The AMNESIS amp: a good amplifier, like a gentleman, has no memory.

Hi Max,


Before I show you the Thermal Effects at 60 Hz, I have first of all cleaned up your circuit diagram by removing resistors and caps without an obvious function.
For instance placing caps on a power supply line from a voltage source makes no sense, because output impedance is 0 Ohm from zero Hz to infinity.

I have already explained the changes in the Vas, with an emitter follower and just one dominant cap, in this case 50 pF.
Your common base solution can only sink an not source current, giving the funny square wave response as shown in my previous posting.

Furthermore have I removed the 100pF cap in the input pair's tail current plus both 1 K resistors.
This results in a huge 33dB drop in IM distortion at 1KHz, with a 19+20 Khz 1:1 test signal.

In the output section I have altered the 1K values around the Mosfets into 2k to improve stability. To prevent the Mosfets to be cutoff when the other one is conducting, resistors have been added to always allow a minimal amount of current.

Last but not least, I have put all models in Amnesis.lib.
In the .asc file you have to add ".include amnesis.lib" after having copied this file into the LTSpiceIV LIB directory with your the file handling system of your computer.
The .asc file already has the square wave generator installed.


Hans
 

Attachments

  • New Amnesis 3.jpg
    New Amnesis 3.jpg
    427.3 KB · Views: 345
  • AMNESIS QUAD BOOTSTRAP BUFFER TEST2.asc
    16.2 KB · Views: 48
  • Amnesis.zip
    2.9 KB · Views: 57
I wish you all read Dr. Kolimunni's book as it analizes all this in detail and proposes some original input arrangements for his own circuits that are far more complex than ours.
M.
Max,

I have read Kolinummi's interesting book but when simulating some circuits in his book, I found large deviations from statements he made, pushing you in the wrong direction.
Showing some deviations in a comprehensive explanation step by step , never resulted in any reaction from his side.


The morale is: when properly done, Spice simulations do deserve the credit to validate with great technical confidence, so don't take everything for granted.
But at the end, only your ears can make the final decision.
Stan Curtis, former head of Cambridge Audio once mentioned: give three designers the same circuit diagram and let them design their own PCB.
The result will be three amps sounding differently. :D


Hans
 
Dear Max,


Since I wanted to know for once and for all the electrical benefits of a Szkilai input pair, I invested quite some time in looking at the thing from different perspectives.
I monitored the power absorbed by Q3, the input transistor, by multiplying Vce * Ie.
Assuming a time constant for a small SOT transistor, a low pass filter with an RC time of 0.5 sec was used to filter this power signal.


To start with I simulated with CFP and without CFP in fig 1 and 2 below.
This resulted in a big difference in heating Q3.
However, when changing R7, the current source resistor, from 100 Ohm just by 0.1% into 100.1 Ohm, fig 3 appeared, showing that this 100 Ohm resistor was an extremely lucky shot, not possible to realise in practice with such an high sensitivity.
When feeding the Non CFP version with an ideal current source, fig 4 appeared, proving the current source is causing trouble.
This is because the VAS is modulating the power rail, feeding the current source with a fluctuating supply.
That's why I modified the current source being almost independent of supply voltage.
See second image below showing the original next to the modified version.


As a next step simulations were made for 100 Hz, 1 Khz and 10 Khz, with and without CFP, but in all cases with Q3 carrying a current of slightly below 1 mA.
Results are visible in fig 5 to fig 10, all speaking for themselves.


To conclude: the slow changing temp curve results in a subsonic Vbe change that is completely harmless, but as can be seen, the Non CFP version is doing even better as the CFP version in that respect.


If a transistor could follow in heating up and cooling down a 100Hz signal, which of course is absolutely not the case with a 0.5sec time constant, but even if this were true, I showed in posting 273 and 276 that the resulting heating up for a 0.1Veff input signal this would result in Vbe change of less than 1pV, or some 80dB below the input noise of Q3.


As can be seen in the Figures, after 0.5 sec RC filtering, the peak to peak values are even much lower, making the Vbe changes into the femto region, drowning completely into the input noise.
When audible differences are noticed with a Szkilai input pair, I don't see how this can anything having to do with thermal effects.

Hans
 

Attachments

  • Total.jpg
    Total.jpg
    433.3 KB · Views: 295
  • Current.jpg
    Current.jpg
    89 KB · Views: 315
  • Total2.jpg
    Total2.jpg
    590.4 KB · Views: 276
Guys, thank you very much for your contributions. For some reason I didn't get the notification...now I need time to digest these informations that are probably beyond my present ability to cope with :eek:

So I must test and revert to my first arrangement for the cascoded input CCS.
I forgot to test ferrites at the gates.
No need for extra gate/base resistors then. I forgot that the extra base resistors were only to test RC decoupling to ground for oscillation taming. Now the C are not needed.

As I said before, I prefer the sound of the bootstrapped-CFP input. What is the cause? I don't know. The late Hephaïstos performed single impulse tests on transistors to prove TMD and gave much importance to complex signal behavior in contrast to periodic signal...

Meanwhile, take a look at this:

Hybrid Cascodes Simplify SiC Adoption in Popular Power Circuits | Electronic Design

I hope we can find a stable potential circuit to build a nice PCB where we can experiment for ourselves the pros and cons of this little monster, step by step if you want. :D :Pirate: Many heads think better than one...

With a totally stable circuit I will test my VFET bootstrapped output. :cool:

Cheers,
M.
 
Last edited:
Dear Hans Polak,

Thank you very much again for your teachings, especially on how to simulate the square waves. That would have saved me some time.

So, I did many experiments as per your recommendations. I wired your CCS input, which is even simpler than that of the Blame's that I copied. I did retain the C from the Vref of the cascode to PS (sort of a bootstrap) as this way it sounded rounder to me. My other Amnesis (now on rest) is also wired this way.
I also got rid of the 1K base resistors for the driver's cascodes and used some ferrites there, but on one of the modules the negative side cascode did not like it (middle level, +/-2V; VHF oscillation) so I retained the 1K and did not explore further.

The mods increased clarity a bit. I think making two separate Vrefs for the cascoded CCS was a bad idea. Unfortunately, none of these mods make the periodic oscillation disappear. :(

Next I got rid of the PNP buffer (pity, it sounded so good) for the current output of input section, the one causing the square wave problem and I replaced it with the dreaded emitter follower mod for the VAS, whit BC546 which I had on hand. It sounded nice, open and detailed, I would say even more open than with the buffer. But my suspicion of instability, from a too long "high current" phase on power up, with protection lamp, was confirmed when powering up with no protection: instantaneous kaput of the positive side cascode (a BD139) for the driver. As before, the emitter follower VAS is potentially the better technical solution, but risky. I re-tried with 1K from collector to ground but it also oscillated...so, now it is wired as the original simple cascoded common emitter VAS which sounds a little duller by comparison at first, less detailed and open, but with nicer midrange/midbass, with more flesh to it, sweeter and relaxed. The simulated square wave test is OK too. I did not perform real square wave: I forgot. Even the buffer should have been tested, sorry.

So, the VAS is the critical section and it takes a more clever man than me to perfect it. As was planned, an emitter follower VAS option is needed on a future PCB.
The VAS-driver interaction is probably the one causing the periodic oscillation, which is still there, I'm afraid.

It is astonishing how sometimes a little mod can totally change the flavor of the sound of the amp. Maybe we choose our preferred distortion...

About Miller cap, as I understand a too low value will make it more easily charged but will reduce global safety margins and vice versa. I kept 100p for now but could use lower values in the next experiments. Other option is increasing current for the diff. amp, which was not successful on prior experiments: reducer CCS resistor from 100 to 75Ohm or so.

I did not had the time to address output section. Maybe its time to reevaluate single bootstrap for driver and output.

Cheers,
M.
 
Dear Max,


Thanks for your feedback.
I have a few questions.

1) You are referring to periodic oscillations.What is their period and what is roughly the HF frequency.
In general, this sound like a wiring problem where ground connections are not made properly. Do you have a groundplane on your PCB ?

2) The other question that I have are the resistors between the sources of the two bootstrap mosfets at output.
These resistors turned out to be very vital for stability. Did you insert them, see R52 an R58 in my circuit, but this can be just as well one single resistor.

3) You are mentioning a BD139 output resistor, that was not in your LTSpice model.
What else is differing from the model, because all this can make a big difference in stability.


The miller cap is there for stability. When a larger version does not cure oscillations, the cause is most likely to be found in the wiring.
The only secondary effect this cap has is on slew rate. But that is a minor problem at this moment.

When having an update of the current situation, preferably in an .asc file, I can redo my simulations to see what can be possibly wrong.


Hans
 
Dear Hans,

I am sure my prototype has a big influence on bad outcomes, but some iterations were very stable at first. I was too lazy to transport everything to a proper case with +/-44V supplies.

1) No ground plane, save at front of PCB. I can post the pics of my cuadratic PCB just to have a notion of how bad it is :D.

Here's a pic of what the periodic oscillation looks like. The scope setting is 1uS and 50mV from memory. The freq. between bursts is +/-200KHz and the base line was flat but now also oscillates, but I think this may be a true fluctuation of the base line (from bad grounding) and not a proper oscillation. The bursts themselves have around 5MHz from what I see.

2) I did not had the time to experiment with the 1K+1K resistors. Maybe in a couple of days.

3) No extra resistor on the driver's cascode safe the emitter to collector between BJTs. I'll post a new .asc file, which will certainly have new numbers on it. Since I don't have the models for all the BJTs, I use what I have on hand. Medium power BJTs are all combinations of BD139/140 and 2SC4793/2SA1837 in the real amp.

Edit: I updated the .asc file to what more or less was my last (sort of stable) experiment.
Cheers,
M.
 

Attachments

  • Oscillation 10R100nF.jpg
    Oscillation 10R100nF.jpg
    892 KB · Views: 184
  • AMNESIS LED BUILD.jpg
    AMNESIS LED BUILD.jpg
    640.9 KB · Views: 89
  • AMNESIS QUAD BOOTSTRAP BUFFER TEST2.asc
    15.8 KB · Views: 84
Last edited:
I noticed that you have changed the current mirror below the input pair from a double into a single version, making the extra emitter follower abundant.
This improves stability.


On the other hand, you haven't added the two 1K resistors between the Sources of the bootstrap Mosfets, which is very bad for stability.
From a lesser degree, but also important are the 470 Ohm Gate resistors.
Removing them improves stability, at least with the given Mosfets, but I do not know what Mosfet's you are really using, so it depends.
Last but not least, R47 of 330 Ohm does not improve stability and R32, now being 20K should be of a lower value like 5K6 to let more current than just 0.1mA flow through the zener.


See the Images below with Green for the collector voltage of Q13 plus in Blue the output voltage scaled down and shifted to have both in one graph.
The upper two images are for Miller Cap C10 = 100pF, left with your .asc file and right with the modifications above.
To show this even more pronounced, the two images below are with C10 = 30pF.
Left is a nightmare and the right version with all the modifications is still stable.

It is hard commenting on your scope image. This could be caused by a ground loop or by an oscillating Mosfet.
If with the above mods, especially the two 1K resistors, you still experience oscillations, try to shortcut the drain and source of both mosfets to switch them off.
If oscillations are still there after this step, it is most likely to be found in the wiring.


Hans
 

Attachments

  • 2x1k.jpg
    2x1k.jpg
    599.3 KB · Views: 86
Dear Hans,

As I insinuated, all your recommendations will be explored, in due time.

The 1k+1k mod is next. And without gate resistors...and probably with ferrites also.

The scope oscillation bursts persist even with MOSFETs (which are exactly those on the schematic BTW) disconnected. I checked this long time ago.

The modified Wilson current mirror (as seems to be the name) was added to the simulation because it caused THD to go a little down, without affecting safety margins...I never had the time to implement it in the real amp. I understand your opinion is that simple BJT mirror is better for stability(?) IF so, maybe simple resistor is still better?

I will try lower values for Miller cap, but wouldn't this jeopardize whole amp stability?

Thanks,
M.
 
Dear Hans Polak,

I forgot to thank you for returning me to the common sense path in my workings. :D

See, the amp was much simpler a while ago, but, after experiencing bursts of oscillations at power peaks (not the one I'm seeing now) I went into adding resistors and caps everywhere, which complicated things a bit. Talking about which, after doing the 5K6 mod on input's CCS and adding the S to S 2K, I removed the gate stopper resistors for the MOSFETs. With these mods the sound regained clarity and HF detail as my original iterations, in a very significant way. BUT, one of the channels showed signs of potential oscillation at power peaks in the form of a widening of the trace. I added 10R gate stopper resistors and that prevented the symptom even during abuse.

The amount of inner detail is evident: I am listening to a CD called "violini di Cremona" or something alike, with Salvatore Accardo playing several famous historical violins (including "il Cremonese") and the timbral and texturial differences are notorious between violins. :cool:

I did not had the time to swap the 1K resistors from the Mosfet's Vref to the recommended 2K, nor did I increased the same for the driver's cascodes to 3K or 3K3.

Anyway, making a little housekeeping and making things simpler is working so far. Later I will try to update the present state of the machine.

I add a pic about the mods where you can see the big 2K between sources, and a lot of spare parts that came and go during these experiments. :D

Cheers,
M.
 

Attachments

  • AMNESIS 2K SS.jpg
    AMNESIS 2K SS.jpg
    427.4 KB · Views: 177
Oscillations

Hi Max,

some months ago I made one PCB for the Blameless Amp including the input section of the Amnesis Amp. In the mean time I cascoded the driver and the output transistors with the 8.2V zener diodes but no split resistors for bootstrapping (short air wiring).
The amp is supplied with +-32V from a labratory supply (analog no switch mode). The amp works stable without heavy oscillations (no listening tests).
Yesterday I tried to reproduce the small oscillations you mentioned.

I connected the scope (without probe for input sensivity of 5mV/scale.) via a shielded cable to the speaker output of the amp.
It took me some time to find a trigger to catch a burst signal which is similiar to yours. I tried many actions to get rid of this burst but without success.
What makes me wonder is that the same burst signal is still present if I disconnect the Lab supply from the amp (Ecaps unloaded). This means at least in my case that the burst signal could not be generated by the amp. I guess it is induced into the amp from outside or is a matter of the measuring setup.
Can you check if the burst oscillation is really generated by your amp.
 

Attachments

  • Burst.jpg
    Burst.jpg
    312.9 KB · Views: 178
Dear Hans Polak,

I forgot to thank you for returning me to the common sense path in my workings. :D

See, the amp was much simpler a while ago, but, after experiencing bursts of oscillations at power peaks (not the one I'm seeing now) I went into adding resistors and caps everywhere, which complicated things a bit. Talking about which, after doing the 5K6 mod on input's CCS and adding the S to S 2K, I removed the gate stopper resistors for the MOSFETs. With these mods the sound regained clarity and HF detail as my original iterations, in a very significant way. BUT, one of the channels showed signs of potential oscillation at power peaks in the form of a widening of the trace. I added 10R gate stopper resistors and that prevented the symptom even during abuse.

The amount of inner detail is evident: I am listening to a CD called "violini di Cremona" or something alike, with Salvatore Accardo playing several famous historical violins (including "il Cremonese") and the timbral and texturial differences are notorious between violins. :cool:

I did not had the time to swap the 1K resistors from the Mosfet's Vref to the recommended 2K, nor did I increased the same for the driver's cascodes to 3K or 3K3.

Anyway, making a little housekeeping and making things simpler is working so far. Later I will try to update the present state of the machine.

I add a pic about the mods where you can see the big 2K between sources, and a lot of spare parts that came and go during these experiments. :D

Cheers,
M.
Dear Max,


Glad to hear you made such a step forward.
Forget the four 2K Mosfet resistors and keep them at all at the current 1K, that's just fine.
And the tip from JOSI1 also seems quite interesting.


Hans
 
could this help?

Hi Max!
I have one more thought,
if you got sq improvement after cascoding output EF transistors then it means it was not due to minimise the memdist (because EF has very snall memdist) but due to linearisation or reducing juncion capacitances or other facor which we overlooked.
BTW attached is stable (not practical tested) circuit which might be helpfull in your efforts,

.MODEL BF245B NJF
+ VTO = -2.3085E+000
+ BETA = 1.09045E-003
+ LAMBDA = 2.31754E-002
+ RD = 7.77648E+000
+ RS = 7.77648E+000
+ IS = 2.59121E-016
+ CGS = 2.00000E-012
+ CGD = 2.20000E-012
+ PB = 9.91494E-001
+ FC = 5.00000E-001

cheers
 

Attachments

  • memdist.asc
    11.5 KB · Views: 40
Guys,

Thanks very much for your encouragement.

Pada, I always stumble upon the same obstacle, being spice-limited, I lack the package cordell-models :(

JOSI1, that is great news. I also considered an interaction between my second hand scope and the SMPS. I bought new probes to no benefit. Maybe I have to open the scope and check the PS and regulators. I also noticed that this burst pervades ground and input...maybe it is just spurious and does not pertain to the amp itself which is demonstrable stable, at least with my 16Ohm test speakers. Today I will make some test to discard the amp as origin of the troubles.

I have to find the time to make the big box transplant: that one has a good quality linear power supply with "charge-transfer" circuit. It used to power a quad UCD100/400 amp.

I attach the future amp with higher supplies Vs for the next project.

Hans, would it be beneficial if I place a resistor (R52) between both emitters for the BJT cascodes for the drivers, as we did to the MOSFETs??? Or will it be irrelevant? I will try to get rid of those emitter-collector little resistors (R8-R28) but the last time I tried the oscillation bursts at-power-peaks reappeared...

Guys, please opine about bandwidth, stability margins and slew rate if you can. I decreased rise/fall times to 0.1uS for the square wave test. I don't know if there is any standard...

Thanks very much again.
M.
 

Attachments

  • AMNESIS QUAD BOOTSTRAP TEST4.png
    AMNESIS QUAD BOOTSTRAP TEST4.png
    54.8 KB · Views: 184
  • AMNESIS QUAD BOOTSTRAP TEST4.asc
    15.9 KB · Views: 37
How stupid can I be??? :eek: :D

Attached are pics of the SMPS power supply (which has extra capacitance that does not filter this) without the amp connected. 20mV-1uS.

The second one shows the extra power supply for the speaker protection circuit.

This does not goes away with a sort of Faraday cage...

I have to test other PS though.

Something bad occured to the SMPS at some point, I imagine, since I don't recall that being there...any ideas to repair it?

Sorry to bother...
 

Attachments

  • POWER SUPPLY OSC.jpg
    POWER SUPPLY OSC.jpg
    562.2 KB · Views: 141
  • POWER SUPPLY OSC2.jpg
    POWER SUPPLY OSC2.jpg
    510.6 KB · Views: 134
Guys,

Thanks very much for your encouragement.

Pada, I always stumble upon the same obstacle, being spice-limited, I lack the package cordell-models :(

JOSI1, that is great news. I also considered an interaction between my second hand scope and the SMPS. I bought new probes to no benefit. Maybe I have to open the scope and check the PS and regulators. I also noticed that this burst pervades ground and input...maybe it is just spurious and does not pertain to the amp itself which is demonstrable stable, at least with my 16Ohm test speakers. Today I will make some test to discard the amp as origin of the troubles.

I have to find the time to make the big box transplant: that one has a good quality linear power supply with "charge-transfer" circuit. It used to power a quad UCD100/400 amp.

I attach the future amp with higher supplies Vs for the next project.

Hans, would it be beneficial if I place a resistor (R52) between both emitters for the BJT cascodes for the drivers, as we did to the MOSFETs??? Or will it be irrelevant? I will try to get rid of those emitter-collector little resistors (R8-R28) but the last time I tried the oscillation bursts at-power-peaks reappeared...

Guys, please opine about bandwidth, stability margins and slew rate if you can. I decreased rise/fall times to 0.1uS for the square wave test. I don't know if there is any standard...

Thanks very much again.
M.


Dear Max,


I just found a last thing for you to change.
R35 and R47 are now 10K each. This is too much to supply Q17 with enough base current.
See image below for the difference it makes when changing these values to resp 3k3 and 1K.

Placing a resistor (R52) between the drivers as you did with the Mosfets seems to make no sense.

Testing an amp can best be done with a slew rate ate the input of 2V/usec.
So when offering a 0.8V square wave, use a rise and fall time of 0.4usec.
This is far outside the audio range, but to test stability and resistance against spikes from the mains, this gives a good indication.
With a Miller cap of 100pF and the filter at the input of your amp, Slew Rate at the output is now 25V/usec and with a 50pF Miller cap it is 35V/usec.
Both values are very good.

BW with 100pF is now 380KHz, which is also very good because phase margin within the audio band is within 5 degrees.

Hans
 

Attachments

  • VAS.jpg
    VAS.jpg
    238.4 KB · Views: 59
Hi Max,
I noticed in hte pictues of your prototype that the output transistors are MJ15023 and MJ15024 which are much slower and are less linear than the devices in the simulation. Also the Drivers in the prototype are TO126 types BD139 / BD140? not the fast SMD devices in the simulation.

I wonder if the reason you head a difference with Cascoded output is just improving the performance of devices in the prototype, and may not be required with the actual sustained Beta devices specified in the simulation.

Might be useful to list all the devices in the prototype so a more accurate simulation can be built.

Regards,
Symon
 
Simulation

The driver/cascode transistors and the transistor for the VBEmultiplier have to be mounted on a heat sink, so no SMD devices can be used. This should be considered for simulation. I suggest to use BD139/BD140 devices (TO126)
since I found simulation models on the ONSEMI website.

Simulation Models: BD139
Simulation Models: BD140

Can anybody adapt these models to the simulation schematic?