Connect with us

Reducing EMI

Discussion in 'Electronic Design' started by galapogos, Jul 5, 2007.

Scroll to continue with content
  1. galapogos

    galapogos Guest


    I'm designing a USB storage device that has to pass FCC/CE tests, and
    as such I'm concerned with the radiated/conducted emissions of this
    device. I've roughly checked out the radiated emissions of the device
    with a spectrum analyzer and a near field probe and I see some spikes
    at the harmonics of USB's frequency of 480MHz, notably at 240, 480,
    720 and 960MHz. I also see a few more spikes that are irrelevant to
    USB such as 250, 266MHz.

    I don't have much experience with RF engineering, but my circuit
    conforms to USB 2.0/ATA specs as far as SI goes, i.e the stackup,
    trace impedance, track spacing/width, trace lengths etc are all
    correct, along with other guidelines such as traces not crossing a gap
    in the power/gnd plane, that kinda stuff. I believe EMI suppression
    also has similar guidelines?

    I've tried switching a couple of USB cables, and some are
    significantly better than others, but none are perfect. I also tried
    added clamp type ferrite beads around the cables but contrary to what
    I thought, they didn't seem to do anything at all for EMI.
    filter in series to the D+/D- signals seems to be the way to go to
    reduce EMI while still maintaining USB signal integrity. Ferrite beads
    can also be added in series to Vcc/GND to act as a filter. My question
    is, are these chokes and beads a panacea for EMI? Or at least, will
    they significantly attenuate the EMI without causing any problems with
    the actual USB signals?

  2. Guest

    The important thing to keep in mind when thinking about radiated
    emissions is that radiation comes from current loops.

    In the crudest example, when a logic output changes state, the chip
    driving the output has to deliver a chunk of charge into the track
    carrying the signal to charge up its stray capacitance to ground along
    the trace, and at every input being driven by this signal along the
    track. For a 0-to-1 transition this means that current flows into the
    driving chip at the Vcc pin (or pins) and flows out of the ground
    plane and out of the Vdd pin (or pins) at each of the receiving chips.
    Ideally, the current flowing into the Vcc connection at the driver
    chip comes almost entirely out of the by-pass capacitor (or
    capacitors) at the driving chip (a ferrite bead can help here), so the
    current flowing out of the ground plane and the Vdd pins on the
    receiving chip eventually has to get back to the Vdd connection at the
    by-pass capacitor at the driving chip. If you look at the area
    included in these current loops, you have some idea of the size of the
    transmitting antenna that is radiating the consequent emission.

    If the radiating loop is sitting on top of a continuous ground plane,
    the current circulating through the signal track and back through the
    ground plane will induce a very nearly equal and opposite curent in
    the ground plane, but the loop within the ground plane has a certain
    resistance, so the compensation can't be perfect.

    Differential signal involve more or less equal currents moving in
    opposite directions, and if the tracks carrying the signals run close
    together, the radiating currents very nearly cancel, which helps a

    Differential signals running along twisted pairs radiate in opposite
    directions over every half twist in the two wires, which can further
    minimise the nett radiation.

    In that sense twisted pair cable is better than flat cable.

    Obviously, screened cable is better than unscreened cable, and cable
    which is only screened by wire braid - which only offers about 95%
    screening - isn't a good as cable screen by wire braid on top of of a
    wound layer of aluminised plastic (which offers 100% screening, but
    has a relatively high surface resistance).

    Hope this helps.
  3. galapogos

    galapogos Guest

    Thanks for the theory. I'll have to read it over a few more times to
    wrap my head around it.

    AFAIK, all the signals are run over a continuous ground plane, and the
    USB signals are routed parallel to each other. I have followed USB
    design guidelines as far as I can, EXCEPT the EMI/ESD portion, which
    recommends a common mode choke. That's why I'm wondering if that would

    Since this is a storage device, there's also an IDE/ATA component,
    which is translated to and from USB. As such, I'm wondering if the ATA
    portion also requires ferride beads for the power lines and common
    mode chokes for the signal lines. From my research only serial signals
    need these shokes(such as SATA/HDMI). I haven't found any for PATA,
    and I already have the necessary terminating resistors for PATA, so I
    guess that should also take care of EMI issues on top of SI issues...
  4. MooseFET

    MooseFET Guest

    Check the length of the wire or trace that connects the ground of the
    USB cable. The ideal is a very low impedance on this line.

    Is the box metal? To radiate, you need either a current flowing in a
    loop or a voltage difference between two exposed surfaces. Connecting
    the housing to the ground of the USB via the shortest thickest path
    you can helps on the electrostatic issues.

    If the box has other cables, such as a power line, filter the heck out
    of those connections. Ideally, those connections should be right next
    to the USB connection so that the path for grounds can be near zero
    length. You want to have the last item in the filter on them be the
    lossy inductor so that any RF energy trying to get out on that path
    gets eaten up.

    A common mistake is to allow the return currents of your circuits to
    flow through the case or out one cable and back in another.

    Also: It is better not to make the RF in the first place. Take a
    look at your design and see if you can source terminate clock signals,
    slow down edges, reduce how many line carry the fast signal and etc.
  5. galapogos

    galapogos Guest

    Th USB connector ground on my device is directly connected to the GND
    plane with a via, so that's the shortest length possible.
    The box is partiallly metal(about 40-50% steel, 50-60% plastic). The
    entire box is not grounded nor are they connected to any part of the
    circuit, but I will keep this in mind. Unfortunately in my current
    design the platform that the PCB is mounted on is plastic, so it would
    be hard to connect the metallic part to the ground plane.
    Sorry, I don't get what you mean by the last sentence. The box does
    have another FPC with a low current power line that is just supplied
    by Vcc. By filtering do you mean placing a ferrite bead in series to
    Also, this is a 2 PCB design that is connected by pin headers, with
    Vusb and gnd flowing between the 2 PCBs.
    Since the case isn't connected, I don't think this is possible? As for
    the cable part, not sure if I'm doing that.
  6. Just the one via?

  7. MooseFET

    MooseFET Guest


    Imagine the cables laying like this:

    PC =============! BOX !==================== Power supply

    See how that looks like a dipole antenna. If you put something lossy
    at the power supply connection of the box, it will eat up some of the
    RF before it gets out to the cable.
    How do the signals go with respect to the grounds. You want the
    grounds to wrap all around everything that can radiate. Using extra
    ground pins on the header helps.

    Chances are you are to some degree. It is hard to avoid. At over
    10MHz, the details of things like power supplied stop mattering. They
    are just big lumpy things that provide a path to the mains wires. The
    same is true of a PC. You have a loop fronm the box to the mains
    through the mains to the PC and back via the USB cable. You want
    losses somewhere in that loop.
  8. galapogos

    galapogos Guest

    Oh, you mean external cables...I thought you meant internally. The FPC
    cable is an internal cable and not to/from a power supply. The box is
    powered by the PCB via the USB cables, but there is also an external
    DC power supply jack. I have thought of filtering that with a ferrite
    bead. I will probably do that with the next PCB revision. However, as
    of now the EMI results of the box with or without the power supply
    connected isn't very different, and the spikes at the USB harmonic
    frequencies can't be caused by the external power supply can it? In
    any case they are still there when it is disconnected.
    There are a few ground pins(8 of them) on the header, since this is a
    standard 44pin PATA header.
    OK, that's assuming I'm using the external AC adapter to the mains
    right? Without it, there wouldn't be a loop since there's only a
    single point of contact to the mains through the PC via the USB cable?
    And assuming there's a loop due to the AC adapter, the way to induce
    losses in the loop is via ferrite beads?
  9. galapogos

    galapogos Guest

    another via from 1 of the 4 legs of the USB shield.
  10. MooseFET

    MooseFET Guest

    Yes you want to put a ferrite over all of the power wiring of the
    power jack. This includes the "ground" connection of that connector.
    The reason those clamp on the cable things are so common is because
    people forget to do this inside the box.

    Without the power cable plugged in, you have a much smaller dipole.
    If you can, you'd still want to get some losses into the picture.

    RF is a little like light. If you block most paths with something
    refective, it will bounce around until it finds a way out.
  11. Guest

    The spikes at the USB harmonic frequencies do get into the power
    supply leads. Most of the current involved in the switching spikes
    circulates through the by-pass capacitors up against the chips on the
    board, but these do have appreciable impedances at 200MHz - take a
    look at the manufacturers data on their impedance as a function of
    frequency, and you can see voltage spikes (usually called "grass") on
    the Vcc lines anywhere on the printed circuit card. This generates
    apprecialbe high frequency current in all the power supply leads.
    Ferrite beads in series with the power supply leads can reduce - but
    not eliminate these currents.
    Presumably the power lead in the USB cable is also radiating the same
    Star grounding doesn't work at RF frequencies. Connections that look
    like a short circuit at low frequencies look like inductors at RF.
  12. galapogos

    galapogos Guest

    Gotcha, one in series on the Vcc, and one in series on the GND.
  13. galapogos

    galapogos Guest

    So I tried adding some ferrite beads...1 in series with the USB power,
    another in series with the DC jack power...couldn't find any pads to
    fix them on the grounds nor could I use any common mode chokes on the
    USB signals coz I don't have the footprints for them either. Turns out
    the ferrite beads did jack for EMI suppression. I'm still getting
    pretty much the same graphs, but I did manage to find a better cable
    with an integrated ferrite bead. Weird since the clamp ferrite beads
    did nothing for my other cables, even one brand that is listed on
    intel's USB design guide.

    Anyway, with the new cable I'm able to reduce the EMI problem range to
    the 200-500MHz range, with another peak at 800MHz. The rest are within seems that at least some of the EMI is radiating out from
    the unit itself rather than from the cable. I've got some EMI absorber
    mats that I just placed over the unit(didn't peel and stick em), but
    they seem to do absolutely nothing when measured with the near field
    probes that I have.

    I also did another mini experiment. On my older design there's a metal
    plate that sorta covers almost the entire PCB area, but it isn't
    grounded. I tried to ground it by touching it with the USB
    shield(which is ground), but strangely, the EMI peaks went UP instead
    of down! Aren't grounded shieldeds supposed to make things better? Man
    this EMI suppression is turning into more of a black art than
    science... :(
  14. Guest

    You should know that incomplete shielding can be as bad as no
    shielding at all. Check out "slot antennas"

    A narrow slot or or slit in the shielding that is half a wavelength
    long acts as a dipole antenna. The wavelenth of you 800MHz component
    is 37.5cm in air (15 inches) so a 19cm (7.4 inch) slot would be a very
    efficient antenna.

    Standard shielding completely encloses the printed circuit board. If
    you want to put a metal box on top of a multi-layer printed circuit
    board including a ground plane (which can work just as well), you need
    to couple the box to the ground plane every cm or so all around the
    periphery of the box. A sort of fringed skirt with lots of beryllium
    copper spring fingers is one way of doing this.

    Howard S. Johnson and Martin Graham wrote a book "High Speed Digital
    Design: A Handbook of Black Magic" ISBN 0-13-395724-1 on the joys
    designing printed circuit boards for GHz logic. The index doesn't say
    anything about slot antennas, though they do have a decent sized
    section on cross-talk and do mention "ground slots".
  15. joseph2k

    joseph2k Guest

    Never forget, lead length makes a hugh difference.
  16. Guest

    Ferrite beads aren't identical - Farnell have a page of or two of
    ferrite beads and chips and the impedances they present at 100MHz
    range from 60 ohms to about 1000 ohms.

    And a ferrite bead just puts an series impedance into the connection.
    To reduce the high frequency current travelling around that loop, you
    also need a parallel impedance to local ground - a high frequency
    capacitor - to divert most of the high frequency current around a more
    compact loop.
    One thing about shields tht you may need to know is that any extended
    slit in you shielding acts as a slot antenna.

    A narrow slot that is half a wavelength long (even when quite narrow)
    functions as an efficient dipole antenna.

    Even at 800MHz half a wavelenth is about 19cm or about 7.5 inches, but
    shorter slots can still be bad news.

    Openable screening boxes usually come with a skirt of berylium copper
    fingers that make contact every few millimetres around the openable

    Speaking of black magic - Howard W Johnson and Martin Graham wrote a
    book "High Speed Digital Design: a handbook of Black Magic" ISBN
    0-13-395724-1 which does talk about cross-talk and shielding, though
    it doesn't mention slot antennas.
Ask a Question
Want to reply to this thread or ask your own question?
You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.
Electronics Point Logo
Continue to site
Quote of the day