Since my speakers have flat impedance, I am going to remove it and see what happens. Many speakers have impedance rise at the higher frequency, this creates back emf that may effect stability. Which could also be why you are experiencing improvements using different values.
Caddock MP930 resistors seem interesting. I tried one in my active speaker. I know I hear a difference, but I can't quite decide what to make of it until I put in the second channel. Certainly wish they had thicker leads. The resistors that I currently use did not sound the same with thinner leads. Hmm, sounds like I have each ear in a different sized room.
Caddock MP930 resistors seem interesting. I tried one in my active speaker. I know I hear a difference, but I can't quite decide what to make of it until I put in the second channel. Certainly wish they had thicker leads. The resistors that I currently use did not sound the same with thinner leads. Hmm, sounds like I have each ear in a different sized room.
Last edited:
Dario - I'm not sure I understand what the charts shown in your posts are measuring, but sure, I can plug in 100 pF for C32 and take a look at phase margin, THD20, etc.
However, I can confirm your observations about audible sonics of the Rev FE (alt.)/E completely: tighter mid-bass, immersive sound-stage and presence/palpability without losing any of the imaging, micro-detail, dynamics or other sonics of the Rev C. The only regression was that the simulated THD20 was higher than Rev C (mostly due to the higher value of C34), but it does not seem to affect the audible sonics. My ear/brain doesn't seem to care much about distortion at 20 kHz, it appears.
In any event, I have converted or in the process of converting all my Rev C builds to Rev E, with no regrets. Rev C is still great sounding, but so was Rev A. But I've heard Rev E fairly extensively now, and I'm not going back to Rev C except for experiments.
Simply that phase coeherence seems better with lower value C32.
Sims would be important to assess that the amp maintain stability with 100pF in C32.
Neither do I, my brief listening test was with the other 'alternate' (or Rev E as you call them) values in place, only C32 was changed in a direction that I didn't tested before.
And there is something interesting with 100pF in C32.
dynamic loudspeaker Z rarely affects stability - they are easy loads at typical audio power amp loop gain crossover frequency - it is usually cable C loading the output that degrades stability
I think the VCCS feedback around the LM3886 is an odd conceit - does it really work?
the Howland VCCS feedback around the 3886 is ineffective at higher frequencies due to the finite loop gain, required noise gain roll off - the VCCS Z rolls off 1st order above ~ 100 Hz with the falling 3886 loop gain - with perfect R ratio matching - with even the 0.1% speced R the performance below a few kHz is a crap shoot depending on exact ratio, whether positive vs negative feedback dominates
combine with the expected load of 8 Ohms or less over the audio frequency range and I'm pretty sure ditching the Howland feedback arangement, sense R would give lower, more predictable output Z at practically all audio frequencies
I think the VCCS feedback around the LM3886 is an odd conceit - does it really work?
the Howland VCCS feedback around the 3886 is ineffective at higher frequencies due to the finite loop gain, required noise gain roll off - the VCCS Z rolls off 1st order above ~ 100 Hz with the falling 3886 loop gain - with perfect R ratio matching - with even the 0.1% speced R the performance below a few kHz is a crap shoot depending on exact ratio, whether positive vs negative feedback dominates
combine with the expected load of 8 Ohms or less over the audio frequency range and I'm pretty sure ditching the Howland feedback arangement, sense R would give lower, more predictable output Z at practically all audio frequencies
Are you able to post the simulation results graphically? What are you using for simulation? What speakers are you using?
I was wondering whether anyone had considered Duelund resistors for R3. I have used them for test purposes, and they sound very good. The Caddocks are good for the price and small volume. In the other channel, I am going to try and beef up the leads to see whether the sound is a bit smoother. They go into the MyRef at a later time when I get the phase issue improved.
If I remember correctly Suburra did and reported me great resuls but I would never mount a pencil, at the end Duelunds are pencils...
, into my amp... I would not call it a 'issue'... but a choice.
We should remember the goals that Mauro had: obtain classic Class A sound (as the Musical Fidelity A370 from where comes the My_Ref) with chips...
For sure the 150pF value, born in Rev B, it's not casual.
As he wrote me in the few mails we exchanged when I started the Fremen Edition project, he fine tuned every value with both instrumental measurements and earing tests, obviously pursuing his goals.
That doesn't mean that, if we have different goals, as total phase coeherence, we can't select different values.
In fact in the My_Evo, a new project with different goals, he changed C32 value.
Such value is not known or public but since it's obtained with two ceramic caps in series we can guess it's not 150pF.
Last edited:
Well, I don't understand what class A sound means.
But my exploration of the MyRef type design is mainly due to the fact that it's basically a voltage controlled current source. Hopfully, we can obtain flat "current" phase. Then we can see if this is what we really want in terms of fidelity.
The most unclear thing in lots of audio listening type evaluations is that the speaker performance specification is not standard, and that will greatly impact listing results. To make things even more complicated, these reports do not publish the necessary information that may effect interpretation of how the listening test results can influenced by speaker performance.
Anyway, I get a virtual beer. (still prefer 38 year Royal Salute)
Has anyone read this book?
http://www.current-drive.info/
But my exploration of the MyRef type design is mainly due to the fact that it's basically a voltage controlled current source. Hopfully, we can obtain flat "current" phase. Then we can see if this is what we really want in terms of fidelity.
The most unclear thing in lots of audio listening type evaluations is that the speaker performance specification is not standard, and that will greatly impact listing results. To make things even more complicated, these reports do not publish the necessary information that may effect interpretation of how the listening test results can influenced by speaker performance.
Anyway, I get a virtual beer. (still prefer 38 year Royal Salute)
Has anyone read this book?
http://www.current-drive.info/
Last edited:
Take it this way: his initial goal was to obtain the same sound of the Musical Fidelity A370 with opamps.
Did you give a look to the posts I've linked on Mauro's measurements of phase?
Cheers!
Sure it works, there are a thousand working amps based on this project around the world...
And it sounds great too
that’s one of the things about feedback amps - they can "work", even well, but not always for the "reasons" the designer choose the topology or parts for
it is possible to play in spice with independent loop gain tests that can't be readily done on hardware - I just took a quick look at what the feedback local to the 3886 was doing, then varied the feedback R - even with the 0.1% R tolerance you get wildly differing responses
I'm quite aware that the outer loop "fixes up" many possible errors - I think simplifying the 3886 feedback could give essentially the same final result
Well, after hours of listening day and night, I've made up my mind to make the Caddock MP930 resistors my new reference. It is more dynamic and revealing, almost any music you throw at it have positive improvements and not fatiguing at higher levels. I would say replace any series resistors in your speakers with these. These will also reveal any other deficiencies in other parts of your system.
I remember seeing them somewhere, but I could not find them when I wanted to look at them again.
Yup, it does work rather well, and drives a wide variety of real speaker loads with excellent audible sonics.
Although it was originally spec'ed as 0.1%, I actually match the Howland resistors to much better than 0.01%. With Megitt-Holcos, I can usually match them better than 0.001%.
You are right that the LF response is a weakness, perhaps the only one, of the MyRef Rev C. Definition and imaging at a few hundred Hz and below is where the MyRef Rev C lacks authority (it doesn't actually seem to matter much with typical small woofers, and Rev E improves the LF sonics greatly, in any event).
I'm not sure about that, but I'll check. The phase of Zo is fairly tight across the audible range with the nested Howland - the small output resistor is critical to achieve this. It needs to be non-inductive and have very constant V/I character, which is why people have reported good results with Caddocks and similar high-quality resistors.
Certainly, going with a more traditional nested voltage-series feedback arrangement, a la Jung, can give much lower THD20 (through higher loop gain), but maybe not the tightly-controlled Zo and Damping Factor that Mauro sought to obtain.
Last edited:
The MyRef Howland topology is actually about as simple as it gets - it's often called the "Improved Howland" in the literature. The advantage is that the output swings are attenuated minimally (ZL/(ZL+R3)) compared to the basic Howland, and this is pretty important for efficiency in a power amp.
The only "trick" in the LM3886 section that isn't intuitive is a compensation gimmick that uses a 3k3 + 1nF network at the LM3886 input, which is shunting a nullator (LTP of the LM3886). I would never have thought about using that network, but would have gone with a more traditional DeBoo integrator near the LM3886 instead. The trick compensation does work, but I have no idea why or how Mauro chose that.
Sure, I can post the simulation results graphically - I'm using LTSpice. The speakers, as mentioned earlier, are either Jamo Compact 60 2-way bookshelves (mine) or Fostex FE206 full-rangers (with a friend). I also use JBL LX2002s 2-way bookshelves from time to time.
Simply that phase coeherence seems better with lower value C32.
...
And there is something interesting with 100pF in C32.
Just checked in simulation (with all the earlier caveats about the LM318 and an LM1875 model being used instead of the LM3886).
100 pF and lower (47 pF) for C32 seems to be unstable in the transient analysis. 150 pF is stable, 220 pF gives ~2 dB improvement and 330pF gives ~4 dB improvement in THD20 over 150 pF (not that THD20 matters anyway - I'll look at the phase of Zo next).
It's possible that 100pF is just marginally stable on a real board.
Related stuff: A local DIYer has found the TL071 to be stable in the Rev E, but claims that the mids are a bit veiled. I plan to check this out with an LF01/TL071 Class-A biased version. I also checked the LM1641 in simulation, and it is unfortunately unstable - maybe some compensation mods are required.