help check over my split-rail FET boost schematic :)

[darron]

Hi all.

So I've been building this FET booster for a while, very similar to the GEOFEX Mμ-Amp recommendations: http://www.geofex.com/Article_Folders/modmuamp/modmuamp.htm

A while back I made a charge pump doubler version with a 7660S, but thought it was a kind of stupid idea and probably would be better and simpler all-round with a split rail supply.

I've never done this before. My idea pretty much has been to move anything from the VR/2 reference to Earth, and anything from the old Earth now to -V


I also put a buffer at the end which will bias through the volume pot. I noticed the buffer really couldn't handle a 10KB pot without HUGE volume loss so a buffer might not hurt after all.

I put a capacitor (C3) at the beginning because I guess now that the input has -V instead of Earth it would pop otherwise.


So... just wanted to make sure that I didn't do anything stupid, or take suggestions. This will be my first time completing anything in Eagle too!

It's a very clean boost with two very useful knobs if anybody would like to build it.




Maybe the GEOFEX img for quick reference, hope you don't mind R.G.:





EDIT: Oh...  just realised C4 is probably redundant can can be chopped out

[PRR]

> very clean boost

Voltage gain is not even 9. (50K:5.6K) Directly sensitive to pot tolerance, which is "loose", though not enough to hurt. Yeah, gain <10 is unlikely to make "DIRTY!" sounds from guitar-level, even at 17V supply.

> simpler all-round with a split rail supply.

I don't see that.

Q1 Gate is naturally at common.

It is easier to filter G2 Gate than Q1 Source feed.

I suppose it is not terribly different, I just like the aesthetics better single-supply.

If you find that the supply inverter works better than a doubler, that would justify bipolar power.

> C4 is probably redundant

No. There is no way Q2 Source will sit at dead-nuts ZERO. In this case it will likely sit a Volt or so high of zero. The stray voltage will make Level pot noisy, also change Q3 bias (maybe don't-care).

I know the offsets here because Heath used the same DC affair on a 'scope input. Different signal path (in to your Q2 working as source-follower). Output taken at bottom of your R3. Their idea was that is Q1 Q2 matched and R3 R4 matched, the offsets all cancel. Well, they get very small, but not small enough, it was annoying the DC shift with gain knob setting. I paid rather a lot of money for a fancy (in 1990) 20MHz chip buffer with much less offset.

I suppose there is no reason you could not make R3 R4 the same value. But JFET matching is hit-or-miss. I suspect Heath even had the resources (boxes of JFETs) to select semi-match pairs. You might get 50mV offset (at the *bottom* of R3), which maybe is good-nuff for stage work.

Q4 could certainly be 0.2uFd instead of 1uFd. Worst-case response then is 20Hz. I suspect it will be very much better than that, because Q2 source network tends toward very-high impedance. I'd wonder if 0.01uFd sounded any thinner than 1uFd. Sure would save some cents, and fair board space.

[darron]

> very clean boost

Voltage gain is not even 9. (50K:5.6K) Directly sensitive to pot tolerance, which is "loose", though not enough to hurt. Yeah, gain <10 is unlikely to make "DIRTY!" sounds from guitar-level, even at 17V supply.

when i was tying straight to ground it used to clip lightly. this was the lowest value i could use without my hot neck pickup making it fart a bit. but that was using J201's which are also pretty crappy for headroom.

it's nowhere near what you COULD get out of this circuit. if the volume pot was audio taper unity would be about straight up.

> simpler all-round with a split rail supply.

I don't see that.

well... my line of thought was
1) no diodes and less caps on the charge pump
2) no diodes on the pump means higher headroom
3) no need to make a virtual ground and filter it
4) don't need higher voltage, larger electros

> C4 is probably redundant

No. There is no way Q2 Source will sit at dead-nuts ZERO. In this case it will likely sit a Volt or so high of zero. The stray voltage will make Level pot noisy, also change Q3 bias (maybe don't-care).

oops. i'll put that back in! thanks my original with the doubler had it in there... the GEOFEX didn't include a cap. even runs a buffer stage directly connected but I understand this wouldn't quite be ideal:


i read what you're saying.. for the sake of a single component and some clean working i'll just keep it in haha.

Q4 could certainly be 0.2uFd instead of 1uFd. Worst-case response then is 20Hz. I suspect it will be very much better than that, because Q2 source network tends toward very-high impedance. I'd wonder if 0.01uFd sounded any thinner than 1uFd. Sure would save some cents, and fair board space.

agreed. i could probably make that around 22n with no real difference. 1uF is my 'go-to' because I bought 1000x pieces and they are the same cost to me. i'll take your advice and change this to a 100 or 220n MKT for realistic specs and space. I could make a similar evaluation for C1.

I might change the source of Q3 to 10K too.




Thanks heaps for looking over it in detail! It's very re-assuring to get a seasoned experts view. I'm thinking of getting the boards fabbed so would be nice to avoid trial and error. Especially with the split rails which I've only assumed my theory was okay, not having any teaching on the matter. I've also never actually outsourced boards, hence Eagle so everything is a bit new. I feel kinda dirty not etching my own actually, also excited.

[PRR]

Any crap (hum, buzz, whine) on the negative rail enters Q1 Source with the same gain as signal at the Gate. At guitar level, this demands a very-clean negative rail.

C3 could of course be smaller, but if you have a thousand pieces, 1uFd is settled.

I would like a much larger series gate resistor. 1K with a 10mA max Ig FET protects to 10V. You already have 9V at start-up. Some heavy signal happening at start-up will put it over 10.0mA. And the series gate resistor could be *much* larger with no effect on clean tone. 50K easily, unless you have something far fatter than J201. And in overdrive, 50K will delay grid-blocking. (OK, no grid, but same effect.) For higher capacitance vacuum devices, 33K is often used to reduce supersonic reception, particularly AM radio band. (Used to be police/taxi bands up there, and they could be very surprising.)

[darron]

Any crap (hum, buzz, whine) on the negative rail enters Q1 Source with the same gain as signal at the Gate. At guitar level, this demands a very-clean negative rail.

i do like considering noise, and maybe i didn't consider this as much regarding inviter VS doubler.

i get what you mean though. at the moment the negative rail is:

DC socket (only) in   >   Schottky diode   >   100uf filter   >   44KHz-ish (from memory?) inverter   >   100uf filter   >   large inductor (around 30 ohms from memory) > 100uF filter.

would you suggest 1uF MKTs as well right at the end maybe? would this be one of your key arguments to rather a doubler instead?

i always hated the idea of charge pumps, for line cleanliness. but never could actually make a single complaint with my previous setup.

I would like a much larger series gate resistor. 1K with a 10mA max Ig FET protects to 10V. You already have 9V at start-up. Some heavy signal happening at start-up will put it over 10.0mA. And the series gate resistor could be *much* larger with no effect on clean tone. 50K easily, unless you have something far fatter than J201. And in overdrive, 50K will delay grid-blocking. (OK, no grid, but same effect.) For higher capacitance vacuum devices, 33K is often used to reduce supersonic reception, particularly AM radio band. (Used to be police/taxi bands up there, and they could be very surprising.)

you know, i often wonder about this! sometimes i see resistors of 220 ohms, maybe a few K, 10K max. and it just doesn't seem worth it to me. maybe piece of mind to take a few hairs off.

a manufacturer i used to work for had quite a few FET pedals damaged, so i know how touchy they are. probably due to abuse in active loops etc. his trick became to put an LED across.

in my original schematic i used 2x 6V2 zener diodes back to back, shunting to ground after the 1K. i use this value just because i have it on hand for other purposes, and I figure if the signal ever comes anywhere near that then it's not coming out clean anyway! but then with this split rail i confuse myself, because i'd be connecting the -9V DC to 0V if i did the same trick. maybe across from gate to source? in example might be good. otherwise i'll gladly take your advise with a hefty resistor and will make this a better consideration with all future work thanks heaps again there!

really perfect and spot on advise

[PRR]

The first risk with JFETs is the 0.6V threshold of the gate diode. Throwing LEDs or Zeners across that is waste parts.

Gate diode breakdown does no harm. Excess breakdown *current* is the damage.

The second risk is the 25V (40V) reverse breakdown. 40V is less likely, but can happen. Again, breakdown does no harm (in our world), unless it leads to excess current. LEDs/Zeners may take more than 10mA, but in guitar audio it is acceptable to pick a resistor to limit to <10mA for quite insane voltages.

10K limits to <10mA for +100.6V and -125V inputs. If you must plug into 120V wall power, 15K covers that. 33K covers global wall power.

The resistor "cuts the highs". In high-frequency switchers, 220r or 47r may be needed for clean 1MHz pulses. Also MOSFETs can oscillate on any length of gate wire (nanoHenries); a few dozen Ohms right AT the gate damps the resonance. JFETs rarely have enough Gm to self-oscillate in neat builds. The gate C of J201 is 3pF g-s and 1pF g-d. For gain of 47 (unlikely) this is 50pFd. 50pFd against say 33K is 100KHz. Four octaves above guitar. Even the very fat JFETs won't nick guitar with 33K series. 2SK170 will be near 330pFd, 15KHz.

[darron]

Thanks again for the very practical advice. It's exactly what I needed to know.

So here's a revision:


I was messaged about building this, so some notes:
- If you don't have a 500KC, then a 50KB will work with a similar-ish sweep, it just won't cut the bass completely at the extreme end of the rotation.
- For more gain make R4 smaller, or just jumper it out to ground
- As discussed the 220n caps are flexible. Could be 100n, or whatever you'd rather. All MKT will do and are physically almost the same footprint.
- The inductors/chokes aren't really necessary at all. Could put 100ohm resistors there for a similar effect.

[Transmogrifox]

I don't see anything wrong here, so take my comments as just an idea for consideration.

If you were going to use an inductor, here's another relatively simple way to get 18V from 9V for small currents:


This booster circuit contains enough of the relevant parts to indicate how things would be hooked up:


Spice simulation with boost converter as shown yields about 300 uV peak-peak switching noise at the output of the booster, and the switching frequency is around 300 kHz.

It's designed to be about 200 kHz with a 5 mA load attached when "R2" is 12 ohms, so this can be easily extended to a pretty broad set of 18V stompbox circuits.  The 9V input serves as "ground" in these kinds of dual supply circuits. 

If the 9V input was truly isolated from everything else, then this would function as a dual supply circuit with "9V" being positive ground, and ground instead is -9V.

Then of course if you wind your own then the inductor could include 2 output windings giving a truly isolated +/-9V output.  Low power SMPS is not really out of reach for DIY work.

There is also a 555 timer based SMPS floating around these forums for generating high plate voltages for tube work.  Some minor tweaks would make this a better design than I have presented (startup conditions are finicky on this one because it takes a good turn-on jolt to get it started).

[darron]

Thanks Trans. It's a cool circuit and sounds like it will give much better performance than a crappy 7660S chip.

That's a lot of work drawing up the booster. I assume you drew it, otherwise the values would be a real coincidence.

I've already got a DIY layout for this circuit with a doubler. I think I just want to play a bit with the idea of software and overseas fab for experience. As well as split rail as something new.
I'm just buying some tiny chokes like these:

http://www.altronics.com.au/p/l7056-6800uh-rf-inductor-choke/

I always use some resistors, but I figure these must be an improvement and also have some low resistance of course.

Punched this in last night:



I have to get used to it. I drew my own laying down capacitor component.


I really don't like 555's with pedals. They.... just, seem to drag issues with them. They are hefty enough though

[Transmogrifox]

Thanks for sharing the link to the chokes.  Perhaps part of my assumption a 6800uH choke must be huge is my context for inductors is generally with power supplies, so I tend to imagine them as being physically large.  I can see those chokes aren't much more trouble than a resistor

As a rule of thumb I would suggest putting C6, C7 and C10 pins as close as possible to the IC.  Practically this is all probably slow enough it won't make any difference.

A second suggestion when you route this is to put a ground plane on the bottom of your board and punch all ground connections directly to the ground plane.  Also route the IN and OUT traces on the bottom side of the board at the edge of the ground plane and don't route any traces from the charge pump above and/or parallel with IN/OUT traces.  It's best to keep IN as far away from the charge pump as possible.

Layout looks nice and tight and reasonably arranged as far as untangling the Rats Nest goes.

[darron]

yeah i completely forgot that the bypass lines would run on the PCB

anyway, i think i'm getting closer. i've never really worked on non-home-made boards before. is it safe to assume the fab house will do some kind of through-hole plating where the top and bottom layers join? otherwise i'm screwed if an MKT cap needs to be soldered from the top layer and can't be accessed. same with the 3PDT foot switch for example. hmmm :\

i'll do a ground pour/fill on top and bottom layer, but now it's easier to see without it:




if it's successful i'm happy to share the board/project. not sure how long that will take me though as i do around 10 things at a time and like to scatter a bit.

[TheWinterSnow]

I am curious as to why you are going with an inductor based DC-DC converter if your entire circuit pulls no more than 100mA and you want multiples of 9v why not look more into a pump charge converter that creates a -9v rail?  Inductor based converters are usually for higher current generally more than 100mA but usually in the 1-5A range.  Your design seems a little overkill.  I use the LT1054 all the time to get a -9v rail for my designs.  I have yet needed more than 100mA @18v on a split rail so that chip has done me well.  I think if your circuit is pulling more than 1.8W then maybe you do need an inductor based solution but I can't think of any stopmboxes that pull more than 100mW let a lot 1.8W outside of DSP/Micro Controller pedals.

[darron]

it is a charge pump convertor. the 7660S is the same as an MAX1044, but doesn't blow up at 10V input.


i'm using inductors everywhere in place of where i've been using small values resistors (<100ohm) because they already have resistance to help with the filter cap, and yeah, they take up the same space and are not expensive so why not

[TheWinterSnow]

Have you not seen this very simple design that only requires 2 caps and that is all?

http://www.mikrocontroller.net/attachment/55384/7660s.jpg

This gives you -9v from a +9v input.

[TheWinterSnow]

I think that you are filtering way more than is necessary, the use of tantalum capacitors (1uF and 10uF in series) will take care of any noise/filtiering in almost all applications. 

I have made circuits with a 1054 that just had 10uF caps pulling almost 50mA and the rail was more quiet than the noise of the opamps at unity or near unity gain.  I just feel that all your filtering is extraneous.

[darron]

Have you not seen this very simple design that only requires 2 caps and that is all?

http://www.mikrocontroller.net/attachment/55384/7660s.jpg

This gives you -9v from a +9v input.

Yeah. This is identical. But then + a filter stage as there's only the cap to remove the ripple.

And add power protection of course.

I like the zener because these things are SUPER sensitive to the tiniest bit of over voltage or current. So, say, an unloaded wall-wart power supply. Just in case.




Yeah, none of it is needed. It's a sensitive circuit though and I don't want to inject noise with the negative supply. From the switching or otherwise (switching is above audio).


Otherwise I could have just used a single op amp IC with less than half a dozen parts to do this whole job too now that you mention it.



you'll be surprised how much difference in ripple regulation a series resistor makes before a cap.

think about it, if you wanted to make an audio low-pass filter you wouldn't do it without a resistor, right?

[Transmogrifox]

I think aggressive filtering is well justified.  These switchers create not only noise at fundamental, but there is often ringing on the switching edges at much higher frequencies.  One may think, but that's all ultrasonic -- but if there is any jitter or other changes making it anything but perfectly consistent on every cycle this all gets demodulated down to audible frequencies.  Getting rid of high frequency noise and switching transients keeps things from getting noisy.  I would also recommend a parallel small value <= 0.1uF in parallel with the output caps but my guess is because a charge pump usually already uses slower edges and lower frequency it will probably be ok.

[TheWinterSnow]

I was just trowing it out there that in my experience I haven't had instability or measurable ripple both aurally and with my o-scope at its lowest setting.  IN my formative days with pump-charge ICs I found that if you didn't have a large enough cap or the ESR was too high it added what sounded and looked like high frequency white noise more like the inverse of pink noise.  With the right caps even with high output current on the negative rail I see virtually no ripple and no less than what is already coming off my +9v rail.  For the amounts of gain I have used, no more than 15-20dB boosts I still haven't been able to audibly hear any whine/ripple.

I like to over-engineer stuff quite a bit, I just feel that unless you are pulling near maximum current the pump can pull ripple has never been an issue, so if I can save a few pennies and not have to worry about sourcing inductors than I typically go that route, although if ripple and stability were an issue in a particular design I might filter it more.

The overvoltage zener is smart though.  Although I use the LT1054 which has a maximum input votlage of 15v, a failing switchmode PSU could cause some nasty current spikes, might have to be something that I should implement.  I also just realized the price difference between the LT1054 and the 7660S-CPA, yeah meaybe that chip does need excessive filtering as maybe it does put out way more ripple.

[darron]

Although I use the LT1054 which has a maximum input votlage of 15v, a failing switchmode PSU could cause some nasty current spikes, might have to be something that I should implement.  I also just realized the price difference between the LT1054 and the 7660S-CPA, yeah meaybe that chip does need excessive filtering as maybe it does put out way more ripple.

i'd never seen these. ouch, yeah a little pricey. either one must be better than the max1044 (10V) though

FETs and charge pumps are such sensitive parts, always my first suspicion with a damaged pedal before checking