add one resistor, quiet ugly smps ADC noise

If a schematic's worth a thousand words, a schematic with component values is worth even more. :0) FWIW, you leave me with the impression that you're doing something similar to this:

-----------------------+--------------------------------o +14V | +------------+ R1 2.2M | TPS61040 | | | | | | FB +-------+---R3-----< +3.3V Enable | | | 68k | GND +---+ | | + | R2 180k +------------+ | |

-------------------+---+--------------------------------o GND

How well does that work? Stopping the switcher will kill the big switching ringing, the main noise source, but will "freeze" the ripple at random voltages.

Thanks, close, but if a 3.3V logic signal swing is to lower the output setpoint by 3.3V, then R3 = R1. In fact I used R1=R3= 2.49M and R2= 167k.

It works extremely well. The "noise" isn't from inductor-switch node ringing, etc., but rather from fast di/dt ground bounce, etc. This kind of noise is very fast, transient, and instantly disappears.

Hi

I don?t really see the point of your efforts about this

With 16bit, one LSB is 200uV

You most likely have about 1uF output cap and a switching ripple of 100mV. So you will never measure this correctly if you don?t filter the he ck out of it, and the 1us idle makes no sense, since if you have filtered i t will be noise free

By the way, I have done digital switch modes a couple of times and digested many papers, and have never seen a system with more than 12LSB

Cheers

Klaus

No, sorry, there's a high-res ADC on the circuit board near the converter, measuring a TIA-amplifier signal. The converter creates 14V for a 50mA LED current source. The TIA amp, ADC and smps are sharing grounds, etc., and the introduced converter noise looks huge on the amplified signal, when viewed on a scope or with the ADC data. But the noise disappears from the scope and ADC data while the converter is momentarily stopped. BTW, we may only be using 10 to 12 bits of the data, but the max noise seen must have been at the 4 to 6-bit level! Anyway, I added the extra 3rd resistor to the next rev of the PCB and now we're happy. I re-purposed a controller bit that wasn't busy doing anything else during the ADC conversions.

I thought others might like to know about this little trick.

torsdag den 28. juni 2018 kl. 20.49.38 UTC+2 skrev Klaus Kragelund:

. So you will never measure this correctly if you don?t filter the heck out of it, and the 1us idle makes no sense, since if you have filtered it will be noise free

ed many papers, and have never seen a system with more than 12LSB

afaiu the switcher and ADC are not related they just have to live together

I've done similar and also paused things like display updates during burst of sampling

I believe there also are some MCUs that can automatically go into "sleepmode" during sampling

That is slick. We's considering an EOM driver with a 9 Gs/s, 14-bit DAC on the same board with switchers. The active shot is under 50 ns, so we could kill all the power supplies when we fire the DAC.

Here's a board with a 12-bit, 250 Ms/s ADC a few inches from some switchers.

ADC noise is about 1 LSB RMS, dumb luck mostly. Well, everything is differential.

Like, computing trashes the on-chip ADC? Makes sense.

The ARMs that we use can, I think, have the ADC programmed to sample periodically and then interrupt. So the CPU could sleep-on-interrupt.

An external ADC isn't bad; just keep the SPI bus quiet while the ADC is busy. Some SAR ADCs are actually sequenced by the SPI burst, but that's probably OK.

torsdag den 28. juni 2018 kl. 21.40.29 UTC+2 skrev John Larkin:

back in the day when I was doing Bluetooth we did a stacked chip with a transceiver and an SoC, everything worked great until the code was mowed to ROM instead of RAM then the sensitivity dropped

turned out to be running ROM interfered with the PLL, running code in RAM didn't

We like the 14-bit TI ADC141S626, which works like that. It has this weird continuous mode where you leave CS' low, and it takes 18 clocks for a 14-bit conversion. That's a nuisance since the MCUs we mostly use (NXP Cortex Ms) can't do SPI transfers that long in hardware.

However, if you work CS' on each acquisition, it only needs 16 clocks. It has super nice DNL and other good properties, and it's about $3 in thousands.

Cheers

Phil Hobbs

We use ADS7866 for a cheap outboard ADC. SOT-23-6. 12 bits, $1.50 by the reel.

I once built a 10-bit ADC out of parts... transistors, no ICs. 3U rackmount. It sold for more than my Austin Healey Sprite.

I interfaced it to an IBM 1401. It was used to digitize waveforms from monkey brains. Try to top that!

I'm assuming you were already using a quiet ground for the ADC with a single-point tie to system ground? And possibly a resistor in series with its Vdd pin? And this wasn't enough?

Nice trick, have to try out that one! I've got a sensor design with analog side supplied by TPS61030 and enough caps.

Having pushed one through EMC, and seen the ~ns edges from one, I laugh derisively.

Tim

After considerable searching, I settled on MCP3201 for my go-to 12-bit ADC. This part has a pin for the voltage reference, so you aren't forced to use the digital supply. It also has separate + and - analog inputs, nice. Or place a 16-bit ADS8325 on the PCB (same MSOP8 pinout), that sealed the deal.

Nope, sorry, there was no room for best practices on this long thin PCB. Yes, a separate "quiet" analog supply, but was forced to be satisfied with a single slightly-fattened ground trace running the length of the board. And the TIA amp was next to the 14-volt boost converter. Yes, very risky, but perfectly OK, with the single-added-resistor smps shutoff trick.

The TPS61030 has a different soft-start scheme, that looks at the output voltage, so you might be able to momentarily stop conversions with the enable input. But its control scheme might also allow for quick cutoff, with the added resistor trick. Let us know.