See answers inline Am Sa 20. September 2008 schrieb Werner Almesberger: > Joerg Reisenweber wrote: > > See placement of R4408. I think that's enough of a good story. > > Hmm, I'm beginning to understand your MICBIAS theory. Both C3014 > and C3015 invitingly offer their MICBIAS end to any RF coming out > of the R4401 and R4408 pair. > > Luckily, the other capacitors in the area only show their GND end > to this probable RF source. > > No amount of component removal besides JK4401 could fix that, so > that's consistent with experiment results. > > A more specific test for MICBIAS contamination would be to remove > R4305 and to bias the internal microphone from an external source > (through a shielded cable ;-), e.g., across D4301, which is easily > accessible. Has something like this been tried ?
YUP, long ago. Result: fix noise. > > It's a pity that you don't have a proper RF generator. All those > tests with the GSM module or an external phone as the RF source > just introduce so many parameters completely outside your control > that it's almost impossible to get meaningful results. YUP. Also note that a proper signal-generator at +33~+39dB, without auxiliary stuff like faraday-room, very likely will lead to me unintentionally quitting my job, as soon as I start to operate it here in Germany. ;-) Maybe you even could visit me at Guantanamo then, which is more close to Argentina than Germany is. :D > I wonder how hard it could be to build a little 1W RF generator > that's tunable in the 1-2GHz range. See above. > > Anyway, I disgress. What really confuses me is that we still don't > seem to have an explanation for why shorting MIC to GND doesn't > seem to help. Or do we ? Or does it in the end ? :-) GND actually is a part of the antenna-dipole. So maybe GND "is not good" there. Another explanation is around standing wave (which has voltage and current maxima), and impedance of source vs impedance of short-to-GND component + impedance of GND itself. But I agree it's a little strange, at least we would expect different result from our gut-feeling and common experience on similar situations > By the way, there are two goals here: > > 1) Make sure we don't repeat the same mistake in the future. I > agree that you're perfectly right with just keeping RF out of > the can. > > 2) Find a work-around for the GTA02 units already in the field. My efforts are targeted on both, but I don't see the amount of progress and support I'd expect, on either of them. For 1) I just won't stop to repeat to quote my own msg from April: > Basic guidelines: > ================= > If we go for a A7 PCB (please correct if in the following anything is > missing or wrong): > placement (layout) of *all* audio components has to be reconsidered! > Separate shielding cage for audio section SHOULD be implemented. > *All* copper rails *) creating a path from outside to inside of the audio > cage MUST have EMI-filters *close* to the point they cross the cage border > (usually those EMI-filters form 'gates' from outside to inside, by being > mounted exactly ON the cage border line in special breakouts of the cage > metal). > There MUST NOT be any nonlinear component (=semiconductor) of audio > circuitry outside the cage. This does not apply for specially EMI-hardened > tested components like microphone, and for high-Z OVP-components. > There SHOULD NOT be any component of audio circuitry outside the cage, > except OVP. > All audio copper rails SHOULD be inner layer of PCB, shielded on the outer > layers by gapfree GND (or VDD) rails, and MUST NOT run in parallel to any > high-current rail. > All rails outside cage SHOULD be differential, high level (microphone!), > low impedance. > Differential lines MUST run in parallel on same layer. > Layout of GND and supply-planes SHALL follow design guides of Wolfson (URL: > tbd) > All connectors to outside of housing SHOULD have EMI-filters [[edit: as > close as possible to component]] on all lines > except GND, because external cables are feeding remarkably high levels of > RF to the PCB. > > *) that's one of the points why it's more reasonable to have a separate > cage. We don't want the need for EMI-block on lines that aren't related to > audio, but run to same cage. Also other (digital) circuitry may create > noise *inside* the cage. cheers jOERG
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ hardware mailing list [email protected] http://lists.openmoko.org/mailman/listinfo/hardware
