I'm making a large circuit involving multiple IC's both digital and analog. Some components will be running at 20Mhz (processor), some at around 500Mhz maximum (radio), and others at substantially lower frequencies.

I learned that to minimize chances of EMI, I need to decouple the power supply, so noise from one circuit doesn't run to the next.

What I have attached is a small fragment of my design in different versions. The power supply is shared amongst all IC's. The inputs (1st, 2nd 3rd) are address lines connected to the microprocessor. These lines get updated roughly 100 times a second. The data line (D) is updated at least 100,000 times a second.

I'm trying to figure out which version is best to use.
the first one at top-left is coupling I think most people use (+ve runs through resistor and gets shunted by capacitor). If I use this method, the circuit board will not be routable. I'm limited to single-sided PCBs.

The second method (top right) is bad as noise can freely enter the circuit, so I see no advantage to it.

The next one (middle-left) is the same as first except I'm applying the filter to a chip input. (I'm not sure if failing to do this causes the chip to emit noise via inputs?)

The second last one (middle-right) is the same as middle-left but I did the coupling on the negative side as well.

The last one is similar except I connected -ve to the G2 inputs on the chip directly.

If I use the last or 2nd last circuit, routing will be substantially easier on a single sided board without requiring jumpers.

Assume for all circuits that each resistor is 10 ohms 1/4 watt, and each capacitor is an electrolytic 100uF rated for 16 volts, and each battery provides 5VDC of unregulated power (for sake of argument, lets say its 4 1.2V cells in series.)

Apart from what I mentioned, are there any benefits or drawbacks to using any of the five methods I have shown?

 

The way I see it is that your circuits with a resistance in the ground to the IC are going to be a problem. The noise margin of the digital inputs will be decreased and the output voltage from the IC may not meet the logic level requirements of the IC that it drives.

In other words, never put any impedance in the ground leg of a digital IC. In fact, you want to do the opposite of that. You want a very low impedance ground plane to all of your IC's.

The best decoupling is done by putting an impedance in the power supply of the IC. To work the best this should be an inductor. Then, decouple the IC power pins with both a ceramic capacitor and a larger value bulk capacitor. You should also do this with each stage of your RF section to keep it from getting "noise" from the digital section and to keep the RF out of the digital circuits. The ground planes of the digital and RF sections should be kept separate and then joined only at one point. I expect that you will have a very difficult task doing this right on a single sided PCB.


This is just a hint of what you need to know. It would take an entire book to give all of the details.

 

the first one at top-left is coupling I think most people use (+ve runs through resistor and gets shunted by capacitor). If I use this method, the circuit board will not be routable. I'm limited to single-sided PCBs.

Right there is one false thinking.
I will wager that just about any board can be routed on singled side PCB.
We know how to route thru-hole resistors and capacitors, right?
How about routing 0Ω resistors called jumpers?

Here is your best practices for mixed signal boards.

1) Two solid ground planes - one for digital and a second for analog.
2) Decouple the power supply feeds with resistors or inductors depending on the requirement and frequency of each section.
3) Use decoupling capacitors in parallel, e.g. 10nF, 100nF, 1μF, 10μF, depending on the section and frequency.

Remember, more capacitance is not necessarily the correct answer. You have to look at ESR and frequency range.

 

I think you have totally misunderstood decoupling.

These capacitors sit CLOSE to the supply pins of each IC. Some analog circuits will use capacitors on other pins as well. The real key is CLOSE to.

In general follow the manufactures recomendations. These are normally ceramic capacitors.

Various parallel combinations of capacitor types are used sometimes. It's a science and an art, but there are ways to optimally select the capacitance. Leaving them out is bad.

You don't want to add capacitance to inputs of IC;s. that just slows down the slew rate.

There is another way to look at decoupling too. Keep the signal grounds, dirty grounds and analog grounds and chassis grounds SEPARATE. At one point, they get combined at one point. But, sometimes they can be de-coupled in different ways.

So in an audio amp I built, everything terminated in a huge star ground except one thing. The input used an isolated connector and a small resistor of about 5.1 ohms was used to connect the amplifier inputs to the chassis. The 5.1 ohm resistor does decouple in a different way, It helps to reduce the effects of a ground loop should it develop.

In an early design of something I built, I didn't fully understand the problems with ideal and reality. Even a tiney amount of current (say picoamps) across a nearly zero ohm resistor can create lots of voltage. You have to introduce a series resistance so that current has someplace to develop a voltage across. Also in many designs, you have to provide a place for bias or leakage currents to go. This isn't de-coupling.

 

ok I'll try LC instead of RC for decoupling. I'm gonna start with 1mH for inductor and 100uF for capacitor. I may change the values later.