A monster DHT amp
Lately I haven’t had any time for audio work unfortunately. Changing nappies to a 4 week old baby whilst working long hours is tough. I can get the odd 30 minute here and there and every time I try to get upstairs to the workshop something pops up. Never mind, hopefully things will get easier in the near future.
I’ve been asked about the 4P1L pentode driver. It’s been a long time since I did those tests and never got around to listen to the driver sound. Tests were promising but never managed to include this driver on my amp.
Driving transmitting valves is a challenging task. Especially if we want to take them to A2-land (unless they operate in A2 whilst in zero grid bias). Driving big transmitting valves like 211, 805, 845, 813 or GM-70 require a large swing of volts for the driver which should do this linearly. The load is quite demanding in particular when we approach the grid to 0V (or biased positively) and using a triode as driver also puts a daunting task to the previous stage due to the Miller effect. It’s not easy to find triodes that can swing 300Vpp with very low distortion.
There are plenty of interesting designs out there using triode strapped pentodes like C3g, D3a or other Russian pentodes. Or beam triodes like 6HV5A. This post is mainly intended to answers the questions I was made about how to use the 4P1L in pentode mode as driver.
The Spice Model I developed time ago can be definitely improved with a new dataset to make it more accurate. I learned this after developing several models using the uTracer and Derk Reefman method. This may explain the differences in THD between simulation of the driver and what I measured on the bench.
I managed to get 0.3% for 200Vpp swing on the bench and simulating a similar circuit gave me about 0.5-0.6%. Can only blame the model I developed 🙂
The curves for the 813 model I think I took them from Andrea’s site here. I have a stash of GM-70 and 813 and some day I will build a similar circuit to this exercise. You can use your own transmitting valve of your choice here providing you adjust the related parameters. I’m not intending to do any DC-coupling or A2 drive here, so stacked supplies and other nice circuit options are not going to be expanded here. I’ve been running 2 years now with the 814SE A2 amp so if you are interested on alternative topologies for A2 operation then you can read some of that here.
The Spice model I developed for the 813 (see above) seems to match reasonably well the hand plotted curves. Looking at the load-lines you can easily get 20W at 1.4% THD with an 11K Zaa load. There is an interesting H3 contribution to this output beam tetrode triode strapped. More power can be obtain by driving it harder above 1kV or swinging it into A2. Alternatively a lower Zaa can work, but not sure how much additional juice you need from this beast. It can be really loud with just 10W! I have efficient speakers so will never need this crazy power levels. I’d be interesting in the sound signature at 5-6W more than anything else.
So here is the first circuit draft:
The driver is the 4P1L pentode in hybrid mu-follower mode with a sandy load using the DN2540 gyrator. The LND150 provides the voltage reference for the gyrator. M4 can be changed to more robust FETs depending the voltage needs. I like the sound of the DN2450 or alternatively the IXYS IXTP01N100D which can run at 1kV. Here we are not swinging more than 350V positively so the DN2540 with a 400V supply should work fine. V4 is a stable screen bias of your choice. R4 sets the gain and I lowered in the simulation to lower the THD. I wasn’t happy with the simulation results of the driver. I’ve got good THD in a higher gain driver as per my previous post on that. C1 and C2 at 220nF value makes the LF good whilst keeping their values low.
Driver filament bias is simulated above with a current source representing the Rod Coleman regulator. Filaments are in parallel to draw 660-670mA so filament resistor is kept to a minimum value as this is reflected in the anode increasing the output impedance and reduces the stage gain.
The 4P1L is not driven hard at 17mA and with a low screen voltage. I found this to be a good point for lower distortion. Asking more than 200 Vpp to this valve is hard in pentode. As you can see in the graph above, below 80V the curves are crunching so the steeper load (i.e. lower resistance) is good on the lower voltages / higher anode currents as it takes the operation away from the knee. Obviously moving it above 300V should be better but the 4P1L is operating there above its limits. You could do this, but at your own risk on the 4P1L lifespan.
The source follower can use the MOSFET of your preference as well as the source CCS can be made of a resistor or bipolar cascode. You will need a -250V supply which can set the grid bias using a pot to derive R5 instead of the V3 shown above. V5 is left to 0V and should be connected to ground in A1 operation.
The output transformer is a pair of LL9202/90mA I used before in my 814 SE amp. It’s configured on 11K and some of the transformer models used in this simulation are from my LL1623. I just adjusted the Lp and the step down ratio to speed up the simulation process. I can tweak and improve this, but not for now.
Looking at the performance we can see that the higher distortion initially found out on the driver simulations is reflected in the results below. To the 1.3% of H2 distortion provided by the output valve, you have to add the H3 distortion mainly from the 4P1L driver:
The H3 levels are close to the H2 levels at 20W and the overall THD gets up to 1.8%. A 0.4% more than what was predicted with the loadline. I’d be keen to listen to the amp at more reasonable levels like 2 – 6W where the distortion is low.
Definitely it’s worth trying, but implementing very high voltages like in this amp is not for any beginners. This is serious stuff. Even for myself after working many years with 600V-ish and stacked supplies which takes overall level to close to 1kV, I’m still very cautious. The HT supply is another matter as are the DC filament supplies here.