Connecting dual triodes in parallel.

[G8HQP Dave]

If two Class A valves combine to drive an AC load then it doesn't matter whether they are PP or parallel. In both cases they each see twice the load resistance. Class B is a bit more complicated, because it may be that on average only one valve is driving the load so each valve sees the load as it is.

Part of the confusion might arise from a problem of technical English. 'Twice the load' could mean 'twice the load resistance' (i.e. x2) or 'a load which is twice as hard to drive' (i.e. x 0.5). Hence clarity is good.

[Diabolical Artificer]

Thanks for the clarification both and I'll endevour to be more precise Dave.

Andy.

[kalee20]

Originally Posted by Radio Wrangler

One good use where paralleled valves (or transistors) can be helpful is in reducing the required source impedance for optimum noise performance - the tape head preamp mentioned earlier.

Can you expand a bit on this, RW?

I'm rather ignorant and inexperienced where noise is concerned, I'll admit to this straight away! Granted, paralleling transistors can help optimise a low-level stage for the signal source, by effectively reducing base resistance.

But, transistors behave as a current noise and a voltage noise source, so have a 'sweet spot' where the combined effect of the two is minimised. Valves have no equivalent current noise source, referenced to the input, so I'm struggling to see how parallelling-up more than one helps the situation.

In fact, putting valves in parallel gives rise to the possibility of VHF oscillations (as has been mentioned) to to unintended interaction of parasitics, mandating (as good practice) grid stopper resistors for each valve. These themselves will act as a small voltage noise source, which would be absent for the single valve (no grid stopper necessary), so two valves could actually be worse!

Can anyone spot the flaws in my thinking?

[whiskas]

I paralleled the triodes in ECC40's to make a singled ended headphone amp some years ago.
(I think the transformer primary was 5K.)
It worked very well and sounded great!

[VT FUSE]

As do 6j6 b7g vhf double triodes .
Abundant and very versatile,actually become well behaved at audio.

Mike

[G8HQP Dave]

Originally Posted by kalee20

Valves have no equivalent current noise source, referenced to the input,

Grid current is a source of noise, although this will normally be irrelevant for audio unless grid leak bias is used. Even then it may be small.

[Herald1360]

Half the internal Ra, twice the mutual conductance, half the RA = same gain but half the internal noise power. I think. Comments?

[Diabolical Artificer]

Morgan Jones says this in Valve Amplifier's 3rd Edition - " Valves produce noise because Ia is made up of individual electrons that shower the anode, and also because electrons leave the cathode with random velocities to join the space charge... " He goes on to say cathode chemistry has a part to play in noise reduction. It would also follow that construction has some effect as regards noise. Construction is mentioned in the EF86 blurb.

Re paralleling valves .. " If we parallel `n' input devices in a low noise amplifier, the incoherent noise signal begins to cancel, but the wanted signal p remains at constant level, resulting in an improvement in signal to noise ratio of n dB. This technique is feasible for semiconductors where it is possible to make 1000 matched paralleled transistors on a single chip (LM394, MAT-01, etc.), but we are lucky to find a pair of matched triodes in one envelope, let alone more than that" That however contradict's reduction of noise in paralleled valves. Heater cathode leakage can be a source of noise.

There's an article here - http://www.tubecad.com/articles_2002...t_2/page2.html that deals with noise reduction and also some info here on paralleled triodes - http://www.dmitrynizh.com/parallel-triodes.htm .

I haven't been able to do any proper testing so far, so can't say if paralleled triodes are less noisy and also can't vouch for any other benefits. But I've built the LTP as in the earlier attached schematic and this drives the six EL34 OP stage very cleanly and effortlessly. It takes only 3v RMS ish to get a power OP of over 120w. This presents a problem in that the IP stage needs to have very low gain. I'm not using the paralleled 12AU7 IP stage but something different, will post a schematic tomorrow.

One thing I have noticed is that the anode DC voltage are not balanced, one is 10v higher than the other, however AC balance is excellent, only out by 0.01v! Anode resistors are matched to 1%.

I've had no trouble with oscillation with the LTP, but I did have with the IP stage, until I changed a few components and layout, I'm not following the schematic 100% though.

Andy.

[Radio Wrangler]

Originally Posted by kalee20

Can you expand a bit on this, RW?

I'm rather ignorant and inexperienced where noise is concerned, I'll admit to this straight away! Granted, paralleling transistors can help optimise a low-level stage for the signal source, by effectively reducing base resistance.

Andy's covered quite a bit of this already. In valves and in transistors there are numerous different noise mechanisms. The temperature of thermionic emission at the cathode, shot noise due to discrete charges on a finite number of electrons. It is usual to lump all these sources together as a fictitious noise generator at the input of the device. Usually a voltage noise source and a current noise source. The current noise is rather small in a valve, but it's not zero. There is shot noise in the grid current and there is also feedback of the anode noise via the Miller capacitance.

If my valve has very low current noise and significant voltage noise, I can put my input signal through a step-up transformer of some sort and increase the voltage driving the valve. At the same time the current is scaled down. The amplifier gain can this be reduced because of the greater signal voltage presented to the grid. However, the higher source impedance presented to the valve makes any noise current at the grid more significant. Up to a point, the transformer improves the noisiness of the stage so long as the voltage noise is dominant. Once I get to a certain level of transformer ratio, the current noise starts to take over and noisiness starts to get worse again.

Resistors act as sources of noise power density availability.

The word density is important, because the power of a noise signal is spread across a range of frequency. It's difficult to talk of an absolute number of watts, because the bandwidth is uncertain, it depends on what is being perceived when the measurement is made. For a good resistor (and there are bad ones) the noise power density is kT watts per Hertz. T is the temperature of the resistor in Kelvin. k is the Maxwell-Boltzmann constant. To find the power in a chosen amount of bandwidth, just multiply by the bandwidth in Hz. I like to talk of it in density terms though because it acts as a reminder that there may be more Hertz worth of noise than just the bandwidth the measurement was done in.

Once we've multiplied by B we have a power level (Watts) and we can convert that to voltage by Watts =(Volts squared)/Resistance

So if you don't put B in at the Watts stage, you get the volts-per-root-Hertz bit that confuses people.

So this power from the resistor... where does it go if the resistor isn't connected to anything, or is shorted?

The answer is that it is an amount of power which is available - not necessarily flowing. The full available kTB watts only flows from the resistor it it is connected to a load which matches its own impedance. In transmission line terms, all other impedances reflect some back to the resistor. Shorts and opens reflect it all.

This mismatch is critically important, otherwise crazy things would happen. If I take a 1 megohm resistor, it makes kTB watts.

If I now split it into one million resistors in series, each one ohm, then shouldn't each sub resistor should also make kTB watts, and a million of them added will be a million times greater?

This way lies free energy. This way to see the perpetual motion machine. All resistors can be considered to be infinite strings of infinitesimal values, each producing kTB watts of noise. All resistors are therefore infinite sources of power and must explode!

It's the matter of matching to get the available power which keeps things sane. Each of those million 1 Ohm resistors sees the other 999,999 in series with it in addition to what the 1 megohm resistor was connected to. So only a tiny fraction of each sub-resistors kTB watts comes out, and the whole string adds up to kTB if it's connected to a 1 megohm load.

Insulation works in a similar fashion. Modern insulation looks kile many giga-Ohms, and also makes kTB watts available, but only into a load of the same huge impedance. Our circuitry is much lower in impedance in lost cases, so most of that noise is reflected at the insulator/circuitry boundary.

If the input impedance of an amplifier is far too high considering the source impedance of whatever is driving it, then that represents a lost opportunity, the source could drive a step up transformation to put more volts into the valve, and the valve's gain could be turned down by a corresponding amount. Thus noise considered at the valve's input level is not amplified as much, or you could look at the turned down gain meaning a lower anode resistor and in this the shot noise from the valve would turn into a lower noise voltage going into the next stage.

There are several ways of looking at it all and they all work and they all come down to the same thing. Different people come across it explained in different ways.

And yes, the use of wider, larger area devices does bring down the noisiness of a stage where the source Z is less than the optimum noise condition impedance. Parallel devices are just a way of creating wider, larger area devices. It works and it's measurable. Valves are a bit more extreme than transistors, but you'll find RF transistor data sheets with Smith charts of applied source impedance with contours of noise figure plotted on them. In one of the many ways in which the universe seems unfair, the same Smith charts can have contours of gain plotted on them and you see that rarely do the optimum source Z for noise and gain coincide.

You just can't win

But with a bit of care you can lose by not quite so much.

Hope this makes sense. I'm used to doing it in RF terms and that includes thermionic amplifiers like TWTs.

David

[boxdoctor]

Although mainly with reference to solid-state devices, some interesting observations here: https://www.americanradiohistory.com...ld-1963-09.pdf Page 447. Tony.

[barrymagrec]

Leevers Rich used two ECC82s in parallel in their push pull Bias / Erase circuit - worked a treat, only failure I ever came across was because of an open circuit heater on one of the valves.

[Hartley118]

I've always found it mightily impressive that Boltzmann in the 19th century conceived of 'his' constant k such that noise power equals kT per Hz bandwidth. And that was decades before we had any general conception of electrical noise and bandwidth.

Statistical thermodynamics is a remarkable subject!

Martin

[Radio Wrangler]

Remember that kT Watts/Hz is the noisiness of a perfect resistor.

Some resistors, like metal film ones, are very close to perfect. Others can, if there is a little DC around create 'excess noise' by mechanisms like flicker effects. The common thick-film surface mount resistors exhibit noise levels about 20dB worse than kT. Carbon composition resistors are truly 'orrible.

David

[kalee20]

David RW,

Thank you for the detailed essay on noise. Grid-current noise through anode voltage noise fed back via Cga is something I hadn't thought of!

Something fundamental bothers me - granted that a 1kΩ resistor generates noise power according to kTB, as you say, if I connect such a resistor (a good quality, metal film type) at room temperature to a 1kΩ resistive load, I can actually extract this power. I could do this by connecting to another metal-film, 1kΩ resistor which I have cooled till it's mighty cold (so it's comparatively noiseless). I help myself to the 'free' power, the source resistor therefore gets cooler (as it is losing energy), the load resistor warms up (as it is absorbing energy). With the load getting warm, it starts to generate its own noise, and the source resistor getting cooler generates less, until the two resistors are at the same temperature and the system is in equilibrium with the noise voltages, on average, equal.

But, suppose my 'source' resistor is a cheapo carbon composition type, with excess noise? The high-quality metal film load will continue to absorb the excess noise and get a bit warmer, and the source resistor will get cooler until the noise voltages average out equal again. And this will result in one resistor being permanently warmer than the other, which violates thermodynamics as I see it!

Going back to the double-triode, if I have something like an ECC83, with ra = 62kΩ, μ = 100, and I load it with 100kΩ, then I can calculate the noise voltage due to the load resistor, knowing bandwidth and temperature. This will be attenuated by itself being loaded by the valve's ra - but how do I calculate the noise voltage due to this? What temperature is associated with the ra, for instance?

I'm not taking this off-topic, but I'm trying to keep my feet on the ground by working with actual numbers - and then hoping to convince myself either way that doubling-up triodes actually DOES improve SNR at audio frequencies, without cheating and using transformers to get back to the original impedances!

[ms660]

There's some stuff on noise in here:

https://open.bu.edu/bitstream/handle...pdf?sequence=1

Lawrence.

[Radio Wrangler]

If we have two identical and equal resistors connected together, each is sourcing an equal amount of power into the other. We have equilibrium.

If I put one of those resistors in a hotter place, it sources more power into its twin than the twin does into it. We have an energy flow from hotter to cooler. thermodynamics is OK with this and Flanders and Swann don't have to re-write their song.

If I put one of the resistors in a cooler place, it's the same situation.

OK, now if one of the resistors is a nasty one, producing excess noise power, we immediately run into the issue of conservation of energy. Where does that excess power come from? There shouldn't be any.

Excess noise is powered by DC bias currents or DC thermoelectric currents. THe normal Boltzmann thermal noise power is so low that DC powers wandering around the place are appreciably greater, and that only a small proportion needs to be converted to RF noise by virtue of fluctuations in the resistivity of the material.

Think of a carbon microphone. They are noisy sods and the noise is powered by the DC bias normally needed for their operation.

You were quite right to smell a rat. Without the understanding of the mechanisms of excess noise, the jigsaw has a piece missing.

So when the audiophools start getting all lyrical over carbon resistors, that choking/gurgling noise you can hear quietly in the distance is me.

Remember also that this is all modelling. Your triode has several noise generating mechanisms in play, some of them 100% in the anode-load loop. What standard noise analyses do, is they take the stage as a black box, ignore where different noise creators are and blame the whole damned lot on a fictitious source at the input then treat everything in the black box as noiseless.

Mostly, this works. and tends to track most of the change when you change the stage gain. But like any black box theory, you need your wits about you when you try to go into deeper detail.

It's when you look at the noise and gain circles on a Smith chart that you start to realise that the noise figure black box models your stage, but only under the test conditions, and that as you change the conditions, its changes in behaviour are not modelled.

David

[Diabolical Artificer]

Thanks for those links Tony and Lawrence. With David and Paul's posts there's a lot to absorb and think about.

Andy.

[G8HQP Dave]

Originally Posted by kalee20

Going back to the double-triode, if I have something like an ECC83, with ra = 62kΩ, μ = 100, and I load it with 100kΩ, then I can calculate the noise voltage due to the load resistor, knowing bandwidth and temperature. This will be attenuated by itself being loaded by the valve's ra - but how do I calculate the noise voltage due to this? What temperature is associated with the ra, for instance?

I think I read somewhere that you can do an estimate by assuming about half the absolute temperature of the cathode. I suspect that this is an estimate of the temperature of the cathode space charge.