Well, until I find anything better I'm using interstages between my 4P1L stages. I've eliminated all cathode bypass caps and coupling caps by using filament bias all through. Are you thinking of direct coupling or what? Hard to avoid 3 stages with DHTs, though Rod has ingenious solutions with his tube/ss hybrids. This is a bit off topic, except that it's worth expanding the "theoretical advantages" of DHTs to small tubes as well, which should be very on-topic.
It is not necessary to have all DHTs and interstage transformers. Here is my take on this:
http://www.diyaudio.com/forums/tubes-valves/183329-one-more-4p1l-se.html
In the experiment I took 4P1L output and 6J5P driver in triode mode. Only 2 stages were neded. 6J5P biased by LED and loaded on gyrator. As the result got super linearity with all needed swing. Then in order to drive 4P1L in class A2 I used a cathode follower loaded on CCS. Try such solution and compare with 3 stages and interstage transformers. You can't get such results.
The only drawback I found, limited negative supply to drive CCS, so on some overload peaks it clipped (when we listened to Glinka's cannon shot), but it can be increased so the probability of such hard clipping due to lack of voltage would be very low.
http://www.diyaudio.com/forums/tubes-valves/183329-one-more-4p1l-se.html
In the experiment I took 4P1L output and 6J5P driver in triode mode. Only 2 stages were neded. 6J5P biased by LED and loaded on gyrator. As the result got super linearity with all needed swing. Then in order to drive 4P1L in class A2 I used a cathode follower loaded on CCS. Try such solution and compare with 3 stages and interstage transformers. You can't get such results.
The only drawback I found, limited negative supply to drive CCS, so on some overload peaks it clipped (when we listened to Glinka's cannon shot), but it can be increased so the probability of such hard clipping due to lack of voltage would be very low.
With a finite plane (or a finite cylinder) you get edge effects. Unfortunately our rooms are not large enough for infinite valves. Careful design can minimise edge effects. It is better to have an imperfect implementation of an ideally perfect configuration than a perfect implementation of a configuration with known defects.
Plate current varies with the 3/2 power of the _sum_ of grid-cathode and plate-cathode voltage /mu , not a ratio.
Thanks,
Chris
With a sufficiently high (ideally, infinite) anode load the current stays constant. Then the voltage ratio becomes mu. This is how triodes work in the ideal case. This is standard valve theory.
If driving into a short circuit, then an ideal triode has a 3/2 law as you say. Nobody uses them like this, except for the bottom half of a cascode - which is why cascode is best used for small signals only.
If driving into a short circuit, then an ideal triode has a 3/2 law as you say. Nobody uses them like this, except for the bottom half of a cascode - which is why cascode is best used for small signals only.
External feedback loops, usually done with resistor and capacitors and including the output transformer have significant time delays. Making for lots of fun with phase delay mismatches and such. But the internal feedback in a triode operates at something like the speed of light, which would mean a time delay around 50 picoseconds. No phase delay issues with audio there!
At best, the NFB of a triode operates on the order of the transit time, i.e., if the plate voltage suddenly changes, it doesn't "know" until the next batch of electrons arrive to affect its voltage. But in practical cases, lead inductance and plate capacitance put further limits on this. These effects are on the order of nanoseconds.
After all, it's hardly an unconditionally stable phenomenon: why should a high frequency triode amplifier require neutralization?
Rattling off poorly formed intuitions like "speed of light" and "phase delay mismatches" are not useful to the construction of a stable feedback loop. To be completely specific, a "phase delay" is somewhat redundant, since, of course a delay always causes a phase shift, and vice versa. I guess the point is to emphasize the effect of a delay on the loop phase margin, a common basis for evaluation.
Phase shift and time delay are two aspects of the same phenomenon, but a constant phase shift (with respect to frequency) is not the same thing as a constant time delay (with respect to frequency). An integrator has constant 90 degree phase shift (a time delay inversely proportional to frequency), while an ideal transmission line has constant time delay (phase shift proportional to frequency). Over certain ranges, a single pole filter can be viewed as a time delay, but in general, it is not a very good phase shift OR delay.
Since an audio amplifier's bandwidth is small compared to the (pure) propagation delay of the circuit (wires acting as transmission lines, totaling perhaps a few nanoseconds in a large amplifier), it is safe to ignore, so the loop is best modeled with poles and zeroes. This approach does not lend itself to time-domain analyses as time delay, but does find use in the abstract domain (analyzing the transfer function, solving for poles with the appropriate frequencies) and frequency domain (most often as a Bode plot, working backwards from the Barkhausen criterion and desired phase margin).
Whether the feedback goes through more or less circuitry is simply a matter of having a properly compensated loop. If the gain is high and the "delay" is high, the loop gain will necessarily be retarded significantly at high frequencies (on the order of f = 1/delay). If the gain is moderate or the "delay" is short, much feedback can be applied, resulting in a low distortion, high bandwidth, low phase-shift amplifier stage.
Tim
After all, it's hardly an unconditionally stable phenomenon: why should a high frequency triode amplifier require neutralization?
Rattling off poorly formed intuitions like "speed of light" and "phase delay mismatches" are not useful to the construction of a stable feedback loop. To be completely specific, a "phase delay" is somewhat redundant, since, of course a delay always causes a phase shift, and vice versa. I guess the point is to emphasize the effect of a delay on the loop phase margin, a common basis for evaluation.
Phase shift and time delay are two aspects of the same phenomenon, but a constant phase shift (with respect to frequency) is not the same thing as a constant time delay (with respect to frequency). An integrator has constant 90 degree phase shift (a time delay inversely proportional to frequency), while an ideal transmission line has constant time delay (phase shift proportional to frequency). Over certain ranges, a single pole filter can be viewed as a time delay, but in general, it is not a very good phase shift OR delay.
Since an audio amplifier's bandwidth is small compared to the (pure) propagation delay of the circuit (wires acting as transmission lines, totaling perhaps a few nanoseconds in a large amplifier), it is safe to ignore, so the loop is best modeled with poles and zeroes. This approach does not lend itself to time-domain analyses as time delay, but does find use in the abstract domain (analyzing the transfer function, solving for poles with the appropriate frequencies) and frequency domain (most often as a Bode plot, working backwards from the Barkhausen criterion and desired phase margin).
Whether the feedback goes through more or less circuitry is simply a matter of having a properly compensated loop. If the gain is high and the "delay" is high, the loop gain will necessarily be retarded significantly at high frequencies (on the order of f = 1/delay). If the gain is moderate or the "delay" is short, much feedback can be applied, resulting in a low distortion, high bandwidth, low phase-shift amplifier stage.
Tim
I was thinking (maybe not correctly
) of the propagation of electromagnetic fields here. Which IIRC travels at the speed of light. But the charged electrons moving slower than the speed of light may block that. Duh! It's been a while since I studied this stuff way back in college, I graduated BSEE in 1978, I'll have to find my books...
Looks like I got my phase shift and time delay mixed up, anyway, the internal feedback you get between the input grid and the cathode is very fast (nanoseconds, maybe less). The external feedback loops, resistors and caps, poles and zeros, will impose shifts that can make the error cancellation a little off, and if bad enough, it can make the amp oscillate. Neutralization between grid and plate, fun, but I was thinking (maybe I went off topic
here) of cathode followers, which have gains of less than 1, and shouldn't be able to oscillate (at least to a first order approximation).
Last edited:
Cathode followers make excellent oscillators, just put a cap from grid to ground! This is where grid stoppers really count. The guys using high frequency transistors, and PHEMTs and the like, can get into trouble very easily since the figure of merit (transconductance / capacitance) is orders of magnitude better than tubes. Those things like to sing at 30GHz, they might not even see it! The traditional solution? Put a small resistor in series with the base or gate (usually 10-47 ohms).
At least one Tektronix plugin used a 6AK5 cathode follower on the input, with a 27 ohm "stopper" resistor between attenuator and grid.
Tim
This is where grid stoppers really count. The guys using high frequency transistors, and PHEMTs and the like, can get into trouble very easily since the figure of merit (transconductance / capacitance) is orders of magnitude better than tubes. Those things like to sing at 30GHz, they might not even see it! The traditional solution? Put a small resistor in series with the base or gate (usually 10-47 ohms).At least one Tektronix plugin used a 6AK5 cathode follower on the input, with a 27 ohm "stopper" resistor between attenuator and grid.
Tim
The data sheet value (rarely given for 6L6GC anyway) and the grid leak resistor, apply to static leakage.
If Thrower's distortion mechanism really does apply to these (it is a suspicion , rather than a claim.... hard data is not presumed) we will expect the grid emission to vary with Vag1, and instantaneous temperature - giving dynamic current peaks large enough to cause distortion.
What d'y mean I didn't exist? I'm not that young, you know.
ac heat on 300Bs was the norm 10 years ago, but nowadays most builders have confirmed the serious improvement available with properly implemented dc.
With ac, you are not only running 1.3A of 50/60Hz current through the same conductors as the 1mA to 35mA music signal, but the current-noise derived from the mains as well. This is direct coupling of noise!
The filament transformer secondary is directly connected to the cathode of the Triode, so any long wiring from filament to transformer is also an antenna looking for yet more corruption to add to your amplifier.
A hum pot gives an rough approximation to nulling the 50/60Hz currents, but as Dmitry Nizhegorodov showed (see the reference given by Chris earlier in the thread) the 100/120Hz second harmonics ride straight in with little attenuation. Further, with the 300B's substantial 5V rms heating voltage, the region around the cathode is also subject to an unwanted addition of in-band alternating electric field: i.e., a noise source.
These undesirable influences give rise to measurable levels of 50/60 and 100/120Hz sidebands in the music output - and a sound robbed the sound of delicacy and immediacy.
Raw rectified dc, where ampere-level rectifier-recovery pulses are circulating in the supply leads and capacitors (connected directly to the filament), sounds even worse than ac - but that's not saying much.
But there is no longer need to accept such a crude degradation in any DIY design. Properly implemented dc filament heating removes the wholly unnecessary 50/60 and 100/120Hz components, present high insertion-loss to noise frequencies all the way up to VHF, and regulates the heating without corrupting the music-signal (even when the music signal is 60dB or less below the heat current). The improvement in the sound is entirely obvious, as many have already witnessed here on our Tubes/Valves forum.
Grid emission is part of the grid leakage (gas being a much larger contributor). If grid emission is high, that resistance value has to be lower. See pp 315-316 in Valve Amplifiers 4th ed. It's made worse by temperature (Richardson-Dushman)- note that a unipotential cathode will operate at 1100-1200k, whereas a filamentary cathode will operate closer to 2000K, thus having much higher heating. This correlates with the higher grid currents characteristic of DHTs.
We tried DC and AC and preferred AC. AC can be done correctly or incorrectly. Ditto DC. But with DC, you can't get away from systematically uneven emission along the cathode. Unipotential has a huge advantage here.
Our experiments were done in the late 1970s, so I doubt you were selling your DHT heater supplies then.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.