• WARNING: Tube/Valve amplifiers use potentially LETHAL HIGH VOLTAGES.
    Building, troubleshooting and testing of these amplifiers should only be
    performed by someone who is thoroughly familiar with
    the safety precautions around high voltages.

New 6L6GC Project. DC -> LTP

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'll stand corrected for the wiper on the input. If that loads the output from the prevoius stage.

Yes, at minimum, the wiper goes to ground. Also, you get typ. 100k in series with the grid which will do two things - it will shift the bias of the tube somewhat due to grid leakage current developing extra voltage on the 100k of the pot (this is in fact exactly how the grid leak test function is implemented in most tube testers - you check for this voltage!), and, worse, it forms a lo pass filter with the tube's input capacitance, so at the extreme low setting, this filter will have a high cut function inside the audio band. And - most importantly, the pot will actually not attenuate, just provide this filtering function. However, if you try this, you might be fooled that it indeed does work precisely because of the (over)loading of the previous stage. It has an output impedance so this acts as a voltage divider with the portion of the pot between wiper and ground. The attenuation curve will be very odd depending on the source, usually only a bit over minimum and you will get full output, and as I said, the source signal will be overloaded when the wiper is set to zero, i.e. the input signal will be shorted.
 
The pot is in parallel to the grid, just as the original 1M grid leak was. At full volume the grid leak would be 100K and at 0 volume it is still 100K.

You could also just tie the wiper to the grounded end of the pot. Then the wiper would ground the grid at one extreme 0 volume, and be at full volume and 100K at the other extreme. Again, this is if grounding the grid does not also place an overload condition on the previous stage for some reason.
 
The pot is in parallel to the grid, just as the original 1M grid leak was. At full volume the grid leak would be 100K and at 0 volume it is still 100K.

No, at full volume it would be the output resistance (at DC) of the previous stage in parallel with 100k. If you are making an amp that connects to 'a signal source' it is not good practice to assume what the source looks like at AC or DC (for instance assume that the source has a coupling cap at the output, or that it will operate correctly with it's output shorted). You could tie the wiper and ground together and add a series resistor in front assuming you use a linear taper pot, a correctly calculated resistor will get you a usable audio-taper like attenuation curve. However, in this case the variation in series resistance seen by the tube is higher than in the classic connection and there is always some attenuation even at the highest setting of the pot.

All that being said I really see no reason to further elaborate on this, if you like you can draw the circuits themselves (not forgetting about avoiding assumptions re previous stage) and figure it all out for yourself, as I have no intention of runnig through all the circuit variants that save a whopping one resistor for the price of added disadvantages.
 
Your diagram has obvious mistakes.
The 6AG5 is triode connected to a 'current source' that is bypassed to ground through a 1uF capacitor - the way it is drawn, the plate load for this section is the 545ohm resistor. This will provide a lot of attenuation and distortion, not gain.
The bottom half 6N1P grid must not be connected to ground, since the top half is DC coupled to ~143V if you want this to work at all. This way the bottom half of the 6N1P is WAAAAY into cut-off (-140 something volts at G1) so no phase splitting.
The output tube bias network has a common flaw, namely if the wiper of the adjustment pots go open (which may happen during adjustment as they turn even on pots that normally seem to work OK), the tubes will be left with no negative bias, i.e. maximum current will flow.

good analysis as always........i was thinking Mullard 5-20 when i saw his schematic......i was unsure about the CCS....

about the output biasing network, i also use such type, to avert disaster in case of pot failure, i do connect a 100k resistor from the wiper arm to the negative voltage source, this way tube meltdown is avoided.....
 
Any glaring errors before I start ordering parts?

Signal
An externally hosted image should be here but it was not working when we last tested it.


LTP Curves
An externally hosted image should be here but it was not working when we last tested it.


Input Curves
An externally hosted image should be here but it was not working when we last tested it.

It's still a nice project, so I think you should go on with it just correct the mistakes. So:

1) First stage needs to have a proper CCS as the plate load. In your drawing it is blocked to ground through the 1u cap, i.e. for AC the tube plate sees only the 545 ohm resistor. I can't read exactly what the CCS part is but in this position a good CCS can save you a lot of extra 'housekeeping' - for instance, no need to specially filter the power supply for that stage as a good CCS will provide very high ripple rejection (basically it will be the ratio of CCS dynamic resistance and plate resistance of the tube, it's quite simple to get 1000+ i.e. 60dB+ here even with a simple CCS). You are using the full mu of the tube, which might be high for your amp given there is no global feedback. The output tubes need some 100Vpp clean swing at least, and the gain of the input times LTP is well over 600, so about 120mV will get you to full power. If this is too low, using a simple resistor load for the first stage may be a good idea, but then you do need extra filtering for it's power supply.

2) LTP stage has two problems - not enough lean swing and there is an error in connecting the grid of the bottom half of the 6N1P. This should be connected to the top half grid via a resistor of 470k to 1meg, and also to ground via a capacitor - the 1u/630V part you have in the first stage on your original drawing, will do the trick. The purpose of this connection is to give both grids the same DC voltage, but prevent any of the AC output of the first stage go to the bottom half of the 6N1P's grid.
Getting more swing out of a DC coupled LTP means in essence maximizing the available voltage across the tubes + their respective plate loads. Since the amplification factor is not too high, you might consider lowering the plate current in the first stage (this will give you bit less mu there but as i said it may be too high to begin with), which will also lower the plate voltage. This in turn lowers the grid voltage on the 6N1P, and as a consequence the common cathode voltage must go down nearly by the same amount, giving more voltage headroom. Secondly, the power supply for this stage needs to be kept as high as possible. One thing that goes hand in hand with this requirement is that the power supply ripple for this stage does not need to be the apsolute lowest possible because it is common mode and hence rejected by this stage and the output stage. This means that a simple RC filter from the main supply will be more than enough, calculated so that minimum voltage is lost across the R - say about 20V. Both of these measures should give you some 50V more to play with in this stage. Also, you can lower the current somewhat. 15-18k resistors in the plate circuit should do fine, and the output tube grid leaks (100l) should still not present a big enough load to significantly reduce the output swing.

3) Output biassing correction - there is a number of ways you can do this but in general I have found the best way is to make a voltage divider with a resistor from the bias supply, and the trimer (one side connected to wiper to form a variable resistor) to ground. This way, if the wiper goes open, the voltage divider automatically gets the maximum negative voltage, thus saving the tubes. You could also keep it as it is now just add a 100k from the wiper to the bias supply (and slightly correct the value of the resistors on the side of the pot that goes towards ground), but that will need more parts. I have also found that a bias network that has a bias and balance pot is much more practical and easyer to adjust. but this is more a thing of personal preference.

4) A few cosmetic touch-ups, probably drawing errors - the 8 ohm speaker needs to be connected between 0 and 8ohm taps, the input stage security grid leak (1 meg) should be on the tube side of the shielded cable. It may be a good idea to reduce the cathode resistors in the output stage to 1 ohm (make these small, so they also act as fuses) - especially if you want to try triode connection. 1 ohm gives you a direct 1mV = 1mA reading of the cathode current. You also already have the resistance of the output winding in series with the cathodes. In pentode and UL this does make a small but hardly signifficant difference, in triode it increases the apparent plate resistance of the triode connected tube by (Rk + Rwindig) * mu, which can get up to 5-10% extra apparent Rp, and a few W of power lost. Also, perhaps I have missed some other smaller points already mentioned.

In any case I wish you good luck with the project, it looks very promising!
 
First, I'd just like to say that I only have the tubes, transformers, 4 10M45's and 4 LM334's. The rest is all in the air :)

1) First stage needs to have a proper CCS as the plate load. In your drawing it is blocked to ground through the 1u cap, i.e. for AC the tube plate sees only the 545 ohm resistor.
I just followed the example schematic provided with the LR8's datasheet. It did look a little strange but I trusted the manufacturer..?

See p. 4:
http://www.supertex.com/pdf/datasheets/LR8.pdf

The output tubes need some 100Vpp clean swing at least, and the gain of the input times LTP is well over 600, so about 120mV will get you to full power.
I didn't have the formulas handy to calculate the decrease in gain due to the CFB but I wanted to leave room for global FB if needed.


LTP stage has two problems - not enough lean swing and there is an error in connecting the grid of the bottom half of the 6N1P.
Yep, the bottom half of that 6N1P is definitely not getting the right voltage, an easy fix I guess. What do you mean by lean swing though? Sure the curves are a little bunched but that looks like mostly 2nd harmonic, so the push-pull topology should neutralize that right?

Since the amplification factor is not too high, you might consider lowering the plate current in the first stage (this will give you bit less mu there but as i said it may be too high to begin with), which will also lower the plate voltage.
Probably what I'll end up doing. It was just a bit of a juggling effort, between trying to stay within the .76 watt dissipation of the LR8, as well as finding the most linear area of the 6AG5 curves in which to operate.

a simple RC filter from the main supply will be more than enough, calculated so that minimum voltage is lost across the R - say about 20V. Both of these measures should give you some 50V more to play with in this stage.
Well, I'd planned on using an LC filter so there's a good deal more potential voltage to be had. The Hammond 282X' 500V secondary should give over 700V if needed. I just didn't want to have to put electrolytic capacitors in series :p

Also, you can lower the current somewhat. 15-18k resistors in the plate circuit should do fine, and the output tube grid leaks (100l) should still not present a big enough load to significantly reduce the output swing.
You think the 10M45's will be happy at 8mA? If so I'll gladly proceed with that plan. Or is there another IC that'd do a better job in this place? I can save the 10M45's for another project.

Output biassing correction - there is a number of ways you can do this
Thanks for all the options, I still haven't quite decided but the extra resistor doesn't make too much difference to me.

the 8 ohm speaker needs to be connected between 0 and 8ohm taps
Since we are grounding the 4 ohm tap in CFB scheme, I'm pretty sure this renders the 0-16 ohm winding equivalent to 8 ohms, yes?

It may be a good idea to reduce the cathode resistors in the output stage to 1 ohm (make these small, so they also act as fuses) - especially if you want to try triode connection. 1 ohm gives you a direct 1mV = 1mA reading of the cathode current.
Sounds like a plan.

In any case I wish you good luck with the project, it looks very promising!
Thanks very much, especially for the detailed explanations. I've built a few amps previously but it was a few years ago, and it's nice to have things laid out clearly.
 
Quote:
Originally Posted by ilimzn
It may be a good idea to reduce the cathode resistors in the output stage to 1 ohm (make these small, so they also act as fuses) - especially if you want to try triode connection. 1 ohm gives you a direct 1mV = 1mA reading of the cathode current.

Sounds like a plan.
Since you are using the output tranny secondary to ground your cathode through you need to consider the total resistance to ground. You'll have maybe .5 -.75 ohms additional and that is a large percentage of error with respect to a 1 ohm resistor when calculating the cathode current. Better to leave it at 10 ohms. Even then you should still read the total for accuracy.
 
I didn't have the formulas handy to calculate the decrease in gain due to the CFB but I wanted to leave room for global FB if needed.

not much....http://www.diyaudio.com/forums/tubes-valves/60767-cathode-feedback.html


Since we are grounding the 4 ohm tap in CFB scheme, I'm pretty sure this renders the 0-16 ohm winding equivalent to 8 ohms, yes?

no, the 8ohm speaker goes to the 8ohm tap, if you put your 8ohm speaker on the 16ohm tap, the reflected primary impedance is halved....

the reason to connect one cathode to 16ohm tap and the other to the 0 tap while grounding the 4ohm tap is so that the cathodes get equal voltage feedback... the 0 to 4 ohm tap and the 4ohm to 16 ohm tap have equal number of turns.....
 
Administrator
Joined 2004
Paid Member
Clearly the Supertex regulator as a current source is not going to work, it is not stable without the capacitor and with it the CCS implementation has an AC impedance of <600 ohms over most of the audio band. You'd be better off with a cascode CCS using a pair of DN2540 - here is an article written by a close friend that I think many of you might find of interest: The Audio Signal Path; Minimising Power Supply Interaction Richard Sears, Vacuum Tube Audio
 
First, I'd just like to say that I only have the tubes, transformers, 4 10M45's and 4 LM334's. The rest is all in the air :)
I just followed the example schematic provided with the LR8's datasheet. It did look a little strange but I trusted the manufacturer..?
See p. 4:
http://www.supertex.com/pdf/datasheets/LR8.pdf

Ah, I see the problem. The application is geared towards supplying a constant DC current to a simple load, with out the other end of it changing voltage. I did not read the datasheet through but I suspect the 1u cap must be there for stability - or it may be a drawing error and should have been put on the input side as on the other examples. Be that as it may, you do need a bit more sophisticated CCS because integrated regulators tend to be rather bad CCSs at higher frequencies, and I am talking on the order of kHz so well within the audio band. Then there is also the residual capacitance of the CCS to contend with, but I think in your case this is not paramount. Let me see if I can figure out a simple CCS for you to use. You can also always use a simple resistor (but then the power supply for that stage needs to be well fintered) - the advantage here being, you have a lot of voltage to your disposal so the plate load resistor can be quite high compared to the tube's plate resistance, therefore you get more amplification from the tube.

I didn't have the formulas handy to calculate the decrease in gain due to the CFB but I wanted to leave room for global FB if needed.

Yes, that's what I suspected. However, given that you already have two forms of feedback, you need to count on them in form of increased requirements for voltage swing to the grids of the output tubes. You might run into problems with limited amplification of the LTP stage - the current you chose there with the given tube gives you the grid bias WRT cathodes, and this also gives you a limit on peak input signal from the previous stage. Getting more current from that stage automatically implies lowering gain and lowering maximum input signal, before you run into positive G1, exactly the oposite of what you want. Reason being, at thet point, you can't get any more gain reserve for NFB by increasing the amplification factor of the input stage, because the input stage clips the input of your phase splitter. Because DC conditions of the two are linked, you need a careful compromise here.

Yep, the bottom half of that 6N1P is definitely not getting the right voltage, an easy fix I guess. What do you mean by lean swing though? Sure the curves are a little bunched but that looks like mostly 2nd harmonic, so the push-pull topology should neutralize that right?

Typo - CLEAN swing. The problem is, you can't really count on ideal 2nd cancellation because you can't count on the 6L6 outputs being ideal pairs, nor the bias voltage of a given (even tightly matched pair) having no tolerances.
Now, using a LTP with a CCS in the tail fed from one end gives you ideally balanced differential outputs, with exactly equal and oposing 2nd harmonic distortion, referred to the input.
There are two consequences:
1) Some of this will cancel out the 2nd harmonic distortion from the first stage. However, it will also produce uncancelled 4th harmonic components. In your case the inoput stage is more linear so there will be 'overcancellation' i.e. some 2nd harmonic distortion will be left. The reason for this is you are driving the LTP from only one end, the resulting distortion is a combination of the nonlinearity of the swing combine with the differences of the 6N1P halves. This however would largely be mittigated by having very tightly matched 6L6s.
2) Because 6L6 are not ideally mtched, one will operate on one portion fo the nonlinear swing, but the other will operate on a (hopefully slightly) different portion, so the nonlinearity will not cancel completely.
If your swing is limited you may end up operating one output tube close to one end and the other to the other end of the usable swing, which will cancel out the major component but not ideally (that being the 2nd harmonic) but may well not cancel or even increase higher order even harmonics.
The need to take account of potential output tube tolerances is precisely why you need extra swing, so that no matter what combination of tubes you end up with, you still are far enough from the end of available swing so that you do not get into extremes such as I mentioned above. It is common practice to provide at least 30% more swing, given that output tubes will easily vary by 10% even when declared a matched pair, and as Murphy's law will have it, tolerances always tend to add up the wrong way, in this case that being 'they tend to not cancel'.

One important point hre is that you should not forget you will potentially be running this amp without global NFB, so there is no way to correct distorsions from the two stages of the front end after the fact. This is whay it's so important to introduce the least possible amount to begin with.

This also brings me to another matter I forgot to mention in the previous mail, and that's the requirement for AC balance setting. I will leave this for the end of this post though.

Well, I'd planned on using an LC filter so there's a good deal more potential voltage to be had. The Hammond 282X' 500V secondary should give over 700V if needed. I just didn't want to have to put electrolytic capacitors in series :p

More importantly, there is a limit to the voltage across the 6N1P in a cold state and you are running fairly close to it already. I would suggest just trying to get as close as possible to the main power supply without having to scarifice too much in terms of filtration. Today this is not too difficult, larger capacitance electrolytics are not that expensive (I'm talking on the order of 47uF here) so a simple RC filter can have lowered R and increased C to still provide decent filtering without burning off too much of the available voltage.

You think the 10M45's will be happy at 8mA? If so I'll gladly proceed with that plan. Or is there another IC that'd do a better job in this place? I can save the 10M45's for another project.

Now, this might be a problem, yes. In this case, being in the cathode leg, the CCS does not need to be highly sophisticated, so you could construct your own out of a few parts without a limitation to the tail current. As I mentioned, the LTP part of this design is the most critical because of the compromises one needs to make in order to get the desired swing (and this is imposed by the local feedback mechanisms used in the output stage), so it would be unfortunate to base this most important part of the amp on a weak point of a single component. IMHO one should look into replacing that component with something more suitable. Are you familiar with what is often called a 'ring of two' CCS? This could work for you both for the input stage and the LTP.

Since we are grounding the 4 ohm tap in CFB scheme, I'm pretty sure this renders the 0-16 ohm winding equivalent to 8 ohms, yes?

No, the 0-16 ohm winding is always the same since this is determined solely by the turns ratio. In principle you should not even need to ground the output winding unless you want a ground referred signal for the rest of the amp, such as for a CFB or global NFB scheme. Your load still goes between the 0 and 8 ohm taps if it's 8 ohms, the only thing is that one of it;s ends is not grounded, instead it swings some voltage with respect to GND - but there is actually no current through GND due to the load - the load current goes only through the windings. Now, the cathode current also goes through the windings to ground and partially shares it's path with the load current, but that is all.

One thing to remember is that 'an 8 ohm secondary' is really a notional thing. It means that if you put an 8 ohm load there, then you will get the specified Raa on the other side. From that standpoint you could put it across the 16 ohm taps, and then because it's 2x smaller than nominal, the Raa on the other end will also appear 2x smaller. However, other aspects of the transformer, such as maximum power, frequency band, and the interleaving required to get it, are designed based on the expetation that an 8 ohm load will indeed be connected to the 8 ohm taps. Doing otherwise also changes these parameters, even if Raa is only determined from which tap you use and the winding ratio. In fact, even though 0-4ohm taps should provide the exact same winding ratio as the 4-16 ohm portion of the winding, it is NOT advisable to connect a 4 ohm load between the 4 and 16 ohm taps precisely because that portion of the winding may be quite differently made compared to the one between the 0 and 4 ohm taps.

I still owe you something on AC balance. Whereas it is fairly easy to get the PP halves precisely balanced for DC, i.e. have equal standing current for both tubes, simply by properly adjusting bias, amps mostly do not have an AC balance preset. Mostly they rely on pure chance and the output transformer cancelling out even harmonics as best it can, BUT this approach has a flaw in form of a net DC current through the transformer, which is not what you really want in a gapless PP transformer.
Adjusting precise bias current for each of the output tubes usually shows a difference between the required negative bias voltage between the tubes, to make it so. This means that one tube may well always be conducting a bit more than the other, since at the very least one will run into positive grid earlyer - the one with less negative bias voltage. At that point you get one-sided clipping which somewhat limits your maximum output power. However, what you get ALL the time is one side conducting more current than the other, which means that you also get a net DC magnetisation proportional to the output current. At some point this may start running the transformer core close to saturation, reducing apparent inductance and resulting in loss of power especially in the bass, and distortion.
Designs with a LTP splitter are very easily augumented with an AC balance control that deliberately unbalances the output of the LTP phase splitter in such a way to counter the disbalance in conduction of the outout tubes. The way it's achieved is really simple - you make one of the plate oads of the LTP slightly variable compared to the calculated value. There are a number of ways to do this, mostly that side uses one reisitor value down on the standard set, and adds a trimmer pot in series to make it about equally adjustable around the fixed value of the other side. For instance, if the fixed resistor is 15k, they on the other side you put 12k in series with a 5k trimmer. This gives you a -3 +2k variance to play with, that introduces proportional disbalance in order to correct for the disbalance in the output stage. Other more complex schemes can be used that reciprocally vary both sides around a nominal value, usually by connecting the wiper of the trimmer to the power supply, and the left and right ends of the trimmer resistive elements in series with the left and right fixed plate load resistors of the LTP. Although sound in principle, this approach may suffer from trimmer wiper losing contact, which is usually guarded from by adding still 2 more extra resistors. In any case, a balance control is a great bonus to have, since it gives you the option to finely trim out all imbalances of the output stage, so much so that in ost cases you can get the same performance from unmatched output tubes by adjusting balance, that you would have with matched tubes and no balance - and then, you can still fine tune it and get the most out of your output transformer. I have never seen anything but improvement (even if slight) by including it, so I recomend it, especially where it's easy to implement such as in a LTP PI.
 
Last edited:
Ah, I see the problem. The application is geared towards supplying a constant DC current to a simple load, with out the other end of it changing voltage. I did not read the datasheet through but I suspect the 1u cap must be there for stability

Not clear what the OP intent was, but it looks to me like a simple voltage regulator for the stage, with the plate load resistor missing.

If, on the other hand, it was intended to use the LR8 as a CCS, then that's a different story.
 
Not clear what the OP intent was, but it looks to me like a simple voltage regulator for the stage, with the plate load resistor missing.

If, on the other hand, it was intended to use the LR8 as a CCS, then that's a different story.

Yes, using the LR8 as a CCS was the intention. The circuit was copied from the LR8 datasheet. The chip itself looks like a high voltage cousin of the venerable LM317, so a CCS can be had by simply connecting a resistor R in series with the chip output and the ADJ pin to the other end of this resistor, making the chip maintain a fixed reference voltage across the resistor, hence passing constant current of Vref/R. However, there is a capacitor to ground from the chip output in the datasheet drawing, necessary to insure stability, in which case it makes it a current source only at DC, the impedance will drop to the value of R as frequency goes up since the output of the chip is not capable of following any output voltage change because of C. This would make it suitable for something like a battery charger but not a real CCS. Kevinkr pointed this out a few posts up.
 
Last edited:
Since you are using the output tranny secondary to ground your cathode through you need to consider the total resistance to ground. You'll have maybe .5 -.75 ohms additional and that is a large percentage of error with respect to a 1 ohm resistor when calculating the cathode current. Better to leave it at 10 ohms. Even then you should still read the total for accuracy.

Yes, why don't we all assume that the other end of a bias setting resistor is always tied to ground even when it's not and add more disadvantages to our design so that heaven forbid we should not change our assumptions by say, reading the manual, even if they are wrong and potentially dangerous.
If one can look up what value they should measure one should really not have a problem moving that one test clip to the actual resistor pin, which is actually the correct way to do it regardless of where it's connected.
And never mind the assumption that the 0 to 4 and 4 to 16 ohm taps almost as a rule do not have the same DC resistance.
I'm all for 'alternative' but not when it has inferior results at same or added cost and especially not when it reinforces bad practice and sloppy standards.
 
I've just had a crisis of conscience and am thinking I'm going to go the 700V LTP supply route after all. It just makes more sense and should make for a much cleaner amp. I have a spare 10H choke I can use, and stacking capacitors is probably worth it. I can also give the 6AG5 a nicer amount of current. ]
 
Explanation: I'm using a choke-input power supply off a 500-0-500V PT, netting around 460V B+. However I can use another rectifier with a CLC supply to net around 700V and give this to the LTP, which will help a lot with getting clean swing. I'll post both signal and PSU schematics this time.
 
Thanks to everyone for all their input. I finally got a new schematic up. Have yet to figure out exact resistor values, and a proper CCS implementation, but the bare bones are there. For the bias supply I was thinking 150K from the wipers to ground, a 25K trimmer, and 33K for the other four.

There should be ample swing from the LTP this time, on the order of +/- 200V. The 700V D+ is derived from a separate CLC supply from the Hammond 282X, 500-0-500V. Turns out a CCS is hardly necessary on the first stage, which is operating at 4v g-k by the way.

SIGNAL
An externally hosted image should be here but it was not working when we last tested it.


6AG5 Input
An externally hosted image should be here but it was not working when we last tested it.


6N1P LTP
An externally hosted image should be here but it was not working when we last tested it.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.