"Ville Voipio" <vvoipio@kosh.hut.fi> wrote in message
news:i3k4qt1zwhc.fsf@kosh.hut.fi...
> Max <mtj2@btopenworld.com> writes:
>
<snip>
> You are absolutely right with higher frequency components. For example,
> a 50 MHz clock producing 100 MHz may be a problem, if you are listening
> at the 100 MHz region. But in this specific case, the odd peaks under
> measurement are at 30 010 000, 30 030 000, ..., etc. It does not really
> matter whether there are even peaks at 30 020 000, etc.
>
> And -- actually quite unfortunately -- the EMC tests do not specify
> the limits by the designated use of a frequency band. So, even if
> the peaks are in the wrong place, the device will pass the tests.
>
> - Ville
>
> -- 
> Ville Voipio, Dr.Tech., M.Sc. (EE)

Dear Ville,

Many thanks for your feedback on this. I appreciate it is a difficult area
to make predictions, but you have reassured me somewhat that this is
unlikely to be a problem with regard to EMC regulations. I was focussing on
it since the lines from the PIC to the LEDs are 5" (12cm) long and carry
some of the highest currents in the design.

Can I please ask then: We also have digital 5V TTL square-wave lines between
ICs, the highest frequency of which is 6.2MHz worst case (01010101...). We
intend to keep track distances as short as possible. But how short do they
need to be to be likely to avoid EMC issues? Is there a way of calculating
this?

Many thanks,

Jim

"Steve at fivetrees" <steve@NOSPAMTAfivetrees.com> writes:

> Hmmm. I wasn't even really considering repeat rates; I was simply thinking
> in terms of a single sharp edge, parasitic capacitance and hence a current
> (and RFI) spike of indeterminate bandwidth - probably mostly related to
> track length.

This may be a problem. But while the single transition has a lot of energy
on all sorts of frequencies, the average power is low as the repetition
rate is low.

I don't have any idea of the actual structure of the PIC's output
pins. Usually the small micros have rather high-impedance (dozens
of ohms) and slowish output signals. In this case the round-trip
time of the signal is maybe 1.5 ns, and the pin is probably
slow compared to this. So, in this sense the cable is not a
transmission line and the load is close to the driver.

> I tend to be concerned about edge rates of *any* digital signal, no matter
> what the frequency. Perhaps I'm oversimplifying ;).

I do shear you concern. Modern digital families produce all sorts
of surprises with their fast edges. Not only RFI, but also problems
with slow edges, etc.

Anything apart from actual measurements is oversimplification in
EMC... But in this case the margin seems to be large enough.
 
- Ville

-- 
Ville Voipio, Dr.Tech., M.Sc. (EE)

Max <mtj2@btopenworld.com> writes:

> >of the LEDs is 10 kHz. The worst-case signal is a perfect square
> >wave, as there are only sharp odd harmonics.
> 
> I'm not clear why you say that's the "worst" case.
> 
> Anything other than a "perfect" square wave (i.e. 50% duty cycle,
> infinite dV/dt) will radiate even harmonics as well as odd.

True. But they do not increase the energy of the signal at all.
So, all the energy on the even harmonics is away from the odd
harmonics. Further, if there is jitter in the frequency, the
peaks will get flatter and lower. The EMC tests measure peaks,
not averages. (Even though, in this case, read on...)

In practice, the signal is not a perfect square wave (0101010101)
at 10 kHz (corresponding to 20 kHz update rate, not 10 kHz, my bad).
It is more a random bitstream (0010111011000101...), where the number
of transitions is smaller (half in the case of a truly random stream).
This reduces the amount of radiated power by half. There may also be 
considerable amount of jitter due to processor interrupt latencies.
This jitter is small, but possibly enough to smear the nthousandth
harmonic over its neighbour.

In practice, the possible duty cycle variations and jitter do not
change the measurement results much. This is due to the fact that
the peak spacing (10 kHz in my example) is narrower than the
spectral analyzer bandwidth (30 kHz, IIRC). So, it really does not
matter whether the power is on one or two or n peaks within the
detection band.

This is good news in the sense that the worst case is not much
worse than anything else. It is bad news in the sense that
you don't win anything be spreading the peaks in software.

> While the
> even harmonics are typically much lower in power, they can be awkward
> if they happen to fall in the wrong place in the spectrum (say within
> a medical-telemetry band).

You are absolutely right with higher frequency components. For example,
a 50 MHz clock producing 100 MHz may be a problem, if you are listening
at the 100 MHz region. But in this specific case, the odd peaks under
measurement are at 30 010 000, 30 030 000, ..., etc. It does not really
matter whether there are even peaks at 30 020 000, etc.

And -- actually quite unfortunately -- the EMC tests do not specify
the limits by the designated use of a frequency band. So, even if 
the peaks are in the wrong place, the device will pass the tests.

- Ville

-- 
Ville Voipio, Dr.Tech., M.Sc. (EE)

"Ville Voipio" <vvoipio@kosh.hut.fi> wrote in message
news:i3kllmermff.fsf@kosh.hut.fi...
> Just a rough guesstimate for the application. The update frequency
> of the LEDs is 10 kHz. The worst-case signal is a perfect square
> wave, as there are only sharp odd harmonics.
>
> The radiated emissions are not measured below 30 MHz. Thus
> the first significant peak is at 30 010 000 Hz (3001st harmonic).
> Its amplitude is 2xVcc/(3001 x pi). This is equivalent to a 1 mV
> sinusoid at the same frequency. Hardly a problem when fed to a
> cable whose length is <<0.1 wavelength.

Hmmm. I wasn't even really considering repeat rates; I was simply thinking
in terms of a single sharp edge, parasitic capacitance and hence a current
(and RFI) spike of indeterminate bandwidth - probably mostly related to
track length. The repetitive nature of the signal makes it worse, of course,
but to my way of looking at it the fundamental problem is the fast edge.

I tend to be concerned about edge rates of *any* digital signal, no matter
what the frequency. Perhaps I'm oversimplifying ;).

Steve
http://www.fivetrees.com
http://www.sfdesign.co.uk

On 06 Mar 2004 18:05:08 +0200, Ville Voipio wrote:

>Just a rough guesstimate for the application. The update frequency
>of the LEDs is 10 kHz. The worst-case signal is a perfect square
>wave, as there are only sharp odd harmonics.

I'm not clear why you say that's the "worst" case.

Anything other than a "perfect" square wave (i.e. 50% duty cycle,
infinite dV/dt) will radiate even harmonics as well as odd. While the
even harmonics are typically much lower in power, they can be awkward
if they happen to fall in the wrong place in the spectrum (say within
a medical-telemetry band).

-- 
  Max

"Steve at fivetrees" <steve@NOSPAMTAfivetrees.com> writes:

> "Ville Voipio" <vvoipio@kosh.hut.fi> wrote in message
> news:i3kbrna5g88.fsf@kosh.hut.fi...
> > Radiated emission problems come from high-frequency clocks (e.g.
> > a 50 MHz clock may be a problem at 150 MHz) and switcher power supplies.
> 
> Forgive me, but this is a bit of an oversimplification. It would be
> perfectly true if we were dealing with sine waves. Hence the OP's
> (well-placed) concern with limiting edge rates.

It is an oversimplification, true. Almost anything said on EMC
(and especially EMI) is an oversimplification, unfortunately.
(BTW, if you look at my numbers you quoted, they are not for a
sinusoid :)

Just a rough guesstimate for the application. The update frequency
of the LEDs is 10 kHz. The worst-case signal is a perfect square
wave, as there are only sharp odd harmonics.

The radiated emissions are not measured below 30 MHz. Thus
the first significant peak is at 30 010 000 Hz (3001st harmonic). 
Its amplitude is 2xVcc/(3001 x pi). This is equivalent to a 1 mV 
sinusoid at the same frequency. Hardly a problem when fed to a 
cable whose length is <<0.1 wavelength.

Without significant inductances it is difficult to think of
ways the low-frequency energy would be transformed into higher
frequency energy, and there is very little of that high-frequency
energy.

The situation is dramatically different with faster signals, such
as bus or clock signals. And by bad design (inadequate bypassing)
it is possible to make the clock signal creep into the LED cable.
This may be a problem, but it is unrelated to the LED signal
transition rate.

- Ville

-- 
Ville Voipio, Dr.Tech., M.Sc. (EE)

"Ville Voipio" <vvoipio@kosh.hut.fi> wrote in message
news:i3kbrna5g88.fsf@kosh.hut.fi...
> Radiated emission problems come from high-frequency clocks (e.g.
> a 50 MHz clock may be a problem at 150 MHz) and switcher power supplies.

Forgive me, but this is a bit of an oversimplification. It would be
perfectly true if we were dealing with sine waves. Hence the OP's
(well-placed) concern with limiting edge rates.

Steve
http://www.fivetrees.com
http://www.sfdesign.co.uk

"Jim" <jim@nospam.com> writes:

> This is my first consume product, and I'm getting paranoid about CE
> compliance requirements (required for sales in Europe - I'm based in UK).

Most probably you won't face any problems associated with the LEDs.
The frequencies are so low that EM emissions are small. The standard
does not specify any radiated emission limits in the low-megahertz
region. Only conducted emissions are measured, and your LEDs should
not affect them any way.

Radiated emission problems come from high-frequency clocks (e.g.
a 50 MHz clock may be a problem at 150 MHz) and switcher power supplies.

> The PIC lines are connected to the dumb front panel PCB via 5" of IDC wire.

If you are still worried about the emissions, use twisted-pair
IDC cable, where every signal is paired with GND. Twisted-pair
cables radiate significantly less than non-twisted.

> In practice then, these measures to limit slew-rate may well not be
> necessary, which would be great. Unfortunately I don't have any experience
> in the area to make that decision, so I'm thinking it is best to play it as
> safe as possible at this stage.

Then use ferrite/capacitor combination. Ferrite in series and
capacitor to ground. The ferrite prevents current spikes and
capacitors eat voltage spikes. You may either put a ferrite on
each line (SMD ferrites are small) or use a clamp-on ferrite on
the IDC. The latter is cheaper but less effective.

But before using time and resources to this, have a good look at
the signal shape with an oscilloscope. If there is no ringing,
then it is very unlikely you'll be radiating anything significant.

(Yes, EMC makes designers run in circles. On the wall, usually.
EMC labs are full of worried-looking engineers :)

- Ville

-- 
Ville Voipio, Dr.Tech., M.Sc. (EE)

On Sat, 6 Mar 2004 11:03:30 -0000, the renowned "Jim" <jim@nospam.com>
wrote:
>For anyone who is interested, I'm not driving the decimal place of each
>digit so I'm driving 3 x 7 segments with 8 PIC pins. All the other I/O pins
>are used for other jobs so it seemed the most cost-effective way to drive
>the LEDs (larger PICs are quite a bit more money percentage-wise, from our
>suppliers anyway).

Yes, for some reason. Maybe it's Microchip's marketing strategy at
work.  

>Spehro's suggestion would only require 6 lines for the same job (actually up
>to 30 segments) That would be even better, but would brightness suffer
>maybe, with only one LED on at once?

Look at it this way. Your limitation on average current is the *digit*
current. Say that is 15mA. You can thus drive 15mA/7 = 2.1mA/segment
or 714uA per segment average. If you drive only one LED at a time at
15mA you'll get an average current of 15mA/21 = 714uA per segment, so
the apparent brightness will be roughly the same. 

This current level is just sufficient for good displays such as this
one (plug plug, but you can't mux them this way) that we sell: 
http://www.trexon.com/leds/trexon_led_display.pdf , but you're only
getting a fraction of the potential brightness of course. Run them at
10-20mA average per segment and they are almost blinding by
comparison. 

Unfortunately, the brightness is pretty much fixed by the display,
given the PIC limitations. 

The big problem is that if you minimize it to 6 lines you'll not be
able to use standard LED digits. Unless your quantities are truly
humongous that is an unpleasant limitation, IMHO. Just getting a
sample display or two would be an expensive hassle.  

Best regards, 
Spehro Pefhany
-- 
"it's the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

"Spehro Pefhany" <speffSNIP@interlogDOTyou.knowwhat> wrote in message
news:qi4j40tbdahb2afag6v1lqctf7ui5ge1pd@4ax.com...
> On Sat, 06 Mar 2004 06:14:43 GMT, the renowned CBFalconer
> <cbfalconer@yahoo.com> wrote:
>
> >I've multiplexed displays since Nixies, but what in the devil is
> >"Charlieplexing"?
>
> He's got a link to an explanation. It's a variation on a general
> method of multiplexing that allows you to drive more LEDs on n + m
> lines than n  * m by driving port pins in both directions.
>
> In fact, if you're willing to drive only one LED at a time (which
> could be muxed fast so some or all appear to be on, of course), you
> can drive (k)*(k-1) LEDs with k port pins. Eg. 12 LEDs with only 4
> port pins. You just connect them both ways across all possible
> combinations of pins, and use firmware that tristates all but the pair
> of port pins that are actually driving current through a given LED.
>
> The method Maxim uses allows you to use conventional digits (but NOT
> standard multiplexed displays) and drives 8 8-segment digits (64 LEDs)
> with 9 lines (less than the 72 possible), allowing all the segments in
> one digit to be on at once.
>
> If you are driving at high current, it's a technique with some real
> limitations, and I've only occasionally used the technique in real
> products. My stuff tends to have analog things going on..  it also
> mucks up the ability to share display driving lines with keypad
> scanning.
>
> Best regards,
> Spehro Pefhany
> -- 
> "it's the network..."                          "The Journey is the reward"
> speff@interlog.com             Info for manufacturers:
http://www.trexon.com
> Embedded software/hardware/analog  Info for designers:
http://www.speff.com

For anyone who is interested, I'm not driving the decimal place of each
digit so I'm driving 3 x 7 segments with 8 PIC pins. All the other I/O pins
are used for other jobs so it seemed the most cost-effective way to drive
the LEDs (larger PICs are quite a bit more money percentage-wise, from our
suppliers anyway).

Spehro's suggestion would only require 6 lines for the same job (actually up
to 30 segments) That would be even better, but would brightness suffer
maybe, with only one LED on at once?

Jim