You are effectively listening to the PSU when amplifying, the Listening room noise floor , negates Digital's supposedly superior dynamics, hence why most dont hear it....
The 'phenomenon' is trivially obvious, it's distortion. Trouble is, most people are so used to hearing a constant hash of low level distortion in playback that they believe that what they hear is 'correct' sound - but it's a long way from it. Perhaps they need some therapy sessions, of being exposed to first, conventional hifi quality, and then correct sound, back and forth - then they might "get it" , 😛 😀.
This distortion is typically dynamic in nature, not a simpleton as most measuring tends to assume. You might as well assess a car for handling performance, from the driver's perspective, by sticking a vehicle on a stationary rig, and tugging and probing it in various ways ...
Dead easy to pick with ears ... if the sound from a speaker driver working strongly is quite unpleasant or unsettling when your ear is close by then there's a very technical term for that: "audible distortion" ...
I would think a problem like you describe would be obvious with something like this: Audio DiffMaker In practice it has shown very subtle and reproducible errors in the presence of large dynamic signals. I would think it could show what you are describing very easily.
Two problems there: DiffMaker is a poor program, very poor in fact - as a programmer I have very little respect for the implementation of it; and it fails to do its job in an intelligent way, it can't handle the real issues that 2 seemingly similar versions of a track most times will present - as an example, that the speeds will vary through the playback, it's not a trivial exercise to synchronise at each point of the track through the full length of the piece.
I'm currently working on doing a Better DiffMaker, for my own benefit in the first instant - mentioned in my blog here ...
I'm currently working on doing a Better DiffMaker, for my own benefit in the first instant - mentioned in my blog here ...
If easily picked and discernible by ear at loud levels , then why have your ears close to driver, as far as i know this is all detectable by measurements and test equipment developed over the last 3 decades ...
Did i miss something ....... ?
Did i miss something ....... ?
Yep - the dynamic shunt reg I use is the simplest one - more caps to lower the impedance at LF, ceramics for HF. No power is wasted just the extra electrolytics are really, really bulky and I'd like a portable solution.....😛
Maybe (like JC does on his preamps), a high current capacitance multiplier?
I can't see how a capacitance multiplier helps when the noise we want to minimize is load-induced. They work great for filtering line noise though.
I'm sure measurements would pick it up, which is not the main point - the all-important aspect is whether it's audible, subjectively does it interfere with hearing "what's going on" in the music. And part of listening to drivers working is not concentrating, if you have to mentally focus, strongly, to unravel the sound when doing this then there is too much distortion present - you need to be able to listen out of "the corner of your ear", so to speak, and the clarity be fully present, maintained ... this is an essential ingredient of "convincing" sound ...
Personally, regulators work for the LF stuff - but there also has to be the "take no prisoners" filtering of high frequency muck, which is where overdosing on capacitors works, 🙂 - the two approaches go hand in hand, IMO ...
No question that it is a difficult problem. Bill acknowledged as much when I met with him at the presentation of the paper. It was targeted at acoustics, not comparing variable playback systems. And you have not pulled together a better solution over a weekend either. I hope you can address the issue and do a better version. Maybe easier to use as well.
Never the less any tool of this type should be able to highlight issues with noise obscuring low level details in the presence of high level signals. I have used a test signal that has 12 tones that are not harmonically or IM related. In the digital realm (DA:AD) I can usually get down to the noise floor between tones. Add an transducer (headphone) and lose about 30-40 dB. Speakers are worse. The loopback test is below. The only open question is whether there is a change from running to stop to running which would be most likely a thermal effect. I can provide a wave file of this as well.
Attachments
Yep regulators work, but there are some kinds of regulators that work better than others. I'm experimenting with a regulator at the moment which uses a regular 3pin device in series with an emitter follower. Sounds better to me than the 3pin regulator by itself in terms of how much LF ambience information is revealed.
Last edited:
Regulators or caps, if the reference/return is not worked out very carefully you are just moving noise around. I prefer bridge tied loads and floating supplies for these reasons. No ground currents to worry about. Done right you don't need heroic efforts in the supply filtering for the outputs.
In the first instant I'm using Octave, because it has all the tools inbuilt, effectively. But I haven't used this type of software before, so I'm still getting up to speed on how to drive it well - looks extremely promising so far. Once it gives me the right answers, then I'll look to see how to take it further ...
The difficult part is that speakers get lots of obvious things wrong, gives DiffMaker a nightmare - but IME they can get the critical things right, what matters subjectively. Which is part of the challenge, sorting that out ...
In my approach I see most of the "muck" originating from the load varying requirements of the circuit being fed, the output impedance of the PS is not low enough at very high frequencies - rather than noise coming through from elsewhere. So I see the job to be in not allowing noise to be created in the first place, by having the rails behaving as close to ideal voltage sources as possible. Ground returns are obviously part of the design approach, in my mind every piece of wire or copper plane is always a resistor, an impedance - helps to keep one's thinking on track.
This is the easier end of the spectrum to fix up provided one is using chipamps where the rails in question are already in very close physical proximity. Ceramic caps aren't very bulky nor particularly expensive, just keep on adding them until the "muckiness" sinks beneath audibility.
That sounds promising, many of the 3 terminal regulators have poor transient response and would benefit from being relieved of large current variations.
When I did my gainclone the simulations didn't show any significant benefit in having ceramics as well as, decent electros effectively at minimal distance from the power leads, so I never used them. Audibly, I didn't feel they were needed - maybe they would have made a difference, maybe not ...
I have my doubts that the transient response is the issue as regards audibility - simply because I'm feeding the output of the reg into a huge cap bank via a series inductor. So the transients are going to be handled by the caps, leaving only the LF response of the reg making any difference at the pins of the amp by virtue of being filtered only minimally by the LC filter.
With the aim of minimizing LF noise in mind, I'm using the emitter follower rather as a cap multiplier fed from a reference engineered for as low LF noise as I can practically get. Seeing as RC filtering does little for LF noise I'm averaging the output of multiple references to augment the LF noise reduction.
@Frank - as you adopted rather large copper sheets for power (are we talking about the same instance?) then you were not so constrained by space as I was with my chipamp experiment. Hence lytics will doubtless do a fine job, particularly so at higher supply rails where ceramics lose a huge amount of their capacitance. My rails for the ceramics experiment were below 18V.
Last edited:
Yes, I was using decent sized copper sheets, so the impedance from any cap to a chipamp lead was minimal, everything was physically squashed together as much as possible to optimise this. Obviously where this is significant stray inductance in those paths a smaller cap, typically a ceramic, would be a better fit. I played quite a bit with including parasitic inductance in my sims, and in my case there was barely any benefit by throwing even the best performing, by the spec's, multi layer cap into the equation.
- Status
- Not open for further replies.