an evening with the ICL7660

[cellularmitosis]

The ICL7660 is a switched capacitor voltage converter for e.g. making a negative rail from a positive rail.

I decided to fool around with one, specifically to see how bad the switching noise was, and to try out some circuits to tame the noise.

[cellularmitosis]

Circuit A: straight from the datasheet.

Schematic:



Circuit construction:

[cellularmitosis]

Circuit A into a 10k load.

Input voltage: 9.26V



Output voltage: -8.39V



Probing technique:



Noise:

[switcher]

The Fluke 8060A uses one. I've recently repaired two 8060As, and had to change to 7660 in each of them, both producing +0.5VDC (yes, positive half a volt! hahaha!)

[cellularmitosis]

and Circuit A into a 1k load:

Input voltage: 9.25V
Output voltage: -7.84V

Noise:

[cellularmitosis]

Circuit B is the simplest attempt possible at eliminating the high frequency switcher noise: a ferrite bead is added before the reservoir capacitor.

Schematic:



Construction:

[cellularmitosis]

Circuit B into a 10k load:

Input and output voltages were identical to Circuit A into a 10k load.

Noise:



















Surprisingly (to me), in this arrangement the ferrite bead basically did nothing.

[cellularmitosis]

and Circuit B into a 1k load:

again, input and output voltages were identical to Circuit A into a 1k load.

Noise:















Again, the ferrite bead does very little in this arrangement.

[cellularmitosis]

Next for Circuit C we try to improve the behavior of the reservoir capacitor by adding a 1uF polyester film cap ("box" cap).

Schematic:



Construction:

[cellularmitosis]

Circuit C into a 10k load:

Input voltage: 9.25V
Output voltage: -8.39V

Noise:

















Now we are starting to see this noise budge a little!

[cellularmitosis]

and Circuit C into a 1k load:

Input voltage: 9.25V
Output voltage: -7.85V

Noise:













Again, the box cap makes a very apparent difference here.

[cellularmitosis]

for Circuit D we get a bit more serious with our output filter, by adding a 100uH inductor and a second pair of capacitors.  Additionally, the ferrite bead is moved in between the two capacitor stages.

Schematic:



Construction:

[cellularmitosis]

Circuit D lowered the noise to the point that the noise floor of my measurement setup became problematic.

To overcome this, I moved the setup into my home-made faraday cage, which is an aluminum dutch oven which has been fitted with a BNC connector.





The lid is held down with 3 bags of lead weights, each weighing 5 lbs.  This ensures a good EMI seal.



The BNC cable used is 1.5 feet of Canare L-5CFB RG6 double-shielded coax.  I had to move to this cable as the cheap-o BNC cables allowed too much noise in.

[cellularmitosis]

Scope noise floor with all BNC connections capped:





Scope + faraday cage + powered-off circuit noise floor:



As best I can tell, these are switching spikes from Rigol's internal power supply.

[cellularmitosis]

Circuit D into a 10k load.

Input voltage: 9.26V
Output voltage: -8.39V

Noise:



















Here we see a HUGE change in the amplitude and waveform of the noise!

[Vgkid]

That is a really informative post, thanks.

[cellularmitosis]

and Circuit D into a 1k load.

Input voltage: 9.25V
Output voltage: -7.84V

Noise:









Our noise is now mostly just a 24kHz sine wave, which should be relatively straight-forward to filter out with a regulator.

[cellularmitosis]

So... that's where I'm at right now.  Tomorrow night I'll try tacking a regulator onto this circuit to see how much of the 24kHz ripple I can get rid of.

I think I'll try two approaches: a simple LM79XX, and also a shunt regulator (TL431), and see how they compare.  I suspect the PSRR of the LM79XX will be lower than the TL431, as I believe their performance starts to fall off above 1kHz or so.

Until next time!

[tautech]

Nice thread and a good amount of work you have done. 
But for these result to have any meaning you must list some characteristics of the components used.
Namely the Electrolytics, low ESR or GP??? 

[cellularmitosis]

Good point.

The 10uF electrolytics are basically bottom of the barrel (cheap, general purpose).  They either came with my Jameco capacitor assortment, or were ordered from e.g. Tayda / ebay (they've become mixed together at this point).

The film cap is a Kemet ($0.56): http://www.digikey.com/product-detail/en/R82DC4100DQ60J/399-5447-1-ND/1930840



The LM7809 is also ebay / tayda (cheapest I could find).

The inductor is also from ebay.

The ferrite bead is (surprisingly, the only) through-hole axial 0-turn available on digikey ($0.16): http://www.digikey.com/product-detail/en/28L0138-40R-10/240-2439-1-ND/806799

[tautech]

If you did some tests with ML ceramic X7R caps as are often specified with switchers we would then have real results we can use. 

[LukeW]

This thread is worth it just for the DIY Faraday cage alone! 

But I would use a 50-ohm coax, which seems more "standard" as what you'd expect in test/instrumentation applications, as opposed to 75-ohm RG6 TV antenna cable.

Great post.

[AndyC_772]

If you did some tests with ML ceramic X7R caps as are often specified with switchers we would then have real results we can use. 

That was exactly my thought. Electrolytic caps have essentially no ability to filter high frequency noise and transients (for some definition of "high"), so there was no way the circuit was ever going to be quiet with just an electrolytic output cap. Stretching out the circuit over quite so much board area isn't helping either, even though the dead-bug style of construction is otherwise pretty good.

I've used the 7660 and generally found it does a good job, but I'd always use it in surface mount form with X7R (or maybe X5R) ceramic caps located right next to the device.

I'd suggest starting again, but instead of leaded caps, solder some SMT caps directly between the pins of the 7660 and the ground plane. No leads == minimal series inductance. Something like this should work well and shouldn't be too fiddly to work with:

http://uk.farnell.com/murata/grm31cr71e106ka12l/cap-mlcc-x7r-10uf-25v-1206/dp/1828837

I'd expect the result to be not too far from the multi-stage filtered method in terms of performance, and about 1/4 the size.

[Andreas]

Hello,

nice test: never thought that a ICL7660 is capable to produce such spikes.
Some annotations:

From the first zoomed picture you can see that the rise time is around 100ns.
This is a equivalent frequency of around 3 MHz.
So you should select your components that they have maximum filtering at the 3 MHz point.

A ferrite bead has its best filtering around 100 MHz or above. -> you will not need it.
When I calculate the XL = j*2*PI*f*L = around 2k Ohms for 3 MHz and 100uH.
So I would expect a dampening in the range of 2K divided by the ESR of the capacitors.

If you select the capacitors for optimum filtering you will have to use either a (4 wire kelvin connected) X7R with 1uF for the 3 MHz together with a low ESR 100uF electrolytic. Or a 100 - 220nF foil (higher inductance) together with a around 10-22uF + a larger electrolytic.
But be aware that the component values are only valid for short leads (every half mm counts). Otherwise the self resonance will be shifted.

Edit: if you do not need the full output current you could try a small resistor 10-100 Ohms in series to the floating capacitor C3.

With best regards

Andreas

[tautech]

To his credit cellularmitosis stated in the first post "as per datasheet"or words to this effect.
However "switchers" are noisy and one must always follow the minimum guidelines.
Low ESR caps are a must and even a bead tantalum (as much as I dislike them) would have been a far better choice than a GP electrolytic.
I had to use a 100 uF SMD Tant before I was happy with the ripple for a MCP1640 in SOT23/6 drawing 3 mA and 20 mA peaks for 70 mS.
It can be done as I'm sure cellularmitosis will show.

[bktemp]

Did you try to measure the noise at the input?
The noise at the input is as bad (or even worse) than at the ouput. Therefore you have to add the same amount of filtering also to the input to avoid the switching noise getting into your supply rails.

Because of the noise generated, I have banned 7660 and similiar ics from almost all of my circuits.
Using a simple pwm controlled buck regulator (like LM2574) in an inverting configuration often produces less noise than a switched capacitor voltage inverter.

[Kjelt]

Thanks for this very nice post. The faraday pan :

[BravoV]

Thanks for this very nice post. The faraday pan :

+1  thanks for sharing this, really informative thread.

[rdl]

I have used this device in a couple of projects to generate a negative rail. I always supplied it with a well regulated input voltage and never found noise on the output to be much of a problem. Yes, there is some, but it is controllable and can be filtered out. Generally I would also regulate it's output, that's where I had the most problems. It simply has no output current capacity. Even a small load of 10-20 mA will cause it's output to drop by many volts.

[cellularmitosis]

Did you try to measure the noise at the input?
The noise at the input is as bad (or even worse) than at the ouput. Therefore you have to add the same amount of filtering also to the input to avoid the switching noise getting into your supply rails.

I was wondering about that!!

[cellularmitosis]

Here's the noise measured at the input (e.g. at the AA batteries) for Circuit D into a 1k load:















Looks like we certainly need to worry about spikes propagating back up into the source rail!

[cellularmitosis]

and here's the input noise of Circuit D with the 1k load:















(and here's the zoomed in portion of the tail end of that plateau):

[tautech]

Measurement at the batteries was wise. 
A "stiffer" source would also have less ripple.
Keep up the good work.

[cellularmitosis]

Circuit E:

Ok, first stab at post regulation.  I decided to just try a simple LM79L05 with a 10uF cap, and the results surprised me (I expected PSRR to be low at 24kHz).

Schematic:



Construction:

[cellularmitosis]

UPDATE: Oops, the 79L05's input voltage is only about -4V, so it isn't in regulation.  These results should be ignored.

(also, I changed the load resistors to 4.7k and 470R to try and keep true to the original intention of taking data around 1mA and 10mA loads).

Circuit E into 4.7k:

Output Noise:



At first, this appears to be just the noise floor of the scope.  However, by zooming out a bit and using an FFT, we can see that there is a tiny (<1mV?) component of ripple around 24kHz, which the scope spikes are riding on top of:

[cellularmitosis]

UPDATE: Oops, the 79L05's input voltage is only about -4V, so it isn't in regulation.  These results should be ignored.

Circuit E into 470 ohms:

The output noise is similar to the 4.7k load, with a little bump at 24kHz.

[cellularmitosis]

Ah, whoops.  It turns out this 79L05 is not actually in regulation.  I forgot to measure the 79L05's input and output (DC) voltages:

4.7k: -4.45V in, -0.891V out.

470R: -4.12V in, -0.541V out.

Drat, I am not regulating at all.  Interestingly, the PSRR still seems to be in effect, which seems strange...

I'm not sure how I was able to get -7V out into 1k, but the regulator pulls that down to -4V with only a 4.7k load...

[Andreas]

For me it looks as if you exchanged Gnd and Output pin on your 79L05.
Caution: often bottom view in the data sheet.

On the other side. quiescent current is around 2(-5) mA for the 79L05.
I prefer using low drop regulators with lower quiescent current.
LT1964 / TPS72301.

With best regards

Andreas

[theatrus]

I've favored the LM2662 in this application, especially due to its high frequency mode (250kHz).

[rdl]

The ICL7660 has a linear voltage drop from 0 to about 40 mA. You can't get -5 volts from +5 volts unless there is almost no current being drawn. How much power does the 7905 use just sitting there? You can power a low current reference such as an LM385 from a 7660 but I don't think you'll be able to run a 79L05 with it.

[cellularmitosis]

For me it looks as if you exchanged Gnd and Output pin on your 79L05.
Caution: often bottom view in the data sheet.

Bingo, we have a winner   I reverse the lead order.

After correcting the lead order, I measured -4.95V and -4.97V on the scope into 4.7k and 470R loads, so I am definitely regulating now.

Here's the output noise into the 4.7k load:











So it looks like there is still about 1mV of 24kHz ripple, with some spike riding on top of it, which are from the scope or from my environment.

[cellularmitosis]

and here is Circuit E into the 470R load.

Output noise:



If you look closely you can make out the 24kHz ripple:



Zooming in one more step makes this more obvious:



and the magnitude of the ripple minus the spikes seems to be just over 1mV:



Here's a closeup of one of the 130kHz spikes:

[cellularmitosis]

For comparison, here are a couple of scope shots after disconnecting the battery.  This establishes that the 130kHz spikes are not coming from the ICL7660.

[dannyf]

my home-made faraday cage, which is an aluminum dutch oven

Nicely done,

[dannyf]

Our noise is now mostly just a 24kHz sine wave, which should be relatively straight-forward to filter out with a regulator.

PSRR at higher frequencies are usually low, 40-50db @ 20Khz would be typical.

If you want to generate a negative voltage from a positive voltage, you may want to try a 555 timer + charge pump. Fairly good to 20ma or so.

[cellularmitosis]

Wow, its been almost a year since I last messed around with the 7660.  Time to revisit this thread!

A comment from the previous circuits expressed interest in seeing the results of using ceramic capacitors.  I decided to investigate that using some through-hole 1uF ceramic caps I had on hand.

Circuit F:

For circuit F, we start over using 1uF ceramic caps.


Schematic:




Construction:

This time around I tried to focus on keeping the critical leads as short as possible to reduce stray inductance.




Measurement Setup:

All of these circuits will use the same as before, using the aluminum dutch oven faraday cage:






Performance into 4.7k load:

DC-coupled probe indicates an output voltage range of -8.64 to -8.0V.



AC-coupled probe indicates 400mV ripple.




Performance into 470R load:

DC-coupled probe indicates an output voltage range of -6.16 to -4.32V.



AC-coupled probe indicates 1700mV of ripple.




Comments:

Wow!  Just switching to ceramic caps and focusing on keeping the leads as short as possible on the flying capacitor has completely eliminated the inductive spike behavior we were seeing in the previous batch of circuits.

The 470R load sees a huge 1700mV swing with such a tiny 1uF flying capacitor.

Interestingly, all of these newer circuits seem to be switching at ~6.5kHz, quite a bit lower than the nominal 10kHz from the datasheet.

[cellularmitosis]

Circuit G:

For circuit G, we add an LC output filter stage, using a 100uH inductor and another 1uF ceramic cap.

Note that these LC values calculate to a filter cutoff frequency of nearly 16kHz, so we shouldn't expect much yet...


Schematic:




Performance into 4.7k load:

DC-coupled probe indicates an output range of -8.64 to -8.08V:



AC-couples probe indicates a ripple of 356mV:




Performance into 470R load:

DC-coupled probe indicates an output range of -6.72 to -4.88V:



AC-couples probe indicates a ripple of 1760mV:




Comments:

A cutoff frequency of 16kHz just isn't enough to put a dent in this waveform yet.

Note: I didn't actually do the cutoff calculation until several circuits later, so adding a larger value cap in the LC filter won't appear until the last two circuits.

[cellularmitosis]

Circuit H:

For circuit H, we add a capacitance multiplier low pass filter, using a 2N3906 with a 1k resistor and 1uF ceramic cap.


Schematic:




Construction:




Performance into 4.7k load:

DC-coupled probe indicates an output voltage range of -7.68 to -7.36V.



AC-coupled probe indicates a ripple of 11mV.




Performance into 470R load:

DC-coupled probe indicates an output voltage range of -4.48 to -4.16V.



AC-coupled probe indicates a ripple of 37.6mV.




Comments:

The capacitance multiplier had a huge impact on the output ripple, at the cost of some additional dropout.

[cellularmitosis]

Circuit I:

For circuit I, I wanted to try a technique I read about in an article at radio-electronics.com, where a voltage divider is used to bias the capacitor multiplier slightly further into its linear region.  I tried this by adding a 6.8k resistor.


Schematic:




Performance into 4.7k load:

DC-coupled probe indicates an output range of -6.48 to -6.16V:



AC-couples probe indicates a ripple of 13.6mV:




Performance into 470R load:

DC-coupled probe indicates an output range of -4.00 to -3.84V:



AC-couples probe indicates a ripple of 34.8mV:




Comments:

Additional biasing on the capacitance multiplier didn't significantly improve our ripple, and it cost us some additional dropout.

[cellularmitosis]

Circuit J:

For circuit J, we beef up the LC filter by adding a (cheap) 100uF electrolytic in parallel with the 1uF ceramic.  Additionally, we remove the 6.8k resistor, which reverts the capacitor multiplier to its previous configuration.


Schematic:




Construction:




Performance into 4.7k load:

DC-coupled probe indicates an output voltage range of -7.60 to -7.36V.

(I am starting to see that this isn't such an accurate way to measure the output voltage.  The Vmin and Vmax are measured at lower resolution and don't line up with the Vpp ripple).



AC-coupled probe indicates a ripple of 1.44mV.




Performance into 470R load:

DC-coupled probe indicates an output voltage range of -4.40 to -4.28V.



AC-coupled probe indicates a ripple of 1.84mV.




Comments:

OK!  We are finally getting near the noise floor of our scope.  The combination of an LC filter and capacitor multiplier appears to be effective.

Keep in mind though that the capacitor multiplier isn't a regulator.  An unregulated but heavily filtered negative rail should be fine for many op amp applications, but not appropriate for situations which need true voltage regulation.

[cellularmitosis]

Circuit K:

For circuit K, I wanted to see the performance contribution of the 100uF electrolytic without the capacitor multiplier.


Schematic:




Performance into 4.7k load:

DC-coupled probe indicates an output voltage range of -8.40 to -8.17V.



AC-coupled probe indicates a ripple of 8.80mV.




Performance into 470R load:

DC-coupled probe indicates an output voltage range of -5.68 to -5.36V.



AC-coupled probe indicates a ripple of 41.2mV.




Comments:

Here we see a similar level of performance as the 100uH / 1uF LC filter combined with the capacitor multiplier.  That would imply that the the 100uH / 100uF LC filter contributes about the same level of filtering as the 1k / 1uF capacitor multiplier.

[Kalvin]

Very nice and informative posting! I too found the Faraday cage very intuitive and clever

The input of the ICL7660 can be filtered with the similar kind of LC filter as the output in order to reduce the ripple seen by the power supply feeding the ICL7660.

Looking forward for more measurements from cellularmitosis. Well done!

[cellularmitosis]

The input of the ICL7660 can be filtered with the similar kind of LC filter as the output in order to reduce the ripple seen by the power supply feeding the ICL7660.

Definitely going to give that a try soon!  Thanks for the suggestion

[ignator]

I didn't see this thread last July 2014.
I'm curious why you don't have any small value decoupling capacitors. Something like 27 to 47 (or 270) pico farads that can do something with the fast nano second pulse edges. When I was working, we compiled Vector Impedance measurements for common stocked capacitors from engineering stock. 1 micro farad if I recall self resonated below 3MHz, and was so inductive above this, that it did nothing to decouple noise. You need to keep lead length short as possible, if you put any capacitor of these small sizes in your circuit.
Sorry, I retired a few years ago, and I didn't keep a copy of various ceramic capacitors and their self resonance frequencies. These were leaded parts as my job was to bandaid avionics that failed emission tests, and the fix needed to be  added to existing circuit assemblies, as there was no time to respin the PWB before certification of the product.
Your circuit board ground plane should be connected to the faraday shield with a wide brass/copper/aluminum strip so the inductance of your ground wire connecting the board to your earth reference is a low as possible. My rule of thumb is every 25mm of wire looks like 25 ohms at 100Mhz (22AWG typically) because of its inductance. This impedance allows RF noise to 'float' on the ground.

[cellularmitosis]

When I was working, we compiled Vector Impedance measurements for common stocked capacitors from engineering stock. 1 micro farad if I recall self resonated below 3MHz, and was so inductive above this, that it did nothing to decouple noise.

Sorry, I retired a few years ago, and I didn't keep a copy of various ceramic capacitors and their self resonance frequencies.

Thanks for the reply.  Man, I would love to compile such a set of measurements myself!  What sort of equipment did your department use to make those sort of measurements?

I been contemplating looking up schematics for older / simpler instruments to see if it would be feasible to produce reasonably priced similar circuits with today's components.

Your circuit board ground plane should be connected to the faraday shield with a wide brass/copper/aluminum strip so the inductance of your ground wire connecting the board to your earth reference is a low as possible. My rule of thumb is every 25mm of wire looks like 25 ohms at 100Mhz (22AWG typically) because of its inductance. This impedance allows RF noise to 'float' on the ground.

Thanks for the tip.  Maybe I can tap a short screw directly into the bottom of the dutch oven and mount a copper braid.

[paulie]

I found this thread very interesting from the start because of using the close cousin 7662 quite a lot. 7660 voltage limitation too restrictive for me. Early on I settled on inductor and couple cheap caps to get ripple below 100mv using 1k resistor load. Also playing with the pump caps had some benefit to minimize drop. IIRC one was much larger than the other and contrary to datasheet recommendations.

[ignator]

cellularmitosis:
I looked at ebay, here's a broken one, but it's exactly what we used HP4815A:http://www.ebay.com/itm/Vector-Impedance-Meter-4815A-RF-Hewlett-Packard-/111671742452?pt=LH_DefaultDomain_0&hash=item1a00278ff4

Note the crank handle, this controlled the frequency generator, and the meters, one indicates ohms, the other phase angle. You would crank the frequency up until you find a minima of ohms, and for a cap. any frequency above this was inductive (lead length and internal paths inductance took over). If you get one, make sure it comes with the probe.

The biggest problem in packaging avionics, was to find a way to get the PWB boards ground planes connected with a wide sheet of connection to the box chassis (the infinite sheet conductor in field theory). Screws were sometimes used to multiple standoffs that were pressed into the aluminum chassis. Sometimes we would make clamps from aluminum bars to sandwich a PWB with edge plated, as well EMI gaskets (spring brass with nickel plate). 
As both the radiated emission, conducted emission, and then the radiated susceptibility, conducted susceptibility compliance evolved over the years, the ability to make your PWB grounds low inductance to LRU (line replaceable unit) chassis, and to make the LRU mount to the aircraft mounting tray to have extremely low impedance was the game to pass tests.
Susceptibility became a big deal, as airframes went from 100% aluminum, to composite, so test levels of HIRF (high intensity radio/radiate frequency) went from 5 v/meter to 1000s v/m.
Everything became low phosphorus nickel plating between the chassis and PWB. Original plating was just zinc chromate on the aluminum when I started working in 84. Aluminum grows oxide almost immediately, and this is non conductive, hence some sort of conductive plating on aluminum. PWBs use either Nickel or tin lead still. 

Note the screws needed to be spaced 20mm minimum (1/4 wave length of max frequencies HIRF was up to 18GHz when I retired), the point being a single screw probably will work with just your single switching circuit, but put a whole high speed processor with lots of parallel address and data buffers all switching at the same time, and you need to have a real good ground connection to prevent the PWB from being an antenna and radiating. As well, if you have noise on your ground, instead of decoupling capacitors quieting the noise, you coupled noise from the ground to the circuit signal (very painful lesson when your out of time, and the emission testing gets worse with every part you add to decouple). 
Also the point about a single screw path is an inductor, as all current flows through just this. The larger the diameter, the lower the inductance. So keep this in mind. Multiple in parallel is better.

[Tombs Balsam]

A serious post, thanks for sharing

[cellularmitosis]

cellularmitosis:
I looked at ebay, here's a broken one, but it's exactly what we used HP4815A:http://www.ebay.com/itm/Vector-Impedance-Meter-4815A-RF-Hewlett-Packard-/111671742452?pt=LH_DefaultDomain_0&hash=item1a00278ff4

Note the crank handle, this controlled the frequency generator, and the meters, one indicates ohms, the other phase angle. You would crank the frequency up until you find a minima of ohms, and for a cap. any frequency above this was inductive (lead length and internal paths inductance took over). If you get one, make sure it comes with the probe.

Wow, very interesting!  It didn't even occur to me that you could make such measurements with an instrument which doesn't have a cathode ray tube or LCD screen!  What's even more interesting is that, since you are just looking for the minimum amplitude value, I could potentially cobble together an instrument to do this with a DDS which performs a sweep, and an ADC which looks for a minimum value.  Hmm...

Aluminum grows oxide almost immediately, and this is non conductive, hence some sort of conductive plating on aluminum.

Funny you should mention that, it had been months since the last time I used the aluminum dutch oven, and I had to use a file to rough up the lip where the lid meets the pan in order to get a good noise floor .  I might also consider bolting a copper strap to connect the lid to the pan...

Note the screws needed to be spaced 20mm minimum (1/4 wave length of max frequencies HIRF was up to 18GHz when I retired), the point being a single screw probably will work with just your single switching circuit, but put a whole high speed processor with lots of parallel address and data buffers all switching at the same time, and you need to have a real good ground connection to prevent the PWB from being an antenna and radiating. As well, if you have noise on your ground, instead of decoupling capacitors quieting the noise, you coupled noise from the ground to the circuit signal (very painful lesson when your out of time, and the emission testing gets worse with every part you add to decouple). 
Also the point about a single screw path is an inductor, as all current flows through just this. The larger the diameter, the lower the inductance. So keep this in mind. Multiple in parallel is better.

Thanks again for sharing your tips, you've obviously got a wealth of knowledge here!

[gildasd]

Sorry to re-spawn this topic, but it has been real useful:

The 47puff capacitor is especially marvellous to kill off the noise that gets dumped into the ground by the 7660...
I used a simple RC filter on the output and I have reached the floor of my 15 Mhz oscilloscope. The flat screen 1m away is more noisy...
This is a couple of magnitudes less noise than a 555 negative supply.
I can reach this level of low sound with the 555, but the output is too weak to even drive the -Vcc of a 741 for a -2.5 follower...
Even dumping unfiltered 12V into the 7660 only generates marginally more noisy should the headroom be needed.

A big thanks!

Pictures:

Easy to spot the inspiration, the output filtering is minimal as this is to create a negative voltage reference via an Opamp.


And I'm getting these results on a breadboard


Top: 8.88V rail bottom 7660 output.


And with the holy 47 puffs (with the scope set at 5mV per division, it's max)

[cellularmitosis]

A small update to this thread:

I turned this into a PCB design, which can be used as a "daughter board" (it can be mounted on a standoff):

https://github.com/pepaslabs/NegativeRail

The performance seems to match the prototypes.

Shown below is the noise into a 1k load, and the noise with the power disconnected (i.e. the noise floor).  I'm guessing the ~750kHz spikes are coming from the scope itself.

[ignator]

Cellularmitosis:
It looks like your PWB is just 2 sided. So you don't get any benefits from a ground plane, where if the plane has a .004 inch dielectric spacing from traces to gnd/pwr plane, the capacitance of this provides LC filters (the L being the intrinsic inductance of the circuit traces).
Your measured noise is pretty good considering that you are not using a true earthed ground plane in your PWB that is bonded with low inductance to earth. You have lots of impedance between your circuit card and the shielding of the dutch oven. The red/black wire connecting the output to your external scope is a very long inductor and antenna, have you tried a coax for this?

A photo of the bottom side would be nice to see. That standoff plated hole, is it connected to ground on your board? That would be the path of providing a low inductance ground if you solder staked a solderable standoff in that hole. Keep in mind every wire is an inductor. In aircraft wiring, where twisted shielded pairs are implemented, the ground drain wire that is soldered to the shield, and connected to aircraft ground, represents 25 ohms of impedance per inch at 100MHz. And aircraft installers like to make that wire longer then needed. I've seen them a meter long. The ME and flight line mechanics don't understand the RF issues with this drain wire being excessively long. I've always tried to have my installation drawings show this at 4 inches maximum (and it's ignored by the airframe manufacture). As this is where I did my qualification testing at with the product test wire bundle. The shielding is more for preventing EMI upset from external noise. As if your box leaked RF out, it would fail the emissions test. The game was, to do as you are doing, and filter the noise at it's source.
Keep having fun.

[Zero999]

for Circuit D we get a bit more serious with our output filter, by adding a 100uH inductor and a second pair of capacitors.  Additionally, the ferrite bead is moved in between the two capacitor stages.

Schematic:

Be careful with LRC circuits. If the circuit is underdamped (low ESR capacitors and inductors) then it's possible to excite its resonant frequency, causing oscillation. Adding some resistance in series with the capacitors or parallel with the inductor can help damp any resonance.

Try connecting a load to the output of the filter via a transistor and switching the transistor on and off at a few hundred Hz. Is there any ringing on the output of the filter when the load is connected/disconnected?

[schmitt trigger]

Very interesting thread! The idea of using the dutch oven as a Faraday cage is brilliant!
Two thumbs up:
 

Back in the days (late 1970s) where good audio opamps required mandatory dual supplies, I used an ICL7660 to generate the negative voltage from a single 9V battery, which I used in a small portable project.

I did notice the whistling, but as I did not have a scope at the time, I never knew what was going on.

Thanks for sharing, it has brought fond memories.

[ignator]

Quote from: Hero999 on May 20, 2017, 12:27:10 PM

Be careful with LRC circuits. If the circuit is underdamped (low ESR capacitors and inductors) then it's possible to excite its resonant frequency, causing oscillation. Adding some resistance in series with the capacitors or parallel with the inductor can help damp any resonance.

Try connecting a load to the output of the filter via a transistor and switching the transistor on and off at a few hundred Hz. Is there any ringing on the output of the filter when the load is connected/disconnected?

______
Very good advice. I know back in the early 90s, when we were transitioning from 6MHz 80186 to 25MHz 80486 processors, we would use 3 capacitors on each chip that would drive octal or more width bus interface drivers. That indeed was our worry that we would make a tank circuit amongst the parallel capacitors.
Your test method is a neet method to try to induce ringing. The Radiated emission test curve was tough to pass. And harmonics in the communication band had notches in those bands, that made it tougher to qualify the product. They were modules that slid into a card cage, and only 6 pins on the 130 pin backplane connector provided ground to the board.  We were stuck trying to decouple each part to prevent them from radiating.

[VanitarNordic]

The designed low pass filter (LC) seems that it does not cut below the ripple frequency. the L and C values must be changed to make it more effective.

[dudester]

...

...

...

...

Guys, is this design or parts of it still a best practice given the new (sort of) ICL7660S with it's "Boost Pin (Pin 1) for Higher Switching Frequency"?

What are the best parts to keep to preserve maximum output voltage/current while keeping any noise at lowest?

[Fronberry]

...

...

...

...

Guys, is this design or parts of it still a best practice given the new (sort of) ICL7660S with it's "Boost Pin (Pin 1) for Higher Switching Frequency"?

What are the best parts to keep to preserve maximum output voltage/current while keeping any noise at lowest?

Looking at the ICL7660S datasheet, it suggests that the boost pin provides about 3.5X increased oscillator frequency.  Seems like this would only lead to having lower allowable capacitance values on the LC filters, with the tradeoff of having lower power efficiencies at the lower end of the frequency band.

Also, does anyone have any info on the mechanism of the boost pin upon the RC oscillator?  Based on my understanding of RC oscillators, the output frequency is independent of supply voltage, so hopefully there are some interesting guts going on here.

[GigaJoe]

why don't use 79LXX on output ? ( ... 08 06 05 )

[Zero999]

why don't use 79LXX on output ? ( ... 08 06 05 )

This was investigated by the original poster and does work, but it's better to use a voltage regulator, with a lower quiescent current.