Splitting the Rail Versus Pos/Neg Supply

I'm designing an isolated current feedback chain for a digitally controlled amplifier. This circuit is powered from an isolated 12V to 12V 1W dc-dc.

My original concept used an LTC3260 charge pump + dual +/- regulator to generate linear +5V/-5V supplies from the single 12V. Unfortunately it looks like the 50mA available from that chip isn't going to cut it.



The alternative I'm exploring here is to generate a regulated 10V and split it, using a buffered 5V midpoint as the reference for the series of differential amplifier stages that are powered from the 10V. This seems reasonable and not terribly unusual, TI makes the TLE2426 specially designed for this (though I don't see a need to use it).

But am I missing any pros and cons here? One detail is the reference to the outside world (the shunt). This whole circuit is isolated so my choices are to tie the shunt to GND or to the midpoint. Choosing the 5V midpoint centers the shunt at the center of the first amplifier's common mode range, as it would be if I had split supplies, which seems advantageous.

For the midpoint amplifier I'm thinking I don't need precision resistors setting the midpoint, small drift would be insignificant, nor do I need a particularly high performance amp. I'm showing it with large caps on the midpoint so I'm not relying on amplifier bandwidth alone to regulate it.



The rest of the picture is that the output goes to a 5V differential ADC with a 4.096 reference. Hence the final differential stage. That ADC will have its own 5V regulator from the 12V. I'm using the NXJ1 series 1W isolated supply from murata because of its size.

Dual output 1W isolated supplies exist of course, but real life availability points to either +/-5V, which doesn't leave room for a linear regulator, or +/-12V which cuts available current to ~40mA on each rail.

Hi,

There are a couple of things I´d change:

* C10 acts as a low pass filter in combinaton with R18, R16. But as soon as you add C9 you don´t have a low pass anymore. Any noise on the 10V line comes to the 5V node (as with no capacitor).

* U1, the second stage is about useless. You could achieve the same with modified values around U5 and/or U2.

* all the power supply capacitors are to small in value for my taste. I´d use at least 1uF ceramics.

* U3 output current may be unpredictable. Therefore I find the value of R1 too high.

* instead of generating 10V/2 I´d use the 4.096 V reference voltage of the ADC. It simplifies the circuit, and all is referenced to this voltage. Especially U2 has relaxed signal conditions, because it doesn´t need to regulate AC difference voltages between 10V/2 and ADC.

* I don´t see the need for an extra regulator for the ADC. Use the same supply but RC decoupled as you do with the OPAMP supplies (nice!).

***
You don´t give informations about signal frequency, signal range, ADC sample rate and ADC resolution.

Klaus

Added comments:
I just reviewed the OPAMP datasheets.
The LT1013 is 1MHz GBW. With a gain of 5 expect 200kHz of signal bandwidth.
But the LTC6362 is a 180MHz OPAMP with gain of 1 resulting in 180MHz bandwidth.

They differ by a factor of about 1000. This is not useful.
If you don´t need the input frequency bandwidth, then you could omit the difference amplifier (U5) totally. Resulting in an even more simple design.

I like a lot the TLE2426 precision rail splitter. It simplifies the solution significantly.
But it is an acquired taste and a little expensive.

Here's another approach that uses two regulators from an unregulated single supply.

Thanks for the thorough critique. There are a few things I didn't show or say. First, the overall picture is that I'm trying to deliver about 750Khz bandwidth to a 5mhz adc which goes back to an FPGA which implements a control loop. The 750khz is reasonably below the ADC sampling frequency and far enough above my planned ~10khz digital control loop bandwidth that the analog stages should contribute minimal phase delay.

Also needing good DC performance, the LT1013's shown are actually OPA727's which are the best compromise I arrived at for DC/AC performance/cost (0.3uV/C typ drift, low bias currents and 22MHz). With this bandwidth the extra stage may not be necessary (and will have zero ohm bypass options), but I wanted design margin generally and wanted the first stage 750khz filter to dominate without the following stages encroaching on the overall transfer function (which they do unless they're significantly higher).

Not shown is a middle stage which uses a quad analog switch to implement X1/X2/X5 gain options by switching in different feedback resistor packs (I compensated the X1 and X2 stages so their bandwidth matches the X5). The need for this stage, which is easier to implement with single ended amplifiers, is why my first stages are single ended at all rather than using more LT6362's.

Klauss:
The LT6362 is compensated unusually. Note that at a gain of 1 it really only has 35mhz of bandwidth which is more in line with the OPA727's. This amplifier and ADC is actually on another board which is why its conceptually simple to use it as a 1:1 buffer and leave the gain settings to my more application specific front ends. Though I realize I'm wasting its bandwidth by not stuffing some gain in there.


*Ok so back to the splitter, the C3/C10 combo sort of makes it the best 'splitter' it can be. I agree that I'm not sure if that's what I want or not.

*Ok I may go 1uF, it may depend on package size.

*Agree R1 may be high

*I dont' see the benefit of using the 4.096. I'd want to buffer it so I wouldn't save an amplifier and it's not really relevant to the first stages. The point of the mid-scale is to center things within the 10V rails which is good for the amplifier input stage and maximizes output swing (planned to be 8V peak-to-peak).

*I think I was unclear. The +5V the last stage is on will be the same as the ADC supply. I was making it clear that I'm not using the mid-scale 'split' rail, which is also 5V typically, as the ADC supply which would in theory be possible.


Schmitt:
Yeah that's what I thought about it and its large. And why would I need precision here? Precision for the mid-scale is no more important than precision for the power rails normally.

I also found the BUF602 which comes in SOT-23 which seems like a decent alternative.



Crutschow:
That's interesting. I'll look into that further.

crutschow said:

Here's another approach that uses two regulators from an unregulated single supply.

Click to expand...

Cool, I didn't really know this existed. A few other manufacturers make them too.

https://media.digikey.com/pdf/Data Sheets/Semtech PDFs/SC1118.pdf

In theory I could use this as the 5V mid-point reference and the supply for the ADC and final buffer. Still not sure if I'd want to do that or not.

Hi.

Yes, I think the voltage regulator - that generates a clean voltage related to it's GND - is better than the voltage divider solution.

Klaus

Unfortunately that regulator is a rare LT part with no industrial rating and the Semtech version isn't as good and isn't really available. So I'm still on the fence here with just an opamp buffer still for now.


Klauss, a while back we were talking about resistor packs and temperature coefficients and you said resistors from the same batch would have near zero difference in tempco.

I'm looking at this again and part of me wants to accept that but why does no one spec it. Take the ACAS series resistor pack which is advertised as TCR matched. They spec only 25ppm overall/15ppm match. If they could go lower on that match spec, even just for the same value packs, I feel certain they would.

Glancing at non-matched packs I don't see any language implying the match is better than the overall.


I had the idea of placing cheap packs in areas where low PPM dividers would be nice but not critical, but what's worse is that some of the cheap/small packs are only 5% overall with no match guarantee. It doesn't seem like I can go that way.

Hi,

but why does no one spec it

Click to expand...

If I were a manufacturer, then I wouldn´t spec it, too.

The problem is: on thos simple resistors there is no batch number.
So if a customer complains about bad matching ... the manufacturer has a problem.

If I want a low drift voltage divider with 25% output voltage, then
*I´d prefer a 1% 10k-10k-10k-10k combination (all from same batch)
* against at 0.1% 30k-10k combination. (they can not be from same batch, because they are different values)

Fur sure it is not guaranteed.. so it´s my own risk if it fails.....

Klaus