Huge issues with TPS62122

For your convinience, copy of plots I posted in the Wireless Forum:

All measures are taken direct at CC112x.

Noise at 6V Vin (well looking PA output), time base 2us:

Noise at 15V, about 900kHz freq:

Noise at 3.6V (there seems to be a resonance, 0.2V above and below it's quiet), time base changed to 10us -> 400kHz signal:

Schematic (there is a bead between 3V3 output and RF-VCC; this bead reduces noise already ~30%):

Board (Red: Top; Magenta: Layer2, GND; Cyan: Layer3, PWR; Blue: Bottom):

Thanks for the plots and design-files, I requested in the Wireless-Forum.

Do you know the current drawn into your CC1175?

I just try to understand your ground connections. Is the GND-Pin in the lower left corner connected? How about the thermal pad?

Thanks,

Florian

Florian,
current is appr. 50mA over the full Vin spectrum, constantly.
If we change the buck's output from 3.3V to 2V, behaviour above 9V is almost the same but the most ugly 'resonance' point moves from ~3.6V to ~2.5V.

We tried all kind of different C/L/R combination, used even Tantal or a serial R to increase ESR of output C (with negative effect), changed bead to inductor (up to 500nH), doubled C's, put small/medium/large C's at the output in parallel, changed R's at Fb to 1/10 of the original values, changed 22p-C at Fb; with almost no effect.

If helpful I would like to send you the full board data via PM, you can open it with Eagle (free as viewer) and fade in all layers as you like.
But I attach some more pictures; the second pin (from top) on the left is GND, the third one VCC; others are data pins, not used/connected in test bed. Thermal pads are fully soldered to ground, connected through vias to ground plane. CC12xx is in the upper right corner, an MCU down right. Fragment size is appr. 16 x 20mm.

Top only:

L2/GND:

L3/VCC:

L4/Bottom:

Hi,

thanks for the drawings. I see - the first layout had the planes skipped. Now it's getting clear for me.

One thing to mention here is the routing of the feedback. You took quite long way routing this trace to the output.

Could you check if the power stage is stable by its own? Meaning check the DC/DC converter operating without the controller? With this test we could verify the stability of the converter itself respectively in combination with the whole system.

Could you please check the waveform of the input voltage as well?

Thanks,

Florian

Florian,
without the RF chip the output is quiet (beside some other lower noise that might result from 'no load').
What we tried out in addition is

a. put 50mA more load (sum 100mA) at the bucks output. Result: 10dB less peaks at PA output. Means that with increasing load the output ripple becomes lower.

b.  to double the inductor by putting two of the 22uH in series - this reduces the ripple significantly (not at the Uout=Uin point).
I put two short movies to show the ripple as a function of Vin, upper curve is Vin, lower is the buck output.
Please forgive quality and size:

Ripple-MOV

1. How come that using a much to big inductor ripple calms down
2. Even more important, what's the price tag of using a larger inductor?
3. For the future: what's a reference design of using a buck for up to ~12V Vin, that does not ruin PA output spectrum

thx.