| smoking-amp |
I'm afraid the Hafler and Keroes paper is a piece of "!%*!*?*" that keeps getting quoted endlessly. It's a little overdue that this mess got cleared up.
I just worked out the output Z formula for UL and it essentially gives smooth variation between the triode Zout (or Rp) and pentode Zout as 1/U% ie. (43% UL => U%=0.43 => Zout = Rp_triode/.43)
The effective Mu for the circuit also varies as 1/U% from the triode to the pentode case.
All H & K's chart shows is the effect of a bad mismatch between the tube and output transformer impedance. If the transformer had been properly matched to Zout for each U% on the bottom axis of their chart, one would just get the normal expected performance for a triode with that effective MU and that effective Zout. There is NO optimum U%.
There is nothing special about the 43% or 20% in their charts except for the particular transformer and tube they happened to use for that test. The optimum point on their graph is where it just happens to be matching that tube.
Another serious fault launched by that paper is that one can get away with using the same DC voltage for the screen as for the plate B+. The sharp dropoff in plate current on the left side of all
pentode characteristic curves is due to screen capture of plate current when screen voltage is near plate voltage.
Just using a tap off the output primary for the screen causes distortion to set in seriously just around idle conditions. This means the first watt is literally ruined. Hence the bad rap UL got for sound.
The screen should be lowered by at least 50 to 100 volts below the plate to avoid this distortion. A separate winding for the screens is one way to do this, but now-a-days one can just use a resistive divider off the plate driving a low capacitance MOSFET follower for the screen.
Sorry to sink a few sacred ships, but this needed to be corrected.
Don |
|
|
| Cobra2 |
| quote: | Originally posted by smoking-amp
----
Just using a tap off the output primary for the screen causes distortion to set in seriously just around idle conditions. This means the first watt is literally ruined. Hence the bad rap UL got for sound.
---
Don |
Ouch...We all know wich Watt is the most important...
Arne K |
|
|
| SY |
| quote: | | Just using a tap off the output primary for the screen causes distortion to set in seriously just around idle conditions. This means the first watt is literally ruined. Hence the bad rap UL got for sound. |
Don, do you have any distortion measurements to support this? It seems plausible for sure, so I'd think it would be pretty simple to set up and demonstrate or dispose of (depending on experimental results). |
|
|
| smoking-amp |
Hi Sy,
I haven't done any measurements yet on this, been busy calculating Mu's and Rp's for UL and CFB. Quite illuminating though. I do have the parts handy however to do some measurements. Give me a few days to set up and I will get back with some spectra. (The weather just cleared up here, sunny and warm thru the weekend, so I may be delayed till next week on this!)
I will use the new Fairchild FQP1N50 MOSFETs I discovered recently to buffer the screen voltage from a resistive plate voltage divider. (only 115 pF input C which will drop to 10pF or so effective in follower mode)
I'm thinking that the best test would be in SE mode so the even harmonics will show up too.
Don |
|
|
| EC8010 |
| quote: | Originally posted by smoking-amp
The screen should be lowered by at least 50 to 100 volts below the plate to avoid this distortion. A separate winding for the screens is one way to do this, but now-a-days one can just use a resistive divider off the plate driving a low capacitance MOSFET follower for the screen. |
I wondered if it was as easy as all that, so I found an envelope and scribbled on it.
For PP EL84 (convenient example). We would have 10W into 8ka-a. That means 283VRMS across the primary, or 141VRMS at one end. 141VRMS = 200Vpk and that 200Vpk is sitting on the 300V of the HT, so the anode swings to 500V. If our source follower has its drain connected to the 300V HT and we don't like the way input capacitance rises when Vds drops <30V, then it can only swing to 270V. 270V/500V = 0.54. Or to put it another way, we need a potential divider that effectively gives us a maximum of a 54% tap, setting Vg2 = 162V.
Nothing looks unworkable so far...
Edit: From where do you get your Fairchild MOSFETs? |
|
|
| EC8010 |
| A further thought is that it might be nice to split the AC and DC conditions of the source follower... |
|
|
| SY |
| For a first watt sort of test, where screen current is constant and low, a bypassed voltage divider might be a quick way to test the hypothesis. Not as elegant as the follower schemes, but will certainly show if they are worth pursuing... |
|
|
| SY |
| One more thought- it strikes me that the distortion from Don's proposed mechanism ought to be second (or at least even) order, yes? |
|
|
| smoking-amp |
Hi EC8010,
I think I bought mine from Mouser: FQP1N50 500V
http://www.mouser.com/search/refine.aspx?Ntt=fqp1n50
Reasonably priced too.
There is also a P-channel marvel too:
FQP1P50 500V
At least an order of magnitude improved specs over any other HV
P-channel part I've seen (there's only two others)
Yes, a good idea to keep at least 30 V across them for low capacitance.
I think I have one EL84 around, I mostly have 6HJ5 horizontal outputs which require lowered screen voltages anyway. I'll try and find it for testing SE UL EL84 case. I do have some 6L6GC's around too.
Don |
|
|
| EC8010 |
Experiment beats theory every time.
My lab still isn't up and running - otherwise there would be FETs under the soldering iron. |
|
|
| smoking-amp |
"nice to split the AC and DC conditions of the source follower..."
Yes, exactly what I had in mind.
I've got 100 FQP1N50's on the bench ready to fire up. No problem going MOSFET follower mode here. (and 100 FQP1P50's on the bench for making hybrid Positronic tubes later on!)
Don |
|
|
| EC8010 |
| Thanks for the source of source followers. They have 250 less than they had a moment ago. International postage being what it is, there's little point in ordering 100 (my default quantity for silicon fuses), and 250 allows for a few cock-ups. |
|
|
| EC8010 |
| I've just had another thought. Assuming that experiment confirms smoking-amp's theory, varying the % of the g2 tap should change ra and thereby change the optimum load for maximum power. Looking at it from the opposite direction, you just pick whichever output transformer is handy, work out the % tap required to match your output valve to your load reflected through that transformer and you're away. You can now use any old transformer with any valve... |
|
|
| Wavebourn |
| Think of feedback. |
|
|
| jkeny |
I'm following with interest the recent posts experimenting with a MOSFET Follower.
Does this approach have significant benefit over the LED or TL431 techniques?
I'm still confused about what sort of voltage drop I should be using on the screen - should it be 50 to 100V below plate as Don says?
John |
|
|
| smoking-amp |
Hi John,
I guess we have gotten a little off your topic here, sorry. For the original triode trick, I think the idea of lowering the screen voltage below the plate is so the screen will absorb less current and dissipate less power. That way the tube can operate with more power overall if its screen dissipation spec was threatened. Maybe some sonic benefits too, don't know. For this trick a fixed voltage drop should be fine and is easy to do.
The MOSFET follower is useful for the case of dropping the screen some % of plate voltage (for Ultra-linear operation), and for this an active follower is needed from some attenuated plate voltage reference. One COULD use a follower to drop a fixed voltage, but it would be overkill.
Hi Wavebourn,
Yes, this is the crux of the matter. Pentode is like open loop gain and triode is like closed loop gain, with lower Mu, lower Zout and lower distortion. This would indicate that full triode, U% = 1, would give the lowest output Z and lowest distortion due to maximum neg. feedback. So H & K's chart would have shown monotonically decreasing distortion as U% increased toward triode (had the transformer impedance been continuously/or equivalently optimised for each Zout).
Hi EC8010,
Yes, I think it provides some new flexibility. It is interesting to look at UL (&CFB too) as a means of changing the Mu and Rp parameters of a triode. One will still be constrained by the max voltage and current limits for the tube, so not all combinations will work out.
And as Wavebourn hinted, lower Mu is equivalent to higher neg. feedback, so this will affect distortion levels. One thing I am curious about is whether distortion will scale as expected with neg. feedback level or whether a given tube will really have some preferred Mu and Z for best results (assuming best or equivalently matched xfmr Z in all cases).
Not sure there is really an optimum xfmr Z either except for power matching. Seems that distortion and power should go down with a higher Z load. And of course, Rp or Zout varies with bias current for any tube too, so I guess we have to specify operation at similar dissipation level to compare apples to apples with any given tube.
My take is that the screen voltage for pentodes provides a means of biasing up current for higher power levels without exceeding plate voltage limitations or drawing g1 current, giving them a practical power advantage over triodes. This seems to still hold in UL mode, even at U%=1 (straight triode) if the screen is rated for a lower voltage than the plate and we can separately bias it up DC wise.
Hi Sy,
Seems likely the distortion would be an even harmonic generator, so this may account for why P-P UL is not painful to listen to. But P-P class AB does convert some even distortion to odd distortion.
Don |
|
|
| Johan Potgieter |
SY,
Gee, don't do that to me at the start of a week-end! (Your post #33) The graphs do indicate the tendency though. I have somewhat different ones for the KT88, but there is a difference in mu(g1-g2) for 6L6 and KT88 - I will still post those when ready.
Smoking-amp,
I am wondering what the most elegant way is to state that you are really taking on a large cut by simply dismissing test results in a sort of back-handed way (your post #36). There has indeed been several quite independant tests (including my own) to verify the tendency shown in SY's graphs. In your rather simplified theoretical analysis I suspect that some of your parameters do not stay constant under dynamic conditions. But I did not analyse in depth so you may fault me on that. Still, to find that practical measured results disagree with your conclusions to the extent that it does, does raise the issue that all is not as simple as simple maths - unless you bluntly want to call all tests deception!
I also have a problem that you state that UL ruins "the first watt". If I understood you correctly; you also seem to couple that to the finding that Vg2 must be some 100V lower than Va to eliminate that. What I read there is then that all pentode/triode (connected pentodes) must also ruin the "first watt", unless you maintain that something magical happens as soon as g2 is connected to whatever % of output transformer tap, that suddenly disappears when g2 is either connected to B+ or anode.
As you know the Quad topology does allow working g2 at a lower potential than Va, and I can categorically state that that does little worth mentioning to the distortion figure (other than the difference in electrode voltages would normally be accountable for.) I hope you will understand my difficulty. We do not require test results; we have them: Apart from Hafler and Keroes, from Mullard, GE, Peter Walker (never mind myself); for EL34, EL84, KT66, KT88, 6L6 .......
So .....? :) |
|
|
| smoking-amp |
Hi Johan,
I am not disputing the charts/data that Hafler and Keroes present, but rather their interpretation of them. They present the results as verifying an optimum %UL for minimum distortion. This is true if the transformer is not optimised for Zout versus %UL as their example shows. (ie. they use a fixed primary Z transformer)
My point is that if one recognizes that the tube Rp changes with %UL due to neg. feedback, then the transformer primary Z must be modified versus %UL too, to compare apples with apples. If that is done, then no optimum %UL will show up (other than the expected lower distortion as triode mode is approached due to increased internal neg. feedback). H & K seem to have missed this control on their experiment, leading to wrong conclusions.
Regarding the lower Vg2 (DC), when AC plate current gets captured by the screen grid, it is being put into the primary at say the 43% tap rather than the 100% tap for the plate. This causes a measureable difference in output power and hence distortion.
As the screen is connected closer to the plate % and finally to the plate itself for full triode, this diverted plate to screen current gets put into the primary at a closer tap % to the plate connection, so less and finally no effect is produced. So, yes, the distortion does indeed magically disappear for triode wired configuration.
As Sy pointed out, the distortion from screen current is likely even harmonic and would disappear for near class A operation. Not sure what the biasing was on the Quad.
Don
|
|
|
| Johan Potgieter |
Smoking-amp and I have posted at similar times; I have only now read his latest contribution (at this time!). Perhaps a few further comments.
Firstly, I get the idea that we are "mixing up" to a certain extent UL and triode operation. This thread started off with essential triode operation with the main aim of getting more power by lowering the Vg2 (dc) but not decoupling it from the anode ac-wise - fine, where that will be considered usefull. But that is not UL (and I have not reread all posts); from what I saw (on rereading), the operation stayed with g2=anode, ac-wise. The first diagram I saw of an effort to make a g2-tap was that by EC8010 (post #41). Incidentally, EC8010, part of the UL topology is driving the g2 with a low impedance, since there is a g2 internal resistance (similar to ra) which is not high - remember it is essentially triode. So it will be difficult to drive g2 ac-wise as you suggested, from a resistive divider small enough to drive g2 but high enough not to load the output transformer. (This is part of UL output transformer design problems - I mean leakge to the g2 feed etc.)
What must also be remembered is that changing the gm and ra with feedback (i.e. as measured externally), is not necessarily changing it "inside" the tube e.g. in order to get lowest distortion. In a manner of saying, the tube does not know that there is external feedback, and will maintain its specific (internal) performance whether there is feedback or not. Logically NFB will lower whatever distortion is generated inside the tube. But it must be understood that we have here 2 separate factors both to be optomised if one wants to analyse that finely.
Then with regard to output impedance matching, define "matching"! It is common knowledge that for maximum power output the load must equal the internal (generator) impedance. As it happens that is also the condition for maximum distortion. Tube graphs will indicate that usually somewhere just before the output reaches a maximum under specific conditions distortion is a minimum. But when one analyses the distortion, 2nd harmonic has a dip there, while 3rd harmonic keeps on rising, etc. (this for a pentode - triode is different). The scene is rather complex and in that sense there is not a simple optimum. The best is to use manufacturer's graphs or draw one's own and make up one's mind.
Some good news with tubes is that there is mainly lower order distortion, so the solution is likely to lie somewhere where the rate of change of the totaln distortion graph is still lower than the rate of change of the output power (both with regard to load impedance). But again, depending .....
Regards |
|
|
| Johan Potgieter |
Busy night!
Your last mail understood, Smoking-amp; few problems there. Perhaps some comments later - it is 04:00 in the AM here, and I want to try to get to bed before the sun comes up. (Off-topic: wickedly hot daytimes here in RSA these days, thus my night owl activities.)
Thanks for interesting debate thusfar! |
|
|
| smoking-amp |
Hi Johan,
I agree that the mixing of two (actually three, I think "enhanced" or "super" triode with boosted Vg2 even got mentioned too!) similar sounding but totally different schemes in this thread is causing considerable confusion here. My fault for commenting on the H & K paper in an inappropriate discussion, I should have started another thread. We have probably left John (the thread starter) in a serious brain spin, my appologies.
And yes, I agree on the confusion over "optimum" matching impedance. An optimum only showing up for power transfer, and at a bad match for distortion at that. What I should specify here regarding the H & K paper and UL discusision, is some factor say "Tau" times the tube Rp to be conserved for apple to apple comparisons.
Maybe Sy could split this UL/ H&K issue off into a new thread?
Maybe call it "Heretical Ultra-Linear thinking" or "Holy Cow under attack!"
Don :) |
|
|
| tubelab.com |
I just got home and saw two pages of new posts on this topic. I have been quietly experimenting with a technique that allows a continuously variable knob that goes from triode to pentode (variable UL percentage). I started with a circuit that bears a striking similarity to EC8010's napkin. In fact it is exactly the same.
There is a minor problem with that circuit. The plate is working into an inductive load (the OPT). Because of this the plate voltage will swing above the B+ voltage. If the pot is in any position other than pure pentode the mosfets gate will swing above the B+ voltage (the drain of the fet). This will cause extreme distortion, and possibly blowing the silicon fuse (mosfet).
There is a solution to this issue, and it involves sprinkling a little more sand into the circuit. I am including a screen shot of the schematic. This has been added into an LTspice simulation of a SimpleSE amplifier. Everything works fine in the simulation world, but I haven't had the time to build it yet.
I tend to give circuits like this dumb names like PowerDrive. I call this one VariLinear. I have LTspice simulations of the unmodified SimpleSE and a SimpleSE with VariLinear. I have another simulation that seperates the DC and AC signals in the manner shown on the napkin. I haven't found it yet. It is on my old computer. I can post these if there is any interest.
There are no potentiometers in spice, so the pot is modeled as two resistors. You can play with the values to see the effects. If the resistor values are too low they will dissipate some of your valuable output power. If they are too large they will interact with the fets capacitance to cause a phase shift at high frequencies, leading to higher order distortions (ugly).
|
|
|
| smoking-amp |
Hi Tubelab,
Excellent point about the B+ being exceeded. I like your diode commutating solution too. I was originally thinking of just using a higher voltage source for the drain, but that would stretch the voltage rating of my 500V MOSFET to the limits. Voltage excursions for the MOSFET are likely to be a problem no matter what approach with only a 500V rating available.
One can also make a second voltage divider with a 50% divide point that splits the voltage from the drain supply to screen voltage in half. Then using two MOSFETs in series, the second one has its gate connected to the halfway point. Each MOSFET then drops half the voltage at all times. This idea can be extended to more series MOSFETs too, but two should be fine here.
Another problem for this test will be to come up with a variable range of load resistances for the tube plate. My output transformer stock here is very limited, and it might be best for this distortion spectra testing to avoid a transformer altogether. Using a bunch of power resistors could work, but will require twice the B+ supply voltage to simulate the voltage doubling of a transformer load. I have 600V available for B+. Also, it would be nice to be able to continuously vary the load resistance.
Still working on it:
There is a technique for making a variable power resistor using a MOSFET, source resistor, pot, and OP Amp. The Op amp compares the voltage on the MOSFET source current sampling resistor with an attenuated sample (using the pot) of the total voltage across the MOSFET. The Op Amp controls the gate of the MOSFET so as to maintain the pot attenuated sample of total voltage drop equal to the voltage on the source current sampling resistor. That way the total resistance is 1/(atten. ratio) times the source sampling resistor. This whole circuit has to float around at the plate voltage. This may require some effort to get debugged and stable. The Op Amp is not going to like flying around 500 V or so.
Another approach is to make a CCS for the plate that has some variable resistance across the CCS MOSFET to its gate reference voltage. (another resistor in the MOSFET source to B+ to set current) This makes the CCS current increase with voltage across the MOSFET, which looks like a resistance then. This is simpler, but does not give a perfectly linear resistance versus drain voltage swing due to the MOSFET gate drive being summed with the voltage reference. Putting an OP amp in to fix this ends up with the previous circuit. A bipolar transistor would come closer to linear due to its nearly constant 0.6V Vbe drop, but requires a lower value resistor divider from the collector (ie. 7W 100K Ohm Pot). Maybe a Darlington to reduce current drive.
Don |
|
|
| poynton |
The discussion so far seems to be concentrated on SE. I assume that this technique is equally applicable to PP, PSE etc.?
I had heard that this technique was developed by Tim de P. True??
Andy |
|
|
| EC8010 |
Poynton: It' simply easier to do the tests on SE and the results are more easily interpreted. Distributed load was invented by Alan Blumlein.
smoking-amp: The variable resistance anode load isn't a problem at all. There's nothing to suggest that frequency affects this particular distortion in any way, so all tests can be done at 1kHz. Any output transformer is perfectly capable of accurately reflecting a 4:1 ratio of secondary loads at this centre frequency. So all you need is an adjustable dummy load...
This one is made of non-inductive (essential) metal film resistors. There are three parallel pairs of 10R 50W resistors wired in series with taps at each junction. That gives, 0R, 5R, 10R, 15R. After that are four 1R 20W series resistors tapped at each junction, giving 0R, 1R, 2R, 3R, 4R. Finally, a pair of 1R 20W in parallel, to give 0.5R. The upshot is a load that's variable in 0.5 Ohm steps from 0.5 Ohm to 19.5 Ohm, with a current rating of 4.47A on all settings. You merely need two leads to go to the load, and one link lead to obtain any setting you like. |
|
|
| Bandersnatch |
Hey-Hey!!!,
Don't neglect the higher voltage MOSFET's. FQP2N90 is a 900V D-S beastie, with a 30V g-s tolerance.
I have not measured my amps, but they're all U-L output and with efficient speakers I live around the 1W neighborhood most of the time. I can say that the first watt is *NOT* ruined, and when a few SE amps (which would not be efected by this effect ) are substituted, the PP comes out on top.
Class A operation, conservative operation at ~75% of dissipation ratings may help. The E-Linear driver/FB scheme (also pentode/faux pentode ) may play a part in it.
I have got my output TX with multiple taps for U-L/E-L opertation. The Peerless S-265-Q I had copied by Heyboer happens to have easy end-of-layer access in 10% increments. 20, 30 and 40% taps were installed. I can run the E-Linear front end from one set of taps and g2 from another. It is nice for doing science experiments...:)
cheers,
Douglas |
|
|
| jkeny |
Hey all,
I just logged in to find the continuation of the interesting posts on the "Triode Trick" thread & I found all you big boys have taken my ball and gone to play over here.
So I'm here to just look on from the sidelines & see if I can pick up anything - I'm very interested in the outcome of all your experiments.
My original intention for the thread was to experiment with triode-strapping but this outcome is more than I expect - the possibility of an adjustable UL operation using existing OPTs
Keep up the good work.
John |
|
|
| smoking-amp |
First, thanks goes to our moderator for splitting this thread out. This will eliminate some reader confusion hopefully.
I do have some ST P2NK90 900V MOSFETs on hand, fairly similar to the FQP2N90. Since we are powering the gate from an output tube in source follower mode, the ~400pF gate capacitance shouldn't be a problem. Or I can just use two FQP1N50's in series mode, 115 pf gate Cin then.
As Sy suggested, the monotonically increasing screen current with lower plate voltage swing in SE should produce mainly even harmonics. Operating P-P mode in class A will eliminate these even harmonics without generating much odd harmonic. So this is likely to explain Douglas's amplifier experience. Also there is the effect of global feedback to consider too.
My comment about the first watt being the worst watt was meant for the typical class aB operation of most UL amplifiers marketed. These will cancel the even harmonics, but will convert some to odd harmonics in class aB mode. What may save them is that these "even harmonics converted to odd harmonics" effects in P-P generally show up at larger amplitudes. A SE test should be good at revealing what is going on here. One also has to wonder what effect smoking the g2 grid has on tube longevity.
I have a Hammond 1640SE transformer, but it has a lowish 1.25K OHm primary.
I also have a Hammond 1650T P-P transformer, again lowish 1900 OHM CT primary, 120W. Could put a CCS on the other half of the primary to balance the current.
(note: Hammond and many other xfmr manufacturers rate their xfmrs wattage at close to magnetic saturation. And guitar amp ones at beyond that. When one backs off to where the maximum permeability occurs, these wattage ratings get cut by 1/2 to 1/3, so this is more like a 50W "reality" Hi-Fi transformer)
And, I have an Edcor CXPP100-SP1 P-P transofrmer, 5K OHM CT primary with isolated 20% UL/CFB windings (the other 80% windings are split/isolated too), 100W. CCS on the other half again to balance current. (Again, 40W "reality" Hi-Fi transformer)
I have an 8 OHM load resistor, 100W, with 1 OHM taps.
Also, have a 24 bit M-Audio sound card for spectrum measurements, and AudioTester software V2.0
Probably the Edcor is the way to go here for this testing.
OH, and welcome to the discussion John, I didn't mean to leave you out, I just thought we were hi-jacking your original thread.
Don:) |
|
|
| smoking-amp |
I have rounded up the Edcor CXPP100-SP1 (5K OHM CT P-P) output transformer for the testing. I will use an ST P2NK90Z MOSFET for the balancing current source on the second half of the P-P primary. (so SE mode for the tube)
I have three pentodes that allow high screen voltages: 6BQ5/EL84 (it has a suppressor grid), 6HB6 (have both suppressor and beam versions), and 6L6GC (beam ). I will use an Optimos SS amplifier to amplify the sound card signal to grid1 drive levels. And I will use a 2nd ST P2NK90Z MOSFET (900V) to buffer (source follower) the g2 drive from an attenuated plate signal with separate DC level adjust. (I will use a 600V supply for the MOSFET drain supply, explained below.)
Some comments:
The screen current is not re-inserted into the UL primary tap with this adjustable %UL scheme. Hafler and Keroes' article correctly comments that re-inserting the screen current into the UL tap reduces distortion. The BEST (perfect actually) solution would be to re-insert screen current from the MOSFET drain to the plate/primary connection, but voltage levels do not cooperate on this.
(This is an area that requires some additional work for this variable %UL scheme to succeed in the longer run. I was going to use Tubelab's diode cummutation scheme for the drain connection, but later realized that this will cause abrupt distortion as the drain current switches from B+ to plate/primary connection. So some re-thinking called for on that approach. )
This all points towards the necessity of reducing the DC level on the screen to avoid this non-linear screen current capture of plate current whoes in the first place. I might also add that the usual CFB setups also suffer from screen current distortion problems, but at least they provide for easy lowering of the screen DC level to minimize this.
I will take a spectrum and scope trace of the screen current while testing, so we can see just what we are dealing with.
I just read thru H & K's article again and also Williamson & Walkers article "Amplifiers and Superlatives". It is now clear to me how H & K arrive at their "optimum" %UL.
Looking at their graph (fig. 2):
http://www.tubecad.com/Classic_Articles/page4.html
The High Level distortion curve is the main curve really confining their selection, although the power curve does some too. This huge upswing in distortion is caused by g1 grid current drawn, trying to maintain the power output curve (which is really an arbitrarily selected curve set by their input signal level). Does anyone really believe that triode distortion is off the chart as they present it? They simply overdrove the triode to get this. (I don't see how they could keep a straight face and present this data!)
This is all due to the fact that the fixed transformer primary impedance is too high at the triode end of the chart where there is lower tube Mu, to get much power output.
The power curve as shown should really be dropping off faster toward the triode mode end if they had avoided overdrive, since the fixed impedance primary is becoming progressively too high to get much power out of the triode.
The Low Level distortion curve presented shows the more realistic case for distortion trends when overdrive is avoided.
The rapid rise of low level distortion toward the pentode end is due to the fixed primary impedance causing the high-Mu pentode to enter 3/2 power current operation into a low impedance (relative to the pentode's Rp). One can get more power out in this regime, but it depends on additional circuit gain elsewhere and global feedback to fix it then.
If one were to recognise that the tube's effective Rp (see the internal impedance curve in fig. 2. The flat top on the pentode end is due to the small plate effect taking over from the larger screen effect when %U goes to zero, the rest of the curve goes as 1/%U) and the tube's effective MU, varies closely as 1/%U, then a diiferent experimental setup would be indicated.
Normal triode design uses a plate load of some factor, say Tau, greater than the plate resistance. This is a compromise, as lower Tau (till Tau = 1) gives more power, and higher Tau gives less distortion. (The distortion is due to the plate Rp varying with signal) If we were to compare different Mu triode tubes, the only fair comparison is to use the same Tau factor loading for each. Apples to apples as they say.
Re-doing the H & K's chart using constant Tau (ie. the primary impedance Z would have to vary as 1/%U on the bottom axis ) should give constant power out, and constant distortion (assuming the same tube idle current setup) for the same power level across the graph. The pentode end then does not jump to higher power since we are not entering 3/2 power current mode either. (of course, infinite Z primary impedance would be impractical for pentodes, so one would not use low %U taps in practice.)
(sorry to be so long winded, but this UL stuff is complex)
Don |
|
|
| SY |
| quote: | | The High Level distortion curve is the main curve really confining their selection, although the power curve does some too. This huge upswing in distortion is caused by g1 grid current drawn, trying to maintain the power output curve (which is really an arbitrarily selected curve set by their input signal level). Does anyone really believe that triode distortion is off the chart as they present it? They simply overdrove the triode to get this. (I don't see how they could keep a straight face and present this data!) |
That's an interesting point, BUT... for true apples to apples comparison at high levels, their approach is valid, I think. If I have a UL amp that shows 1% nonlinearity at 10W, but the triode amp is already clipping, then indeed I've gotten a significant distortion reduction. Power amps supply power, and for the same power supply and output tubes, if one amp is clipping away and another is still well below grid current, the latter has performed better in that regard. The thrust of their graph is just that- there's a screen ratio where one can get the majority of pentode power, a point at which the triodes are already in full collapse.
To your point, it would have been interesting to pick a power just below the triode clipping point, adjust the p-p loading for each, and use that for a reference, just to see how the distortions compare. But we can get that off of datasheets (especially the excellent ones from Mullard), and at least in the examples that come to mind, the high-level performance of UL still trounces triode, even when operating within the clipping limits.
MUCH more interesting (to me) is what happens at low levels. |
|
|
| smoking-amp |
Hi SY,
Your point is taken. Clearly the H&K paper is from the era when max power output per cost reigned supreme. And the UL scheme does squeeze out the most power from a given output tube. However, this extra power comes at a hidden cost here. Their curves are apparently for their complete amplifier, including input stages. The higher power in UL necessitates greater gain in the global feedback loop to overcome the heavier 3/2 power distortion generated in UL (or pentode) (compared to a non-overdriven triode). The triode circuit might save an input tube for the same distortion spec. and so allow the purchase of a slightly larger output tube.
But I agree, the first watt quality is really the more important criteria. So we test...
Don |
|
|
| smoking-amp |
It just occured to me that the MOSFET follower (for the screen drive) can have its drain terminal connected to the plate/primary connection IF the DC level of the screen is reduced (from the plate DC level) slightly more than the UL% reduces the peak screen voltage swing.
Example: lets say the plate is swinging 100 to 300 V (200V DC average) and the screen UL% is 50%, so the screen will be swinging + and - 50V around its DC level. If its DC level is reduced by more than 50V below the plate DC level, then the MOSFET has some headrome to operate. Let's say DC screen set at 140V, so the screen varies 90 V to 190V. The 90V minimum staying below the 100V minimum of the plate. And the 140V max. staying below the 300V max of the plate. (In P-P the DC MOSFET current component will cancel in the xfmr.)
This scheme won't help for my present testing, since I need to check for screen operation up at plate voltage levels, but this will work for other practical variable %UL amplifiers that use output tubes with reduced screen voltage operation, like TV horizontal and vertical outputs. Ironically, their reduced screen DC level already mostly fixes the screen current distortion problem anyway, but this zeroes the effect altogether. Hi-end specs!
Hmmm...., alternatively, one could put a floating positive boost supply on the plate(s)/primary(s) to power the MOSFET drain, again a little more boost than the %UL reduces the peak screen swing. This WOULD work for my present experiment I think. Have to check this out. Got some nice Power-one 150V, 200V, & 250V linears and some similar Acopian linears setting on the shelf here. Tempting, have to check their float voltage ratings. Where's that isolation xfmr. at?
Don |
|
|
| Johan Potgieter |
I am still hunting for the promised GE UL graph set for KT88 and will not say much now, but the graphs I mentioned earlier gave max. output available (i.e. just before onset of g1 current) at several different primary impedances (also my own tests). These do not at all change the behaviour with different g2-taps greatly, apart from obviously giving different output figures.
Then I am a little perplexed by the concern for what happens at low output. Why would we expect anything disasterous there? Both the triode and pentode operation give expectedly low figures; UL shows a smooth transition from pentode to triode at higher output powers, so why this fear of a sudden inconsistent aberation? We have some Ia/Va characteristics for power tubes at UL operation. Is anything evident there? I fear I fail to see why e.g. distortion products would be say 20mV at pentode side, but suddenly jump up to 100mV or whatever we fear somewhere inbetween, only to come down again to say 10mV. The whole theory as well as tests go against the expectation of any discontinuity.
What am I missing? Forget about H and K if they are unpopular; is there also a reason to doubt Peter Walker, later graphs from GE (KT66 and KT88) and Mullard (EL84 and EL34)? I always both presumed and found that manufacturers will give moderately accurate data regarding their products. I am all for experimenting (I earned my salary as a researcher), but with respect, why do I feel that we are re-inventing the wheel here?
________________________________________________
(Saw I took long to say "not much"! Moderator allowing: It is a little like Churchill: The "Times", once outraged at something Sir Winston did, wrote a 2 column report stating that they are speechless. Said Sir Winston: "The Times is speechless and takes 2 columns to express its speechlessness.") |
|
|
| Johan Potgieter |
| quote: | Originally posted by smoking-amp
Example: lets say the plate is swinging 100 to 300 V (200V DC average) and the screen UL% is 50%, so the screen will be swinging + and - 50V around its DC level. If its DC level is reduced by more than 50V below the plate DC level, then the MOSFET has some headrome to operate. Let's say DC screen set at 140V, so the screen varies 90 V to 190V. The 90V minimum staying below the 100V minimum of the plate. And the 140V max. staying below the 300V max of the plate. (In P-P the DC MOSFET current component will cancel in the xfmr.) |
Smoking-amp,
I totally follow the above (to give a comment closer to what you intend doing than the previous post). And you state this only as an experiment. But how far can we decrease Vg2 in favour of decreasing its dissipation etc, before we loose what we gain by a higher Va? I do not have much data for power tubes at many different g2 voltages, but just off the memory I have an impression that one cannot lower the Vg2 a lot before loosing any output gained by a higher Va. Or are you arguing transmitter tubes and say 400Vg2 and 800Va or such (sorry if I missed this earlier)?
Regards. |
|
|
| Johan Potgieter |
Sorry for over-exposure,
but just before ducking I found the GE graphs for KT88 in UL operation on the following:
www.webace.com.au/~electron/tubes/ulo.html
Go down past the first circuit diagram, and the first graph set is it. It does not state in the caption that it is p.p. KT88, but those are the ones.
Regards
Edited: Spelling in www-address |
|
|
| Bandersnatch |
Hey-Hey!!!,
At the mention of Walker, and if you'll take an IIRC for references...:) his amp was running ~11% U-L thanks to it's 1/9 CFB/tertiary winding. It was also good for reducing 3rd HD more than the amount of FB in the cathode winding could account for.
I wish I could account for all these with good ref's but 'I read it somewhere whilst researching tube circuits' is the best I can do today. The source looked good enough to trust when I read it.
The good performance of Walker's amps matches my own building with only CFB in the output TX. Dynaco, Acro and most recently Chicago Iron has yeilded some nice amps. I have moved to DH amps, so low capacitance filament Iron will be needed to execute the next one, built with 813's and the B0-14.
cheers,
Douglas |
|
|
| smoking-amp |
Hi Johan,
We changed the name of the thread so no one could find us! (Just kidding)
I think I need to explain my position some since some questions keep coming up that I didn't think were applicable to my complaints. First of all, I am not doubting the Hafler-Keroes paper as far as their results for getting more power in ultralinear or the manufacturers data sheets on UL either for that matter. In fact I don't think I ever mentioned power at all in the beginning. Some confusion is likely due to my dropping into an existing thread without having read it first too, to see where everyone else was coming from, and it seems power was a big concern in the earlier discussion.
I simply had finished working out the formulas for the effective Rp and Mu for the UL and CFB cases, and they come out quite interestingly and even beautifully. The results show that UL and CFB designs are absolutely equivalent to triodes. The UL scheme giving effective MU's that are between the triode wired Mu rating and the Pentode effective MU, depending on the %UL tap position.
Effective Rp's also being between the triode wired configuration and the pentode one. Ignoring a small plate term, the UL results are Mu = Mu_triode / (U%) and Rp = Rp_triode / (U%).
The CFB case gives similar results for effective Mu and Rp but for LESS than the triode wired Mu and Rp. So CFB and UL are really complimentary techniques for tuning a tubes MU and Rp.
The Hafler-Keroes paper is annoying (to me anyway) because it does not give any indication of the underlying simplicity of the results. In fact they seem to be at a loss to explain them. And their method of over driving the tube makes any %UL tap above 50% look like a horror show for distortion. They don't give any results for other primary impedances, so the reader is left with the impression that only a 43% tap (18.5% impedance) will ever work. As a result, we have endured years of UL tapped transformers with only 43% taps on them.
So #1 item to be tested is whether other tap % are useful and whether they obey the effective Mu and Rp formulas. Maybe H-K will be right on the 43% only, but I doubt it. If other % taps work as expected, then variable tapping via active screen drive will be an interesting new hobby for some.
My second complaint about the H-K paper is their condoning of just using a tap on the primary instead of a tertiary winding that would allow lowered screen voltage. No doubt one will get more power by scorching the g2 grid at B+, the manufacturers loved it.
My complaint is about likely distortion with this scheme. Screen current skyrockets as it approaches the plate voltage, as can be seen on virtually any pentode plate curves (that's why the plate curves drop off suddenly on the left side, the screen swallows all the current). This screen current is captured plate current that is then put back into the primary at a reduced # turns, so it lowers output power, and will produce distortion if it varies as a percentage of plate current. (and it certainly will from inspecting any of those plate curves again)
Setting the screen DC equal to the plate DC at the idle condition, by using just a primary tap, means that this mechanism is at maximum effect for the first watt. Now it does seem likely that this is monotonically varying even harmonic distortion, and so will be nulled out by P-P operation. But depending on biasing (class A or class aB) Some of this even harmonic distortion will get converted to odd harmonic. For SE operation, I would expect horrible results. I would guess some purists will want to know if P-P is cleaning up some horrible mess and increasing odd harmonics as a result.
So test #2 is to check for grid2 current distortion (spectrum) versus DC level vis-a-vi the plate DC level. And it also would be interesting to see if the earlier mentioned scheme of connecting the screen driver MOSFET drain to the plate/primary point will remove it.
That about covers it I think.
Don |
|
|
| gingertube |
I am about to start build of a couple of monoblocks using Plitron VDV2100-CFB/H Output Trannies I've had sitting around for a few years.
These are:
2000 Raa
33% U/L taps
Separate 10% cathode feedback windings
5 Ohm secondary which is center tapped (another cathode feedback option)
100W rated (-3dB power @ 23Hz)
-3dB High frequency at 290kHz
Am I right in thinking that if I use the 10% cathode feedback windings in the cathodes of the output tubes and the 33% Ultralinear screen taps then I will effectively get 43% Ultralinear anyway (cause the screen to cathode volts will be 10% + 33% being opposite phase)?
Any advise on ouput tubes?
KT88, 6550 would give more power output than the transformer is rated for BUT I have Chinese and JJ KT88 and Sovtel 6550 on the shelf
EL34 is too "Old Hat" although I have heaps of them.
Perhaps KT77? I have some JJs.
Also have a pair of Mennos VDV70/100 monoblock amps (100W version with PAT4006) using triode strapped El34s which could serve as test beds. Now if only I had a distortion analyser!!!!
Cheers,
Ian |
|
|
| smoking-amp |
10% + 33% = 43%
Sounds right to me. This 43% magic # from UL designs is likely only applicable to 6L6 or related tubes (KT88 etc) in UL. Probably not meaningful for CFB particularly. For any tubes that have high Gm2 / low screen voltage ratings, I'm pretty certain this 43% number is too high for UL, but who knows what it does for CFB. The grid1 CFB effect is much greater than the screen UL effect due to the much higher Gm1 than Gm2.
Don |
|
|
| isaacc7 |
OK, this seems like the best time to ask this, I've read in the past Grimwood's ideas about so called optimized ultra linear operation. His claim (and he has no number to support this) is that the "optimum" ratio of UL operation to use with any particular tube depends on the physical geometry of the tube itself. Read it here:
http://www.webace.com.au/~electron/tubes/ulo.html
I have heard that some of his other ideas are a bit questionable, but this article seems reasonable enough. He also claims (once again, with no data) exceptional performance out of tubes that are notoriously difficult to work with in HiFi operations (2e26, 6146, etc.) Is there a way in your testing to see if the so called optimums fall where he says they should base on tube geometry?
Isaac |
|
|
| ilimzn |
| quote: | Originally posted by smoking-amp
I just worked out the output Z formula for UL and it essentially gives smooth variation between the triode Zout (or Rp) and pentode Zout as 1/U% ie. (43% UL => U%=0.43 => Zout = Rp_triode/.43)
The effective Mu for the circuit also varies as 1/U% from the triode to the pentode case.
...If the transformer had been properly matched to Zout for each U% on the bottom axis of their chart, one would just get the normal expected performance for a triode with that effective MU and that effective Zout. There is NO optimum U%
Another serious fault launched by that paper is that one can get away with using the same DC voltage for the screen as for the plate B+. The sharp dropoff in plate current on the left side of all pentode characteristic curves is due to screen capture of plate current when screen voltage is near plate voltage... The screen should be lowered by at least 50 to 100 volts below the plate to avoid this distortion. A separate winding for the screens is one way to do this, but now-a-days one can just use a resistive divider off the plate driving a low capacitance MOSFET follower for the screen.
Sorry to sink a few sacred ships, but this needed to be corrected.
Don [/B] |
Don, all i can say is THANK YOU for this! Without wanting to sound insulting to many here, I am simply not old enough to be that experienced with tubes, especially since there is a large non-practitioning gap with me and tubes between my HS days and perhaps 4-5 years ago - but what you state here, and i have emphasized, is alsmot exactly along the lines of my own thoughts re UL - which, given the H7K paper and the many existing UL constructions past and present, I simply kept to myself, thinking there is probably something I am missing.
It has been advocated here before, that pentodes get more linear with lower screen voltages, which would of course seem almost obvious given what G2 does in a pentode. Various 'kinky' troubles begin whenever t he plate goes near or, worse, below G2 potential. In the classical Ul this is the default case at low power. I think a search would reveal several instances where it was said, with tube curves at hand, that G2 should be kept well away in potential from the plate - this is a simple corroboration of what you are saying. I am looking forward to your measurements.
I also agree with your assertion that there is no universally acceptable 'ideal' UL ratio, and that it depends on the combination of tube and transformer, if you model the various feedbacks as a different kind of triode, this follows automatically. There have been things on the web that deal with this, and one of them was based on the electrode distance from cathode, and (supposedly) resulting Rp per unit area of G2 and plate - both, in a sense being 'plate'. i will try to reference a link, can't find it now, but would love to read your comment on this.
EDIT: someone else was typing about it as i was typing my own relply - see post above mine, or go here:
http://www.webace.com.au/~electron/tubes/ulo.html |
|
|
| Bandersnatch |
| quote: | Originally posted by ilimzn
It has been advocated here before, that pentodes get more linear with lower screen voltages, which would of course seem almost obvious given what G2 does in a pentode. Various 'kinky' troubles begin whenever t he plate goes near or, worse, below G2 potential. In the classical Ul this is the default case at low power. I think a search would reveal several instances where it was said, with tube curves at hand, that G2 should be kept well away in potential from the plate - this is a simple corroboration of what you are saying. I am looking forward to your measurements.
|
Hey-Hey!!!,
If we're to believe the curves, the plate can go a fair distance below g2 before the screen current begins to climb. It is dependant on the type of pentode of course, and some are better than others.
I don't see any decrease in linearity as the plate goes below g2 either. If the load line is reasonable drawn, the 'kinky' behaviour isn't anywhere the load line. Now if one accounts for an inductive elipse, the line begins to blur(so to speak ).
cheers,
Douglas |
|
|
| smoking-amp |
Hi ilimzn, and isaacc7,
Thanks for the encouragement, I will continue to plow on along these lines.
I read the Grimwood Optimised UL article last night as Johan cited this same article for a UL characteristic chart for the KT88. It seems to be a very LONG winded suggestion that the DC fields in the tube should be partitioned in the same way as the AC fields are naturally due to element spacing or geometry. This seems fine as far as avoiding tube voltage breakdown problems, but this will certainly greatly limit the maximum power that can be derived from it's operation. But I think his emphasis is mainly on the quality or linearity of amplification and avoidance of instabilities, particlularly for difficult tubes. He could be right, I have no idea. Some spectrum/distortion tests would seem to be called for to resolve this hypothesis. Grimwood does not provide this (well I didn't read the other UL pages he referenced, maybe he does?), but depends on listening results apparently. I would say this would be a good little test project for someone interested, just need a sound card.
Anyway, I woke up this morning and realized that I have been blab.. blab ...blab.. about my new effective triode formulas, but haven't applied them to H-K's article to see if any insight could be gained about their magic 43% tap. so here goes:
They mention that the power supply current for their circuit reaches 130 mA peak current. Looking into the GE databook for the 6L6 (for two tubes P-P in class A), they conveniently list 134 mA operation for two tubes as giving a gm1 of 5700. The triode wired Mu for the 6L6 is listed as 8. Since Mu = gm1 * Rp for triodes, this would mean the triode wired 6L6 config has an Rp of 1403 OHMs.
Next, they hint that the idle two tube current is 100mA as can be seen from the schematic with 35V across the 350 OHM common cathode resistor. This is clearly an amplifier that is mostly operating in class A operation, if not fully. Even for class A, the power supply draw increases a little at peak power due to the residual 3/2 power law nonlinearity of the loaded output tubes.
Next, they use a 6600 OHM CT output transformer. Which would give 1/4 that Z for each tube in class B but 1/2 that Z in class A (in class A, both tubes are operating simulataneously, so each tube provides 1/2 the current for the observed voltage swing, so each sees the load as 2x)
Using the class A operating mode we then get 3300 OHM seen by each tube.
Now let us apply the effective triode formula for Rp and Mu using UL% at .43 (the winding tap is at 43%, their chart gives impedance ratios on the X axis which is .43 * .43 = .185 or 18.5%)
effective Rp = triode_Rp/ .43
so eff. Rp = 1403/ .43 => 3263 OHM
and eff. Mu => 8/.43 => 18.6 (if we had the grid1 drive signal level, we could compute the maximum power output using this)
3263 OHm eff. Rp sure looks like a close match to 3300 OHM loading Z, this is the condition for maximum power transfer from a triode when load = Rp. I think we have seen the smoking gun here!!
We can now likely compute the optimum %UL for other tubes (with different triode MU and GM1 and Gm2) used with any Z primary transformer. So I think this 43% magic tap is just an artifact of using a 6L6 with a 6600 OHM P-P transformer.
Now if H-K had just done this analysis, I would not be "%*/!^*" on their paper today.
Don
:) :) :) :) |
|
|
| smoking-amp |
And... we would not be stuck with all these 43% tapped output transformers on the market that are totally meaningless!
Don :D |
|
|
| SY |
Care to explain the EL84 datasheet? There certainly seems to be something good happening at 43% (as opposed to 20%).
BTW, I had read through Grimwood's stuff a few years back and could find zero analytical or experimental evidence in support of his assertions. |
|
|
| ilimzn |
| quote: | Originally posted by Bandersnatch
Hey-Hey!!!,
If we're to believe the curves, the plate can go a fair distance below g2 before the screen current begins to climb. It is dependant on the type of pentode of course, and some are better than others.
|
Of course, you are right, i was generalizing too much... after all, heaps of work have been done on this phenomenon, resulting in, amongst other, shadow grid tubes and as a subset, beam tetrodes. but i am sure you understand the gist of what I was saying - in the classical UL with DC G2 and plate voltages the same, i.e. just an UL tap on the output transforme, and not a separate winding, whenever the plate swings under the supply, it is going below G2 potential - and above G2 potential when it is swinging over the supply. The fact that UL tap is by definition <100% makes it so.
In general, pentodes and beam tetrodes tend to be less kinky over the useful plate voltage range (just look at sweep tubes!), the lower the G2 voltage - but of course, you sacrifice gm by lowering Vg2. In an implementation where G2 is driven via a separate winding with it's own G2 supply, it follows that one could optimize sensitivity vs distortion for a given percentage UL, by chosing the right DC Vg2, which would alow the designer to move the point at which the plate swings below G2 and G2 starts capturing more current, in esence, pushing as much of any kink to higher swings. This is particulairly important with your excellent point about drawing a complex load ellipse - in such an opimized system, there would be less 'blur' (good description, BTW). |
|
|
| Bandersnatch |
| quote: | Originally posted by smoking-amp
And... we would not be stuck with all these 43% tapped output transformers on the market that are totally meaningless!
Don :D |
So get yer winders to build at something other than 43%. IIRC, from looking at a test of output TX's few had a tap at 43%. 25-50% was the published spread...:) I read one of H&K's papers on the taping point, and they said that for the 6V6 22.5% was the sweet spot. Now the Z565 of theirs came in at 25% taping.
There are of course limits on the taping, as end-of-layer points are the most convenient places to tap. Get several! The last pair I ordered of the Peerless S265Q came with 3. Being able to hook up the g2 at another spot is quite useful, and my circuits sometimes use more than one pair of taps anyway...:)
cheers,
Douglas |
|
|
| smoking-amp |
"Care to explain the EL84 datasheet? There certainly seems to be something good happening at 43% (as opposed to 20%)."
OK. I assume you mean the Mullard data sheets. Using page C14 for 43% with 6000 OHM CT P-P xfmr. For the 10Watt output max shown, Ik is at 94 mA. I am assuming class A operation here (the GE data book shows 48 mA idle per tube for class A) so at peak output, one tube would be at this 94mA peak and the other at 0 mA. For 94 mA, page C6 of the Mullard sheets shows:
Ra = 1360 Ohm
Mu_triode = 21.9
Gm = 16.1 ma/V
Using the effective triode formula: eff. Rp = Ra/.43 => 3162 OHM
The 6000 OHM P-P CT xfmr in class A would be loading at 3000 OHM per tube. This is a pretty good match. 45% UL would be perfect.
Mu effective would be: Mu_triode/.43 => 50.9
Vin on the Mullard chart C14 is 16.5 V g-g rms => 11.66 Vpk on each grid.
Output voltage swing then would be 11.66 *50.9 => 593 Vpk for unloaded or 593/2 for matched Z load (Rp = Zload) => 296.5 Vpk which matches "roughly" with the 300V B+ available.
Mu_triode on chart C6 is actually varying between 14 and 21.9 between 0 and 94 mA, so if we take the average of 17.95 for Mu_triode, and times 1/.43 to get the effective triode average MU:
we get 11.66*17.95/.43 => 486.7 Vpk unloaded and 486.7/2 => 243 Vpk loaded (Rp = Zload) which is more plausible with a 300V B+.
Calculating power out then: 243 *243 / 3000 = 19.68 W pk power or 9.8 Watt average. Not a bad match to the 10W shown on the graph.
Seems to me the effective triode #s hold up.
The EL84 is not much different in Rp from the 6L6, (1360 vs 1403 Ohm) so not too surprising to get near 43% again with a 6K xfmr.
The 20% UL case would like to see more like a 7K or 8K OHm loading (per tube) which would require a 16K Ohm P-P Ct xfmr (in class A, 32K Ohm in class B). So the Mullard charts for 20% are grossly overloading the tube. Hence the high distortion.
Anyone have data for UL with a very different tube?
Don |
|
|
| smoking-amp |
"So get yer winders to build at something other than 43%."
Well, I did special order the Edcor CXPP100-SP1 output xfmr with 20% separate windings. Was more for CFB experimenting, but now I'll have to find a tube to try 20% UL too.
I should point out however, that this "optimum" %UL for power is just making Zload = Rp effective. This gives maximum power out, but not necessarily lowest distortion. The H-K UL data is really "rigged" in my opinion since they test at the maximum power out drive for all cases.
This causes serious overdrive or clipping for the other "non optimum" %UL in their test. Making everything else not at the "optimum %UL" look horrible distortion-wise in their test.
A more sane test would be to test at below clipping and G1 current overdrive and just measure clean power out and distortion.
Almost certainly, lower distortion than at the power "optimum %UL" will result with Zloading > than Rp effective, just like any normal triode design.
Don |
|
|
| Johan Potgieter |
Yipes,
When I post on my lonesome own everybody else is sleeping, and as soon as I go to sleep everybody else posts, quietly move threads elsewhere, etc. :( [oh, and Smoking-amp is introducing new maths by dividing by 0. At 0% tap the rp of a pentode is not infinity! :) :) ]
But there are still 3 things that perplexes me (OK, now I must use :( There) I see the webace article has been posted 3 times, but with little cognisance or at least mention of what the GE characteristics for the KT88 tell us.
Sorry to sound like a skipping CD, but I thought it clearly shows that impedance is not critically related to all of this subject matter, apart from obviously influencing the max. available output power. Why do I get the impression we are still hunting for an optimum here, at least regarding the UL topology?
Secondly it seems to be clear why 43% is popular as an "optimum" (sic) tap. Maximum output power stays about the same, but rp goes from 8K for 20% G2 taps to 4,1K for 45% taps in the KT88 graphs. Distortion is slightly lower (maybe not worth mentioning, but still). This is by the way very similar to EL34 behaviour (the Mullard graphs are on the internet, as are those for EL84). Ditto for my tests with 6L6, although I use 33% taps - windings come out conveniently that way.
So praythee, what do we have against 43% or thereabouts?? I have not seen different information, and I do not contest the maths of Smoking-amp - neither has he shown that there is a substantial advantage at lower taps (if I did not misread somewhere in the dead of night). I get the impression that many of us are simply "wondering", again as if this is the first time someone has thought of this.
One could notice something else that is different, if not much. The EL34 graphs indicate max. available power output for pentode operation, immediately starting to drop somewhat as the screen is shifted, while for the KT88 (and other beam tubes) there is an increase (for RL=6K) from 47W for pentode mode to 50W for early tapped mode - obviously the contribution of the screen grid to usefull output power. (BTW, it would also seem that we have nothing to worry regarding the first watt, judging from both figures 2 and 3 in the Williamson-Walker article, apart from other references.)
Finally I detect here an unexpected nervousness concerning the undesirability of operating the anode and screen at the same dc voltage.
Hey???
Am I missing something after 50 years in the business? For ever amplifiers have worked under most strenuous conditions in this mode - one is not talking about a few volt drop across a power supply choke - and tubes have lasted for years ... Then there are the manufacturers graphs - a 6L6 say, is operational down to below 100Va with the screen still at all of 400V. (Of course handling an ac signal; the next moment the plate shoots up to 700V - nobody is talking of dc amplifiers). Is somebody missing the fact somewhere that the screen grid is not solid, and can never collect nearly a large percentage of the anode current especially in a beam tube? As I have respectfully asked before, are we challenging manufacturers' data? For most of the regular power output tubes anode and screen current/dissipation are given at zero signal as well as maximum output conditions. The rated screen dissipation takes this into account! Surely nobody can suggest that tube manual compilers have forgotten that sometimes a signal must also be handled by the tube.
I am ready to be informed of where I am missing something vital. But I hope to be forgiven for being sceptical about the sudden questioning of practices that existed successfully for more than a half century. It is really unlikely that errors in basic tube operation will now come to light, that failed to do so in the hey-day of the product.
(I see someone else has posted while I was typing - ah! I'm not alone! - Hope not to conflict or repeat.)
Regards all. |
|
|
| Bandersnatch |
It isn't new math. Just like the field intensity right next to an electron is not infinite...:) the 0% U-L collapses to the pentode plate resistance. Send your your quantum mathematician.
cheers,
Douglas |
|
|
| SY |
| quote: | | Anyone have data for UL with a very different tube? |
OK, 6973. I've got UL data, but no triode data (I'll see if I can extract something from the curves I've got). The UL data calls for 12k5 plate to plate fixed bias and 50% UL taps. For cathode bias, 13k plate to plate and 43% taps.
edit: at Vgk = 0, rp seems to be about 1k7 for triode. |
|
|
| smoking-amp |
Hi Johan and all,
Firstly, the full effective triode formulas do not go to infinity at %UL = 0, I just said that the plate term in the formula was small so it could be ignored for most cases and the formulas simplified to 1/UL% form. With the plate term included, the Rp does indeed converge on the pentode Rp and the Mu effective does converge on the pentode effective Mu. Here are the actual formulas:
Mu_eff = -dVp/dVg = 1/ [ (U%/Mu_scrn) + (1/Mu_plate) ]
Rp_eff = dVp/dI = [K*Ik^(-1/3)] / [ (U%/Mu_scrn) + (1/Mu_plate) ]
or: Rp_eff = K * Mu_eff *Ik^(-1/3)
and Gm = Ik^(1/3) / K ( K is a constant for units and tube geometry, Ik is cathode current )
which nicely gives: Mu_eff = Gm * Rp_eff ( just like any triode! )
OK, next on the KT88 curves. The tube Rp is affected by the plate (or cathode) current (see formula above, Rp proportional to Ik^-1/3). These KT88 curves are scaled power-wise so that the same Rp (or Zout on the graph) occurs for each case, hence the inclusion of only one Zout curve on the graph. Due to this, one expects all the curves to peak at the same %UL. This does not tell us much. The info we want has been factored out.
Now it could be that the power curves shown are the maximum power that can be achieved at some max tube dissipation, and then we could conclude that Ik^1/3 conspires to keep the peak at nearly the same %UL for all cases of xfmr Zload for this tube.
But this favored %UL will still be specific to the particular tube chosen because it depends on tube Rp and geometry. Other tubes will cluster at a different %UL. One will notice that the distortion increases for the lower impedance xfmrs with higher power output. One might want to scale the power curves for constant distortion at max power, then the %UL's for this tube will move around.
If max cathode current is kept the same (another option), then these curves will move toward the triode %UL as the xfmr Zload is lowered in order for the Rp to match the Zload.
Next, I have not advocated lower %UL, perhaps the 20% taps on my Edcor xfmr are confusing the issue, they were for CFB experiments.
If anything I would advocate higher %UL taps in the interest of lower distortion. But max power will likely come down as triode mode is approached. This is a tradeoff for the designer to decide.
On the screen current distortion issue, several factors likely have prevented noticeable distortion problems. First, the distortion is likely to be even harmonic so is eliminated by P-P.
Secondly all the papers quoted so far are using class A operation, which will eliminate most of the generation of odd harmonics by full overlap of the complementary monotonic screen distortion effects. My concern as stated earlier, is that most comercial UL amps do not run in class A, so likely will have increased odd harmonic distortion. Also, there are some SE xfmrs around that have UL taps, I would expect the screen current distortion there to be quite noticeable.
Thirdly, the traditional audio tubes used have HV screen grids with lower transconductance, these generally have much more forgiving screen current curves. The later designed, low voltage screen pentodes are much less forgiving. Many DIY designs use these low cost (well they used to be!) horizontal and vertical TV scan tubes.
And, well, we just want to test this screen current issue to see if it is a problem nowadays, with the greater emphasis placed on avoiding odd harmonics.
And, I wouldn't be surprised if someone has come up with the same effective triode idea before, I just haven't seen anything published yet. Its not really rocket science, I really thought it was obvious, it really is just ordinary feedback theory. Thats why I cannot see H-K not picking up on it in their paper.
Don |
|
|
| Bandersnatch |
ummm, Smokingamp...your quote:
Secondly all the papers quoted so far are using class A operation, which will eliminate most of the generation of odd harmonics by full overlap of the complementary monotonic screen distortion effects.
This overlap in Class A PP does in the even. The odd stuff is symetrical( v. the asymetrical 2HD ), is not effected by that operating method.
Cheers,
Douglas |
|
|
| smoking-amp |
"This overlap in Class A PP does in the even."
Umm, I said that in the line above what you quoted:
"First, the distortion is likely to be even harmonic so is eliminated by P-P."
What I meant by:
" class A operation, which will eliminate most of the generation of odd harmonics by full overlap of the complementary monotonic screen distortion effects. "
was a second order effect in P-P, where placing even order distortions back to back produces some odd order distortion. The more it is overlapped, as in class A versus class B, the less symmetrical or odd order residual remains.
Basically, averaging the gain of two devices smoothes out the variation. For class B, there is no overlap so no averaging.
Ie, two J shaped curves when placed back to back make an S shaped curve. The more we overlap the J's, the less of an S remains. Overlapping doesn't get rid of all the odd residuals however, particularly not the high order ones. As in most distortion cancellation schemes, the low order cancellations are effective, the high order ones are not.
It's like the LTP or differential stage, the even order distortions cancel, but an S shaped odd order one remains. It just saturates on BOTH ends now.
Don |
|
|
| smoking-amp |
6973
Ummm, here's what I came up with:
gm1 = 4800 @ 46mA
gm2 = 30mA/50V = 600
Rscrn = 1/gm2 = 1666
Mu = gm1/gm2 = 8
Hmmm, looks like its a 9 pin version of a 6L6!
6L6: gm1 = 4700
Mu = 8
gm2 = Mu/gm1 = 1700
Rscrn = 1/gm2 = 587.5
Been there, done that!
Well, might be interesting to see if the 6973 UL data matches the 6L6 UL data.
Any UL stuff available on one of the horizontal or vertical output tubes?
Don |
|
|
| smoking-amp |
Ahh, I just found the Mullard data sheets on the EL34 and they have UL mode data. The triode rp is 910 Ohm, so it's a little different from the 6L6 and EL84 at 1700 OHM. Will look at it tomorrow.
Don |
|
|
| ilimzn |
| quote: | Originally posted by smoking-amp
Any UL stuff available on one of the horizontal or vertical output tubes?
|
Not unless someone traced the curves. It should be possible to calculate the relevant data for sweep tubes that have published triode data, EL504/PL504 would be one like that, but of course, there would be nothing to check them against :(
As an aside, given that most sweep tubes have (often vastly) reduced maximum Vg2 WRT maximum Vp, it is no surprise 'classical' UL data is not provided, as it is unlikely that the tube can be used with a tapped transformer, unless the power supply is lower than the usual standard values - and we all know that separate G2 windings are not at all common. I think your approach with a MOSFET follower would be of great benefit here. |
|
|
| SY |
| quote: | | Well, might be interesting to see if the 6973 UL data matches the 6L6 UL data. |
It doesn't. That's why I suggested it. |
|
|
| Johan Potgieter |
| quote: | Originally posted by smoking-amp
And... we would not be stuck with all these 43% tapped output transformers on the market that are totally meaningless!
Don :D |
To summarise the above: 43% tapped output transformers are
totally meaningless.
Care to explain, in the light of practical measurements (to try to get some practice in here edgewise as you do not seem to have constructed anything so-far - one does want to make one's amplifier produce music, doesn't one?). And you have mentioned several times that the manufacturer's graphs do not give everything - that is not an explanation - they never do! Point is: do they give useful designer's data, or are we busy here with a mainly academic exercise?
But could one for once get to the nitty-gritty before continuing an otherwise very informative subject: If I use manufacturer's data as a first approximation in circuit design, and I find that a spectrum analyser (which is not out of order) gives roughly corresponding data ....
AND... being a researcher, I find that changes from that do not significantly improve my design .......
how must I arrive at the conclusion that "43% taps are meaningless"?? (I think it is the 3rd time I have asked this question in some way or another. I am not trying to be disrespectful in any way, Don, but allow me to also say: Been there, done that! (Didn't get the T-shirt; in those days there were none). Still doing it (still no T-shirts).
I am also not disputing the value of what you are presenting here (I understand that; just for the record, did pass Maths IV at varsity). But either the manufacturer's graphs are misleading or they are useful. If the former, see previous paragraph. If the latter, why do I need mathematical gymnastics in order to design a successful amplifier (and why do you, to my perception, dismiss everything that doesn't agree with you)?
(More on graphs later)
Kind regards. |
|
|
| Wavebourn |
Hmmm... Johan, may be I am wrong, but the essence on the whole discussion is that feedback on screen grids must be delivered from separated windings and voltage on it must be less than on anode according to the tube geometry.
Another option is to use the same UL taps, but do not connect grids directly to them, instead use zeners or some another way to shift voltages, such as power MOSFETs.
If power MOSFETs are used we do not need taps anymore and can play with variations using potentiometers. |
|
|
| smoking-amp |
"It doesn't. That's why I suggested it."
I have the RCA datasheet for the 6973, is this the one you have?
It does not give any graphs, but just a single operating point for 50% and 43% UL. What they state is class AB1 operation, and a very high primary resistance xfmr compared to the pure pentode data case. Operation at 95 or 84 mA for both tubes at peak signal.
I came up with this data from other parts:
Gm1 = 4800 at 46 mA (2*46 mA => 92 mA so applicable for here)
Gm2 = 30mA/50V = 600
Mu_triode = gm1/gm2 = 8
Rp_triode ~ = 1/Gm2 = 1666 OHM
Rp_pentode = 73000 OHM
this would give Rp_effective = 1666 / .43 => 3874 OHM
and Rp_effective = 1666 / .5 => 3332 OHM
If we also include the small pentode plate factor in the above calcs
these numbers drop to 3788 and 3258 OHM respectively
For class AB1 the tube sees about 1/4 of the P-P xfmr impedance.
So 13000 / 4 => 3250 OHM (43% case)
and 12500 / 4 => 3125 OHM (50% case)
So these seem to be roughly in line with load Z
matched to the effective plate Rp again. With some overlap in conduction (class AB) the load Z might be averaged up a little too for the higher Z seen in class A.
They don't give any drive signal amplitude, so I cannot check the
effective Mu #. Without any graphs, one cannot tell whether these points are minima to distortion or maxima to power or what. Limited conclusions here.
Hi Johan,
None of the papers on UL seem to treat the case for any other triode plate Rp other than the 6L6. Nor do they treat different xfmr primary Z.
They simply did not finish the analysis. That's why we are all still here scratching our heads today.
The manufacturers data on tubes in UL seem to all be for tubes that are extremely close to the 6L6 in triode_plate resistance (KT88 6550 ...). In fact, most of the common audio tubes around are closely related to the 6L6. The only exception I have seen with UL data so far, being the EL34.
The Maths on the other hand DO give suggestions that the %UL would be different for a different triode configured Rp or primary Z.
They are simply not constrained enough. Now there COULD be some additional constraint to be uncovered yet, but no one has anounced it yet to my nowledge.
Being DIYer's, we are NOT obliged to use the common audio tubes recommended by the manufacturers historically, and like to experiment outside the box. Having no satisfactory historical theory on this, one feels a certain need to check this out. Do I sense some concern that the Emporer (UL 43%) might be found lacking in the clothes department?
On the screen current distortion issue, another concern has not been mentioned yet. P-P will only cancel even harmonic distortion if the two tubes are matched. Now how many tubes get matched for identical screen current? (probably none, and certainly not dynamically over the signal range) I'm sure the H-K tests used the best matched tubes they could find, plus class A operation, to present the best results possible.
The screen current drawn by pentode tubes depends critically on the alignment of the g2 grid wires behind the g1 wires. Little windows often being present in the plate for visual grid alignment purposes. Having checked this same window on a number of tubes, I can often see visually obvious misalignments. Just dropping an amplifier or tube on the table may cause grid wire misalignment too. So lab results are not sufficient to characterise practical UL amplifiers.
Now, don't get me wrong on this, I am not dumping on UL, I just think more needs to be characterised, so that good designs can be attained with confidence. And whether any new flexibility in good design exists. And whether any problems identified need to be addressed, which I'm confident fixes can be found for IF necessary. (the screen current issue already has a fix previously noted)
Well, I need to spend some time on getting the test arrangements set up here.
Don |
|
|
| Johan Potgieter |
| quote: | Originally posted by Wavebourn
Hmmm... Johan, may be I am wrong, but the essence on the whole discussion is that feedback on screen grids must be delivered from separated windings and voltage on it must be less than on anode according to the tube geometry. |
Anatoliy,
I understand that part of it; we are on slightly different paths somewhere. What you say above is part of the discussion, and I was trying to state that I can not see the necessity for "voltage on it" must be less! It is not clear how I failed to register my point, but apology where needed. Some did mention tube geometry, but it is the same tube for pentode/triode - why not the necessity for separate voltages there?
Unless I got mixed up with the sweep tube affair! Yes, surely there. But unless I am obtuse (and please do not hesitate to point that out! :o ), much of the discussion was also about normal audio tubes. After all KT88, EL34, 6L6 et all frequently came up. I cannot make it clearer than before: I simply see no reason for this sudden decree that Vg2 must be lower than Va in UL, but it is fine for, let us say, 0% and 100% g2-taps. The question is stil begging: Are all the manufacturer's specs suddenly wrong - or where am I missing what?
To (try to) be clear, I have no objection to a lower Vg2 than Va at high voltages. In my 100W Quad type circuit I use an effective Vg2 of 460V and Va of 560V. But that "it must be" (i.e. under all conditions) - that is my question.
Kind regards |
|
|
| SY |
Your point is interesting- most audio pentodes/beam power tubes cluster around the same general characteristics. And there's an evolutionary reason for that.
The 6973 data I got was from the sheet on Frank's.
http://frank.pocnet.net/sheets/049/6/6973.pdf
No UL curves, but plenty of beam power data. |
|
|
| Johan Potgieter |
Not to make this personal or that I am of particular importance, but I did promise certain responses. Thus only to state that I may be "out" for a day or two. I hope I will be able to catch up later!
Regards. |
|
|
| smoking-amp |
Seeing as how this %UL factor is still unresolved, let me suggest some possible constraints on it.
(The equivalent triode model does not care about the %UL. Setting %UL to some # simply causes it to crank out a primary xfmr Z to use with the given tube so that Zload = Zplate_effective _triode for maximum power transfer to the load.)
Usually I see low Mu triodes specified for regulators with minimum conduction loss (ie. best efficiency). This I believe is due to the plate being useable for high current down to lower voltage drops. This efficiency criteria would constrain %UL's on the low side (high MU's) as too inefficient. Maximum power output cannot afford to waist power in the tube.
The Ik^(-1/3) factor in the triode effective_Rp formula (it's also in the normal triode formula too) may constrain %UL's on the high side. I don't know why yet. Workin on it. Maybe it causes the cathode current limit to be reached?
I think it is important to understand the nature of any constraints on the %UL, so that if a designer does wish to vary this for some objective, he/she will understand the tradeoffs. Many would be comfortable with the loss of some power or efficiency if lower or more benign distortion, for example, could be gained for small change.
Don |
|
|
| smoking-amp |
Gee, the constraint on the higher %UL has been staring me in the face. It's obvious.
The effective triode model shows that the effective MU drops with increasing %UL, or alternatively, more triode % means more internal feedback, so less gain. The input signal to drive it simply gets so big that it runs into grid1 current.
(10 lashes for not seeing that one a few days ago!)
The previous Hi-MU/lowefficiency constraint on lower %UL is not sufficiently convincing yet.
Don |
|
|
| Johan Potgieter |
I should have been off already - this thread is addictive (and that from a body that was born in a wine-producing area!)
This runs fast; I was typing when Don's last post came through and cancelled to read first.
Don, I do agree with your, well then, two last posts (if you are not already typing further!) You make stimulating remarks; rest assured that I am also thinking further despite some considerable previous experience (the scourge of researchers - they always have questions). I also owe you some explanations that I have not forgotten about. Kindly see my previous post.
But then already a further question now! Yes, the possibility of overdriving (too large an input signal). But how much of a probability is that? I do presume that that constitutes such obvious sign of overload - mostly there will be NFB that will then "discontinue" etc, that it signifies the amplifier output limit (i.e. max. output) without any room for doubt. When testing one looks with a scope at all signal points anyway (at least I do). Do you mean that some people will do this without recognising it? (Oops, let us not dwell on what some will do. Anything is possible these days.)
Kind regards (I am now off for a while.) |
|
|
| smoking-amp |
Well, its clear from H-K's paper that THEIR constraint on lower %UL was low level distortion rising. But this is because their xfmr load Z becomes too low compared to the effective triode Rp, causing 3/2 power distortion. A higher primary impedance should allow %UL to continue to drop without a distortion penulty. And so no limit appears to be really stopping lower %UL.
My conclusions from this would be that a broad range exists for %UL. On the high %UL end, power must be sacrificed as effective Mu lowering runs into g1 current. But distortion does not shoot up as long as one stays below the g1 current level.
On the low %UL end, practical transformer primary winding problems for high Z limit the %UL tap. Also, screen current distortion will increase as %UL lowers since the screen current will be using less of the winding. (the closer the screen current enters the winding to the plate current, the less the alteration in power output due to it siphoning off plate current)
Sounds good to me. Anyone see any problems with this?
Don |
|
|
| smoking-amp |
"When testing one looks with a scope at all signal points anyway (at least I do). Do you mean that some people will do this without recognising it?"
Well, H-K's paper (figure 2) show's the high level distortion skyrocketing as the triode end of the chart is reached. Probably what is happening is the distortion begins to rise rapidly as g1 signal peaks get close to 0 volts, with out-right grid current occuring somewhere where it goes off the chart. Obviously, triode wired pentodes for normal amplifiers do not cause distortion like they show it, so they are clearly overdriving it.
That was my complaint earlier about their data, they should have lowered the drive on the triode end and so lowered the power out curve more toward the triode end too. Its the old trade-off of more power and more distortion - or - less power and less distortion.
An observation would be that as one increases the %UL toward triode, it pays to increase the screen DC voltage (well, up till B+ is reached). Since the tap is moving closer to the plate, where screen current will do less dist. harm, and higher screen voltage will hold off g1 overdrive (can lower g1 DC level, ie larger bias).
Conversely, toward lower %UL, it would pay to lower the screen DC voltage, due to the tap moving further from the plate, and the increasing MU at lower %UL allows g1 DC voltage to come up to compensate the lower g2 DC (less g1 drive signal required, so more headroom).
AS far as testing %UL effects, its clear that 100% UL works fine, just less power there. The low %UL with higher primary Z to compensate is really the issue lacking data support.
Don |
|
|
| smoking-amp |
Well, I woke up this morning and realized that my previous conclusion that grid1 overdrive limits output on the triode end of %UL was a little off the mark. Right answer, wrong reasons actually.
Both effective MU and effective Rp formulas have a %UL term acting identically in each. So if the xfmr primary Zload is matched to Rp effecitive in all cases, the voltage output required drops at the same rate as Mu drops, and no g1 overdrive occurs. They track.
The 1st outlaw factor turns out to be the Ik^(-1/3) term, which appears in only the effective Rp equation. This causes a mistracking between the two (Rp and Mu). Ironically, it favors the triode end. Lower Rp and Zload at higher %UL increases current causing the IK^(-1/3) term to lower Rp a little further (and so too our matched Zload ). This requires LESS g1 drive for the same Mu and power out, and in addition, the gm1 formula has Ik^(+1/3) in it to produce more current out for the same g1 drive. This term i | | | |
