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

Hi!
will you try to cascode also output transistors?

Hi padamiecki,

Yes, I said before I vertically cascaded (or maybe cascoded; at least Pass' article says it is indeed a cascode) the outputs of many of my amps. Yesterday I added Sziklai pair to the cascoded outs:

Second, I finally did the cascoded-CFP mod for the already modded cascaded/cascoded output. Since it was already modded, I just added a PNP (D45H11; different encapsulation) to the 5200 NPN, as follower of the Sziklai pair, with a 10R from 5200's C to D45H11's E. It worked flawlessly,

Obviously, the other side has the complementary PNP Sziklai implemented, just in case it was not evident from my description.

I am still evaluating (and hopefully also shall be able to show it on LTspice) the mod to see if it can be improved...I need to improve my LTspice skills quickly!

Sorry for the photo...my quick and dirty mods...but the diagram may be useful.
Edit: don't know why the photo is allongée...

Cheers,
M.

PS: please comment about my 10R feedback resistor...is it too Big? is it too small?
 

Attachments

  • Blame cascoded CFP cascoded CFP.jpg
    Blame cascoded CFP cascoded CFP.jpg
    991.4 KB · Views: 713
Last edited:
This is the present stage of my experimentation, from memory (no pun intended)...

Again, Heil to DestroyerX, to Hephäistos, to Peufeu and to every one who helps with advice and constructive criticism...

I am beginning to understand this cascode thing :-( slowly...


Edit: there may be some mistakes in the diagram...

PS: I added an extra 1K resistor to the input "power cell" to have low DC offset. I know think that it might be detrimental in noise performance. I shall experiment further with it...
 

Attachments

  • Blame cascoded CFP input cascoded VAS cascoded CFP out.png
    Blame cascoded CFP input cascoded VAS cascoded CFP out.png
    187 KB · Views: 920
Last edited:
Hi maxlorenz,

your latest schematic differs from the schematics
used for simulation / showed in 'Memory distortion theory'
especially in the VAS section:

no use of VAS buffer, VAS connected directly to bias adjustment
no current source in VAS and in bias adjustment

insteadt insertion of R14, R15, C9/C10 to FB

Can You explain the functionality and the reason for this change.
What is the function of R11 in the current source and R3 in VAS.

Thanks
 
Hi JOSI1,

Thank you for your interest.

As I stated, this is not exactly the proposed amp from Peufeu.
It is the Blameless amp from our dear Destroyer X, which I hapen to have here and which is willing to sacrifyze itself for a good cause. It has originaly a enhanced VAS with bootstrap (I am trying to figure it out how it works and if it can influence the putative TMD of AC signals) which I am modifying at the same time that I study (and hopefully, understand) the theory behind, when time allows. This is my strategy...not very clever but it is the best I can do: if the amp doesn't oscillate than I am happy and leave it there for a while, until I find a better aproach, doing the critical listening tests.

I plan to test many mods and evaluate what is best, sonicaly. I cannot do it otherwise. These are very interesting and, IMHO, worthwhile experiments and I am shocked that few have attempted them. I am reluctant to post my listening experiences for fear to get too much negative feedback but believe me I would not be doing further test if I had not seen the potential it has...

For the future (and forgive me if this is insane) I want to test and "all cascoded-CFP" amp, meaning in every critical position, I will replace a transistor for a "power cell" formed by a "cascoded-CFP", and with higher PS voltages. :eek:
I imagine a "cascoded-CFP" driver would eliminate the need for a VAS buffer...am I wrong?

So far my attempts for simulating the experimental amp have failed due to my lack of expertise on LT spice...

R11 is in the original amp, help biasing the CCS I supose.
R3? Maybe R32? This is 3ohm (small degen) remanent of the original enhanced VAS, practical for soldering purposes...

Best wishes,
M.
 
Last edited:
Hi Max!
probably the emmiter follower does not suffer from memdist, so no need for cascoding it.
Attached is corrected output section, don't you think it should be look like attached?

Ah! You're right! My diagram is bad but the Sziklai (you can see my pic with a big resistor to the collectors of 5200/1943 transistors) is wired as you posted. Sorry.
I have higher bias for the common-base BJT to avoid premature clipping.

About output section, I want to test the most promising mods. Even if it is not TMD the cause, I like better the sound as it is now. We'll see if it is satisfying in the long term...
I want to mod the driver and see what that makes.

Please take a look at the VAS. What is needed to make it an Hawksford cascode???

Many thanks for your input.
M.
 

Attachments

  • Blame cascoded CFP input cascoded VAS cascoded CFP out.png
    Blame cascoded CFP input cascoded VAS cascoded CFP out.png
    188.8 KB · Views: 742
Apparently, when I added the CFP output the former drivers became pre-drivers, or so it seems...

Anyway, apart cascodying the tail CCS for the input, the next move should be exploring a CFP mod for the cascodying transistor on the output, he, he. So four transistors instead of one, per side. :D

I hope I will be able to simulate before moddying, this time. :eek:
 

Attachments

  • Blame cascoded CFP input cascoded VAS CFPcascoded CFP out.png
    Blame cascoded CFP input cascoded VAS CFPcascoded CFP out.png
    192.8 KB · Views: 700
Dear Max!
I wrote before that EF has a little thermal effect regarding memdist, as Peufeu stated in his analyses,
so this means that cascoding of cfping output transistors will not lower memdist,
however that might be better sound due to better linearity...

And better handling of power peaks without simply paraleling devices. ;)

The objective should be to explore "power cells" formed from different types of transistors, hopefully even different encapsulations, and its influence on sound and electric or thermal behavior, compared to simple transistors in the same function...I am not saying I can do that but this is the ideal... :D

Cheers,
M.
 
OK. Following with my crazy experiments for no apparent logic reason (which is probably the belief of our linear logic inclined mates :D), I managed to get the time to mod only one channel with the CFP-cascoded-CFP output. I increased the feedback R to 100 everywhere in the output, as this is the minimum I saw in other circuits...and still no luck with simulations.

The channel works properly. I experimented an increase of bias current from +/-45mA to 104mA. I am trying to figure out how this happened...maybe the increased feedback R?
Anyway, I still have one figure DC offset. ;)

Why on H**l would anyone use 4 transistors where only one is needed? :confused: :mad:

Maybe sound quality??

BTW, feel free to critisize and advise.

Cheers,
M.

PS: Q9 and Q10, formerly drivers, now probably pre-drivers, look very lonely IMHO. I think they deserve good company, hehe.
 

Attachments

  • Blame cascoded CFP input cascoded VAS CFP cascoded CFP out.png
    Blame cascoded CFP input cascoded VAS CFP cascoded CFP out.png
    192.9 KB · Views: 615
Last edited:
Hi Max!
if it did not blow away, this means that the mods are "nontoxic". from the pov memdist philosophy there is no need for use cfp pair in cascoded transistors pair D44/45H11.

Since we are exploring new realms of the possible, why don't we try experiments and see/hear what it gives?
(our universe is as narrow as our frame of reference allows...)

The transistors, the circuits, the currents, the energy flow, do what they do, not what we think they do. :D

The present paradigm is flawed anyway. ;)
 
Quick update.

Since I am heavily modifying an existing amp, the practical execution of the output section is a mess...the amps started oscillating as I increased current bias or, with lower bias, when I increased volume :eek: I managed to get one channel stable at lowest bias (without crossover distortion) at all volume levels.

The other channel is stable at idle but, as I increase volume it starts short bursts of oscillation at the crest of the positive waves! Even with added capacitance at power pins...
I will have to re-arrange the whole thing to get it stable.
Future PCBs will have decent layout for stable behavior.

I am a little frustrated because the sound is even better that with the "simple" cascoded-CFP output. It seems nude of all artifacts. Instruments' timbres and textures are fantastic, as are attacks...but I may have said that before... :D

Anyway, wish me luck.

BTW, I wanted to understand better the behavior of the CFP an I started investigating, so I found our mate's Ian Hegglun's PAKproject: wow! The mats involved are serious. I'd better follow the empirical way...
M.
 
OK. Apparently I managed to stabilize the second channel. It was a trial an error exercise...

Briefly, one of the first steps after noticing the 1MHz oscillation was to increase Cdom from 100pF to 220pF. That appeared to make a little improvement. By improvement I mean the burst of oscillation of the positive crest of the sine wave appeared at increased power, in mW range!.

Then I tried improved positioning of the output section Transistors, with shorter leads, this made also a little improvement, similar to increase Rfb for the CFP cascode from 100R to 370R.

Adding parallel capacitance (5 to 39pF) to the main FB resistor, R13, made things worse!...which made me think...

Increasing base stop resistors from 2R2 to 22R for the CFP output deteriorate both amplitude and threshold for the oscillation...

Then I thought probably the pre-drivers were not having anough slew-rate to accomodate for the OS and did the mod that I was thinking of anyway: I cascoded the pre-drivers. That improved significantly the things but failed to stop oscillation. Then I though; what if is the VAS that is too slow (slew rate limited) after the increase of Cdom? I reverted to 100pF and voilá! Finally, I could listen to the stereo amp, undistorted and stable, at least with about 4W output, which is the limit my neighbors would accept on the patio system.

YouTube

YouTube

All in all, good exercises to increase the insight of the way the amp works...

I think now I must try increasing current for the differential input pair a little more...AND cascodying the CCS for it.

Today I plan to do more listening.

Wish me luck.
M.
 

Attachments

  • Blame cascoded CFP input cascoded VAS CASCODED PREDRIVER CFPcascoded CFP out.png
    Blame cascoded CFP input cascoded VAS CASCODED PREDRIVER CFPcascoded CFP out.png
    203.4 KB · Views: 486
Last edited:
Updated schematics with cascoded CCS for the diff. input "pair".

Successful long term listening test :cool: You guys do not know what you are missing...transforming the cascodying elements (Tr13 and Tr14) at the output from simple transistors to CFP (Tr13+13'; Tr14+14'), IMHO, produced a major improvement. Sorry but I have to explain: the sound opened up and became more transparent (nude), and smoother while more detailed; soundstage became wider, taller and also deeper; instrumental textures improved further. The musical lines became more evident, which, added to timbral and spacial discrimination, makes a better separation of performers. If the band has seven members, you probably will hear the seven individual instruments.

The not so good part is that I lost the "dark" presentation of previous iterations to which I was used. At present it sounds more neutral, in a natural (not clinical) way...

I am still considering this a prone to oscillation, experimental amp. I will today reduce the corner F of the input filter. ;)

Cheers,
M.
 

Attachments

  • Blame casc CCS casc CFP input casc VAS CASC PREDRIVER CFPcasc CFP out.png
    Blame casc CCS casc CFP input casc VAS CASC PREDRIVER CFPcasc CFP out.png
    204.5 KB · Views: 442
Last edited: