I've previously shown that my SA-11s2 and Panasonic UB9000 sound similarly and measure similarly with DeltaWave.
In this post, I'm sharing some preliminary data showing
a) a measurable PK Metric difference between two reasonable DACs with real music
b) the DAC with a ~10 dB worse 1 kHz SINAD unit seems to be more accurate to the recording using PK Metric!
Please consider this preliminary data, since I am hoping that others will try to reproduce the results with their own gear and own music too. There is an excess of data presented to "show my work" since this is an unusual result.
The reasonable DACs? The Topping D90 MQA and Marantz SA-10. Both have been characterized by @amirm. The original D90 is still reference class by today's standards except for XLR phase inversion from convention, and the SA-10 does not measure as well, but to be fair, is a 2016-era product which was in development since 2014 or so and still "should" be good enough to threshold of audibility.
My 2018-production SA-10 does give slightly better PCM 1 kHz SINAD results than Amir's unit, though the test environment is different, and I got these results for DSD64. My Topping D90 gets to the -120dB 1kHz SINAD, so it's not defective either. I kept the D90 away from other transformers.
Test Track / Methodology
Concerto symphonique No. 4 in D minor, Op. 102: Scherzo from SFS0060 - Masterpieces in Miniature (Physical SACD) & DSD extract
I measured the DSD playback on the Topping D90 versus physical disc playback on the Marantz SA-10.
E1DA Cosmos ADC was running 32/176 to capture everything the SACD should reproduce. XLR balanced cables were used and L/R channel was verified each time before recording. For the SA-10, I played the disc, and then pressed the previous track button on the remote and then recorded using Audacity. For the D90, I used Foobar/ASIO4All to send a DSD stream to the D90. I confirmed that the LCD display indicated a 2.8MHz sampling rate, ensuring that my software configuration was valid for transmitting DSD. I pressed play in Foobar and then quickly used ALT-TAB and SHIFT-R to start the recording. When the 1 minute mark on the SACD player or the Foobar status bar was reached, I stopped the recording.
Control Experiments
This shows how repeatable the source devices are, and how repeatable the test environment is. The E1DA Cosmos ADC itself is subject to thermal effect. The PK Metric is super low, indicating the testing conditions are pretty good.
Comparison
Now I compare the Topping D90 as reference against the Marantz SA-10
- the phase is inverted on the D90; DeltaWave corrects for these
- DeltaWave will treat the "reference" as reference and then modify the comparison file. Since the SA-10 outputs a higher voltage than the D90 (2dB worth), DeltaWave will reduce the volume of the SA-10 recording to the D90. This might help the SA-10 by helping to reduce its inherent noise. I'm not sure. (@pkane?)
Quality of the Match
I crop out the first 6 seconds and last 10 seconds of the recording since it throws DeltaWave off. Here are the files as-is, showing correct phase matching between the two recordings.
and the final match
We can zoom in and see how tight/good DeltaWave's level matching and alignment are. Blue is the Topping D90 and White is the Marantz SA-10. The increased noise and increased ultrasonics of the SA-10 are clearly seen in these ultra-zoomed-in views.
The aligned spectrum shows a slight roll-off in the bass for the SA-10 below 20Hz and the higher noise level in the high frequencies which does drop down into the audible range. The big difference is the higher ultrasonic noise of the Marantz SA-10.
Looked at a different way, we can see the delta of the Spectra. Here, you can see that the ultrasonics are really the main difference between the two files. No surprises yet. The spectra looks at the recording as a whole, not for specific portions of the music.
I don't know how best to interpret the phase, but I've shown it here. It seems like everything is pretty close to 0 degrees in the audible range, ignoring the bass I mentioned.
The important slide: PK Metric of -56.9 dBr rms and -48.4 dBFS for the transients that generate the biggest difference. -48.4 dBFS meets the PK metric threshold for audible.
This suggests that both sound similar but during that transient, it should be possible to hear a difference between the two.
(and a close-up of the spike at ~16 seconds shows that there's a lot going on in the audible frequencies. Note that I clip the first 6 seconds of the recording of you want to listen to the music yourself.
The PK Metric is a perceptually weighted value that tries to capture if a null comparison should be or should not be audible. Greater than -50 dB means that it's clearly audible. At -56.9 dBr, it's probably still really hard to hear the difference, but if you use the cool split window feature that @pkane put in, and hold CTRL and then move your mouse on the left to reach the area of the peak spikes around 15 seconds, you can see that the level there is -48.4 dBFS and in the audible frequency range.
So, here, we have a flagship DAC and a flagship SACD player that *should* get into the range of indistinguishable, but the PK Metric is showing us that there may actually be a difference.
Different, yes. But which is correct? Maybe, the Marantz!
For true digital comparisons, I took the DSD source file and then used the TASCAM Hi-Res Editor to convert the DSD 2.8MHz file into a 32-bit FP / 176 kHz PCM version. 176 kHz is how I made my E1DA recordings. I then redid the DeltaWave tests, using that Digital Source as the reference.
Referenced to the digital source file
Topping D90 gets a PK Metric of -73.3 dBr (rms)
Marantz SA-10 gets a PK Metric of -80.3 dBr (rms)
Let's look at the details because if anything, the Marantz performance is under-estimated.
Here are the RAW waveforms before matching
and then after matching super zoomed in
and the PK Metric is -80.3 dBr and you can see a good amount of time is spent below -80 dB.
The big spike at 35 seconds? It's obvious in the delta, but not obvious in the source recording.
The matched spectra shows the effect of the digital filter used by TASCAM. The slope in the 30 to 40 kHz range looks similar.
And here is the Topping D90
Original unmatched waveforms show that the Topping D90 has a phase inversion
Thankfully, DeltaWave identifies and corrects that. After matching, we see a well aligned waveform:
And silly zooming in gives us this, confirming a great match:
But the PK Metric is not as good at -73.3 dBr. In fact, it never really gets below -80 dB.
And you can see that the way that the Topping D90 processes DSD (white) results in a slightly lower ultrasonic noise than even the digitally source (blue).
Discussion
The biggest differences are in the ultrasonics. PK Metric will take this into account, but if the ultrasonics induce IMD in the audible range, it could make a difference. The Marantz does better when compared against the reference DSD digital source, so the Marantz is overall more accurate to the recording when using the perceptually weighted PK Metric. This argues against IMD being the reason for differences. The PK Metric really does show a lot of differences in the audible frequency range.
The E1DA is a great 1 kHz SINAD measurement device, but it's not so great of an ADC for recording music. It has a huge amount of DC, which DeltaWave corrects. It's also very thermally sensitive, so I may very well be seeing a glitch/defect in the way that the E1DA records things and variability in temperature. The E1DA has no buffer, so the input impedance doesn't reflect what you see with an actual pre-amp/integrated amp. I don't know how to control for this without spending more money than I really want to.
You can get false positives with null testing (suggesting that there is a difference between two recordings when there is none). I have seen this when DeltaWave makes an error in matching the waveforms and it's possible that trimming the audio files more precisely improve the match. That said, the silly zooming into the waveform shows that the matching is pretty good in my opinion.
I haven't really done listening tests. No good way to do ABX testing since I cannot level match closely enough. I could try a nearfield measurement of a speaker playing back the music, but then I'd be introducing the amplifier into the chain and I'd need a UMIK-2. All of this may still be academic, but it is interesting to see how the measurements and null testing surprises.
In this post, I'm sharing some preliminary data showing
a) a measurable PK Metric difference between two reasonable DACs with real music
b) the DAC with a ~10 dB worse 1 kHz SINAD unit seems to be more accurate to the recording using PK Metric!
Please consider this preliminary data, since I am hoping that others will try to reproduce the results with their own gear and own music too. There is an excess of data presented to "show my work" since this is an unusual result.
The reasonable DACs? The Topping D90 MQA and Marantz SA-10. Both have been characterized by @amirm. The original D90 is still reference class by today's standards except for XLR phase inversion from convention, and the SA-10 does not measure as well, but to be fair, is a 2016-era product which was in development since 2014 or so and still "should" be good enough to threshold of audibility.
My 2018-production SA-10 does give slightly better PCM 1 kHz SINAD results than Amir's unit, though the test environment is different, and I got these results for DSD64. My Topping D90 gets to the -120dB 1kHz SINAD, so it's not defective either. I kept the D90 away from other transformers.
Test Track / Methodology
Concerto symphonique No. 4 in D minor, Op. 102: Scherzo from SFS0060 - Masterpieces in Miniature (Physical SACD) & DSD extract
I measured the DSD playback on the Topping D90 versus physical disc playback on the Marantz SA-10.
E1DA Cosmos ADC was running 32/176 to capture everything the SACD should reproduce. XLR balanced cables were used and L/R channel was verified each time before recording. For the SA-10, I played the disc, and then pressed the previous track button on the remote and then recorded using Audacity. For the D90, I used Foobar/ASIO4All to send a DSD stream to the D90. I confirmed that the LCD display indicated a 2.8MHz sampling rate, ensuring that my software configuration was valid for transmitting DSD. I pressed play in Foobar and then quickly used ALT-TAB and SHIFT-R to start the recording. When the 1 minute mark on the SACD player or the Foobar status bar was reached, I stopped the recording.
Control Experiments
This shows how repeatable the source devices are, and how repeatable the test environment is. The E1DA Cosmos ADC itself is subject to thermal effect. The PK Metric is super low, indicating the testing conditions are pretty good.
Comparison
Now I compare the Topping D90 as reference against the Marantz SA-10
- the phase is inverted on the D90; DeltaWave corrects for these
- DeltaWave will treat the "reference" as reference and then modify the comparison file. Since the SA-10 outputs a higher voltage than the D90 (2dB worth), DeltaWave will reduce the volume of the SA-10 recording to the D90. This might help the SA-10 by helping to reduce its inherent noise. I'm not sure. (@pkane?)
DeltaWave v2.0.8, 2023-03-27T20:43:12.1452921-07:00
Reference: D90 Repeat2.wav[?] 7892736 samples 176400Hz 32bits, stereo, MD5=00
Comparison: SA10 repeat 2.wav[?] 7905362 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -2.747dB Comparison: -0.791dB
Initial RMS values Reference: -25.778dB Comparison: -23.844dB
Null Depth=7.747dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Phase inverted
X-Correlation offset: 17670 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (1.01μs)
Trimmed 17963 samples ( 101.831066ms) front, 6976 samples ( 39.546485ms end)
Final peak values Reference: -2.747dB Comparison: -2.764dB
Final RMS values Reference: -25.83dB Comparison: -25.884dB
Gain= 2.049dB (1.2661x) DC=0 Phase offset=100.162876ms (17668.731 samples)
Difference (rms) = -44.88dB [-65.5dBA]
Correlated Null Depth=43.02dB [47.78dBA]
Clock drift: 12.64 ppm
Files are NOT a bit-perfect match (match=0.11%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0059% when reduced to 6.95 bits
---- Phase difference (full bandwidth): 15.2586814087706°
0-10kHz: 0.94°
0-20kHz: 0.81°
0-24kHz: 0.75°
Timing error (rms jitter): 2.2μs
PK Metric (step=400ms, overlap=50%):
RMS=-56.9dBr
Median=-59.2
Max=-49.2
99%: -50.56
75%: -55.79
50%: -59.16
25%: -61.21
1%: -66.19
gn=0.789857158565091, dc=0, dr=1.26390395685614E-05, of=17668.7312619653
DONE!
Signature: 5f180304410d4c7fb8d5a1775e8f2b5f
RMS of the difference of spectra: -103.878891018632dB
DF Metric (step=400ms, overlap=0%):
Median=-20.1dB
Max=-15.9dB Min=-31dB
1% > -29.41dB
10% > -25.18dB
25% > -22.79dB
50% > -20.13dB
75% > -18.97dB
90% > -17.65dB
99% > -6.78dB
Linearity 15.9bits @ 0.5dB error
Reference: D90 Repeat2.wav[?] 7892736 samples 176400Hz 32bits, stereo, MD5=00
Comparison: SA10 repeat 2.wav[?] 7905362 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -2.747dB Comparison: -0.791dB
Initial RMS values Reference: -25.778dB Comparison: -23.844dB
Null Depth=7.747dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Phase inverted
X-Correlation offset: 17670 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (1.01μs)
Trimmed 17963 samples ( 101.831066ms) front, 6976 samples ( 39.546485ms end)
Final peak values Reference: -2.747dB Comparison: -2.764dB
Final RMS values Reference: -25.83dB Comparison: -25.884dB
Gain= 2.049dB (1.2661x) DC=0 Phase offset=100.162876ms (17668.731 samples)
Difference (rms) = -44.88dB [-65.5dBA]
Correlated Null Depth=43.02dB [47.78dBA]
Clock drift: 12.64 ppm
Files are NOT a bit-perfect match (match=0.11%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0059% when reduced to 6.95 bits
---- Phase difference (full bandwidth): 15.2586814087706°
0-10kHz: 0.94°
0-20kHz: 0.81°
0-24kHz: 0.75°
Timing error (rms jitter): 2.2μs
PK Metric (step=400ms, overlap=50%):
RMS=-56.9dBr
Median=-59.2
Max=-49.2
99%: -50.56
75%: -55.79
50%: -59.16
25%: -61.21
1%: -66.19
gn=0.789857158565091, dc=0, dr=1.26390395685614E-05, of=17668.7312619653
DONE!
Signature: 5f180304410d4c7fb8d5a1775e8f2b5f
RMS of the difference of spectra: -103.878891018632dB
DF Metric (step=400ms, overlap=0%):
Median=-20.1dB
Max=-15.9dB Min=-31dB
1% > -29.41dB
10% > -25.18dB
25% > -22.79dB
50% > -20.13dB
75% > -18.97dB
90% > -17.65dB
99% > -6.78dB
Linearity 15.9bits @ 0.5dB error
Quality of the Match
I crop out the first 6 seconds and last 10 seconds of the recording since it throws DeltaWave off. Here are the files as-is, showing correct phase matching between the two recordings.
and the final match
We can zoom in and see how tight/good DeltaWave's level matching and alignment are. Blue is the Topping D90 and White is the Marantz SA-10. The increased noise and increased ultrasonics of the SA-10 are clearly seen in these ultra-zoomed-in views.
The aligned spectrum shows a slight roll-off in the bass for the SA-10 below 20Hz and the higher noise level in the high frequencies which does drop down into the audible range. The big difference is the higher ultrasonic noise of the Marantz SA-10.
Looked at a different way, we can see the delta of the Spectra. Here, you can see that the ultrasonics are really the main difference between the two files. No surprises yet. The spectra looks at the recording as a whole, not for specific portions of the music.
I don't know how best to interpret the phase, but I've shown it here. It seems like everything is pretty close to 0 degrees in the audible range, ignoring the bass I mentioned.
The important slide: PK Metric of -56.9 dBr rms and -48.4 dBFS for the transients that generate the biggest difference. -48.4 dBFS meets the PK metric threshold for audible.
This suggests that both sound similar but during that transient, it should be possible to hear a difference between the two.
(and a close-up of the spike at ~16 seconds shows that there's a lot going on in the audible frequencies. Note that I clip the first 6 seconds of the recording of you want to listen to the music yourself.
The PK Metric is a perceptually weighted value that tries to capture if a null comparison should be or should not be audible. Greater than -50 dB means that it's clearly audible. At -56.9 dBr, it's probably still really hard to hear the difference, but if you use the cool split window feature that @pkane put in, and hold CTRL and then move your mouse on the left to reach the area of the peak spikes around 15 seconds, you can see that the level there is -48.4 dBFS and in the audible frequency range.
So, here, we have a flagship DAC and a flagship SACD player that *should* get into the range of indistinguishable, but the PK Metric is showing us that there may actually be a difference.
Different, yes. But which is correct? Maybe, the Marantz!
For true digital comparisons, I took the DSD source file and then used the TASCAM Hi-Res Editor to convert the DSD 2.8MHz file into a 32-bit FP / 176 kHz PCM version. 176 kHz is how I made my E1DA recordings. I then redid the DeltaWave tests, using that Digital Source as the reference.
Referenced to the digital source file
Topping D90 gets a PK Metric of -73.3 dBr (rms)
Marantz SA-10 gets a PK Metric of -80.3 dBr (rms)
Let's look at the details because if anything, the Marantz performance is under-estimated.
DeltaWave v2.0.8, 2023-03-27T23:02:26.6131642-07:00
Reference: Master-Digital.wav[?] 7811807 samples 176400Hz 32bits, stereo, MD5=00
Comparison: SA10 repeat 2.wav[?] 7905362 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -0.862dB Comparison: -0.791dB
Initial RMS values Reference: -23.929dB Comparison: -23.844dB
Null Depth=13.749dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
X-Correlation offset: -10531 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (0.81μs)
Trimmed 42353 samples ( 240.096372ms) front, 35254 samples ( 199.852608ms end)
Final peak values Reference: -0.862dB Comparison: -0.776dB
Final RMS values Reference: -23.902dB Comparison: -23.832dB
Gain= 0.0259dB (1.003x) DC=0 Phase offset=-59.704791ms (-10531.925 samples)
Difference (rms) = -40.55dB [-69.79dBA]
Correlated Null Depth=37.87dB [68.06dBA]
Clock drift: 15.35 ppm
Files are NOT a bit-perfect match (match=0.07%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0024% when reduced to 6.23 bits
---- Phase difference (full bandwidth): 5.48353378963904°
0-10kHz: 3.00°
0-20kHz: 3.84°
0-24kHz: 4.25°
Timing error (rms jitter): 15.8μs
PK Metric (step=400ms, overlap=50%):
RMS=-80.3dBr
Median=-83.7
Max=-64.5
99%: -70.77
75%: -80.38
50%: -83.73
25%: -86.47
1%: -93.91
gn=0.997023879532064, dc=-7.47465367171731E-08, dr=1.5353617707155E-05, of=-10531.9251125162
DONE!
Signature: 27331bba9f6fa02bd5743a3125da2397
RMS of the difference of spectra: -99.5984246212413dB
DF Metric (step=400ms, overlap=0%):
Median=-17.7dB
Max=-13.9dB Min=-28.3dB
1% > -27.01dB
10% > -23.16dB
25% > -20.04dB
50% > -17.68dB
75% > -16.28dB
90% > -15.09dB
99% > -5.65dB
Linearity 14.5bits @ 0.5dB error
Reference: Master-Digital.wav[?] 7811807 samples 176400Hz 32bits, stereo, MD5=00
Comparison: SA10 repeat 2.wav[?] 7905362 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -0.862dB Comparison: -0.791dB
Initial RMS values Reference: -23.929dB Comparison: -23.844dB
Null Depth=13.749dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
X-Correlation offset: -10531 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (0.81μs)
Trimmed 42353 samples ( 240.096372ms) front, 35254 samples ( 199.852608ms end)
Final peak values Reference: -0.862dB Comparison: -0.776dB
Final RMS values Reference: -23.902dB Comparison: -23.832dB
Gain= 0.0259dB (1.003x) DC=0 Phase offset=-59.704791ms (-10531.925 samples)
Difference (rms) = -40.55dB [-69.79dBA]
Correlated Null Depth=37.87dB [68.06dBA]
Clock drift: 15.35 ppm
Files are NOT a bit-perfect match (match=0.07%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0024% when reduced to 6.23 bits
---- Phase difference (full bandwidth): 5.48353378963904°
0-10kHz: 3.00°
0-20kHz: 3.84°
0-24kHz: 4.25°
Timing error (rms jitter): 15.8μs
PK Metric (step=400ms, overlap=50%):
RMS=-80.3dBr
Median=-83.7
Max=-64.5
99%: -70.77
75%: -80.38
50%: -83.73
25%: -86.47
1%: -93.91
gn=0.997023879532064, dc=-7.47465367171731E-08, dr=1.5353617707155E-05, of=-10531.9251125162
DONE!
Signature: 27331bba9f6fa02bd5743a3125da2397
RMS of the difference of spectra: -99.5984246212413dB
DF Metric (step=400ms, overlap=0%):
Median=-17.7dB
Max=-13.9dB Min=-28.3dB
1% > -27.01dB
10% > -23.16dB
25% > -20.04dB
50% > -17.68dB
75% > -16.28dB
90% > -15.09dB
99% > -5.65dB
Linearity 14.5bits @ 0.5dB error
Here are the RAW waveforms before matching
and then after matching super zoomed in
and the PK Metric is -80.3 dBr and you can see a good amount of time is spent below -80 dB.
The big spike at 35 seconds? It's obvious in the delta, but not obvious in the source recording.
The matched spectra shows the effect of the digital filter used by TASCAM. The slope in the 30 to 40 kHz range looks similar.
And here is the Topping D90
DeltaWave v2.0.8, 2023-03-27T23:19:19.0023470-07:00
Reference: Master-Digital.wav[?] 7811807 samples 176400Hz 32bits, stereo, MD5=00
Comparison: D90 Repeat2.wav[?] 7892736 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -0.862dB Comparison: -2.747dB
Initial RMS values Reference: -23.929dB Comparison: -25.778dB
Null Depth=14.57dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Phase inverted
X-Correlation offset: -28201 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (1.23μs)
Trimmed 24996 samples ( 141.70068ms) front, 15742 samples ( 89.240363ms end)
Final peak values Reference: -0.862dB Comparison: -0.877dB
Final RMS values Reference: -23.916dB Comparison: -23.925dB
Gain= -1.8904dB (0.8044x) DC=0 Phase offset=-159.868703ms (-28200.839 samples)
Difference (rms) = -51.06dB [-67.26dBA]
Correlated Null Depth=46.36dB [64.76dBA]
Clock drift: 2.78 ppm
Files are NOT a bit-perfect match (match=0.23%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0072% when reduced to 7.96 bits
---- Phase difference (full bandwidth): 5.84587571327181°
0-10kHz: 2.62°
0-20kHz: 4.38°
0-24kHz: 4.74°
Timing error (rms jitter): 9.4μs
PK Metric (step=400ms, overlap=50%):
RMS=-73.3dBr
Median=-76.2
Max=-63.3
99%: -65.62
75%: -73.15
50%: -76.16
25%: -78.12
1%: -81.09
gn=1.24313312684506, dc=1.46502821251829E-06, dr=2.78400992464591E-06, of=-28200.8391638284
DONE!
Signature: 06bd2487c490ccfe122634a8bd61ea15
RMS of the difference of spectra: -103.52586412889dB
DF Metric (step=400ms, overlap=0%):
Median=-28.5dB
Max=-24.4dB Min=-39.1dB
1% > -37.84dB
10% > -33.32dB
25% > -31.18dB
50% > -28.45dB
75% > -26.97dB
90% > -25.73dB
99% > -10.17dB
Linearity 18.6bits @ 0.5dB error
Reference: Master-Digital.wav[?] 7811807 samples 176400Hz 32bits, stereo, MD5=00
Comparison: D90 Repeat2.wav[?] 7892736 samples 176400Hz 32bits, 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: -500dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:False
Upsample:False, Window:Kaiser
Spectrum Window:Kaiser, Spectrum Size:32768
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Kaiser, taps:262144, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=6s, End=10s
Discarding Comparison: Start=6s, End=10s
Initial peak values Reference: -0.862dB Comparison: -2.747dB
Initial RMS values Reference: -23.929dB Comparison: -25.778dB
Null Depth=14.57dB
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Phase inverted
X-Correlation offset: -28201 samples
Trimming 0 samples at start and 0 samples at the end that are below -90.31dB level
Drift computation quality, #1: Excellent (1.23μs)
Trimmed 24996 samples ( 141.70068ms) front, 15742 samples ( 89.240363ms end)
Final peak values Reference: -0.862dB Comparison: -0.877dB
Final RMS values Reference: -23.916dB Comparison: -23.925dB
Gain= -1.8904dB (0.8044x) DC=0 Phase offset=-159.868703ms (-28200.839 samples)
Difference (rms) = -51.06dB [-67.26dBA]
Correlated Null Depth=46.36dB [64.76dBA]
Clock drift: 2.78 ppm
Files are NOT a bit-perfect match (match=0.23%) at 16 bits
Files are NOT a bit-perfect match (match=0%) at 32 bits
Files match @ 50.0072% when reduced to 7.96 bits
---- Phase difference (full bandwidth): 5.84587571327181°
0-10kHz: 2.62°
0-20kHz: 4.38°
0-24kHz: 4.74°
Timing error (rms jitter): 9.4μs
PK Metric (step=400ms, overlap=50%):
RMS=-73.3dBr
Median=-76.2
Max=-63.3
99%: -65.62
75%: -73.15
50%: -76.16
25%: -78.12
1%: -81.09
gn=1.24313312684506, dc=1.46502821251829E-06, dr=2.78400992464591E-06, of=-28200.8391638284
DONE!
Signature: 06bd2487c490ccfe122634a8bd61ea15
RMS of the difference of spectra: -103.52586412889dB
DF Metric (step=400ms, overlap=0%):
Median=-28.5dB
Max=-24.4dB Min=-39.1dB
1% > -37.84dB
10% > -33.32dB
25% > -31.18dB
50% > -28.45dB
75% > -26.97dB
90% > -25.73dB
99% > -10.17dB
Linearity 18.6bits @ 0.5dB error
Original unmatched waveforms show that the Topping D90 has a phase inversion
Thankfully, DeltaWave identifies and corrects that. After matching, we see a well aligned waveform:
And silly zooming in gives us this, confirming a great match:
But the PK Metric is not as good at -73.3 dBr. In fact, it never really gets below -80 dB.
And you can see that the way that the Topping D90 processes DSD (white) results in a slightly lower ultrasonic noise than even the digitally source (blue).
Discussion
The biggest differences are in the ultrasonics. PK Metric will take this into account, but if the ultrasonics induce IMD in the audible range, it could make a difference. The Marantz does better when compared against the reference DSD digital source, so the Marantz is overall more accurate to the recording when using the perceptually weighted PK Metric. This argues against IMD being the reason for differences. The PK Metric really does show a lot of differences in the audible frequency range.
The E1DA is a great 1 kHz SINAD measurement device, but it's not so great of an ADC for recording music. It has a huge amount of DC, which DeltaWave corrects. It's also very thermally sensitive, so I may very well be seeing a glitch/defect in the way that the E1DA records things and variability in temperature. The E1DA has no buffer, so the input impedance doesn't reflect what you see with an actual pre-amp/integrated amp. I don't know how to control for this without spending more money than I really want to.
You can get false positives with null testing (suggesting that there is a difference between two recordings when there is none). I have seen this when DeltaWave makes an error in matching the waveforms and it's possible that trimming the audio files more precisely improve the match. That said, the silly zooming into the waveform shows that the matching is pretty good in my opinion.
I haven't really done listening tests. No good way to do ABX testing since I cannot level match closely enough. I could try a nearfield measurement of a speaker playing back the music, but then I'd be introducing the amplifier into the chain and I'd need a UMIK-2. All of this may still be academic, but it is interesting to see how the measurements and null testing surprises.
Last edited: