frequency-, phase-, distortion (and noise) response enough to determine sound quality?

[Jazigo]

The correlation of measurements from the same amp is a good check for some of these problems. And the B100-B100 also shows slight oscillations in the treble, while the PA5II-PA5II looks pretty much flat.

This was actually reproducible over several takes. The PA5 II was super consistent, while the B100 showed more variations. Why that is, I can't tell, but it is there.

Actually, I uploaded the sources here: https://drive.google.com/file/d/1qtwrhEBCuwYW_EeacA6F23XQEF54nbNE/view?usp=sharing, so that anyone can check themself.

 

[Julf]

For the time being, I have been researching bandwidth limitations of class D amps. This is a area that Bruno Putzeys points out being an inhered drawback of class D amps and have been "downplaying" the importance of this. I may be completely wrong here in the "downplaying" part, but I am investigating for the moment. For the time being I am looking at step responses, as I suspect they will show something, since they have infinite bandwidth in a mathematical sense.

I am interested in hearing why you think the bandwidth limitations are an issue. You are bandwidth-limited by the source material, speakers and your ears anyway.

 

[Jazigo]

I am interested in hearing why you think the bandwidth limitations are an issue. You are bandwidth-limited by the source material, speakers and your ears anyway.

The main difference from a class a/b/ab and a class d is that the class d samples the input and reconstruct an amplified version of it on the output. The sampling rate directly sets the bandwidth for the rest of the system. This becomes important for the feedback circuit, which is bandwidth-dependent.

I will fully admit that this may be a fools errand, but it is part of my research on way or the other.

 

[Julf]

The main difference from a class a/b/ab and a class d is that the class d samples the input and reconstruct an amplified version of it on the output.

No, it doesn't. Pretty much all commercial class D amps are analog pulse density modulation systems. Yes, it is bandwidth-limited, but so is the (digital) source material, and so are your ears. Again, why would it have an audible effect?

 

[RandomEar]

I downloaded the files and looked at them with DeltaWave. After tuning some settings to get slightly smoother plots, I analyzed all combinations. For easier naming, I will call "B100 No.1" = B1 and "PA5II No.1" = "P1" from here on.

A couple of things to note:

• The PK Metric (sort of a similarity score, the lower the better) for the B1-B2 pair is -52.8 dB, which is slightly worse than that of any other pairing (B1-P1, B1-P2, B2-P2, P1-P2)
• The PK Metric of the P1-P2 pairing is the best at -61.3 dB. For comparison, the PK Metric for a direct loopback recording on an RME ADI 2 vs. the original file is about -80 dB and the loopback should be indistinguishable from the original.
• Even though I initially thought so, I could see no clear pattern in the PK metric that would indicate that certain points in time within the song trigger clear differences between the amps
• The Spectrum of Delta metric is usually below -70 dB, often below -75 dB upwards of 2 kHz, indicating extremely small differences
• The clock drift is measurable and non-linear, but the non-linear correction does not significantly improve any metric vs. the default linear correction (I used non-linear correction for all evals nonetheless)
• You can also see the wavy pattern from the Delta of Spectra plots in the Delta Spectrograms (for the Bx-Px pairings)
• The wavy pattern in the Delta of Spectra plots starts above 12 kHz, which is exactly 1/4 the sampling rate (odd, but could be random)
• The wavy pattern looks a bit like comb filtering, indicating the influence of delayed reflections. But if that were the case, the pattern should also appear for the P1-P2 pairing
• The Delta Phase plots look pretty random to me and I don't think we can extract any useful info from those
• The linearity plots look identical for all pairings

Overall, I generated lots of plots, but I don't feel any closer to an explanation


Example plots
Delta Phase P1-P2 and B1-P2

Clock Drift B1-B2

Delta Spectrograms B1-B2 and B1-P1

Linearity B1-P1

Self-correlation plots
B1-B2

P1-P2

Cross-correlation plots
B1-P1

B1-P2

B2-P1

B2-P2

 

[Jazigo]

That is super cool! I only glanced at the spectrum delta when I did it. Hard to tell why. On one hand, maybe the PA5 II is the one sounding correct. Maybe the sound should be a bit leaner? Maybe the dynamics from the B100 is just a loser grip?

For the time being I am looking at the global feedback in class d amps, which include the output filter in the feedback (PFFB). This in turn also include the back-pressure from the driver(s). Sort of interesting in its own right, but also connected to the lesser bandwidth at higher frequencies.

 

[solderdude]

Maybe the dynamics from the B100 is just a loser grip?

The notion that certain amplifiers have 'grip' on drivers is erroneous.
Amplifiers can not and do not 'grip' drivers, that is all in people's heads.
I understand the ratio behind the thinking usage of wording.
Now if there is actual MFB it is another thing then the correction signal compensates for the excursion errors.

Amplifiers can have a different output resistance (and thus damping factor) but there is no difference in actual damping between a DF of 500 and 50 in reality.
When a driver with say 8ohm internal resistance generates counter EMF and it is connected to an amp with 16mΩ output R (DF 500 at 8ohm) and it generates 1V counter EMF the damping current will be 124.8mA.
When a driver with say 8ohm internal resistance generates counter EMF and it is connected to an amp with 160mΩ output R (DF 50 at 8ohm) and it generates 1V counter EMF the damping current will be 122.5mA.
So the difference in damping is just 1.9% = 0.16dB difference.

So a higher damping factor above say ... 50 does not lead to more 'grip' on the driver. And this is assuming there is no inductor in series with the driver which will worsen the actual DF of a high DF amplifier even more.

When a driver with say 8ohm internal resistance generates counter EMF and it is connected to an amp with 1.6Ω output R (DF 5 at 8ohm) and it generates 1V counter EMF the damping current will be 105mA, 18% less damping current opposite DF500 = 1.5dB difference in current.

 

[RandomEar]

That is super cool! I only glanced at the spectrum delta when I did it. Hard to tell why. On one hand, maybe the PA5 II is the one sounding correct. Maybe the sound should be a bit leaner? Maybe the dynamics from the B100 is just a loser grip?

For the time being I am looking at the global feedback in class d amps, which include the output filter in the feedback (PFFB). This in turn also include the back-pressure from the driver(s). Sort of interesting in its own right, but also connected to the lesser bandwidth at higher frequencies.

I think currently, my main takeaway would be that looking at the Delta of Spectra is misleading in the sense that the Spectrum of Delta - so an FFT of the actual differences between recordings - shows that they are very small and very far down at -70 to -90 dB. So if you look at the Delta of Spectra, you may forget that small differences of up to 0.5 dB could technically be audible, but they won't be if they are essentially happening around the noise floor of the microphone (and the room!).

Also, the PK metric shows that on average, the correlation or similarity of the recordings isn't perfect at around -55 dB. But at the same time, the similarity of the tracks is clearly limited by the fact that it is a microphone recording and considering that limitation, they are all matched pretty well.

If you remember your ABX results after DeltaWave matching from multiple runs, it would be interesting if you could post them. That would add another data point.