The power supply for example for the DHT22 sensors I'm currently using in my "Home Measurement Project" has a big influence on the accuracy of the readings. The graph was getting a bit "nervous" and unstable.
I removed the huge 1000µF electrolytic capacitors and added a few capacitors to the ESP board and the sensors: 100nF, 10µF, 100µF ceramic capacitors and one to two 100µF tantalum capacitors. This looks way better than before!
With those tantalum capacitors and 0.1+10µF ceramic capacitance the result was looking better when looked upon with the oscilloscope. But you can see bigger 10ms areas where the voltage significantly drops to below 2.9V.
Thus adding 100µF ceramic capacitor was helping a lot, looks way better already! But still the voltage drops to ~3.0V for those 10ms slots which is still not optimal.
The USB-to-serial-boards are nice for programming the ESP boards, but for permanent power supply, something more special is needed. I found a few RT9166A in my LDO drawer. They can deliver up to 500mA - but in the SOT23 case can only stand 0.25W of heat. 1W of heat on the other hand leads to a temperature rise of 250°K. With 500mA at 5V to 3.3V I was calculating with 0.85W of heat ... thus I added fat lines of solder tin so the dissipation would be improved at least. To my surprise, the RT9166A stays really cold!
The result is very satisfying. You ca see the transiants on load changes, but no sustained voltage drops on the plot! Thus it is a good idea to use a proper decoupling with capacitors for the sensors, the ESP board itself and use a proper power source!
Two sensors are running with TD6810 switching regulator and one with MCP1825S LDO. They all run stable and don't need the USB-to-serial-converter anymore.