Generating +/- 14V from 3V supply

[MagicSmoker]

...
Thanks, I'd like to see how far I can push the boost based design at the moment. It's helping me understand the limitations of boost converters anyway I think.

Okay, then to minimize switching noise you'll want to use a large value of inductance (to ensure the converter operates in continuous conduction mode [aka - CCM]), a Schottky output diode (almost mandatory for CCM) and two stages of filtering comprised first of a small value, good quality ceramic capacitor (like X5R or X7R dielectric) in parallel with the usual large value electrolytic followed by a lossy ferrite bead* and another good quality ceramic capacitor to integrate away any spikes.

* - acceptable at low current; not a trick that scales up well, however.

[edit - corrected parenthetical definition of CCM as "continuous conduction mode" rather than "...current..."]

[rs20]

Okay, then to minimize switching noise you'll want to use a large value of inductance (to ensure the converter operates in continuous conduction mode [aka - CCM]), a Schottky output diode (almost mandatory for CCM) and two stages of filtering comprised first of a small value, good quality ceramic capacitor (like X5R or X7R dielectric) in parallel with the usual large value electrolytic followed by a lossy ferrite bead* and another good quality ceramic capacitor to integrate away any spikes.

+1, and also make sure that you oscilloscope probe ground lead is physically connected directly to one side of the output capacitor, and the probe itself as close as possible to the other.

[new299]

Okay, then to minimize switching noise you'll want to use a large value of inductance (to ensure the converter operates in continuous conduction mode [aka - CCM]), a Schottky output diode (almost mandatory for CCM) and two stages of filtering comprised first of a small value, good quality ceramic capacitor (like X5R or X7R dielectric) in parallel with the usual large value electrolytic followed by a lossy ferrite bead* and another good quality ceramic capacitor to integrate away any spikes.

+1, and also make sure that you oscilloscope probe ground lead is physically connected directly to one side of the output capacitor, and the probe itself as close as possible to the other.

For the next revision, I've put an SMA connector on the board and go from the board to the scope directly. Would this be suitable? Or are there potential issues.

[rs20]

Okay, then to minimize switching noise you'll want to use a large value of inductance (to ensure the converter operates in continuous conduction mode [aka - CCM]), a Schottky output diode (almost mandatory for CCM) and two stages of filtering comprised first of a small value, good quality ceramic capacitor (like X5R or X7R dielectric) in parallel with the usual large value electrolytic followed by a lossy ferrite bead* and another good quality ceramic capacitor to integrate away any spikes.

+1, and also make sure that you oscilloscope probe ground lead is physically connected directly to one side of the output capacitor, and the probe itself as close as possible to the other.

For the next revision, I've put an SMA connector on the board and go from the board to the scope directly. Would this be suitable? Or are there potential issues.

I'd call that overkill -- you're ultimately trying to get this power to a whole bunch of op amps and DACs and so on, so making a circuitboard that delivers clean power to just one SMA port isn't particularly useful for your end goal. Arranging for a good shielding solution instead, if that's deemed necessary, will give an improved waveform on your scope, and better performance for your end product.

[new299]

Okay, then to minimize switching noise you'll want to use a large value of inductance (to ensure the converter operates in continuous conduction mode [aka - CCM]), a Schottky output diode (almost mandatory for CCM) and two stages of filtering comprised first of a small value, good quality ceramic capacitor (like X5R or X7R dielectric) in parallel with the usual large value electrolytic followed by a lossy ferrite bead* and another good quality ceramic capacitor to integrate away any spikes.

Thanks, from my reading of the datasheet it looks like I should use ~100uH to run in CCM, so I'll try that.

For the filtering should I also try adding a pi filter? I tried this using a 10uH inductor but it didn't seem to help much.

Does it also make sense to separate the ground plane of the switching and non-switching parts? Some of the noise I'm seeing seems to be coupled through the ground as far as I can tell (as I still see noise on the linear regulators if only the ground is connected) would coupling the switching/linear parts of the circuit via an inductor be a sane thing to do?

[new299]

Magic, are you referring to the type of DC-DC converters that require the use of small transformers with primary and secondary windings?  The secondary is then DC rectified and filtered?

Yes. More specifically, the original Royer is a self-oscillating voltage-fed push-pull converter that delivers a nearly-pure DC output voltage even before any filtering. The biggest downside of the original Royer is its low efficiency (due to the high peak current in the transistors when the transformer saturates each half cycle), but this can be fixed by simply inserting an inductor in series with the primary center-tap (and flyback diodes across each BJT), which turns it into a current-fed self-oscillating push-pull converter. Then for the ultimate in noise reduction you can place a resonating capacitor across the primary which changes the voltage waveform from a square wave into a sine wave.

The resonant current-fed push-pull was very popular for CCFL backlights in laptops (which have since been rendered obsolete by white LEDs) so a lot of literature will be aimed at that application, but the principles are the same whether the output is in volts or kilovolts.

While I wait on parts/boards for the switching design I plan to try out the Royer circuit too. I think I understand its basic operation seems fun! So now I'm trying to search out a transformer to use.

Digikey has some listed as "for CCFL power supplies". But these tend to be quite low current (30mA). Do you have any particular recommendations?

[MagicSmoker]

...
+1, and also make sure that you oscilloscope probe ground lead is physically connected directly to one side of the output capacitor, and the probe itself as close as possible to the other.

This is a really good point that I should have mentioned! The usual 10cm ground lead with alligator clip is all but useless when scoping switchmode converter circuits because it picks up radiated magnetic fields. The OP really needs to use a spring tip, like in the picture below:

[MagicSmoker]

...Thanks, from my reading of the datasheet it looks like I should use ~100uH to run in CCM, so I'll try that.

The primary factors that determine whether a switchmode converter operates in CCM or CrCM (Critical Conduction Mode, aka BCM, or Boundary Conduction Mode) or DCM (Discontinuous Conduction Mode) are the load current and switching frequency; there is no single value of inductance, then, that guarantees CCM. If a converter has to contend with a wide range of loads then it will be difficult if not impossible to ensure CCM (exception: the synchronous variants of the classical switchmode topologies).

For the filtering should I also try adding a pi filter? I tried this using a 10uH inductor but it didn't seem to help much.

No, I already gave you a super secret, hard-learned tip before which is to split your filtering up into two stages - the primary filter, which consists of the switchmode inductor and the output capacitor, followed by a secondary "spike/ringing" filter which consists of a lossy ferrite bead and a really low-loss capacitor. Using a good quality inductor for the second filter often results in worse noise.

Does it also make sense to separate the ground plane of the switching and non-switching parts?...

Separating ground planes (and joining them together at one point via a trace, a lossy ferrite bead, etc.) is kind of an advanced topic that is rarely necessary and exceptionally difficult to do properly. Usually it is sufficient to make the 2nd layer of a 4 layer board (or the bottom layer of a 2 layer board) a solid ground plane and then physically separate the noisy stuff (like a switchmode power supply) from any low-noise analog stuff.

[new299]

For the filtering should I also try adding a pi filter? I tried this using a 10uH inductor but it didn't seem to help much.

No, I already gave you a super secret, hard-learned tip before which is to split your filtering up into two stages - the primary filter, which consists of the switchmode inductor and the output capacitor, followed by a secondary "spike/ringing" filter which consists of a lossy ferrite bead and a really low-loss capacitor. Using a good quality inductor for the second filter often results in worse noise.

Ok, thank you for your help, it's extremely useful. "lossy" would mean higher resistance? 30Ohms? I don't have a good feel for that.

I kind of feel like the ferrite bead is absorbing the EMI? I guess I need to read up more of higher frequency transmission, but my current model is that significant potion of the transmission occurs in the EMF around the wire and the ferrite bead absorbs this.

[MagicSmoker]

...
Digikey has some listed as "for CCFL power supplies". But these tend to be quite low current (30mA). Do you have any particular recommendations?

You can certainly reuse the core from a COTS (commercial off-the-shelf) CCFL transformer, but all of the windings (except, maybe, the feedback) are going to have the wrong number of turns for your application and the bobbin will be divided into several sections. The nice thing about the Royer is that almost any transformer will work as long as it doesn't have too much of an air-gap (that is, does not have too-soft of a saturation knee). The frequency it will oscillate at is determined by the core area, number of turns and saturation flux density. I would buy a small ferrite toroid and some #30 wire-wrap wire (and a stripper for it, if necessary; pro-tip: most wire-wrapping tools come with a stripper) and start tinkering. Begin with about 1 turn per volt for each winding (this also applies to each half of any center-tapped windings), assume 4V of base drive for the BJTs and tweak as necessary. So, for example, you might have a primary that is 3T x 2 (ie - 6CT - for 6 turns center tapped) a 4T feedback winding, and a 30CT secondary.

..."lossy" would mean higher resistance? 30Ohms? I don't have a good feel for that.

"Lossy" refers to the ferrite core material (and the resistance of the windings) which results in an inductor with low Q, which is good for filtering; bad for power conversion or tuned circuits.

One other thing with regards to the boost converter - the inductor must be a closed-form shape. For example an E-core or toroid is good, but a rod (or air core) is bad. This is to prevent spraying magnetic fields everywhere.