Modern Balanced Tube Amp Build
Jun 16, 2017 at 5:57 AM Thread Starter Post #1 of 189

SonicTrance

Member of the Trade: Custom Amp Builder
Joined
Dec 27, 2014
Posts
1,554
Likes
1,608
Location
Sweden
Hi,

I thought I'd start a thread on my DIY project I've been building.
Comments or questions are welcomed!

For the last year or so I've been experimenting and building @MrCurwen 's LTP design. It has been many ups and downs along the road but the amp is finally finished!

Granted, I've not heard many top dollar amps but this LTP is the best amp I've heard!. It's very resolving, transparent and non-fatiguing. The LTP sounds clean, crisp yet very dynamic and real.

Tubes do only voltage amplification and cheap russian tubes sounds just as good as the premium audiophile tubes in this circuit!
All tubes have gyrator plate loads and CCS tails. B+ and B- are regulated.

Circuit goes like this:

Input LTP - grid drivers - output LTP - power buffer - OT/parafeed caps.

Pics of the completed amp:
Input stage:
9949617.jpg


Grid drivers:
9949618.jpg


Gyrators/CCS.
9949619.jpg


Output stage:
9949620.jpg


Power buffer:
9949623.jpg


Gyrators. I doubled up on their heatsinks in output stage as they need to dissipate about double wattage compared to input stage.
9949624.jpg


9949625.jpg


9949626.jpg


9949627.jpg


9949628.jpg


9949629.jpg


9949630.jpg


9949631.jpg


9949632.jpg


Like you see from the pics I use separate chassis for input and output stage. That's because the original plan was to only build the input stage and hook that up to the output stage of my LD MK6. But plans changed and I decided to build the complete amp instead. Hence the separate chassis. If I was building from scratch today I'd have input and output stage in one chassi and a separate chassi for the psu.

I encourage people to build it. It sounds amazing!
:)


Final pics of version 2 with 6C8G's and 47's (full build gallery on page 4):

DSC_0539.JPG


DSC_0548.JPG


DSC_0549.JPG



Pic of the SE version with C3G input and 6BG6 output (more pics and schematic on page 8 and 9)

20180719_202533.jpg



Here's the latest edition with SE input stage using 6AT6 and balanced LTP output stage with EL81. This amp is called Oblivion.
DSC_0603.JPG
 

Attachments

  • IMG_0942.JPG
    IMG_0942.JPG
    2.1 MB · Views: 0
Last edited:
Jun 16, 2017 at 11:18 AM Post #2 of 189
Nice! Lots of Russian cold war caps in there :D The chassis looks great! Love the lighted VU meters too.

Over the heatsink, you might try mounting a couple of these old school phenolic terminal strips to hold the resistors and transistors, may prevent a short if you plan on traveling with the amp.

mdhH_ISZ0HfsKYbauOYe16A.jpg



Did you wire up the stepper yourself? Nice job!!
 
Jun 17, 2017 at 7:28 AM Post #3 of 189
Nice! Lots of Russian cold war caps in there :D
Yes, all film caps are Russian NOS caps. They're huge but does the job. Good caps.

The chassis looks great! Love the lighted VU meters too.
Thanks!
I spent some time on aesthetics. Glad you like it! The VU meters look cool but are kind of useless TBH. I've not managed to calibrate them correctly. Sure, the needles are bouncing but not the way they should, if that makes sense, lol.
:)

Over the heatsink, you might try mounting a couple of these old school phenolic terminal strips to hold the resistors and transistors, may prevent a short if you plan on traveling with the amp.
I understand your concern, but I'm afraid that would complicate the assembly of the CCS's/follower's. But I assure you the construction is quite solid. This is not a amp to travel with anyway.

Did you wire up the stepper yourself? Nice job!!
Yes, I bought the attenuator and resistors as a kit from hifi-collective. I did the soldering. Some therapeutic work right there!
:)
 
Jun 20, 2017 at 5:10 PM Post #5 of 189
Do you know what the reasoning for the grid drivers is? Are they driven A2 or just there to buffer the previous stage?

The output tubes are not driven to A2 at least in any meaningful way. Maybe some transients; however on the other side of the waveform, there is not that much B+ available.

The buffers serve two functions:

1) Relieve the previous stage of any drive responsibilities. This includes the coupling capacitor; the capacitor is basically completely transparent if placed in front of a FET gate.

2) Properly drive all the parasitics of the tube grid. All tubes draw some amount of grid current much before grid going positive. For best results, this must be accounted for, and the way to accomodate it is not thru a capacitor.

A FET followers gate is a nonexistent load (provided there is enough voltage between gate and drain, easy to accomodate in a high voltage tube circuit), whereas a tubes grid is in fact not a nonexistent load, not at all.

Sonic has listened to this amp both with and without the grid drivers, I'll let him provide his own comments on that.

Very interesting design with the mosfet sandwich and transformer coupled output.

Thank you. My primary direct influences are Wavebourn, Tubelab-George and pmillett. Rod Coleman's work was also important for the directly heated version of this amp

What do you mean with MOSFET sandwich?

The transformers in my own and also Sonic's build are Hammond 125D. Smaller ones work just as well, like 125A. There is no real reason to use 125D or even 125E.
 
Jun 21, 2017 at 10:45 AM Post #6 of 189
Hi MrCurwen,

Thanks for your thoughts. I've also been reading up on TubeLab's A2 experiments and have seen Millett's similar CCS loaded mosfet implementation. Even if not driving the grids to A2, it sounds like your arrangement serves a similar purpose here: dealing with the impedance inconsistency due to grids drawing current. Of course presenting the high impedance to the previous stage in the process is another advantage. Stating for the benefit of others as I'm sure you know all this.

"MOSFET sandwich" is just referring to the CCS loaded MOSFETs between stages and at the output. Not using it pejoratively :)

Couple of questions since I'm curious and you're willing to discuss:

The power supply isn't shown; how are you biasing the MOSFETs following the LTP? I see trim pots in the pics but wondering if you figured out some other kind of arrangement (servo or otherwise).

Have you had the chance to compare the MOSFET approach to the old school cathode follower in a similar configuration? On paper a MOSFET wins in most ways, but would love to hear any thoughts.

Have you done any experimenting with "Super Triode" outputs? That's something I'd like to try in the not too distant future and judging by this design I imagine you might have looked into it. TubeCAD and JC Labs are both websites that touch on it pretty frequently but I don't see many others actually giving it a go.
 
Jun 21, 2017 at 11:38 AM Post #7 of 189
Thanks for your thoughts. I've also been reading up on TubeLab's A2 experiments and have seen Millett's similar CCS loaded mosfet implementation. Even if not driving the grids to A2, it sounds like your arrangement serves a similar purpose here: dealing with the impedance inconsistency due to grids drawing current. Of course presenting the high impedance to the previous stage in the process is another advantage. Stating for the benefit of others as I'm sure you know all this.

Yes. The grid drivers improve transient response, they make the sound (in my opinion) more 'fast' and dynamic. The technical reasons for this are (in my guesstimation) 70% relating to the grid and Miller capacitance and 30% related to the coupling capacitor. The follower deals with both of them in a brute force manner.

"MOSFET sandwich" is just referring to the CCS loaded MOSFETs between stages and at the output. Not using it pejoratively

No pejorative inferred. Hmm, most people are more puzzled with the LTP configuration at first glance; high impedances at both the tail and at the top.

The power supply isn't shown; how are you biasing the MOSFETs following the LTP? I see trim pots in the pics but wondering if you figured out some other kind of arrangement (servo or otherwise).

The output followers are biased with R1/R2 and R7/R8 voltage dividers to halfpoint between B+ and ground, minus 4.5V FET drop of course. This leaves the follower with maximum amount of symmetric voltage swing available. Bias current is determined by the bottom CCS.

The power supplies are very simple RCRC or similar (doesn't really matter) followed by a simple 2 FET regulator.

Have you had the chance to compare the MOSFET approach to the old school cathode follower in a similar configuration? On paper a MOSFET wins in most ways, but would love to hear any thoughts.

Not in a similar configuration. I have built some circuits with tube cathode followers, most notably tube plate loads (plate load is a follower). I am not pleased with using tubes in roles other than voltage gain. They can do ok, but not for best results in my opinion.

I used to do all tube regulators back in my tube purist ideological days. Crazy business.

Have you done any experimenting with "Super Triode" outputs? That's something I'd like to try in the not too distant future and judging by this design I imagine you might have looked into it. TubeCAD and JC Labs are both websites that touch on it pretty frequently but I don't see many others actually giving it a go.

I haven't built any Super Triode circuits. I've played around with them in sims some time ago.

Not a bad idea, but I don't think it would bring any benefits above my current strict roles (tubes do voltage amplification, FETs provide current). Super Triode is a FET or transistor that has a feedback loop around itself, and the tube is in the feedback loop. This way the tube 'lends' some characteristics to the FET.

Only upside would be efficiency, but as far as I'm concerned that boat sailed away a long time ago.

Serious problem would be providing a good load for the Super Triode. I guess it's supposed to be used in series feed mode, which is not for me as it requires expensive OT's and doesn't properly deal with OT parasitics.
 
Jun 21, 2017 at 2:56 PM Post #8 of 189
Thanks for the lengthy responses! All good info.

I'd encourage you to revisit the Super Triode topology when you get bored because I'd love to see what you come up with. I don't believe there is any reason they can't be used OTL, differential, or parafeed. With the lowered effective Rp and multiplied load impedance they should also be able to make use of lower turns ratio transformers (more power, less parasitics). The tube determines voltage amplification (still Mu) by way of the feedback scheme.

But if one thing is true in the hobby, it's that there are plenty of ways to skin a cat. Thanks again for sharing yours.
 
Jun 21, 2017 at 11:36 PM Post #9 of 189
So basically it has Mu follower driver stages with a parafeed output and some buffers in between. Overall pretty standard design but still pretty good. Chassis work is amazing. I gave up on P2P construction about a year ago..too many fried mosfets..too many power supply accidents. After a long search I've found solderless breadboards to be the bees knees for prototyping overall. I can build an amp in 30-60mins after conception, its great.
Personally I would ditch the parafeed, OTLs are easy to make with solid state no need to deal with transformer parasitics as a factor at all and you'd be able to ditch that second set of tubes in the output stage.
I'd change the RC ratio of the mu followers to something that allows a lower and higher quality cap to be used, such as 10meg/0.1uf. I'd also switch to a faster mosfet than the IRF830 for better HF response and better bandwidth. I'd also use a current balancing bias scheme to ensure the tubes run at matching currents for better H2 cancelation.

For the output stage I'd go with a lower distortion stage like a sandman or an aleph-esk design, or at the very least I'd cascode the source follower to make it much more linear. I prefer to know where my distortion is coming from, easier to tune that way. Those source followers will introduce considerable distortion under load, depending on the load. Not that increased distortion is necessarily a bad thing but in terms of amp design I think it's better to add distortion after the fact.

As far as interstage buffers go the standard current source fed source followers are fine, the load is negligible as long as the idle current is at a much higher ratio than the expected grid draw.
I'm curious about your findings on the interstage buffers, I'm not convinced the grid current is enough to make a difference in sound if the previous stage is drawing a reasonable current, especially when you're driving the next stage with a mu follower already.

You'd need quite a few microamps or miliamps to delinearize the preceeding stage depending on the current draw of that stage. The mu follower should keep the triode at a stable current, the mu follower output itself could become slightly delinearized depending on the ratio of current delta at its source pin but I don't see how a triode under standard operation could draw enough current to matter. Maybe I'm wrong and you could enlighten me? I've never considered grid current as a factor. I'd just assumed it drew a few microamps at most depending on grid capacitances.

In either case I'm a believer in results over theory, what is your impression on the sound with/without the buffers in place?
 
Last edited:
Jun 22, 2017 at 8:41 AM Post #10 of 189
Wew, nice build!
Thanks Sodacose!

Do you know what the reasoning for the grid drivers is? Are they driven A2 or just there to buffer the previous stage?
MrC got that covered much better than I ever could. I have merely been given his design and built it. When I started I couldn't even read a schematic, lol.

Sonic has listened to this amp both with and without the grid drivers, I'll let him provide his own comments on that.

The grid drivers improve transient response, they make the sound (in my opinion) more 'fast' and dynamic.
I agree, definitely a faster sound with more weight behind the tones. Very noticeable.

The power supply isn't shown;
I've updated first post with psu schems. Those are exact values that I've used.

The output followers are biased with R1/R2 and R7/R8 voltage dividers to halfpoint between B+ and ground, minus 4.5V FET drop of course.
And the grid driver followers are biased through the voltage reference at R1. It's just B+ -> 220k -> 10k trimmer -> ground. I have the grid drivers output 0VDC to output tube grids. The other trimpots, on top of the B+ decoupling caps, are to set the gyrator output voltage (anode voltage)

Chassis work is amazing.
Thanks!

I gave up on P2P construction about a year ago..too many fried mosfets..
I think P2P is great. I too have fried lots of FETs. It became a pain to change them since I glued them to the heatsink everytime. I now have drilled and tapped M3 holes in both heatsinks, grid driver and power buffer. If something should go wrong now its very easy to change them.

Personally I would ditch the parafeed, OTLs are easy to make...
That would make this an entirely different design. I already have an excellent sounding OTL amp as you know. The modded LD MK6. Great thing with having the tubes do only voltage amplification is that you can use any linear tube and it'll sound great.
As far as using better caps there's no need. I've tested the input stage with Mundorf S/G/O .68µf coupling caps. Absolutely no difference compared to the russian KBG's when followed by the grid drivers. I'd rather spend a few dollars on FET's than hundreds on audiophile caps.

I've built MrCurwens design without questioning it or try to change it. I am extremely happy I did that too. The price to performance ratio is very, very good. You can, if you don't care much about aesthetics like I do, build this amp for little money.

In either case I'm a believer in results over theory, what is your impression on the sound with/without the buffers in place?
I've listened both without the grid drivers and power buffer. Also with only power buffer and no grid drivers. I've already mentioned the grid drivers above. The power buffer sounds very clean. The sound is "muddier" without it (without sounding bad, still sounds good) I think I notice more of a difference with the use of power buffer than MrC does.
Anyway, the buffers do improve SQ by a significant amount.
 
Jun 22, 2017 at 10:06 AM Post #11 of 189
.
I already have an excellent sounding OTL amp as you know. The modded LD MK6. Great thing with having the tubes do only voltage amplification is that you can use any linear tube and it'll sound great.
Well, what I meant was simply removing the parafeed stages and using a good solid state output stage. The parafeed is effectively acting as a distortion generator in comparison, which isn't a bad thing at all in itself but personally I would forgo the mess of parasitics involved with transformers unless it offers a specific sound you like.

I don't agree with the statement that having tubes only amplify voltage makes them all the same, my experience has been far different and I've tested this a lot.

The power buffer sounds very clean. The sound is "muddier" without it
I've noticed this too when using solid state in place of tubes where they are supposed to drive current.

So far I've only built 2 stage amps though, I try to avoid more. Effectively speaking in comparison my "grid driver" simply drives the headphones instead of a bunch of extra stages. I usually direct couple the whole thing, no capacitors or transformers anywhere in the signal path, talk about clean sound.

Although my DAC blew up along with my laptop during the winter due to a mishap when my usb oscilloscope accidentally whacked against an exposed 350v power supply rail of a CLC filter I built which caused a chain reaction of destruction so I can only really compare my designs to each other now as amplifying my laptop headphone jack sucks in comparison to a real dac, there's this permanent "haze" in the sound.

That was the last time I ever touched P2P, that accident was a multi thousand dollar mistake right there in itself. Hindsight I should of spend a grand or 2 on lab power supplies and I would have saved twice the money in the long run and a whole lotta burnt components. (I've spent way to much time/money on trying to be cheap, ironically)
Only recently I discovered the ease and speed of simply using solderless breaboards, I guess the best solutions are often the simplest ones. What normally took me days now takes me less than an hour.

I will say that my designs wipe the floor with the modded MK6 when I compared, I swapped the input stage in my modded MK6 with my latest one and it improved the sound in every way possible by a very large degree. I did the same for the MK6 with a SS output stage I had at the time, it also wiped the floor with the MK6 output stage. Since then I've made those designs twice as good, I just need to get that darn dac fixed for once or buy a new one.
 
Jun 22, 2017 at 11:00 AM Post #13 of 189
But if one thing is true in the hobby, it's that there are plenty of ways to skin a cat. Thanks again for sharing yours.

That is indeed true!

It all depends on

1) the design goals and priorities that are set (they cannot be vague like "good sound" but specific in technical terms, like a certain output impedance or a certain voltage gain or a certain THD etc.)

2) the practical limitations; what kinds of components will be used, funds available etc.

3) the design principles and philosophies to be followed.

These all to an extent interact with each other in very complex ways.

My design has been arrived at during several years. I thought I was the first to come up with it, but later found out some others had made similar efforts. Specifically the LTP configuration, the grid stuff came from George.

I assure you it's quite well thought out, the individual parts complement each other. For example the output buffer; you can say the simple one FET follower with CCS bias is too weak and will produce distortion if loaded. Well, the load for it is quite low, it can handle it with no distortion. Confirmed by sims and listening. It's super clean, no distortion, either low order or high.

As well one could think that the OT being there at all would be "less than perfect" since a real coil will always behave a bit nastily. Well, it is driven by such a low impedance, that the nasties simply are not there. However if you were to take the OT out and direct drive your load, you would end up with a situation where the follower would indeed produce a lot of distortion and would need to be a lot more complex for good results.

Here you have two imperfect solutions ('weak' follower and an OT) that in effect cancel out each others weaknesses. The OT primary is such an easy load the weak follower is overkill for the job (even a tube can manage it!) and produces no distortion, and the imperfect OT is so 'overdriven' (you get my point) that it's parasitics disappear. It doesn't produce any distortion. The idea of OT's as distortion producers is a relic from ages past where you only had very compromised ways of driving the primary.

The follower doesn't need to be more stronger or more complex, it is overkill as it is. The OT doesn't need to be of higher quality because it is driven to transparency as it is. Going OTL would destroy this design balance and introduce a whole Pandora's box of problems on it's own.

Perfect can be an enemy of the practically (doably) very good.

Same thing with the grid drivers. On another more engineering oriented forum I get a few private messages every year from people who are interested in my direct coupled circuits from years ago (back when I was a direct coupled purist). I spend a lot of time explaining the reasoning behind how and why I get better results with proper grid drive than with DC.

DC is in the abstract 'better' but not in practise.

The price to performance ratio is very, very good. You can, if you don't care much about aesthetics like I do, build this amp for little money.

I'm very very very cheap. I find it a virtue not a vice. This circuit can be built much simpler while achieving 85% of it's (in my opinion quite exceptionally high) sound quality. Ditch all the PSU's except one for B+ and one for B-. It can even be arranged so there's no need for B-. Output buffer can be omitted. For an even simpler version use a SE input stage; this introduces a little bit of H2 but all the dynamics, all the punch is still there.

Very very cheap tubes can be used with sound just as good as an "audiophile tube". Just as Sonic said;

you can use any linear tube and it'll sound great.

Quite true. Emphasis on the linear there. But, many dirt cheap unknown tubes are super linear. One of my design philosophy rules is if you hear a noticeable difference between tube types in a voltage amplification context, your design is poorly thought out. Sorry rollers...

As far as using better caps there's no need. I've tested the input stage with Mundorf S/G/O .68µf coupling caps. Absolutely no difference compared to the russian KBG's when followed by the grid drivers. I'd rather spend a few dollars on FET's than hundreds on audiophile caps.

My thoughts exactly. The cap game is for audiophools in my opinion. Sorry rollers again...

I've listened both without the grid drivers and power buffer. Also with only power buffer and no grid drivers. I've already mentioned the grid drivers above. The power buffer sounds very clean. The sound is "muddier" without it (without sounding bad, still sounds good) I think I notice more of a difference with the use of power buffer than MrC does.

Yes. I think it's the bottleneck thing. Sonic has much better headphones and a much better source. I do hear a very noticeable difference with regards the output buffer, but I'm quite sure Sonic hears more, because it's more of an bottleneck for him than for me.

Also I'm at the point where I'm very lazy. I have a decent enough workhorse system (no chassis... don't do this at home guys) that the effort of getting my ultimate system builts is too much to actually start doing. I've been busy with studies and work and other hobbies.
 
Jun 22, 2017 at 11:19 AM Post #14 of 189
Very provocative thoughts on the 'weak' buffer driving the cheap OPT :) I was wondering the same thing as coinmaster, but you make a good point. Proof is in the listening, I think.

Have you tried any similar experiments allowing the MOSFETs to provide a fixed bias to the output tubes rather than the CCS bias? I'd be interested to know how it might compare in sound.
 
Jun 22, 2017 at 4:18 PM Post #15 of 189
Have you tried any similar experiments allowing the MOSFETs to provide a fixed bias to the output tubes rather than the CCS bias? I'd be interested to know how it might compare in sound.

That's the place where I started my balanced building originally. I then moved to a resistor tail and then a CCS tail. The CCS tail is an integral part of the way my LTPs are set up. Without the tail the signal balance (DC balance is not important within normal variation) needs to be achieved some other way. The tail is the easiest and most obvious way to achieve this balance.

And in a balanced amp the actual signal balance is one of the most important things.
 

Users who are viewing this thread

Back
Top