-
I'd like to make a battery-powered linear power supply. In my current design (harrisson-like, see [1]), I need a floating point aux supply. What are my options? It will power opamps, reference, ADC and DAC. Something like 12V->12V or 12V->15V would be great. I'd prefer an integrated module that I can just "plug and pray".
I'm not terribly concerned about efficiency, the power supply is not meant to be used for extended periods of time. But noise is another matter.
References:
- [1] http://hparchive.com/Journals/HPJ-1962-07.pdf -
How about an isolated 12V to 15V/17V DC/DC converter followed by a linear low-drop low-noise regulator to 12V/15V?
The same applies if you want a dual supply; many such converters you can follow up with a linear regulator.
To get the noise even lower, you can use either a capacitive multiplier or a Pi (C-L-C) filter between the converter and the regulator. -
I'd like to make a battery-powered linear power supply. In my current design (harrisson-like, see [1]), I need a floating point aux supply. What are my options?
If you make battery-powered linear power supply than it is already floating. -
Agreed, either there's something you didn't tell us, or this is a non-issue, as you shouldn't need to isolate anything if it's strictly battery-powered.
But, again, there may be some missing requirement, like use cases where it could still be connected to a charger while operating, or another power source than a battery.
Now, maybe the "isolated" part is a byproduct of the quest for finding a DC/DC converter solution?
If you want a "ready-made" module kind of thing, then I'd suggest picking any DC/DC converter that fits your requirements in terms of in/out voltage and max power rating, and add proper filtering (as usually suggested in the DC/DC converters' datasheet, often something like pi filters at both the input & output), and yes, then followed by linear regulator as suggested by Nominal. Downside then is that often, those ready-made, encapsulated modules are not adjustable and you'd need to select an output higher than the final output voltage due to the linear reg dropout. So that may be the annoying part if you look for "standard" output voltages such as 12V or 15V, but some of these modules are adjustable.
If noise is the primary concern and efficiency, not so much, my second suggestion would be to forget about DC/DC conversion altogether, and use a multicell Li-ion battery with the proper nominal output voltage, followed by a linear regulator down to your target voltage. For instance, for 12V output, you could choose a 4S or 5S Li-ion battery and then a linear reg down to 12V. That won't be super efficient, although with a 4S (nominal output 14.8V), you would have a "nominal" efficiency of 12/14.8 = 81% which you would hardly beat using a DC/DC converter followed by a LDO anyway. Just my 2 cents.
-
What sort of power levels are you thinking about for the whole power supply?
I am seriously considering building something similar myself, but starting from a 24Vdc power supply, for a starter, and later mayby higher voltage and current. but my first attept is modest, If I can get a neat 20V 500mA out of it I'm happy for the first try.
Because auxillary mains transformers are also no option for me, I am probably going to use an isolated 24Vdc to 2x8V auxillary power supply.
With a 1Watt module (costing about EUR 3) you have about 2x 60mA, which is enough for a uC, some opamps and a bit of bias current for the power section, although with a discrete darlington or Sziklai, you can use the main power supply to set the (minimum) operating point of the main power transistors. I prefer a symmetric power supply, because I don't like messing with "rail to rail" opamps, which never quite deliver what they promise.
I know these small modules are not very clean. I am going to attempt to clean them up with RLC filters, probably a capacitor multiplier and proper design of the GND plane. And how clean does this power supply have to be? Opamps generally have a quite high PSRR, and the voltage reference needs very little power, which makes filtering easier if it's needed.
My first attempt is not going to be a designed PCB, but built on matrix board. For the SMPS module + filtering I'm going to go dead bug / Manhattan style. My main goal is not to design a power supply, but to experiment with ADC and DAC techniques from a uC, and to gain more insight in analog circuit design. So I'm don't know yet what I'm going to build exactly yet, and it's likely to go though a bunch of iterations. All good reasons for going to matrix board.
Another option for you: What about using separate batteries for the auxiliary power?
And also: Realize that your auxiliary power supply does not have to be floating. If it's floating it can't deliver base current for the power section. Usually it's in series with the main power supply. The GND of the auxiliary power supply often is riding on top of the main power supply, If you want a +/- voltage for the aux supply, you can steal some power from the main power supply.
Also, how is your Dutch? (Firefox has a built in (experimental?) translator these days, and in general, translating is becoming easier over time anyway. About 10 years ago "Blackdog" on the forum: https://www.circuitsonline.net/forum/view/110029 has given his own swing on this harrison design. He had been working on details for about 3 years (form 2013 to 2016) on his design, and it started a few years earlier by improving the power supply he had (ELV-22532) There are a lot of interesting design details on that dutch forum, but there is also a lot of crud and distractions. It's inherent to how the 'net works I guess. If you want to skip all the Dutch, You can find his final version (42 of course) of the schematic on:
https://www.bramcam.nl/NA/NA-01-PSU/NA-PSU-SCH-42.png
Or, if that link gets broken someday, a direct upload attached below.
The older thread about the ELV power supply is:
https://www.circuitsonline.net/forum/view/109400
-
Why I need isolation in a battery powered power supply: the harrisson design needs to power rails: one "main", and another one "floating" on positive output.
How about an isolated 12V to 15V/17V DC/DC converter followed by a linear low-drop low-noise regulator to 12V/15V?
I don't have much experience with SMPS, so not sure if that's will be sufficient. That's why I wanted a ready-module that would optimized by professionals.What sort of power levels are you thinking about for the whole power supply?
Very modest, 12-15V, 200mA output.The older thread about the ELV power supply is:
https://www.circuitsonline.net/forum/view/109400
Yess, that's the one I use for inspiration).
PS Mijn Nederlands is niet goed) -
Small pulse transformers with 12:12 and 12:15 ratios are available which can be used with sine wave drive to make very low noise isolated outputs.
-
How much power is it switching, buck or boost, and what is your noise budget?
Edit: oh, and what is the concern: radiated or conducted noise? Those can both be significantly mitigated with minimal effort by using various standard techniques - like inductor orientation. Do those, then see where that puts you.
-
How much power is it switching, buck or boost, and what is your noise budget?
Edit: oh, and what is the concern: radiated or conducted noise? Those can both be significantly mitigated with minimal effort by using various standard techniques - like inductor orientation. Do those, then see where that puts you.
That's the thing, I don't know(. Let's say, I'm looking for less than 1mV p-p noise output. May be higher is ok if there is no "high-speed" content in it (which is quite likely the case).
As of radiated vs conducted, I'd love to not have any of them.
I was hoping to find something like LTM8080: https://www.analog.com/media/en/technical-documentation/data-sheets/ltm8080.pdf .
There are plenty of integrated modules from TI (https://www.ti.com/power-management/dcdc-power-modules/isolated-power-modules/products.html). But TI doesn't provide noise specs, not just spectrum plot missing, any numbers related to noise are missing in DS for a few modules I checked :/ -
It's elementary. Use converters with any noise, just turn them off for the duration of measurements, powered by charged capacitors.
-
The Harrison topology does not need a high voltage to power the controller part. Chances are some +-5 V would be enough and not much power needed either. So one could use a chip like LT3999 to drive a small transformer. If low noise is more important than efficiency an audio amplifier like LM4991 could be an option too. For a compact ciruit consider a ready made DCDC converter (e.g. AW) or SN6501 hard switching driver.
The main point of the Harrison topology is having the output stage as a controlable current source. One can get this also without an extra floating supply. For battery supply or starting with a single voltage this may be the easier option. -
If you’re not concerned about efficiency, you can deliberately slow down the turn on/off of the main power Mosfets.
Another trick is a resonant push pull, with a cap across the primary. You then trim the switching frequency to obtain maximum amplitude. (Resonance). -
Small update, I did a bit more research on the topic. Particularly I read "the art of electronics, x-chapters", there is a discussion on the topic. I'm a bit disappointed that it takes so much effort to make a clean SMPS. Like, everything matters, every tiny-tiny detail, like layout, slew rate, component selection, etc. Since designing a quiet SMPS is a big topic on its own, I decided not to push it this way and try to build my power supply without the need for an isolated power rail.
Having said that, for the sake of science, I'm building an isolated power supply on the breadboard just to verify that it's difficuilt). So, I'm building something like on the image attached. Unfortunately, I don't have ideal parts to build this, I'll just use the parts I have in my "trash bin".
One thing that already caught me off-guard is how much slew rate I need from opamp. I wanted to use LM358 because of its slow edges, but, turned out, its slew rate is not enough to drive even 5KHz 15-30V p-p trapesoidal signal ("power bandwidth").
-
Why not use a 555 timer for your oscillator? Surely you have one in your junk parts pin!
You can also make an oscillator from a schmitt trigger CMOS inverter 74HC14. -
I was just trying to get a slow edge. Otherwise 555 is good).
So, I get my trusty MC33072 opamp produce a trapezoidal signal. It was nice and clean, and I could get some 3us rise time.
But then I had a brillian idea that I can simplify circuit and use a power opamp to generate signal directly. So, I picked UTC2003 which I got from LCSC for like 20 eurocents and I got... nothing. Scratching my head, I opened the schematic, and it doesn't look like an opamp to me (where's long-tail pair?). So, I replaced it with LM1875 from HGSEMI which costs whooping 60cents... Pricey upgrade I would say).
And I worked. Kinda. I have horrible oscillations which I can't get rid of. Decoupling doesn't help. The opamp is presumably requires minimum gain of 10, but I thought in oscillator mode this is irrelevant. Anyway, it badly oscillates and I don't know why and how to fix this (. Randomly placing capacitors didn't change anything.
May be the issue is that I only place 100nF ceramic caps, but this is a beefy IC, and it needs some bulk electrolytic caps for proper functioning... I'll try to add those too.
-
Why not use a 555 timer for your oscillator? Surely you have one in your junk parts pin!
The NE555 is quite noisy, with a relativey large current spike at the supplies, even though it is BJT based.
You can also make an oscillator from a schmitt trigger CMOS inverter 74HC14.
In a different thread about powering the VFD filament someone suggested a LM4871 audio amplifier that could be suitable and also provide some power up to maybe 2 W. -
The cleanest DC/DC you may get with harmonic signals only, imho. A colleague of mine (talking mid 80ties) made, as a part of his thesis, a DC/DC converter for a 1/f noise measurement system (afaik a higher voltage one). The idea was the filtering will be much easier, no high energy spectral mess generating by any switcher..
Also its frequency components would be better defined considering the spectral analysis measurement.. He spent most time manufacturing the transformer
as I can remember, he was running it at 10+ kHz (we measured the noise up to 10kHz), driven by a low distortion "power" amplifier (for a well defined sine signal).
-
The NE555 is quite noisy, with a relativey large current spike at the supplies, even though it is BJT based.
In a different thread about powering the VFD filament someone suggested a LM4871 audio amplifier that could be suitable and also provide some power up to maybe 2 W.
Yeah, original NE555 is noisy. I have much better results with a CMOS version of the chip.
LM4871 is a nice suggestion, adding to my shopping cart. I like that it is unity-gain stable and that it seems it provides symmetrical output drive.The cleanest DC/DC you may get with harmonic signals only, imho.
I agree. I was just thinking that by controlling slew rate I could simplify circuit a little bit. According to X-Chapters, they found that in their isolated power supply, there is not much difference between a clean sine wave and a triangle wave. They also suggested that if triangle waveform works fine, then trapezoidal signal should work fine too (with obvious advantage of higher RMS voltage, afaik). That's why I'm experimenting with trapezoidal signal. -
You can try an open loop LLC converter.
Edit:
-You can increase the separation between prim and sec. winding without any problem. The leakage inductance can be used for the LC part.
-Efficiency in this case is not a concern. so you can use small resistors in series with the sec. rectifiers.
Disadvantage is that for low power you will need to reduce the magnetizing inductance such that the converter always stays in ZVS. A core with an air gap will be required. -
If the main goal is to have no noise, then you can cascade thermoelectric elements and heat them.
-
For low power one-off lab solution, LED array module driven by good linear CC and optically coupled to a solar cell is both isolated and clean. At a bit bigger distance the capacitance is also very low. It's also funny to show to people, cause it's a bit ridiculous.
-
For very low power there are even ready made PV opto-couplers. There can deliver some 10-50 µA at some 7 V. That is enough to power some OP-amps if really needed, though the main intended use is to drive a MOSFET gate.
The efficiency is not great (e.g. ~ 3% for a TLP3906), but not as bad as one may think as IR LEDs are quite efficient and PV cells have there best efficiency at around this wavelength. -
For very low power there are even ready made PV opto-couplers. There can deliver some 10-50 µA at some 7 V. That is enough to power some OP-amps if really needed, though the main intended use is to drive a MOSFET gate.
The efficiency is not great (e.g. ~ 3% for a TLP3906), but not as bad as one may think as IR LEDs are quite efficient and PV cells have there best efficiency at around this wavelength.
Yeah, I once used VO1263 in a design (both channels in parallel), works well if only small current is needed. -
LM4871 is a nice suggestion, adding to my shopping cart. I like that it is unity-gain stable and that it seems it provides symmetrical output drive.
Whoa hold on.....
I thought you were just using parts available in your junk bin?
If you're just going to purchase low noise parts from online retailers like Digikey, then that opens up an entire set of possibilities! -
> that opens up an entire set of possibilities!
You're right, if do so, I'll never finish anything)
I considered photovoltaic cells, but that's too much, imo). I need a big array to have any meaningful current. At such efficiency peltier elements or, maybe, piezo-effect something doesn't sound too bad)
Anyway, I did the experiment. I just assembled the whole thing without much thinking and tuning. This is what I get: p-p noise around 4mV (~5MHz BW, probe is 1x, oscilloscope set to 20MHz BW limit). Baseline noise seems to be around 800uV p-p when everything around is powered off and probe shortened (but with a long ground lead).
Setup: Trafo is some random audio trafo (EI14-600:600). Thenschottkydiode bridge (DBL206G), caps 100n+low-esr electolytics. Final smoothing is AMS1117-3.3. Load is 330 ohm resistor (10mA load).
Current problems:
- trafo (EI14-600:600) has resistance of ~120 ohm each winding.
- diode bridge (DBL206G) is not schottky
- opamp wildly oscillates with base freq about 1MHz, but there are some high-freq components because it's a saw-tooth oscillation.
- frequency seems to be 35KHz instead of 5KHz as I wanted. I messed up while messing up the opamp.
- voltage headroom is minimal. Just on trafo it seems voltage drop is whooping 20V p-p!
I consider this a mild success. I might do a bit more experiments, not sure. I think, if driven slowly and gentle, sub-1mV noise is achievable without much trouble. But trafo size will be significant.