Precision 3V6 Voltage source @ 200mA for ADC

[Stiive]

Hi,
I'm designing a board that will require very accurate ADC conversions. I will be using the STM32F4 DSP chip with 12bit ADCs.
Now the chip itself is $17, and it comes in a demo board for $20 with on-board programmer. Since I am running out of space on my PCB anyway, I thought id just make use of the demo board and connect via header pins.
Only problem is, the Vref for the ADC is connected on the board to the Vdd pin (via RLC), and Vref is not broken out to the header. So I have decided to provide Vdd with the most accurate voltage possible.

The DSP chip needs about 150mA max, so I think 200mA is a good conservative figure.
Anyway, what is the best way to provide a precision 3V6 power source to Vdd? I have 15V and +-5V already on my PCB which can be used. Efficiency is not an issue, and to a certain extent circuit complexity.
There are plenty of LDO regs that have 1%, and I found a precision shunt reg that has 0.5% and 200mA (AP432A). Or should I build a circuit using a precision reference and a power buffer amp (such as the LT1010)? I'd really want to get the voltage as accurate as possible while keeping noise/ripple to a minimum.

Suggestions? I will consider downgrading to 3V3 too (more available) if the benefits are there.

Cheers

[Psi]

Only problem is, the Vref for the ADC is connected on the board to the Vdd pin (via RLC), and Vref is not broken out to the header. So I have decided to provide Vdd with the most accurate voltage possible.

The problem of getting an accurate 3.6V source is only half the issue. The other is that the vdd current draw changes as the chip does stuff.
This will cause a voltage dip even if you have a super accurate/precision 3.6V supply.

Ideally you should do it properly and separate the pin somehow, but if that's impossible i would put equal effort into getting vdd as smooth as possible using lots of ultra low ESR caps right near the chip on the VDD rail.
(not too large though, you want high ripple current rather than high capacity as this may cause issues for the power supply)

[Stiive]

Well, I could de-solder the resistor (of the RLC) and solder on a shielded wire to bring Vref down to my PCB. This would leave just the decoupling cap near the Vref pin.
Still don't really like it though even with shielding...

[Psi]

If the official demo board connects Vref and Vdd to the same point perhaps it's not really much of an issue.
Maybe test it and see what the ADC noise is like

[Stiive]

hmm yeh but I've spent a small fortune designing all my signals up to the ADC using extra precision components and ultra low offsets.

[hans]

I have got an STM32 board myself, and what you could do is desolder resistor R23 and wire one side to your voltage reference. The analog supply seems decoupled with a small inductor, so they have done that properly. R23 is reasonably to get to, it's next to the crystal oscillator.

I am interested to hear what the noise is like. I am planning on using this board (that's why I got it) for some high-speed projects as well.. possibly utilizing it up to 7.2MSPS (maximum).

Oh btw, if you're doing this, I wouldn't use 3.6V, as it's higher than VCC (I don't think the device can handle that - haven't checked the datasheets). 2.5V or 2.048V is maybe more ideal. 

[Stiive]

I have got an STM32 board myself, and what you could do is desolder resistor R23 and wire one side to your voltage reference. The analog supply seems decoupled with a small inductor, so they have done that properly. R23 is reasonably to get to, it's next to the crystal oscillator.

Yeh thats the resistor I meant, this will leave the decoupling caps still close to the pin. This would however remove the inductor from the Vref line, but I guess a wire will probably add more inductance (maybe too much?) than L1 anyway

I am interested to hear what the noise is like. I am planning on using this board (that's why I got it) for some high-speed projects as well.. possibly utilizing it up to 7.2MSPS (maximum).

Oh btw, if you're doing this, I wouldn't use 3.6V, as it's higher than VCC (I don't think the device can handle that - haven't checked the datasheets). 2.5V or 2.048V is maybe more ideal.

The chip is capable of 3.6V (4V max), the dev board however uses 3V. I wanted to upgrade to 3.6V to get a wider spread on my ADC (scaling 0-400V to 0-3.6V). I will also be trying to use the ADCs at 7.2MSPS


Would it be best to just ditch the demo board entirely? I'd prefer not to though, I like the modularity incase I let the smoke out, and the on-board programmer. Not to mention the space saving.

[Stiive]

Still tossing up whether to drop a shielded cable down to my PCB below. Its worth noting though that this setup will be in a particularly noisy EMI environment.

Anyone?

[free_electron]

put chip on the pcb. forget the demo board sandwich. you're looking for trouble ...

if you want to get any kind of precision out of that adc you need to use a reference diode. something like a LT1004 / LT1790 / ADR366 these will give you 3.3 volt reference.
power these from an adequately filtered supply , make usre you connect AGND and AVCC of the STM32 chip in the correct way , provide filtering , ( PI filter using two caps and choke/ferrite)

and take care of not having digital return signals running through your analog return.