Skip to content

For many years, I have been designing and building vacuum tube audio equipment. The site I have is built using MS FrontPage that is no longer supported so over the coming days, weeks and months, I will be transferring content here. You are encouraged to make full use of the information presented providing what you do is for your personal use.
These days, my attention is focussed on restoring and documenting vintage test equipment at

58dB Hybrid Phono Stage

Several years, I built a quick and dirty phono stage and it had promise but I set it to one side until I came to sell my AR turntable that is fitted with a Sumiko Blue Point Special Cartridge. I decided to sell the two together. The BPS is a high output moving coil design however, I felt that higher gain than the 44dB or so that my typical design (two stage with passive RIAA) provides, was desirable. This phono stage actually started out in 44dB form. Now, the BPS has a mixed reputation and this was a challenge to me. I had realised that it probably does not tolerate capacitance well being a high impedance moving coil and thought that perhaps a cascode input phono stage might help. And so I modified the design to incorporate a fet / triode cascode in the front end, using a 6GM8 battery type tube for the triode. I have used the 6GM8 in a couple of other projects and it sounds quite good altogether. The new design worked, all of a sudden the BPS went from having a congested and rather lifeless sound to being immensely open, articulate and above all, musical. I had worked the design out further in a re-build of a commercial integrated amp (link) and continued to be pleased. And so come the time to sell the AR, I cleaned this one up and the new owner reports experiencing much listening pleasure with the combination. It does hail from my tube voltage regulation days, now I would use current regulators. Here is the schematic:

58dB Hybrid Phono Stage Schematic

I haven’t bothered to show the raw B+ gubbins, anyone who has the ability to use the schematic offered can sort that out easily. Note the +6.3V is from a LT1085 regulator that also feeds the heaters. It must be superbly clean. Oh, the stain at the top right is from  chocolate ice cream (every detail matters).

And here is the thing:

58dB Hybrid Phono Stage 1

It is still on its wooden baseboard, I added proper terminals, a copper ground plane under the phono stages and a power switch.

58dB Hybrid Phono Stage 2

Here’s the AR, with Audiocraft unipivot arm and Merrill platter, the phono stage and my 2A3 line stage are behind:

AR, Merrill & Audiocraft 1

AR, Merrill & Audiocraft 2

It’s a damn fine sounding set up and caused me to do some maintenance on my Benz – Sota – planar triode set up which was not comparing well! It turned out that the arm ground was flaky, phew! It is a good idea to treat audio connectors with Deoxit once in a while. Even if there is not audible noise, lousy connections sound just that, lousy.


Class A 807/EL34 PP with transformer gain stage.

I have pretty much done all I intend to do with tube audio and decided to do a “left overs special” using some good parts I have on on hand including a very “bling” commercial chassis from an amp that had been dropped by Fedex. In particular, I have a pair of Sowter 9045 microphone transformers that are reputed to be excellent. They can provide a gain of 1:5+5 thus creating accurate push-pull phases. Not surprisingly, they can’t swing enough voltage to drive the output stage directly, even using easy to drive 807s and EL34s. However, the additional gain and swing needed is not great and I decided to use 6BX7s for a common cathode push-pull drive stage. As with the transformers, the 6BX7 is known for its audio quality. I also had a suitable One-Electron power transformer on hand, all that was missing was output transformers and I bought a pair of good-quality-for-the-money James 6225 HS units. At $275 for the pair, they do represent good value. The choice of tubes came about because fellow vintage oscilloscope custodian Volker Klocke gave me four good 807s in return for repairing a Tektronix HV transformer for him.

Here’s the bare chassis with the original sub-chassis:

BLAMP Bare Chassis & Old Sub Chassis

It is basically a polished stainless steel cake pan which is not good for tubes, something mechanically dead is required so I placed strips of double sided foam thus:

BLAMP Ready for new Sub Chassis

And then having removed the backing tape replaced the original sub-chassis with a glass epoxy copper clad panel:

BLAMP New Sub Chassis in place

This is also secured by a multiplicity of screws that hold the tube socket bling rims in place. The resulting sandwich arrangement significantly damps the liveliness of the cake tin. I also made a cocobolo plinth for the transformers. (I took the precaution of wearing latex gloves and a face mask while working this wood.):

BLAMP with Transformer Deck

Very gaudy hence “Blamp” as in bling amp. Still it ends up looking quite attractive, here it is with a nice set of EL34s showing off the Sowter microphone transformers:

807 : EL34 Class A PP 2

Here it is with the 807s, using 5 pin to octal adaptors that do make them rather excessively tall:

807 : EL34 Class A PP 3

And so to the design narrative:

If you have read my paper that I published earlier on this blog, you will know that I am a proponent of cascode (depletion mode) mosfet current regulators, largely because of the near total isolation of the audio path from the power supply that they allow when used thoughtfully (Class A only of course). I refer to them as regulators since they are two terminal and can be used either as sinks or sources. They are available as kits from K&K Audio.

Starting at the front end, the first challenge was to come up with a neat way to drive the microphone transformers. I knew a cathode follower was the way to go and because there are not enough tube locations on the chassis, I decided to use 5840 sub-miniature pentodes. The use of a cascode current regulator in the cathode circuit was a given. The novel aspect is that I know that the way to make a pentode really perform as a cathode follower is to dc couple the screen grid to the cathode because this provides constant transconductance. Taking my lead from the tube instrumentation world, I decided to use NE-2 neons. Any noise that they introduce will be non measurable, at least with any equipment in my lab. For the plate current I wanted, 6mA, the screen needs to be at 100V or more (relative to the cathode of course) otherwise the plate voltage will rise towards the limit which I think (senility doesn’t help while writing this stuff) is 160V. So I used two NE-2s in series, fed by a J509 3mA current diode that having a high impedance, leaves the screen grid completely free to jive with the cathode. I also placed a 47V zener across the diode to prevent the 50V limit for the diode being reached in case of some kind of trouble. In normal operation, the zener is effectively out of the circuit. For good measure – and no better reason than that – I bypassed the two neons with a 1nF cap. After some screwing around with a signal generator, dB meter and oscilloscope, I determined that a 10μF capacitor would be correct for driving the paralleled primary of the transformer. I just happened to having some of those good sounding Russian pink paper in vodka caps, or is it paper in pink vodka? I used a low voltage PSU kit from K&K audio to provide a -20V return for the cathode current regulators. The cascode current regulators also come from K&K Audio.

So, now we have 20dB of gain (less 0.6 dB due to the cathode follower and transformer insertion loss) and do not need a great deal more so I chose a 6BX7 for the push-pull driver that has a mu of 10 providing a further 20dB (referred to the push pull output, not one phase). Since the 6BX7 is current fed and the load is due to the 220k grid resistor only, the measured gain actually is 20dB. I chose a bias of 20mA for the 6BX7 sections that with a common cathode resistor of 270Ω results in a plate voltage of around 170V for a new tube. The plates are fed by current regulators from a 230V choke input supply that, further filtered, also feeds the cathode follower. I think there is a real sonic benefit in using such a powerful drive stage even with easy to drive output tubes, I say more about this aspect at the end of this post.

The output stage is also somewhat unusual. Because I didn’t have room for lots of conventional high current B+ filtering, I chose to use, once again, current regulation feeding into a 40μF motor-run type bypass cap. The effective impedance of the cascode regulator is so high that there is no measurable ripple at this location, again using the fairly comprehensive test equipment that I have. This arrangement effectively limits operation to Class A and that too is intentional. To assist in good current sharing between the output tubes, I stood them on separate 1k resistors and this means that I could ac cathode couple the tubes which is a real sonic win for a Class A push pull output stage. The grid bias is variable and fed by a 75V stabilised point that also provides the positive heater supply bias that ensures hum free operation of the drive and output stages. The grid resistors are 220k in value and the output stage is coupled to the drive stage by 100nF Russian paper in oil caps, this time green though so perhaps we are onto Absinthe rather than Vodka? Since the cathode current is fixed, varying the grid bias voltage will set the plate voltages and with the EL34s, I set them at +350V relative to the cathodes. On replacing the EL34s with the 807s, the voltage dropped between 20 and 30V. As the tubes age, the plate voltage will rise to compensate for falling emission. With the tubes I have, the size of the cathode standoff resistors provides current equalisation better than 2mA. Yes, it’s a power hungry topology however the whole amp consumes less than 200W and I am happy with that. Once again, screwing around with a signal generator e.t.c. I arrived at a cathode coupling capacitor value of 4μF. Enter pink ruskies again, see the picture of the “works” below. The James 6225HS output transformer has a plate to plate reflected impedance of 5k however, I found that things got a LOT better when I loaded the 4Ω tap with 8Ω thus reflecting 10k plate to plate. To get the output Z below 1Ω, I added some negative feedback, introduced at the ground end of the microphone transformer primary. With the EL34s the output stage gain is -14.7dB and with 5.4dB negative feedback, the resulting loop gain is 19.4dB, perfect for my system requirements. With the 807s, the output stage gain falls to -16.7dB and the negative feedback to 4.4dB giving a loop gain of 18.3dB, still more than sufficient for me. I did experience a problem with unstable plate voltage on one channel. This turned out to be due to one of the green caps being leaky. I’ve got loads of them so finding a good replacement was not a problem, they do sound good……

Power with the EL34s is 1w at 0.05% THD, 12w at 2% THD, with the 807, the power is a bit less at 10W. Remember, it is Class A. Some commercial push pull tube amps (especially Chinese) that are claimed to be Class A are not.

The power supply uses a One-Electron BFT-1 transformer run well with in its rating. The highest voltage taps are used for the +500V supply for the output stages at 240mA while the lower voltage taps are used for a 100mA choke input 230V supply for the drive and input stages. I used the spare 5V and 6.3V windings in series into a voltage doubler and then a LT1085 configured as a 150mA current regulator to power the 5840 heaters that are connected in series. A 2A slo-blo fuse protects the whole thing.

Oh, the front end has a nice 20k switched attenuator. It doesn’t have to be 20k, I design for relatively low interface impedances because interconnect cables like that and it keeps noise down. This amp is silent using my 95dB/W/m Tannoys.

If you’ve made it this far, here is the schematic:807 : EL34 Transformer Coupled Class A PP


Talking of keeping noise down, I made a soldered tinplate box to shield the power switch that is bang in the middle of the front ends! I also connected the power switch using shielded wire.

Here’s a picture of the “works”:

807 : EL34 Class A PP 1

Please take the time to click on it to see it properly!

As for the sound, I think it is a good as anything I have done and this being no time for false modesty, I design and build excellent sounding equipment. In particular, this amp has what I can only describe as kick, transients are extremely clear and since a lot of music is transient, this may explain to some extent the good sound. It resolves very deeply, and has marvelous clarity, it just gets out of the way. I think this may be, at least in part,  due to the use of a power drive stage combined with easy to drive output tubes. Whilst on the subject of output tubes, good EL34s are excellent. I have built equipment using the much vaunted 845, 300B, 50, 2A3 and 45. Frankly the EL34 doesn’t fall far short if at all, and is MUCH less expensive. I haven’t listened to the 807s enough to form a good impression yet. I have them playing as I am writing this and at the least they make music that way I want to hear it, clear and in the room, not stuck to the speakers. Reproducing music well in a domestic environment is difficult and 100s of watts hinder more than help! Driving my Tannoys this amp is powerful both with the EL34s and the 807s, if loud is what you want. I can’t hear deeply into the music if it is loud.

I then got around to repairing my HP 3580A audio spectrum analyser and since the amp can accommodate both EL34s and 807s, I put el34s in one channel and 807s in the other then ran a spectrum comparison at 1W and 8W. The analyser can store one scan digitally then it can be restored once a second scan has been taken, allowing direct comparison.

Here’s the comparison at 1W:

807 : EL34 PP Spectrum 1W

The EL34 is showing 2nd only at -61dB. The THD was 0.055%

The 807 is showing 2nd -60, 3rd -62, 4th nil, 5th -78. The THD was 0.11%

Here is the comparison at 8W:

807 : EL34 PP Spectrum 8W

EL34: 2nd -50, 3rd -75, 4th -77, 5th -65, 6th -71, 7th -70, 8th -74, 9th -72, 10th -74, 11th -73. The THD was 0.256%

807: 2nd -49, 3rd -45, 4th -70, 5th -51, 6th -81, 7th -52, 8th -80, 9th -56, 10th -81, 11th -62. The THD was 0.72%

I “should” have repeated the tests after swapping the output tubes between the channels, however, I am not going to! The results are shown purely for the sake of interest though they do seem to reveal that the EL34 actually is cleaner than the 807 at low levels and much cleaner at high levels.

The HP 3580A can also perform a log frequency sweep over the range 20Hz to 43kHz. Here are the sweeps using the EL34s at 10W upper and 1W lower:

EL34 PP Log Frequency Sweep, 20Hz to 43kHz, 10W Upper, 1W Lower

It reveals about a 2dB rise overall and notably, the HF performance does not drop off at the higher power. The frequency scaling is logarithmic starting at 20Hz, then 43, 98, 200, the same repeating for the next decades ending at 43kHz. So the dead flat response is roughly over the range 40Hz to 10kHz then rising gently to 43kHz.

Main System

The Monument Nov 2015
A wag dubbed this system “The Monument” which amuses me. This system is the result of about 15 years experimentation starting with KT88 parallel PP amps driving Energy C8s. At this point, I am satisfied so I maintain that I am thankfully ineligible for being accused of the apparently heinous sin of being being an audiophile! (Most EE’s would say that “I can’t hear the difference” meaning that if THEY cannot hear a difference, then we must be fooling ourselves. I say that clearly you cannot hear the difference and that explains the truly horrible sounding early solid state amps and most cheap commercial gear.)

This is my main system with the 845 amp which is described on this blog.
The alternative amp on the table above the 845 amp, is a 50 design that can also accept 300Bs the design principles of which are described in the “The Audio Signal Path” post on this site. The UPS looking thing is a line conditioner that uses the secondaries of the UPS transformer as a common mode choke. It also provides a common power switch for the amps e.t.c.

(The secondary system still uses the C8s with a variety of small amps and a Meridian 508 CD player.)

The heart of the system is the line stage. Many years ago, I decided that with SE amps, it is a good idea to take the low bass via a separate amp so I designed a line stage that has 70Hz high-pass filters in both main channels and two bass channels that feed the plate amp in the woofer. (It is the polished aluminum unit on top.) The most important part of a music reproduction system is the speakers. The C8s are actually quite good and I spent several years in speaker hell, trying to better them. Many years ago, on the advice of my friend John Dahlman, I bought a pair of re-coned Tannoy 3836 drivers. These are professional units that were used in the Met production of Phantom of The Opera, Tannoy re-coned them they were re-sold. Finally, I took his advice and tried them in open baffles. Err, why oh why didn’t I do this sooner? Suffice to say that I am done with experiments. The baffles are large enough to work superbly with the 70Hz crossover. To me the bass crossover is seamless and this is particularly tricky to accomplish so I either got lucky or I know what I am doing. I claim that it is a combination of both.

CD / SACD player is an unmodified Sony SCD-777ES
There is also a DV-563A DVD player that I have modified with a transformer coupled output stage. This is a good-sounding unit that can play some (few) discs that the Sony unit will not digest. Full details at:

The turntable which is hidden behind the left speaker is a Sota Star Sapphire with a Benz L2 cartridge and AudioQuest arm.
The phono stage on the shelf above the turntable is my MKII WE 416 planar triode based design, full details at:

The line stage is a 2A3 based design, full details on the line stage at: The switched attenuator now has motor-drive and a remote.

Here is the back of one of the Tannoys:
Richard Tannoy 3836 Open Baffle
The straps on the bottom secure 40lb bags of concrete mix (in plastic shopping bags to prevent leakage) to weigh the speakers down. Sand would be better since the concrete will set over time and lose it’s ability to dampen vibrations.
The dimensions are based on the golden ratio with respect to the vertical location of the driver and the ratio of the width of each of the wings taken from the edge of the driver.
The 1.3KHz crossover is a factory recommended design I believe, that consists of a first order 4 microfarad / 6.8 Ohm high pass and a first order 2mH / 30 microfarad low pass filter and a 30 microfarad / 10 Ohm Zobel across the main driver. Here is a closer view of the crossover (Zobel not yet fitted):
Tannoy 3836 and crossover

The Audio Signal Path; Minimising Power Supply Interaction


The comment “the sound of the rectifier” is always an alert to me. I assert that if the rectifier is audible, then there is signal current flowing in the least linear part of the circuit, the power supply and in particular, the rectifier! I have been on an explorative and experimental path looking to minimise the degree to which signal currents flow in the power supply.  The first place to look is the output stage since the current requirements of that stage are the most demanding (usually). In considering the output stage, the first issue that comes up is the class of operation. In this respect, Class A has an advantage and so in this discussion we are limiting ourselves to Class A operation. However, simply limiting our designs to Class A is not enough. The problem is that the current demands of even a Class A stage will change according to the signal level. For example, consider the data sheet case of a 45 triode operating single-ended in Class A at 250 V with a bias current of 34 mA, delivering 1.6 W into 3.9 kΩ:

Placing a 3.9 kΩ load line on the curves passing through 250 V at 34 mA, we can note that the plate currents at the ends of the line as defined by a grid swing of ±49.5 V are at 67.5 mA and 6 mA, the mean of which is 36.75 mA.

We can see that this is elevated 2.75 mA compared to the quiescent current of 34 mA; this is due to the departure of the tube from perfect linearity and is sometimes referred to as rectification; that is to say, the current in one direction is greater than the current in the opposite direction; a diode simply takes this difference in “back-to-front” current to the limit.  When reproducing music we rarely if ever, encounter steady state conditions (except for silence) and so this example is somewhat unrealistic however, it does serve to illustrate that even in Class A, the signal will place demands on the power supply. So, given that no amplifying device is perfectly linear, what can we do to minimise audible sonic interaction of the signal with the power supply?

The traditional approach is to use a combination of chokes and capacitors acting as an energy reservoir to satisfy the current demands resulting from the signal acting upon a less than perfectly linear output tube. That experimenters and observers often note “the sound of the rectifier” indicates that such traditional measures are not adequate in the quest for sonic fidelity. This article shows how the signal may be isolated from the power supply using a simple and practical technique.


First, let us review the signal current loops in a cathode biased, single ended output stage, referring to figure 1:

The first loop is the input loop (Dashed);

What is not commonly stated is that an amplifying device responds to a voltage difference in this case, the difference between the grid voltage and the cathode voltage. If the grid voltage falls relative to the cathode voltage, the current passed by the tube falls and vice-versa. With a common cathode stage, the cathode is held at signal ground potential by the cathode bypass capacitor while the grid ‘sees’ the input signal. The tube grid input impedance is not infinite, in fact RDH4 has this to say with regard to grid input impedance: “When a valve is used at low audio frequencies, it is sometimes assumed that the grid input impedance is infinite. In most cases however, this assumption leads to serious error, and careful attention is desirable to both its static and dynamic impedances.” For the purpose of our discussion, the serious error referred to is that the input signal has no current associated with it such that the input loop is usually not considered carefully; the point being that if a cathode bypass capacitor is used this current flows from the source to the grid, to the cathode and via the cathode bypass capacitor back to the source and vice versa, alternating with the signal: the cathode bypass capacitor is in the input signal path.

The second case is the output loop (Chain Dashed);

Similarly to the input case, the output voltage develops across two terminals of the amplifying device, in this case the plate and the cathode. In the case of a transformer coupled load, the voltage acting across the load reflected to the primary causes current to flow from the plate to the output transformer primary, to the PSU bypass capacitor to ground and via the cathode bypass capacitor to the cathode and vice versa, alternating with the signal current. Note that there are two capacitors in series with the output signal current! (We will address this issue later in this article.)

The third case is the power supply (PSU) loop (Dotted);

In the case of a perfect Class A amplifier, there would be no power supply signal loop, the power supply would merely be feeding the circuit with energy to balance the energy dissipated in the load. Furthermore, if the PSU bypass capacitor were perfect (zero impedance at all frequencies) then the power supply would not be in the signal path however, there is no such thing as a perfect Class A amplifier or a perfect capacitor and so the power supply is in the signal path. To make matters worse, the energy supplied by both the power supply and PSU bypass capacitor has to pass via the cathode bypass capacitor.

We have identified two signal path issues that audibly impact the sonic results of the circuit: One; Signal current flowing in the power supply and Two; The cathode bypass capacitor that appears in all three loops; the input loop, the output loop and the power supply loop.

Taking the power supply first; the traditional approach is to simply increase the energy storage thereby increasing the period of time averaging. The obvious issue here is that the equivalent series resistance (ESR) of a capacitor is not flat with respect to frequency and even more, the impedance of the power supply as a whole is far from flat with respect to frequency. This means that the impedance of the power supply signal path is not flat and so the power delivered into the load is not flat. Note, even if the measured performance appears flat, our ears are quite capable of distinguishing micro-effects that limit the potential musical beauty of an amplifier and that is, I assert, the point of this hobby so long as we are having fun. Worse still, many L-C or C-L-C power supplies exhibit a severe resonance (impedance peak) in the region 5 to 15 Hz. In fact, all power supplies are to some degree reactive (i.e. they resonate) and this has a disastrous impact on the sonic results that some people may actually like, however, it is not conducive to fidelity.

When I started designing tube amplifiers (around 12 years ago), I was aware of the power supply resonance issue and my approach was to employ tube based voltage-regulated power supplies and this is quite effective. The advantage of a regulator is that there is a substantial voltage drop across it that corresponds to available stored energy, much lessening the degree to which signal current flows in the rectifier. I then started using a simple mosfet series regulator with a reference voltage on the gate. (I also put a long RC time-constant on the gate to obtain slow turn-on.) This is highly effective, affording notably flat power source impedance in the 6 Ω region. What I noticed is that the sound improved over that of relatively sophisticated error compensating tube power supplies. However I felt I could do better. The problem is that the power supply signal loop is now a composite of the regulator and the bypass capacitor. The tube regulators I built have source impedances in the 1 to 4 Ω region. This means that up to 1000 Hz or so*, I was listening to the regulator while above 1000 Hz, the PSU bypass capacitor dominated the sonic results. “Listening to the regulator” is a common and valid objection to voltage-regulated power supplies. The series mosfet may help by eliminating the error amplifier. Even so, we are still left with the problem of the rectification effect of the amplifier tube causing current demands that are not adequately managed by the power supply in sonic terms.

*For a PSU source impedance of 4 Ω, the bypass capacitor would have to be 4000 µF to be effective down to 10 Hz. Apart from not being practical at tube power supply voltage levels, such a large capacitor (or bank of capacitors) is likely to sound gruesome.

Many of you may find the use of solid-state devices bothersome for various reasons. I assert that there are a few ways in which solid-state devices free us to explore topologies that, while not necessarily new, are not practical when using tubes only. In fact, used with due attention to detail, there is a particular solid-state device that permits fuller realisation of the inherent sonic beauty of vacuum tubes; The solution to the power supply signal interaction problem that I am proposing is to harness depletion mode mosfet technology to create truly effective isolation of the signal from the power supply.

I did a simple investigation of the performance of the two-terminal depletion mode mosfet cascode CCR, see figure 2. Note, I am using the initials CCR, viz; constant current regulator, since this device is equally effective as a source or a sink.

What I found was that using the Supertex DN2540 for M1 and M2, the measured DC current variation was less than 0.5 mA over a range of impressed DC voltage from 50 V to 200 V and the variation was immeasurable over a 100 V range. This is extraordinary performance for such a simple two-terminal circuit.

Some people report that for plate or cathode “load” applications, the IXYS 01N100D is superior sonically, I have not attempted such a comparison; I did repeat the DC current test using the 01N100D for M1 and M2 and the performance was not acceptable. I then replaced M1 with the DN2540 and found the resulting DC performance to be acceptable but not quite as solid as that obtainable when using the DN 2540 for both M1 and M2.

The AC test set-up used one section of a 6AS7 as a cathode follower, with the CCR (using the DN2540 for both M1 and M2) in the cathode circuit, see figure 3.

A test load of 500 Ω was inserted between the CCR and the negative rail. I drove the 6AS7 such that 20 VRMS (VIN) was measured at the cathode and then measured the voltage (VOUT) at the load. I ensured that Vin was constant for each test frequency. The AC impedance of the CCR is very sensitive to the value of gate stopper resistor; I chose to implement the design using the same value chosen by Walt Jung, 100 Ω1. (It is worth noting that I have not experienced oscillation of the CCR devices I have built using 100 Ω gate stopper resistors.) My results are very different than those of Walt Jung, probably due to a different test method and much higher current combined with a large input voltage; however, my set-up represented my application; at a test current of 65 mA, the cascode CCR exhibited an impedance greater than 1.5 MΩ over a range of 10 Hz to 40 kHz.

This circuit is quite remarkable and opens the door to new, extreme-performance tube audio implementations that were not possible before the advent of depletion mode power mosfets. My current (ahem) preferred choice of devices is the DN2540 for M1 and the 01N100D for M2. Note that the 01N100D is rated for 1000 V, whereas the DN2540 is rated for 400 V; Since the VDS of M1 is limited by the VGS of M2, the voltage withstanding capacity of the cascode CCR is determined by the voltage rating of M2.



Given that a CCR has an impedance of 1.5 MΩ or greater over and beyond the audio spectrum, placing a CCR in series with the plate supply will result in almost perfectly constant current; that is, the power supply will not be modulated by the signal and will simply feed energy to the circuit to balance the energy dissipated in the load; this is the ideal we are striving for. To justify this statement, let us take a look at some numbers:

We will continue with our 45 triode operating single-ended in Class A at 250 V with a bias current of 34 mA having a plate resistance 1.75 kΩ and a reflected load of 3.9 kΩ.

We first must establish the resistance of the stage at the lowest frequency of interest, say 5 Hz. Let us also say we have an output transformer having primary inductance of 40 H.

The impedance at 5 Hz is given by 2 x π x f x L = 2 x π x 5 x 40 = 1256 Ω.

In parallel with this we have the reflected load, in our 45 case, 3.9 kΩ resulting in a composite load of 1.256 kΩ in parallel with 3.9 kΩ = 0.95 kΩ.

In series with this we have the plate resistance, again for our 45 case; 1.75 kΩ and the cathode bypass capacitor, let’s say for the purpose of discussion 2 Ω, that is negligible and so we will ignore it. The resulting stage resistance is 0.95 kΩ + 1.75 kΩ = 2.7 kΩ.

Let the plate signal current be IP. If there were no PSU bypass capacitor, the voltage due to the signal at the PSU end of the output transformer would be IP x 2.7 kΩ (neglecting the Thevenin effect of the CCR and PSU impedance since 1.5 MΩ >>> 2.7 kΩ).

The PSU source impedance and the CCR impedance form a potential divider, the top leg being the CCR impedance and the bottom leg being the assumed PS source impedance of 2 Ω.

For a potential divider, VOUT = VIN x RBOTTOM ÷ (RTOP + RBOTTOM) and so we have the voltage at the CCR / PSU node is;

VCCR/PSU = IP x 2.7e3 x 2 ÷ (1.5e6 + 2) which simplifies to IP x 3.6e-3.

To calculate the power supply rejection ratio, we simply divide the voltage at the CCR / PSU node by the voltage at the PSU end of the output transformer giving:

PSRR = IP x 3.6e-3 ÷ IP x 2.7e3 = 1.333e-6 or -117.5 dB, an excellent result!

Note; If you are following closely, you will have noticed that the stage resistance cancels and that the PSRR may also be calculated by dividing the PSU source impedance by the CCR impedance. If we are pedantic, we will include the PSU source resistance in the denominator but 1.5 MΩ >>>> 2 Ω so we will neglect it. This gives the power supply rejection ratio as 2 ÷ 1.5e6 = 1.333e-6 or -117.5 dB, the same figure we had calculated from first principles and so it can be seen that the power supply rejection ratio is dominated by the impedance of the CCR. This analysis illustrates and confirms the effectiveness of CCR as a powerful tool in our quest to isolate the signal from the PSU.

In fact, 117 dB is close to or beyond the dynamic range of our hearing so it is possible that we have achieved our ideal, any signal current modulation of the power supply will now be inaudible and the power supply is now doing the only thing that we need it to do, that is supplying the energy dissipated by heating of the circuit elements and in the load. Note, even if the power supply impedance rises to 100 Ω (quite possible if resonance is present), the power supply rejection ratio is dominated by the CCR and will still be -83 dB.

Now that we have effectively removed the power supply loop from the signal output loop, the output signal loop completion is simplified to the PSU bypass capacitor; what value should this capacitor be?

The signal output or load loop is comprised of the stage resistance (already calculated as 2.7 kΩ), the PSU bypass capacitor to ground and then via the cathode bypass capacitor to the cathode. (It should be noted that the PSU capacitor and the cathode bypass capacitor appear in series the consequences of which we examine later.)

The minimum value of the bypass capacitor is given by C = reciprocal (2 x π x f x R).

Again, our lowest frequency of interest of is 5Hz and so we have

Reciprocal (2 x π x 5 x 2.7e3) = 11.78 say 12 µF.

This value is available in high quality film devices; we no longer need to horribly compromise our lovingly created signal path with an electrolytic capacitor. I usually parallel this with an excellent coupling capacitor such as an Auricap of not more than 1/100 value; in this case 0.1 µF would be a good choice.

Now, let us take a look at the cathode bypass capacitor:

To bias the 45 according to our example, the cathode bias resistor will be 50 V / 45 mA = 1.11 kΩ. Thevenin tells us that this resistance appears in parallel with the cathode resistance, given by 1/Gm. In our case, the data sheet gives Gm as 2.175 mA/V; taking the reciprocal, we get the cathode resistance as 0.46 kΩ. 0.46 kΩ in parallel with 1.11 kΩ gives 0.325 kΩ. Again, taking the lowest frequency of interest to be 5Hz, the value of the cathode bypass capacitor is reciprocal (2 x π x 5 x 325) = 100 µF.

As we have already noted, this capacitor appears in series with the PSU bypass capacitor thus reducing the PSU bypass capacitance to reciprocal (1/12 + 1/100) = 10.7 µF. Not too bad and increasing the PSU bypass capacitor to 13.6 µF (say 15) will serve our design intentions. Please note however, that the cathode bypass capacitor not only appears in the signal input loop but also in the signal output loop in series with the PSU bypass capacitor (again, refer to 1). No capacitor has flat ESR with frequency and the larger the capacitor, the worse the ESR behaviour and connecting two large capacitors in series makes matters worse. This is why I am a staunch advocate of grid bias* (or, in the case of voltage gain and drive stages, LED bias). The sonic impact of cathode bypass capacitors is far from theoretical; it is very audible.

(*Note, I distinguish grid bias from fixed bias: Grid bias is simply applying a bias voltage to the grid. Combining grid bias with fixing the plate voltage results in fixed bias. As I describe later, I set the grid voltage, set the current using a CCR and let the plate voltage float.)

The PSU bypass capacitor will increase the power supply rejection ratio since the instantaneous signal current will be sourced by the bypass capacitor. The effect of the capacitor is to drastically reduce the voltage due to the signal at the PSU end of the output transformer; this will now be given by the plate signal current multiplied by 2.7 kΩ in parallel with the ESR of the capacitor, say 2 Ω; since 2.7 kΩ  >>> 2, Ω let us say 2 Ω. To be pedantic, we should also take account of the CCR resistance and the power supply source impedance that is, 1.5 MΩ + 2 Ω. Again, since 1.5 MΩ  >>>> 2 Ω, the effect is negligible and so we will use 2 Ω giving the voltage at the PSU end of the output transformer as IP x 2. This can be compared with the result previously obtained of IP x 2.7 kΩ.

As before, the PSU source impedance and the CCR impedance form a potential divider, the top leg being the CCR and the bottom leg being the assumed PS source impedance of 2 Ω.

Again using the potential divider equation, the voltage at the CCR / PSU node will now be (IP x 2) x 2 ÷ (1.5e6 + 2) which simplifies to IP x 2.667-6.

Again, to obtain the power supply rejection ratio we have:

PSRR = IP x 2.667-6 ÷ IP x 2 = 5.334-7 or -125.5dB.


Figure 5A shows a parallel feed CCR topology that is a Western Electric concept I came across in Lynn Olson’s presentation “Loop Distortion”2. In this case the CCR is acting as a gyrator while the output transformer is parallel fed by a capacitor that closes the output loop directly to the cathode. The classical version of this circuit uses a choke in place of the CCR. The choke provides a large degree of isolation from the PSU; this may be one reason why parallel feed amplifiers have a strong following. However, a choke may not provide the same level of signal isolation and noise attenuation from the PSU that a CCR can, both at the LF end due to limited reactance and at the HF end due to shunt capacitance. It is worth noting that when this WE topology was conceived, the CCR would have been tube based that, in requiring a much elevated B+ voltage together with added parts cost, would have had limited commercial viability. Now, with the advent of depletion-mode mosfets, this topology is not only attractive, but practical too. My implementation uses direct feed, see figure 5B; I am considering making it switchable on the fly between parallel feed and direct feed for the sake of comparison and fun! Practically speaking the switching is a simple matter of moving the CCR connection between ends of the output transformer primary perhaps using wetted relays.

You may observe that both the above topologies look like the Ultrapath topology trademarked by Jack Elliano3, the difference is that Ultrapath has no specific provision to isolate the signal from the power supply; it relies on a conventional PSU bypass capacitor. It is also worth noting that a cathode feedback capacitor is not shown in either case and may not be necessary; it is possible however, that audible levels of noise may be injected at the cathode. Fixing this gets us back to having a big capacitor in the input signal loop and I would advocate the use of grid bias with a coupling transformer to avoid this pitfall.


Figure 6 shows the topology of my recent direct-feed single ended 50-300B project. It is extremely simple, particularly with respect to the signal paths since the power supply has been removed from the signal path by the CCRs. CCR power supply technology may be combined with grid bias for current gain / output stages or LED bias for voltage gain / drives stages to address both the signal path compromises identified at the beginning of this article; accordingly this design uses LED bias at the drive stage and grid bias at the output stage. Moreover, a CCR power supply combined with grid bias allows us to set the quiescent current of the tube and we can now vary the grid voltage to obtain the best sonic results without changing the quiescent current. The plate voltage will float to the necessary value for the tube to conduct the set current at the set grid voltage.

I took advantage of the CCR / grid bias arrangement to make changing the operating conditions between a 50 or a 300B a simple matter of turning a switch. I adjusted the grid bias voltage values for the 50 and 300B such that the distortion residual (monitored on a ‘scope using a HP339a analyser) was cleanly 2ND harmonic across the full dynamic range of the amplifier. The filament current draws of the 50 and the 300B are very nearly equal so I used current regulation for the filament also, thereby avoiding the need to change the filament supply setting when changing tube types. I intend to write this design up fully in a future article so I won’t say more other than it sounds excellent, more open and able to resolve far more deeply than my previous designs and without false modesty, than takes something special!


Figure 7 shows two push-pull topologies that use CCR technology. Referring to figure 7A, the CCR is located in the plate circuit so there is no cathode coupling; the circuit exhibits current balance but not current and voltage balance. Taking an extreme case, if we apply the input to one grid only, the result will be gross distortion. We can do much better if we place the CCR in the cathode circuit, see figure 7B; the circuit is now a long tailed pair that is completely self balancing, fully compensating for any imbalance in the input signals.


The first approach I came up with is the “Current Balanced Push-Pull” amplifier, published in the May 2006 edition of Audioxpress. When I created the current balanced amplifier, I was primarily exploring the potential of inexpensive toroidal power transformers as output transformers. The result is one of the best sounding amplifiers I have built or have heard. This implementation originally used a simple (single) mosfet type CCR located in the cathode of each output triode to assure near perfect DC current balance so as to avoid DC saturation of the toroidal output transformer; the cathodes are coupled together with a capacitor to obtain perfect signal balance in the output tubes and thus true Class A; I dubbed this arrangement a “Power Long Tailed Pair”. Because the push-pull signal path loop impedance is higher than the cathode resistance of the triodes, (refer to the chain-dashed loop on figure7) the capacitor does not need to be in the range >100 µF, in fact in this case it is a 10 µF film type. The effectiveness of the CCR technology is revealed by the fact the output and drive stages share a common supply; I and my friend, bassist John Dahlman found it difficult to distinguish if a PSU bypass capacitor was present or not, suggesting that the true Class A push-pull behaviour made possible using CCR technology frees us from PSU bypass capacitor quality concerns. I recently updated this design, replacing the simple mosfet CCRs with cascode CCRs, see figure 8.


Many years ago, John Camille developed a single-ended 845 amplifier. He courageously shunt regulated the B+ supply using another 845 controlled by the signal via an op. amp. I am clear that well implemented shunt voltage regulation is superior to series voltage regulation; for one thing, the shunt may be fed using a CCR thereby isolating the shunt signal loop from the PSU. Another characteristic of the shunt regulator is that it can both feed and absorb current, perfect for output stages. However John’s design and until recently, other implementations I have seen are asymmetrical; the shunt circuit does not complement the signal circuit.

I am beginning to wonder if true Class A push-pull may actually be superior to single ended, gasp! I suspect that many so-called Class A push-pull amplifiers actually cross into Class AB during more of the program material then is generally recognised with the result that the not only is the PSU modulated audibly but also, the output resistance of the amplifier is continuously switching. In past days, 20 dB of negative feedback would most likely have swamped this issue (and the music) but with ‘modern’ zero feedback triode designs; I suggest that this issue might horribly compromise the sonic results. Whether this is valid or not, true Class A push-pull opens up the possibility of a self-shunt regulating amplifier.

The point I am making about a true Class A push-pull amplifier is that given that it is fed via a CCR, (preferably in the cathodes to assure near perfect signal current and voltage balance) then it is self-shunt regulating, refer back to figure 7B. And so for me, the next level is to take what I have started with the 6AS7G output and drive stages and apply this self shunt-regulation, true Class A push-pull technology to all the stages. In fact, Kevin Carter of K&K Audio is doing this and his GM70 PP design is the best sounding amplifier I have yet heard. Furthermore, Kevin has nice PCBs for building CCRs available.


After a year or so, my 6AS7G PP amplifier failed due to overheating of the DN2540 mosfets used in the CCRs: The current levels went far out of specification resulting in the amp going from extremely musical to horrible! Part of the problem was due to inadequate ventilation. However, I had been considering replacing the simple CCRs with cascode units for a while and this was the opportunity! Referring back to figure 2, you will see that in this topology, the VDS of the critical control mosfet M1 is limited by the VGS of the cascode M2: Clearly this topology has the further advantage of causing very low heat dissipation of the critical control mosfet, M1. If the voltage drop is large, I take some of the thermal load off the CCR by putting an appropriate value of power resistor in series with it. In this case, I spread the 6AS7G cathode voltage 50/50 between the CCRs and resistors. I mount the two mosfets back to back on a single heatsink, however I now use larger heatsinks than I think are necessary and ensure that they are in free air. Moreover, given that the mosfets are back-to-back I use two mica insulators and no thermal grease on M1 to reduce the heat soak from M2 and a single mica washer with thermal grease on M2. So far, this is working well, the dc current in the output tube push-pull sections has remained stable and the current balance between the push-pull sections remains better than 0.5 mA.


Other possible applications of the depletion mosfet two-terminal CCR technology are:

1/ Norton dc signal level shifter;

Sometimes it is desirable to shift the DC level of the signal, for example when DC coupling. Figure 9 shows a Norton level shifter that simply superimposes a clean DC current on the signal; the DC current flows through a resistor interposed between the plate of the first stage and the grid of the second stage causing a DC potential difference that can be arranged to match the potential difference between the plate and the grid. Since the CCR has an impedance of 1.5 MΩ or higher, it adds little load to the previous stage. The operating conditions of the circuit need to be extremely stable (as with all DC coupled circuits) so PSU voltage regulation is a good idea. Also, the value of the interposed resistor R will interact with the Miller capacitance of the second stage. All in all, I think this device is mostly of curiosity value, especially given that interstage transformer coupling sounds so excellent. Still, one of you may find a use for it!

2/ Precision variable ultra low-noise DC voltage reference;

I still use voltage regulation in some areas, for example, my SE 50-300B design uses mosfet series regulation ahead of the CCRs. I wanted a simple way to adjust the output voltages so zener diodes were not the best choice for the gate voltage reference. Since the DN2540 cascode CCR is so extremely stable, I simply built an adjustable one (a pot for the SOT resistor in figure 2) feeding current through a thermally stable metal film resistor (having an adequate voltage rating) to obtain a precise, ultra clean reference voltage that does not drift around as things get warm, (unlike a zener).

3/ Tuning the load of a pentode or cascode to obtain the best sonic results and precisely define the driving impedance for a Lipshitz (or other) passive RIAA equalisation network;

This is a nice one and I almost hesitate to give it away, see figure 10.

I have found that a hybrid, jfet–triode cascode works and sounds extremely well for the front-end of a phono stage.

So far, I am not alone with this experience. However, the next issue is providing well-defined source impedance for the following Lipshitz (or other) passive RIAA equalisation. The trick is to use a CCR to feed the current to the plate of the cascode; it only needs 15 V to work. Then put a resistor in parallel with the CCR to set the load. The resistor value may be adjusted to provide the best harmonic characteristics from the stage without changing the plate voltage or B+ voltage required unless the resistor is so small that it starts to conduct appreciable DC current; in that case, a CCR is redundant. You can use a pot for the resistor while experimenting; when the best value has been found, replace it with a fixed resistor. Now, the plate load resistor value controls the cascode output impedance; the cascode plate resistance in my designs at least, will be around 3 MΩ so any variation in the cascode RP with age will have negligible impact on the accuracy of the equalisation since 3 MΩ is likely to be significantly greater than the value of the plate load resistor.

4/ Shunt regulator current feed;

If voltage regulation is desired – and it can be a superior way to implement ultra high-resolution, small-signal stages such as phono-stages – then arguably the best way to do it is shunt regulation. Figure 11 shows the topology of the stage specific shunt regulators I created for my WE 416C MKII phono-stage.

The shunt path is shown by the chain-dashed loop and again, the CCR isolates the shunt path completely from the PSU with sonic benefit. Since this is a no-compromise design, I used a negative supply to permit the shunt triode to be directly across the phono-stage supply with nothing in the cathode circuit. Note; in this case, the shunt regulator is in the signal path; a corollary to this is that an individual shunt regulator must be used for each stage to realise the resolution benefits of shunt regulation. I have found that the effect of bypass capacitors in my WE 416 phono-stage are inaudible to me (I actually can switch them in and out of circuit to compare), so I am listening to the regulators; this is the benefit of ensuring that a shunt regulator design is a match for the amplifier stage design, not a generic design. To reiterate, I suspect that the best shunt-regulated audio circuit is a long tailed pair having a CCR as the tail.


My intention in writing this is to illustrate the potential for the thoughtful application of solid-state devices to more fully realise the sonic beauty of vacuum tubes than was possible without such devices. It is noteworthy that Western Electric had envisaged (maybe originated) the use of a CCR in the PSU current feed to a parallel feed, single ended output stage and now, we can easily experiment with this concept.

I hope you find the ideas expressed in this article stimulating and applicable. I am a huge fan of experimentation; that is where the fun is so I encourage you to play!

Thank you for your attention.

Richard Sears, April 2010

1/ Walt Jung; Letter to Audioxpress, April 2009

2/ Loop Distortion; Lynn Olson,

3/ Ultrapath, Parallel Feed and Western Electric; Lynn Olson, Vacuum Tube Valley, Issue 16

Source for CCR kits; K&K Audio (Kevin Carter),

Source for Supertex mosfets; Mouser Electronics,

Source for IXYS mosfets; Arrow Electronics, http://www

845 Triode SE Amplifier.

I designed and built this frankly as an ego exercise, it is too heavy and puts out too much heat. Having said that, it sounds very good, maybe even excellent and the design definitely breaks the mold.

I am very glad I made it as a double decker. Having 1000v on the umbilical, it is a good idea to have the power supply and amp physically connected yet if I could not separate them, I could not move it. The transformers on the left and right are the interstage units while the box in the centre contains both the output transformers. So there.

Showing the amplifier “works”.

Showing the amplifier schematic. In brief, the first stage is a choke assisted mu-follower providing adequate and very linear gain together with low source impedance to drive the fixed bias 45 transformer coupled drive stage. The output stage is a fixed bias series fed arrangement having a 11k double C core transformer developed and wound by yours truly. The interstage transformer is by Bartolucci.

View of the power supply tubes.

Showing the power supply works. The high voltage and heater/filament transformers are custom units from Bartolluci. They were worth waiting for the shipping time being perfectly, perfectly to my specifications and, as I recall reasonably priced for what is the best of the best.

Showing the power supply schematic. As they say, “don’t try this at home”. You have been warned.

I wound my own output transformers using 2mil hypersil double C cores that I located in a surplus store in London. I wound two prototypes which worked quite well but I was not satisfied since in both cases there was a mild resonance present in the HF roll-off. Such resonances are due to leakage inductance combined with capacitance. I then hit on the idea of winding on all three limbs, this vastly improves the coupling to the core; this idea came to me from noting the super-wide passband of Vanderveen’s toroidal transformers, resulting from near perfect coupling to the core due to the winding occupying the full circumference of the core. This method also reduces distributed capacitance since the surface area of the juxtaposed primary and secondary sections are reduced. It would be prohibitively expensive to produce commercially however, I now obtained -3dB beyond 100kHz at full power (25W) with no resonances, not bad for a big transformer with an 11k primary. The interleaving insulation is a combination of teflon and craft paper. Teflon, to get the lowest capacitance and craft paper to create a stable surface for the next winding. The transformers have a long-narrow shape and I chose to pot them together in one can (from Bartolucci) using micro-crystalline wax.

I re-discovered the following Q&A dialogue with TC of Hong Kong, it may be of interest:
TC: I intend to build an over 40W SE amplifier based on the KR-T1610 (an audio tube made by KR audio). My preliminary concept, which is basically based on your design, will be: “Super Mu Follower (76 or 6P5G or 6C8G) – Direct Couple (if possible) – 45 driving tube (or 6W6 triode-wired or 2A3) – LL1677 Interstage Transformer (80ma primary, ratio 1:2) – KR-T1610 output tube – KR 1.25K single-ended transformer”
Richard: I am aware of this tube. It is attractive. It requires 140Vp-p to drive it, a bit less than the 845 at the operating point I am using. The 45 will do this, just. I think a 2A3 will be hard-pressed to develop this much swing. A 300B would have more headroom but sonically, a 50 or even a 10 could be better. When I rebuilt from the EL34 drive, I chose the 45 because I like the sound of this type and I could easily adapt a filament supply from the existing resources. [I made an error in my response: Yes, the T1610 requires 140Vp-p drive. However, the 845 requires almost 300Vp-p. When I commented on the 2A3 as a driver I was recalling my thought process on the 845, the 2A3 will drive the T1610 just fine.]
TC: In this connection, much grateful if you could clarify my queries as follows: (1) Can you elaborate why you change your Super Mu Follower from “76 – LL1168 – 76 – 680 ohm” to “27 – LL1168 – 1k ohm – 76 – 1.2k ohm”? What are the benefits ofsuchachange?
Richard: OK. A compromise, what I did was to change my mind about what compromise I wanted. Originally, I chose the 27 to be the input tube and the 76 to be the follower. I could see from the 76 curves that the 76 would bias nicely using the 680 Ohm DCR of the choke and so the 27 became the input tube by default. However, I had some hum from this tube. The heater current for the 27 is 1.75A and I do not have enough room to include a DC heater supply for these tubes. Additionally, I also wanted a little more gain; mu for the 27 is 9 while that of the 76 is 13.8. I already had DC heater supply for the 76 and so I decided to change the tubes around. The 680 Ohm DCR of the choke does not fit well with the characteristics of the 27 and so I re-jigged the operating points, increasing the bias resistance for both tubes and increasing the B+ level to what I felt resulted in a good operating compromise between voltage, current and dissipation.
TC: (2) Why don’t you consider a Direct Couple design of Super Mu Follower to the grid of 45 so as to omit the RC coupled capacitor 0.68 uf (2.2 uf in previous version)? What do you think about the DC design here?
Richard: Good question and I did in fact consider this. The reason I chose capacitor coupling was to avoid having to provide a ‘stand-off’ voltage for the 45 cathode. My design philosophy includes as few components in the cathode path to ground as possible. So I had a choice: DC coupling and a voltage stand-off for the 45 (that may be gas tubes, a mosfet or even a resistor bypassed by a capacitor. If I were rebuilding the amp I would consider gas tubes but as-is I do not have the room. The mosfet was not attractive to me; I wanted to keep this amp (but not all my amps) as tube-pure as I reasonably could. The bypassed resistance approach breaks all my rules because it requires that most elusive of all audio components, a transparent large-value capacitor. And so, the compromise for this aspect of the design was to use ac coupling and a small, high quality capacitor.
TC: (3) Have you considered using ultra-sonic AC (around 100k Hz, square wave) heating of filaments of DHT tubes (modified from an electronic transformer for powering halogen lamps at the Audio Asylum forum, e.g. )? How do you compare it to the DC current regulated one?
Richard: I general, I think it is a good idea to avoid RF circuits in audio equipment, especially square waves with the broad harmonics. Most of us have heard an improvement in sound when a good line- conditioner is installed. So why go to a lot of trouble to provide extremely clean B+ (and ac) supplies then introduce a noise generator? I have compared constant V and constant I supplies and I strongly prefer constant I. AC may be better still but then there is the hum issue. A couple of years back, I think some listening test were done at the European Triode Festival; if I recall correctly, the conclusion was that constant current was best.
I also would ask why do you need so much power? Even with 90dB speakers, my 845 is a monster, relegated to winter use to keep the room warm! I have learnt (along with many, many others) that the audio signal is complex and fragile so the best way is to use efficient speakers and do as little to the signal as possible (least number of stages, low power). My next move is into the sub-4W area with Tannoys, most likely bi-amped. I see 50s for the low end and perhaps 71As for the tweeters. I already have built several sub-4W amps and they all sound incredibly open and lucid. The 845 does match them but lower efficiency speakers mean that more compression takes place in the voice coil (most of the power is dissipated as DC resistance heating) and so the openness is not fully realized.
TC: For power tubes, I prefer the sound of 2A3 to 300B. I have heard a 250 amplifier, it is much better than a 300B one But 2A3 has lower plate resistance than the 45, so that’s why I want to try 2A3. 6W6 is a sleeper tube of low plate resistance when triode-wired. Very cheap. I don’t know the reason why it is so under-used.
Richard: I’ve not heard a T1610 but in order of what I have heard, my preference for driver or output is (good to better) 300B, 2A3, 45, 50/71A/845 (The last three are all different but offer to me comparable satisfaction.) To drive an 845, a plate resistance of several kOhms will not cause slew-limiting. The Rp of the 45 is around 1700 Ohms which is plenty low to drive the transformer shunt C and the Miller capacitance of the 845. I cannot speak for the T1610. On the 6W6 there is a ‘snob’ factor around multi-grid tubes, one that I am not immune to (see comments on pentode follower below).
TC: other approach that I will consider is Allen Kimmel’s mu-stage (but choke assisted). The implementation will be “D3A – LL1168 – 6P5G – a resistor”. Very similar to your Super Mu Follwer, except that the top tube is a pentode of very high transconductance (~30 mA/V) running at pentode mode. One more difference is that at the cathode of the D3A there is a also a resister (Rk) connecting to the ground. I am considering using a tube current sink to replace the Rk. Do you have any comment on this choke-assisted mu-stage design?
Richard: Electrically, the pentode follower is superior – I try to avoid multi-grid tubes in filamentary triode designs because if I am going to lengths to provide clean, DC current filament supplies, I don’t want to “corrupt” the design any more than I can. Having said that; multi-grid tubes can sound excellent. So I use them in projects where I am willing to compromise in the interests of experimentation.
TC: It sounds very interesting. I agree with your approach to try sticking to a tube-pure design first, unless it is apparently overwhelmed by the solid state one. What’s meant by the gas tubes that you are mentioning? Is it the VR tubes such as 0A3, 0C3 or 0D3? Could you kindly elaborate more or sketch
the schematics? For the Direct Couple approach, the designs that I have seen are quite dangerous as the 2A3 or 45 will blow up when there is something wrong with the front tubes.

Richard: Yes, I mean something like 2 0D3s in series ‘under’ the cathode, with around 10nF across them to attenuate the HF noise. People get fairly emotional about gas tube noise but I have never been able to detect sonic degradation resulting from their use, even in my ultra high-resolution 416D phono pre-amp. It is important not to impress too much current swing onto gas tubes so the load resistance on the drive tube (if any) should not cause more than a few mA of current swing at full drive. This again is a compromise; we have to run the gas tube hard enough to prevent it (or them) from extinguishing at current troughs, but not so hard as to cause flaring at current peaks. DC coupled designs are tricky, I (usually) do not DC couple more than 2 stages. As I said, the clear best compromise for my design was to RC couple the Super Mu Follower to the drive tube.
TC: They mentioned about the Common Mode Choke (CMC) design. Can you elaborate what’s the difference between CMC and the chokes that you are using?
Richard: The common mode choke effectively shorts any common mode noise that is on the (filament) supplies, within the frequency range of the choke. It does this more efficiently than separate chokes. What I am trying to do is to 1/ attenuate line noise and 2/ isolate the end-to-end ac signal on the filament from the DC supply circuit. I used choke values that do this well down into the audio band, I am not sure where the cut-off is though. A common mode choke will try to short the end-to-end signal as well as incoming common mode noise. Usually the inductance of such chokes is too small for this to be significant within the audio band. However, in the same way, the inductance is not sufficient to keep the end-to-end signal out of the DC supply. So, once again, a compromise. I have not tried common mode chokes so this is a philosophical issue for me and not proven, whatever that may mean; in fact, I find that what seems to be proven in audio all too often turns out to be the flavour of the moment. The ONLY line of thought that I find to be solid is to do as little as possible to the signal in increasing its energy to a transducible level.
TC: Also, I find that some TV damper diode tubes are very good sounding rectifying tubes, but cheap too. Some examples are 6CJ3 (half-waved) and 6BY5G (full-waved). I have tried 6BY5G personally, very very good. Have you tried this? I will use two 6BY5G (with choke) to feed the B+ of T1610: both plates of 6BY5G will be connected together (to increase the current) to act as a half-wave rectifier tube and two such 6BY5G will be used together as a full-wave rectification.
Richard: I maintain that if you can hear the rectifier then the PSU is not efficient in keeping the audio signal out of the least linear portion of the circuit, the rectifie;, see the notes on PSU design on my site.
TC: For the B+ of input and driver stages, I will use another 6BY5G (full-wave) with choke and am considering of using the simple VR tube shunt-regulated design to feed them, such as 0A3, 0C3, 0D3, etc. Anycomments?
Richard: It should work well. Ensure the DC to the gas tubes is well filtered, try to minimize any resonance due to smoothing chokes or better, use an C-R-C filter and a mosfet current feed to the gas tube. This will provide an extremely high level of isolation of the regulated B+ from the rectifier and line noise. Put a good (Teflon) 10n cap directly across the gas tube to minimize the gas tube hash, keep the gas tubes say 3 inches or more away from the audio components.
TC: For your Super Mu Followers, how do you compare the 76-choke-76 and the 76-choke-resistor-27 in terms of sound quality?
Richard: On the 27-76 and 76-27, I cannot really comment since I lost my aural memory as I made the changes. I do know that both ways sounded good and I did not get any sense that I lost anything when I made the change.
TC: In comparing various types of driver tube, you mentioned that your order of preference is 71A/50/845, 45, 2A3 and 300B. So I will choose 71A, 45 or 2A3. Can you elaborate more the brands and types (e.g. clear or smoked glasses, single-plated or double-plated) of the 71A, 45 and 2A3 that you have used in the comparison for my reference please. One more point is that the plate current of 71A seems to be too small for my case, isn’t it?
Richard: On tube types, I have not had the privilege to evaluate a range of brands. In general RCA tubes sound good. On the 2A3 I did get a chance to review a range of types but my memory does not serve me well. I do know that of the several we tried, the RCA bi-plates were clearly superior; I am told that the rare mono-plate is better still. The 71A is capable of generating up to 3/4W and as an output tubes is miraculous. Nothing I have heard quite gets the balance, resolution and transparency of a 71A. If you are operating in Class A1, the plate current will be more than sufficient. I would suggest that you do not allow the issues of plate resistance and current for the drive stage to constrain you overly unless you want to consider the 6C33 (and why not?).
In conclusion, I believe that the art of this field is in finding the magic compromise that works for the individual.