6H13C OTL using 12 parallel tubes

[sampleaccurate]

The topic of how much power is needed will be debated forever, but the bottom line for me is that I have 3 sets of JBL studio monitors, I have used them for over 20 years, my ears are accustomed to them, and the titanium tweeters reveal high frequency detail that even much more expensive speakers I've tried simply don't deliver IMHO. I'm not buying new speakers, so the change will need to be in the amplifier. I'm not ditching my 300B SE or my KT88 PP, but after listening to a 120 watt per channel amp for several days now I realize I WANT more power. My speakers can handle it, and I also want a low output impedance for solid bass.

I purchased a 24" by 24" 1/8 inch aluminum plate for the top of the chassis, and the sides will be 6" high oak. The amp will have 12 output tubes per channel - 24 output tubes in all. I'll be using the Russian 6H13C (a close equivalent of the 6AS7G). The amp will be huge, but the chassis will allow any future modifications or improvements to be made, and provide ample room for the 24 large output tubes. This will not be a portable amp. If it sounds good it will have a permanent home as my main reference amp. If it sounds just "OK" I'll sell it off on ebay.

I'll post the first draft of a schmetic I modeled in PSIPCE. The simulation results are impressive.

 

[sampleaccurate]

It's "working" beautifully so far. I got the offset at the output down into the pico-volt range, and it's not a super sensitive adjustment. In other words adjusting the bias to null DC at the output looks like it will be relatively easy without resorting to 20 turn pots.

That's one channel, but it's only 12 tubes - each tube is represented on the schematic by two circles because they're dual triodes so it looks like more than it is. If you want a picture of what the final stereo amp will look like though, that's about how the 24 tubes will be laid out on the chassis.

The 24 power tubes will use 6 volts times 2.5 amps times 24 tubes equals 360 watts for all the heaters. That's not that bad when you look at some other high power amps. Actually I'll be using 12 volt SMPS units and running two 6 volt heaters in series. I could use one SMPS to power 4 heaters in theory, but I think 5 amps of constant load on a cheap Chinese SMPS is probably a bad idea. With only 2.5 amps of load the SMPS should coast. So I have 12 (10 right now, more on the way) $10 SMPS units to power the heaters. 360 watts won't roast me but it's a lot of heat.

It's ruler flat from 10Hz to almost 1MHz. The 3dB down point at the low end is about 3 Hz, but take that with a grain of salt. It all looks good so far. I'm sure in "real life" these numbers wouldn't look so perfect, but that's OK. Before all is said and done I'll have played around with the component tolerances to know what's critical and what's not.

I'm jumping right on this amp. As soon as I mark my aluminum with the tube spacing I'll start drilling it out. I picked out my oak side panels at Home Depot and my Dad is going to cut the wood so the corners of the chassis will look nice. The sockets are on the way too. Here's the first draft of the working schematic:

An externally hosted image should be here but it was not working when we last tested it.

 

[sampleaccurate]

222 WATTS!!!

I'm showing 222 watts of power into 8 ohms. This assumes identical output tubes, which won't be the case when I build it. I expect less but I expect at least 150 watts. The distortion looks very low.

The drivers are parallel 6SN7 triodes, one tube with the triodes in parallel for each phase for a lower output drive impedance. Each channel will have one 12AX7, three 6SN7s, and 24 6H13C power tubes. The stereo amp will have 56 tubes in all.

The only drawback is the number of coupling caps, but I'm using very high quality caps for this build.

2 volts p-p input gives 150 watts into 8 ohms. Frequency response is flat past 100k. The 3dB down point for the bass is about 5 Hz. I measure about +/- 0.5 dB from 10Hz to 100kHz at 222 watts RMS with no visible clipping. I don't know the distortion figures but superimposition of a perfect sine wave on the output waveform and visual inspection indicates that it's very low.

WOW! I'm very happy at this point with circuit performance. This will be my first OTL, but if I'm going to build one why not go all out? The tubes are cheap. The heater supplys and sockets cost more than the tubes!

I'll get a higher resolution schematic up soon so the component values can be read.

 

[Alastair E]

Personally for drift and reliability issues, I would drive the power bottles via a Follower, this allows better control of the bias as the G1 circuit of the output tubes is of a lower impedance....

With SO MANY tubes I think I would go for MOSFET followers!

But, Thats just my experience.....

 

[SemperFi]

Very cool...actually very HOT

I take it those numbers you come up with are estimates from simulations? No way you'll get that low offset in this world, but a few 100mV wont hurt JBLs.

I cannot read the numbers with the resolution I get on my screen. Is that +-150volts or +-190V?
These tubes do not come matched, and although I believe matching is over rated, you really ought to use cathode rseistors. With this many tubes I'd even use something like 10ohms, but everone freaks out at the thought, so I'd say at least use 3ohms.

Also, these tubes need hard and steady bias with a relatively low resistance on the grid. 100kohm per tube gives you a input resistance to the output tubes of 8333ohms. The drivers need to be able to drive that. (You can of course go crazy in the real world and use something more practical, but it shouldnt be too high).
Ralph knows this much better than me...What' u think???

Great project, I'm looking forward to the results, good luck.

 

[sampleaccurate]

Latest design

Here's the latest design I came up with. It has an output impedance of 0.033 ohms, and can output 250 watts RMS into 8 ohms at very low levels of distortion. 300mV input will drive it to full power.

You should be able to read it now.

An externally hosted image should be here but it was not working when we last tested it.

 

[alexberg]

Is it just simulation or real prototyped thing?
You need at least 140V amplitude from the driver and be still shy of getting 8A peak current into 8 Ohm load.

 

[sampleaccurate]

Building Soon

Is it just simulation or real prototyped thing?
You need at least 140V amplitude from the driver and be still shy of getting 8A peak current into 8 Ohm load.

The driver can swing a lot more than 140V! The power section will clip first.

Right now it's a simulation using PSPICE. However, the parts are on the way, and I anticipate constructing this thing in January. There is still a lot of work to do, especially failure analysis so when a tube shorts or some other catastrophic event occurs, the amp doesn't blow up my speakers or itself. There is also the issue of start up. The voltages used in the LTPs are large, in excess of what the tube can handle if it's non-conducting.

For these problems there are a number of solutions such as gas discharge tubes, MOVs, and inrush current limiters.

Here are some response plots from the first schematic. The second schematic has better specs and very low distortion. It turns out it was the 12AX7 gain stage that was creating 90% of the distortion. Once I replaced it with two SRPP 6SN7 stages the sensitivity went up and the distortion went down. The bandwidth also went up.

These simulations are generally very accurate, with the caveat that they are only as accurate as the tube models used in the circuit model. So far I've been very pleased with the models I've found online. I find computer modeling an indispensible tool for designing amps or anything else for that matter. I may need to adjust the resistors by 5% or even 10%, but once I get the bias voltages and currents close to where the simulation shows they should be (and for which the amp has been optimized for) the amp will work - maybe!

There are still some pitfalls and traps to fall into. The amp could oscillate if it isn't constructed properly. I attempted to get the amp to oscillate by introducing simulated parasitic positive (as well as negative) feedback. I was able to get it to oscillate, but the coupling required was far in excess of what I expect to see in the actual build. The circuit seems very stable.

The response:

An externally hosted image should be here but it was not working when we last tested it.

The simulation uses a model of a 6AS7G, not the Russian equivalent 6H13C that I will be using. I scored the tubes very cheap in sealed boxes from Ukraine and Romania.

The output voltage with a sine wave input into an 8 ohm load. This looks better in the latter circuit revision.

Also note there is a DC offset. The DC offset is proportional to the amplitude of the output signal, but the time constant is such that it should be inaudible (it takes several seconds to develop) and the amplitude is such that the power wasted in the speakers will be less than 10% of the total input signal to the speakers. At idle with no input signal the DC offset at the speaker is less than 1mV and is easily adjustable:

An externally hosted image should be here but it was not working when we last tested it.

 

[6L6]

This amp is really interesting!! I am looking forward to seeing all your progress!

Why do you want to build it as a Stereo amp? The amp itself is going to be huge, the PSU is going to be huge, and it seems like making it monoblocks will let you actually pick the thing up without requiring a forklift...

 

[el156]

That amp would be very nice as amp for my ALL house , but also as a central heating system
How many watts will it consume?
Silvino

 

[SemperFi]

I will be following this I could use an amp like this in my house which is much too cold these days.
I agree simulations can be surprisingly accurate and your amp may very well perform. However, the simulations do not take into account the actual layout of components and how that interfers with an otherwise ideal assumption. You should give all tubes some random difference in mu or Gm to simulate offsets, as well as assume you'll have some unspecified temp drift.
I see you get farely symmetrical output swing even though you drive the output without compensating for the difference in the upper and lower tubes in the totempole. Only massive amounts of NFB will do that. That is my only caveat to real world performance.

Are you not going to use cathode resistors to help balance current sharing as well as offset drift?

 

[sampleaccurate]

Layout and Current Balancing

I will be following this I could use an amp like this in my house which is much too cold these days.
I agree simulations can be surprisingly accurate and your amp may very well perform. However, the simulations do not take into account the actual layout of components and how that interfers with an otherwise ideal assumption. You should give all tubes some random difference in mu or Gm to simulate offsets, as well as assume you'll have some unspecified temp drift.
I see you get farely symmetrical output swing even though you drive the output without compensating for the difference in the upper and lower tubes in the totempole. Only massive amounts of NFB will do that. That is my only caveat to real world performance.

Are you not going to use cathode resistors to help balance current sharing as well as offset drift?

You bring up some good points that I haven't neglected. Certainly the amp must be constructed to minimize the potential for problems, and the layout is important. I've introduced some parasitic capacitance (but not inductance) in some key spots, and the circuit behaves rather well under simulation. It's difficult to get it to oscillate or misbehave without adding a lot of coupling in areas where there will be almost none due to careful layout.

There will be cathode resistors (not shown yet). Initially I intend to make those 1 ohm simply to allow measurement of the bias currents. I'll start with a lean bias and check each tube to see how much variation there is. The value of resistors required to keep the tubes reasonably well balanced will need to be experimentally determined. I considered changing gm and mu in the models to simulate "real world" tubes as someone suggested, but I have no idea how much variation to expect so I concluded it would be a waste of time and is best done experimentally.

The circuit is designed to bias the tubes well under their max power rating, and it's not that important for them to be balanced. It's more important to make sure that none of them exceed their dissipation rating and burn up. That I will be certain to do.

 

[atmasphere]

There will be cathode resistors (not shown yet). Initially I intend to make those 1 ohm simply to allow measurement of the bias currents. I'll start with a lean bias and check each tube to see how much variation there is. The value of resistors required to keep the tubes reasonably well balanced will need to be experimentally determined. I considered changing gm and mu in the models to simulate "real world" tubes as someone suggested, but I have no idea how much variation to expect so I concluded it would be a waste of time and is best done experimentally.

I recommend 5 ohm 5 watt devices. If you use 1 ohm 5 watt units, if a tube arcs that will be the end of it. The 5 ohm resistors also sound better. The only issue is that if a tube shorts, it may not blow out so you do have to think about speaker protection. IMO, rating the B+ fuse properly should do the job.

The 100 ohm grid resistors should be rated at least 1 watt. That way they can survive a tube shorting/arcing.

 

[sampleaccurate]

The 100 ohm grid resistors should be rated at least 1 watt. That way they can survive a tube shorting/arcing.

I didn't consider that yet, but you're right, using small resistors there would be asking for damage.

I think I'll throw some 5 ohm cathode resistors in the simulation to see how it affects the performance. I'll post the results when I get them.

 

[atmasphere]

If all the tubes are rated the same, I doubt it would help in simulation, but in practice you can hear it easily. These resistors would appear to raise the output impedance, but in practice they actually lower it, because the output impedance otherwise has a lot to do with which tubes are hogging the current.

 

[sampleaccurate]

If all the tubes are rated the same, I doubt it would help in simulation, but in practice you can hear it easily. These resistors would appear to raise the output impedance, but in practice they actually lower it, because the output impedance otherwise has a lot to do with which tubes are hogging the current.

That makes sense. If only some tubes are conducting fully and others are near cutoff the output impedance would be higher than if all were conducting.

I'm in the process of modeling a circlotron version of the amp. I ran across some interesting reading material and now I'm seriously considering the circlotron topology instead of a totem pole output. I still want to use 12 tubes per side. Believe it or not the parts count isn't much different. I was even intending to use lots of parallel 100VA isolation transformers for power, 6 to 8 on each side of the +/-150V supply. Instead I'll use 6 -8 transformers per channel, 3 or 4 in parallel for two 300 or 400VA matched supplies on each channel. The circlotron ties both ends of the load to a cathode, and the perfect symetry of the design insures the negative signal excursions don't exceed the positive ones as they do with a totem pole.

If it works in PSPICE this will be a 12 parallel tube OTL based on the Circlotron circuit.

 

[alexberg]

The farther is goes the more wishful it gets.
Unless you get trafos with exact turns ratio or
connect primary/secondary in series
while secondary/primary in parallel at the same time
you will find balancing power between transformers really frustrating.
Or...
Are you going to use separate power source per tube?
Power supply tubes are "very particular":
datasheet shows that the fixed bias current is all over the scale i.e. 40...112 mA.
Attached is the speadsheet with cathode resistors to use per number of tubes in parallel
along with power dissipation per anode.
5 ohm resistor in cathode would be of no use, at very least.
So, as mentioned before, do you have hundreds of tubes on avail to match... some?

 

[sampleaccurate]

The farther is goes the more wishful it gets.
Unless you get trafos with exact turns ratio or
connect primary/secondary in series
while secondary/primary in parallel at the same time
you will find balancing power between transformers really frustrating.
Or...
Are you going to use separate power source per tube?
Power supply tubes are "very particular":
datasheet shows that the fixed bias current is all over the scale i.e. 40...112 mA.
Attached is the speadsheet with cathode resistors to use per number of tubes in parallel
along with power dissipation per anode.
5 ohm resistor in cathode would be of no use, at very least.
So, as mentioned before, do you have hundreds of tubes on avail to match... some?

The power trannys are close enough in turns ratio to use with no balancing problems, although I wouldn't recommend mixing transformers from different manufacturers or of different sizes.

I'm well aware the bias will be all over the place. A less than 3:1 ratio (112 to 40mA) is perfectly acceptable. 5 ohm resistors are for current limiting more than balancing.

What's the problem?

 

[alexberg]

The power trannys are close enough

SKIPPED

5 ohm resistors are for current limiting more than balancing.

What's the problem?

Alrighty then.
Please disregard what I've posted.

 

[sampleaccurate]

Alrighty then.
Please disregard what I've posted.

I appreciate the warnings about the pitfalls of constructing and balancing an OTL. What I'm attempting has already been successfully accomplished by others with less experience than I have.

One thing I intend to do is individually bias each triode grid to balance the bias current. That's an easy thing to do. It takes a lot of pots and caps, unless the tubes are all biased lean and pull up resistors are used on the grids to raise the bias to what it should be. I've seen several methods for balancing tubes that don't come close to matching.

When using parallel triodes, diversity and probabalistics tell us that the more triodes we use the closer the balance will be. Some tubes will conduct as you point out up to 3 times as much current as others, but if none are allowed to exceed their power rating there should be a close balance when using many tubes, or the tubes could be graded and chosen so that the sum of the currents will match.

Nobody said this would be easy.