It can be made to work, but you have to be able to adjust the current setting of the CCS in the tail very finely. Essentially, the two CCSs up top are fighting the CCS down the bottom. It works in simulation, but in the real world it's finicky. It's only really worth using if you don't have enough HT to be able to afford the voltage drop caused by resistive anode loads.
I was thinking of the fact that the top and lower current sources are going to fight eachother and was wondering how can I adjust them.
On the other hand, I do have as much voltage as needed, but I thought the active loads will be able to lower distortion at very low levels.
I have to add that this circuit is built in order to drive a 5998A PP final stage.
Do the calculated part values seem o.k?
On the other hand, I do have as much voltage as needed, but I thought the active loads will be able to lower distortion at very low levels.
I have to add that this circuit is built in order to drive a 5998A PP final stage.
Do the calculated part values seem o.k?
On the spot
In theory, the LED drifts with temperature in the same direction as Vbe, so they tend to hold the voltage across the current programming resistance constant, holding current constant. In practice, there's quite a variation between LEDs and the LED might not be at the same temperature as the transistor. If the LED is rectangular and the transistor a TO92 package, then you could epoxy them together to assist in temperature tracking, but the fact remains that there's a significant thermal resistance between the transistor die and the outside surface of the package and the LED die and its outside surface, and that will always cause errors.
Positioning: Obviously, it is not a good idea to put the circuit near localised sources of heat like transistor heatsinks or where the temperature can change rapidly. It's a good idea to put the tail CCS and the anode CCSs reasonably near to one another so that they are all at the same temperature. That way, the drift that wasn't cancelled between the LED and Vbe may be cancelled between the anode and cathode circuits to keep anode voltage approximately constant (the absolute value of current isn't important, but the anode voltages are).
Does it matter?: If the circuit has 20V across the anode CCSs but only swings 10Vpk-pk of signal, then a few volts of drift won't matter. It's all down to how closely you approach the limits. My experience is that it's more stable than you'd expect.
Test it: Wave a hairdryer at it whilst monitoring anode voltages with full expected signal applied.
In theory, the LED drifts with temperature in the same direction as Vbe, so they tend to hold the voltage across the current programming resistance constant, holding current constant. In practice, there's quite a variation between LEDs and the LED might not be at the same temperature as the transistor. If the LED is rectangular and the transistor a TO92 package, then you could epoxy them together to assist in temperature tracking, but the fact remains that there's a significant thermal resistance between the transistor die and the outside surface of the package and the LED die and its outside surface, and that will always cause errors.
Positioning: Obviously, it is not a good idea to put the circuit near localised sources of heat like transistor heatsinks or where the temperature can change rapidly. It's a good idea to put the tail CCS and the anode CCSs reasonably near to one another so that they are all at the same temperature. That way, the drift that wasn't cancelled between the LED and Vbe may be cancelled between the anode and cathode circuits to keep anode voltage approximately constant (the absolute value of current isn't important, but the anode voltages are).
Does it matter?: If the circuit has 20V across the anode CCSs but only swings 10Vpk-pk of signal, then a few volts of drift won't matter. It's all down to how closely you approach the limits. My experience is that it's more stable than you'd expect.
Test it: Wave a hairdryer at it whilst monitoring anode voltages with full expected signal applied.
The question about dueling current sources (plates versus cathode CCSs) in a diff amp is an interesting conundrum that I’ve faced in the past. It is a potential problem for both DC stability and for AC signal performance.
I think of it this way: If we apply a signal only to one grid, then the same triode’s cathode must drive the other cathode, as well as the cathode CCS. We want the signal current to feed only into the other cathode, and not into the cathode CCS. This requirement ensures that the diff amp acts like a true diff amp – dual but opposite outputs, with low common-mode response. This requirement means that the resistance looking into the other cathode must be much lower than the resistance of the cathode CCS. The resistance looking into the “other” cathode will be (Rx+rp)/(mu+1), where Rx is the resistance of the “other” plate’s CCS. This cathode resistance must be much less than Ry, the resistance of cathode CCS. CCSs can have rather indeterminate values of resistance. Or, said another way, the values are determined by rather sloppy, and non-linear, active component parameters, such as hfe and 1/hoe, if we are forced at gunpoint to use BJTs here. Usually the resistance value is so high that it is easily swamped by some other parallel resistance, such as a plate resistance, but that assumption cannot necessarily be made here. Even if Rx and Ry were predictable and exactly the same, the ratio of Ry to (Rx+rp)/(mu+1) would then only be about mu, hardly enough for ideal diff amp behavior. So you almost want the cathode CCS to be a “better” CCS than the plate CCS, and that stray thought should indicate a fundamental problem.
I have tried common-mode feedback circuit in the past to address these issues. If you tie a large value resistor to each plate and connect the other ends together, the voltage at the joined ends will be Vplate1 – Vplate2, which is the common-mode output voltage. Differential (desired signal) voltages are ignored. The catch is that these resistors must themselves have much larger values than the plate CCS resistance to avoid degrading the high plate load resistance, and that’s tough. In tube designs where there is already a buffer on each plate, such as a differential mu stage/SRPP or a diff amp followed by a pair of CFs, the low impedance outputs can safely feed the common-mode summing resistors without degrading the plate CCS effect. If we sample this summed voltage and then apply it (in proper polarity and with adequate gain) to a voltage-controlled CCS in the diff amp’s cathode, then that CCS is forced to react in a way that ensures that there is no common-mode signal present in the output plates, or said another way, that the output voltages are equal and opposite. Back to true diff amp behavior.
If this common-mode-only feedback loop is low-pass filtered at a very low frequency, only DC stability is addressed, although usefully. If the loop runs wide open across the audio range, then the AC signal balance across the plates is improved. But now we’ve added more active parts influencing the audio signal, although that influence is at a second order (reduced by the native CM rejection).
In my trials, I used triode-based CCSs at the plates and a pentode-based CCS in the cathode. As I recall, I buffered (with a CF) and level-shifted (with zeners and yet another CCS in the CF cathode circuit) the sample common-mode output voltage and applied it to the control grid of the pentode in the CCS. I recall that it did function, but the added complexity had exceeded its net value for that project. Also, as all the node resistances go to astronomically high values, stray capacitances now have a open playing field, especially those ugly non-linear capacitances that you get for free if you use (at gunpoint) solid-state CCS components.
Still, for the adventurous tube designer, the common-mode feedback concept might find a useful home. I’ve seen similar circuits in entirely solid-state designs, so I know that the concept has a history.
In the end, for my “simple” diff amp project, I reverted to choosing “good old” plate RESISTORS with large, but tractable, values (RL >> rp) and made a decent cathode CCS. With this standard approach, you can easily design for Ry >> (RL+rp)/(mu+1), as required for good diff amp behavior.
I think of it this way: If we apply a signal only to one grid, then the same triode’s cathode must drive the other cathode, as well as the cathode CCS. We want the signal current to feed only into the other cathode, and not into the cathode CCS. This requirement ensures that the diff amp acts like a true diff amp – dual but opposite outputs, with low common-mode response. This requirement means that the resistance looking into the other cathode must be much lower than the resistance of the cathode CCS. The resistance looking into the “other” cathode will be (Rx+rp)/(mu+1), where Rx is the resistance of the “other” plate’s CCS. This cathode resistance must be much less than Ry, the resistance of cathode CCS. CCSs can have rather indeterminate values of resistance. Or, said another way, the values are determined by rather sloppy, and non-linear, active component parameters, such as hfe and 1/hoe, if we are forced at gunpoint to use BJTs here. Usually the resistance value is so high that it is easily swamped by some other parallel resistance, such as a plate resistance, but that assumption cannot necessarily be made here. Even if Rx and Ry were predictable and exactly the same, the ratio of Ry to (Rx+rp)/(mu+1) would then only be about mu, hardly enough for ideal diff amp behavior. So you almost want the cathode CCS to be a “better” CCS than the plate CCS, and that stray thought should indicate a fundamental problem.
I have tried common-mode feedback circuit in the past to address these issues. If you tie a large value resistor to each plate and connect the other ends together, the voltage at the joined ends will be Vplate1 – Vplate2, which is the common-mode output voltage. Differential (desired signal) voltages are ignored. The catch is that these resistors must themselves have much larger values than the plate CCS resistance to avoid degrading the high plate load resistance, and that’s tough. In tube designs where there is already a buffer on each plate, such as a differential mu stage/SRPP or a diff amp followed by a pair of CFs, the low impedance outputs can safely feed the common-mode summing resistors without degrading the plate CCS effect. If we sample this summed voltage and then apply it (in proper polarity and with adequate gain) to a voltage-controlled CCS in the diff amp’s cathode, then that CCS is forced to react in a way that ensures that there is no common-mode signal present in the output plates, or said another way, that the output voltages are equal and opposite. Back to true diff amp behavior.
If this common-mode-only feedback loop is low-pass filtered at a very low frequency, only DC stability is addressed, although usefully. If the loop runs wide open across the audio range, then the AC signal balance across the plates is improved. But now we’ve added more active parts influencing the audio signal, although that influence is at a second order (reduced by the native CM rejection).
In my trials, I used triode-based CCSs at the plates and a pentode-based CCS in the cathode. As I recall, I buffered (with a CF) and level-shifted (with zeners and yet another CCS in the CF cathode circuit) the sample common-mode output voltage and applied it to the control grid of the pentode in the CCS. I recall that it did function, but the added complexity had exceeded its net value for that project. Also, as all the node resistances go to astronomically high values, stray capacitances now have a open playing field, especially those ugly non-linear capacitances that you get for free if you use (at gunpoint) solid-state CCS components.
Still, for the adventurous tube designer, the common-mode feedback concept might find a useful home. I’ve seen similar circuits in entirely solid-state designs, so I know that the concept has a history.
In the end, for my “simple” diff amp project, I reverted to choosing “good old” plate RESISTORS with large, but tractable, values (RL >> rp) and made a decent cathode CCS. With this standard approach, you can easily design for Ry >> (RL+rp)/(mu+1), as required for good diff amp behavior.
Thanks everybody for answering!
I tried the circuit last night and as most of you predicted the thermal drift is so big that blowing with a hair dryer cold air, changed everything dramatically.
Thus I guess it mast be very hard to stabilise such a circuit.
What about a simpler form though? ( I will try to test it tonight)
I tried the circuit last night and as most of you predicted the thermal drift is so big that blowing with a hair dryer cold air, changed everything dramatically.
Thus I guess it mast be very hard to stabilise such a circuit.
What about a simpler form though? ( I will try to test it tonight)
Attachments
Hi ,
1n4148 is drawn the wrong way round . I built something similar using ZTX558 anode loads with 6SL7 with the lower CCS set to 1.6mA . It was a bitch to set up and intermittently distorted at first . It took a bit of tweaking of the tail current to get it stable . Also I had problems when connecting my meter to the anodes . 90m ohm of 10m resistors was connected in series with the meter , otherwise once the meter was disconnected the amp would distort . With the top loads , the amp sounds much better than with resistors at the limited HT available and at least I now have enough swing from the driver stage to drive triode connected PL508 to full power without the driver clipping
cheers
316a
1n4148 is drawn the wrong way round . I built something similar using ZTX558 anode loads with 6SL7 with the lower CCS set to 1.6mA . It was a bitch to set up and intermittently distorted at first . It took a bit of tweaking of the tail current to get it stable . Also I had problems when connecting my meter to the anodes . 90m ohm of 10m resistors was connected in series with the meter , otherwise once the meter was disconnected the amp would distort . With the top loads , the amp sounds much better than with resistors at the limited HT available and at least I now have enough swing from the driver stage to drive triode connected PL508 to full power without the driver clipping
cheers
316a
Panos29,
So long as you have current sources above and below, there will be DC bias point instability (unless you implement a servo scheme or settle on plate resistors). Your simpler CCSs probably won't help much, or if they it do, it is only because they are not behaving as a very good CCSs. Think of it this way:
A current source, by definition, will force (or “demand”) a certain current, which you set by choice of parts. A CCS will allow the voltage across it to vary, but the current must stay constant, to the extent that it behaves as a true current source. In overload, the voltage across the CCS can go either too high or too low and the CCS will cease to behave as a current source.
Let’s say that each of your upper CCSs is set to force exactly 8.0mA of DC current into each plate (for sake of example) That DC current has no where to go but through the tubes and into the cathodes. The DC cannot exit through the plate coupling caps at DC, and it won’t go out the grids if the tube is biased in the normal way. From the joined cathodes, the currents (now adding to 16.0mA – exactly, in this example) must go through the bottom CCS and into ground. Since current can only circulate in a complete loop, the loop is completed through the power supply back to the B+. No current has been added or subtracted in this loop; it must be 16.0mA all the way around (a basic law of physics).
Now suppose that you designed your lower (cathode) CCS for 16.0mA, expecting to exactly grab all of the current coming at it from the upper CCSs. But due to component variations, it actually wants to draw in 16.05mA (or maybe 15.99mA) instead of 16.0mA. It will fight for 16.05mA even over a large range of voltage shifts. But the upper CCSs (together) force out just 16.0mA and not a bit more or less. Something has to give. This is where the term “dueling” comes in. Voltages will shift dramatically (and immediately) in a vain attempt to equalize these currents, until one or more of the current sources go into overload, ceasing to perform as a current source, at which point the circuit satisfies the laws of physics. But this overload situation is not what you wanted.
You can delicately tweak one or more of the CCS to momentarily achieve the desired balance, but as soon as you walk away, the currents will drift ever so slightly and the voltages will go awry again. As you waved your hair dryer over the circuit you forced these currents to drift more quickly, in unpredictable ways, and the battle between the CCS resulted in moving voltages.
If you use plate resistors instead of plate CCSs, these resistors will pass varying currents in response to varying voltages across them (or vice versa) according to Ohm’s law. Circuit conditions can be made to settle more predictably. A CCS can and should still be used in the cathode circuit.
Hope this helped.
So long as you have current sources above and below, there will be DC bias point instability (unless you implement a servo scheme or settle on plate resistors). Your simpler CCSs probably won't help much, or if they it do, it is only because they are not behaving as a very good CCSs. Think of it this way:
A current source, by definition, will force (or “demand”) a certain current, which you set by choice of parts. A CCS will allow the voltage across it to vary, but the current must stay constant, to the extent that it behaves as a true current source. In overload, the voltage across the CCS can go either too high or too low and the CCS will cease to behave as a current source.
Let’s say that each of your upper CCSs is set to force exactly 8.0mA of DC current into each plate (for sake of example) That DC current has no where to go but through the tubes and into the cathodes. The DC cannot exit through the plate coupling caps at DC, and it won’t go out the grids if the tube is biased in the normal way. From the joined cathodes, the currents (now adding to 16.0mA – exactly, in this example) must go through the bottom CCS and into ground. Since current can only circulate in a complete loop, the loop is completed through the power supply back to the B+. No current has been added or subtracted in this loop; it must be 16.0mA all the way around (a basic law of physics).
Now suppose that you designed your lower (cathode) CCS for 16.0mA, expecting to exactly grab all of the current coming at it from the upper CCSs. But due to component variations, it actually wants to draw in 16.05mA (or maybe 15.99mA) instead of 16.0mA. It will fight for 16.05mA even over a large range of voltage shifts. But the upper CCSs (together) force out just 16.0mA and not a bit more or less. Something has to give. This is where the term “dueling” comes in. Voltages will shift dramatically (and immediately) in a vain attempt to equalize these currents, until one or more of the current sources go into overload, ceasing to perform as a current source, at which point the circuit satisfies the laws of physics. But this overload situation is not what you wanted.
You can delicately tweak one or more of the CCS to momentarily achieve the desired balance, but as soon as you walk away, the currents will drift ever so slightly and the voltages will go awry again. As you waved your hair dryer over the circuit you forced these currents to drift more quickly, in unpredictable ways, and the battle between the CCS resulted in moving voltages.
If you use plate resistors instead of plate CCSs, these resistors will pass varying currents in response to varying voltages across them (or vice versa) according to Ohm’s law. Circuit conditions can be made to settle more predictably. A CCS can and should still be used in the cathode circuit.
Hope this helped.
And finally, the winner is...
Exactly!
So, after much fiddling with both types of anode load ccs I have to note that the second behaved a lot better.
Still, distortion measurements showed that the distortion did not go down as, falsely, predicted.
The simplest solution was indeed the resistive loads (50K) with the current source located only at the tail which gave marginally better distortion figures than the second drawing.
Thanks everybody for the very explanative posts!
Exactly!
So, after much fiddling with both types of anode load ccs I have to note that the second behaved a lot better.
Still, distortion measurements showed that the distortion did not go down as, falsely, predicted.
The simplest solution was indeed the resistive loads (50K) with the current source located only at the tail which gave marginally better distortion figures than the second drawing.
Thanks everybody for the very explanative posts!
Here's an idea, less tongue in cheek than the last one I posted
If there are cathode followers after the LTP, it would be possible to use bootstrapping for the plates of the LTP, which effectively solves the DC operating point conundrum and the requirement for precisely half the current from the top CCSs. The bootstrap only 'multiplies' the plate resistors for AC, leaving the current intact. Of course, this relies on usable output from the followers - clipping behaviour should be carefully investigated, as well as sourcing signifficant current from the followers (i.e. A2 or AB2 operation).
If there are cathode followers after the LTP, it would be possible to use bootstrapping for the plates of the LTP, which effectively solves the DC operating point conundrum and the requirement for precisely half the current from the top CCSs. The bootstrap only 'multiplies' the plate resistors for AC, leaving the current intact. Of course, this relies on usable output from the followers - clipping behaviour should be carefully investigated, as well as sourcing signifficant current from the followers (i.e. A2 or AB2 operation).
Re: What happemed to Geeks idea
I tried it and it didn't work well at all. The problem was that it has only a very limited output voltage swing. That's fine in a transistor amplifier where the gain comes from the VAS stage and not the input differential pair, but it's not so hot with valves.
gingertube said:
I tried it and it didn't work well at all. The problem was that it has only a very limited output voltage swing. That's fine in a transistor amplifier where the gain comes from the VAS stage and not the input differential pair, but it's not so hot with valves.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.