This amplifier was inspired by Steve Bench's experiments with the Talema small power toroids as output transformers. I built it in an empty Bombay Coy tea sampler box. This design was published in the May 2006 edition of AudioXpress. Since then I have taken a much closer look at the constant current sinks (CCS) I used to prevent DC saturation of the toroids, notes on this are at the bottom of this page.
It produces 4W pc in Class A. The output tubes are biased using DN2540 (depletion mode mosfet) constant current sinks, carefully trimmed to get the unbalanced DC current in the toroidal output transformers under 0.1mA: this is very important to ensure good LF response from the ungapped toroidal transformers. The cathodes of each output pair are coupled using a (10µF) film capacitor to create power long-tailed pair output stages. This technique results in pure Class A operation and near perfect signal current balance in the output transformer primaries. The even order harmonic cancellation of such an output stage is superior to that of a conventional cathode or fixed biased push-pull output stage. The toroids used are the largest in the range, 50VA: At $18 each (at the time anyway), the Talema units cannot be beaten for value as output transformers. Furthermore, they are the only transformers I have tested which can reproduce a 20kHz square wave with absolutely sharp corners, no ringing or undershoot! (I think square wave performance is an important aspect, so much of music has a fast attack or transient nature.)
The input stage uses a 6GM8 low-voltage triode, one section in each channel. This tube has a screen between the sections that helps to ensure good channel separation. The low plate voltage (20V) permits DC coupling to a long-tailed pair drive stage resulting in topology very similar to the classic Mullard circuit. The circuit departs from the Mullard design by using a DN2540 constant current sink (CCS) in the tail, resulting in near perfect balance. (The Mullard design uses a slightly larger resistor in the second triode of the long-tailed pair to compensate for the cathode signal leakage in the cathode resistor. The CCS is a benefit of later technology.) The long-tailed pair uses E88CC's or 6922s operating at a little over 6mA per section. (The plate voltages are around 130V, this means that the 6DJ8/ECC88 type is a bit marginal in terms of plate voltage rating.)
The combination of extremely accurate self-balancing drive and output stages means that the amplifier is very undemanding of the power supply; a big bypass capacitor is redundant. I used a 10µF film capacitor that is common to both the drive and the output stages; this economy does not appear to compromise the sound of the amplifier. In fact even 10µF may be larger than necessary, all that may be needed is a very small capacitor to attenuate any HF hash that may be riding on the supply. I may get around to testing this hypothesis.* The input stage being unbalanced, does have RC decoupling from the drive/output stage supply rail.
*Since I wrote this, I did some auditioning with John Dahlman, a friend and professional bassist. I trust his hearing as he really knows what music sounds like, as a performer. We concluded that this amp actually sounds most musical and accurate with no B+ capacitor!
The 6AS7s being high perveance triodes need protection from cathode stripping during turn-on. B+ Series-pass fets that are ramped-on using a long time constant filter on the gate reference voltage provide the required slow turn-on.
The sound is fine; full, detailed and fast. I have converted three of my PP amps to power long-tailed pair topologies and in every case, the sound became more detailed and lucid. This design produces fine sound yet is very economical to build, given the low cost of the output transformers.
Click to see the schematic:
NOTES ON DN2540 Constant Current Sink PERFORMANCE.
July 2008: The amp had failed a couple of years ago, I inadvertently left it on when I went out. When I got back, the sound was horrible. The CCSs had cooked and gone WAY out of adjustment. I had been mulling over the pro's and con's of the single mosfet devices I had used vs cascode devices. I thought (but had not confirmed) that the Z offered by a single device was excellent. Well it is however, the cascode proved to be much superior, as you will see. However, in this case, what caused me to look at the cascode is that the cascoding mosfet dissipates most of the heat, the dissipation of the control mosfet being limited by the Vgs of the cascoding mosfet.
If you look at the data below, the outstanding aspect is the inherent superiority of the cascode. It is notable how much impact the value of the gate stopper resistor has. I have never experienced oscillation using 100Ω so I do recommend that value. The other aspect that is not revealed by the data below is the DC performance. The simple device exhibited a variation of current with voltage (V on the schematic below) of around 0.025mA/V (i.e. 40k) which is not especially impressive. On the other hand with the cascode, I could not detect a change in DC current with impressed voltage (V) even over a range as wide as 150V. My conclusion is that the cascode design is an extremely effective and simple CCS in both AC and DC respects. It is still important to ensure that the heat from both the mosfets is properly dissipated.
I have not tried the IXCP 10M45S. A couple of folks I have spoken with say it sounds superior however. In AudioXpress, 5/07, Walt Jung has this to say: "The DN2540 is measurably better in terms of AC rejection at the higher frequencies. This is apparently due to the lower parasitic capacitance of the DN2540 versus the IXCP 10M45, bit I cannot precisely confirm this (the latter isn't specified for capacitance)."
Please note that my tests were carried out to support this application at 65mA; I cannot comment on what differences may, or may not, be evident at lower currents. Suffice it to say that I am converting ALL CCS devices in my designs to the cascode design, as I get to them. I was able to match the 6AS7 currents to better than 0.1mA easily, AND this is stable! Finally, a BIG thank you to Walt Jung for generously sharing his work.
|F||SIMPLE Rg =100||CASCODE Rg1 & Rg2 = 1k||CASCODE Rg1 & Rg2= 100|