Roland DeLuca said:
Thanks Paul
I will try building it. Here is the link if others are interested
http://www.reed-electronics.com/ednmag/archives/1998/060498/12DF_04.pdf
As far as putting everything in a metal box, I tried it with limited success
. I think my problem is that the RF is conducted in the wires and gets in
that way. I could add ferrite beads/ caps to all lines and better shielding
but it gets costly. So I want to try and find the worst offenders first.
Roland DeLuca
Hi Roland,
In practice it is very difficult to measure such RF interference (be it self
inflicted or otherwise). As soon as you attach a dangly wire (read as:
aerial), you will change the behaviour of your circuit, often making it
worse.
In general you would be better off to read a book or two on EMI suppression
techniques, then implement all the easy-to-do solutions. Henry Ott's book is
excellent, but I prefer these two:
Controlling radiated emissions by design, Michel Mardiguan, and
Controlling conducted emissions by design, John Fluke jr (all this book
really says is include all of the parasitics in your design, often easier
said than done
Step 1 is generally to throw out your pcb layout, and do it again with a
solid 0V plane - absolutely no slots! this is usually quite achievable on 2
layers - I once had a 1500W PFC pcb re-designed in this way. we deleted the
existing 0V traces, which freed up lots of room on the pcb. we then moved
every bottom-layer trace to the top layer that we could - about 90% of them
were simply layer changes! Then we poured an 0V plane, and spent about 4hrs
reducing the size of the inevitable slots (where there are either traces on
the bottom layer, OR adjacent vias which form a slot (these just get moved
apart a bit, to keep the slots small, say 1cm max). The entire re-layout
took our PADS guy 1 day, and the PFC worked beautifully afterwards (it used
to explode), all without changing a single component!
This is usually enough to solve your problem completely, all without
measuring a damn thing.
Step 2 is then to band-limit all input and output signals, right at the
connectors - capacitors are the go here. If a signal is low-speed, then
filter the shit out of it - dont provide more bandwidth than you have to.
Then you can look at cables - twisted pair for H-field attenuation, an
electrostatic screen for E-field attenuation, etc.
Then start thinking about ferrites etc, which are usually used to try and
compensate for lousy pcb layout - invariably it is far cheaper to just re-do
the pcb layout.
Hell, often a 4-layer PCB is a cheaper solution than wanky cables, ferrites
etc - I did an EMI analysis on a bearing temperature monitor for a 400kW
rock crusher (AC motor drives generate metric shitloads of EMI, they werent
using $80/m screened cables etc) a while back. The controller was a 2-layer
pcb with a long, skinny 0V trace, and it sucked majorly. The component
routing was sufficiently complex that a 2-layer pcb was unlikely to work
well, and would take a bloody long time to do. We deleted the 0V and Vcc
traces, added 2 internal layers and made mid-layer1 = 0V plane, mid-layer2 =
Vcc plane - again, only took an hour or two. It added about $10 to the cost
of the controller (2x original pcb price) but instantly solved dozens of
weird intermittent problems. Again, no measurements were taken. It was also
about $10,000 cheaper than re-wiring the ac motor drives with appropriate
cabling to reduce the noise source. Needless to say, the customer was happy.
You are dead right about the cables - internal radiated noise invariably
enters your cables, shoots out the box as conducted EMI, then radiates away
again - this is why the filters etc. should be placed as close as possible
to the edge of the box. Often when you place a ferrite on a cable, it turns
conducted emi into radiated emi, which leaps around the "filter" and back
into the cable - especially if your cable then runs across your pcb.....
cheers
Terry
