You are on page 1of 72

Project Robert 02

Radiated Emission and Its Troubleshooting Process

https://emccompliance.co.uk
https://mach1design.co.uk

Mach One Design - Public


How to reduce Radiated Emission
of an LED Lamp
Methodology - Pre-compliance Radiated Test

Device Under Test -


* Rather than test it in an anechoic chamber, we test and x
Mains Powered LED Lamp - A the problem in a normal work space

gift from a very good friend * How do we know eventually the product will pass EMC
Test? - Compare with another product that passes EMC
test

fi
Let’s rst have a brief look - Test 1

Video cannot be played on


the downloadable PDF
fi
Let’s rst have a brief look - Test 2

Video cannot be played on


the downloadable PDF
fi
Troubleshooting EMC Test Set-up
Log Periodic Antenna - Siglent SSA 3032X-R
DUT Kent Electronics

Wood Table

1 meter
Troubleshooting EMC Test - Some details
Spectrum Analyzer - Tekbox EMCView
Siglent SSA 3032X-R Software

Can test both horizontal


and vertical. Previous
demos are vertical, this
DUT
one shows horizontal

Markings are important


for measurement
consistency
Troubleshooting EMC Test - Comparison Results
It’s important to note that we are only doing a comparative study here. We measured 1
metre distance with an un-characterised antenna (i.e. there’s no coe cient antenna
factor, no coe cient curve t for 1meter distance, etc). The limit doesn’t mean
anything meaningful here, it’s just a reference.

DUT
Benchmark Unit

Noise spikes such these are


caused by ambient noise. Note
we didn’t test the DUT in a
chamber.
ffi
fi
ff
Troubleshooting EMC Test - Cable Length Effect

Assumption: The cable of the


LED radiates.
If true, by shorting the cable, we
Video cannot be played on the downloadable PDF should be able to reduce the
noise/shift the resonant
frequency.

Mach One Design


Troubleshooting EMC Test - Cable Length Effect
Noise is prevalent in the
frequency range of 100 Mach One Design

and 200 MHz, shortening


the cable does bring
down the noise in the
frequency range between
30 MHz-50 MHz. Why is
that the case?
See this page.

Ambient Scan

Long Cable Scan

Short Cable Scan


Troubleshooting EMC Test - the Ferrite Approach

Choose the right material ferrite core is


important. In this case, we chose 61 material
from fair-rite. As it can be seen from the
impedance Table, between 100 MHz and 250
Mhz, the impedance is quite high. For this
test, we used a three-turn con guration to
increase the impedance, as the impedance is
proportional to the square of number of
turns.
Mach One Design
fi
Troubleshooting EMC Test - the Ferrite Approach

Mach One Design

Three-turn 61 ferrite core helped,


in the peak, we can see 25dB
reduction. But the noise level is
still quite high, >60dBμV.

Ambient Scan

Long Cable Scan

Short Cable Scan

Ferrite core 61 material 3 turn


Before we go into further details - Some theory
rst, what causes radiation?

A home appliance product demonstration

Di erential-mode current
Common-mode current
Typical applications, hand Surprisingly, human
drill, hair dryer, mixer, etc body has a capacitance

Class II double insulated


product that does’t have an
earth wire

L Dut

230Vac
N

Gnd/Earth
fi
ff
From the eld point of view
Essentially, energy travels from the source, to the load, via a transmission line. So
many mismatches here, so there will be many rounds of wave re ection. The more
wave re ections, the worse the EMC.

Radiation is through space. The space below shows


where the eld is. Lots of activities are going here. As
a result, you will see common-mode current and
voltage, see next slide.

L Dut

230Vac Main eld area, this area determines radiated


N emission performance
Gnd/Earth
fi
fl
fi
fi
fl
There’s only one rule to follow
RF noise only follows one rule, and that rule is physics. Power diodes only conduct when forward biased, but
this is NOT a rule for RF noise. From RF noise point of view, power diodes are simply a path with the
impedance determined by its intrinsic capacitance Cd.

Di erential-mode current
Common-mode current Cps
Cd
The load could be a noise
source, such as an electric
motor drive circuit

There will be common-mode


voltage, Vcm
L

Parasitic capacitance to
230Vac
earth
N

Gnd/Earth
ff
Common-mode current on the cable
From previous discussion and the tests we did, we know that:
1. Cable does have an impact on radiated emission.
2. Putting ferrite on the cable suppress the radiated noise.
Spectrum Analyzer - Tekbox EMCView
Plant for decoration
Siglent SSA 3032X-R Software

DUT RF Current Probe -


TBCP2-750
Common-mode current on the cable

We also need to benchmark the performance.

We also placed the probe in di erent parts of the cable to compare.


ff
Common-mode current on the cable
Since the noise is dominant below 300 MHz range, we will only measure the common-
mode current up to 300 MHz.
We are measuring CM
current here. So the
limit line here is NOT a
line indicating pass/fail.
We only care about the
comparative study in
this case.

We also swapped the colours of


the trace for the units later. In this
case, green is DUT, purple is
benchmark. In later slides,
colours are swapped.
Common-mode current on the cable
Below 30 MHz

DUT
Benchmark Unit
Common-mode current on the cable
30 - 300 MHz

DUT
Benchmark Unit
Common-mode current on the cable
Since the previous measurement only measures the DC side, we also need to extend
the AC line cable to measure the AC side noise.

Output DC

Extended AC line wires so


we can measure the
current

Input AC
Common-mode Voltage

See Appendix
Simple circuit diagram here

IC
Cable length ≈ 1.9m

PCB
PCB

A quarter-wave
=λf λ/4 perpendicular to a
Where is the speed of light 3×108 m/s, for re ecting plane is
λ/2
dipole antenna such as this, it resonates often Re ective plane
electrically the same as a
half-wave dipole.
strongly at λ/4, so that’s about 0.5m, we can
work out at which frequency it tends to radiate
the most, the answer is 40MHz (at 1.875 meter λ/4
cable). Back to slide 10
𝒗
fl
fl
𝑣
In Summary
Most of the time, radiations are via cables We all learnt this in high school, remember?

Cable as antenna

Noise source
In Summary
Radiation of course can also occur directly on the PCB. There could be
1. Noise source directly radiate into space
2. Noise source energise a PCB structure, as a result, the PCB work as an antenna

Remember this board? The switch


PCB
node radiates a lot!
Slot creates antenna. See Robert’s other videos
Deep Dive - Let’s see what is inside
Back side
Warning: Don’t repeat the
same, HV safety rst!!!!

Output DC

Diode

Front side - A transformer + An electrolytic capacitor (400V, 4.7μF)

Switcher

Input AC
fi
A very basic yback SMPS

Cable length ≈ 1.9m


IC

PCB

For more info about basic operation of a yback converter see


https://training.ti.com/understanding-basics- yback-converter
fl
fl
fl
Quick bench top troubleshooting
Plant for decoration
Oscilloscope - Rigol
DS7054
We use an open Transverse Spectrum Analyzer -
Electromagnetic Cell Siglent SSA 3032X-R
because it is convenient. You
can also use a current probe
as we demonstrated before
for trouble shooting. The key
is after each step of
modi cation, you should see
some noise reduction in your
Tekbox TEM CELL
measurement (TEM Cell/ DUT
TBTC1
Current probe/LISN/
Antenna).
fi
Recommended Reading

Although this paper talked about radiated


immunity. But thanks to Reciprocity theory,
it is also true for radiated emission.
Antennas work equally well as
transmitters or receivers, and speci cally
that an antenna's radiation and receiving
patterns are identical.
fi
Open TEM cell

You will need to test 3


orientations. Find the
worst radiation
1

2
3
Open TEM cell
Mach One Design

The fact that all


emission pro les are
similar indicate the
main radiation
mechanism is through
Ambient Scan the cables.
Orientation 1

Orientation 2

Orientation 3
fi
Step 1 - Optimise the DC link
The purpose is to reduce the
impedance in this loop area. To
contain more differential mode
Change 4.7 μF to 10 μF, also note noise, so less common mode
this cap sits directly on the PCB, noise

rather than bend the cap leads, this


is to reduce ESL caused by
connection.

IC

Original Cap, bend the lead to reduce the height of the cap
Step 1 - Optimise the DC link
Mach One Design

10 μF cap
Original 4.7 μF cap
Step 1 - Optimise the DC link
Put X type lm cap across
the DC link, since lm cap
has smaller ESR.

10 nF X cap

IC
Cable length ≈ 1.9m

10 µF cap
fi
fi
Step 1 - Optimise the DC link
Same X type often is used
directly across the AC line.

10 nF X cap

IC
Cable length ≈ 1.9m

10 µF cap

But neither of these xes had any impact on the noise pro le, in fact, they
might have made the EMC performance worse, why is that the case?
fi
fi
Step 1 - Optimise the DC link
Mach One Design

10 μF cap
Added 10nF X2 cap in DC link
Added 10nF X2 cap in AC lines

Performance is getting worse!


Step 1 - Optimise the DC link


The lm cap we used is a 10 nF part. According to the data sheet, it should give me
good attenuation in the frequency range between 10 and 15MHz. But it didn’t.

The layout of this X2 cap is


important, since we had to
solder this on the board, the
connection alone will introduce
about 10 nH inductance.

Rule of thumb, 20nH/inch for


extra length.
fi

Step 1 - Optimise the DC link

10nH ESL+10nH extra due to layout

10nH ESL

This is already the Inductive region of


a capacitor.

Step 2 - Shield the transformer

Shield
Shield is bonded to the VDC-(HV)
using a very short cable.
10 nF X cap

10 µF cap
IC
Cable length ≈ 1.9m

Sadly, no impact as far as we can see…


Step 3 - Adding Y-caps


Shield

10 nF X-cap

10 µF cap
IC
Cable length ≈ 1.9m

Our intention is to guide the common mode


current back to its source using this low
impedance path.

2×2.2 nF Y2 cap

Sadly, no impact as far as we can see…


Step 4 - Adding more caps on the


secondary side
Shield

IC
We are adding both electrolytic
and ceramic caps here.

2 MLCCs piggybacked. One 35V 330 μF


for 10-50MHz range, one for
100MHz range

Sadly, no impact as far as we can see…


It seems that we have hit a wall…


Mach One Design

It seems whatever we do, we are


stuck with this unchanging noise
pro le. 47dBμV is still too high!
Note we also had a 10dB transient
protection connected to the RF
input of the spectrum analyzer, so
whatever reading we had, we
needed to add 10dB.
fi
At this time, what is important is
We need to revisit the circuit board! C3

R1
C4
R2 HV
FB
HV
Vdd IC
FB
CS GND

C2 R3

I don’t understand the purpose of


C3. But from an EMC point of view, C3 R3 E cap

it seems a bad design. The anode Diode C2

of the diode is very noisy as it is a Recti er


IC

switch node, why did they connect C4

the C3? R2 R1
fi
Compared to a normal design
Normally, you will have something
like this
The design also has the following issues

• The layout is terrible.

• Especially the DC link cap negative connection, it goes through top and bottom layer
with thin and long trace, through hole vias.

• Safety design on this board is also terrible.


C3

• No snubber circuit, no Feedback R1


C4
R2 HV
FB
HV
Vdd IC
FB
CS GND

C2 R3

Huge inductance due to the poor layout. There’s really no impedance


path for the hot loop area.
Removing C3
Mach One Design

In fact, C3 is already broken, not


a surprise.

Remove cap C3

Where we were stuck


Remove the capacitor C3
The new starting point
Looks like we are in good path now, so let’s repeat Step 2. But we need a new reference.
We can now turn the pre-ampli er in the spectrum analyser on, and do a scan between
9kHz and 1GHz. So we can see noise pro le more clearly.
Mach One Design

This new reference level is now 57


dBμV

Ambient scan
New starting point, without C3

Frequency stops nows at 1 GHz


fi
fi
Step 2 - Adding Y-cap between the
transformer primary and secondary ‘ground’
Previously
added LV caps,
we keep them
here
In fact, this is the Game Changer.
See next page.

IC
Cable length ≈ 1.9m

Vishay Y2 cap 1nF


Step 2 - Adding Y-cap between the
transformer primary and secondary ‘ground’
Mach One Design

That’s what we expect! Two


important things to note here
1. The cap needs to be Y class rated
for safety reasons!
2. Due to the leakage current
requirement(also safety related),
one cannot use very large
capacitance value. Check the
leakage current requirement
according to your product
application!
Ambient Scan
Without C3
With the added 1nF Y-cap

Why this Y?
This Y cap is so important because, it provides a ‘virtual ground’ for the system, in the
same time, provides galvanic isolation as per safety requirement.

For
Foraaterribly
terriblydesigned
designedsystem
system
such
suchas
asthis,
this,this
thisYYcap
capprovides
providesaa
quasi-ground
quasi-groundplane.
plane.
IC

The previous failed attempts are like trying


to x the little pipe here. We won’t be able
to see any effect until we x the main
pipe!
In this case, the Y cap acts like the main
thick pipe, energy is like water here! Note,
it is energy, not current in this analogy!
Current never ows in the wire!
Now we should be able to see the effect
when we try to x the little pipe!
fi

fl
fi
fi

Step 3 - Adding a shield around the


Transformer and ‘ground’ the shield
Previously added
LV caps, we keep
Shield them here

IC
Cable length ≈ 1.9m

Vishay Y2 cap 1nF Shield needs to be bonded


to HVDC-

This connection is not ideal! Ideally, you would need a


very short connection to the Gnd plane and have
multiple ground points. But this is what we have and
we need to apply the most effective method based on
this very bad design.
Step 3 - Adding a shield around the
Transformer and ‘ground’ the shield
Mach One Design

That’s what we expect!


Adding a shield so that the leakage
magnetic eld of the transformer is
now contained. The shield needs to
be bonded to a low impedance point.
Unfortunately, we don’t have a good
low impedance path due to the poor
layout design. The least impedance
path in the current design is the
VDC- (HV). So we bonded the shield
Ambient Scan to this point.
Without C3
With the added 1nF Y-cap
Now adding a shield around transformer
fi

Step 4 - Adding Y-caps on the AC lines


Previously added LV
caps, we keep them
Shield here

IC
Cable length ≈ 1.9m

Vishay Y2 cap
Vishay Y2 cap 1nF

2.2 nF×2

This has limited impact as far as I can


see… See the result here.
Step 5 - Adding CMC on the AC line
Previously added LV
caps, we keep them
Würth 450μH Shield here

CMC

IC
Cable length ≈ 1.9m

Vishay Y2 cap 2.2 nF×2 Vishay Y2 cap 1nF

Again, ideally, we would like to do this! But we are limited


to what can do due to the layout constraint.
Previously
added LV caps,
we keep them
Shield here

IC
Cable length ≈ 1.9m

Vishay Y2 cap 1nF


Step 5 - Adding CMC on the AC line
Mach One Design

These are the noise spectrums need attention, but since their level is low
enough, we are going to leave them as they are.

These points are of no


concern as they are the
ambient noise, as trace 1
(yellow) also shows spikes in
these frequencies.

Ambient Scan
Without C3
With the AC Y caps
With the CMC choke on the AC line
Step 5 - Adding CMC on the AC line
Mach One Design

These are the noise spectrums need attention, but since their level is low
enough, we are going to leave them as they are.

These points are of no


concern as they are the
ambient noise, as trace 1
(yellow) also shows spikes in
these frequencies.

Ambient Scan
Without C3
With the AC Y caps
With the CMC choke on the AC line
Pack it up!
Of course we can do more things to suppress these noise, but we want to do the
minimum change that would allow us to still keep the original form factor. We know
the 330μF electrolytic cap added on the LV side is having very limited impact, so in
the nal version, we removed that cap and we were able to pack the circuit back to
its original box.
I will not be surprised if an off the shelf 2-stage lter
(as the one shown on the right)will do the job as
well. But such a part itself will be larger than this
plug box. In the world of EMC, the challenge is
never in buying a lter that does the job, but rather,
how to pass an EMC using minimum lter material.
Customers don’t buy your product because it’s EMC compliant, they buy your
product because of its performance. So the skills of excellent EMC engineering is
enough to be e ective, but not too much. That is our philosophy.
fi
fi
ff
fi
fi
Moment of truth - nal test
Using exactly the test set-up as shown here. Note that it is good practice that we marked
where we put our DUT and antenna, so the set-up error is minimised. With one
exception, since now we improved the EMC performance, we were able to remove the
attenuator on the RF input, this results in reduced noise oor as can be seen below:

DUT before the x


DUT after the x

Noise oor was reduced due to the fact that


the attenuator connected to the RF input of
the spectrum analyzer is removed.
fl
fi
fi
fi
fl
Comments on the nal test
We also did a comparison test between the DUT and the bench mark unit. Results are
shown below:

Note: We cannot reach a conclusion that both units performed the same, because the
small antenna we used is only e ective from 400 MHz to 1 GHz. Below 400 MHz, it has its
limitation picking up sensitive signals. We can, however, draw a conclusion saying that the
DUT’s noise up to 400 MHz range were reduced by 30dB at least!

DUT after the x


Bench mark unit
fi
ff
fi
So, in real life, what does it mean?

Video cannot be played on


the downloadable PDF
What have we learnt?

• Radiated emission, its source, noise propagation and its impact

• A bad design is the source of the problem

• With a wrong design, often all the right choices of x fail to work

• When it comes to EMC, understanding the fundamental is key

fi
Appendix
Common-mode Voltage Demo
Demonstration of common mode voltage
10× passive probe, with probe tip connected to Vdc 0V
(secondary side of the SMPS), gnd end oating. See
videos on next slide.

Mach One Design Mach One Design

fl
Videos cannot be played on the downloadable PDF

Before the plug is plugged in After the plug is plugged in


Common-mode voltage and current Common-mode voltage and current
before Y-cap is placed. after Y-cap is placed.
There’s another one…
Here, a very quick demonstration of ESD radiation. Sparks are generated by this
simple spark igniter. This kit can be home made. Check https://www.emcesd.com/
tt2004/tt050104.htm
Video cannot be played on the downloadable PDF
My homemade spark gap EMI source

Spark captured by iPhone in the dark


ESD generated noise
Mach One Design

Note, the noise spectrum is


way beyond 1 GHz. That’s
something to take with you
from this session!
Would a shield box work?
A shield such as this will never work, but such practice is
common in the eld. We will show you how to make an
e ective shield later.
ff
fi
Shielding - but this time, let’s ground the
shield
Shielding - but this time, let’s ground the
shield

Still no effect. Even


when we ‘ground’ the
shield. But don’t be
disappointed, we will
come back to this
point later.
Test set-up to measure VDC (HV), Vdc(LV) and IAC.
Vdc (LV)

VDC (HV)

IAC
Common mode current is reduced by a
factor of 10, that’s 20dB improvement.

You might also like