This design was a completely new direction for me, having eschewed pentodes up until now. I came up with a fully differential topology using a combination of output stage cathode-coupled feedback and global feedback. I hope the following may prove both informative and fun, I had a lot of fun debugging my re-design.
To drive a load, pentodes require negative feedback and the (poor) quality of the output transformers that came with this unit challenged me to learn how to effectively damp ringing at source, not merely attenuate it, also how to properly lag-compensate the circuit so that it will remain stable under all conditions into a severely reactive load. To summarise what was done to obtain stability: I identified and damped any resonant circuits (in this case due to output transformer leakage inductance and parasitic capacitance) and I slugged the dominant pole of the amplifier. Coupled with this is the use of adequate value grid stopper resistors. On the LF end, I used a completely separate power supply for the driver stage, which in the words of Morgan Jones, will kill motorboating “stone dead”.
I saw the Mini P1 push-pull 6P1J
(Russian 9 pin equivalent to the 6AQ5) on eBay, it was cheap and liking the look
of it, decided to buy it. Upon looking inside I had my doubts, so I tested it
(before coupling it to my speakers) and it was - wait for it - junk; it didn't
even come close to meeting the promotional specs, more like 1W, not the claimed
10W “Class A” (yeah right...). I sent a detailed report to the vendor and heard
– you got it, nothing. (I did take the trouble to draw out the existing
circuit, which was compromised but actually workable. The limitation was due to
a combination of worse than awful construction and insufficient feedback
for
the pentode output stage to drive more than 1W into 8Ω.) Still,
physically, it is an attractive piece and so having let it languish in my
basement for about 3 years, I have worked up an interesting circuit for it.
While tuning my design, I spent a lot of time sorting out damping of the ringing
of the cheap output transformers and compensating for the double pole roll-off
to get absolutely solid stability, even though there is ‘only’ 10dB of feedback.
This most likely explains why the feedback on the amp as designed was
insufficient. Perhaps the cost of sorting the thing out properly was not
acceptable to the builder.
I took some pictures of the circuit board,
which unfortunately, are not clear. It was a peeling mess of dry joints. I
removed all the components (and threw most of them away), sanded the copper mess
off it and attached some aluminium foil for some reason then used the board as a
base for the new wiring. Actually, I put the heater wiring between the board and
the chassis so that the aluminium foil would afford a measure of shielding to
the audio circuitry.
After
a lot of thought and fooling around in MCap (I am NOT an EE), I came up with a
new design for it. While dismantling it, I had discovered that I could get at
the OPT secondary section connections and to my delight, the exposed connections
proved to be centre-taps. I am (was) not familiar with working with pentodes as
output tubes and I had not properly appreciated the role of negative feedback to
get a workable output Z so that the things will actually deliver some power into
a load. So, the centre-tapped secondaries are perfect as I have used them to
provide cathode coupled local negative feedback, thereby reducing the level of
global feedback required. This gets the Zout down to around 10Ω; only 6dB or so
additional (global) negative feedback is required to get the Z into a workable
range. The global feedback is applied to the negative input grid of a
long-tailed-pair driver, resulting in a Zout around 2.7Ω when using a 5751 for
the driver or 2.5Ω when using a 12AX7, definitely workable for 8W speakers.
Maximum output is 7W,
actual Class A; the sensitivity for the full glorious 7W (smile)
using 5751s is 1.2Vrms and 1.1Vrms using 12AX7s. At the full glorious 7W,
THD at 1kHz is 1.5%. I did not bother to make any other THD measurements.
The transconductance of a pentode is extremely sensitive to the screen grid voltage, so it is clear to me that screen regulation (coupled with fixed bias) is a good way to get the best out of pentodes. To this end, I included a simple series mosfet regulator, one for each channel*. In a first for me, I did not use regulation for the plate supply since a pentode is largely unaffected by small plate voltage changes and I could not afford to drop any B+, it just makes 250V on a good day. All this combined with fixed bias (to simplify the cathode feedback connections and to maximize use of the available B+) results in extremely stable bias current that I set at 36mA per tube.
*As a side note: I have designed and built many variations of series regulator including some splendid tubed efforts; They all work well however, I have found that the simple series mosfet sounds as good as the best of them and it is easy to include soft start by putting a long time constant on the mosfet gate. I am sure that shunt regulation is superior, however to implement well requires more parts and effort than I wanted to put into this design, err, well perhaps I just ran out of space under the tiny chassis!
The differential driver stage uses a current sink for the tail; I did try a current-regulating diode however the shunt capacitance caused poor phase accuracy above 10kHz. I then went to a single DN2540 mosfet design, finally going to a cascode mosfet design. I did some measurements on single and cascode designs and the cascode is markedly superior. (I have published these results with the commentary on the 6AS7 PP update.)
The voltage available from the power transformer is not adequate to use a 5751 or 12AX7 and low voltage triodes such as the 6DJ8 would not provide enough gain to allow adequate negative feedback. I resorted to using two small transformers (stuffed under the now full chassis), back-to-back (120V:36V-36V:120V) running into a voltage doubler with separate mosfet series regulators for each stage. These two transformers are visible in the picture showing the bare board, above. (I used the low voltage between the two transformers to provide a source for the bias supply.) The final design is carefully worked out to allow both 5751s or 12AX7s to operate with excellent linearity driving the 26V pk-pk swing into the combined plate resistor/output stage grid leak load of 69k without making any changes. The operating current is 1.05mA per section and the supply voltage is set to result in a plate voltage of around 180V.
The voltage doubler setup was causing the second of the back-to-back transformers to radiate emi spikes followed by ringing as each diode ceased conducting. To stop this I put an RC snubber across each diode that completely damps the excitation of the power transformer leakage inductance/parasitic capacitance. The way I did this was to set up my scope* so that I could measure the period of the ringing and then put small caps across the diodes until I got a frequency reduction of roughly 2 (i.e. the ringing period doubled). This turned out to be 33nF. I then tried different values of damping resistor in series with each diode until the spike and the ringing was suppressed, the best value turned out to be 10k. I used carbon composition resistors to ensure that there is no (well vanishingly small) inductance associated with the resistors. My approach is to use as small a cap and as large a resistor as possible that will accomplish complete damping.
* I simply set the scope on 5mV/cm and put the probe (on X10) next to the transformer, the spike and ringing energy will couple inductively with the scope probe. I set the triggering on ‘line’ and the timebase fast enough to clearly see the ringing cycles.
So the main features of the new circuit are:
> 5751/12AX7 long tailed input/drive stage with cascode current sink.
> Cathode coupled local feedback on the output stage (around -3.6dB)
> Global feedback to the -input (around -6dB).
> Mosfet regulated input and screen supplies, separate for each channel.
> Unregulated output stage B+ supplies, a first for me!
> Fixed bias (which makes the best use of the available output stage B+ voltage).
> Regulated adjustable bias supplies, separate for each channel.
> Extremely stable output stage bias currents, The screen regulation coupled with fixed bias pretty much prevents any variation of the unregulated B+ affecting the bias current in the output tubes.
> 7W Class A at 1.5%THD, measured at 1kHz
> -3dB at 53kHz referred to 7W at 1kHz.
Debugging this design has been fun! Even
with 'only' just under 10dB of total feedback, getting solid stability has been
a game. The output transformers have two poles, around 80k and 115k.
Accordingly, the task was t
o damp the ringing due to the output transformer
leakage inductance and parasitic capacitance. I investigated this using the same
technique as I had learned on the diode snubbers, i.e. looking for a tank
(resonant circuit) where the frequency would drop when I put a capacitor in
parallel with it. (If the frequency does not change then the loop under
investigation is not a tank and the snubber will merely attenuate the ringing,
not damp it at source.) I ended up with a RC snubber across each half of both
the primary and the secondary. I acknowledge Volt_Second for providing a
detailed explanation on how to identify tank circuits and work out the RC
snubber values at:
http://www.siteswithstyle.com/VoltSecond/Damping_ringing_XFMRS/Damping_ringing_in_xfmrs.html
This work started with the output transformer secondary as the most sensitive tank to capacitance. When I had that squared away, the tank associated with the primary became evident and so I developed RC snubber values using the same technique. Even with all this, I found that when I put my scope probe (on X1) on one of the drive plates, oscillation at 115kHz would erupt. Also, shorting the output would have the same effect. I can't say I fully understand what was going on, other than it is related to transformer phase lag affecting both the local and outer feedback loops. Turning the output tube current down a few mA would stop it, which sent me on a detour, investigating screen grid parasitic oscillation (more on that later). And so, what was actually next was to slug the circuit dominant pole (which is the drive stage output Z looking into the output stage Miller capacitance) with a 82p cap across each output tube grid leak resistor. Looking back, this is actually a properly ordered stability process, taking each issue in turn of severity, dealing with it and moving on, down the chain. However, I suspect that the redoubtable Norman Crowhurst would consider all this 'faking it': Reading Crowhurst, in his view with the issues I experienced, the only way would be to replace the output transformers. Not for this boy!
Crowhurst notwithstanding; the final touch, to minimize squarewave overshoot, was to put a phase advance capacitor across the series global feedback resistors: If you listen to squarewaves, at 1kHz the shape is pretty clean into all loads I tried. At 20kHz, the shape still shows signs of having been related to a squarewave! In all seriousness, an amp that does not remain stable when driving squarewaves into a reactive load is junk, period. Now, it is stable into no load, short and capacitance (I tried 100n to 2u) across 8 ohms. The final overall frequency response for the whole show at 7W is -3dB at 53kHz. Roll off beyond that is well behaved, nice and smooth with no signs of resonance.
During the stability investigation, at a point of desperation, I tried all sorts of screen stopper stuff, however that was not the source of the problem (which of course, I really knew, I was just getting a little flaky). Whatever, Mullard recommended a common screen stopper for PP pentode output circuits in order to minimise the modulation of the screen voltage by the signal. In the past, I have fooled around with diodes in the screen circuit (of triode connected pentodes) as suggested by Dennis Grimwood. This does seem to yield a lessening of any mid-bass muddiness but who knows really... Whatever, instead of a common screen stopper, I am trying a common screen diode and for emotional security, ferrite beads located right at each screen pin. We will see and maybe even hear?
The other wierdness I am trying is a diode before the final B+ bypass cap (TSHB) to perhaps, maybe, probably not, prevent back emf current from the speaker flowing around the rectifier circuit. Eye of newt and toe of frog......
To add to the fun, I found that the heater supply was causing HF coupling between the stages, which made the HF related issues harder to diagnose/rectify. Grounding using two equal carbon (non L) resistors really helped. Hum and noise is as low as I could wish.
One good thing I have to say about the iron this unit came with; the power transformer runs really cool, despite running the bias current a bit harder than was intended with the original design.
I have learnt a few things:
1/ Pentodes do reproduce music well.
2/ My fetish for regulated supplies is mitigated with pentodes since the operating conditions are driven by the screen voltage which only requires regulation capable of supplying a few mA.
3/ Pentodes are tricky and really good iron would help. Much skill is needed with cheap iron.
4/ I have a new-found respect for designers of good pentode amps, though I guess the commercial guys for the most part worked with decent iron.
5/ How to identify resonant (tank) circuits in an amplifier and damp them effectively.
6/ Don't know how to do this without being a 'scope jockey.
On the sound, well, given that my ears are pretty lame, I really like it. It is musical, resolves well and surprisingly does not get congested with orchestral stuff. It is much, much better than I expected and frankly, more than good enough to enjoy all music. If I was not interested in tubes 'n stuff, I could call this it. It also has plenty of power for me. (Playing with tubes is my version of a ‘train-set’, I prefer full size steam locomotives, must be steam, and I don’t have enough money for that……)
And finally; speaker cables are effective antennas and any RF will arrive at the input stage of an amp having global feedback. I did not get too obsessed with this, remaining content to provide 4.7k grid stoppers on the drive stage grids. This value acting with the Miller capacitance will attenuate anything much beyond 500kHz, which in hindsight could be lower, say 200kHz.
Click here to see Amplifier Schematic: