Yes, but what we also do know is that by changing the grid-to-cathode voltage by only 1 Volt a 4 mA change in plate current occurs. How come?
It is because the field strength perpendicular to the surface of the cathode (which ultimately determines cathode current) is determined by the plate voltage and the grid voltage. Only the grid voltage has a factor mu more influence on this field strength than the plate voltage. The cathode does not have a clue wether its in a diode, a triode or a heptode for that matter, it only sees one electrical field strength at each spot. So in our typical 300B, where the grid has 3.85 (WE figure for 300 V on plate, -61 V on grid) more influence on this field than the plate, from the center of the cathode (V=0) it looks like it is in a diode with 300-3.85*61=65.15 Volts on the plate. At the lowest extreme (negative terminal) it looks like a diode with 300-3.85*(58.5)=74.78 Volts on the plate, and at the highest extreme (positive terminal) it looks like a diode with 300-3.85*63.5=55.53 Volts on the plate. This should already give a hint of significance.
As a vacuum diode's current is proportional to V^1.5, the ratio between the current densities carried by both extremes of the filament is (74.78/55.53)^1.5=1.35^1.5=1.56. So according to this (fairly exact) calculation, the negative extreme is working a whopping 56% harder than the positive extreme of the filament! Even worse than in my quick calculation a few posts earlier.
It is because the field strength perpendicular to the surface of the cathode (which ultimately determines cathode current) is determined by the plate voltage and the grid voltage. Only the grid voltage has a factor mu more influence on this field strength than the plate voltage. The cathode does not have a clue wether its in a diode, a triode or a heptode for that matter, it only sees one electrical field strength at each spot. So in our typical 300B, where the grid has 3.85 (WE figure for 300 V on plate, -61 V on grid) more influence on this field than the plate, from the center of the cathode (V=0) it looks like it is in a diode with 300-3.85*61=65.15 Volts on the plate. At the lowest extreme (negative terminal) it looks like a diode with 300-3.85*(58.5)=74.78 Volts on the plate, and at the highest extreme (positive terminal) it looks like a diode with 300-3.85*63.5=55.53 Volts on the plate. This should already give a hint of significance.
As a vacuum diode's current is proportional to V^1.5, the ratio between the current densities carried by both extremes of the filament is (74.78/55.53)^1.5=1.35^1.5=1.56. So according to this (fairly exact) calculation, the negative extreme is working a whopping 56% harder than the positive extreme of the filament! Even worse than in my quick calculation a few posts earlier.
The space charge is formed by thermionic emission regardless of the presence of plate or grid.
It is reasonable to expect a variation in the geometery ( thickness, radius) of the space charge with respect to the voltage gradient along the grid. Does this variation in geometry cause a variation in emission? SY's reasoning would suggest not.
Emission feeds the space charge and is limited by the field of the space charge itself... not the grid. The space charge feeds the plate... via the grid.
All of this suggests that DC will not preferentially wear-out one end of the cathode. Were this the case, thermionic diodes would be "stripped" whenever forward biased.
It is reasonable to expect a variation in the geometery ( thickness, radius) of the space charge with respect to the voltage gradient along the grid. Does this variation in geometry cause a variation in emission? SY's reasoning would suggest not.
Emission feeds the space charge and is limited by the field of the space charge itself... not the grid. The space charge feeds the plate... via the grid.
All of this suggests that DC will not preferentially wear-out one end of the cathode. Were this the case, thermionic diodes would be "stripped" whenever forward biased.
I can not follow this reasoning at all. It doesn't matter what exact structures surround a cathode. The rate at which the space charge around the cathode is being depleted depends on the external electric field, which is simply the sum of all external fields. Due to the fact that the directly heated cathode is not unipotential, these fields will be different in strength for different locations of the cathode surface. Without this mechanism, controlling the current would simply be impossible.
But its getting late, I am thinking of an experiment that should settle this argument. That will have to wait until tomorrow.
I have thought about DC fiaments with DHT's and the voltage gradient along the filament with reference to the grid. On a tube like the 45 it shouldn't be an issue dut to the small 2.5 volt filament voltage compared to the bias voltage. The 300B is a little bit worse off and the 211 should be terrible. The all time worse case has to be the 811A. The tube has a Mu of 160 the 6.3 volts of the filament could mean 50 mA difference in plate current. One end of the filament could be supplying most of the plate current, while the other end is cutoff.
On the other hand I find better sound using DC. When developing the TubelabSE amps I noticed a "fogginess" in the sound on some vocal music with AC heating. There was no noticible hum in the speakers (87 db speakers that are probably 82db at 60Hz) but an FFT analysis revealed intermodulation products from the AC filament voltage. In short this means that when you put a 1KHz tone into the amp you get back the 1KHz signal and some 940Hz and some 1060Hz. This was adding about 1/3 % of non musically related distortion to the amp. Further testing revealed that the distortion level was tube related, some tubes were worse than others. All tubes were 45's. DC heating completely removed the distortion. I have used DC heating on all my DHT amps since.
I believe that a clean sinewave an the 100KHz region may be useful to heat the filament on some DHT's, but my initial experiments were inconclusive. I was using a home made MOSFET audio amp and a 1:1 toroid transformer to light the filament on an 811A. My FFT analyzer did not pick up any spurious beat notes, but it only goes to 48KHz. The MOSFET amplifier did not like putting out 25 watts at 100KHz and rewarded me with smoke. END OF EXPERIMENTS.
On the other hand I find better sound using DC. When developing the TubelabSE amps I noticed a "fogginess" in the sound on some vocal music with AC heating. There was no noticible hum in the speakers (87 db speakers that are probably 82db at 60Hz) but an FFT analysis revealed intermodulation products from the AC filament voltage. In short this means that when you put a 1KHz tone into the amp you get back the 1KHz signal and some 940Hz and some 1060Hz. This was adding about 1/3 % of non musically related distortion to the amp. Further testing revealed that the distortion level was tube related, some tubes were worse than others. All tubes were 45's. DC heating completely removed the distortion. I have used DC heating on all my DHT amps since.
I believe that a clean sinewave an the 100KHz region may be useful to heat the filament on some DHT's, but my initial experiments were inconclusive. I was using a home made MOSFET audio amp and a 1:1 toroid transformer to light the filament on an 811A. My FFT analyzer did not pick up any spurious beat notes, but it only goes to 48KHz. The MOSFET amplifier did not like putting out 25 watts at 100KHz and rewarded me with smoke. END OF EXPERIMENTS.
Hmmm... I think kevinkr has me confused with somebody else....
The filament converters I made generate a 100kHz sine wave. I didn't have any real problems with them. I just posted some info on them at http://www.pmillett.com/hf_fil.htm
The supposed advantage of HF heating is that you get AC on the filament but don't have any audible hum to worry about.
As to whether there are any problems using DC, I suspect not. Some people worry about the potential differemce from end to end of the filament but I doubt that the effect is enough to worry about myself.
We did a "shoot-out"" at ETF04 (or was it 03?) where we compared several heating methods by listening to them on an test amp. We tested the HF filaments, CCS DC, constant-voltage DC (a lab supply and batteries, I think), and normal AC.
The audience (60 or 70 geeks like me) didn't agree 100%, but as I recall constant-current DC and the HF drive were rated highest. The differences between each method were subtle but audible.
I think Tubelabs comment about the intermod with the 60Hz filament is valid. That's also my experience. I guess with a 100kHz heater the intermod products are also all ultrasonic...
Is it worth the complexity to do HF heating? Probably not. I designed and built it mostly as an excersize, and to contribute to the shootout, not because I think there's anything wrong with DC (or even AC).
If I were building a 300B amp, I think I'd go with a CCS, like the TentLabs boards. Or AC and a hum pot - I'm not hypersensitive to a little hum and LF IMD. Gives it that ole' homey tubey sound
Back to the original issue - the noise - I would guess that the RF is getting into the circuit and getting amplified and maybe detected somewhere? The description is a lot like what you hear when a circuit is oscillating...
If you have a scope it should be obvious; there should be RF all over the place.
Pete
The filament converters I made generate a 100kHz sine wave. I didn't have any real problems with them. I just posted some info on them at http://www.pmillett.com/hf_fil.htm
The supposed advantage of HF heating is that you get AC on the filament but don't have any audible hum to worry about.
As to whether there are any problems using DC, I suspect not. Some people worry about the potential differemce from end to end of the filament but I doubt that the effect is enough to worry about myself.
We did a "shoot-out"" at ETF04 (or was it 03?) where we compared several heating methods by listening to them on an test amp. We tested the HF filaments, CCS DC, constant-voltage DC (a lab supply and batteries, I think), and normal AC.
The audience (60 or 70 geeks like me) didn't agree 100%, but as I recall constant-current DC and the HF drive were rated highest. The differences between each method were subtle but audible.
I think Tubelabs comment about the intermod with the 60Hz filament is valid. That's also my experience. I guess with a 100kHz heater the intermod products are also all ultrasonic...
Is it worth the complexity to do HF heating? Probably not. I designed and built it mostly as an excersize, and to contribute to the shootout, not because I think there's anything wrong with DC (or even AC).
If I were building a 300B amp, I think I'd go with a CCS, like the TentLabs boards. Or AC and a hum pot - I'm not hypersensitive to a little hum and LF IMD. Gives it that ole' homey tubey sound
Back to the original issue - the noise - I would guess that the RF is getting into the circuit and getting amplified and maybe detected somewhere? The description is a lot like what you hear when a circuit is oscillating...
If you have a scope it should be obvious; there should be RF all over the place.
Pete
Having played a bit with ultrasonic filament supplies in the past, I have some half-baked and random thoughts (or pseudo-random thoughts – see below) to toss on the table, FWIW:
1.) A discrete frequency source presents the *possibility* of intermodulation with other signals present, thus producing in-band tones. This is even an issue with the standard 60 Hz supplies, of course.
2.) Not only could there be gm modulation at this discrete frequency, but these higher discrete frequencies can pass readily through stray capacitances and interelectrode capacitances.
3.) If a digital sound source is used, digital clocks and other spurious tone leakage could intermodulate with an ultrasonic heater supply frequency generating in-band products.
4.) If two or more ultrasonic heater supplies are needed in an amp, they should be phase-locked or derived from common frequency source so that there is no difference frequency. If one tube leaked some of its 100 KHz supply frequency and this leakage mixed with another tube’s leakage at 101 KHz, a 1 KHz tone would be produced in the presence of any non-linearity, the presence of which we can be assured.
5.) In theory, the PDF (probability density function) of the supply’s voltage should be as “compact” as possible to limit voltage extremes from one end of the cathode to the other at peaks. A tight PDF minimizes modulation of the tube’s gm into more non-linear regions (see the valid 811 worry above). DC or an ultrasonic square wave would have the best PDF (since peak = rms), but a square wave also produces a spray of harmonics that could play havoc. However, a very efficient square wave oscillator is easy to make with switching FETs. A little judicious low-pass harmonic filtering might be a good compromise, while still preserving much of the tight bi-modal PDF. This is where a wide band thermal true-rms meter comes in handy. The ultrasonic sine wave that Pete uses, or the standard 60Hz, has a worse PDF (peaks are 1.414 times rms), but harmonics are largely absent. It might be interesting to compare both schemes.
6.) Being a techno-geek, I had started to sketch out a pseudo-random ultrasonic noise generator as a heater supply. My thought was to avoid the use of any discrete frequencies so that intermodulation could only produce imperceptible in-band noise at a very low per-Hertz-of-BW level. My thought would be to create a shaped bandwidth above the audio band not unlike the noise shaping of DSD/SACD. Perhaps noise shaped from 100 KHz to a few hundred KHz. The goal would be to create a band of noise to heat the filaments, safely above the audio band, but not allowed to stray too high in frequency to avoid excessive capacitive coupling. If you start with a pseudo random number generator, you get a square-wave-like PDF. With noise filtering, the Central Limit Theorem predicts a morphing of the bi-modal discrete voltage PDF into something approaching Gaussian, which has high peak to rms PDF (infinite actually for a perfect Gaussian distribution). The PDF here would be a compromise between the tight square wave PDF and the spread-out Gaussian, maybe statistically closer to a middle-of-the-road sinusoidal PDF. And no discrete frequencies are present to cause problems. If more than one pseudo-random supply is needed, it would be best to sync them to avoid noise-noise intermodulation, although these effects would be less deleterious than intermodulating sine waves. If you are thinking this is all to complicated (well, maybe it is), you can make a decent PN generator with just a few CMOS ICs from the parts bin. Noise bandwidth shaping would be accomplished with a passive filter followed by a stage or two of high-speed opamp filtration. The output stage could be FETs in a quasi-linear amp design, but some non-linearity and even some clipping could be allowed. Not too hard. Once the proper rms voltage was set by a pot with a good rms meter, the stability of the heating voltage could be excellent. Is all this worth the effort? I dunno…
7.) Stray thought: a little ultrasonic noise might be a good thing for amps and tweeters, like analog tape bias.
Presented for your further thoughts and/or derision
1.) A discrete frequency source presents the *possibility* of intermodulation with other signals present, thus producing in-band tones. This is even an issue with the standard 60 Hz supplies, of course.
2.) Not only could there be gm modulation at this discrete frequency, but these higher discrete frequencies can pass readily through stray capacitances and interelectrode capacitances.
3.) If a digital sound source is used, digital clocks and other spurious tone leakage could intermodulate with an ultrasonic heater supply frequency generating in-band products.
4.) If two or more ultrasonic heater supplies are needed in an amp, they should be phase-locked or derived from common frequency source so that there is no difference frequency. If one tube leaked some of its 100 KHz supply frequency and this leakage mixed with another tube’s leakage at 101 KHz, a 1 KHz tone would be produced in the presence of any non-linearity, the presence of which we can be assured.
5.) In theory, the PDF (probability density function) of the supply’s voltage should be as “compact” as possible to limit voltage extremes from one end of the cathode to the other at peaks. A tight PDF minimizes modulation of the tube’s gm into more non-linear regions (see the valid 811 worry above). DC or an ultrasonic square wave would have the best PDF (since peak = rms), but a square wave also produces a spray of harmonics that could play havoc. However, a very efficient square wave oscillator is easy to make with switching FETs. A little judicious low-pass harmonic filtering might be a good compromise, while still preserving much of the tight bi-modal PDF. This is where a wide band thermal true-rms meter comes in handy. The ultrasonic sine wave that Pete uses, or the standard 60Hz, has a worse PDF (peaks are 1.414 times rms), but harmonics are largely absent. It might be interesting to compare both schemes.
6.) Being a techno-geek, I had started to sketch out a pseudo-random ultrasonic noise generator as a heater supply. My thought was to avoid the use of any discrete frequencies so that intermodulation could only produce imperceptible in-band noise at a very low per-Hertz-of-BW level. My thought would be to create a shaped bandwidth above the audio band not unlike the noise shaping of DSD/SACD. Perhaps noise shaped from 100 KHz to a few hundred KHz. The goal would be to create a band of noise to heat the filaments, safely above the audio band, but not allowed to stray too high in frequency to avoid excessive capacitive coupling. If you start with a pseudo random number generator, you get a square-wave-like PDF. With noise filtering, the Central Limit Theorem predicts a morphing of the bi-modal discrete voltage PDF into something approaching Gaussian, which has high peak to rms PDF (infinite actually for a perfect Gaussian distribution). The PDF here would be a compromise between the tight square wave PDF and the spread-out Gaussian, maybe statistically closer to a middle-of-the-road sinusoidal PDF. And no discrete frequencies are present to cause problems. If more than one pseudo-random supply is needed, it would be best to sync them to avoid noise-noise intermodulation, although these effects would be less deleterious than intermodulating sine waves. If you are thinking this is all to complicated (well, maybe it is), you can make a decent PN generator with just a few CMOS ICs from the parts bin. Noise bandwidth shaping would be accomplished with a passive filter followed by a stage or two of high-speed opamp filtration. The output stage could be FETs in a quasi-linear amp design, but some non-linearity and even some clipping could be allowed. Not too hard. Once the proper rms voltage was set by a pot with a good rms meter, the stability of the heating voltage could be excellent. Is all this worth the effort? I dunno…
7.) Stray thought: a little ultrasonic noise might be a good thing for amps and tweeters, like analog tape bias.
Presented for your further thoughts and/or derision
Or like dither. Helps linearize suspension hysteresis. Yeah, that's the ticket.
I thought a bit about the interesting calculation on current density along a DC filament. It bothers me because it makes use of equations that are derived with certain assumptions, then violates those assumptions. I'm thinking that it's at best a second order effect, a tilting of the zero-potential surface in the space charge.
Now it may be that the use of that equation is correct and I'm all wet, but I'd like the throw the question out, how could one measure that current density remotely?
I thought a bit about the interesting calculation on current density along a DC filament. It bothers me because it makes use of equations that are derived with certain assumptions, then violates those assumptions. I'm thinking that it's at best a second order effect, a tilting of the zero-potential surface in the space charge.
Now it may be that the use of that equation is correct and I'm all wet, but I'd like the throw the question out, how could one measure that current density remotely?
Perspective.
This is interesting stuff. Let's assume that there's no signal going into the amplifier. That means the anode is at a constant voltage and current drawn from the cathode is determined by Vgk. Somehow or other, we adjust Vgk to achieve our required Ia. If we have DC on the heater, then for a typical bias point of -45V for a 2A3 (2.5V heater) we have 48.75V at one end and 46.25V at the other.
The question is whether a different Vgk at one end of the cathode to the other matters in terms of emission from the cathode. It has been suggested that it doesn't because Ia is drawn from the space-charge which is never stripped. But the space-charge has to be continuously replenished and if there's a different voltage between the grid and the cathode at one end to the other, then surely that means that one end must give up electrons preferentially? AC heating (whether 60Hz or RF) means the ends are continuously swapping over, and the effect averages out. With DC, that won't be the case. Does it matter? In the 811A example given earlier, it might. Does the hum from AC heating matter? No question about that, it does. So we're fretting about a possible second-order effect being traded for a definite first-order effect...
If you're really worried about uneven wear caused by DC heating, periodically swap over the heater connections. This could be done with a substantial relay and some very simple battery-powered logic so that each time the amplifier is switched on, the heater is powered the opposite way to the way it was powered the last time. That should average out over the lifetime of the valve...
This is interesting stuff. Let's assume that there's no signal going into the amplifier. That means the anode is at a constant voltage and current drawn from the cathode is determined by Vgk. Somehow or other, we adjust Vgk to achieve our required Ia. If we have DC on the heater, then for a typical bias point of -45V for a 2A3 (2.5V heater) we have 48.75V at one end and 46.25V at the other.
The question is whether a different Vgk at one end of the cathode to the other matters in terms of emission from the cathode. It has been suggested that it doesn't because Ia is drawn from the space-charge which is never stripped. But the space-charge has to be continuously replenished and if there's a different voltage between the grid and the cathode at one end to the other, then surely that means that one end must give up electrons preferentially? AC heating (whether 60Hz or RF) means the ends are continuously swapping over, and the effect averages out. With DC, that won't be the case. Does it matter? In the 811A example given earlier, it might. Does the hum from AC heating matter? No question about that, it does. So we're fretting about a possible second-order effect being traded for a definite first-order effect...
If you're really worried about uneven wear caused by DC heating, periodically swap over the heater connections. This could be done with a substantial relay and some very simple battery-powered logic so that each time the amplifier is switched on, the heater is powered the opposite way to the way it was powered the last time. That should average out over the lifetime of the valve...
SY,
I agree with the notion that the space charge is just getting pushed around a bit...
Check this out... way down toward the bottom... some loose confirmation.
If one turns this whole question upside down for a moment, you might wonder that AC is actually worse than DC. Whatever harmful effect present is spread evenly over both ends of the cathode. But is it possible that the cumulative effect is worse because the voltage peaks at 1.414 times the equivalent DC?
I agree with the notion that the space charge is just getting pushed around a bit...
Check this out... way down toward the bottom... some loose confirmation.
If one turns this whole question upside down for a moment, you might wonder that AC is actually worse than DC. Whatever harmful effect present is spread evenly over both ends of the cathode. But is it possible that the cumulative effect is worse because the voltage peaks at 1.414 times the equivalent DC?
Maybe. Since the defining equations for cathode current (and that tag-along gm) are non-linear functions of Vgk, a line frequency modulation of gm will occur in the AC driven case (actually at 100 Hz or 120Hz if the filament is physically symmetrical). When the heater voltage peaks the change in emission and the gm contribution at one end of the cathode (compared to the cathode midpoint) will not exactly oppose the change in emission and gm contribution at the other side of the cathode. Compare this situation to a few milliseconds later, where there is no heater voltage at all – the tube still “works” due to thermal inertia of course, but all sides of the cathode are now contributing equally.
If the peak AC heater voltage drives the characteristic toward an incipient limiting non-linearity, say onset of grid current, then it is possible to imagine that sine wave heating may be worse than DC due to its more stretched-out PDF of voltage.
Yes, indeed – very slow square waves after all.
I’m not sure exactly what you’re saying here. But I think that, while the long-term build-up of space charge is an averaging kind of thing, the emission from a portion of the cathode is instantaneously responsive to the Vgk in that vicinity (assuming that the cathode dimensions are much larger than the thickness of the space charge reservoir). Otherwise any tube, IDHT or DHT, would not respond instantaneously to cathode potential, and we know that’s not the case.
Can I have a whiteboard, please?
Yes, that was what I meant, the emission at the cathode surface must track any signal waveform and if we superimpose a heating waveform along the cathode, we must draw more electrons from one end of the cathode than the other.
Meanwhile, back to the lathe to make the wand for my microphone capsule...
Brian Beck said:
Yes, that was what I meant, the emission at the cathode surface must track any signal waveform and if we superimpose a heating waveform along the cathode, we must draw more electrons from one end of the cathode than the other.
Meanwhile, back to the lathe to make the wand for my microphone capsule...
I thought of the following way to determine if the cathode is used unevenly or not. It may sound a little barbarian, as it involves sacrificing a 300B to science. (I have one lone specimen that I am willing to annihilate). It works like this:
Connect a 300B grid-to-plate, as if it were a diode and heat it with DC. Now connect rectified but unbuffered AC across the negative filament terminal and the plate so it is forward biased. This will cycle the current from 0 to max 100 (or 120) times per second. Now crank up the AC until the current peaks just begin to level out. This is the point at which the space charge surrounding the cathode is completely stripped away for just a split second, every half mains period. This will ensure a good deal of damage to the cathode within a reasonalbe timespan. After a while, the emission will be severely degraded.
Here comes the trick: if the space charge levels out the emission across the cathode, the entire cathode suffers about the same amount of damage. If it doesn't, then the negative end of the cathode will suffer more damage than the positive end. In the latter case the emission will be rendered asymmetrical, and because of this the polarity of the filament will make a difference in plate current at a certain setting of the tube. If the space charge does level out emission, then this difference will not be observed.
I think that this experiment will give a reliable answer to the question posed here, whether AC is better for tube life than DC, or that it simply doesn't matter. Someone has called it a second-order effect, but to my wallet, this is a first order matter. Comments please!
Jurgen
Connect a 300B grid-to-plate, as if it were a diode and heat it with DC. Now connect rectified but unbuffered AC across the negative filament terminal and the plate so it is forward biased. This will cycle the current from 0 to max 100 (or 120) times per second. Now crank up the AC until the current peaks just begin to level out. This is the point at which the space charge surrounding the cathode is completely stripped away for just a split second, every half mains period. This will ensure a good deal of damage to the cathode within a reasonalbe timespan. After a while, the emission will be severely degraded.
Here comes the trick: if the space charge levels out the emission across the cathode, the entire cathode suffers about the same amount of damage. If it doesn't, then the negative end of the cathode will suffer more damage than the positive end. In the latter case the emission will be rendered asymmetrical, and because of this the polarity of the filament will make a difference in plate current at a certain setting of the tube. If the space charge does level out emission, then this difference will not be observed.
I think that this experiment will give a reliable answer to the question posed here, whether AC is better for tube life than DC, or that it simply doesn't matter. Someone has called it a second-order effect, but to my wallet, this is a first order matter. Comments please!
Jurgen
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.