Attack of the Redbox Preamp

[ced1]

And its now late and I'm out for the next few nights, so I'm going to get this mutha thread off the ground. Its about time after 2 weeks f****ing around with my abouttobecomeex friends trying to get me up on their servers. But we broke though in the end. So here goes:
And I hope you enjoy the thread.

Design intention:

< the Redbox Preamp is a bash at a serious no-holds-barred, world-class high-end preamp>​

No pressure then.

Context:
So, to set the context, the Redbox amp project started with the Redbox poweramp here (v6.0):
http://www.pinkfishmedia.net/forum/showthread.php?t=71372
and the continuation thread here (v6.1):
http://www.pinkfishmedia.net/forum/showthread.php?t=133120
THAT however will, errr, unfortunately still be a thread in progress for a while. Anyway, it led on to many little things, including a more thorough way of modding for me .


Preview:
lets have a quick cheat preview of the Redbox Preamp system before sinking into the details.

Just after xmas, the regulator and preamp boards arrived and are top quality. Way better than the mis-etched rubbish the last guys I used produced. Sweet! Those regulator boards are indeed pretty busy. Each board measures just under 175mm x 100mm.

Base assembly: The core of the preamp together with the regulator system. The simplest build version. Yep, its a sandwich construction again.

Up and running:

Trouble shooting was relatively completely painless this time. Just had to change 1 resistor value, doh.

Second assembly, currently running:

Fully loaded with decoupling, but still only 2 engines out of 8 switched on so far. Yeah yeah, the wiring trains are slopped around I know. shoot yourself in the foot doing them properly before the intention is to close the lid for good.

And that's as far as I've got to so far. The really interesting stuff starts now for me.


History:
Enough of peeking at the back page. Rewind back to the beginning with a little more history to this project.
So, what I learnt from the Redbox Poweramp project, together with ideas I've wanted to try more fully, led on to the development of this, the Redbox Preamp System 5. Composed of my Line preAmp Regulator System 5 (LARS5) and matching Line PreAmp 5 (LPA5).
*** God I really do come up with completely crap names, arf arf. Should call these things something macho-military or aristocratic like The Stealth Dark-Hawk Preamp or The Verdi Valeroso Conductore Poweramp. Actually those are astonishingly rubbish and pretentious, but frighteningly likely to be sitting on some marketing desk in Italy or the US right now. I shall rename them Fido and Tabitha.***

Anyway, in its turn, the poweramp project was based on my prior fumblings back in the preamp arena with, I guess I would have to call in Preamp System 4. It looked like this in the end:

which admittedly was a piss-poor contender for Bemused's heroic crown:

over which i wet myself with laughter. ...but nevertheless still getting electrically scary.

And therefore required a 'more elegant' solution or reconstruction.

Hope that's clear, and apologies again if this all sounds preposterous and/or Teutonic. In my defence I have hundreds of schematics, sims, variations of themes and circuit fragments I've played with. It's got to the point I've have to use a numbering convention for identifying my numerous versions of fiddlings to actually know what I'm going on about, in my own mind, never mind explain anything. I swear, I open a sim- I don't even know what I was analysing. Then I go into the project I find fourteen more variations on the theme.

Anyway, Preamp System 5 is more sophisticated than Poweramp System 6. Its my 5th major development of a preamp, each one a based on the modded-to-hell remains of the previous mess together with new things I've wanted to try. The DNA stretches all the way back to the original Air Guitariste of so many years ago. (still doing good on those names, buddy).

What was the plan?
So with cretinously named system 5 I wanted to optimise every aspect of audio design as best I knew how. Basically build the best freaking preamp I have imagined, based on the (variations of) the Naim/RCA circuit. And 18 odd years of modding gear by trial and error and pink fish and loafing around other DIY audio forums and research and learning about electronics and a bit of luck.
To this end the preamp board is also a development board in that I've built it to be easy to experiment with several configurations.
Regulators? I've thrown the kitchen sink at 'em. Then watched 'em land, signed it and sold it to the Tate for £3m.
Design wise, I've done everything I've imagined- I still look at the boards, and with 2 minor bug fixes, I wouldn't even begin to know where to look for anything significantly better. After this I'm out of gas, washed up, don't care anymore, nowhere left to go.
So I've tried to make everything absolute tip-top class across the board; component selection, regulation, layout, grounding, multi-railing, circuit design, power supply. There is nothing better in the way of sound quality that I can crow-bar out of a bunch of transistors and copper than this. Or I've really heard or read that anyone else can convincingly do.
Sure, anything can be refined, and I'm no genius, but from my listening experiences, limited understanding and just sitting back and enjoying toones, I am supremely happy. And as I write, I'm only half way to exploring what these boards can do. They are working, sounding utterly awesome, and ready to ramp up. Of course, might all go down the swanee yet in a screetch of unlistenable and unentertaining detail and 'holographic imaging', but am at a great place right now sonically, so its a win whatever happens.

Sure other circuit topologies may deliver more. I might one day try a folded cascode or an ultra simple zero feedback class A topology. But my choice has been to push the Naim/RCA configuration as far as I know how. I love the essence of the sound and presentation that Naim gear epitomises. And actually, just about everything here is really about optimising the conditions for a circuit to operate in- It should be applicable to any circuit.

Either way, so far the sound I've already got is so fantastic, that I am happy in my own mind that this is world-class high-end league. Cock on the block? Big words to back up, Mr bigger mouth, but I would be honestly bowled over if even this basic implementation of the Redbox preamp system 5 would do anything other than blow a 552 preamp into the weeds.

Watch this space as my pants and go up in flames, he he. Who cares, this is for pleasure. But anyway, why do I say so? Many reasons but here's a good un: Old news I'm sure (I've been away), but Naim are making big song and dance about their new discreet super regulator. Well I've taken a gander at their 'white paper'.

http://www.naimaudio.com/sites/default/files/products/downloads/files/Naim-DR_Whitepaper_May2012_FINAL.pdf

Even from the block diagram can't see it's anything other than a variant on the ALW/Jung bootstrapped series superreg. Article here:

http://waltjung.org/PDFs/Improved_PN_Regs.pdf

with a few tweaks to save a scrap of dignity e.g cascoded current source feeding the voltage ref. Even there, I'm not so sure that particular implementation is actually a sonically significant step. This shit floated pink fish ten-fifteen years ago, and the knowledge has moved on in leaps and bounds since then. Je ne suis pas impressed. The amplification signal-path, components and layout of what I have seen of a 552 is piss average and safe, Electrolytics, huge inches between components etc etc, and unless there is a box of magic under the pcb, I'm yawning. Last time I listened to a top flight Naim (laptop+dac/500PSU+552/psu+300/psu) on the front of £5k+ PMCs, yeah it was good, but not hyperspace. For big league money. As always, out of home territory, trickier to assess, but I stand by my impression. I say average from a point of view of having optimised a particular technical aspect and found a sonic gain. e.g moved electrolytic to film cap, ground layout gone from loose to tight, cascoded x, y or z, improved decoupling etc. Therefore not optimising that aspect= known lower performance. So, unless there is that 'box of magic' below decks...? I guess a strict A-B comparison at home would be the only way to know for sure. Anyone want to bring their 552 round, I will happily and publicly eat humble pie should the results merit. Otherwise, I'm happy to make wild assertions based on experience.

Enough willy-waggling. I'm ashamed at being so mouthy, so onwards.

First of all, its a preamp system. That means regulator board + preamp board designed and honed to match each other.

Design Features- LARS 5 regulator system.
24 regulators per channel. 48 for a 2 channel preamp.
Dual (+/-) polarity rails.
2 x 8 independent output rails (16 rails/channel, 32 rails for a whole preamp) feeding the audio circuits only.
2 x regulation paths feeding output gyrators independently at transistor collectors and bases.
Quadruple regulation/isolation path: gyrator feeding current source feeding shunt regulator feeding gyrators.
Both star grounds and ground planes.
Star supplying.
Critical feedback regulator sense points.
tight dual layer layout.
premium industrial components.
'Sandwich' mating to LPA5 preamp board.

Design Features- LPA5 preamp boards
Designed to work with LARS5 regulator board.
'Sandwich' mating to LARS5 regulator board- super-short supply paths of approx 10mm between boards.
Multi-railed power supplies.
Premium industrial and audiophile components.
Both star grounds and ground planes.
Short signal paths.
Short ground path layout.
8 possible circuit configurations.
Tight dual layer component placement.
On board decoupling.
High quality decoupling, footprinted for alternatives.
Dual mono.


Circuit configuration options:
a. Select between classic Naim/RCA and Bootstrapped circuit topologies
b. Select between normal and cascoded output transistor topologies.
c. Select between normal and cascoded VAS transistors topologies.
d. Select between normal and cascoded input transistors topologies.
e. Select between decoupling strategies.

There are also a couple more options that are interesting though I wouldn't class them as strictly circuit configs- choice of coupling capacitors, ditching the output coupling cap, choice of input transistor (jfet/bipolar).

All sounds like a pompous lot of old toot, dazzle dazzle, when put into words, but I shall explain these verbal fumblings. Its not actually rocket science and a lot of the preamp board stuff is pretty established electronics stuff. The regulators not so.
A lot of work also went into selecting operating points so the whole horrible mess would actually work- This is emphatically not a plug and play production.

Enough for now, its 12.30 and time for bed. More in a few days.

 

[GWM]

Welcome back Ced hope your not going to leave us all up in the air wondering how it turned out again have an airguitar around here some where very nice it was to.Agree with you about high end Naim stuff been a couple of years since I heard a 5 series system it was very nice but hardly jaw on the flaw.
Geoff

 

[timH]

Yes very interesting - I'm wondering if I should get some shares in Wima

 

[Bemused]

Ced what a marvelous post, I think you disappeared before I started bodging (2008) and can only recall stumbling across your threads by accident.
Looking forward to your further posts.

Tickled thst you laugh at my most serious effort, link was not working but I assume it's my Buffalo

Tony

 

[natzev_i]

Ced, great project - will the schematics, PCBs, BOM, etc. be available for the general public here, please?
Cheers,
Ivo

 

[ced1]

Well vicissitudes of life and all that, but short of that this thread and the poweramp thread will be going to the grizzly end. Glad I'm not the only one underwhelmed by the high end Naim gear though. I sometimes think Im missing something. And yeah, definitely; I would advise buying into Wima- I hear stockholders get 10% discount on container load orders, heh. Schematics, part selections etc will all be explained as the threads run their course but that's for next week. off to work.

 

[sq225917]

Colour me interested.

 

[Telstar]

Subscribed

And wondering if the preamp would be suitable to drive headphones.

 

[martin clark]

Excellent, Ced's back

 

[demotivated]

Cripes, one more in the spaghetti club.

 

[ced1]

Hah, cheers Martin. And yep indeed we must have that pint some time soon. You still in Bath I assume. And still inveterately posting here by looks of it.
Headphones? Now that is a question. I have some AKG701s which are a dog to drive.

Anyway, onwards:
LARS 5 REGULATOR SYSTEM
Design in detail
So lets start with the regulators. This is my 'next gen' on from the Redbox poweramp regulator board 6.
This is the preregulation section, and has 3 parts:

http://www.naimmods.com/2 Attack of the Redbox Preamp/0 LARS preregulation.pdf

Raw DC
Raw DC in from my psu floats around 34.5 - 36V (pretty close to a Highcap and actually the ideal raw DC for this thing. Max input voltage for the board is 45v, determined by the gyrator pass transistor's Max Vce rating. Min is ideally more than 33V (I want the zener to be active), but one can go a bit lower.

The regulation/isolation starts with

The gyrator.
I have ramped this up from the poweramp version to a 'heavier duty' 2 pole one from my prior lightweight choice. Pass transistor is a C class BC550. I wanted a high gain transistor to reduce the current draw through the base filter resistors because I have increased their value somewhat from earlier iterations- as I really wanted to keep a low dropout gyrator as the reg board as a whole gobbles voltage headroom. But I've found the 2nd pole of the filter works best around f3=2 Hz or lower. I have experienced significant impact on sound quality with higher f3 so I consider less simply under-filtered. So my choice for the 2nd pole of the filter was 20k/4.7uf
A 33v zener fixes the base voltage (though its specific value is dependent on current load). R42 was chosen to limit current through the zener at worst case scenario of 45-33=12V. The zener is a 1.3W part, so a max current of 5ma is acceptable as this gives 33x5=150mW power dissipation. C20 forms the first pole of the filter; 5 to 20uf is good in this position (I had the poweramp oscillate at a 47uf value once so I'm uncomfortable anything above 22uf and this has long been touted as a good value by dual pole gyrator folks).
I wasn't worried about zener noise as it gets viciously filtered.
Anyway, I wanted the output voltage at the transistor emitter to be stable throughout the frequency spectrum. Gyrator high frequency performance of this implementation is excellent but regulation is generally shit at low frequency, because gyrators are intrinsically pretty high output impedance circuits- Their output voltage rises and falls (they sing along [in phase]) with load current demands. Changing back to an old school 2 pole filter slugs as much of the low frequency noise as possible.

Next comes a
Constant Current Source (CCS).
Now the beauty of following a gyrator with a current source is that the gyrator has a relatively lousy output impedance, but a CCS obviously has such a huge impedance that the gyrators output voltage becomes rock solid (stops singing along). This is a virtuous circle where the stable load current of the CCS allows a very stable output voltage of the gyrator and the stable voltage output of the gyrator allows superb current regulation of the CCS. Its a real dream team combo. Remember, the output of the CCS is also fixed by a shunt voltage regulator as we'll see shortly, which is the perfect operating condition for a CCS- sandwiched between 2 fixed voltages. Furthermore, I also ramped up the CCS by cascoding it. This improves current regulation and frequency response massively. Note the CCS biasing string is itself only loaded by a resistor not another CCS which is customary. Because if you think about it its a very stable operating space. It doesn't need one IMO. I have a sneaking suspicion its also more effective without one. It is selected for a draw of 1-2ma. Ive just used green LEDs for biasing, the actual CCS decoupled with a decent sized capacitor to mop up any low frequency noise that might make it through the gyrator.- the stable voltage across R22 to base of Q24 is what matters for the CCS.

At this point I think a quick review of regulation in general is required.
I'd like to explain the whole design principle behind the LARS5 regulator boards. In a sentence this is:

<High frequency line regulation>​

Brutally filtering the **** out of the mains supply into the ludicrously high Mhz region. Maybe as much as a thousand times beyond the audio frequency range, maybe more. Out to a measly Mhz just doesn't cut it for the top quality stuff. I've come to the conclusion that that is the single dominant factor that determines the sound quality of a regulator. Load regulation is relatively irrelevant as long as its in phase. I may of course be wrong but bear with for a bit.

Some musings on regulators that led to that conclusion.
a. What is a regulator?
A regulator sits between the power supply and the circuit. It reduces (rejects) both line and load noise- Noise coming in from the power supply line, and noise created by the varying load current demands of the audio circuit it drives.

b Noise
This obviously covers the whole frequency spectrum from 0 hz - Well, to what indeed? At what frequency does noise cease to affect an audio circuit. I'm not so sure there is a limit. My experience suggests at least 10-20Mhz. If the Vertex AQ stuff is correct, and the gear seems to get damn good reviews, well check out the datasheets for EMI shielding materials e.g copper tape. That's really only specced in the VHF frequency range.

So possibly much higher than 10-20Mhz. Anyway, line noise from the mains covers a damn wide spectrum. I have a graph somewhere which I'll post if I find.
Noise into the circuit can really only come from the signal in path, the signal out path, the power supply and intrinsic noise of the circuit itself.Signal in noise is really not going to be much beyond the audio frequency. There's the source music signal and perhaps environmental noise e.g from oversampling dacs. In a Naim circuit however, the input is filtered by a polystyrene cap RC filter and starts rolling off beyond 60khz IIRC. That will kill anything dead higher up the range.
Then I guess noise can inject from the output of the circuit? I'm gonna make a punt that it relatively low level audio frequency stuff working its way back from the poweramp.
Then there's load related noise. So, what is most significant source of load noise going to be? Noise from circuit current demand mainly-The music signal. Hold that thought.

C. Regulation methods
broadly, there's 2 ways of regulating noise- active and passive.
Active uses some form of feedback to measure output voltage against a reference voltage and adjusts it to match usually via a voltage divider. Being active its output impedance is low, but consequently doesn't normally reach high frequency. They also have phase shifts, overshoots and a transient response time lags.
Topologies and variations include series, shunt, bootstrapping, single ended, opamp and LTP error amps and various tweaks to circuit optimisation e.g. filtering, current sources etc. Not going to go further into the details here.
Passive methods use things like capacitors, inductors, filters either alone or in conjunction with silicon e.g. gyrators. More characterised by high output impedance, but higher frequency performance depending on construction.

d. Regulator circuits and sound
So how do things look in terms of regulator circuits and sound?
LM317s: Ok back at the dawn of the internet, I started with LM317 regulators. They were utter shit. Never got a single bit of joy out of 'em despite building several PSUs. Maybe it was me but remember the 'mystery' surrounding the supercap LM317s, the special selection testing bla bla bla. Chasing spectres. I auditioned a Naim Hicap at Grahams (fab dealer) but thought he was sadly having a laugh- £800+ for something I couldn't even hear the effect of? He swore blind I had to trust him on this, and in my system at home I would begin to hear what it did. It probably would have. A bit. But I trust ME. So I spent the money on holidays instead.
ALW superregs: Then the wonderful ALW superregs arrived. A boostrapped linear reg based on Walt Jungs design.

http://waltjung.org/PDFs/Improved_PN_Regs.pdf

And that was a quantum leap forward. Oh my god, this was where we should be at. Just a clear no brainer improvement. All that 'does it sound a bit better? I'm not sure.' nonsense swept away under a blazingly better sound you could hear within 3 notes. The numbers blew away the LM317s too, way lower output impedeance, way higher frequency response. Then to improve matters, there was the C5 mod; output impedance rose and frequency response dropped, which didn't bode well, but oh lord did the sound improve again. In fact sound changed so much it took me about 3 days to convince myself something wasn't broken. ALW's conclusion was linearity was much better- no capacitor in the feedback loop, and this was important. So the conclusion I drew was that in terms of load regulation, phase/linearity was important.

Gyrators- Then came John Luckins and the VBE regulator/gryator/capacitance multiplier. I kept the post: I removed the 1086 tracking reg and replaced it with a Vbe reg on my 62 power supply. Not sure I've got the optimum balance of filtering and output impedance of the Vbe yet, but listening tests are very promising. I was up till 3:00 am listening last night and no fatigue. More detail and beauty to the sound. No downsides so far... John.
Now to be fair Les Worstenholm had and does use these and the circuit has a long history in electronics generally, but the trick was in the 2 pole filter. This circuit got played around with a fair bit on pink fish, by Teddo Pardo amongst others and a popular implementation was the darlington transistor arrangement with 2 pole RC filter. Relatively good output impedance due to the high composite HFE but also good filtering.
Popped that in to test and well well. To be honest, yep was just plain better than the ALW superreg. Not dramatically so, but most notably high frequencies were just smoother and no downsides. Bass still stonking, great clarity, etc etc. So really why make a complex ALW when a simple darlington twin pole gyrator would do? It also prompted some implications- The ALW had way better output impedance but the gyrator still sounded just as good/better. Therefore maybe output impedance wasn't that important? Having stuck a hig power gyrator in front of a poweramp output transistor with no ill sonic effect then modelled how the rail voltage plunged under load, I have to conclude output impedance is not a big issue, though I have some caveats to that statement.

I played around with lots of different components and values over a period of years and found the darlington 2 pole to be remarkably robust. Really as long as the filtering was around 10uf/100K/0.1uf they sounded good. My slight preference ended up being for polypropylene filtering caps as they just sounded a bit better, but there was really nothing in it. Do what you might they just sounded a flat great. couldn't get to sound worse, couldn't get em to sound better.
Then I went to multi-railing topologies- every part of the circuit got its own power rail. That was a big win, sound wise.

Low current Gyrator;
After a while I started reconsidering the gyrator again, and looking at the preamp circuit and starting to work out how they worked in more detail. It struck me that the preamp uses tiny amounts of current yet here I was using these massive 25watt D44Hxx series transistors so supply really uA or the odd mA on each rail. So I cut out and bypassed the D44s with a bit of wire and boooom a massive leap in sound quality. Now that was really interesting, because one of the things about darlingtons is you trade bandwidth for gain and power handling. And when you model preamp current loads in a multirailing context you don't need much gain or power handling. But by my reckoning I was getting a whole heap more bandwidth. So as a ballpark if your transistor FT is 400mhz and its hfe is around 400, you're getting full gain out to around 1Mhz, dropping slowly. And then I started musing over how hi freq line noise would get through a transistor and it seemed to me really via collector base capacitance. Which is way smaller in a small signal transistor than a big TO220 though some Sankens are impressive. And that's really what goes on in the redbox poweramp 6. The values have been played around with a lot, as well as decoupling and circuit values and component selection, I keep resistance up and capacitance down, but that's the heart of it

High bandwidth gryator
So then on reflection, it was really starting to look like high frequency performance was what was making this sound so good. So the question became how to increase bandwidth of the gyrator transistor. And so I looked at some very interesting Sanyo transistors with FT nearing or above 1GHz but the gains were pretty low at around hfe of 30. and they had reps for being a bit unstable. And you do need a bit of gain otherwise you start having to use large filtering capacitors which are expensive and tend to be slower so you're staring to lose high frequency performance again. I look for around hfe150 min. And settled on the cascode configuration as a way of getting a boost to high frequency performance.

http://en.wikipedia.org/wiki/Cascode

http://www.naimmods.com/2 Attack of the Redbox Preamp/0 output gyrator.pdf
And that's the heart of the idea behind the LARS5 regulator board. So the question is, does it work? Read on. Another day.
cheers
Ced

 

[Markus S]

Excellent, Ced!

 

[PigletsDad]

Ced,

From the day job, shielding above VHF becomes more nearly optical in behaviour. Even very thin metal gives complete shielding, all the way up to visible light, and so residual leakage is entirely to do with gaps and holes. Once a hole is about half a wavelength across, it is effectively open.

There are lots of big UHF and low microwave signals round, from TV, mobile phones, Wi Fi, DECT cordless phones and leakage from microwave ovens. This are in the range from say 800Mz to 3GHz or so; if anybody use 802.11a, then you get 5GHz too, but that is less common in Europe.

Anything above about 10MHz will not be effectively filtered by the Naim RC arrangement; the traces are way too long, and the series inductance stops all the filtering. Keeping VHF and above out needs ground plane construction, and feedthrough capacitors, rather than leaded ones.

Something like

It is also worth noting that closed metal boxes make high Q UHF or microwave resonators; these will couple strongly at their modes.

 

[ced1]

Hi PG,
Ahhhh, I was actually wondering about that, thanks very much. An idea I have been mulling over is putting the boards in a faraday cage and running a line direct to mains earth to try out this Vertex AQ type stuff. The cages I've seen are mainly screens though, and wondered why. e.g.
http://en.wikipedia.org/wiki/File:Electromagnetic_shielding_inside_mobile_phone.jpg
Your explanation is very concise. Also good to have some ballpark numbers. 10 Mhz? Not so bad but does now beg the question, if I lifted one lead of that cap and went to ground plane instead of star earth.... Actually would be more than pleased if you or others are able to inject some likely real world numbers. For example I have been wondering about at what point bypass capacitors cease to function effectively. For example I love the non-inductive Wimas and they have slew rates ranging from say 60V/us for an MKP4 here:
http://www.wima.com/EN/mkp4.htm
to up to above 11,000v/us for FKP1s here:
http://www.wima.com/EN/fkp1.htm.
Seems to me the FKP1s should bypass much higher frequency noise, but on an actual board, what can one expect real world? Those are numbers I'd like to have, to see how my lines of reasoning hold up.

And in fact it brings up a wider epistemological discussion point:

To be clear, I am far from proving my lines of reasoning by measurements and that's not my intention or purpose. Digging a bit deeper into the cascoded gyrator for example- it works without doubt from a sonic point of view. (Well, I say that. I seem to have a reasonable track record from comments and takeup here, and the whole point of posting here is to give people likely successful mod ideas to try out. Don’t take my word for it, try it.)
Anyway I have tried various combinations of preregulation as I'll describe later with the cascode and non-cascoded configurations so the cascoded configuration is the key factor. The actual reason why is not necessarily so clear and it would have to be tested and measured by experiment to narrow it down. For example I groundplane C32 and starground C33. The former is an smt ceramic the latter a polyprop. Why, because I wanted max bandwidth filtering on the casode transistor Q41 and the cap is the link to common. But I have chosen stargrounding to eliminate intermodulation between output transistors Q42 which is the point of multi-railing. SMT for very low inductance, stacked polyprop for still low inductance but also the 'sound' of polyprop. To me, you are more likely to get intermodulation on a ground plane as the currents on a groundplane are a shared path. Jonathan Carr also recommended ground planes for high impedance nodes (which I've chosen to treat as decoupling) and star grounding for low frequency (signal level), though the difference was slight. From his description he had tried, measured and listened to both cases in an A-B comparison which was good enough for me. Either way I'm covering my bases.
But alternative explanations might be that that the lower inductivity of a ceramic and groundplane is what matters, or the lower filter knee of the cascode transistor gyrator. Mabe it would be slightly better if both caps were groundplaned. Or stargrounded. Maybe none of those would make a difference. So many combinations possible.
My rationale on how to deal with this disconnect is that this stuff is time consuming enough as it is and one would have to try out, measure and then listen to probably 6 or 8 combinations to work out if there was one particular aspect that was critical or the whole lot. That’s a whole lot of scopes, prototype boards and time and effort etc. It may be satisfactory intellectually because one then has measurements and theory to back up or refute a line of reasoning. Practically it might mean I can drop a couple of unnecessary components. But on a cost/benefit analysis, I'm wasting my time. At the end of the day, it doesnt matter because what I'm trying to do is build great audio gear not submit a research paper. I try stuff out, more often it doesn't work, sometimes it does. If it doesn't sound better, doesn't matter what is happening in theory or in measurements; it doesn't sound better therefore is of no use. It is slightly nagging to think that maybe reducing that value there, or a slightly different routing there might make a huge difference, because I don't really understand whats happening at a very deep level. But that way lies insanity. So I get a win, I take it and enjoy it for what it is.
The more I read around the subject, the more I get ideas and possible reasons for what I hear. Like all humans, I try to make sense of it. The reason I tried cascoding for example is I read posts, again by Jonathan Carr who to my mind is a serious heavyweight audio designer- he has the EE education, he has experimented and measured, he has listened, he doesn't have a dogmatic approach, he makes a living out of this shit and he sounds like a perfectionist. Anyway, he discussed cascodes and poo-poos the nay-sayer who claim they sound ‘steely’. He interestingly states that he has measured oscillations as high as 500Mhz IIRC which he thinks may be why some people don’t like em. Now that sort of attention to detail is way more than I can invest nevermind the scopes that can measure that. So I just tried it briefly on the input transistor on a preamp a few years ago and it made a huge sonic difference-incredible clarity and precision. Various reasons I didn't pursue it in any depth at the time but it has sat there ticking away. Then from a schematic for a poweramp I pull the idea of a damping resistor for the cascode transistor as the designer also considered oscillation. And a couple of years later, I'm mulling over why gyrators might sound better than superregulators; Bandwidth... cascode... that worked on the preamp... thats gotta be worth trying.
So although my modding cuts corners and is certainly a bit speculative from an EEs point of view, it’s not randomly throwing components at a board, spraying with solder and seeing if anything happens either. And it has one benefit- I only go with what I try and am happy clearly sounds better. And yeah, I try not to delude myself, so try to go back and forwards a few times between mods to ensure I am clear in the differences I hear. If that were to be a fairy cake with 2 wires stuck in it, I would feel more than a tad uncomfortable about it, but I trust my judgement when I listen.
I guess I would say I mod on the principle that an ounce of practice is worth a ton of theory. Nevertheless if you can't marry theory, experiment and experience, you can't claim to truly know what you're doing. However, ditto audio engineers that don't listen, which from what I hear is not uncommon in the industry. I read some great anecdote somewhere about an audio brand owner hiring an electrical engineer and having to teach him how to listen to gear before letting him loose to design. That’s my way of saying, don't shoot me if some of my reasoning has the odd crack in it; at the end of the day I'm building audio gear to listen to for entertainment. If I don't enjoy it, if it doesn't grab me and move me, it aint no good.
But conversely It would be very informative if anyone qualified spots and fills in gaps, makes interesting points, can share similar experiences etc.

OK for now
Ced

 

[PigletsDad]

If you look at the spec sheets for capacitors, they will often have an impedance against frequency curve, which goes through a minimum (the self resonant frequency), and then rises when the lead and internal inductances take over.

See http://www.avx.com/docs/techinfo/cap_sel.pdf for some more details, and scary graphs.

 

[ced1]

Hah, bloody wima datasheets don't though. Not enough scary graphs if you ask me the slackers. I've worked to a ballpark of 60V/us slew rate will be the quarter wave slope, so equivalent to good decoupling up to 15Mhz, 1v amplitude before the caps run out of steam, turn inductive whatever. Could be quite wrong.

Anyway, because it could be getting a little philosophical, technical and dry, and its about to get numerical in a beginners kind of way looking at component values and operating points, here's a few on-the-ground listening impressions 2 and a half weeks in. Make of my ramblings what you will.
So running Redbox amps, pre and power.
Well, turn-on thump is a thing of the past. I mean no crack, no whoomp, no nothing. Switch on the poweramp and there is the faintest little 'bpmh'. If you weren't paying attention you wouldn't know if the amp were on or off- A benefit of the soft start afforded by the slow rise time of gyrators.
Speaker hiss? Gone. I guarantee you. Well, you have to put your ear OVER the tweeter to get the faintest sssss. Couple of inches away, 5 cm no more, and nothing. Anyone starts twittering on about Johnson noise, hmmph; that is least of your worries IMO. Powersupply, powersupply, powersupply.

Listening- what I am finding weirdest and most wonderful is the synesthetic affect the amps have brought to music. The lines between listening, feeling and seeing seem to have gotten blurry. On some tracks, good bass lines come out of the soundscape and literally dance and writhe on the carpet in front of me.
Crank up Leftfield's El Cid off R&S to 8.30 on the volume. I know I can get up, walk over and literally lie on that deep synth bass line as it suddenly detaches, comes out of the sound stage at me from 2 o'clock high, swings to and sits on the rug in front of me for a moment, vibrating. Before melting away... and already the next run is coming in, this time from 10 o'clock high, again swinging round to sit in front of me, vibrating like the engine room of a big ship. Back and forth, back and forth, eerie I say.
Music is so... so... tangible? Perhaps more as if reality almost gets a little spongy and malleable instead. A mate came over last weekend and I gave him a quick blast to explain why I'd been off radar and what all the geeky 'electrics' was lying around. 30 seconds in, he turned round, gave me a big grin and said 'you've really got to get stoned and listen to this'. He is a peasant.
I spent yesterday evening going more jazzy/acoustic, bit of Khaled, Ragu Dixit, Blue Note, Nor**ahem Jones, and so on. And one of my long term fave albums is Tourist by St Germain, if you know it. First track, Rose Rouge is a wonderful overkill of brush work on cymbals along with sax and other stuff going on. But the wildest thing was the brushwork- so (sur)real, tangible; it started creeping out and before I knew what was happening, I was sitting in a pasture of metallic pampass grass, hearing the swoosh and sizzle of the wind brushing a thousand grass-like cymbal-leaves all around me, completely immersive and touchable. And no, I was completely straight! But maybe the mains were good.
Tonight I've been carrying on and wheeled out the old Buena Vista Social. Now that is a great recording, you can hear the room behind the speakers in classical audiophile, 'get up walk around the stage' holographic imaging sort of stuff. Nice, but doesn't particularly float my boat as an effect. Nice because the clarity of the instruments makes them really easy to follow though.
But this? This is more like ultra-Naim. Forward, not in an in-your-face sort of way, but to the point of taking on real corporeal form. So I spin The Egg's Travellator, a brilliant funky, jazzy, electronic mish mash. Very groovy, very cool and loungey. My sister's mate had them playing at their wedding; damn I missed that. I guess not being invited might have been a problem.
Anyway, so the instruments and sounds are coming out of the soundstage. I want to touch them, I can feel the heft and physical shape of the notes of the slap guitar or whatever the hell that instrument and pedals is- fat chunks of pyramid shaped yellow wood, jabbing in and out of the sound stage with each string pluck. But of course I can't because its sound, not matter.
Like I said; synesthetic. That's the best I can explain it anyway. Completely addictive, keeps me transfixed and captivated in my listening. That sort of realism don't come for free though. You have to imagine how utterly perfect and present and burnished the music is. It's just already there, whatever should be. Too fast to have any sense of speed or pace, too lithe and articulate to be rhythmic, too finely honed to have any edge or sense of distortion or artifice. Whatever the signal demands, is. No fuss, no effort. Very very nice. Very very enjoyable to listen to.

And enough touchy-feely for the moment. Back to the circuits next and tie that up before going on to the build and testing. Then we'll look at the preamp, circuits and build. Stick em together, and then more toot- listening notes.
Midnight so over and out.
Ced

 

[orangeart]

Sounds brilliant, alas all digital these days here so no preamp needed.

Looking forward to the amp write up though. Maybe we'll all replace our hackernaps with cednaps? Although there seems to be a slow creep toward the new generation of class d amps.

Stefan

 

[ced1]

Hi Stefan, yeah indeed your points are well taken. Technology moves on. But really there are many ways to skin a cat. I see digital as just another possible avenue for doing so.
So for example part of my day job includes managing a design studio. Now digital has completely changed working practices from 15 years ago; days of paste up, camera ready copy and so forth. But actually the finished product is not materially different and the creative process remains the same. The production process is merely much faster, needs far fewer people, is more flexible and is way cheaper. But certain exotic products, such as may be purveyed to her majesty, are still made the old way, on a letter press for example. Hifi wise I would expect similar. The primary real benefit I can see of switch mode and class D, is the ability to make relatively cheap PSUs (no big transformers) and produce high output watts cheaply- v efficent, small heatsinks, output devices etc. So interestingly, in last months HiFi+, they reviewed the new Densen integrated. And that was compared as equal in sound quality to the Devialet D premier which has been touted as the 2nd bloody coming. From the described comparison I would probably end up buying the Densen on sonic character.
If one wants to go for all-in-one box solution, I see no difficulty in building a really premium integrated. The trick would be an appropriate DC powersupply- big tranny with secondary 4 windings and 8 reservoir caps is how I would do it.
But so far, Ive really rambled on about the regulation system and that has been my primary interest. My view is that is applicable to anything line level. So later in the year I will get some universal bessel filter boards made up which a LARS5 should be perfect to drive. So, a real good discreet component CD output section a la CD555 is then viable to pop into my modded 3.5. I haven't bothered with 729 cards in a preamp for years but that would be another. But my other aim is active filters to go direct in my preamp and go active- I have all the parts necessary for 4 mono amps already. Then I have sketched a sub design which should be awesome. My real sticking point is high current delivery for the sub. I have a sweet idea for an output board I could drive using the exisiting poweramp but it hasn't simmed as expected so far. I'll need to work that out at some point. That's all for a fair way down the line however.
I guess point is; for me a preamp, and this project, makes complete sense. If the cascoding on the pre sounds good and is balanced in terms of listening pleasure, the Bessels I make up will also use cascodes.

Anyway, I digressed. Was going to start looking at operating points but that will have to wait for the moment. Happily I finally found the graph I was looking for:

Brings me back to the theme that I believe the ability to filter way beyond 1Mhz is critical to real high quality sound. I remember back in the dawn of the internet Julian Vereker posting a comment on the original (and best- ah fond memories of that) Naim Forum to the effect that
'rectifier switching spikes were a devil to control!'

Not 'arf! Nevermind SMPSs and running a DAC via a computer!

One begins to understand the excellent thinking Naim's construction strategy in their DAC (the only product I now lust for in their range):

http://www.naimaudio.com/sites/default/files/products/downloads/files/naim_dac_august_2009.pdf


ok for now
ced

 

[ced1]

The shunt pre-regulator
The last bit of the puzzle, which comes from the PRS6 regulator, is the shunt pre-regulator. Shunts are pretty de rigeur these days. I have about 20 versions of circuits on file including the huber, salas, paul hynes, John Lindsey Hood, Twisted Pear's placid, etc etc etc. They've been around.
Walt Jung did a nice article, here:

http://waltjung.org/PDFs/Dont_Shun_the_Shunt_Regulator.pdf


In my mind I wanted the shunt as a pre-regulator on the LARS5 board to do 3 things:

1) ground current regulation.
A key benefit of the shunt; it forms an extremely small current loop with the load, bypassing a lot of the the influence of the psu, and effectively 'regulates' the ground line:
"a constant current always flows through the ground line even if the same or different current changes take place in the load, hence the potential [voltage] of the ground line exhibits no change"
If you want to get your head more around shunts, read US patent 4,366,432. Its a shag reading patent speak, you almost have to translate it, but is I feel well worth the effort.
This is what you can expect:

It acts as a virtual battery, but with none of the drawbacks of a battery such as being noisy under active load. I view the CCS as effectively acting as a local psu, working in much the same manner as Mr Tibb's Tracos, but without placing switchmode circuitry into the mix.

2) low output impedance to at least 100khz
The output impedance was important for me because this is a multi-rail design and I wanted to kill any intermodulation between output gyrators and therefore the circuit fragments they feed; brutally so at lower frequencies until the gyrator filters could really ramp up effectiveness at higher frequency and look after themselves. The point of multi-railing is that no circuit fragment can modulate any other through current or voltage. In the pre for example this would mean that the input transistor biasing string is unaffected by large voltage or current swings of the output transistor (Well I say large, I mean relatively large differences- we're talking uA Vs ma). In practice the input transistor biasing string receives its own independent power supply. So does the input transistor, so does the VAS (well partially), so does the Class A current sink for the output etc. In the case of the LPA5 Preamp board so do the cascode biasing strings.
Sure, we're talking about uV and uA here but damn, it makes a huge difference to sound quality.

Naim have done this with the higher end PSUs though at a board level (gain, buffer, 729 filter etc), but the Lyra Connoiseur amps took this to a higher level with 24 regs in the 4.2L.
The starfish preamp also if I remember? Anyone?

Anyway, this is what can be expected from the shunt reg in terms of output impedance:

Very very good, depending on pcb layout*, and just what was needed.
*Minimising output impedance depends on tight layout and accurate sense point positioning. I laid out the LARS5 boards with a lot of consideration for these issues- to the mm, sensed at both rail and ground star points, exactly where the loads come together.

3) low noise to at least 100khz.
I wanted low noise out to around 100khz as this is the region where the output gyrators are weak:
Some ballpark numbers for the gyrators to put this into perspective:
The output gryator RC filter values I have selected attenuate voltage noise by roughly these figures:
At 100khz by at least 60,000x That's 1V amplitude attenuated to about 16uV. Nice.
At 10khz the gyrators only attenuate by around 1/6,000- that's no longer great.
At 1kHz by only around 1/600- pretty grim in the scheme of things.
At 100Hz, by 1/60. Terrible.
So at these lower values I needed the shunt to slug line noise. But like I said, above 100khz, I'm happy the output gyrators get into their stride and keep going right up to frequencies where parasitics roll them off as piglets dad clearly identified.
Happily this shunt is brutally efficient at low frequency being still around 100db down at 100khz:

Apart from that, I was not per se interested in its direct line/load rejection capabilities (though they are probably inseparable) as those are rendered irrelevant by the output gyrators never mind the input gyrator and CCS.

 

[Bemused]

Excellent posts Ced, most interesting.