Power supply for voltage references

[miro123]

 Hello,
I have start experimenting with few nanocrystaliine cores and i compare it to old low/MF frequency ferrite material 3C90. There are much better martials than 3c90 nowadays.
Reading the parameters on simple rlc meters look ok, Thinks get worse when i start using Huoky and HP RLC meters  at 1k,10k,10k,1MHz.
Before i making final decision I want to ask you guys does somebody look at impedance of the transformer? - Magnetizing /leakage inductance vs frequency and other parameters
Do you have datasheets of materials used on your design's?  I am interested on graph representing the complex magnetics permeability  real and imaginery part.

Look at the attached pictures.
My preliminary toughs are
- material is not suitable for LLC converter using UCC25800
- material is great for feritte beads or common mode chokes.
- maybe material is good for old fashion push pull - all harmonics get absorbed in core losses. See at core loses graph attached
- such material should mitigate the hard switching of push pull converters - anyway the parasitic capacitance should still play a role.
-my experience with 3c90 is that it match the specified up to 400KHz application area. The losses at 100...400khz range are acceptable and matched the specified complex ur.

[Overspeed]

Hello

Could be nice to open a subject in the forum as your need / question do not look to be linked with the main thread

Regards
OS

[gamalot]

This is the USB-powered ADR1399 prototype board I made in May 2024. The isolation boost chip is TI's SN6505.

[faraday]

There is company iFi they make low noise power supply, model  Audio iPower2 claim noise 1uV

[miro123]

This is the USB-powered ADR1399 prototype board I made in May 2024. The isolation boost chip is TI's SN6505.

Nice and neat design I fully agree that modern COTS solution works just fine or just beter than many handmade solutions. I have used SN6505 with wurth transformer like your solution in the past. I find the isolation performance excellent. The noise can be belter. I have one question what are the four pin devices at the bottom right corner. Devices that cross the isolation barrier.

[miro123]

Hello

Could be nice to open a subject in the forum as your need / question do not look to be linked with the main thread

Regards
OS

Maybe my description is too cryptic. The design of the transformer is a major task in building a power supply for voltage references. I was wondering how other builders measured or characterized the performance of their solutions

[Kleinstein]

The nonocrystaline and amorphous cores are a thing for relatively low frequency (e.g. 15 kHz). There they allow higher magnetization than ferrites. At the high frequenicy as used with the UCC28500 or SN6505 the eddy currents cause extra loss and the usable magnitization may already be limited by the loss, not by saturation. A nice point for the nanocrystalline cores is that they have low magnetostriction, so little acoustic noise even of used at 10 kHz.

The layout for the SN6505 output side looks good - kind of 4 wire contacts to the tantalum capacitor  and a small area for the first diodes and capacitor.
I would consider an extra common mode choke - it may help a little with the higher frequency CM part. The choke could be at the output or in the supply to the SN6505.

[gamalot]

This is the USB-powered ADR1399 prototype board I made in May 2024. The isolation boost chip is TI's SN6505.

...
 I have one question what are the four pin devices at the bottom right corner. Devices that cross the isolation barrier.

They are HCPL-181 photocouplers, used to control analog switches to select different output voltages.

The components that are not populated on the PCB are the nRF52805 MCU and its peripheral devices, including Bluetooth, temperature sensors, LEDs, and a push button.

[gamalot]

The nonocrystaline and amorphous cores are a thing for relatively low frequency (e.g. 15 kHz). There they allow higher magnetization than ferrites. At the high frequenicy as used with the UCC28500 or SN6505 the eddy currents cause extra loss and the usable magnitization may already be limited by the loss, not by saturation. A nice point for the nanocrystalline cores is that they have low magnetostriction, so little acoustic noise even of used at 10 kHz.

The layout for the SN6505 output side looks good - kind of 4 wire contacts to the tantalum capacitor  and a small area for the first diodes and capacitor.
I would consider an extra common mode choke - it may help a little with the higher frequency CM part. The choke could be at the output or in the supply to the SN6505.

I think I will try your suggestion if I attempt to make a new version.

[Kleinstein]

With only a single voltage, I would consider the CM choke on  the output side. The CM choke would than also be used to have the filter capacitors on both sides as an addition normal mode filtering. An additional MLCC costs very little.

The old PCB could still work OK. If needed there are clip on ferrites to go to the USB cable and also a few turns through a ferrite torroid could work.

[exe]

Which components would you guys recommend for LC filter on output of SN6505?

AN-70 recommends COILTRONICS CTX100-3, but it seems it's a transformer, not a single coil. Quite a big device. I wonder if I found a wrong device, or they do use a transformer and connect it as a T-filter, or they put two devices? I'm not sure.

[Andreas]

Hello,

I think they used 2 single devices. (otherwise naming would be L2a + L2b).
And it makes no sense to transform the current ripple from first to 2nd stage.
100 uH and iron powder makes sense for low frequencies around 100 kHz.

with best regards

Andreas

[Kleinstein]

The COILTRONICS CTX100-3 could be used as a common mode choke = current compensated inductors. However the inductance is still relatively low. How much CM inductance is wanted depends on the capacity to ground. Also the quality (symmetry in coupling capacitance, maybe shield windings) of the transformer used with the SN6505 makes a difference in how much common mode signal is generated. If more than a single voltage is generated, one could have the CM filter also on the primary side.

The rectifier and filtering for the normal mode ripple could also use some inductor. However this would be a more normal inductor. Here also the layout on the PCB is important - learned that from a poor layout. Instead of an inductor one could also use RC filtering, especially if the power is relatively low.

[dietert1]

Yes those small parts may not be enough. They are specified at 100 KHz, yet the switching spikes are at a much higher frequency, where the parasitic capacitance of an inductor gets more important. I remember some Marco Reps youtube video where he studied a transformer with much larger parts, with air distance between primary and secondary coils and spacers between coils and core. Each coil had only 2 or 3 turns.
Today i made some measurements on a 30 VA toroid mains transformer with shield between primary and secondary. Even when using a 100 mH common mode yoke on the primary side the shield injected high frequency noise of about 0.5 mA rms into the guard net. Capacitance between primary and shield is about 265 pF.
The special switcher transformer mentioned before has a coupling of less than 5 pF.

Regards, Dieter

[Kleinstein]

The coupling capacitance gets larger with a larger transformer.
An example for a ready made transformer for the SN6505A is here:
https://www.we-online.com/components/products/datasheet/750315240.pdf
They give 18 pF of coupling capacitance.
For the higher frequency SN6505B the transformer can be smaller and less capacitance is possible.
The exact construction can also make some difference, but here the effect is somewhat limited.

[dietert1]

The "low coupling" transformer has about 4 or 5 times less coupling. That difference doesn't matter, of course, if somebody adds another 10 pF or so by the PCB design.

Regards, Dieter

[EC8010]

Even when using a 100 mH common mode yoke on the primary side the shield injected high frequency noise of about 0.5 mA rms into the guard net. Capacitance between primary and shield is about 265 pF.

Yes, toroids have high interwinding capacitance compared to EI, so you should expect significant current to the shield. But that doesn't matter because the shield is (should be) connected directly to chassis. What matters is the capacitance between primary and secondary with the shield earthed. In general, a single crude E/S screen reduces C_PS to <10pF. But it is possible to do much better (as I described in post #137); I have achieved C_PS = 8fF on a mains transformer.

But reducing C_PS isn't the end of the story. You need significant capacitance from the secondary to chassis to form a potential divider. 4n7 from each end of the secondary directly to chassis works. Assuming C_PS = 10pF, you now have a 1000:1 potential divider to common mode interference, constant with frequency. If you add 470 ohm from each end of the secondary directly to chassis, you add a lot of low frequency attenuation, but depending on your rectification etc after the secondary, that might not be feasible.

[David Hess]

That SN6505 is slick.  Linear Technology had some parts which could be configured to do the same thing with voltage and current slew rate limiting, but they are not cheap.

I have been thinking of trying a resonate design to reduce noise, and tuning the frequency to match an LC notch on the secondary side.

Shown below is how Tektronix was doing it in the 1970s.

The inverter is operating at 40kHz.  The pot core transformer has special construction where the centertapped primary (left side red insulated wires) is led in through a fixture separate from the secondary leads, so I assume the primary and secondary are physically separated and electrostatically shielded from each other.  The transformer is suppose to support 1500 volts of isolation.

This particular one had some tantalum capacitor failures and I am messing with it.

[Kleinstein]

The SN6505 (alt least the A) version has one detail that can be anoying. To make it easier with EMI regulations it has some frequency modulatoin. The speed of that modulation is at some 50 Hz (a bit hard to measure) in combination with mains ripple at the input side this can cause low frequency modulation at the output. For a precision measurement the modulation with some 50 Hz can be an extra issue  If really needed one could feed it an external stable clock.

For the SN6505 the idea to reduce ringing after switching seems to be a rather short dead time. So switching is well faster than the self resonance in the transformer and avoid ringing by driving it hard. This does not work well with a resonant secondary.

[David Hess]

The SN6505 (alt least the A) version has one detail that can be anoying. To make it easier with EMI regulations it has some frequency modulatoin. The speed of that modulation is at some 50 Hz (a bit hard to measure) in combination with mains ripple at the input side this can cause low frequency modulation at the output. For a precision measurement the modulation with some 50 Hz can be an extra issue  If really needed one could feed it an external stable clock.

Doesn't the datasheet say that the frequency dithering is disabled when an external clock is used?

Spread spectrum clocking reduces the measured EMI because of how it is measured, but the amplitude is the same.

[Kleinstein]

The datasheet tells that the SN6505 used spread spectrum, but I have not found the typical modulation frequency. Having this close to the mains frequency has pros and cons.

[exe]

There are many similar push-pull ICs on the market. I'm gonna try NXP's equivalent of those. Anyway, my mini-research showed at least these devices. Prices are for single quantity, from my mouser account.

- max5075 (€ 12.94, just driver, no mosfets, no spread spectrum, programmable frequency)
- LT3439 (€11.5, slew rate control, wide supply voltage)
- LT1533 (€ 13.91, all-you-can-eat, 20-250kHz, slew rate control, 2.7-23V, etc)
- LT3999 (€ 5.32, no spread spectrum, wide supply voltage)
- MAX253 (obsolete? no spread spectrum, 200kHz or 350kHz)
- MAX485 (€ 4.73, no spread spectrum)
- MAX13253 (€ 8.04, spread spectrum, slew rate 200V/us, 250kHz or 600kHz, 5V)
- NXF6505 (€ 1.62, A and B versions, 5V, slew rate 57/155 V/us, spread spectrum, differ from corresponding TI parts)
- SN6505 (€ 2.14, A and B versions, 5V, 48/152 V/us, spread spectrum)

I like AD/LT parts for their versatility and voltage range, but NXP is cheaper...

[exe]

Btw, what do you guys think of a snubber?

I originally saw that in one of LT documents (don't remember which), and I found this: https://www.ti.com/lit/an/slla566a/slla566a.pdf .

[Kleinstein]

A snubber could make sense, but one could well get away without it. It would only be for the rather short break before make time. Other than that the input drive FETs and the output diodes would dampen any ringing effectively.

[dietert1]

Don't quite understand the slope ratings, e.g. 48 V/usec for the TI part. If the converter could run at 20 KHz, that is a 50 usec period time, can't we assign some microseconds to switching? That would be about 1 V/usec slope. So another factor 50 less EMI amplitude, or 34 dB. The snubber studied in the TI paper gets you 7 to 12 dB.
I think the TI part is designed for applications where efficiency matters. When it's about some 200 or 500 mW for a voltage reference one can do better.

Regards, Dieter