Attack of the Redbox Preamp

[ced1]

How the shunt works
Here it is again to refer to:
http://www.naimmods.com/2 Attack of the Redbox Preamp/0 LARS preregulation.pdf

Its not a particularly complex circuit, though there are a fair few factors you need to balance if you start tweaking it. For example, reduce the gain of the shunt transistors too much and you can see how noise elimination collapses in sims and noise creeps into the ouput waveform.
***And note another benefit of using a shunt regulator- it has current limiting, not as an add-on protection circuit that detracts from overall performance, but as an inherent part of its functionality and actually part of its high performance. The preamp and regulator are protected even in case of a short, as long as the relevant parts are rated and heatsinked to handle worst case scenarios- full on and off load.***


in operation
Overall its an error amp composed of long tailed pair which 'compares' the output voltage as measured across voltage divider (R53/R54) on one side against a filtered reference voltage (U3) on the other.
If the output voltage goes high, the darlington shunt transistors (Q30/Q31) conduct more current to ground until the output voltage drops again such that voltage across reference U3 and R53 are equal again.
If voltage goes low, the shunt transistors will conduct less, allowing more current to flow to the load instead of ground and so increasing voltage again.
Voltage drops and rises because available current is limited to that set by the CCS.
Should the load impedance drop such that load current demand exceeds that supplied by the CCS, the circuit ceases to regulate- the shunt transistors no longer pass any current and the output voltage will continue to drop until a point is reached where the load current demand has reduced to that which can be supplied by the CCS.
Offload, the full current of the CCS shunts to ground.
Pretty high gain is required of the shunt darlington pair as the drive current is also an error current to the error amp. I've chosen composite hfe of 10,000.

Overall, I've kept the shunt simple and avoided peppering it with current sources, because I'm not convinced its needed, and the LARS 6 board is complex enough as it. In a direct a-b comparison on the PRS6 it clearly beat an ALW superreg as a prereg and that's all I was looking for.
As an aside, In fact this shunt is in some ways a more elegant implementation of a superreg anyway- In the ALW/Jung, bootstrapping (using the low noise output to regulate the error amp) is achieved via a quite temperamental bolt-on current source and messy level shifting with decoupled zeners. As I remember, pre-regulation on the ALW/Jung made a huge difference to the ALW- due to the poor current regulation of that current source IIRC? Here it's intrinsic to the shunt circuit, and the current source is not a bolt on but improves isolation. I digress.


LARS5 Operating points and component values

So at this point, let’s look at some operating points, component values etc. for the regulator boards and how to set them up.

a. set the shunt output voltage
Shunt output voltage is set by the voltage divider R53/54.
The voltage reference (U3) is a 6.9v buried zener (very low noise, low dynamic impedance).
That means you get approx 6.9v between the base of Q27 and V+out- the base sits 6.9v below V+out.
As the Vbe of Q27 is about 0.7v you get approx 6.9 - 0.7=6.2V below Vout at the emitter of Q27 and therefore across R51, whatever the output voltage might be.
R51 therefore behaves as a current source, and is chosen to give 1.3ma of current.- One doesn't need to build an extra full CCS with transistors, biasing etc. as the current source for the LTP. Very elegant circuit design. But means you can't just plop in jfets willy nilly instead for example.
As the emitter of Q27 and Q29 are directly connected, they must be at the same voltage and therefore their bases must also be at the same voltage. So the voltage across R53 will always be 6.9V - the same as the voltage ref U3. That's the equilibrium state for the error amp, and why the regulator works; it always tries to maintain this equivalence.
6.9V across R53 generates a current. As the base of Q29 effectively gobbles no current that same current will pass through R54 and generate a voltage across it also. So by choosing the value of R54, one can determine the voltage across it is and therefore what V+out will be.
For example, if both R53 and R54 are 1k, then the output voltage across R54 must also be the same, 6.9V, so the total voltage across both resistors must be 2 x 6.9= 13.8V.
If R54 is 2x R53, so R54 is 2k then the voltage across R54 must be 2x 6.9V so the output voltage will be 3x 6.9V=20.7v.
However, DC impedances of both sides should be matched as this is a bipolar error amp so significant currents flow through the bases of Q27/Q29. Imbalances add to the 'error budget'.
Impedance of R48+R49 should equal that of R53//R54. R49 is a damping resistor to ensure no oscillation of Q27. Its probably not necessary as C24 is polyester which I've not had troubles with, and is in some ways undesirable, but rather be safe.

b. What output current should one set the CCS for?
The max that the CCS can provide is determined by the limits of the transistors in the current path.
The gyrator pass transistor is a BC550 which has a max Ic of 100ma if you look at the datasheet.

http://www.datasheetcatalog.org/datasheet/fairchild/BC550.pdf

The CCS is composed of another BC550; and a 2sc2705 cascode transistor which has a max Ic of 50ma from the datasheet.

http://www.datasheetcatalog.org/datasheet/toshiba/3203.pdf

So the max current the CCS can supply is limited by the 2sc2705 -50ma. Obviously it’s always sensible to go below max ratings so I’ve set an upper limit of 45ma.

Is this enough?
To answer that we need to know that the max load current demand of the preamp board will be. I’ll look at that later in detail, but the answer is nuclear worst case scenario- the LPA5 preamp uses around 30ma each for the +ve and –ve rails.
[Overkill worst case scenario; +/- 5v output swing out of the preamp driving the 18k input impedance of a normal Naim poweramp e.g Nap250- see page E37:

http://www.naimaudio.com/sites/default/files/products/downloads/files/amplifiers_reference-manual_english_issue5_0.pdf

Obviously that is way more than is required but good to know there is plenty of margin. Drive current is always useful to have in reserve.
Back of a fag packet: Naim poweramps can’t really take much more than +/- 50v rails without almost certain thermal runaway or just plain blowing the output devices. Done that a few times so I know.
45V are really the max I’d be comfortable with and NAP250s run at 40V rails. Voltage gain of the poweramps is about 27x IIRC (have to check that) so max input voltage the amp would sanely need is 45/27=1.66 rounded up to 2V]

more shortly.
ced

 

[ced1]

Now if you read around, the general recommendation is to be shunting at least 25% of the total CCs current to ground in operation. Based on that the CCS should output at least 30 x 125% = 37ma. BUT the shunt error amp also uses current. How much?
Well, from the datasheet the voltage reference has a minimum operating current requirement of 0.6ma, max 15ma.
http://www.ti.com/lit/ds/symlink/lm329.pdf

Obviously we want to be safe but not waste valuable current so I set its load resistor R47 to deliver 1.5ma at 20V regulator output voltage though the reference. So even if the shunt is set for an output voltage of say 15v not 20v, load current drops but is still fine. So that’s 1.5ma for that circuit fragment.
Then the Long Tailed Pair (LTP) is set to 1.3ma as discussed previously. 1.3ma is a nice ballpark. Bear in mind the base of Q30 is fixed at 2Vbe drops, so around 1.3v, above ground- R56 is just a 1 ohm test resistor, to check shunt currents. R55 any sensible value for a darlington.
The voltage divider gobbles another 4ma (thereabouts - will see how we came to that figure later), so we can say the shunt itself requires another 1.5+1.3+4=6.8 say 7ma to run.
Therefore our current requirements are now 37+7=44ma. Well that works hey? Lucky.
So we set the CCS for 45ma (to feed the output gyrator collectors of course).
In practice 45ma is a nice default value for the LARS5 board and the above calculations are not really necessary, but it illustrates the point that you need to know the current draw of the load to work out if your shunt can deliver.

Knowing that you can now calculate the required CCS current-set resistor accordingly; knowing the Vak of the Leds are 1.7V each and the Vbe of the 1st CCS transistor is about 0.65V.

c. What is the max output voltage of the shunt?
With that you can work out what the max voltage might be that the shunts can output.
Starting with the front end gyrator:
Having set the CCS to draw 45ma, let’s say the gyrator pass transistor (a BC550c) has a nominal hfe of 450. My batch are around that anyway. Class C hfe range is 420-800 from the datasheet, and it’s a convenient number to work from. So if hfe is 450, the CCS transistor is passing 45ma, the base current draw Ib of the transistor will be 450/45=0.1ma. Now if the filter resistor R43 is 20k, then the voltage across it will be 2v. As the zener sets the voltage of R43 at one end at 33v (it’s a 33v zener), the base of the transistor sees roughly 33-2=31v and therefore the emitter will be sitting at one Vbe lower: 31 - 0.65= 30.35V. So we can say the gyrator will output about 30.35V, which is therefore the input voltage of the CCS.
So what is the voltage across the CCS?
That’s determined by the 2 LEDs on the CSS which have a Vak of about 1.7v each across them. So the base of the cascode transistor of the CCS will sit 2 x 1.7=3.4v below the gyrator output voltage: 30.35-3.4=26.95v. That means the emitter will sit 1 Vbe higher at 26.95+0.65=27.6v. The max Vce of the transistor is 1v according to datasheet, so the max possible voltage at the output of the CCS is around 27.6 - 1= 26.6V. Obviously I wouldn’t at all want to run things that close to the bone, and an output voltage of 20V is what I have selected for this implementation- plenty of headroom. I would leave at least a 2 - 2.5 volt margin.
So formally I would not set the shunt regulator to an output voltage higher than 24V.

d. thermal/power limits.
Next to check are that the thermal and power limits of components are ok.
1)what is the thermal dissipation requirements of the principle shunt transistor?
So assume running the CCS at 45ma and the output voltage at 20V. If the shunt is off load, ie no load connected, then full current must pass through it and it will dissipate 20x45=900mW. (For this we can ignore the current draw of the driver transistor as a trivial amount). This suggests at least a T0126 transistor rated at 5w is required. To92 and similar sized transistors can’t handle the power dissipation. Naked T0126s are usually rated to a nominal 1W, but to be safe it should still be heat sinked. A small heatsink is fine, something around 20-30C/w.
2) What is the thermal dissipation requirement of the input gyrator pass transistor?
This depends on the raw DC in. Now at max allowable raw dc is 45W determined by the Max Vce of the transistor on startup as per datasheet. As the base is fixed at 33V the max Vce the transistor would see in operation is 45-33=12V while passing the 45ma set by the CCS+ the 1ma required to bias the CCS LEDS. This means power is 12x46=552mw. This is over max power dissipation Pc of 500mW. So the reality is I would be uncomfortable running the regulator at more than 40V raw DC. Hicaps output around 37V and my psu about 36v which is warm but safe. In that instance 36-33=3Vx45ma=135mW power dissipation- fine.
3) What is the thermal dissipation requirement of the CCS cascode transistor?
Well , it passes 45ma set by the CCS. Now the thermal limit for a 2sc2705 is 800mW as per datasheet. With that we can play with 800/45=17.7V across it absolute max. The max the CCS/shunt will ouput at no load is about 26.6V as previously calculated. Therefore the lowest voltage one could run the shunt at is 26.6-17.7=8.9V.
*** Again it’s sensible never to run even close to thermal limits; comfortable is half thermal limit in my book due to temperatures inside a chassis building up, hot weather etc etc. you want good safety margins. ***
So I wouldn’t run the LARS5 board at less than about 9V lower than max voltage so about 26.6-9=17.6V.
Oddly enough close to the voltage I designed my filter shunts to run at...

So part of the process for selecting transistors with different thermal ratings and operating points. Also a reason I added the front end gyrator Zener to the circuit- to fix a lot of operating points. It’s not actually necessary (though should improve low frequency performance from an academic standpoint). In practice if the psu were to run at 30v, the shunt would still work absolutely fine; simply the max voltage it could output would be 3v lower at around 23v, and the zener would be in a state of non-conduction.

It’s really instructive working through this process because one realises that commercial audio, far from running with magical and highly technical operating points and components has to be very much driven by hard requirements on the ground- Juggling a load of parameters to make the whole bag of bits fit together, actually work and not blow up when it does start working. Wizardry, wonder components, audio foo, luck and ‘golden ears’ listening might have a place, but a lot is actually just working though numbers and practicality- no point designing in a component that has been discontinued for example.

And those are the workings in the design of the LARS5 regulator board.

Top level structure of the LARS5
Both the filters and the collectors [ports] of the output high bandwidth gyrators are independently pre-regulated by the above system, so there are 2 of the pre-regulation circuits for +ve and -ve rails; 4 in total per channel.
Grounds use a mix of split ground planes for high frequency performance and star ground to eliminate intermodulation of the multiple rails.

And thats about it for the regulation system.
On to the preamp --->

 

[ced1]

The Line Preamp Board 5

I'm going to start with a bit of history again.

http://www.naimmods.com/2 Attack of the Redbox Preamp/20 Air guitariste LPA1- currents.pdf

Here is the basic Naim preamp circuit (near enough- the old Air Guitariste so just a bit of extra decoupling and a few tweaks to values to reflect the more modern implementation of the circuit as was found in the Nac82/52 rather than the older 32s etc).

And the currents each circuit fragment draws when simmed with a dummy load of 18k, with a 1Khz0.1V input signal. Yep the load would be active in real life, but I don't want to look at the extra complexity for this exercise.

Interesting on a couple of levels.

 

[cubeasic]

Hallo Ced,
This are very interesting posts, and very impressive as well. Not that I understand all the details, some things are beyond my knowledge.
I am still listening with your air-guitariste preamp most of the time. It was a very beginner friendly project from a few years ago when I came into diy-hifi. For about a year or so, I am working on an advanced version based on the starfish pcb, with symetric psu and seperate secondaries for every regulator. I have to admit, it feels like a kindergarten project, compared to what you are doing.

I am looking forward to hear more from your project
Florian

 

[ced1]

Hi florian,
glad you're enjoying it.
And holy cow, sounds like you're creating a monster. Separate transformer windings for each regulator? I think the starfish was a multi-rail design too was it not? Though using LM317/337s which are dogs IMO. You must have a forrest of capacitors there! Very impressive.
So Neil McBride used this strategy on this earliest designs, though only at a board level, i.e. a winding for each 729, 321 and phono board. The only person I know of who used your planned strategy with so many windings was, If I still remember, Jonathan Carr again, on the Connoiseur 3 or 2 Preamp I think it was- this is going back many years now. Though he didn't continue on the 4 or 5 so I guess he found better ways of doing it.
Well my take on your plans are, for what its worth:
symmetric PSU
I assume you mean +/- rails. Definitely worth while. The second Preamp I built after the Air Guitariste (I called it the Axe-God, argh will I ever learn?) but never published it, used +/- rails. This was definitely worth while doing. The sound became a lot more sophisticated - detailed without brightness. I would describe it as more musical access.
LM317s/337s?
Get rid of them- they are rubbish. I'm pretty sure with some bodging you could replace these with low current gyrators for a huge leap in performance.
Your multiple windings
I can't comment on. I'd be curious to hear your listening notes on it.

Ive tried a few transformers in my time, different sizes, windings and types:

My feeling now is that with regards to transformers in general, low capacitive coupling between windings is critical to getting a lift in performance. R-cores put the windings on separate bobbins. I only use R cores now. I rate Toroids as indifferent things at best. Perhaps with a electrostatic screen they might deliver.

Preamps: I can say that, using an R core in a preamp psu, there was also a big leap in performance running left and right channels on separate windings. But you also get isolation between secondaries as well secondary/primary on R-cores, so this may be a reason why.
Others on pink fish have not noticed a difference, but they seemed to have been using run-of-the-mill toroids.

Poweramps: But, to put the cat amongst the pigeons,
I've run a poweramp from 2 windings, and also 4 windings, both on R600 R-cores. Found this pic of the blighter here:

Got to sell off a couple of these buggers actually- they've been sitting around for about 4 years now.
Anyway, I noticed bugger all difference on the poweramp between both, so I didn't bother with using 4 windings any more.
BUT that's when running a poweramp in standard Naim configuration with the whole amp board run off the windings, with separate winding for both + and -, both channels.
Change the configuration to run the front end and the back end of the poweramp boards off separate windings and its a different story again. The isolation between low current front end and high current output is definitely worth doing. The Avondale NC200 boards lend themselves very well to this strategy, though I don't think that's how Avondale does their Voyager amp.

Transformer size: Lots of iron can also deliver... but like many of these things, without exhaustive testing its hard to pin down which variables are key. But I have noticed an improvement in weight and body and meatiness to the sound going from a 600VA to a 900VA R-core tranny.
Annoyingly I've also not noticed this effect going from a 160VA here:

to a 600VA tranny here:

so in my mind its not so clear exactly whats happening.

Your milage may vary.
Oops gotta go.
ced

 

[GWM]

Hi Ced where did you find a 900va Rcore that's a monster was that powering a stereo pair? Know what you mean about toroids I replaced a good quality toroid with shield on the front end of my hackernaps with a c-core it was an obvious improvement.Keep it coming very interesting stuff.
Geoff

 

[john.luckins]

Hi Ced,

It is great to have you back and being so open and frank with your analysis and ideas, most if not all of which gell with what I have found but which take things further, a great deal further.

I have struggled to find anything constructive to contribute but there is one thing that I have found which might be worth a try. There is a potential weak link in your constant current source in your pre-reg shunt circuit. In a nutshell it is R46, the resistor to 0v. It provides a common path for both load and line currents from device operating point currents and parasitics in the devices used in the CCS. Walter Jung has suggested a fully floating 2 terminal cascode current source that removes this path. I will try to dig out the link to his original article but here is a diagram from it (all due credit to Mr Jung):


I've compared this to a variety of grounded current sources and found it to be an improvement though to give of its best with the best quality cascode capacitors you can muster. You can also use a paralleled up jfets to achieve a similar result at lower parts count if output impedance at HF isn't deemed as critical as the error currents due to parasitics.

BTW have you ever tried choke regulation?

John

 

[ced1]

Hi John,
thanks for the excellent post. And pleased my waffle is gelling with one person at least, heh.

Also, love that you've seem to have referenced a lot of the same sources I have.
If I'm not mistaken, that circuit is one of the variants of the Gary Pimm depletion mode mosfet CCs is it not? And just had a peek online- seems he now has a website. Will have to see what he's updated now you've reminded me:
http://www.pimmlabs.com/

Walt Jung? Are you thinking of these 2 excellent articles on current sources:
http://waltjung.org/PDFs/Sources_101_P1.pdf
and
http://waltjung.org/PDFs/Sources_101_P2.pdf

But, yes, I did consider what you are suggesting. And annoyed that you have tried it and found it to be an improvement, damn, heh heh. I mulled it over it for some time, and then decided to make my life easier and just use the resistor for 3 reasons:

1) Used as a sole in-line CCS replacing my current bipolar arrangement, the dropout of your (Pimm?) CCS arrangement would have taken another 5 or 6 volts headroom again. I have it in my head one stares down the barrel of 9 or 12 volts headroom with the things? Also the DN2540s have pretty high output capacitance I seem to remember
(yep they do:
http://www.supertex.com/pdf/datasheets/DN2540.pdf
and I was keen on very low output capacitance for high frequency performance. The LND150s
http://www.supertex.com/pdf/datasheets/LND150.pdf
couldn't handle enough Idss current by a long way, though look very desirable from just about every other parameter. Jfets in some senses were a lot more desirable, but then you are dealing with the problems of finding something which can handle a high Vds, high currents up to say 50ma, has a low output capacitance and a high Rds(on) for good current regulation. I searched for a while.

***In context, I originally chose my arrangement for the redbox poweramp, to fit in a NAP140. The poweramp would have been left running at about 12-15V rails with a Pimm CCS which was unacceptable***

2. the second option of course would have been to CCS the LED biasing string (D8, D9, C23, R46) via a self biasing jfet or mosfet CCS to the output (collector of Q25) of the CCS instead of ground. Effectively 2 CCSs in parallel. Then you can use things like the LND150 or some Jfets. However you are still left with the voltage headroom issue. I worked out I probably still loose perhaps 4-6v extra headroom. This was the other way I considered doing it.

** In context I run my preamp at around +/- 17V i.e around 34V instead of the Naim standard 24 volts. I have found higher voltage rails to be an improvement. Perhaps to do with the lower capacitance of bipolars as Vce increases?- look at a spec sheet. Also run the preamp even at 22-23V and the sound completely collapses. Very bizzare but a naim preamp goes to shit even only slightly under 24V in my experience. ****

3. I simmed what happend to R46 as follows (its actually R14 on my sim). The below are based on a raw psu of 30V, with +/-1V amplitude, 1khz or 10hz noise on the line. The load current noise is a hefty 5ma with +/- 2.5ma amplitude current noise at 1khz. Just to get a feel for how things were looking.

1. Output gyrator voltage, 1khz line noise:

2. R46 current, 1khz line noise:

3. Output gyrator voltage, 10hz line noise:

4. R46 current, 10hz line noise:

And apologies, I've just had to replace the above sims. They looked a bit wrong when I went back to look at them, and durr, no s*** Sherlock, they are completely so! Dunno where that went wrong. Much better now.

In sim 1 above, the input gyrator is keeping the +/-1v, 1khz line voltage noise down to around +/- 0.5mv at its output. In sim 2 one can see how this translates into a current noise of about 0.1uA though the biasing resistor R46. That basically means we have the biasing resistor R46 across a constant voltage which will produce a constant current to ground. The CCS transistors will use constant base currents more or less.

Sims 3; At +/-1V, 10hz (bear in mind this is subsonic) line noise, well this is actually looking a bit uglier. Gyrator output line regulation is poor at about +/-80mV which does reflect line related current noise of 5uA through the load resistor to ground as per sim 4.
But I made a call that its effectively constant current at higher frequencies through that biasing resistor R46. And also a judgment call that an effectively constant current through to ground would not impact audibly on ground regulation.

So yep hands up, it was a judgment call. Not ideal but my transformers are all rated for around 35V dc. It came back to juggling components and numbers on the actual board. Actually a perfect example of the point I was making in my previous post on operating points- the design decisions are often made on the basis of what is practical or practicable.

But John, you seem to have put the cat back amongst the pigeons. Looks like my lovely boards are up for some bodging in the next month or so. I do now want to try this out. I will have a mull on the best way forwards but I've been looking for an excuse to use my stash of LND150s for some time.

 

[ced1]

And crap, yep your other excellent point. I really do love where you're going. Same wavelength I feel.

Have I ever tried choke regulation? You are truly the prince of darkness.

So I have had these looking reproachfully at me from my parts bins for over 5 years:

My only defense was my chassis was so goddam full its only now I've been considering them again. Specs are Shaffner common mode, 1.8mH, mains rated, DCR less than 0.1ohm- Too low to measure on my meter anyway. The lowest I could get, rated to 10A. Ah it say 0618R so I guess 0.06 ohms DCR.

So my question to you is.... should I get off my ass and put them in immediately? Now?

If so, I have 3 return questions for you-
1) where?
a) CLC arrangement in the psu. This is what I would try first.
or....... here's a thought:
b) why mains rated choke? Well, the sucky thing with inductors is the impedance, no? They do emphatically not have a 'no-brainer' reputation. The avondale capacitor board is designed as a CLCLC arrangement for example. But I remember posts where folks took out the L's as because they were killing the dynamics.
So Possibly in the mains? clean up the mains without affecting the transformer secondaries in terms of impendance?

2) What sort of inductor?
I made some custom crossovers for my Credo speakers about 10 years ago which were an utter failure. In retrospect I think because the Audyn caps I used were shit (actually I have no time for Audyn caps at all) and I threw all my money at inductors. In the end I modded the stock Naim x-overs with decent Wima and Evox polypropylene caps, which worked nicely (smoothed out the edginess of the sound and made it more sophisticated) and left it at that. Active seemed the most sensible next step rather than throwing hundreds of £ at big boutique caps when my experiences with fancy caps were so-so to say the least.
BUT anyway, the point I'm making is I have some fat air core inductors here to hand- though the largest at 700uH is smaller than the schaffners. Would air core inductors be better?
Here we go; the cursed xover:

3) what size inductance and DCR?
Do I need to go to 100s of mH rather than a couple?

Would be keen to hear any thoughts or listening impressions on this issue.

cheers
Ced

 

[john.luckins]

Hi Ced,

Sorry if I set the cat amongst the pigeons with the two terminal current sources. You rightly spotted that voltage headroom is a problem with them, I give my Jung style cascodes at least 10 volts though they beneifit from 12and I found that they are very fussy about the capacitor choice, only polystyrene really cuts it and 1uf is a lot of Polystyrene cap (thank god for Russian Military surplus).

I had already resigned myself to higher voltage R cores to get enough headroom for my choke regulation so a few extra volts for current sources wasn't a problem. My gut feeling is that prevention is better than cure and the gyrators do a pretty good job of prevention, particularly on line noise up towards 100kHz, for roughly the same amount of voltage headroom as the extra for these snazzy CCS's. I may try some gyrators in front of it all on my Buffalo DAC feeds to see if the improvement is discernable. If it is then perhaps they are complementary to some extent.

BTW I have found to my immense frustration that even when I regulate to the greatest degree I can to prevent line noise, I too can still hear the difference between a small say 50VA R Core and a large 432 VA R core, in spite of less than 100mA load currents. I think this has more to do with how the transformer behaves when trying to deliver high current spikes and how this is all fed back into the mains and other parallel audio supplies.

When I suggest choke regulation then I mean a choke in series and connected to the rectifier diodes and before the first smoothing cap in the supply. This choke should be of high enough value for a given (CCS regulated) load current that it remains in continuous conduction, i.e. its current load > choke critical current. As a result the maximum rectified voltage is roughly 0.9 X the RMS AC voltage of the secondaries, which is a big loss of volts, and the choke values are in Henrys and 10's of Henrys. The one I mostly use is a Hammond 156R rated at 1.5 Henry at 200mA. Any such choke will produce voltage spikes in a rectifier at switch off as it tries to sustain current flow. These voltage spikes can by fairly easily removed (as described in Valve Ampifiers by Morgan Jones) with two small caps to earth on either side of the choke. I see it as the lesser of two evils. It is much harder to remove the high current spikes from a non choke supply at diode switching points than to remove the voltage spikes of choke regulation and the current spikes induce a lot of interference alll over the place. The latest edition of the Morgan Jones book is definately worth a read if you fancy trying some of this. I have found it beneficial to use choke regulation like this on every supply to my system that I can, and that has means a separate ring main and choke regulation on Squeezebox touch, Network switcher and power amp front end. Doing this on the power stage of a class AB power amp has defeated me so far so i may end up with class A or valve amplification.

It all makes for a great deal of iron and a lot more boxes than the wife would like in the lounge so not for the faint hearted.

John

 

[ced1]

And this is the big 900VA beast in the poweramp:

R800 core with 2x100VA windings + 2x350VA windings, 900VA in total. Originally intentioned to power 1 channel of a huge poweramp, until I decided that actually going active made a lot more sense. As I mentioned, seperate windings for + and - of 1 channel of a poweramp a la Naim made no difference to sound. I got quite cross and went back and forwards several times to make sure. Why do Naim do it this way? Maybe it works on toroids.
Now it's powering a Redbox dual mono poweramp. The 350VA windings power the high current output section of each Redbox amp board, the 100VA windings power the front end of each board. Front end runs to 4-pole jensens, output big 22,000uf Kendeils. Simply swapping out the earlier R600 600VA (2 x 300VA) for this monster, (the 2x 100VA windings unused) so actually only running 700 VA (2 x 350VA) made a significant improvement to sound. I can only surmise its more iron in the core being the main significant difference.

Running the front end of the poweramp boards separately was also a nice little improvement.

Later this year (famous last words) its brother will be fitted in the chassis to go quad-mono with active crossovers.
Where can you get em? A lot of research, a lot of phone calls and emails, a lot of costings, a custom order specced and commissioned and a lot of crossed fingers and praying. Cost me about £1k to make the unit cost even remotely sane and this was years ago. I must have been nuts.
cheers
Ced

 

[Bemused]

Ced this is a marvellous thread, quite beyond my capabilities but never the less most interesting and educational.

Yes I also think your probably nuts but if you were boring and sane we wouldn't have these great threads.

John, a 101 choke thread would be nice, I understand the principle but not sure how I would calculate.

 

[ced1]

Well my observations after long breaks are; there are far too many, very modest people on this forum. Its turning into an absolute pleasure.

Bemused, anyone that can build your dac is far from having found capability (or sanity) limits yourself. I hope to start delving into your adventures in the near future. Right now a couple of months free to tinker and catch up would be very attractive, nevermind twisted pear actually having the little swine for sale ever.
John, you have delivered 2 quite excellent and loaded posts. I've had to reread several times, go away and muse for a bit. And I am quite happy to do a bit of forelock tugging to the gyrator man. None of this preamp project would have been possible without your lead on this subject many years ago.

Anyway, I'd like to second Bemused's request for a chokes 101 thread.

I will admit to having had the Connoiseur 5 preamp as a kind of model to deconstruct, understand and Naim-ify for probably 10 years. That little sucker sold for about £25,000 in Japan if I remember. The choke was 1 of 3 things left 'flapping in the wind' for my part. Stare as I might, it seemed to be on the primaries, but what you have said resonates completely with other comments around it and choke regulation.

I'm now doing some back of fag pack calcs for my own selfish ends. So if I serial my primary windings on my preamp tranny, I will get about 70-72V out of my preamp. Waahh, insane! I think my reservoir caps are good to 60V? I bought what, 60, 80 and 100V items for pre and power? How many volts does your choke drop? I'm running 45ma CCSs. Tell me the chokes drop more than 20V? I'd like to understand the math so would be grateful if you could explain.
The headroom would hopefully give headroom to mod the CCS in the shunt. It would be easy to lift a resistor and cobble together the gubbins. Will have to do the sims and check all works out. I'd still base it on bipolars though- I stumbled across a whole book on CCSs a while back which was brilliant and which I therefore naturally managed to loose while relocating back from Oz. So many great circuits in that, clear concise explanations. One of the things I came away with are that bipolar CCSs are better by just about every measure if well implemented. And bipolars definitely to be used as the cascode component.
OK its late.
Ced

 

[ced1]

Here we go. One of the more adventurous current sources in that book, for a laugh.

I think it was for aviation or something. Please don't try to build it or use it unless you know what you're doing and can build from first principles. I copied this years ago, probably made some mistake and lord knows what. Looks like a triple CCS at a glance. Wilson mirror feeding a complemtary jfet feeding a cascode mirror or something. My head is buzzing.

 

[ced1]

And after a small break for an intensive course, back to this.

So despite some v interesting bits on current sources and inductors (of which I answered my own knee jerk questions with 10 minutes of research, lazy boy), I'm going to leave those to one side for the moment so this thread actually gets finished off. Spring is coming and the soldering iron will soon lie fallow.

So back on track, the basic Naim preamp circuit again (Air guitariste tweaks):

http://www.naimmods.com/2 Attack of the Redbox Preamp/20 Air guitariste LPA1- currents.pdf

and the actual sim plots to refer back to in case I made a typo transcribing the values onto the schematic.

http://www.naimmods.com/2 Attack of the Redbox Preamp/20 Air guitariste LPA1- current plots.pdf

As I said, this is interesting for a couple of reasons.
1. The most blazingly obvious is not a single circuit fragment uses more than 1ma,
EXCEPT the output transistor and output CCS that biases it Class A which both come in under 10ma.
So I have to reiterate again, chunky regulation is contraindicated for the preamp as high current regulation indicates large and lossy transistors and correspondingly large therefore electolytic and lossy capacitors.

2. the second is the actual AC component of each circuit fragment in the in lowish uA range. Here the Output transistor and VAS string are the biggest offenders at 60uA and 170uA. Some are in the PicoAmp range for crying out loud.
This implies that
a) regulator output impedance is not particularly important and we can play pretty fast and loose and not get crazy voltage rail noise, in phase or not.
b) AND that very moderate capacitance can slug voltage noise on these lines as it doesn't have to cope with much AC.
This is both excellent news as it implies that a low current gyrator is not only acceptable but positively desirable.

And for your delectation, the -3db points of the preamp:

around 3Hz and 150khz

Now to understand the LPA5 without getting overwhelmed, lets look at one of the missing links; my old Funk Monster preamp (3rd gen preamp) of about 7 years ago, which is a split rail, bootstrapped preamp.

http://www.naimmods.com/2 Attack of the Redbox Preamp/24 LPA3 preamp.pdf

Notable points are:

1) An additional -ve rail, taking a lot of current out of the ground line a la Nac552. Hardly cutting edge technology, but of a very significant sonic benefit. As I said before the best way I can describe it is it makes music more sophisticated and musical. More detailed and coherent without any brightness A no-brainer mod though a pain in the arse in terms of cost, as it affects the PSU where the bulk of a preamps cost is. The only bits grounded are the components referencing the signal to ground (filtering capacitors C17, C19), decoupling caps C18, C30 and C14) and R27/R31 providing cap discharge paths to ground.
2) rails running at around +/- 17V, 34V it total. As I referred to earlier, going much under 24V kills the preamp sound horribly and transistors exhibit less internal capacitance the more Vce they bear. The difference is slight as I remember, but do it if its not a hassle.
3) The Class A biasing CCS (Q11, Q12, D1, D2, D3) has been changed to a Cascode from the traditional Naim configuration, for improved high frequency operation. To be brutally honest, I'm not so sure there is much of a difference sonically, but I never did the A/B comparison carefully. Its certainly not a step back, and as I remember, the sims were microscopically better. Picoamps in it, but what the hell.
4) Input cap C15 is 3.3uf not the traditional 10uf- this does not impact the 3Hz roll off by more than a fraction of a Hz, but opens up the world of film caps signifcantly. Film caps are obviously desirable here for more clarity and transparency unless you really dig the sound of those tants. Its not a deal breaker by a long way, and I wouldn't class this cap (or C16) as affecting musicality, but beyond a certain level of quality, Its really a bit hard to justify tants unless a house sound is part of your marketing strategy and a brake to ultimate sound quality.

*** As an aside, My heartfelt belief is that the heart of the naim 'sound' comes from the input/VAS topology, essentially Q9, Q6, R18. You can chuck in R24/C17 if you like too. Just about anything else can be piddled around with.***

And then the bootstrapped topology (thanks to ALW and Walt Jung). Compare the Air Guitariste to the Funk Monster and you will see essentially points A + R15 and B are joined at different points. The main difference is it reduces the AC component though the Vas transistor massively by referencing R25 to the output rather than the -ve rail so the voltage swing across R25 becomes far smaller, but beyond that I don't pretend to really understand the technical basis for implementing it.

SIGNAL CASCODING
Finally lets look at a cascode before we go to the LPA5 circuit. There are several ways to implement a cascode, but in view of the principle of multirailing- keeping circuit fragments as separate as possible- the only topology that came up to scratch is what I call a buffered cascode.

This is the cascode around the input transistor Q11 (a jfet):

http://www.naimmods.com/2 Attack of the Redbox Preamp/26 input cascode.pdf

Q8 is the cascode transistor for Q11 and eliminates Miller effect which massively increases effective transistor input capacitance and therefore decreases the working bandwidth. It makes Q11 very low capacitance instead of very high capacitance in action (compare perhaps 2-3pf to several hundred pf depending on gain and feedback capacitance, to get an idea).
And as fooling around with filtering and coupling capacitors will quickly learn you, capacitance unless for decoupling, is not desirable and can muddy the sound significantly.

So although it may look like a hell of a lot of circuitry wrapped around Q11, what it actually does is make Q11 work more ideally.

The supposition then becomes, if you can stop this additional support circuitry from interacting with the core circuit, you win. If you can't the question becomes do you still gain overall, trade benefits or loose? Obviously safest and least boring is not to find the answer to this question, but rather ensure it doesn't interact with core circuitry. The way to do this is by .... multirailing and careful design. Thus the effort gone into the LARS5 regulator board.

In summary, 2 of my design criteria based on experience, are high bandwidth and low capacitance. The cascode achieves this.

So in theory it should sound good, but as ever the real question is.. does it sound good? We'll be finding out soon but early investigations indicated its not a small difference in sound. Not a small difference at all!

IN DETAIL:
Q12 is the buffer transistor. It measures the voltage at the drain of Q11 and using only a couple of uA of constant current, transfers this voltage to the cascode string via the Vbe of Q12. We know the emitter of Q12 will be about 0.65V above the source of Q11. U3 is a fixed voltage zener which biases Q8 and ensures that the Vds of Q11 is the same constant voltage as U3. This is nice as you can get cascodes referenced to ground or to the source/emitter of the input transistor. The latter are theoretically better as they also ensure a constant Vds for Q11 and therefore constant input capacitance. The grounded sort don't.
Q1/Q4/R7 therefore form a cascoded constant current source to ensure a constant current through U3 and Q12 (and pretty much Q8/Q12 so unmeasurably small modulation of the actual core circuit). D1/D2 and R13/R17 bias the cascoded constant current source and C11 a bit of decoupling from any noise on the rail due to Q12. In theory this biasing string should also be isolated by multirailing, but the design would get to unwieldy and the ac currents are unmeasurable, so I've decided not to.

Now for the LPA5, the Vas and output transistor are also cascoded in a similar manner, and in total we arrive at the LPA circuit:

http://www.naimmods.com/2 Attack of the Redbox Preamp/25 LPA5 preamp.pdf

So before panicking, to be absolutely clear: This is still the Naim circuit.

All the extra bits around it are all almost completely isolated from the core circuit and all for the purpose of making the core circuit work in a more ideal manner, less plagued by parasitic capacitances, parasitic inductances, leakage currents and power supply intermodulations.

One can actually say, this is more the essence of the Naim circuit than Naim themselves manufacture.

and that's all for tonight.

cheers
Ced

 

[bivalve]

Wonderfully stimulating thread Ced, perhaps even a bit challenging after dinner and a couple of glasses of shiraz. Erno Borbely was very big on FET input stages for their low capacitance. I never heard one of his later pre amps, but they seem to be well regarded.
His projects were in Audio Amateur during the 80s before he went commercial.

 

[GWM]

Interesting post Ced but why bother with an input cap anymore I assume you have no DC offsets on your sources so C15 can be dropped.I used to use a 3.3MKP10 in this spot.
Geoff

 

[binnie]

Yep, now it is really interesting, thanks CED !

Concerning the input/output tants: I removed them from my B4 pre (but I still have one at the output of my pre-phono and in the input of my 160).
Before removing them, I switched On and Off my B4 several times and couldn't detect any DC peaks on my USB Oscillo.
I would perform this simple check before removing any I/O caps.

 

[YNWOAN]

Colour me interested.

Yes, absolutely - me too .

 

[ced1]

cheers guys, glad you enjoying.
Yeah the output dc offset and input coupling cap. I remember conversations to this effect a while back, puzzling over this, working it out and it didn't calculate.
Basically you have to bias the input transistor base/gate away from 0v to get 0v output offset. The feedback cap reduces gain to unity at DC, it doesn't eliminate dc. With a bipolar this is not so much (when base is 0v, emitter is -0.65v ) but with jfets the voltage difference is considerable. A quick and dirty sim to demo this here:

and waveforms here

The gate of the input jfet needs to be biased -4v below 0v to get 0v at the output. For that, you need the biasing string and input coupling cap to be able to shift the bias point away from the 0v that a 0dc offset source would impose. Stop me anytime I'm wrong?
So I guess the choice is output or input coupling cap, unless you want to depend only on your poweramp input coupling cap to protect from dc offset blowing the speakers. Then you can probably get away with dumping both which is probably the optimum, though a dirty, solution sonically. I know I'd forget, do a major rebuild a couple of years down the line and blow the whole shebang to hell. My choice is the option to dump the output cap because its biggest, and a bottleneck for headphones, though this may end up being a risky strategy due to thermal drift etc. If I'm left holding smoking cans and burnt ears down the line, well that will be the answer right there.
That is for the RCA/Naim topology anyway, unless I missed something? The B4 I don't know. Different topology. If I ever do a LPA6 Naim style, off the cuff I would try a dc servo driving the input biasing string and still keep the input coupling cap. Not convinced it would be better sonic choice than the LPA5 though.

Ernesto Borbely, he is indeed a fan of jfets, not low capacitance I don't think though. Most of the jfet guys seem to dig low noise, which is invariably high capacitance. I have one of borbely's project papers anyway where he recommends 2sJ74 (Ciss 105pf) and 2sK170 (Ciss of 20pF). Those are very high, and from experience i'd guess they wouldn't be to my taste naked. Though they might work if cascoded. I might yet end up having to try them! When my write up is up to date I'll go live, so not sure. For the moment my choice is J304, Ciss 2.2pf.
Personally I find the jfet thing smacks a bit of dogma though. Tube lite. To me one should choose what sounds best or what is appropriate for a specific reason. I fall back on dogma (I like to call it rule of thumb), when its too much of a hassle to do the listening test. So I use a jfet on the input for a reason as the next post will explain. I use bipolars for cascodes because there are a couple of places that give the technical rundown on why they are superior in this position even though jfets are MUCH easier to implement as cascode devices. Off the cuff they don't maintain the Vce across a transistor nearly as well as another transistor for example.
cheers
Ced