planet10 said:
I've heard this notion before, but I'm afraid I don't get it. Maybe if you could define "blurring" in math terms it would help. I'm not saying there is no such phenomenon, just that I haven't observed it, heard it, nor can I imagine a mechanism. If there is one, let us know.
There was a fine article published in either Glass Audio or AudioXpress several years back by a group of authors about paralleling tubes. They showed, using both math and some tests (if memory serves) that the combined transfer functions of paralleled triodes have no peculiar traits. If I recall it wrong, I hope someone will correct me. I've got my back issues stored in boxes. Maybe I can find the article.
Hi Max,
The SEWA needs around 20. Gives some headroom for quiet sources.
You can use it with anything. Just reduce the gain a wee. This can be done by either a -6 to -10dB pad or by replacing components.
By replacing the LED's and 470 ohm resistor with 680 ohms and making the 10K resistor 1K. Adjust power supply resistor accordingly. Disadvantage of the component change is is higher output Z and increased noise. Personally I'd use the atten. pad.
maxw said:
The SEWA needs around 20. Gives some headroom for quiet sources.
maxw said:
You can use it with anything. Just reduce the gain a wee. This can be done by either a -6 to -10dB pad or by replacing components.
By replacing the LED's and 470 ohm resistor with 680 ohms and making the 10K resistor 1K. Adjust power supply resistor accordingly. Disadvantage of the component change is is higher output Z and increased noise. Personally I'd use the atten. pad.
I'm not arguing with anyone’s subjective impressions, but I don't think we've got the technical reason in what Dave has said. Transfer functions usually relate Vout to Vin (or sometimes Iout to Vin). The slope on any point of the transfer function is the gain (dVout/dVin). Curve, bends and ripples in the transfer function create distortion. But the only way that low-level information can be completely lost is if there is a flat spot (zero slope = zero gain), such as with cross-over distortion or hard clipping. Below hard clipping, triodes typically exhibit smooth and monotonic transfer curves with a gentle curve. The curve gives rise to the distortion character dominated by second and third harmonic that is easy on the ears, with only a bit of higher harmonic content. If you add together four similar but slightly triode different transfer curves, the result is a new transfer function, also fairly smooth and monotonic, which will have a subtly new distortion character. This is what is likely to be audible, since the harmonic series may have shifted. But I don’t see how there could now be a flat spot, which is required for the output to disappear altogether for low level signals. Dave, I think that maybe your concept of an envelop which is somehow defined within the spaces between the 4 drawn curves is throwing you off. When the four curves are combined into one new transfer function, there is no envelop. I’m really not trying to be relentlessly argumentative, but either I want to be corrected and learn something new, or dispel what I suspect might be one of those “urban myths”.
Brian Beck said:
try it ,and let your ears-brain combo decide - what's better for you.
you don't really need technical explanation and argumentation.
listen stage with one toob in setup ( with x Ua,xx Ia and gate stopper) ,then make 4 or 6 paralleled stages ,with exactly same x,xx ,and gate stopper per system.
you can try mutual and separate biasing schemes (common cathode resistor,separate cathode resistor) to deepen test a little....
been there ,done that
in case where one system struggled with load , ART (say it like that) approach is clearly better sounding;
in case where one system already have enough muscle for load,one system is clearly better.
from analytical point, or even speculative or even philosophic point-there are numerous ways and analogies which we can use for
either "side" ,pro at contra .
but my engrish really sucks for this task
EDIT:
hehe....not "gate stopper".......grid ......naturally
Brian Beck said:
A transfer function also has a frequency axis, so you don't really have a curve you have a 2D manifold in 3-space.
When you average 4 TF together you get a single transfer curve but it now has a downward dynamic range limited to those signals larger than the difference between the outer edges of the envelope defined by the original TFs. Each tube still has its own unique TF and outputs according to that... then it passes to the summer which blenderizies everything below a threshold defined by that TF envelope. The closer you can match the tubes over all the points in that 2D sheet, the better your DDR... but then come back in a month and that matching no longer holds because the tubes have aged at different rates.
On a gross scale as usually extracted by measuring kit, you won't detect it by the usual means... but it is still there... you need to imagine (or measure) what is happening 30 or 40 or 50 dB down from the reference level.
A grosser analogy that might make it clearer... in big SS amps the outputs have to be matched to even run...
Zen Mod,
As I said, I am NOT debating the subjective impressions. I am interested in the technical explanation that Dave offered for them however.
I have tried multiple parallelled tubes in MC front ends. I liked up to four in parallel that we tried. Also, fairly recently, I tried first two pairs of 7119s in a headphone amp output stage and then compared that sound to just one pair. I thought the two pair sounded a bit punchier, but not by a lot. I ended up with just one pair for simplicity. There are a lot of other variables floating around that confound making absolute conclusions about this. I also know that the ART is well received by many others.
As I said, I am NOT debating the subjective impressions. I am interested in the technical explanation that Dave offered for them however.
I have tried multiple parallelled tubes in MC front ends. I liked up to four in parallel that we tried. Also, fairly recently, I tried first two pairs of 7119s in a headphone amp output stage and then compared that sound to just one pair. I thought the two pair sounded a bit punchier, but not by a lot. I ended up with just one pair for simplicity. There are a lot of other variables floating around that confound making absolute conclusions about this. I also know that the ART is well received by many others.
planet10 said:
I hope we can be excused for hijacking this thread for a bit. I figure after 10 pages the original thread has had its run anyway
...I found the reference I was looking for. In Glass Audio 2000/5, authors Kamna, McDonald and Boehlke methodically explore the claim that paralleling tubes does not give good sonic results. They used a three pronged approach: test data, listening tests and mathematical analysis. While not exhaustive, their work was pretty complete and well thought through. Their conclusion was that paralleling tubes caused no harm – just increased the power. No “envelops” or “blenderizing” were noted. They also note (significantly) that any one tube can be considered as a parallel combination of smaller sections of tubes. Imagine the long 6SN7 plate structure covering its underlying grid and cathode. Imagine slicing it into 4 smaller sections through planes perpendicular to the longitudinal axis of the plate. Each section is a smaller triode, with nominally the same mu, but with 1/4 the gm and 4 times the rp. Each section would be slightly different, however, since mechanical tolerances are not perfect across the length of the structure. But since they are butted together (in parallel) we get one tube that combines these slight differences into one working 6SN7. Nothing untoward is observed in the combined result. If paralleling separate tubes is inherently bad, then we’d have to explain how one tube can work.
I think that if there are any parallel tube effects, these would be at the large signal extremes, near cut-off or at the onset of grid current, where the “worst” tube of the group stubs its toe before the others. However, for low-level signals, you can zoom in to look at any smooth curve (a curve that can be defined with a finite set of polynomial coefficients) very closely, to see what a low level signal sees, and you’d see only an increasingly linear tiny piece of a bigger curve.
Dave, you certainly could develop transfer functions at all frequencies and make a 3D shape out of them. Then you’d have surfaces for each tube that can still be combined by summing/averaging. And the result is just another surface, with no “envelops” or “blenders” that I can imagine. Perhaps if you could sketch out what you’re thinking and label the graph axes, or develop some more rigorous mathematical explanation, it would be easier to understand what you mean, especially with regard to frequency effects and “blurring’.
Again, I won’t fuss with those rushing in to claim that they tried paralleled tubes and didn’t like the sound. There are a lot of variables in those kinds of trials besides transfer functions. But I am interested when a new technical explanation is offered which deserves further exploration.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.