That's not my experience with a triode-connected EF86, see http://www.diyaudio.com/forums/tubes-valves/173443-low-noise-low-microphony-low-hum-valves-3.html posts 19, 24 and 25.
I believe radiotron agrees with you too on noise and current.
Btw ... would anyone be interested in a LCR group buy? Lundahl has 0.45H and 1.8H chokes available ... we just need a very basic PCB (to make soldering easier and for a better look) and source a couple caps and few resistors.
they go for something like 110euro per pair, wound on mu-metal core and shielded. maybe we can get a special price for higher quantities.
Btw ... would anyone be interested in a LCR group buy? Lundahl has 0.45H and 1.8H chokes available ... we just need a very basic PCB (to make soldering easier and for a better look) and source a couple caps and few resistors.
they go for something like 110euro per pair, wound on mu-metal core and shielded. maybe we can get a special price for higher quantities.
Hi SY,
I hope we both agree that in a conventional pentode circuit, where the screen grid is connected to some sort of voltage source (such as a decoupled voltage divider), most of the signal current flows through the anode and a small part through the screen grid.
In Frank Blöhbaum's circuit, the screen grid voltage is fixed by a voltage source and the current follower (common-base or common-gate stage). Applying the substitution theorem, as the screen grid voltage is fixed like it is in the conventional pentode circuit and the anode voltage is also similar, the signal current should still be mainly flowing through the anode and only a small part through the screen grid. The only difference is that the screen grid current passes through the current follower and then joins the anode current again, which eliminates partition noise.
The fact that you can get similar transconductance and noise with a triode-connected pentode with a current follower (cascode) on top of the whole thing does not mean that the circuits internally operate the same. According to Blöhbaum, a practical advantage of his circuit over triode connection plus cascode is that the power rating of the current follower can be smaller, as it only conducts a small part of the current.
Best regards,
Marcel
It is important not to draw too many conclusions from only four samples. Some of your data is consistent with the theory, and the remainder may be due to experimental error or simply odd valve samples. Flicker noise is highly variable between samples.
Radiotron cannot be trusted for noise at audio frequencies. It only quantifies noise at radio frequencies, and was written before a lot of the low frequency work was carried out.
SY is right. It doesn't mater that more current flows in the anode than in the screen, all that matters is the transconductance, and that is almost entirely determined by the screen grid. Hence you can disconnect the anode and the circuit still functions basically the same. Thinking about it the other way around, Blohbaum's circuit is a hybrid cascode, but where a whole lot of current is wasted by allowing it to flow through the anode, in parallel with the transistor. I guess you could argue this is an advantage of the circuit, since you can have little dissipation in the transistor but maintain a good slew rate.
Last edited:
It's also fixed like it is in a cascode, right?
Let's take an example (which is the one I experimented on), a D3a and the "Bloehbaum" connection done via an NPN transistor. With Vg2 = 140V, Vg1 = -1.2V, for a D3a Ip = 20mA, rp = 120k, gm = 30mA/V . Ip/Ig2 ~ 3.5, so screen current will be slightly uder 6mA. I use a 10k plate load, so that gives a gain of 300. The base of the transistor, an MPSA42, is held at AC ground. hfe is ~100. Impedance looking into the emitter is 10k/100 = 100R in series with 1/gm, which is negligible. The gm at g2 is 1/3.5 gm to the plate (the current splitting ratio) = 30/3.5 = 8.6mA/V, so tube's gain to the transistor's emitter is 8.6 x 0.1 = 0.86V/V. The transistor has gm = 35*Ic ~210, so its voltage gain to the anode is 2100, whereas the D3a's voltage gain to the anode load is 300. The transistor is apparently doing all the voltage amplification, and if you disconnect the plate and run 6mA to the screen, that's exactly what you see- the circuit performance hardly changes.
The math is very straightforward. It doesn't take much voltage swing between base and emitter to get a large current swing. If we use the usual cascode gain approximation (gm1*Rp), gain with the 10k load would be 860, lower than the more involved calculation. We need to use the more involved equation because the collector impedance and bipolar transconductance are hugely higher than a triode used on top. My way of calculating is a bit simpler in this case, recognizing that the bipolar transistor only reacts to AC voltage between base (at AC ground) and emitter (driven by g2, which acts as the anode). That gain calculation is straight textbook.
Remember, the transistor's emitter is ONLY driven by the screen.
Chris, another way to look at it is that the voltage gain to the screen is slightly less than unity. So the transconductance splitting as far as the plate load is concerned in Frank's circuit is between the tube (gm ~ 30) and the transistor (gm ~210). It's not surprising then that the transistor dominates.
Great explanation SY! 
But, but, but.... The transconductance is split between the transconductance of the g1 to anode and that of g1 to g2. G1 to anode is obvious, but g1 to g2 cannot be made arbitrarily large by loading with a transistor. That's a second law violation.
I think the source of the error lies in the impedance looking into the transistor's emitter. It's pretty much independent of collector load and is very small. Remember that emitter and collector _currents_ are essentially the same.
Thanks,
Chris
I think that effective collector load divided by hfe is the correct expression. What would you use instead? If the looking-in emitter impedance were actually independent of collector load, what you'd have is a simple pentode circuit with a regulated screen.
Indeed, the g1-g2 transconductance can't be made arbitrarily large (it's about 80 for a D3a). But it IS actually pretty large, which is why the gain to g2 in this circuit is fairly small (slightly less than unity).
I must be missing something in SY's logic. You have two signal-controlled current sources feeding the same load: the pentode anode, and the g2 transistor collector. Both CCS will have high output impedance; the anode because that is what pentodes do, and the BJT because it has a high impedance feeding its emitter. The current, and gain, splits 3.5:1 in favour of the anode but the currents simply add in the load. To talk about the g2+BJT having more voltage gain than the pentode is meaningless when they are connected, and irrelevant when they are not?
Without his circuit in front of me, I'm unclear where Stuart gets some of his figures from. In particular, the input impedance at the BJT emitter. Where does the 10k/100 come from? 10k is the joint load, being fed by another signal current too. 100 is the BJT beta? To calculate the emitter impedance correction due to the collector load you surely have to take account of Early effect, not current gain? 10k/100 might be the effect on emitter impedance of a 10k resistor in the base circuit, not the collector circuit. You can't just insert current gain of a BJT into formulas meant for voltage gain of a valve! The dimensions will be correct, but the result will be wrong.
Without his circuit in front of me, I'm unclear where Stuart gets some of his figures from. In particular, the input impedance at the BJT emitter. Where does the 10k/100 come from? 10k is the joint load, being fed by another signal current too. 100 is the BJT beta? To calculate the emitter impedance correction due to the collector load you surely have to take account of Early effect, not current gain? 10k/100 might be the effect on emitter impedance of a 10k resistor in the base circuit, not the collector circuit. You can't just insert current gain of a BJT into formulas meant for voltage gain of a valve! The dimensions will be correct, but the result will be wrong.
Hi SY,
I don't deny that a conventional triode-connected pentode with a current follower on top behaves rather similarly to Blöhbaum's circuit. I'm just pointing out that the internal operation is different, as a large part of the current (DC and signal) does not flow through the current follower. That's all.
Best regards,
Marcel
What matters is the trend with current: white equivalent input noise voltage rather consistently drops with increasing current while 1/f noise first drops and then increases. These trends are consistent with the theory in Van der Ziel's book Noise. (Van der Ziel spent most of his life studying noise, both low and high frequency, so I assume he understood what he wrote about.)
That may be true for some of your samples, but in general flicker noise increases with current continuously (which is what the theory tells us).
Whose theory? Certainly not Van der Ziel's!
(See post 15 of http://www.diyaudio.com/forums/tubes-valves/173443-low-noise-low-microphony-low-hum-valves-2.html for a very short summary of what Van der Ziel writes about 1/f noise, or excess noise as he calls it.)
Last edited:
I have read his works. There was some revision to his theories as time went on, and a lot of extra work on low frequency noise wasn't carried out until the 1960s, after most of van der Ziel's books on valves. There are several 1/f mechanisms, some of which appear to be constant, but the main one (in oxide coated cathodes) increases with current.
You can get a noise minimum at a particular current, but it is due to flicker noise falling and shot noise rising, until they become equal.
For example, see:
J. J. Brophy, "Minimizing Flicker Effect in Low Level Vacuum Tube Amplifiers," Review of Scientific Instruments, vol. 32, pp. 204-204, 1961.
Bare in mind that your comparative measurements were made with a large resistance in the grid circuit, so grid-current noise could have been a significant contributor to error. This is not a significant source of noise in phono stages, however, since the source impedance is small.
Last edited:
Thanks for the reference to the article, I'll look it up.
What I stated in post 36 of this thread is incomplete. I only saw excess noise going down and then up on the Amperex devices, not on the second Trigon, there it only went up. Still, as white noise and 1/f noise are both far from negligible at audio frequencies, you have to determine your noise optimum experimentally.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.