Power supply for voltage references

[exe]

Could it be because most of the noise comes from dead time? We'll see, I'm building a small proto board to play with. I'll try different parts with different slew rate.

Question: does transformer's V-t product depends on the load? I mean, can I violate it if there is always a load? My logic is that some energy will go to the output, so less chances to saturate the core.

[Kleinstein]

The required V-t product does not change much with loading. Essentially all of the output current will come from the input side as additional input side current. It is only the small voltage drop at the swich and winding resistance that reduces the effective voltage.
One can reduce the input side voltage if absolutely needed.

[DeltaSigmaD]

Power supply with low leakage with LT1683

I'm missing the LT1683 in the mentioned list of push-pull controllers. I developped a power supply with this LT1683 since it has a slew-rate control feature to minimise noise. My application is a supply for external test circuits, e.g. a controlled voltage reference, which are working inside a current acquisition system operating in the pA to uA range with fA resolution. The entity "leakage current" (LC) must be defined: this is not the differential noise generated by the power supply (which may be actually very low), but the current flowing through the ground connector (not protection ground) of the power supply when it is connected with the protection ground of the power line system. Many manufacturers do not specify this LC. A good value for LC is less than 1 uArms (broadband), but even this is much too high when you want to measure fA (current between Gnd of supply to Gnd of acquisition system).

I tried a design with the double ferrite core topology (coupled by a short-circuit winding) as dc/dc transformer to reduce the coupling capacitance and hence the LC. It was hard to find a useful loop compensation. The slew-rate control of the LT1683 is working insufficiently with this transformer type. Therefore, the switching noise spectrum is unexpectedly high which can't be accepted. I had to reject this design.

A dual core transformer has a considerable stray inductance, so that many push-pull controller ICs will have problems with this topology. In general, it is difficult to suppress the noise of push-pull controller ICs : the duty cycle and the switching frequency are not stable enough to attenuate the noise by digital filtering performed by the data acquisition. Any nonlinearities mix varying switching noise down to the base band. Spread spectrum is useless for these applications.

Therefore, I'm just developping a power supply on base of a microcontroller. The focus is directed on low leakage by a dual-core transformer and constant frequency by a quartz. The duty cycle can be fixed to obtain a stationary noise spectrum, taking a worse power efficiency into account. I will report here the results, hopefully in a few weeks.

[trobbins]

Another option to suppress coupling capacitance between input source and destination supply is to use multiple small commercial dc/dc, such as from vintage Newport, but now C&D etc.  For example a 5V to 5V dc/dc NMA0505S has 100mA output capability with 28pF coupling.  Two to three of those in series, along with an isolated plug-pack, should suppress coupling to <10pF in a convenient, relatively cheap, off-the-shelf, space-saving manner.

That style of isolation has many model variations to allow different input and output voltage rails, and power rating, to suit an application, and some have lower coupling cap.  The final dc/dc in the chain may benefit from some L-C filtering to suppress residual hf ripple, and a final linear regulator IC.  I haven't personally tried this to confirm no anomalous behavior from interaction between converters.

[Kleinstein]

The LT1683 is not made to directly drive a transformer, but for external MOSFET switches and thus for higher power. So it would be an odd choice for a low power solution. The extra area and traces alone can be an issue with radiated EMI.
The weak point with the push pull converters is the fast switching and related EMI issues at rather high frequencies (e.g. ~ 100 MHz in the SN6505 snubber appl. note). This part can be hard to predict and depend on resonances in the layout / wireing.

For the capacitive coupling it is not only the capacitance that matter, but also the effective voltage at the transformer. Here symmetry can help and shield in the transformer can help with symmetry. An easy point to keep the voltage low is to use low voltages at the transformer. So more a 5 V to 5 V transformer and if needed a charge pump for higher output voltages. The overall capacitance to ground also has the case and cable capacitance for the outputs. The transformer is only a part of it.

[dietert1]

Yesterday i looked at a Wien oscillator with a TL082 that i made some years ago. I removed the jfet gain control, so it produces a clipped sine wave of about +/- 12 V from a +/- 15 V supply. Second opamp is used as inverter. Adjusted frequency to 25 KHz.
At the time i wired it with four larger bipolar transistors as output buffers, but i think the opamp might handle the supply of a reference all by itself. At the time i hand made a small toroid 1:1 transformer with 2.2 mH coil inductance, coupling of 2.5 pF and a shield. It's impedance is low for use without the buffers.
Then i found two pieces of VAC 4097X011 transformer in a drawer. Those are 1:1:4 transformers, roughly the size as the WE part mentioned above by Kleinstein, except the "4" coil has 200 mH inductance and 2 Ohm resistance. That appears to be a good solution. Coupling capacitance is 25 pF. A RRO opamp with good drive capability can improve the circuit.
Looking to buy that transformer, i couldn't find an offer. Yet i found another interesting part: VAC 4099X011. That is a 1:1:1 transformer with 0.95 mH per coil and a low 2.5 pF coupling spec.
Another useful part may be the black Schaffner 100 mH common mode choke. I measured a coupling of 21 pF. Two of those can be used: as transformer and as common mode filter.

Regards, Dieter

[Kleinstein]

The idea with a clipped sine looks OK. As a driver I would consider a small audio amplifier like LM4871 (5 V brigde for 3 W). Full bridge drive would ideally keep the average votlage at the coil constant, which could help with the effective voltage for coupling.

For the waveform one anyway has to compromise between the voltage waveform and current waveform. A sine voltage tends to cause pulsed current with a simple rectifier.

[David Hess]

Then i found two pieces of VAC 4097X011 transformer in a drawer. Those are 1:1:4 transformers, roughly the size as the WE part mentioned above by Kleinstein, except the "4" coil has 200 mH inductance and 2 Ohm resistance. That appears to be a good solution. Coupling capacitance is 25 pF. A RRO opamp with good drive capability can improve the circuit.
Looking to buy that transformer, i couldn't find an offer. Yet i found another interesting part: VAC 4099X011. That is a 1:1:1 transformer with 0.95 mH per coil and a low 2.5 pF coupling spec.

Also look for gate drive transformers.  They have about the right characteristics.

[exe]

So I built the push-pull circuit. It supposed to boost voltage from 3.7 to some 21V. I think I made a mistake somewhere, as output voltage (unloaded) is 26V at only 3.1V input. Trying to increase the voltage I got current quickly increasing. I guess either diodes give up, or I overload the transformer's VT. I might miscalculated something. Circuit main components:
- trafo: wurth 750317072 https://www.we-online.com/components/products/datasheet/750317072.pdf
- driver: NXF6505B https://assets.nexperia.com/documents/data-sheet/NXF6505_Q100.pdf

I think it's a moderate success, given the fact that trafo has quite some long leads inside the enclosure, and there is no snubber yet.
But I'm struggling to measure the noise, because I got 6-7mV p-p noise (full scope BW, >100MHz), regardless of leads connected or not (measured at the output, after LC-filter).

If I use 1X probe, and limit BW to 1MHz, I got 220-300uV p-p noise, according to my scope, regardless the probe is connected or not. Go figure 

Question: is it normal in trafo I got 4X the input voltage? If you look at the input, you'll see 12V (plus ringing) p-p, but supply is only 3.1. I kinda expected it to be 2X 

[Kleinstein]

If one used the full winding at the secondary 4 x the input voltage is normal for the peak to peak value. Half the winding at the primary side see both a positive and negative voltage. So 2 x the input voltage as peak to peak for the primary (half winding) and thus 4 x for the full secondary.

The unloaded output voltage can be higher from the ringing part on top.

[exe]

I put 220 ohm resistor on output, got voltage dropped to ~16V (72mA output current). Input current is 0.5A.

The output noise because 36mv p-p (full bandwidth)  . I didn't investigate further, but spikes are ~1us apart, roughly matching switching frequency (~450kHz). Also the drop is much more severe than I expected, I lost 10V on the output.

[EC8010]

I've just looked at the PCB photograph in post #183. This is to be a clean supply for a voltage reference, right? It won't be with all that flux left on the PCB. Trust me; deflux properly and you will see circuits work better. "Properly" means holding the board vertical and scrubbing with defluxer from top to bottom to release contamination, shaking as much liquid off as possible then immediately rinsing and scrubbing from top to bottom with isopropyl alcohol to dilute and lose the contamination, and immediately drying with a pre-warmed hair dryer. Then do the other side. Then the first side again. Then the second side again. Not only does it make your soldering look much nicer, but it all works better, and with low noise or low current stuff, that's important.

[dietert1]

I see only one buffer cap close to the driver IC while the original schematic in post #175 had two of them, maybe 100 nF and 10 uF. Also it appears like the two primary side snubbers vanished.
Did the driver IC change from ..A (160 KHz) to ..B (420 KHz) due to transformer saturation?

Regards, Dieter

Edit: I think the 16 V is roughly what one expects, as the turns ratio of that WE 75037072 is 6.17. So if you start with 3.1 V you end up with 19 V minus driver saturation and diode voltage loss. Note that the primary inductance is 200 uH but in your circuit that divides by 4 as you drive each half separately.

[Kleinstein]

The higher frequency of the B version is kind of needed to avoid saturation with that transformer. The scope confirms the higher frequency.
An asymetric load on the output side (one side rectification) can also drive the core slightly into saturation and this way increase (maybe double) the current needed. In the extreme case diode self heating may shift a rectifier to asymetric operation.