Valve Distortion 1: element shape

Introduction

I will assume that you have some idea about how thermionic valves/tubes work. If not, there are plenty of good descriptions on the web. A very clear, but rather long, account is given by John Harper. He describes the factors which determine the characteristics of a valve.

Roughly speaking, the amplification factor (mu) of a triode valve is set by the ratio of the cathode-grid spacing and the cathode-anode spacing. A large ratio (i.e. a grid very close to the cathode) gives a high value for mu. The mutual conductance (gm) and anode impedance (Ra) are then set by the shape of the valve. A small cathode, with the anode a long way away, gives low gm and high Ra. A large flat cathode with a nearby flat anode gives high gm and low Ra. There are lots of combinations of shapes and sizes which give a particular set of characteristics at the normal operating point. This is why valves of the same nominal type from different manufacturers can look so different internally.

All this assumes that the control grid has small wires, spaced uniformly with a pitch much greater than their size, and at a distance from the cathode which is greater than the grid pitch. The valve should then behave something like the standard model says, with a 3/2 power law. Given a sufficiently high anode load, local feedback will then ensure linearity of amplification i.e. low distortion.

In reality things are not that simple. As the grid voltage goes negative and the valve approaches cut-off, parts of the cathode directly under a grid wire get cut off first; this is the "island effect". The result is a reduction in mu and gm at low anode currents, and so nonlinearity. This effect can be minimised by using very fine grid wires, as in a frame grid, provided that the grid is then not placed too near the cathode. Unfortunately, frame grid valves almost always have their grid very near the cathode in order to get high amplification so the potential linearity advantage is thrown away.

Valve Shape

Another cause of nonlinearity at low currents is non-uniformity of the valve construction. The simple theory assumes a simple electric potential distribution, which can come from either a cylindrical construction or a planar construction. In either case the electric field at the grid due to the anode is the same for all points of the grid. If this is not the case then different sections of the valve will have different gain, and the higher gain sections will reach cut-off first as the valve current is reduced. So for low distortion a valve should have self-similar construction i.e. the cathode, grid and anode should all be the same shape. Two special cases of this are cylindrical, where all three electrode cross-sections are concentric circles, and planar, where all three electrodes lie in parallel planes.

Early valves often had a cylindrical construction, then the cross-section went from a circle to a square. It turns out that the electric field from a square anode is almost the same as that from a cylindrical anode as the corners get filled in by the field, so this is just as good.

By the 1950's, many small signal valves were being made with a slab construction: a small cylinder for the cathode and a large slab for the anode, with the grid somewhere in between. This is about the worst possible shape for a valve for the purpose of linearity! Different parts of the grid will see very different electric fields, and so there will be considerable island effect near cut-off. As a result, people complain that recent (i.e. 50's onwards!) valve designs are inferior to earlier ones when used for zero overall feedback audio. However, it must be remembered that for some purposes (e.g. RF mixers) non-linearity is essential.

Most valves have anodes which are not completely uniform. Many have holes (to help with element positioning during assembly?), or some form of patterning such as 'ladders' (to increase rigidity?). These will cause local variations in electric field. A bump coming out (i.e. away from the cathode) will probably have less effect than a bump going in, which may be why most anode patterns are 'bump out'. The effect depends on how much the anode-cathode distance is changed by the bump, and also what effect it has on the electric field at the point occupied by the grid. The further away the anode is (i.e. high mu) the less effect a bump or other variation will have, as the field will smooth out at higher distances. This suggests that low mu valves will be more sensitive to the details of anode shape than otherwise similar high mu valves. Could this be part of the reason* why the ECC81/12AT7 is generally more highly thought-of by audiophiles than its cousin the ECC82/12AU7?

* Another issue could be grid design. The ECC82 is derived from the EC90/6C4 power triode. It is likely to have a more robust grid that can cope with significant grid current. As well as being further away from the cathode than in the ECC81, the ECC82 grid may have a coarser pitch and thicker wire.

An example - twin triodes

Valve internal layout
Photo of three low mu valves: Ei 6FQ7, Mullard ECC82, Mazda ECC804
From left to right: Ei 6FQ7, Mullard ECC82 and Mazda ECC804.
Type mu gm at Ia, Va, VgCa-gIhtrVa maxIa maxPa max
6CG7, 6FQ7 20 2.6mA/V 250V, 9.0mA, -8.0V4.0pF0.6A300V20mA3.5W each, 5W total
ECC804, 6/30L2 18 3.4mA/V 200V, 10.0mA, -7.7V 2.5pF0.3A250V2.0W each, 2.5W total

Two valves with quite similar characteristics are the 6CG7/6FQ7 and 12AU7 (ECC82), yet many people report that the 6CG7 is much more linear and so better suited to high-end audio. It has been speculated that the difference is due to size, and miniaturisation somehow introduces distortion. My suggestion is that the difference is due to shape, not size. The 6CG7 has a box anode (although the cross-section is not quite square), while the 12AU7 has various shapes of slab anode. Both have round (more accurately, cylindrical) cathodes.

Another similar valve is the ECC804 (6GA8, 6/30L2). This has the same pinout as a 6CG7 but is not the same valve. Like the ECC82, the ECC804 was intended for general purpose use in TV sets (timebase oscillator etc.). It too has a slab anode, but the cathode is a small slab rather than a cylinder. It is possible that this could give it better linearity than the 12AU7, although probably not as good as the 6CG7. The cross-sections are shown on the right (not to scale!) and below that is a photo of a sample of the valves.

I have recently noticed that dealers are selling 'Mazda 6CG7', and people seem to like them. These look suspiciously like ECC804, and have small anodes. The ECC804 could be a substitute for 6CG7 in some low anode current circuits but it is definitely not an equivalent - see the comparison table. It seems likely that Mazda, or someone else, just remarked ECC804 as 6CG7. These 'Mazda 6CG7' are said to be rare (and hence expensive), yet ECC804 (6/30L2) are quite common (and cheap) in Europe. I suspect it is the mark that is rare, not the valve inside. If these 'Mazda 6CG7' sound as good as people say, then perhaps that confirms my theory about element shape. It would be interesting to know how much heater current they draw#, and what etch codes are on these valves: 1576/1876 or 2195? I suggest that an audiophile who likes the sound of 'Mazda 6CG7' should try an ECC804 instead, and perhaps save a lot of money! These often turn up on eBay in assorted boxes of 300mA TV valves. The triode in some Mazda TV audio or frame output triode-pentodes (e.g. 30PL13 = PCL800) is identical to half an ECC804.

Brand Type AnodeGain2nd @ 5VVinIP23rd @ 5VVinIP34th5th
Brimar 13D3 long 6.1-60.2dB (0.098%)841V-53.3dB (0.22%)17.7V-81dB-92dB
Mullard (Gf1)ECC82long3.8-53.0dB (0.22%)589V-56.5dB (0.15%)34.1V-95dB-104dB
RCA (USA) 12AU7Along 4.2-52.2dB (0.25%)486V-58.6dB (0.12%)34.8V-103dB-103dB
RFT ECC82 short4.5-49.0dB (0.35%)314V-58.2dB (0.12%)31.8V-97dB-104dB
Tungsram (Hungary)ECC82short4.6-47.1dB (0.44%)247V-57.1dB (0.14%)29.2V-104dB-101dB
RCA (Brazil)6FQ7long 6.5-44.0dB (0.63%)122V-59.3dB (0.11%)23.4V-85dB-103dB
Sylvania 6FQ7 long 6.2-43.7dB (0.65%)124V-58.8dB (0.11%)23.9V-89dB-105dB
Mazda ECC804short5.5-41.2dB (0.87%)105V-59.0dB (0.11%)27.2V-99dB-105dB
SMS ECC804short5.5-41.2dB (0.87%)105V-58.1dB (0.12%)25.8V-96dB-102dB
Ei 6FQ7 long 6.4-39.9dB (1.01%)77V-57.2dB (0.14%)21.1V-99dB-103dB
Test amplifier circuit
Test amplifier circuit

I have done some tests comparing ECC82, ECC804 and 6CG7 distortion. I included the 13D3 because some people use this as an ECC82 substitute - it had very low 2nd but much larger higher harmonics. The circuit& was a simple cathode-coupled amplifier, which tends to cancel out even-order products. Output was 5V rms at 1kHz. I have shown the second and third-harmonic input intercept voltages - the hypothetical input for which the harmonic is equal to the fundamental. It is interesting to note that, with the exception of 13D3, the valves generally favoured by audiophiles have significantly higher second-harmonic and slightly higher third-harmonic in this circuit.

These results that the ECC804 and 6CG7 give similar performance. The ECC804 draws a little more anode current than 6CG7, and has slightly lower gain. For those 6CG7 circuits which run at less than 2W anode dissipation a cheaper$ substitute is thus available. A minor advantage is that the ECC804 always seems to have an inter-section screen connected to pin 9, whereas most '6CG7' have no screen so are really 6FQ7.

& The circuit is not optimised for any particular valve type. The two 56K resistors were intended to be 560K; I misread the colour code! Output A was used to measure the fundamental (1kHz) and 2nd and 3rd harmonics, and Output B (after the notch filter) was used for 4th and 5th harmonics. This made best use of the limited analyser resolution.

# If a 'Mazda 6CG7' is in fact an unmodified ECC804 then it will be fine in most 6.3V parallel heater circuits, but could cause problems if naively substituted into other arrangements. For example, if a 12.6V paired parallel heater supply is used, then the ECC804 will get too hot and the paired valve could be too cold and may stop operating. This is because the 300mA ECC804 will have 8-9V on its heater, while its 600mA partner will only have 3-4V.

$ I bought a box of twenty 'SMS ECC804' on eBay for 50p each; they look like Mazda manufacture but some are a bit wonky and two had internal shorts so I suspect that they are rebranded factory rejects. Compare this with the $25-40 being charged for 'Mazda 6CG7' by US dealers! By the way, every 'Mullard' ECC804 I have seen has come from a Mazda/Brimar factory; I do not believe Mullard actually made any of this valve.

Another example - 12B4

The 12B4 triode is regarded as having low distortion. This may be partly due to its low mu (about 6), but its construction is essentially two back-to-back planar triodes. The anode plate facing the grid/cathode is flat, and the cathode is a flat slab. Thus it meets my criterion for low distortion - a self-similar structure.


Notepad made
Valid HTML 4.0 Transitional

Back to audio home, Brimar factory codes, the effect of grid current

updated 27 April 2011: comment on heater arrangements