive treatise on questions of audibility.)

Sound is a waveform (in time domain). Distortion of a sound is distortion of the waveform first of all. This is fundamental. If further analysis of the distortion in freq. domain reveals that some types of distortion are less important from hearing perspective (?) then it might be reasonable to give them different terms. But anyway, this distinction is not fundamental. Its validity is limited by the framework of your analysis/method. In df-metric I analyze the waveform mostly with statistical methods and so use the general definition of the term "distortion". Such broad definition of the term is fundamental because it refers to the fundamental fact that sound is a waveform (of pressure or voltage).

 

[xr100]

Such broad definition of the term is fundamental because it refers to the fundamental fact that sound is a waveform (of pressure or voltage).

There are 100,000s of words in the English dictionary. I am certain that suitable alternatives exist, the use of which does not require adherence to a kind of postmodernist epistemology.

 

[xr100]

OK, might as well try out this "DF Metric" as currently implemented in (the nifty bit of software that is) "DeltaWave"...

Reference file--(16-bit/44.1kHz PCM) 80's pop music, modern master, looks like some "brickwall" peak limiting was used for the re-release.

First test--A 16x oversampled "soft" clipping (static waveshaper) process with no action until ~-4dB. "DF Metric" = -21.8dB.
Second test--8th-order all-pass filter. "DF Metric" = -3.7dB. (BUT see error message below!)

I had actually intended for the "soft" clipping to be an example of a process the audibility of which would be somewhat obviated by only acting on the transient peaks of percussive sounds, particularly kick drums. Alas, the choice of file with its brickwall peak limiting meant that the clipper's action must have affected rather more of it, because "tonal" sounds--vocals, etc.--are harsh.

The 8th-order all-pass filter processed file sounds awful (interestingly reminiscent of vinyl?)

I have no idea what to make of the "DF Metric" results.

As mentioned, the error message for the all-pass filter processed file:

 

[pkane]

OK, might as well try out this "DF Metric" as currently implemented in (the nifty bit of software that is) "DeltaWave"...

Reference file--80's pop music, modern master, looks like some "brickwall" peak limiting was used for the re-release.

First test--A 16x oversampled "soft" clipping (static waveshaper) process with no action until ~-4dB. "DF" metric = -21.8dB.
Second test--8th-order all-pass filter. "DF" metric = -3.7dB. (BUT see error message below!)

I had actually intended for the "soft" clipping to be an example where the audibility would be somewhat obviated by only acting on the transient peaks of percussive sounds, particularly kick drums. Alas, the choice of file with its brickwall peak limiting meant that the clipper's action must have affected rather more of it, because "tonal" sounds--vocals, etc.--are harsh.

The 8th-order all-pass filter processed file sounds awful (interestingly reminiscent of vinyl?)

I have no idea what to make of the "DF" metric results.

As mentioned, the error message for the all-pass filter processed file:

View attachment 46665

Upload the reference and the distorted files or post the details from DeltaWave results tab.

This error is displayed when DW can’t find a solution to the clock drift that produces a residual error of less than 5 samples, which is huge.

 

[xr100]

This error is displayed when DW can’t find a solution to the clock drift that produces a residual error of less than 5 samples, which is huge.

OK. Then what happens when clicking on "YES"?

I've performed another test using Softube's "Trident A-Range" plug-in with the "SATURATION" setting at the middle position. (It's an "emulation" of a hardware EQ so I've left all controls at 0dB and turned off LP/HP filter options, although it probably won't measure flat. Also, it probably uses internal upsampling to keep aliasing at bay, resampled back to base rate on returning the audio to the host software.) No idea what non-linear process happens internally to the plug-in; however, Softube's key patent* basically relates to the dynamic selection of waveshapers depending on the system state. "DF Metric" = -23.4dB.

I've pasted the results from DeltaWave for this and the "soft clip" process below. The "Trident A-Range" processed file sounds very different to the "soft clipping" processed file.

(*"System and method for simulation of non-linear audio equipment" -- Assignee: Softube AB.)


Results: DeltaWave v1.0.50, 2020-01-20T16:58:12.2132149+00:00 Reference: Reference.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Comparison: Softube Trident.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Settings: Gain:True, Remove DC:True Non-linear Gain EQ:False Non-linear Phase EQ: False EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB Correct Drift:True, Precision:30 Non-Linear drift Correction:False Upsample:False, Window:Hann Spectrum Window:Hann, Spectrum Size:32768 Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048 Dither:False Trim Silence:False Enable Simple Waveform Measurement: False Discarding Reference: Start=0s, End=0s Discarding Comparison: Start=0s, End=0s Initial peak values Reference: -0.028dB Comparison: 0.133dB Initial RMS values Reference: -12.546dB Comparison: -12.979dB Null Depth=34.693dB X-Correlation offset: 0 samples Drift computation quality, #1: Very Good (2.58μs) Trimmed 0 samples ( 0.00ms) front, 0 samples ( 0.00ms end) Final peak values Reference: -0.028dB Comparison: 0.554dB Final RMS values Reference: -12.546dB Comparison: -12.578dB Gain= 0dB (1x) DC=0 Phase offset=0.009089ms (0.401 samples) Difference (rms) = -32.85dB [-36.23dBA] Correlated Null Depth=55.68dB [44.49dBA] Clock drift: 0.01 ppm Files are NOT a bit-perfect match (match=1.1%) at 16 bits Files are NOT a bit-perfect match (match=0.85%) at 32 bits Files match @ 50.0014% when reduced to 5.5 bits ---- Phase difference (full bandwidth): 66.2367741670488° 0-10kHz: 4.41° 0-20kHz: 63.69° 0-24kHz: 66.24° Timing error (rms jitter): 4.9μs RMS of the difference of spectra: -77.4629143388485dB DF Metric (step=400ms, overlap=0%): Median=-23.4dB Max=-6.7dB Min=-36.2dB 1% > -35.74dB 10% > -24.51dB 25% > -24.01dB 50% > -23.41dB 75% > -22.75dB 90% > -21.71dB 99% > -18.67dB gn=1.0000000584211, dc=0, dr=7.535E-09, of=0.4008309672 DeltaWave v1.0.50, 2020-01-20T17:03:52.2972309+00:00 Reference: Reference.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Comparison: Soft Clip.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Settings: Gain:True, Remove DC:True Non-linear Gain EQ:False Non-linear Phase EQ: False EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB Correct Drift:True, Precision:30 Non-Linear drift Correction:False Upsample:False, Window:Hann Spectrum Window:Hann, Spectrum Size:32768 Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048 Dither:False Trim Silence:False Enable Simple Waveform Measurement: False Discarding Reference: Start=0s, End=0s Discarding Comparison: Start=0s, End=0s Initial peak values Reference: -0.028dB Comparison: 0.439dB Initial RMS values Reference: -12.546dB Comparison: -12.532dB Null Depth=22.897dB X-Correlation offset: -4 samples Drift computation quality, #1: Very Good (3.21μs) Trimmed 0 samples ( 0.00ms) front, 0 samples ( 0.00ms end) Final peak values Reference: -0.028dB Comparison: 0.5dB Final RMS values Reference: -12.546dB Comparison: -12.525dB Gain= 0dB (1x) DC=-0.00001 Phase offset=-0.097199ms (-4.286 samples) Difference (rms) = -31.24dB [-33.72dBA] Correlated Null Depth=44.4dB [40.17dBA] Clock drift: 0.01 ppm Files are NOT a bit-perfect match (match=0.16%) at 16 bits Files are NOT a bit-perfect match (match=0%) at 32 bits Files match @ 50.0072% when reduced to 5.24 bits ---- Phase difference (full bandwidth): 72.1442638643905° 0-10kHz: 6.40° 0-20kHz: 66.71° 0-24kHz: 72.14° Timing error (rms jitter): 1.3ms RMS of the difference of spectra: -76.3933324582667dB DF Metric (step=400ms, overlap=0%): Median=-21.8dB Max=-6.5dB Min=-38dB 1% > -35.93dB 10% > -22.88dB 25% > -22.43dB 50% > -21.82dB 75% > -21.18dB 90% > -20.23dB 99% > -17.68dB gn=0.99999996205188, dc=-6.54159689219462E-06, dr=9.411E-09, of=-4.2864698309

 

[xr100]

Upload the reference and the distorted files or post the details from DeltaWave results tab.

This error is displayed when DW can’t find a solution to the clock drift that produces a residual error of less than 5 samples, which is huge.

And the results for the 8th-order All-Pass Filter:

DeltaWave v1.0.50, 2020-01-20T17:17:12.3255467+00:00 Reference: Reference.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Comparison: 8th Order AP Filter.wav[L] 10469928 samples 44100Hz 64bits, stereo, MD5=00 Settings: Gain:True, Remove DC:True Non-linear Gain EQ:False Non-linear Phase EQ: False EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB Correct Drift:True, Precision:30 Non-Linear drift Correction:False Upsample:False, Window:Hann Spectrum Window:Hann, Spectrum Size:32768 Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048 Dither:False Trim Silence:False Enable Simple Waveform Measurement: False Discarding Reference: Start=0s, End=0s Discarding Comparison: Start=0s, End=0s Initial peak values Reference: -0.028dB Comparison: 4.151dB Initial RMS values Reference: -12.546dB Comparison: -11.792dB Null Depth=13.117dB X-Correlation offset: -156 samples Residual error too large 5.19394123. Trying alternative drift correction method Drift computation quality #2: Very Good (3.5μs) Trimmed 0 samples ( 0.00ms) front, 0 samples ( 0.00ms end) Final peak values Reference: -0.028dB Comparison: 3.135dB Final RMS values Reference: -12.546dB Comparison: -12.677dB Gain= 0.4147dB (1.0489x) DC=0.00008 Phase offset=-3.537415ms (-156 samples) Difference (rms) = -13.17dB [-13.09dBA] Correlated Null Depth=21.13dB [31.85dBA] Clock drift: -0.01 ppm Files are NOT a bit-perfect match (match=0.03%) at 16 bits Files are NOT a bit-perfect match (match=0%) at 32 bits Files match @ 49.9985% when reduced to 2.32 bits ---- Phase difference (full bandwidth): 102.76506643725° 0-10kHz: 101.46° 0-20kHz: 102.41° 0-24kHz: 102.77° Timing error (rms jitter): 7.8ms RMS of the difference of spectra: -67.8216708011731dB DF Metric (step=400ms, overlap=0%): Median=-3.7dB Max=-0.4dB Min=-10dB 1% > -9.01dB 10% > -5.68dB 25% > -4.59dB 50% > -3.68dB 75% > -2.91dB 90% > -2.12dB 99% > -1.07dB gn=0.953378748051051, dc=7.74697922495239E-05, dr=-1.1484499646724E-08, of=-155.99999998926

 

[pkane]

OK. Then what happens when clicking on "YES"?

By default, DeltaWave computes clock drift by measuring offsets over multiple points and fitting a curve to it. In the case when this default method produces a large error, the offset is measured at the beginning and the end of the track and a simple line is fit through those two points to try to get a simple linear drift fit.

 

[Serge Smirnoff]

@xr100: Please, provide reference and processed files in some lossless format. I will also check them.

 

[xr100]

@xr100: Please, provide reference and processed files in some lossless format. I will also check them.

OK, I suppose an excerpt would suffice (in view of copyright) rather than the entire track?