Determining Decoupling Capacitor Value for EMI Reduction

[cyclone24]

Hello:

How does one determine the decoupling capacitor value in a circuit that is prone to EMI? For example, I need to remove the noise at 400 MHz. I use the resonant frequency formula

f = 1/ (2pi*sqrt(L*C))

What should be the value of inductance L in the equation? how to determine the L value in the cables?

Thanks!

 

[vk6kro]

You could use a "notch" resonant filter (series tuned L/C circuit to ground) for this if you knew there was only one frequency involved, but generally decoupling involves a low pass filter.

So, you work out the reactance of the capacitor in relation to the other components in the circuit. Decoupling is often used in power rails in a circuit and capacitors to ground from this line may be placed between inductors in series with the line.

At 400 MHz, very few capacitors work exactly as capacitors, so the choice is mainly between surface mount or feedthrough types. The leads on other types of capacitor can become more important than the capacitor itself at these frequencies and the device may function as an inductor rather than as a capacitor.

 

[cyclone24]

Thank you for the reply.

Can't I use ceramic capacitors that do not have leads on it. Also, could you elaborate more on the effects of cap leads at high frequencies?

I gathered that cap value doesn't matter at high frequencies. Use of multiple caps of same value might reduce the inductance.

 

[Averagesupernova]

Cap value can matter at higher frequencies. If you choose a cap that gives you an Xr of .001 ohms at 400 Mhz for instance, you will have a cap that is around .39 uF. Generally leads on caps of this size will introduce a significant amound of inductance. Much more than the .001 ohms you shot for. If you have wideband noise you want to reduce, choose several different value caps and put them in parallel. Choose your components in such a way to minimize the effect of their leads.
-
Edit: Even surface mount caps have inherent ESL. Check the data sheets.

 

[vk6kro]

Thank you for the reply.

Can't I use ceramic capacitors that do not have leads on it. Also, could you elaborate more on the effects of cap leads at high frequencies?

I gathered that cap value doesn't matter at high frequencies. Use of multiple caps of same value might reduce the inductance.

Ceramic capacitors can have a significant length of wire lead inside the body of the capacitor and even the plates of the capacitor can have inductance.
Reducing the wire length to an absolute minimum is a good start if you have to use ceramic capacitors, but it probably isn't a real solution.
Surface mount capacitors seem to be the best available at present.

The leads of a capacitor appear in series with the capacitor and will series resonate at some frequency. Above that frequency, the inductor has increasing reactance and this makes the capacitor useless as a bypass component above its resonant frequency.

 

[skeptic2]

I used to design amplifiers for 880 MHz. At that frequency, 39 pF surface mount capacitors were series resonant. We used them at many places in the circuit to keep the RF off Vcc.

Try to find the data sheet for the capacitors you will be using. It should give the series resonant frequency for each capacitor value.

 

[cyclone24]

Thanks again,

I suppose I was referring to surface mount when I said ceramic (0603). I have 3 specific queries.

1. There is almost an inch distance between Vcc and GRD. Somehow I have to connect the surface mount cap between them. If I solder it using short leads it will increase the impedance. How to deal with this?

2. At 880 MHz, you are using 39pF? I was expecting the cap range between microF to nF. But pF cap value seems too low. Did I miss something here?

3. The datasheets from different companies give me different ranges at the frequency that I am looking at, say 100 MHz to 400 MHz. I see an "insertion loss" plot. What is it do? And how different is it from attenuation plot in dB? I am thoroughly confused.

Any pointers to the above three would be helpful.

Thank you

 

[vk6kro]

You can't have 1 inch leads at 880 MHz or 400 MHz. You could cut a piece of copper foil so that it can be soldered to the ground lead and almost touch the V+ lead. Glue it solidly in position and then solder the capacitor between the new piece of copper and V+. This foil should be much wider than the wires you would otherwise use, so it will have less inductance.

Slightly better is to cut the V+ line, put a small coil of wire in there and have a surface mount capacitor on each side of the coil to the copper sheet. You would have to model the coil/capacitor combination to optimise it. This may have so many unknowns to it that it would be better to do it experimentally.

39 pF would be OK at that frequency. It only has a reactance of 4.64 ohms and even this would largely cancel out if it was self resonant.

Also, don't forget that bypass capacitors are a vital part of the circuit. They are the return path to ground for signals and tuned curcuits won't work properly unless the "cold" end of the circuit is grounded for RF. So, getting them right is very important.

 

[cyclone24]

Thank you for the detailed reply.

I think that is a good idea. Will try that. The thing is I am using long cables (over a meter) before it reaches the PCB and then to V+ and grd. Right now I put a 100 nF cap between V+ and grd, it didn't help at all.

Since then I started thinking to compute the reactance on the cables and what cap value I need to use and how/where to use it.

 

[sophiecentaur]

I would agree with most of the above but:
Ferrite beads can be very effective in reducing high frequency EMI. They are very lossy at high frequencies (appearing as a high series resistance at the unwanted frequency) but have no effect on lower frequency signals. As they are not 'tuned' the values involved are not critical - mainly depending on the particular ferrite used, which can be ascertained from manufacturer's spec.
Ferrites are frequently used on computer equipment leads for supression - in fact I am looking at one at this very moment.

 

[skeptic2]

All of our RF circuits were built with a nearly solid ground plane on the other side. The 39 pF capacitors were connected to ground via vias or feed throughs to the ground plane. We measured the inductance of the vias at 0.4 nH.

Yes, the 39 pF is correct. Once when we were qualifying a new vendor for capacitors I saw on their spec. sheet the series resonant frequency was I think about 1.1 GHz. I mentioned this would be a problem because their capacitors wouldn't be interchangeable with the ones we were using. It turned out, so they said, that they had used a formula to calculate the resonant frequency instead of measuring it. When they measured the resonant frequency it was about 880 MHz.

 

[cyclone24]

Thanks for the reply.

Is measuring the inductance on the cables pretty straightforward? I think I am going back to the same question as to what is "L" in the resonant equation when you are figuring out the cap value. My supervisors suggest to use caps in the order of few hundred nF at 400 MHz. I want to reason that.

Do you know the formula that the vendors used?

Thanks.

 

[vk6kro]

If you have two capacitors which each have similar lead lengths, which one would have the lowest resonant frequency? The one with the most capacitance.

Bearing in mind that a capacitor used above its resonant frequency is very nearly useless, you should use as small a capacitor as you can.

Here are some reactances for capacitors at 400 MHz:
25pF 18 ohms
50 pF 8 ohms
100 pF 4 ohms
220 pF 1.8 ohms
500 pF 0.8 ohms
1000 pF 0.4 ohms
You can see that the capacitance doesn't have to be very high for the reactance to be low.

If you look at the power rail from the RF signal's point of view, it has two choices. It can go along the power line or it can go down through a capacitor to ground.
It doesn't really choose. It will split according to the relative impedances in each direction.

So, you need to make the inline impedance high (by adding inductance in series with the line) and make the path to ground lower impedance by adding capacitors or series tuned circuits.

You asked:
what is "L" in the resonant equation when you are figuring out the cap value
but this has no meaning. Mostly we are not talking about resonant circuits because we want the bypassing to be effective over as wide a range of frequencies as possible.
What is important is to get as low an impedance to ground as possible.

If we don't get this bypassing right, you can get oscillations at frequencies away from your operating frequency and you may not even know about them.

 

[cyclone24]

Got it!

Thanks a lot for patiently answering my queries...

1. The cap value need not be high.
2. Extend the ground plane to solder cap close to V+.
3. Add caps in parallel for high impedance loads.

 

[skeptic2]

No I don't know the formula they used. I once tried to determine the series inductance of the SMD capacitors but got different inductances for different values even for the same series of capacitors.

Are these capacitors being used in a transmitter or a receiver?
If a transmitter, are you doing impedance matching?
Frankly, using a few hundred nF at 400 MHz seems like overkill. Using such a large value permits lower frequency fluctuations perhaps from the power supply or other parts of the circuit to contaminate your signal. In fact in RF it is quite common to use the decoupling capacitor as part of a filter that eliminates frequencies you don't want.

I second vk6kro's comment that you can easily get oscillations by using the wrong value of decoupling capacitor.

 

[cyclone24]

The device I am working with is a DC circuit. But I have an RF coil very close to it. This coil is tuned and impedance matched. So when I transmit RF at 400 MHz, it couples to the DC circuit.

so in the hunt of a good decap, I tried 100 nf at the V+ and grd of PCB. I am fresh from a test now, where I put a 470 pF decap. Didn't help either.

So will try some other value now...

 

[vk6kro]

That sounds like you need shielding rather than decoupling.

Decoupling is filtering of AC or RF signals on a DC power line.

For shielding you need to provide a grounded piece of metal between the source of the RF and the DC circuit.

 

[cyclone24]

I am doing both. Just posted a query under 'Of shielding and apertures and seams' thread.

 

[skeptic2]

Can you re-orient the RF coil to minimize coupling?

We also used RF filters on our DC lines that essentially were low pass T filters. They consisted of a ferrite bead, a high dialectric bead and another ferrite bead. All three were in a conductive sleeve which was connected to ground.

 

[cyclone24]

No...reorienting didn't help. And I cannot use ferrite beads due to space constraints. So it has to be SMD caps.

I noticed another thing. I am using decaps at the V+ and grd terminals on the PCB. I shortened the leads etc. Now I see that the pickup has increased! I used 100 nF cap. What could've happened here? Maybe the cap ground is floating? Or cap value is too much as discussed earlier?