Multiple DC voltages, regulators, and capacitance multipliers reducing ripple

[David Hess]

What is capasitance multiplier advantage?  Why you put it BEFORE regulator?

If the capacitance multiplier is placed after the regulator, then it would spoil the regulator's load regulation.

The advantage of the capacitance multiplier is simplicity and ease of understanding.  But its high frequency line rejection is no better than a good regulator and low frequency line rejection is not usually a problem.  If high frequency rejection is a problem like with a linear regulator following a switching preregulator, then an LCR filter is a better choice.

The two serial  LM317 is quite good too, isnt it?

That helps for line regulation but does nothing for inherent noise of the second regulator.  If low noise and better performance is the objective, then a discrete design will be best.  But it is rare that the power supply rejection of the following circuits is insufficient for a "noisy" LM317 regulator.

[MiDi]

But it is rare that the power supply rejection of the following circuits is insufficient for a "noisy" LM317 regulator.

This mostly is true for low frequencies, for higher frequencies the PSSR for most circuits/ICs will drop.
If we are talking about precision circuits, the alarm bells should ring!
A true 20bit ADC/DAC has a resolution of around -120dB refered to its reference!
You have to do the math and measure the circuit if it meets your specs.

[David Hess]

But it is rare that the power supply rejection of the following circuits is insufficient for a "noisy" LM317 regulator.

This mostly is true for low frequencies, for higher frequencies the PSSR for most circuits/ICs will drop.

At higher frequencies, the output capacitor's impedance is lower and RLC decoupling becomes more effective.

If we are talking about precision circuits, the alarm bells should ring!
A true 20bit ADC/DAC has a resolution of around -120dB refered to its reference!
You have to do the math and measure the circuit if it meets your specs.

Low phase noise oscillators and amplifiers have a need for low noise power.  Low jitter singled ended logic does also but really differential logic which actually has power supply rejection should be used instead if this is an issue.

[T3sl4co1l]

I do wonder why there are no 5VDC (USB; say 4V to 15V input) non-isolated SEPIC modules providing filtered, low-ripple 5VDC and/or 3.3VDC. You could even add some diodes, so one could power the project from USB or a separate power supply. A lower ripple (especially at lower frequencies) would help a lot with all kinds of sensors.

No?  Have you checked uModules and such?

If they're too expensive, well... you should've said so.

Not sure if SEPIC, but flying inductor style I know at least.  They tend to be expensive and noisy (think I've seen a few claiming good EMI).

I believe, but have no proof, that supply ripple or glitches cause a large part of stability issues in hobbyist/maker microcontroller projects. I know they are a big problem with many SBCs.

Power, bypassing, lack of filtering on signal wires, lack of ground plane or stitching, cables just fricking everywhere... take your pick.  When one don't understand what a reference plane is, one is sure to make a dog's breakfast...

It's a common misconception that bulk caps are for storing energy.

Nothing in reality is that simple They do store potential energy in an electric field.

Yes, but not necessarily -- which reinforces the point!

Suppose you had a nonlinear capacitor, where its zero-bias value is small, then increases, peaking at some bias voltage, then falling again.  You'd get very little energy storage in the first so-and-so volts, then a big slug of energy where the capacitance is large, then not much more as voltage continues to rise.

You'd have one of these:
https://www.digikey.com/en/product-highlight/e/epcos/ceralink-capacitors

The trick is, if "some bias voltage" is zero, this describes normal ceramic capacitors, which aren't very helpful on account of losing capacitance under bias.  These things are biased with an electric field while hot, then cooled below Tc, "freezing in" the field -- making an electret.  This basically shifts the zero-bias point to the opposite direction, so that when applied field + built-in field cancels out, E is zero and C is maximum.  Cool, huh?

Downside: besides being large ceramic caps (pricey to begin with), and rather boutique (and that much more expensive yet), they can't be reflow soldered, because Tc is below soldering temperature.  They'd just be annealed to normal zero-bias caps, and you'd have to somehow reflow them at negative bias to keep it.  Or something.  So, they're hand soldered, carefully.

It's also a somewhat special case, because power converters are more likely to need energy storage rather than arbitrarily low ripple.  In that case, both factors (bypass performance and energy storage) are important.

But anyway, yeah, from a physical perspective, if we don't need energy storage, and didn't have any practical limitations, we would gladly choose a capacitor that somehow has zero capacitance everywhere but over the voltage range it needs to operate at!

My own field is computational materials physics, so that is basically second nature to me now. Whenever I find a new formula describing some interesting phenomena, I do a quick dimensional analysis to see if it makes any sense at all. (So much utter garbage in the air these days.) Then I test it with known (or expected) results. If they differ, I start looking at why. If they don't, I'll be surprised and very suspicious.

YES!  Dimensional analysis is such a useful sanity check.  And it's so often /right/, despite being completely unreasonable (e.g., slapping together units arbitrarily, then constructing a possible situation where those units would be related, and what constants, and what physics, would come into play in that case).

Tim

[Nominal Animal]

Have you checked uModules and such?

I didn't know of them until now.

Remember, I am at the hobbyist level, with very little practical knowledge.  So, if I need a DC-DC converter, I go to Digikey, Mouser, my local component sellers, and do a search on each. (When I know a component or type details, I can do a search on eBay, but I'm well aware of the prevalence of fakes and substrandard clones.) I do a web search for the problem at hand, and trawl discussions here, and project logs e.g. on Hackaday, to find practical solutions.  I don't usually do an appnote trawl then, because I'd be too easily distracted; they're actually pretty hard to search for when you have a particular problem at hand.

These things are biased with an electric field while hot, then cooled below Tc, "freezing in" the field -- making an electret.

Yes; it's the electrostatic equivalent of a magnet.  I was trying to say that even when the behaviour of something looks mathematically simple, it tends to be rather easy to use those in very complex ways.  This means that while "rules of thumb" are useful, they never capture all of the behaviour, and often there is a completely unintuitive (from the "rule of thumb" point of view) way to use the same material or phenomena. You can see this in some clever circuits, where jellybean components are used for odd purposes. Prince Rupert's drops is another good example. The way they're made causes huge opposing internal stresses; i.e lots of potential energy stored in the atomic structure. The bulb end of the glass drop can withstand a blow from a hammer, because the internal stresses are even greater, and easily withstand the additional stress of a hammer blow. But a small ping on the tail end creates a wavefront of collapsing opposing forces, and the entire thing explodes.

One of the very first simulations I wrote, modeled Morse copper. (It is a very simple model of the interactions between copper atoms.) I had a a block of atoms in a perfect lattice, and another as a liquid (simulated to move the atoms to random locations, but keeping the total energy constant). Unfortunately, I miscalculated their location by about one interatomic distance, so the blocks basically intersected. This means some atoms were way too close to each other, corresponding to either insane temperatures or ion bombardment. To my amazement, after about ten thousand simulation steps, the simulation stabilized with a drop of molten Morse copper inside a solid lattice, with the entire solid lattice vibrating at wavelengths much longer than any dimension in the system, and was stable like that for tens of thousands of time steps. That was not what one would expect from such a simple model, and it took me quite a while to find out exactly what and why it happened.

To me, it felt like the first time I wrote a program to draw the Mandelbrot set on screen. (Or rather, colored the outside of the Mandelbrot set depending on the number of iterations.) It is just a dozen or so lines in any programming language with floating-point number support and a putpixel() function, but the images are mesmerizing.

Dimensional analysis is such a useful sanity check.

So very true.

I think it was a high school physics teacher, who showed that keeping units with the quantities when applying some formula, ensures you use the correct units. For SI units, this is especially useful, because all units can be expressed in base SI units, and there are very few numeric constants to remember or look up. I've caught my own errors countless times that way. It even helps locate the errors. When you calculate a velocity, and get a result with units [kg m/s], you know you made a booboo somewhere; probably missed a mass term.

Dimensional analysis sounds fancy and advanced, but at the core, it is about dropping the quantities, and applying the formula to the units only, to see what units pops out as the result.

It isn't that useful in electronics, unless you add conservation of energy. The SI unit for energy is joule, [J] = [W s] = [V A s] (where W is watts, s is seconds, V is volts, and A is amperes). So, when a component drops 1 V at 1 A current for one second, it spends 1 J of energy. Some of that will do something useful, but because no component is perfect, at least a fraction of that energy will be converted to heat. One watt-hour of energy is 1 W h = 1 W · 60 · 60 s = 3600 J, so one kilowatt hour is 3.6 million joules.

When you examine the system at a point in time, instead of energy you examine power, usually in watts [W] = [J / s] = [V A] -- but note that the voltage and current units here refer to that particular instance in time, and it does not hold for example for average or RMS voltage and current (see e.g apparent power vs. true power for the difference).

Many of the debunking videos Dave has posted are based on this simple approach. You take the input voltage and current, or power, or available energy; estimate the efficiencies of the components (they're pretty well known, and you can always do min/max estimates based on known technology limitations); and you'll find out what is possible and what is not.

This is so basic, simple, and utterly robust way to examine devices, ideas, and formulae, that I think it is a crime to nowadays not teach this in high schools.

[T3sl4co1l]

Hmm, fascinating

Tim

[David Hess]

I do wonder why there are no 5VDC (USB; say 4V to 15V input) non-isolated SEPIC modules providing filtered, low-ripple 5VDC and/or 3.3VDC. You could even add some diodes, so one could power the project from USB or a separate power supply. A lower ripple (especially at lower frequencies) would help a lot with all kinds of sensors.

There are some switching regulators which support voltage and current slew rate limiting for low noise applications.

My own field is computational materials physics, so that is basically second nature to me now. Whenever I find a new formula describing some interesting phenomena, I do a quick dimensional analysis to see if it makes any sense at all. (So much utter garbage in the air these days.) Then I test it with known (or expected) results. If they differ, I start looking at why. If they don't, I'll be surprised and very suspicious.

YES!  Dimensional analysis is such a useful sanity check.  And it's so often /right/, despite being completely unreasonable (e.g., slapping together units arbitrarily, then constructing a possible situation where those units would be related, and what constants, and what physics, would come into play in that case).

In different schools, I alternately learned this as factor label and dimensional analysis.

Power, bypassing, lack of filtering on signal wires, lack of ground plane or stitching, cables just fricking everywhere... take your pick.  When one don't understand what a reference plane is, one is sure to make a dog's breakfast...