TGM8 - my best amplifier, incredible bass, clear highs, no fatigue (inspired by Rod Elliot P3a)

If I wanted to try two-pole compensation with the simple input stage...

R4 = 0R - correct
Q3 = unpopulated - correct
C4 = 22p - correct (empirical tuning maybe needed)
C6 = 330p - correct
R7 = unpopulated - needs to be installed, use between 1k and 4k7 (I used 4k7 in the end)
A wire link between Q3 collector and emitter pins - No!

Conversely, if I wanted to try the bootstrapped input stage with dominant pole compensation:

R4 = 1k - yes
Q3 = populated - yes
C4 = 68p - No
C6 = wire link - No
R7 = populated - Yes, 1k to 4k7 will do fine

For the dominant pole compensation cap you need to string a cap between the collector and base of Q4; a through hole cap tacked on the underside of the pcb will do it - value 68p to 100p sounds right.

It is possible to use nested feedback with either dominant pole or 2-pole compensation schemes.

Yes. It provides local feedback from the VAS, which has a few benefits and side effects - one of them though is to reduce the feedback around the output stage and with the simple input stage you have less gain to share around.

I thought you might be interested to know that I've experienced none of the output transistor overheating issues on my boards with only the BJTs populated.

I believe it's caused by the presence of the FET gate capacitance. In simulations there are two options to address it, a) increase the gate-stopper resistors R25&R26 to around 470R to 560R, b) replace the FETs with another pair of BJTs.

Option b) is interesting, it creates an all-bjt output which looks smoother on the simulations. The downside is that this doesn't protect the drivers from getting hot if driving low impedance loads because the drivers then have to source more base current for two output bipolars. However, it is viable and may just need a small DIY heatsink on the drivers to make it work.

Another option I thought was interesting, it's not related to controlling cross-conduction but it's an option I considered at the start (simulated but not built). The idea is to use FET drivers to better shield the VAS from the non-linear impedance of the CFP output stage more effectively than the bipolar drivers. You would need to squeeze in TO-220 packaged drivers but I believe they will fit (pinout will be reversed).
 
Last edited:
More progress....

I went back and did some more simulations and more reading. I have brought the cross-conduction under control. I reduced the FET zobel capacitors C13&C14, to 47pF (I wouldh've used 33pF if I could have found them). I had to lift up one end of R31 to access C14!
Next, I increased the FET gate stoppers R25&R26, to 274R (200 to 300 will be fine).
The more significant factor is reducing the driver load resistors R20&R23 to 100R. This doubles the driver current to over 5mA (much better). In fact, if you want to heatsink the drivers you can reduce these resistors further to 56R and push the current higher - the cross-conduction in the simulation is then even lower than the P3A. I felt chicken so I went with the 100R load resistors.
I also made some adjustments to the compensation to reduce triangulation of the -ve waveform at very high frequencies (based on simulation only). I reduced C6 to 100pF and increased C5 to 33pF. I left C4 at 22pF and C10&11 at 100pF.

Powered up fine, green LED. The bias had to be re-adjusted for the new operating points to give me around 19mV across the 0R33 resistors. I noticed noise levels were lower today, only around 5mV p-p random fuzz on the output.
I ran the sinewave test again (into 7R8 load)and noted the rail current draw changing with frequency - the attached plot shows the new version (red) with much lower current than yesterday (blue dotted line). According to my simulations this is now on-par with the P3A. At 200kHz I noticed that the zobel was getting hot, but cooled immediately when I backed off on the frequency. Above 500kHz the rail current was not stable - started to increase while I watched the meter so I terminated the test.
At 1kHz the average rail current was 400mA with 19V p-p into 7R8 (simulated into 8R it should be 396mA). The power devices & heatsink got pretty warm doing this.
1kHz and 10kHz square waves were clean into 7R8 load. No h.f. parastic oscillations visible.
 

Attachments

  • TGM8improvedCross-conduction.jpg
    TGM8improvedCross-conduction.jpg
    36.8 KB · Views: 825
Listening test - I was able to reconfirm all sounds fine. The bass extension is very good. Playing music it hardly got warm.

The next step is probably to optimize compensation (C14) for the best subjective sound (mainly in the treble) but this can't happen until I have a power supply, volume control and some time for the amp to burn-in; this will take a little while.
 
Member
Joined 2002
Paid Member
I found two issues:
1) the silk screen is wrong, 'offset' and 'bias' which were printed near VR2 and VR1 to make it easier to set it up are transposed. Dissapointing to make such a silly error.
2) the circuit design is wrong with regards the LED, preventing it glowing green when the SS relay is 'closed'. I missed it in the simulation because I used a regular LED in place of the Opto. The opto has a much smaller voltage drop - it's in the datasheet but I missed it. As a result, there isn't enough voltage to light up the green LED. I am exploring 3 possible options to fix this.

Hi Bigun,

Just in case you do another PCB....

3) Elliott has 2 t's

regards
 
Hi again

I've been a bit busy with NYE and other stuff the past couple of days. Even though I haven't had time to spend in the workshop, I've still been following your posts with interest.

Looks like you're getting a handle on the cross-conduction issue. While you are playing with that, I've decided to systematically retrofit Q3, the two-pole VAS, and some of the other component value substations, and evaluate the sonic differences at each stage.

I'm still a little ambivelant about whether the use of the second pair of outputs (the MOSFETs) in this configuration. I've been reading Cordell and Self, both of whom make the case for parallel outputs; not so much to increase peak output but to reduce the effects of beta droop at lower output levels also. As I understand it, the MOSFETs only "kick in" once the BJTs are conducting quite a lot of current. I wonder what the distortion profile looks like around this cutoff point.

Rod has shown with P68 that the CFP stage can be stabilised with as many as five pairs of outputs. I've built this amplifier and it sounds pretty good plugged into a full-range loudspeaker, despite the fact it was designed as a subwoofer amp, not really suited to hi-fi use. But others who have built variants of the P3A with two pairs seem to indicate the sonic qualities aren't really very good.

Alternatively, perhaps a pair of lateral MOSFETS as shown in Rod's P101. The drivers would be eliminated to make the board even more compact.

Perhaps a pair of BJTs in a darlington configuration. Rod would probably cringe, and it certainly wouldn't be a P3A anymore. However, I would say that its already been established that your Singleton input stage design sounds very, very good in its own right, and its probably worth experimenting with it coupled to different output stages.
 
Hope you had a good New Year and everything !

I consider the cross-conduction issue as solved now. I learned a lot more about it. With multiple outputs and single driver it gets worse - maybe explains some design choices made by others. What I want to do is listen to the amp in more detail but for that I think it will be easier to add the power supply and put it in a box. Listening will allow some fine tuning of C4 and R8.

For me, the priority was to make the P3a output stage. This means a BJT driver and BJT output device, single pair. I'm not claiming it to be the best and I would encourage you to experiment. The CFP driver and emitter-follower output I used in TGM3 and TGM5 might be worth your time too.

Look at the attached excerpt from the datasheet for the 5200 device. The beta is the same at 5A as it is at 10mA; if you use the MOSFETs the BJTs never see 5A so no beta droop, ever. With the new resistor values (posted above) the FETs kick-in at lower current now, around 1.8A in simulations (with no adverse affect on distortion). They also reduce the current draw through the drivers, which stay cooler; the amplifier is thermally stable. And no device matching is required even though I have two pairs.
 

Attachments

  • 5200 beta.jpg
    5200 beta.jpg
    45.8 KB · Views: 768
Last edited:
Happy NY!!

OK that report is starting to fill me with confidence. I think I'll place another order with Digikey and order some MOSFETS (and other bits and pieces) and try it for myself. The output stage is a mongrel with its BJTs and MOSFETS, but if it sounds good then that's what's important.
 
One thing, I'm not yet sure that the simple CRC filter that was included on the pcb for the power supply is 'good enough' to just hook up to a rectifier. I suspect there will be some gains to be made from finessing the power supply further - such as adding snubbers across the rails.

A cap per rail bolted to the chasis near the rectifier might help to keep the worse of the charging pulses out of the amplifier board and grounding. I know this may be more complexity than you were planning but if you are ordering parts you may want to get some bits to play with when building the power supply - all part of the fun !

Check this out (look up the DX PSU) the snubber is C1&R3 plus C2&R4 in Carlos' schematic and they help damp out h.f. resonances after the rectifier: http://users.tpg.com.au/users/gerskine/dxamp/default.htm
 
Last edited:
Hi again

I spent a few hours in the workshop today on a blistering 37C day (41C tomorrow!) to retrofit Q3 and the two pole compensation on my second board. I think we are getting into the realm of very subtle differences, because I couldn't detect much - if any - difference with the other board using the 'simple' scheme. I suspect The addition of Q3 helped slightly clean up some of the high frequency artefacts I spoke of earlier (more on that in a second). The two pole compensation? Sounds very similar (dare I say identical?) to the single pole comp once I changed the Miller cap to 68p. I will caveat all of that by saying a) all listing tests conducted in my garage with poor acoustics; b) it was hot and uncomfortable, and hence; c) I didn't spend a lot of time doing A-B comparisons.

Ok now here is the interesting part. I decided to swap in one of PMI's cap multiplier boards in place of the simple 4700u "cap per rail" (no resisters, no snubbers, nothing...). The difference was profound - the hiss was gone and so too all my complaints about the high frequency artefacts. Interestingly, I've used the simple bulk cap arrangement with a vanilla P3A to good effect, leading me to conclude that the TGM8 has worse PSRR.

Gareth, I think we are of a similar mindset regarding the PSU. I've been concerned about having the first line of caps onboard with dirty power from the rectifiers in close proximity to the front end. But after today's testing I suspect TGM8 will place more stringent requirements on the PSU than the P3A (and what can be accommodated onboard). A cap multiplier or at least high capacitance CRC/CLC supply will be required for good results. I can't explain why - this is just my gut feeling. It will be worth the effort - to me this amp gives a very lively, dynamic performance at all power levels. The standard P3A, to my ears, while very detailed and refined, sounds a little "thin", which in the past led me to think it was underpowered with its single output pair.

In my view this amp has gone from "very good" to "superb" and plan to move it into my living room to test it with my main loudspeakers and a wider range of material. I'm already thing about a suitable chassis (to keep the wife happy) because I think it's going to replace my P3A in the living room
 
Happy NY to all forum members reading this and other threads!

I must say that even P3A sounds better with MrEvil/PMI cap multiplier. Much cleaner highs and more detail and ambience.

The difference in sound comparing with P3A is due to different topology. Voltage feedback always sound a bit thinner and brighter, it was not that P3A is underpowered. Current feedback sounds more musical.
 
Happy NY to you too!

Well guys, very interesting observations. I'm prepared to do some more work on this amp if you feel it has the potential. Interesting that you didn't get much from the two-pole comp - it should have most impact only at very high frequencies and perhaps this isn't very audible until the power supply is perfected or until you push the amp to high powers. Or it may not add much benefit - I'm happy either way since it's all good learning!

Did you install C17 & C19 ? - I didn't in my case. The idea was to provide a high frequency by-pass to C15 & C18 but I've read conflicting reports about doing this - worse case you can introduce some resonance into the power supply at high frequencies and this could affect the sound. By using the XR7 type this is supposed to be mitigated somewhat.

The PSRR of the TGM8 looks pretty good in simulations but it does depend on some details - such as the sizes of R18 & C8 which I do believe are under-valued. I ended up increasing R18 to 560R in my build because of this which moves the corner frequency down lower. This improves PSRR at low frequencies though, not in the treble.

Funny you should mention the Mr Evil cap multiplier though as I've been looking at this recently for use with another of my amplifiers. I found another design that is very similar by Pete Schumaker. I am thinking of designing my own version with some protection built in.


How should we proceed - do we leave the amp alone and work on a better power supply or are you suggesting we change the TGM8 pcb ?
 
Last edited:
Gareth, this is a fabulous amp; one of the best. I love the sense of energy and the dynamic performance, which I feel is somewhat lacking in the basic P3A. But it loses none of the P3A's best qualities: that beautifully smooth and detailed midrange, and crystal clear highs.

I wouldn't read too much into my comments yesterday; once I get it into the living room my perceptions will probably change. I'm no golden ear either. What I perceive as a negligable improvement or minor nuance, others would regard as significant.

Bear in mind also that I intentionally used basic, argricultural grade components almost everywhere. No 'audio grade' capacitors - aside from the Muse input cap - just Nichicon HE in most locations. C8, C9, C15 & C18 are only 100uF (per my selections in the BOM). No onboard bulk caps and just a single off board 4700uF can per rail.

I didn't install C17/C19 but later tacked some X7R 100n TH caps on my second board and performed some A-B comparisons. Conclusion? Couldn't hear any difference. But I left them there anyway.

I'm not suggesting any changes to the circuit design or board layout at the moment. I have some 22mm snap in caps coming from Digikey, some 1000uF caps that will fit in the same through hole locations, the MOSFETs, and a range of other SMT component values for experimentation.

I plan to focus my efforts on the PSU next, using the cap multiplier sitting on my workbench as the baseline.

I'm really hoping you can find time to build the "basic" version, without Q3, with just the single pole compensation. Start with 68p for the miller cap and 22p for lead comp. You might be able to go lower but I dared not without a 'scope. It is not difficult to make the board changes to add your singleton "bootstrap" afterwards, to evaluate the sound of that version without the 2-pole comp.
 
Gareth, this is a fabulous amp; one of the best.
Thanks for the encouragement.
I wouldn't read too much into my comments yesterday
I understand that initial impressions will change over time, but it's very important you don't worry about 'hurting my feelings' - we will only be able to make this amp great if we are as honest as possible :)
I'm not suggesting any changes to the circuit design or board layout at the moment...I plan to focus my efforts on the PSU next, using the cap multiplier sitting on my workbench as the baseline.
Sounds like a good plan to me - get a good PSU in place then go back and tweak the amp. You are well ahead of me as you have a Mr. Evil boardset whereas I will likely design my own because I need something extra for use with another amplifier - but I'm on it !

I'm really hoping you can find time to build the "basic" version, without Q3
me too, but I feel that I need to get on with the psu for now...
 
Last edited:
Oh don't worry if something sounds rubbish I'll be the first to say so :D

I don't know why you're surprised that I'm ahead of you with the cap multiplier... I've been ahead every step of the way ;) :snail:

Here's an idea: if you are going to lay out a cap multiplier, perhaps you could incorporate optional solid state speaker protection, provided there is adequate room within your 10x10cm limitation. The opto-isolator is the single most expensive part and an "n-channel" circuit only requires one isolator. Some other parts economies could be realised in the detection circuit also.
 
Yes, you are doing very well keeping ahead - please keep that up as it's very helpful !

Funny you should mention dc-protection, my other project has no on-board dc-protection and I do want to add it. But after building TGM8 and having the benefit of it on-board I do prefer what has been done with TGM8, the cost is trivial when I consider other factors (of course for a commercial product I might feel differently).

A potential challenge with doing the protection at the supply is you don't get delayed turn-on and fast turn-off of the speaker and so you have to rely on the amplifier behaving well even if you ramp up the supply rails slowly. And when done at the power supply it needs to be latched since once you cut off power you also cut off the dc error. And if you have capacitors of significant size on the amplifier pcb they still get to dump all their energy into the speaker. All things considered - I like having it on the amplifier pcb. My other project has no caps on board so a psu based system makes sense.
 
Last edited:
Have you entered the basic P3A schematic into your programme to compare PSRR between the two designs? Are the results the same or similar?

Yes, but I only have the simple schematic Rod posted on his website and I don't know if this reflects what he sells when he provides the board. The PSRR of his circuit is considerably worse. Why ? - he has no RC filter on the +ve supply rail for the front end and he has no decoupling cap on his current source for the LTP. Both of these are standard precautions that he would be aware of so you'll have to look at your P3a board and tell me if it deviates from the schematic he has published (but don't post the details on this forum as I assume they are his IP, just say if it's the same or not).

One thing he doesn't have is a zener diode, which could intrude some noise into the amplifier. I've simulated what would happen by placing a dummy signal source in series with the zener and looking at the output. I find that there is good rejection. However, the ear is strangely sensitive so we could try and put a capacitor in parallel with the zener, 10uF would be enough to hear a difference in the treble (and 1206 sized 10uF 50V multi-layer ceramic caps are available and could be soldered alongside the zener if we found it was worthwhile).
 
Last edited:
Yes, you're right, it makes more sense to put it on the amplifier board. It also keeps wiring to a minimum and eliminates the possibility of PSU induced noise in the speaker outputs.

Its a shame that people have been put off because of the SMT devices. Before TGM8 I was kinda on the fence: neither for nor against but somewhat favouring TH for convenience sake. Now I can honestly say I prefer working with 1206/1812 devices.

I can accurately position and solder a 1206 resistor quicker than I can align, bend and solder a TH equivalent. Desoldering a SMT device I find *much* easier: simply remove the excess solder with solder braid, and then gently remove the SMT component by applying upward pressure with a pair of tweezers while applying heat on one pad with the iron. Two irons not required. Getting all the solder out of the through holes is a pain and I usually resort to snipping the leads off at the component and pulling them through individually.