Rail Splitting Chips

You would think there would be one of these but I still take 2 equal value resistors and feed into a unity gain follower op-amp to supply half-Vcc. Usually one half of a dual op-amp.

This would be a good use for a SOT-23 package....

But since it is just for bias, could use a SOT-23 NPN follower biased up Vcc/2 + 0.65V or so.

boB

TI sells a dedicated rail splitter chip, the TLE2426. It costs more than a TCA0372 and doesn't have nearly the oomph. Plus the TCA0372 is a dual, so you can use the other half to do useful work too.

Cheers

Phil Hobbs

I've used TI DDR Vref drivers that create the half voltage for the DDR termination at high amps, see e.g.

Arie

The TCA0372 data sheet only mentions +-15 supplies, but it actually works at 5 volts between rails.

There's no power pad, but lots of V- pins on the 16-pin versions.

Would an LM386 in VSSOP-8 be too old-fashioned? It will naturally bias up to half the supply +/- not highly accurate or stable.

The NPN follower would require the same resistor I would use with an LDO. So that's not much better, drawing the current all the time. In fact, when the load is pulling up, both the resistor and the LDO draw current!

The op amp is reasonable, but it's bigger than what is possible. It just seems like something aught to be available. A zener is not bad, but a three terminal device is ideal. Power, ground and output, no passives. SC-70 package or smaller. This board is going to be seriously tight.

npn+pnp follower,

Both parts have the same problem, far too large. I barely have room for a pair of zeners. A SOT23 would suit, but better something smaller.

If I had a use for the op amp, I would have a spare op amp in the dual packages I'm using.

Interesting ! I would have possibly used one if it could split 3.3V for LV applications.

But, that might be a single-sourced part ? Haven't looked at the price.

boB

We pay 44 cents.

I don't care about "old fashioned". Everything we design is old fashioned, other than the new bits that we have to come up with.

I just want it to be very small. I can get an op amp in a small package, so that's likely what I'll do. But we'll see what turns up. Thanks.

Does your drawing need a couple of diodes? Otherwise the output would have a dead band 1.something volts wide, no?

It's easier to design with lots of power rails, like the old NIM +/- 24V, +/- 12V, +6, GND, and Vref all from the standard power supplies. Scaling to fast low-V logic was just a point-of-load regulator away, but nowadays... everyone wants a tiny box with a wallwart power solution.

The only solutions NOT discused here, are multioutput DC/DC converters (you can get REAL ground, not just synthetic) and tapped battery stacks. Some of HP's old 200 series oscillators used strings of NiCd batteries, and only trickle charged from AC.

Assuming your own power supply onboard, what's the ideal wallwart to start with? Maybe a Cuk AC generator at 10V, 40 kHz or so, three-wire so the GND doesn't carry power current? You can rectify or double/triple to get almost anything with not much extra hardware. Or maybe -48VDC like POE, to keep wires slender and delivered power quiet?

In this case, we are talking about a board, 4.5" x 0.85". Not much room for power converters, or even a wall wart connector. This is a daughtercard on a main board, which is plugged into a backplane that provides power to the rack. I got ±12V, 5V, 3.3V. The -12V is limited power, so it's not usable for the amplifier circuits.

That's why I need to provide a virtual ground at 6V. The previous generation did it simply with a couple of zener diodes, since the signal paths were all low current. But now there are some paths that have 50 ohms. Connecting them to the actual ground draws power all the time. The supply per module is limited and it has to drive 50 ohm loads, so I don't want to stress the main supply. I could add a cap to make it an AC ground, which would reduce the peak current. However, that's a large cap to get a 20 Hz passband.

An active virtual ground draws significantly less peak current, with one virtual ground for both channels on the module. Right now I'm leaning toward the op amp approach. A three pin SC70 would certainly fit the bill though.

OK. That might work. Especially if space was at a premium which can happen.

boB

We use +24V medical-grade wall warts (SL Power ME10A2403B01) that come with a set of international adapters.

We've used the multiple-output things occasionally, but they all seem to have these gruesome large-diameter, not-very-flexible cables with DIN connectors. 'Clunky' doesn't cover it.

Cheers

Phil Hobbs

We have standardized on a 24 volt Phihong wart for most things. And a

Right. Make what you need from 24. That's easy nowadays.

It's less easy when doing wideband ultrasensitive measurements, but yeah.

Cheers

Phil Hobbs

Our critical measurement is jitter, and a typical target is a few ps RMS. Switchers can play hell with that, especially ones that make, say, 400 MHz bursts. It's easier to prevent those from happening on a board, as compared to trying to filter them afterwards.

Another jitter contributor is the modulation of logic device prop delay by supply voltage ripple, which happens at any frequency but can be filtered. FPGAs are horribly sensitive, prop delay vs Vcore.

We designed one delay generator that had the fast trigger path go through an Artix7 FPGA. Big mistake. Delay changed 1 ps for a 70 uV core supply change. I tore up the fast path and made it all discrete logic and cut jitter and insertion delay both.

FPGAs are getting faster inside and slower pin-to-pin, which seems strange to me. So we keep designing with discrete transistors and diodes and gates and flops, kind of like you use discretes for the critical bits. A diode OR gate works pretty well.

There are some logic parts with femtosecond jitter, but they are too expensive for most uses.