Joachim, the mystery of WHY we hear things that we cannot easily measure is the whole point of the discussion. However, one good EXAMPLE of a quality playback that measures about the same as a 'not so high quality playback' is all that we need to demonstrate that there are factors that we are not measuring as carefully as we should. I nominate PIM or dynamic level modulated jitter, as something we do not yet measure very carefully.
For example, 33 years ago, I did an A-B comparison between a 5534 IC and a discrete design that is op amp based, but with more open loop bandwidth, than the 5534. Actually the two circuit's TOPOLOGY was similar. Still, the 5534 failed to sound the same as the discrete design, and more 'homogenized', yet the measurement that I made at the time would not predict any difference at all.
Of course, when I mentioned this to Dr. Lipshitz, he did not consider it a serious comparison. But, when my reputation is at stake, I need to be wary of what double blind testing might not show. I must say, that just shifting over to the 5534 would have made things much easier for me, if it had been possible to do, without an audible difference.
For example, 33 years ago, I did an A-B comparison between a 5534 IC and a discrete design that is op amp based, but with more open loop bandwidth, than the 5534. Actually the two circuit's TOPOLOGY was similar. Still, the 5534 failed to sound the same as the discrete design, and more 'homogenized', yet the measurement that I made at the time would not predict any difference at all.
Of course, when I mentioned this to Dr. Lipshitz, he did not consider it a serious comparison. But, when my reputation is at stake, I need to be wary of what double blind testing might not show. I must say, that just shifting over to the 5534 would have made things much easier for me, if it had been possible to do, without an audible difference.
Here is an example. I made a TIM measurement with extreme resolution ( the floor is at
-150dB ) on an AD797 stage in an INA arrangement and the same measurement on a single ended JFet-BJT cascode. Both designed as an MC pre-pre. The plot of the AD797 is cleaner but a bit more noisy. Later i made stages with the AD797 that had the same good noise performance ( massive paralleling ) then the exceptional low noise Fet solution.
Subjectively the Opamp solution sounded very clean but a bit sterile whereas the Fet Cascode was bouncy and engaging. Non double blind with peaking of cause so totally irrelevant, just for fun.
Pictures later. I have sometimes trouble to upload files that load well on another day without editing. That is a strange mystery too.
-150dB ) on an AD797 stage in an INA arrangement and the same measurement on a single ended JFet-BJT cascode. Both designed as an MC pre-pre. The plot of the AD797 is cleaner but a bit more noisy. Later i made stages with the AD797 that had the same good noise performance ( massive paralleling ) then the exceptional low noise Fet solution.
Subjectively the Opamp solution sounded very clean but a bit sterile whereas the Fet Cascode was bouncy and engaging. Non double blind with peaking of cause so totally irrelevant, just for fun.
Pictures later. I have sometimes trouble to upload files that load well on another day without editing. That is a strange mystery too.
Quoting John Curl: "Homogenized, is a term that I use for 'sterile', but not adding anything negative, just not giving the 'positive' of the music."
Just riffing on this thought, maybe one difficulty we're running into in these days of modern times, when the major sins of commission are >-100db for electronics, is that we're out of practice at searching for sins of omission. We think pretty exclusively in terms of added distortion, added noises, added modulations, etc. Even strange effects like capacitors are modeled pretty exclusively as added delayed signal, even though it's obviously bogus.
It's a way of thinking that's served us so well that we almost can't think about sins of omission. Loudspeaker work, sure; electronics, not so much. And yet, some large amount of what folks complain about in print on the "sound" of electronics could be read as complaining about omitted parts of the sound. But do we yet have any measurement that directly correlates with *loss*?
I'm currently working on upgrading my testing hardware here at the shack. Currently have two (!) Sound Tech's, a 1700B and a 1701A, both with IM, but need to add a PC-based spectrum analyzer behind the Sound Tech. I can probably use my little Tascam US144 and RightMark to get started, but any recommendations would be welcomed. Also any comments on the subject of music loss and measurements.
Thanks,
Chris
Just riffing on this thought, maybe one difficulty we're running into in these days of modern times, when the major sins of commission are >-100db for electronics, is that we're out of practice at searching for sins of omission. We think pretty exclusively in terms of added distortion, added noises, added modulations, etc. Even strange effects like capacitors are modeled pretty exclusively as added delayed signal, even though it's obviously bogus.
It's a way of thinking that's served us so well that we almost can't think about sins of omission. Loudspeaker work, sure; electronics, not so much. And yet, some large amount of what folks complain about in print on the "sound" of electronics could be read as complaining about omitted parts of the sound. But do we yet have any measurement that directly correlates with *loss*?
I'm currently working on upgrading my testing hardware here at the shack. Currently have two (!) Sound Tech's, a 1700B and a 1701A, both with IM, but need to add a PC-based spectrum analyzer behind the Sound Tech. I can probably use my little Tascam US144 and RightMark to get started, but any recommendations would be welcomed. Also any comments on the subject of music loss and measurements.
Thanks,
Chris
No, we don't. And much of the reason is that no-one seems to be able to hear this "loss" from amplification when evaluating by ear alone. Almost any current piece of non-high-end equipment (and much from the high end) will pass a bypass listening test, so the information loss has to be pretty damn ephemeral, if not non-existent.
Tough to correlate measurements to ghosts- rather than spilling ink, perhaps those folks should direct their energies into trying to set up some real listening tests so that they have something to back up their writing.
I'm afraid that the mere mention of isteningla eststa will restart the shouting. Nobody can agree and it goes nowhere. I'm still kinda hoping that some actual hardware measurements that some actual one of us might make might point down a new path.
You're a tube guy, aren't you? Can't be *just* because you have a lot of sockets left over from olden days. There's some there there. But what is it?
Much thanks,
Chris
Indeed, yes. And if I could trade my car for anything, it would be an old Datsun 2000 (or even better, a Nissan Figaro, but I don't think they're legal in the US).
Hi,
Ix-nay on the oopid-stay...
isteningla eststa? Umkay Oninnay, the Otterway's Inefay!
Ciao T
Ix-nay on the oopid-stay...
isteningla eststa? Umkay Oninnay, the Otterway's Inefay!
Ciao T
Chris, you are starting off well, but you WILL run into a dead end. However, IF you can get signal averaging, it is much easier to use the ST equipment at normal working levels for the device under test. You can also improve the front ends of these ST analyzers by replacing the input IC's with something much quieter.
Not that there is much wrong with the equipment that you can afford, but because it will not test for dynamic time errors, do we have a problem. This seems to be the direction we need to go in. Of course, it is difficult to convince any manufacturer to make something that shows dynamic time shift with signal level and frequency, but that seems to be where the problem lies.
Not that there is much wrong with the equipment that you can afford, but because it will not test for dynamic time errors, do we have a problem. This seems to be the direction we need to go in. Of course, it is difficult to convince any manufacturer to make something that shows dynamic time shift with signal level and frequency, but that seems to be where the problem lies.
Last edited:
End of the story: when I was able to replicate the same pulse from my amp like the reference amp, the two amps sounded exactly the same!!
I do believe in your conclusion, in spite that it contradicts to one of the main suppositions in audio science, that linear distortions (individualt output harmonics are time shifted with respect to harmonics of input signal) are not audible.
Correct reproduction of and individual shaped pulse, means that both linear and non-linear distortions are absent.
To add a bit more, we should imply that the individual pulse must be correctly reproduced at various signal levels, including mV and uV, and the last can put difficult questions regarding NFB effects.
I prefer to stay in time domain because, according to my tests, I'm able to better correlate what I measure with what I heard. Matter of taste.
What I wanted to measure was the modifications of the signal envelope; when I started this journey the digital measuring instruments was affordable only to professional engineers. Today it's another story.
At first I wanted to use two notes played on my piano and captured with a mic, but I failed to manage them: the IC based envelope detector I built introduced others defects...
The transient signal I used had some special properties: it was not simmetrical, transient and I made it non periodic by using a single pulse (see image): it was quite similar to a musical note!!
By looking at the time domain shape modification of this pulse I got the "time distortion" qualitative informations I needed. I used the response of the DUT at this signal only on a comparative basis: I needed to get the same response from my device as from the reference. Effectively, I thought, I'measuring the pulse response of the device and this lead me to the first great mistake: to get the same pulse response I have to replicate the frequency response of the reference device, I said me. Nothing more erratic!! This is only theory that works within simulators. But, in real life circuit, the pulse response is affected by the power supply too, by the wiring and all other sort of parasitic components in the real circuit built (capacitances and inductances). The task to replicate a pulse response with a different circuit is not so easy!! Have you ever trayed?
I learnt a lot from this experience.
Well, at the beginning, this was only a sort of theory for me, so I had to validate it. I took a more manageable circuit, a line level pre, and started the tentative to make a "clone". The candidate was the pre that I found as the best sounding in my system: a Threshold Fet Ten (of Nelson design). I built two circuits: one with an ECC88 and one with an LF411. I succeed with the tube circuit only. The reasons are that the open loop topology of the tubed circuit give me more freedom in shaping the pulse response; with the IC I had to fight with its internal constrains and I failed at the end.
Attachments
Did you occasionally try to compare serial reg PS with shunt PS, in respect of their possible influence on an individual pulse reproduction?
No, I don't think so.
I'm only making two devices have the same pulse response on a time basis.
I made an iteresting experiment many years ago: a comparison between two pre with the same pulse response but with very high THD difference (a discrete op-amp with high NFB and a 2 Fet open loop with 1% THD): I was unable to distinguish any differences between the two devices with normal musical signals.
I believe that our ear is more sensitive to minimal time differences instead of frequency differences.
When you consider iteraural distance between the ears and make some simple geometrical calculations to extimate the time differences in the path from a distant noise source (i.e. the direction of arrival of an out of sight car on the road), you will discover that the ear resolution is very high.
You are able to immediately figure if the car is in front or behind you and if it's coming from right or from left.
Yes, of course.
In my experience I prefer to work with No NFB circuts, because it's more easy to shape the pulse response than with NFB circuits where, for stability matters, you have less degree of freedom. But this is only my limit.
No, never. I had the plan to make this test, because shunt regulator always intrigued me, but my free spare time is very limited actually.
I got good results with simple capacitance multipliers, both BJT or Mosfet: by correctly choosing the base/gate capacitor I got the right pulse control.
Saying that non-linear and linear distortions are absent, I meant that they are not present in big amounts, i.e. they are not ignored, like it is usually the case with linear distortions. Simply I wanted to note, that for an individual pulse reproduction, linear distortions are of the same importance, as non-linear.
You also mentioned easily observable effects of power supply, wiring and PCB tracing for the pulse reproduction. In this connection, using of shunt reg PS could give quite predictable improvement, since shunt PS arrangement cuts out to definite instance an influence of PS and wiring.
Last edited:
The cap multiplier is also my preferable solution, but in combination with shunt PS following it. The current source, being a part of shunt PS, palys an important role for minimizing electro-magnetic interference of wiring.
Ivigone, now i understand better. You feed in an only positive going pulse in the form of a raised sine or approximation of a Dirac. In that case i must disappoint you. This is nothing new. I use this test since the late 70th and learned it from Herr Manger. In the old days we made that with an analog pulse generator and a scope. Modern digital analysers like the DAAS have that implemented as a raised cosine waterfall. I will post pictures of other dynamic tests too. Unfortunately that does not give any new information. Frequency and time are entangled by the Hilbert transform so measuring the transfer function ( amplitude over frequency plus phase ) gives the complete information. You could only argue that the measured bandwidth is too small because of the confines of the sampling rate. That is the reason Daniela Manger still uses an analog stimulus generator with wide bandwidth. Provided your opamp stage has much higher bandwidth then the stage you want to copy by simply bandwidth limiting ( although with a particular shape ) you can copy the impulse response of the slower stage so i could destroy your argument about that problem too. Concerning harmonic distortion i agree with most that you say. Even under the most stringent tests the ear is not better then picking up 0.4%. I found a paper somewhere that sums up what we know about that. When i find it i post it.
- Status
- Not open for further replies.