Real audio science - exploring beyond the obvious and bringing back true science and exploration

A bigger issue in metal drivers is ringing, even if subtle, that is longer than the fundamentals in the high frequencies.

Also the imd from the above the audible band hard break that comes into the audible band can be very audible and measurable with many tweeters and speakers even if us deaf guys have hf notched and roll off. It’s pretty much the equivalent of shoving some butt circuits that distort in the highs in unpleasant ways into the audio path.

Proper converter low pass filters are linear phase. Cheap converters, certain audiophile, “special” pro ones, and ultra low latency filters compromise this or have inadequate rejection. Minimum phase filters and NOS dacs are no improvement over linear phase filters for anti alias filters because some frequencies will be delayed. The ringing from the FIR filter is inaudible if done properly.

 

So after reading more about Volterra series, it seems that the order <=2 series is determined by:

• The usual frequency and phase response
• The amplitude and phase of D2 at all frequencies
• For any pair of distinct frequencies at A and B Hz, the amplitude and phase of the intermodulation tones at A+B Hz and A-B Hz.

Is there anything to be got from D2/D3/D4 phase plots? Or IMD phase plots? I imagine most time domain effects are lurking in extremely high order but maybe there are low hanging fruit already in D2/D3/IMD phase.

 

One could try to a distortion test like stereophile sometimes does where the fundamental is notched out and compare for amplifiers and speakers, etc. For example phono carts, I think, are supposed to have a different D2 phase than amps.

In that case 1 and 1 doesn't make 2. Maybe it's 1.5 or maybe 0.5. Maybe 0. I'll try to see if I can rig something up soon.

 

Very probably a phase issue; the latency of different DACs from input to output is not gonna be the same.

 

Thought this was an interesting podcast on the topic if a bit limited in scope: https://podcasts.google.com/feed/aH...d=0CAUQkfYCahcKEwjQ-pXZ-qGAAxUAAAAAHQAAAAAQLA

 

Interesting link in the show notes. I'll have to add Fast Cars to my test tracks playlist.

https://www.linkedin.com/feed/update/urn:li:activity:7084958885526065153

 

At the other end of the fidelity spectrum,
"“Tom’s Diner” was so admired for its simplicity and clarity that it became the test track used by German scientists to perfect the mp3."
https://observer.com/2008/09/suzanne-vega-is-the-mother-of-the-mp3/

 

I looked a bit more into the DAC11001B chip and it looks like they want to push its usage in audio DACs:
https://www.ti.com/document-viewer/...8B5#GUID-2C286BFE-A859-49C0-8B99-2314DF472023


Looks like the team for this chip at TI really like Schiit if they reached out about the revision. @schiit It's probably kept under wraps but were you privy to how they revised it for that much out-of-band performance improvement?

 

I'm working on an amp that I had intended to build with nanocrystalline core output transformers, but recently had the opportunity to pick up some amorphous core output transformers with similar ratio and secondary DCR. I figure what the heck, let's roll some transformers. I throw both onto sockets, and try A/Bing them. The difference was very audible, though both sounded excellent.

I tried measuring the amp in both configurations on an apx515. I am amazed by how little insight THD and single harmonic measurements provided.



The NC core ekes out a bit more power, but this is probably due to the higher DC resistance of the AM primary reducing the amount of voltage headroom, rather than a core effect. The AM core does have a much higher primary inductance, but both are more than sufficient. Both are made by Lundahl on the same size core.

I'm thinking the difference between the transformers might be easily measured in isolation, but lost when measuring the complete system. I'll be taking some more measurements to try to shed light on this. More than happy to try out any suggestions.

 

Square waves.

 

I gave this a shot with a 47 ohm resistor to terminate the secondary, and measured at 200Hz, 2kHz, and 20kHz. Image attached because it's huge.



The leading edge is a bit sharper on the NC unit for the 2k and 20k tones, but I think this just tells us that the NC transformer has a wider bandwidth? This would make sense-- the AM core has a higher ratio, so one would expect the parasitic capacitance and leakage inductance to be higher. They're both darn flat up to 60kHz. Perhaps there's a correlation to the transformer's >20kHz performance, but I'd be hesitant to say the link is causal.

 

Awesome! The differences in the leading edge and the slight saw-toothing on the NC definitely seem different enough that I think there could be audible differences there. I bet an FFT of both of those would highlight some interesting patterns.

I wonder if a reactive load might be a more challenging scenario than the fixed resistor?

 

The leading edge of the square wave is high frequency content and the falling edge approaches DC. If your square wave gen is 50R source impedance, then these are pretty well behaved transformers, e.g. no ringing. But if you've got a higher source impedance and not enough primary inductance, then that'll create the falling sawtooth output waveform which indicates a loss of low frequency content. It's quick and simple to see the effect of rising source impedance (to mock up potential output tube Rp) vs primary inductance with a series pot on the primary side. But if the source impedance is in fact low and you're seeing either sawtooth, then the transformer wants some other load impedance than your resistor. You may need to add a zobel. All of this is pretty easy to converge through trial and error if you've got a couple pots and miscellaneous caps lying around. Good transformers don't require a zobel.

What kind of tube amp are you building?

 

Agreed, I don't think the square waves tell us anything that the frequency (and phase) response don't already tell us. And with the frequency response extending past 80kHz on both of these, there must be some other underlying cause.

The amp is one of the solid state triode amps I've been rambling about in the DIY threads. Each "tube" has a mu of 60, a gm of about 500mS, and an rp of about 100 ohms. B+ is regulated to 100V, and each phase runs around 40mA, and it uses both input and ouput transformers. I don't want to get into too much depth here, but if there's interest I'm happy to share a bit more in the DIY threads.

I knocked the noise down further and redid the distortion analysis. I think we're getting somewhere? It's interesting that the noise had such a masking effect on the THD vs frequency measurements. Note that this is with a 600 ohm load.



The first FR plot aligns well with what we're seeing in the square wave response. THD vs output voltage at 1kHz doesn't tell us much, but the THD vs frequency graph is a lot more revealing now. I've attached the H2/H3 graphs as well. I don't want this to be taken as too definitive for now-- I don't understand by what mechanism transformer THD would rise with increasing frequency, so it may be an interaction with the amp rather than a transformer effect.

 

The premise is and it at least it correlates so that speakers that sound neutral, measure flat, at the limit there can be a slight downward tilt to the highs. A crossover I designed to a speaker made it sound awesome. I then measured the fr response, it was horrible, with a massive droop from 2 to 4 kHz - an inductor was disconnected in both speakers. It must have meant the phase and timing was way off, too. And still it sounded great.
On this day listening test 1, measurement 0.
*I did get the same awesome sound through some sweat and elbow grease with the same speaker measuring flat, eventually.
Further 'proof' to myself that fr response, while relevant, is severely overrated metric. And it is the best of the pool of semi useless measurements we have.

The best objective qualifier to tweeter sound I know is step response.
It correlates to what I deem musical and worthy for tweeter. Very generally speaking alu domes are the worst, CVD diamond the best - others in between.
Even so objective in this sentence is not that objective - you pick and choose what you think matters more from the step resp graph.
Is it already time for the myth do die that bandwidth to 20 kHz is enough? People like myths.

The science worthy knowledge in audio is in a very poor and undeveloped state.
That is why ''science based'', ''engineered'' audio components sound so bad.
The published science we have is good enough for foundation. Then we have simulations that help inform on some decisions.
Then comes the submerged part of the iceberg that we have no science based background.

Behind every great sounding device there is a set of good ears, some massive amount of empirical (usually hidden from public) knowhow, experience and a lot of trial and error. Kick me for producing a cliche - AI perhaps changes this.

 

"Real audio science - exploring beyond the obvious and bringing back true science and exploration"

Along these lines, see my post in this forum about using DOEs (Design of Experiments) to set up and effectively integrate a subwooder in a 2-channel system.

DOEs drive true understanding and insight from a scientific perspective for audio applications and set-ups, and...you not only get a tansfer function (and....everything is a transfer function), but you also get a frickin' p-value that tells you with statistical rigor just how good your model is.

It's often about...INTERACTIONS between "control factors" driving functional responses, and the most effective way to undertand and characterize with statistical rigor the nature and influence of these interactions is by doing DOES.

OFAT, One Factor At a Time analysis usually does not get you there. OFAT is often overly simplistic and usually does not respresent how the REAL WORLD actually works.

Cheers.

 

For you @JK47