Decoupling power supply

Here's how I intend to build my next project:

  1. USB power supply goes into USB MIDI microcontroller, which is soldered on a small PCB
  2. use power supply pins from that board into bigger board containing Arduino Pro Mini, 4051 multiplexer (with decoupling cap), few diodes and resistors for button/LED matrix and a bunch of connectors

Now, should I place caps on power supply line on that bigger board where Arduino is located (as close as possible to connection I assume) to further smooth out the voltage? If so, what type? Regular ceramic ones (0.1uF), electrolytic, or something else?

Small ceramic capacitors are excellent sources of power for digital signals / activity inside a chip. Their respective ends should be placed as close as possible to the power pins of the chip you're using. See "Noise Reduction In Electronic Systems" by Henry Ott, a classic on the subject.

A bigger capacitor is typically placed close to the power supply to help deal with transients. How big a capacitor depends on the power supply, the scale of the transient relative to the capabilities of the power supply, etc. I'd consult the data sheets on the power supply to be sure. For example, most of my power supplies have bulk output capacitors ranging from as little as 1uF to 10uF. However, a GPRS-chip power supply has 470uF input and output capacitors to help deal with the very high transients that occur during GPRS transmissions (very short bursts of 2A draws).

What type of bulk capacitor to choose is again to be determined via the data sheets of your power supply. Some devices like low-ESR aluminum electrolytic capacitors, some prefer Tantalum, and some insist on ceramic X5R types. It all depends on what you're using to regulate the power supply. If you go with Aluminum Electrolytic or Tantalum Capacitors, most of those are polarized and will be very unhappy if you put them in the wrong way round. Also, it's usually a good idea to choose a rating voltage at least 20% higher than the highest you expect to hit your circuit.

Lots of info, thanks. My power source is USB from laptop, so, what's the best choice?

You don't want to start creating problems when the problems don't necessarily exist. See if it works without any decoupling capacitor or if you want to be on the safe side start with a 10uF electrolytic capacitor, 16V rated or better. Feel free to experiment with different sizes or multiple sizes in parallel.

Different types of capacitor have different characteristics. Ceramic capacitors are easiest to make in low capacitances, (though manufacturers are always pushing back the limits). They're made as alternating layers of conductors and dielectrics, and are effectively numerous capacitors in parallel. They're great for decoupling high frequencies. Electrolytic capacitors are made by coiling up foil and electrolyte-soaked layers which produce a very thin dielectric on the surface of the foil. This enables them to be made with much larger capacitances, but because they're coiled up they have inherent inductance. This makes them less responsive to high frequencies. This is why many designers put the two types in parallel, with ceramics close to the sources of high frequency noise, as Constantin says.

kustom:
Lots of info, thanks. My power source is USB from laptop, so, what's the best choice?

Again, it depends heavily on the application. That 2A power supply I mentioned above runs off a USB connection w/o issues. In that particular instance, the designers are throwing large capacitors at the problem to ensure that the computer never has to put out more than the 500mA limit that the USB folk imposed by default on the USB power supply. Granted, many laptops can deliver more, but 500mA is the usual design limit without various enumeration schemes, IIRC.

If you current is small and relatively constant, you may not need any additional capacitors. However, installing some capacitors (one electrolytic 10uF @10V or higher at the entrance, and 0.1uF capacitors to decouple every IC component on board) will likely improve board performance. If you notice funny behavior (MCU resets and the like), pull out a good oscilloscope and see if the voltage supply is the culprit. If so, investigate what you can do in terms of on-board storage to deal with transients, etc.

If you have any analog signals to deal with, you need to take an even closer look at board layout, decoupling, and so on. Especially if you're dealing with very sensitive ADCs, op-amps, and the like. I.e. you'll spend as much time troubleshooting the power supply to make sure it delivers clean power as designing the actual circuit that uses said power.

Some folk also like to run their USB power through ferrites to help deal with transients / high frequency noise on the line. 10uH 0805 chips and the like are pretty common for that. Only a good scope though would tell you how much of a difference such chips actually make.

Chagrin:
You don't want to start creating problems when the problems don't necessarily exist. See if it works without any decoupling capacitor

Every chip can (should) have it's own decoupling capacitor.

Constantin:
If you current is small and relatively constant, you may not need any additional capacitors. However, installing some capacitors (one electrolytic 10uF @10V or higher at the entrance, and 0.1uF capacitors to decouple every IC component on board) will likely improve board performance.

If you have any analog signals to deal with, you need to take an even closer look at board layout, decoupling, and so on. Especially if you're dealing with very sensitive ADCs, op-amps, and the like. I.e. you'll spend as much time troubleshooting the power supply to make sure it delivers clean power as designing the actual circuit that uses said power.

Well, it's really simple 2 button row/2 LED row/8 shared columns matrix with 6 pots on 4051 multiplexer and 3 additional faders connected directly to Pro Mini. Really small currents, but I'm mostly interested in smoothing out the main voltage as much as possible, because of those analog components. Plus I'm eager to learn good PCB design (ground planes, decoupling etc.)

Some folk also like to run their USB power through ferrites to help deal with transients / high frequency noise on the line. 10uH 0805 chips and the like are pretty common for that. Only a good scope though would tell you how much of a difference such chips actually make.

Unfortunately, I don't have one.