That's exactly what I said: if it's about tube (hollow state) MC stage, then transformers are the only option (avoiding SS) to make up a low noise spec. Where is the "rational approach to engineering" in this? Set aside the fact that using tubes today is as much a fashion as the Blowtorch.
And $200 for a pair of decent audio grade step up transformers is not what I see every day (at least on EBay and reputable manufacturers price lists).
SS + transformers for a MC is simply bad engineering and on top it doubles the price. Could be the secret for a successful commercial product, though.
I used Sowters, which are a pretty decent brand. If you have a more rational approach to engineering a tube MC stage (that was the context), I'd love to know about it.
There's some good reasons to use tubes there (undefined max voltages, unlike digital, are well-served by large overload margins), though I'm certain that I could build an equivalent solid state unit (though not more cheaply!).
There's some good reasons to use tubes there (undefined max voltages, unlike digital, are well-served by large overload margins), though I'm certain that I could build an equivalent solid state unit (though not more cheaply!).
Yes tubes are tough for direct MC, although the midband noise is not as bad as some suppose (a 40mA/V supertriode is about 940pV/sq rt Hz). The killer is the 1/f noise, which is terrible given the huge low frequency boost of RIAA.
Far more practical is paralleled low rbb' bipolars, although the assuring of low input bias current is tricky, as they must be run at healthy collector currents, hence fairly substantial base currents, to get the half-thermal of r sub e down at or below the rbb' thermal noise. Also the best parts are passing into the realm of unobtainium, although claims are made by some manufacturers that their medium-power devices achieve comparable rbb' to the old reliable Toshiba and Rohm et al. parts.
For JFETs things are of course worse unless the effective gate width is quite large. A ~1nV/sq rt Hz part will need about 100 of its brethren to do 100pV/sq rt Hz. By itself that situation is not all that daunting from the standpoint of expense, since BF862s are only about 20 cents in moderate quantity. So for FETs alone in stereo we're talking about 40 bucks, before of course the rest of the circuitry and power supply and box etc. (and it is easy to spoil 100pV input noise in succeeding stages).
When I set out to please an audiophile friend who believed he'd identified a niche market for a hybrid phono preamp at a roughly 1500 US MSRP, the target of a nanovolt density was deemed adequate for most uses for the moderate-output MC input. His conviction based on his own listening was that low-cost stepup transformers were inadequate sonically, and much as he distrusts sand state he opted for a JFET front end as the pre-pre. For the MM input to the output it was tubes with sand-state assists. I fabricated a global-feedback-free MC stage (catering to that particular audiophile conceit) and actually listened to it a lot as the stepup to a commercial MM stage before a clumsy attempt to move the strung-out stuff on the table led to a loud bang, and having fallen off my horse I cleared the thing away
I finally had the time and courage to rebuild the power supply in a way that would lessen the likelihood of another accident, and it turned out the pre-pre and supply were intact --- the bang was a big cap discharging as a wire had come loose and touched something else. But by then I had discovered that the primary reason the MC setup had sounded better, compared to a second setup with a SME tonearm and MM cartridge, was that the arm had never been securely tightened down to the turntable
Far more practical is paralleled low rbb' bipolars, although the assuring of low input bias current is tricky, as they must be run at healthy collector currents, hence fairly substantial base currents, to get the half-thermal of r sub e down at or below the rbb' thermal noise. Also the best parts are passing into the realm of unobtainium, although claims are made by some manufacturers that their medium-power devices achieve comparable rbb' to the old reliable Toshiba and Rohm et al. parts.
For JFETs things are of course worse unless the effective gate width is quite large. A ~1nV/sq rt Hz part will need about 100 of its brethren to do 100pV/sq rt Hz. By itself that situation is not all that daunting from the standpoint of expense, since BF862s are only about 20 cents in moderate quantity. So for FETs alone in stereo we're talking about 40 bucks, before of course the rest of the circuitry and power supply and box etc. (and it is easy to spoil 100pV input noise in succeeding stages).
When I set out to please an audiophile friend who believed he'd identified a niche market for a hybrid phono preamp at a roughly 1500 US MSRP, the target of a nanovolt density was deemed adequate for most uses for the moderate-output MC input. His conviction based on his own listening was that low-cost stepup transformers were inadequate sonically, and much as he distrusts sand state he opted for a JFET front end as the pre-pre. For the MM input to the output it was tubes with sand-state assists. I fabricated a global-feedback-free MC stage (catering to that particular audiophile conceit) and actually listened to it a lot as the stepup to a commercial MM stage before a clumsy attempt to move the strung-out stuff on the table led to a loud bang, and having fallen off my horse I cleared the thing away
I finally had the time and courage to rebuild the power supply in a way that would lessen the likelihood of another accident, and it turned out the pre-pre and supply were intact --- the bang was a big cap discharging as a wire had come loose and touched something else. But by then I had discovered that the primary reason the MC setup had sounded better, compared to a second setup with a SME tonearm and MM cartridge, was that the arm had never been securely tightened down to the turntable
Yes the original notion of "shot noise" was based on the particulate nature of charge carriers, after the electron was discovered, and the idea that they slammed into plates and the way that this represented a noise when they were completely uncorrelated and had a small finite charge per particle. A great IRE review article from the late Bernard Oliver and reprinted elsewhere in at least two books is probably accessible online: Thermal and Quantum Noise.
I don't know the etymology for certain, but I'd heard that the "shot" nomenclature was based on very small number of particles phenomena and not named after the first to produce an account of it, Walter Schottky (at least he gets a type of diode named after him). See the very good wiki: Shot noise - Wikipedia, the free encyclopedia
Those guys, including Oliver, that invented "information" are the spookiest of our species. They touched origins and entropy, and lived to tell the tale. Thanks for the pointer; I'll try to find that review article.
For those folks too young to have been raised on the Big Red Bible (Radiotron 4), "shot noise" was used to mean thermal/Johnson/Nyquist noise in vacuum valves, and calculated as the noise of a resistor of 1/gm at 1000 K (although not expressed exactly that way - they used a 2.5 multiplier). Modern use of the term "shot noise" is a statistical thing, arising from the quantized nature of charge. What Brad has discovered for us is the interconnection of the two. Very cool stuff. Must have been out there all along but I never knew it.
Again, much thanks,
Chris
But you say you like the sound of valve amps so presumably that means you like the sound of transformers.
You seem to be saying that they are theoretically worth abandoning ( in most cases ) but subjectively worth including.
Just looked up the Lundahl prices, a pair of LL1678 is USD 159,- for example.
I pulled out my copy of Sze, Physics of Semiconductor Devices, early this AM to see a bit more about Schottky and others, and in particular to make sure there weren't two of them (the shot noise one and the diode one). What was fascinating was the theory of the rectifying metal-semiconductor junction, and the references and some resemblance to thermionic emission. Sze is great by the way, although a bit out-of-date understandably --- but one of those "if you have only one book..." I suppose by now there may be a later edition that the second that I have.
That was a great time to be in the field (npi). It recalls Dirac's remark about physicists of the day, when so many wonderful insights and results abounded that even the not-fully-first-rate folks could get exciting results. Oliver was a wig though to be sure, albeit one who suffered fools less than gladly according to accounts. The review article is in Proceedings of the IEEE, 53 (5), May 1965, pp.436-454, and is reprinted in The Selected Papers of Bernard Oliver, Hewlett-Packard, 1997, and in an earlier and very nice collection, Electrical Noise: Fundamentals & Sources, ed. Gupta, from 1977, ISBN 0471031166.
Thanks again. That's very kind of you to help so much. I've ordered the Selected Papers, and Ginny's working on getting me the IEEE ed. Gupta through inter-library (currently too expensive for me). I'm only self-educated, so finding things like this are especially valuable to me.
Thanks again,
Chris
Thanks again,
Chris
Tubes have several advantages, including: Low and linear input capacitance.
Tubes can also be made VERY LINEAR, per stage, with the right tube, and the right adjustment to the circuit.
Some tubes are pretty darn quiet, but 12AX7's and others of their kind, are NOT!
Tubes will ALWAYS have more voltage noise than the best solid state. It is probably a matter of operating temperature, as well as Gm. Transformers are a necessary 'evil' for power amps, and a useful step up device for phono or mike stages.
Tubes can also be made VERY LINEAR, per stage, with the right tube, and the right adjustment to the circuit.
Some tubes are pretty darn quiet, but 12AX7's and others of their kind, are NOT!
Tubes will ALWAYS have more voltage noise than the best solid state. It is probably a matter of operating temperature, as well as Gm. Transformers are a necessary 'evil' for power amps, and a useful step up device for phono or mike stages.
The 'official line' is that shot noise is not thermal/Johnson/Nyquist noise but genuine shot noise - the statistical thing - but smoothed by the cathode space charge. In the early days of valves it was thought for a while that the noise was thermal noise from the anode resistance, and people argued about it. Eventually it was agreed that it was not thermal.Chris Hornbeck said:
However, in calculating the space charge smoothing factor the term in 'e' (the charge on the electron) disappears and a kT appears so it begins to look a bit thermal. In a noise diode there is no smoothing so the 2eI remains untouched and pure shot noise can be experienced.
The 'resistor' (i.e. equivalent thermal noise) arises because it is convenient to refer the noise to the grid input circuit. Before this is done the expression for noise power has a gm in the numerator. Referring to the grid circuit means dividing by gm^2 (as we are talking power) so we end up with gm in the denominator. The 2.5 on top is an approximation which is roughly true for many valves.
So the gm originally on top comes from the noise generation and smoothing mechanism. The gm^2 on the bottom is stage gain, when referring anode noise to the grid. If we could somehow separate these two different appearances of gm then equivalent noise resistance would not vary like 1/gm but gm(noise)/gm(gain)^2. Exactly this happens in a valve RF mixer: the gm on top stays as gm and is set by the valve bias etc, while the gm on the bottom becomes gc (conversion transconductance) which is roughly proportional to local oscillator voltage level. So we get 2.5gm/gc^2, which in my view contains more useful information and is more accurate than the usual 4/gc which the textbooks quote for a valve mixer.
Please excuse this little diversion into valve RF design!
Self-education is by far the best in my opinion, and you have clearly done well with it. The one area I think for the autodidact that is particularly challenging is mathematics, and although there are a few key books that are really good for explanations and motivations, a good teacher/tutor can be immensely helpful. In fact I wish I knew a few.
Just cascode input low noise jfet(s) with a 2SC3501 (Vce>350V, 10p at digikey), add cooling for the tranny and you got the same overload margin as in hollow state. Is that not cheap enough?
Set aside that any overload margins over 30-35dB (perfectly achievable with low voltage SS) has nothing to do with any SQ, but with fashion. It's impossible to get more (and run into the rails) from a limited dynamic range vinyl, even if you consider the largest ticks and pops that would mistrack. 30-35dB overload margin is considered on top of the RIAA 40dB.
That would be the excess noise, otherwise 1/f noise is generated in any condensed matter system, that's what I was taught in the solid state physics course. Or perhaps it's the high 1/f corner frequency which is very high in tubes?
So let us consider some of what is special about tubes:
Low and nearly-constant input capacitance, not changing much with signals except under extreme conditions. Contrast with solid state.*
Very low input currents at the grid for a properly-biased device.
For most triodes, a wide operating region where the ratio of transconductance to output conductance (or reciprocal "plate resistance") is nearly constant. This means that if the loading at the anode for common-cathode stages is negligible, or the loading at the cathode for common-anode stages ("cathode followers") is negligible, the gain is constant with signal swing. Hence distortion is low.
Very soft overload characteristics, if you are unable to prevent overload, except again under extreme drive (like when you strip away the space-charge layer).
Long thermal time constants of things that may affect the characteristics, usually a good deal larger than ones of the audio signals. Contrast this with solid state devices.
Other cited characteristics are more controversial. Demian mentioned the effects of microphonics as accounting for distortions that some may in fact favor, and I mentioned some stating the importance of large number of charge carriers in play.
Keith Johnson and I discussed the benefits of being forced to use balanced topologies for tube work, to achieve certain optimal results, which then as well ideally enforces equality of positive- and negative-going slewing/settling, to which the ear seems especially sensitive.
There are some truly off-the-wall ideas floating around as well, but I'd rather not have any of this transcribed to appear in the category of metaphysics, though that day may come
*Note that a lot of the capacitances in tubes are extrinsic, i.e. not part of the active region but contributed by structures outside of it. So they are certainly constant (except under mechanical vibration).
- Status
- Not open for further replies.