|
Posted by glen herrmannsfeldt on June 28, 2008, 3:57 pm
If you were Registered and logged in, you could reply and use other advanced thread options
Trebor Mushroom wrote:
> glen herrmannsfeldt wrote:
> <snip>
>>> With regard to my original question, I dig deeper and have found
>>> some problems with MB layout. Basically the network adapter
>>> (Intel btw) is completely integrated in the chipset, with just four
>>> lines on the outside (+ 3 lines to control LEDs). On the motherboard
>>> there is only integrated RJ45 module jack (with transformer & LEDs).
>> It shouldn't be that hard to get right before the transformer.
>> It is a little harder to keep the signals balanced after the
>> transformer through long PCB traces.
> I'm not sure what do you mean by "keeping the signals balanced after
> the transformer" - firstly there are virtually no tracks _after_ the
> transformer, secondly isn't the goal here keeping the tracks equispaced
> and short as possible between the chip set and the transformer to
> to avoid crosstalk and pick-up of unwanted interference? (see below)
Minimizing radiation (RFI) depends on the signals being balanced.
The currents are opposite in the two wires of the pair, such that
they pretty much cancel out as far as a radiated signal.
If you run them across the PC board after the transformer, you
have to be extra careful to keep them balanced. Stray capacitance
differences could easily change that. That effect is not
there before the transformer.
>>> The design violates the rules regarding the length and spacing
>>> of the routes. I still have to find what exactly is wrong with
>>> it but probably too long tracks increase loses to unacceptable
>>> level and it doesn't work with some adapters.
That should only be true for long cables, approaching 100m.
>> It has to last through 100m of cable, so a few inches of PCB
>> trace normally won't have that much loss. It gets somewhat
>> harder for gigabit, but you didn't say that. It is nice to
>> impedance match, though, in any case.
> Same as above, I'm talking here about the signals between the PHY
> in the IC and the transformer, which are normal (i.e. voltage level)
> fast digital signals with all associated problems. After the
> transformer (as you rightly pointed out) it doesn't matter because
> they became differential pair (current) lines up to 100 meters long,
> inherently prone to the problems mentioned above.
> The engineer who laid out the MB was thinking similarly to you
> because all he did was terminating the lines with resistors.
> As it shows :-( it is not enough under some circumstances.
So, you don't have to be quite as careful before the transformer,
but they are still fast digital signals. Normal precautions
still apply.
> Re-designig the MB is not an option for us, so we will have to
> live with it I'm afraid. Given some time I will try to probe the
> lines to learn what exactly goes wrong when using crossover cable
> as compared to connecting it to the switch.
I still don't see how it could depend on crossover.
I could easily believe cable length, though.
Another that I thought of, some use a transformer that
isn't 1:1. The voltage (and impedance) is different
across the transformer. If that was wrong, you could
get unexpected results. Otherwise, it is hard to guess
at all the possible reasons.
-- glen
|