When you do the math, a term that will frequently appear in the equations is jwRC, which determines the so-called pole or zero, which eventually determines loop stability. If we want to tweak either R or C, but want to retain stability (and retain most of the transfer function the same as before), it's prudent to keep that RC product constant. So if you decrease C, just increase R accordingly and it's likely to remain stable. This works with R43/C34, which are in series. In the case of C34, lower values give lower distortion in the sims, right down to the Rev C default of 10 pF. Below that, the loop will probably get into the unstable region.
For the other Cs like C10, there's no easy analytical recommendation to make. It has to be tried empirically with real components and auditioned carefully with a variety of tracks to ensure that's there is no regression. Generally, larger C10 gives you more feed-forward compensation, but also lower stability. 27-33pF is around the maximum practical value in the simulations, anything higher tends to be unstable.
Yup, the BG PK is about the best choice for C9 still easily available. Other alternatives are Panasonic FC/FM, Nichicon Muse FG and KZ, Elna RE3, RJJ, RJH, Silmic II, Cerafine, and Panasonic Pureism. Nichicon Muse non-polars will also work, as will nearly unobtainium stuff like BG NX and BG NX-HiQ (both of which are boutique caps priced in the stratosphere from a few EBay sources). In some cases, bypass cap C21 may have to be populated with a film/foil cap for optimal sonics.
It might also be worth trying Oscons at C9 - maybe SA/SC/SP/SF etc. will fit the bill for an inexpensive BG replacement.
Thanks, C9 is listed at 50V in the BOM, that's what threw me.
Hifi Collective have BG PK 220uF 16V (12.5 x 20)
£6.26 + VAT 😱
Hifi Collective have BG PK 220uF 16V (12.5 x 20)
£6.26 + VAT 😱
Siva, thanks for the explanation, but I still have the basic question: why would we want to change either C34 or R43? If we maintain the same RC product, what aspect of the sound changes as we change the values of R and C within the equation?
More listening last night, swapping amps between right and left sides of the room and system. There is definitely something not sonically correct with alternate comp C10@33/C34@27. 33/22 is an improvement in every respect (it's even louder!), but still not ideal.
Do NOT install MK132 or any other transparent resistor at R10 unless you are willing to adjust compensation or suffer ear bleeding, distorted highs.
Regarding C9 and BG's in general: yes, the BG is the best available 'lytic cap. All others are a sonic compromise, but you can still get decent sound without it. If you already have a BG, treat it as a precious gem because there are no more. Most on ebay are probably fakes. Anyone who has the real thing is smart enough to know its value, so forget finding a bargain anywhere. I suspect BG's will be handed down from father to son until some company gets smart and realizes they can sell any sonically equivalent capacitor for a LOT of money. Mundorf comes to mind.
If you do not have a BG cap already, you can fiddle with a good quality 'lytic and a small film cap bypass until you find the right combo. I plan some experiments after I solve this other issue.
Peace,
Tom E
More listening last night, swapping amps between right and left sides of the room and system. There is definitely something not sonically correct with alternate comp C10@33/C34@27. 33/22 is an improvement in every respect (it's even louder!), but still not ideal.
Do NOT install MK132 or any other transparent resistor at R10 unless you are willing to adjust compensation or suffer ear bleeding, distorted highs.
Regarding C9 and BG's in general: yes, the BG is the best available 'lytic cap. All others are a sonic compromise, but you can still get decent sound without it. If you already have a BG, treat it as a precious gem because there are no more. Most on ebay are probably fakes. Anyone who has the real thing is smart enough to know its value, so forget finding a bargain anywhere. I suspect BG's will be handed down from father to son until some company gets smart and realizes they can sell any sonically equivalent capacitor for a LOT of money. Mundorf comes to mind.
If you do not have a BG cap already, you can fiddle with a good quality 'lytic and a small film cap bypass until you find the right combo. I plan some experiments after I solve this other issue.
Peace,
Tom E
What is so special about BG.. Technology ?
I use BG from pcx for c9 myself so can say I own some sort of religious artefact 🙂
I use BG from pcx for c9 myself so can say I own some sort of religious artefact 🙂
Hello Tom, I maybe wrong but I've got a look to the LM318N datasheet to understand the meaning of that R39,R43,C34 network. The "8" pin is a compensation pin. In the paper, it is used in one "auxiliary circuits" (note 10) as a "feedforward compensation for greater inverting slew rate".
In that example, the negative input is connected to the 8 pin (that is located almost at the "end" of the opamp internal circuit) through a RC series. Then, the input is "feedforwarded" to the output. Now, in the My_ref FE (and rev.C) schematic, the 8th pin is connected to the output 6th pin. Then, it can be simply assumed that it's giving a local "feedback" action. Of course, this should help to "stabilize" the opamp.
The analysis of Siva is correct about the meaning of the series R43-C34. If you change only one value, you are changing the "corner frequency" (f=1/2pi RC) of the network. By increasing the capacitance, the frequency is decreasing.
On the other hand, above the corner frequency, the value of R43, together with R39, is setting the resistive value of this feedback effect because at high frequency R39 and R43 are in parallel (being negligible the impedance of C34).
My 2 cents 🙂
Daniele
I've read and reread Siva's post, and I think I get it. Reducing C34 reduces distortion in the feedback loop.
So why does the alternate comp INCREASE C34 value? And if larger C10 decreases stability, why is the alternate comp value larger? Am I finally getting this correct? What was the original goal of changing comp?
I also understand the RC product concept, that changing either value requires changing both to maintain the transfer function. Changing only C34, and by a huge amount (2.7 X!), means that filter is operating at a very different corner freq. So if we want to try alternate comp, then R43 should also be changed proportionately.
Would that improve the sound? Now I hear excessive brightness with FR anomalies and what sounds like distortion (edge) of highs.
Peace,
Tom E
So why does the alternate comp INCREASE C34 value? And if larger C10 decreases stability, why is the alternate comp value larger? Am I finally getting this correct? What was the original goal of changing comp?
I also understand the RC product concept, that changing either value requires changing both to maintain the transfer function. Changing only C34, and by a huge amount (2.7 X!), means that filter is operating at a very different corner freq. So if we want to try alternate comp, then R43 should also be changed proportionately.
Would that improve the sound? Now I hear excessive brightness with FR anomalies and what sounds like distortion (edge) of highs.
Peace,
Tom E
Reducing C34 gives higher corner frequency in the local feedback loop. This makes the LM318N faster and this reduces the distortion.
Increasing C34 makes the LM318N slower and more stable. Increasing C10 gives more "feedforward" effect (and less total stability). The two effects are opposite then and, in a certain sense, should compensate themselves (from the stability point of view).
If I'm not wrong (Dario can confirm), the goal of changing C34 and C10 was just get "better" audio performances, while keeping stability.
Of course, if C34 goes larger and larger, more distortion should come in (slower circuit).
Regards,
Daniele
Siva,
Good list. I have tried (and I think Dario has also tried) both OSCONs and Niobiums. They look promising from the datasheet, but my experience is that they are somewhat thin sounding at C9. I'm not sure if they are above or below the other 'lytics but I suspect that Dario would say below the 'lytics.
The Oscons I tried are marked "FP" on top, but I've forgotten the rest.
Jac
CLICK!
That's the sound of everything falling into place. Daniele, thank you so much for your patient, detailed explanation. This is all pretty complicated for someone with no formal education in electronics, but I have a much better idea now of what's going on.
Of course, the goal of changing the comp was better sonics. That is always our goal, and Dario and others worked hard to achieve it. But it seems that balancing these two characteristics, stability and distortion, is very delicate.
So we do NOT want to alter R43 proportionately to C34. That would basically have no effect at all. We want to change only C34 or R43 to change the RC product to move the filter corner to slow down the LM318 to increase stability to improve the sound. Whew! But that also increases distortion. So we change C10 to try to reduce distortion and give up some of the stability we just gained. I think when you wrote "should compensate themselves" you really meant "balance each other." What if they don't really balance and it all doesn't work? How will we know if it does or doesn't? Simulation? Measurements? Listening?
Please don't lose patience with me now, guys. I feel that we are just getting to the heart of the matter. Whether this is or isn't the cause of the unpleasant sound I'm getting is important, but it is also thoroughly enlightening to know what is going on with all the circuit bits.
Peace,
Tom E
That's the sound of everything falling into place. Daniele, thank you so much for your patient, detailed explanation. This is all pretty complicated for someone with no formal education in electronics, but I have a much better idea now of what's going on.
Of course, the goal of changing the comp was better sonics. That is always our goal, and Dario and others worked hard to achieve it. But it seems that balancing these two characteristics, stability and distortion, is very delicate.
So we do NOT want to alter R43 proportionately to C34. That would basically have no effect at all. We want to change only C34 or R43 to change the RC product to move the filter corner to slow down the LM318 to increase stability to improve the sound. Whew! But that also increases distortion. So we change C10 to try to reduce distortion and give up some of the stability we just gained. I think when you wrote "should compensate themselves" you really meant "balance each other." What if they don't really balance and it all doesn't work? How will we know if it does or doesn't? Simulation? Measurements? Listening?
Please don't lose patience with me now, guys. I feel that we are just getting to the heart of the matter. Whether this is or isn't the cause of the unpleasant sound I'm getting is important, but it is also thoroughly enlightening to know what is going on with all the circuit bits.
Peace,
Tom E
Something seem seriously wrong...
I can't understand... As I wrote in my system I've no shrill effect, no ear bleeding or whatever using both Rikens and Caddocks.
Rikens are more forgiving, meatier and musical, Caddocks more neutral and detailed but both sounds good.
Daniele listened to my system and I talked with him about that last bit of harshness I was earing and that I later removed replacing True Coppers.
Daniele can you tell us what was your impression? Did you experience ear bleeding despite the presence of true coppers?
I can't understand... As I wrote in my system I've no shrill effect, no ear bleeding or whatever using both Rikens and Caddocks.
Rikens are more forgiving, meatier and musical, Caddocks more neutral and detailed but both sounds good.
Daniele listened to my system and I talked with him about that last bit of harshness I was earing and that I later removed replacing True Coppers.
Daniele can you tell us what was your impression? Did you experience ear bleeding despite the presence of true coppers?
Yup, that's the general intuitive picture I got as well. The Rev E compensation went even further, increasing C10 to 27..33pF and C32 to 330pF - but this combination required a very high value of C34 (100 pF) to maintain stability. That increased distortion by a pretty large factor - the upper mids became edgy and jagged (most noticeable on saxophones). However, the soundstage became very immersive and expansive, and the lower mids became explosive - percussion instruments sounded very taut and tight, and almost like you were in the midst of the orchestra. I used a Class-A biased LM318 driving a Class-A output buffer in the LF01 module to ameliorate the jagged edginess of Rev E in the upper mids (the same effect is also used in the LF07 module, but with a different Class-A output buffer than the LF01). Many listeners liked the up-front presentation of Rev E, although it was colored considerably by the (hugely) altered C34-R43 product.
For FE, Dario can probably give more insight - C32 was reverted to 150 pF to allow C34 to be kept at 27pF (closer to Rev C's 10 pF), and to keep the distortion and corner frequency closer to Rev C. It seems to have worked as intended, though the upper-mids may still be audibly more jagged than Rev C. However, much of that is probably kept under control by the improved (lower-Z) series-shunt regulators for the LM318.
In principle, the same Class-A opamp output buffer trick can be tried on Rev FE also, but requires a respin to revert back to a DIP8 footprint for the LM318. The other alternative is to respin the last Rev C board (with a DIP8 LM318) to incorporate the shunt regulator from the FE. I'll eventually get around to doing the latter, but no timeline as of now.
Simulation is only a general guide as to which direction the tweaks should take. Measurement gives hard numbers for THD, but doesn't necessarily give insight into audible sonics (e.g. Class-A sonics, even with moderately high distortion, still sounds pleasing). The ultimate proof of the pudding is the eating thereof - in this case, extended A-B listening tests with a wide variety of tracks.
Given that things are still gradually being improved 8 years after Mauro's initial breakthrough, I think there's still many more years of tweaking and modding ahead of us.
When I 've submitted what Siva call Rev. E to Mauro (which was against touching compensation) he explaned that such an high value of C32 was way over compensated, a performance killer.
A small extract from Mauro's comment:
With 330pF on the global feedback you defintevely completed the over-compensation work. Probably you have a local oscillation on the LM318, if not you have saturated the intermediate stage for sure with the new C34.
By the way you have an insufficient bandwith and a THD that probably increased by a decade (for the LM318's saturation)
So I've reverted C32 to 150pF and started working on C34 and C10 only.
From what I wrote to Siva at the time in my listening tests increasing C34's value (C10 was 22pF) gave sweeter highs and less control on woofers.
From Evolution Rev A documentation I knew that C10 could be 33pF so, I've started from there and tryed again several values of C34 and settled on 27pF, which sounded best balanced (control vs sweetness) to my ears.
I've also tryed changing C10 but the best value seemed 33pF from start.
Siva kindly simulated those values and confirmed stability in sim, so after months of correct operation I've released them as official for the FE.
I really think FE compensation it's very similar to Evolution one, both from listening comparison and by the fact that on the FE you can experimentally apply Evolution Rev A. mod without problems.
IMHO if compensation values would have been too much different from Evolution ones the Rev A. mod would probably never worked.
Last edited:
If amp is so sensitive to the values as you say (BTW thank you guys for this conversation on RC an compensation, I really learn a lot)could be difference we all hear in FE's be down to the discrepancy -+ of components value ? I do not remember if I measured my c34 and cannot guarantee my compensation network works same as all yours. Possibly what difference in 2-5% of regular value could be big enough to hear the difference.
Okay, maybe ear bleeding highs was a bit of exaggeration, but the highs are very unpleasant and certainly inaccurate, not at all like real music, and no fun to listen to. There are elements of recordings that I listen to regularly that are almost inaudible with Caddock at R10 and C34 at 27pF. I think the FR is no longer linear, and distortion is high. With Riken at R10, not as much of the unpleasantness can be heard, but it's still there, and I've heard it from the very start. My first impression was that the FE sound had more definition but a sharp edge compared to the Rev C. With C34 at 22pF, the highs are better but still somewhat sharp. More values are on the way.
Regarding part tolerance, I don't think +/- 5% is enough to make your amp sound better or worse. However, I do notice some weird phasing effects because right now one amp has 27pF C34 and the other amp has 22pF. The 22pF amp is also louder. It also makes the image slide across the room from one speaker to the other as frequency of a note changes. So if you are unlucky enough to have one amp with lowest tolerated value and your other amp has highest tolerated value, say 25.5pF and 28.5pF, you might hear some images moving across the room as notes slide up or down. There are probably lots of worse things happening with room reflections and so on, plus all the other components have tolerances which tend to randomly cancel such effects.
To strive for accuracy, when doing these kinds of tests comparing the sound of one amp to another, I always switch the amps from one side of the system to the other at some point. If any differeneces I hear move along with the amp, I know it must be the amp making the difference, not the room or my ears or anything else.
Peace,
Tom E
Regarding part tolerance, I don't think +/- 5% is enough to make your amp sound better or worse. However, I do notice some weird phasing effects because right now one amp has 27pF C34 and the other amp has 22pF. The 22pF amp is also louder. It also makes the image slide across the room from one speaker to the other as frequency of a note changes. So if you are unlucky enough to have one amp with lowest tolerated value and your other amp has highest tolerated value, say 25.5pF and 28.5pF, you might hear some images moving across the room as notes slide up or down. There are probably lots of worse things happening with room reflections and so on, plus all the other components have tolerances which tend to randomly cancel such effects.
To strive for accuracy, when doing these kinds of tests comparing the sound of one amp to another, I always switch the amps from one side of the system to the other at some point. If any differeneces I hear move along with the amp, I know it must be the amp making the difference, not the room or my ears or anything else.
Peace,
Tom E
Tom,
we're going nowhere in this way... we need to understand where the problem you're earing lays.
Can you, please, try rev C. compensation with Caddocks in R10 (both channels) and tell us what you ear?
Also can you post some pics of boards and report differences from BOM?