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

First of all, big thanks to Hephaïstos (ascended Master), to Peufeu, to Douglas Self, to uncle Charlie, DestroyerX, and to all those that investigate on thermal memory distortion and finally to everyone who is willing to help me here.

This is not a thread about THD. It is a thread about experimenting with putative Low Thermal Memory Distortion or LTMD circuits.

Edit: just at the Lions Gate of 2022 I found the best output configuration. Check out post
#806 (I attached the diagram down here) for the latest official Bootstrapped Darlington output configuration.

Edit: it also runs very well with VFETs as bootstrap element for the outputs.



I am proposed to implement the mods that Peufeu studied for diminishing putative thermal memory distortion effects on solid state amplifiers, but adding my own experiments on the output section, which Peufeu did not (unless I am not aware of that) address. In fact, I began my research modifying the outputs (Cascode all thy outputs), that is why Peufeu's articles were so interesting to me. I took the liberty to re-post the proposed amp from Peufeu. (it may have a minor error on the cascoded CFP like my experiment proved)

I have no electronic background; only the experience and scarce knowledge I obtained here, so be patient.

The blameless amp is the prototype to modify and since I have a Blame ST from Destroyer X, the idea is modify it section by section as Peufeu proposed and evaluate if there is any improvement in sound to be achieved.

I already swapped differential amplifier BJTs (Q1 and Q2) for cascoded CFP (J103 P channel JFET cascodying BC556-BC546 Sziklai pair) and it worked. Current is 2,09mA by side, though one may have to trim mirror resistor R8 to achieve 0 offset.

The sound, and this confirms Peufeu's claims, became very smooth and relaxed, with soft highs and more prominent midrange and mid-bass (my test speakers are very bright sounding) very soothing experience...the bad news is that the other channel has a problem that I could not solve yet: -27V at drivers bases, without oscillation :( But, hey! That is the way we learn the most!

I let to the theorists the task of discriminating if this supposed sound improvement comes from increased linearity, from decreased thermal memory or from shared thermal task of units with different thermal behavior...

Time ago I had already tried a mod to the output section that I liked: I cascaded the output BJTs with other BJT power devices (D44H11/D45H11) in common-base. Not properly a cascode, since it is not a common emitter output (even if I CAScaded the COllector :D) apparently, but I perceived an improvement in transparency and dynamics. I have also modded another amp, the VSSA, with this trick and I also noticed a worthwhile improvement. My plan is putting, like on the input section, a cascoded CFP output later. You may notice that, absolute and eternal beginners like yours truly, are willing to try mods that reasonable people would never even think of... :D

Then, I will try the mods to the VAS circuit (cascoded-cascoded VAS) that Peufeu recommends.

But the final goal of all this is to gain experience to attack the modification of my Sony TA5650 VFET amplifier to make a real AMNESIS amp. I already cascoded the VFET common-source output section, which not only make it usable and stable, but also made it a very good sounding amplifier, with increased dynamics and punch.

I hope this humble effort awakens renewed interest in this most important topic.
Cheers,
M.
 

Attachments

  • Memory compensated annoy_full_amp.gif
    Memory compensated annoy_full_amp.gif
    23.2 KB · Views: 7,541
  • BLAME CASCODED.jpg
    BLAME CASCODED.jpg
    987.2 KB · Views: 6,723
  • CASCODED CFP MOD.jpg
    CASCODED CFP MOD.jpg
    476.6 KB · Views: 549
  • AMNESIS BOOTSTRAP OUTV5.png
    AMNESIS BOOTSTRAP OUTV5.png
    110 KB · Views: 765
Last edited:
well semiconductors come with memory there is no way to avoid it is just semiconductor physics it is "semi" responsible for memory effects. Informally the carriers of electric current in semiconductors are not "free electrons" as in for.ex. a metal like gold. That transforms to capacity, after all a capacitor is memory ( see sample and hold). Thus there is not only "thermal memory" there is further signal memory . Although the model of billiard-ball electrons flying from cathode to anode in a tube is wrong in physical reality those good old vacuum tubes come almost without memory.
 
  • Like
Reactions: 1 user
well semiconductors come with memory there is no way to avoid it is just semiconductor physics it is "semi" responsible for memory effects. Informally the carriers of electric current in semiconductors are not "free electrons" as in for.ex. a metal like gold. That transforms to capacity, after all a capacitor is memory ( see sample and hold). Thus there is not only "thermal memory" there is further signal memory . Although the model of billiard-ball electrons flying from cathode to anode in a tube is wrong in physical reality those good old vacuum tubes come almost without memory.

Indeed, but we can try to minimize it, to render it negligible...in terms of audio satisfaction I mean... ;)

Cheers,
M.
 
I'm convinced linear amps will yield to switching amps. The plagues are reduced to accuracy of ADC ( not really a big item) and to timing errors ( a big item) of the switches. Eventually the topology will be a sort of current dump where a very linear and very fast linear amp with small power - 1 watt about- corrects errors of the switching amp.
 
see thread How fast should an amp be? broadband topology

I know that one. Thanks.

The higher (if stable) the better...apparently we are most sensitive to transients (attacks) and to harmonic signature of sound sources (instruments; voices) as evolution or The Designer wanted. :)

Maybe that is why I tend to prefer passive current sources (?)

Anyway, now I'm stuck with one of the amps. See, the first board accepted the mod luckily without problem: the VAS is asking for 5mA and I was able to set offset to mV unit level...and it sounds great.

The other board though, maybe because of a little error in the original cascoded CFP diagram (it must be 1K between second BJT (NPN) emitter's and CFP output node) the amp keeps behaving strangely, with high negative bias voltage for the drivers. The VAS apparently is asking 25mA to the passive CCS and that collapses the positive voltage available, or that is what I understand, even when I use single transistor VAS, but I am not sure where that current is going...I couldn't detect any damaged units nor oscillation. I replaced most active devices anyway. If I cannot make it work, I will try an active CCS. If everything fails, I will have to build a new board... :(

Thanks for your interest,
M.
 
hum I think the approach of DIY is at best inconsistent - incomplete anyway. Y ou see those guitar guys they knew the guitar the pickups the various filters the amp AND the speakers is ONE unit. That is the way to go for DIY. Leslie West's (Mountain) classic sustains had only been possible with such a one unit. Not with any other. More advanced HiFi magazines had demonstrated that it matters not so much what the data of an amp and a speaker one by one are for the sonic impression - that is what matters.
For ex. the Rock classic Get ready sounds best in a convertible car with the amp clipping. At least up to 80 mph. As philosopher Russell puts it bluntly "There is no objective Hamlet in the theater and outside neither". It is all about "mood" and that cannot be measured
 
Perhaps you are right.

I recognize two ways to listen to music:

The first is "passive", where you let the reproduced sounds make an "impression" on your vigil consciousness.
In the second, that I call "active", your mind re-creates the music heard instantly as it flows, which requires a permanent awareness of the moment...it is hard to explain. I will try to pick clearer words next time.

Both experiences can be fulfilling, the first being soothing and the second more alike critical listening, if you wish. Some systems make you incline to the first one and some to the second.
 
Finally, your favorite eternal beginner and humble narrator has succeeded. :cool:
It was hard. Like "awakening the Dragon"...

The faulty amp, believe it or not, had the 100V bootstrap cap damaged! Or so it appears...My cap tester kaput (kids, discharge caps before testing) so I cannot confirm leakage but replacing it for a good one made the amp work flawlessly...and I seem to remember Rod Elliot, perhaps (great site) mentioning something about that those caps never fail...or was it my imagination?

I had the chance of a short listening test in stereo until the other amp kaput!

I now always think that these catastrophic events are a blessing in disguise because we learn so much during the trouble-shooting. :D :redhot:

I replaced the corresponding bootstrap cap thinking that the was the fault and it ended being a PNP driver gone.

While I was there I couldn't help thinking that using a big cap in the signal chain must not be good for memory, right? So, apart thinking about running a version without the said cap, I started looking for information about current source plus bootstrap. I stumbled across a very interesting thread about CCS-bootstrap by Dadod, which I am now reading. Must be a worthwhile experiment. He also used the Blame as model.

Please bear in mind that I am not proceeding in a strict scientific way to evaluate my mods: I simply try them and decide, from memory (hehe) if I like it or not. I have not the time to build another pair of amps and do the ABA test for each mod, so you are warned.

So when trying to fix the amp, I decided to make the simple cascoded VAS mod and it worked. I think the cascoded VAS gives the more velvety midrange so now I have it in stereo. One VAS asks for 12,7mA and the other asks for 10,8mA. Offset is steady around 10mV. Voltages for the bases of the outputs are +/-550mV and amps run cool.

So, how does it sound, my cascoded-CFP plus cascoded-VAS plus cascaded output modified blame?
Stunning. First, one notices reduced highs which is intriguing but then one realizes that this is how live music sounds, and the high pitched instruments just sound right. Mids are a bliss. Velvet is the word that comes to mind. Instrument's textures and timbres are very well defined and distinct and this a salient quality of this amp!
Bass is all there. Depth of images is great and so is layering. Musical lines of every instrument are well designed and I can hear a lot of previously hidden details...an anecdote: in my first listening I worried because I started to hear loud clicks and pops and the scope gave a jittery trace before music started, until I remembered that I was listening to a vinyl upload on youtube, "Stone Flute" by Herby Mann. I have listened to this before but never with such loud clicks and pops! :D


Anyway, you know always a child is handsome for his father, so I am probably not being objective, but anyway, I listen to music in an subjective way, hehe.

The good things are that inexpensive, low bias, class AB amps can sound grreat and that I am learning a lot to improve my Sony TA5650.

I cannot upload a decent diagram but I promise I will in the future.

Cheers,
M.
 
I use same input but simple (no CFP).
I get THD = 0.000275% at 31W/8Ohm, 20kHz.

If I uses CFP, THD20 rise but THD1 fall.

Hi guys. Thanks for your inputs.

Bimo, are you sure you need that resistor to JFET gate?
Did you tried or just simulated cascoded-CFP?

I wish I knew how to use simulators...

Anyway, the mod is simple and transistors are cheap, so why don't try it and see/hear what does it bring?

You only need a pair of PNP and a pair of 1K resistors that go between emitter of the new PNP and source of JFET which is also the output node of Szicklai...or at least I wired that way to get even currents (2.09mA) between legs.

Another easy mod is cascodying the tail. ;) But later as now I will continue my listening tests.

Cheers,
M.

PS: now I am mentally listing all the circuits that include differential pair in my arsenal to proceed to modification. :D
 
Hi guys. Thanks for your inputs.

Bimo, are you sure you need that resistor to JFET gate?
Did you tried or just simulated cascoded-CFP?

I wish I knew how to use simulators...

Anyway, the mod is simple and transistors are cheap, so why don't try it and see/hear what does it bring?

You only need a pair of PNP and a pair of 1K resistors that go between emitter of the new PNP and source of JFET which is also the output node of Szicklai...or at least I wired that way to get even currents (2.09mA) between legs.

Another easy mod is cascodying the tail. ;) But later as now I will continue my listening tests.

Cheers,
M.

PS: now I am mentally listing all the circuits that include differential pair in my arsenal to proceed to modification. :D


I just simulated it. I do not have time to make a prototype.
I sent my design to a friend, he will make a prototype. But it take a time.
 
OK. I know have the input section of my Sony TA-5650 properly modified with Peufeu's Cascoded CFP trick. The effect was easily detected as being the same caused on the Blame. Less aggressive highs, more midrange presence (the bodies of the instruments are very pronounced), better delineation of melodic lines, in all, if I knew how to speak Latin, I would say the sound is now rotundum :D

Fortunately, the DC offset was low and I had not to mod the circuit: 10mV and 1,8mV.

It is now decided: I will swap all my differential pairs for the cascoded CFP unit. See attached picture of the units in preparation. I think I reached the limit of reduction. More than this, it would have to be SMD...
I am also studying mixed CFP (JFET + BJT) to replace JFET inputs... :cool:

With a little of effort, the units can be placed in existing space.

Guys. try it. At least for experimentation...remember, all mods must be evaluated not less than 48hrs before verdict. ;)

Cheers,
M.

PS: an "all cascoded-CFP" amp would be interesting...

PS2: I attached the diagram of the mod. The amp has now 100R degeneration on the emitters. The 1K CFP resistor is fitted on the emitter of the NPN BJT instead. This way it worked better.
 

Attachments

  • CASCODED CFP unit.jpg
    CASCODED CFP unit.jpg
    633.6 KB · Views: 1,450
  • SONY TA5650 AMPLI CASCODO-CASCODO2 CFP SCHEMA.jpg
    SONY TA5650 AMPLI CASCODO-CASCODO2 CFP SCHEMA.jpg
    278.2 KB · Views: 1,475
Last edited:
Continuing with the experiments, yesterday I successfully "improved" the differential amplifier input section of my single ended Tokin 2SK180 VFET amp :cool:

Since it has JFET diffamp, I decided to make a "cascoded-compound Sziklai" arrangement. For that I studied some academic articles but they were too much for my limited understanding so I made it the brute force with SK170 N JFET both as the cascodying element and as the "driver" of the Sziklai pair and a BC560 PNP as the "follower" of the same, with a 1K emitter resistance. It worked. And first listening tests are very rewarding. :cool:

Guys, this mod always improve attacks and detail retrieving. Pizziccatti e stacatti, plucking, smashing of the keyboard, or any ambient noise are greatly enhanced, which, summed to the extra wide soundstage, beyond speakers, makes the experience like a "live" event and you keep wandering -is that the recording or the neighbor??? :D

Photo of the small unit and diagram of the target amp.

Cheers,
M.
 

Attachments

  • diffamp TOKIN 2SK180.jpg
    diffamp TOKIN 2SK180.jpg
    60.6 KB · Views: 888
  • diffamp cascodo CFP.jpg
    diffamp cascodo CFP.jpg
    328.2 KB · Views: 757
Hahfran, you are likely correct as I built a class D operating at 2.2 mhz @ 42 v rails
the sound is very class A. the 2 problems are creation of low level noise on FM when in the same case and reliable over current protection speed.
distortion is in the area of 0.003 as the corrective op amp is a opa2132
 
Last edited: