– Hey, everybody.

Welcome to Altium Academy.

I am Zach Peterson,

your host and your local Technical Consultant for Altium.

And today we are gonna talk

about PCB noise reduction techniques.

So, what do we mean by noise reduction?

Well, there’s all sorts of different types of noise

that can arise in your PCB.

We’re gonna talk about a few of those

and three specific strategies that are often brought up

to try and reduce all these different types of noise.

These are filtering, adding shielding,

and something called isolation.

Now, I know isolation sounds pretty vague,

but that’s all the stuff that we’re gonna talk about

in this video.

Let’s go ahead and get started it.

(upbeat music)

Okay, so when we talk about noise reduction in a PCB,

the term noise is pretty ambiguous.

It could mean a lot of different things.

It could mean crosstalk, so like signal-to-signal noise.

It could mean some other form of electromagnetic coupling,

which is usually something like power-to-signal noise.

It could mean radiated emissions

or reception of external radiated emissions.

It could be conducted emissions.

All sorts of different things.

There are even thermal effects that produce noise,

and that’s something that’s a little more advanced

and goes back to what happens inside of components.

But what we wanna do is actually look

at some of these strategies that are usually brought up

as like cures for different kinds of noise.

So some of these noise reduction strategies

will include, number one, filtering.

So when we talk about filtering,

we’re usually talking about literally

a circuit that filters stuff.

Filter circuit,

sometimes something as simple as just an EMI filter,

like a ferrite bead,

is usually brought up as a cure

for all these different types of noise.

So you have to be careful with filtering,

because you wanna pair up the filter that you’re gonna use

with the specific noise that you want to eliminate

while also not affecting the signal

that is carrying the noise with it.

So this is not a cure at all,

it’s something that you have to very precisely plan for,

and that’s something that we’ll talk about here coming up.

The next common noise reduction strategy

that people like to bring up is something called shielding.

So I think if you ask 10 different designers

what shielding is,

you’ll probably get about 10 or 20 different answers.

So most people, when they refer to shielding,

they’re probably referring to a Faraday cage,

meaning they’re basically just gonna put a bunch of metal

around the problem circuit and they’re gonna ground it,

and then that’s going to basically eliminate the noise,

and that’s essentially the idea.

Now, there are components and materials

that you can use to provide shielding in your PCB,

and we’ll discuss when those are basically relevant

and what types of noises they deal with.

The third is something called isolation.

Now, this is even more vague than shielding.

So isolation is usually brought up

in the context of mixed-signal design.

And we’ve talked about mixed-signal design before

when you basically have different frequencies

or digital and analog signals on the same board

and you wanna make sure

that they don’t interfere with each other.

So isolation refers to those strategies,

and we’ll get into those last,

because frankly, that’s the easiest one to deal with,

and we’ll talk about that more here soon.

So first things first, filters,

when should you use filters?

Well, filters should really only be used

when dealing with analog stuff, okay?

You should not use filters with digital stuff

unless your filters have very high roll-off

which is generally not gonna be the case.

Filters are generally used to target low-frequency analog

or you’ll have specialized filters

with like 50-Ohm input impedance

that will be used for RF signal chains.

So filters are useful, but they’re best for analog stuff,

and here’s the reason why.

So generally, when we talk about filters,

we talk about a transfer function.

And if you graph a transfer function for a filter,

H is gonna be as some function of frequency,

and there will be some curve here, okay?

And let’s just take a low-pass filter

just as an example,

which transfer function basically looks like this.

Now, the transfer function tells you the output

versus the input voltage,

and that’s equal to my transfer function

as a function of frequency.

So what this is telling you is that if I have a signal

and it’s operating at this frequency, f1,

and let’s just say, it’s just a spike here.

So it’s just this one signal operating at f1.

What this transfer function is telling you

is that it’s gonna allow all of the power in f1

to reach the destination.

Now, everything above this f3dB roll-off,

everything above that is gonna get heavily attenuated

and will essentially get removed from the interconnect.

And so the point of drawing out a transfer function

is to tell you that you can’t just pick random filters

and pair them up with your particular interconnect

and then expect to remove every noise source.

So you should actually look at what the noise is

on an oscilloscope, or even better,

on a network analyzer or on a signal analyzer,

and you can actually pick out the frequencies

that you need to eliminate.

So if you just wanna transfer one specific frequency

down the interconnect,

maybe you have a driver component,

and you have your trace,

and then you have your receiver,

somewhere along this trace,

it’s picking up some noise.

So if you already know

the frequencies involved in this noise source,

then you can design a filter

and put it somewhere here, usually near the receiver,

for this signal that is going to only allow this signal

while removing this particular noise source.

So your filter options are usually something

like a notch filter.

That’s gonna remove specific frequencies.

You could also have a bandpass filter,

so meaning it’s only gonna allow

the frequency that you want,

so in this example, f1, while removing everything else.

You could also have a band-stop filter.

So maybe just in this example

you have another frequency, f2,

that’s originating from your noise source

and reaching your trace.

And then in that instance,

the band-stop filter could target f2,

while having zero attenuation at f1.

Filters are useful for targeting specific frequencies.

Now, if this were, say, a digital interconnect

and we had a signal that was a digital signal

and it’s traveling this direction

from the driver to the receiver,

well, should we use filtering?

Well, the answer is no,

we actually shouldn’t be using any kind of filters.

The reason we shouldn’t be using any kind of filters

is because, generally, with filtering components

that are operating in practical frequency ranges,

you’re actually gonna attenuate a lot of this digital signal

and you could distort it to the point

where it doesn’t trigger the receiver properly.

And the reason for that

is because a digital signal does not have all of its power

concentrated at one frequency,

its power is actually concentrated,

well, theoretically, all the way out to infinity.

Usually, the cutoff for our digital signal bandwidth

is gonna be very high.

And so we’ve talked about this before,

that actually relates to the rise time.

So if you go back and watch one of our videos

on digital signal bandwidths,

we’ll put the link in the description,

that will tell you why the power spectrum

of a digital signal spans out so far

into the high-frequency range.

And you can’t remove high frequency noise

with something like a low-pass filter as I’ve shown here.

Next, shielding,

what exactly is shielding?

Well, shielding, as I said before,

is pretty much anything that you can use

to create, essentially, the equivalent of a Faraday cage

around your components.

Now, no Faraday cage is perfect,

and you could basically go just as heavy as you want

and make a totally enclosed

shielded metal enclosure around your board

and to try and eliminate any sources of outside noise.

And you could do that,

but you may not be successful.

There are actually other things

that you’ll wanna do in your PCB

rather than just defaulting to shielding

to try and solve all your noise problems.

So just as an example,

let’s take a look at our cross section of a PCB.

And let’s just say for a moment

that I have an integrated circuit here

and it’s receiving a lot of noise from some outside source.

Well, the guys that claim

that you should always use shielding

around these types of components

will basically put some vias around here.

And what they’ll do

is they’ll go find a big metal shielding can

that basically encloses this entire component.

And so the idea here is that

if this component is emitting a lot of noise,

the shield is then gonna block this.

And then what they’ll do

is they’ll attach this to some pads,

they’ll connect it to ground,

usually on the next layer,

and then that basically creates a totally enclosed cage

that prevents any noise from escaping this component

and possibly reaching other circuits.

Vice versa, it also prevents any noise from out here

from entering inside the cage up to very high frequencies

and then interfering with any of the operation

of this component

or any of the other connections that it makes

that are inside of this cage.

So that’s kind of the idea behind shielding.

And people will go to various lengths

to try and create a similar structure

around different elements in a PCB.

So if you’ve seen any of my presentations

on high-frequency design,

one structure that I describe

is actually called a substrate-integrated waveguide

where you essentially shield a signal

inside to the substrate

by enclosing it on all four sides

with copper planes and vias.

It’s essentially the same thing

as like a grounded coplanar waveguide.

Coplanar waveguides, speaking of the devil,

do the exact same thing.

So let’s say that you have your PCB here

and you want to create a coplanar waveguide,

well, what you do is you basically have a trace

being routed like this,

you have some copper pour,

this is ground,

this is ground,

and then you have a bunch of vias lined up along here.

This essentially creates some shielding

by enclosing this trace as much as possible

by grounded copper,

so that if there’s any noise that comes in here,

it has very difficult for this noise

to induce a signal inside of this trace

and create some noise that then gets read out

by a receiving component.

So that’s kind of the idea behind shielding,

is basically to just block noise.

The reality is that shielding is kind of seen

as one of those things that you are just gonna have to do

on some components,

and that’s not really true.

If you do the layout correctly,

usually you don’t have to do things

like adding these shielding cans.

And there are variations of shielding cans

that you can actually implement on the enclosure,

not just on the PCB.

So there are different shielding materials

like meshes that you can put inside your device enclosure.

There are like shielding compounds or gaskets

that will absorb electromagnetic radiation

and will prevent it from entering or leaving the enclosure.

Those are other options.

All of that stuff is useful in different contexts,

but you should always start

by creating the board correctly first.

And if you do that,

then you’re gonna eliminate the need for shielding.

So when I say create the board correctly

or create the stack up correctly,

what exactly does that mean?

Well, let’s just look at an example here.

Let’s say that this is our noisy component

or our noisy circuit.

Let’s say it’s a power regulator.

And we’ve got another trace over here,

and this trace is receiving some noise from this circuit.

Well, there are a few things that you can do

to try and reduce how much noise couples from this circuit

over to this trace

that don’t involve putting any shielding around here.

So one option, if we don’t have this shielding,

is to reduce this coupling strength

between these two circuits, right?

So we’ll have some Cm here,

so that’s our mutual capacitance,

and then some Lm, that’s our mutual inductance.

So we’ve got two levers that we can pull

to try and eliminate the noise coupling

between these two circuits.

Those are, number one, to take the distance here,

so d, and increase it.

So spacing things out is always gonna reduce noise

every single time.

It will always reduce the noise

that gets coupled from here over to here.

So that’s your first lever you can pull,

is just space stuff out.

The other thing that you can do is

if you have a ground plane on the next layer,

which you usually should,

you can take this height,

so this distance between the trace and the ground plane,

and you can reduce it, okay?

So bring the ground closer to the trace

that’s being affected,

and that will also reduce these parasitic capacitance

and parasitic conductance values.

So just recently we did another video on parasitics,

where we actually looked at the graph

that shows what happens when you take this ground plane

and you move it closer to the circuit

that’s being affected by noise.

It actually reduces these coupling capacitances

and inductances,

and that’s gonna reduce the strength of noise

that couples from this circuit over to this circuit.

Now, let’s just suppose that the noise source over here

that you’re about to couple into this trace here

is at a specific frequency.

So maybe this, instead of a power regulator,

is some sort of analog circuit,

and the analog signal is harmonic,

so it’s operating at a single frequency.

So what you could do is you could put stitching vias

around this circuit.

So putting stitching vias around this trace is effective

if the vias are sized properly.

So don’t just pick some random value

like two-millimeter spacing or whatever

and expect that that’s always gonna work

for every frequency.

So we discussed in a previous video

the role of stitching vias in copper pour

and how they can actually contribute to resonances

that aid noise transfer between two circuits,

or how they suppress resonances

and prevent noise transfer between these two circuits.

And so they do that

by creating a resonant cavity between them.

So let’s just say for a moment that I’ve got two vias here

and they’re separated by some distance S.

Whenever S is equal to half a wavelength,

where this wavelength is the wavelength

inside of this particular structure

associated with the decay of the substrate,

then you will have a strong resonance

that builds up in this cavity,

and that strong resonance could aid noise transfer

into this interconnect,

which could then get read out by the receiving component.

So, what exactly is the wavelength

that we’re talking about here,

the half wavelength?

Well, if you remember your basic physics,

the speed of light inside of the substrate here

is just the speed of light in vacuum

divided by the refractive index or the square root of Dk.

And then that is equal to the wavelength

inside of this structure multiplied by the frequency.

So if we just substitute in,

Lambda equals 2S into this equation,

we get f times 2S is equal to C sub zero

over square root of Dk.

Solve for f.

I get f is equal to 2S square root of Dk in the denominator

divided by speed of light in vacuum, okay?

So you can look this up in any old textbook.

S is something that you chose

when you set up your stitching vias.

DK,

that’s just the dielectric constant

for your substrate material.

So this tells you the frequency

that is going to build up a strong resonance

between these two vias.

And so you would like to prevent that.

Because whenever you have a strong

electric and magnetic field built up inside this structure,

that strong field could then couple noise

into a nearby interconnect,

and that’s going to actually increase noise transfer

between these two circuits.

So if you space this out properly,

you will actually prevent all lower-level frequencies,

so frequencies less than this,

from building up these strong resonances

and then coupling noise into this circuit.

All right, last but not least, isolation.

So, what does isolation mean?

Well, let’s just look at an overhead view of a PCB,

just kind of as an example.

I could have, let’s say, a circuit here,

and I wanna prevent it from coupling any noise

into a nearby integrated circuit

and all of its traces and everything.

Isolation says that you essentially prevent the return path

associated with any of the traces

and any of the signals moving around in this circuit

from interfering with the return paths of,

let’s say, this circuit over here.

So this is essentially the counterpart of noise coupling

that I was referring to

when we were talking about shielding.

Now, generally, when people talk about isolation,

this is where you’re gonna see

that 30-year-old design guideline come up that says

you should start splitting up ground planes.

And I still, to this day, see new articles coming out

that say you should split up ground planes

in order to prevent isolation from,

let’s say, a signal over here

from creating a return path that then interferes

with a different signal over here.

It is true that you do wanna prevent the return paths

from appearing to superimpose on each other

and then create noise transfer between these circuits,

however, splitting up the ground plane

and doing something like this

is really not the right way to do it.

So the issue here is that you need to take note

of where your return paths exist

and prevent return paths from crossing each other.

So just as an example,

if I have a return path that exists here

and I have another return path that exists here

along a circuit,

it’s possible that

if there’s no ground plane

or any ground separating these,

that these signals could couple noise into each other

and you essentially have a form of crosstalk

that would then be read out

at whatever IOs you’re dealing with.

That’s essentially what could happen in this circuit.

And basically just changing the layout

is probably gonna fix this problem.

So maybe moving this over, moving this up,

and then re-engineering where you put the actual IO port

or whatever the receiving component is

that needs to interact with these two chips,

that is actually gonna solve this problem for you

without having to do anything

like splitting up a ground plane.

So pretty simple.

I’ve been told in the past that I talk a little too much

about not splitting up a ground plane.

And for most designs,

you don’t need to split up a ground plane,

especially, if you’re dealing

with the standard set of digital buses.

So let’s say you’re doing like a microcontroller board.

You need to access the SPI, I2C, UART,

whatever other standard digital buses are on there.

So if you’re dealing with that case,

save yourself a lot of headache,

go with a solid ground plane.

Don’t try and split it up,

and practice good layout practices

and learn how to track return paths,

and you’re gonna be in good shape,

you’re gonna prevent a lot of your noise problems.

Now, there are some specialty cases

where you actually might need to split up the ground plane,

because the frequencies involved are such that

that is the only way you’re gonna be able

to provide any kind of isolation.

So I’ll give you one great example

that I was just recently reminded about

from someone that works at another company,

which is precision DC measurements.

So with highly precise DC measurements,

again, in that case, the return paths are such

that you actually kind of have no choice,

you may have to split up the ground planes

in order to provide the isolation you need.

Some audio systems,

it’s actually desirable to do this

because they may have a digital section on the front end,

and that’s what the user interacts with,

and there’s gonna be the standard digital buses.

Those could then couple noise into your audio section.

Those audio frequencies are such

that they need to be confined from the digital stuff.

And one of the easy ways to do that

is by splitting a ground plane.

However, those are the exceptions,

those are not the rule.

Give yourself a reason to split the ground plane

if you think you need to do it.

Come up with a compelling reason and then do it.

Don’t split it first

and try and find a reason to use a full-ground plane.

All right, everybody, thanks for tuning in.

And hopefully this gives you some strategies

that you can think about

to deal with your own noise problems in your PCB,

should you ever find them.

Now, one thing I didn’t mention,

which I probably should have mentioned at the beginning,

is if you do have noise problems,

try and track down what the actual source is,

that’s going to give you a reference to try and figure out

how you can actually approach this

and what strategy is gonna be best

for your specific noise problem.

All right, thanks again, everybody.