Grounding

John,

I notice a lot of orange patch cords in your o/e and e/o stuff (nice looking, btw). How do you handle the residual dispersion of your multimode fibres over those distances? I'd have thought that few ps/m would blow your jitter spec out of the water.

Cheers,

Phil Hobbs

Hello Jim,

!

I know, that's why I placed a smiley. On a chip you have no other choice because you can have a lower impedance to board ground than whatever a few bond wires will give you.

However, when I read data sheets and split ground layout recommendations in there I often develop goose pimples. On a few occasions I called in and talked with the authors, usually after something had gone terribly wrong with the layout at a new client. To my surprise the authors were often the chip designers themselves. When I explained the "unwanted antenna" effect I mentioned to Dimiter here in the thread to them you could almost hear a light bulb come on at the other end.

--
Regards, Joerg

http://www.analogconsultants.com

pins !

;~)

Even fat C4 (flip chip) bonds can be a problem.

Yes, though I have seen some ECL circuits (frequency synthesizers)) where the App Notes suggest a cutout under the chip so there is no ground flowing under the chip to the reference ground.

--
  Keith

Try the electrical side of a high-sensitivity EO system--those are 160 dB CNR in 1Hz sometimes, not to mention having to cope with big changes in carrier power. The signal power is quadratic in the optical power, which can be painful.

Cheers,

Phil Hobbs

Hello Jim,

Oh boy, them thar words might send the chairs and beer glasses flying if we were all in a pub...

--
Regards, Joerg

http://www.analogconsultants.com

I can't recall when I have seen a simple question (should I split groundplanes?) lead to such a fearless display of ignorance.

Now I only have to figure out which side is the ignorant one.

It's pretty simple; just pick one and do it. If you have to hire Joerg or Jim to get your ass out of a crack, you chose poorly. ;-)

--
  Keith

I don't know, seems like pins 11(DGND) and 12 (AGND) are tied together right outside the chip

Hello Phil,

Audio does not always mean you won't have to worry about anything above

25kHz. The local AM transmitter can inject enough energy that it gets rectified somewhere and you hear Rush Limbaugh coming out of the mixer. The mechanism can be as innocent as a transistor bias being wiggled a bit.

The most funny experience I had was at a client where a new Doppler receiver design had EMI issues. So I fired it up and heard some telltale squirling. Sounded like SSB transmission with the frequency setting way out of tune. So to make them believe it I asked the SW guys how I can directly write to the register that controls the PLL and then I tuned it in. We could now hear BBC World Service news in remarkable quality (actually sad news about the war in former Yugoslavia).

--
Regards, Joerg

http://www.analogconsultants.com

That'll be where they chose to put the 'zerohm link' An amusing choice

Note the 2 'ground planes' on layer 1 though.

Graham

Aaah! Aaaah! AAAAAAUUUUGGGGHHHH! (I see the problem.)

Yeah, it's hard to control what comes in on an antenna. In the optics business I can usually be sure that what I'm seeing was what was on the laser beam. Not that that's particularly nice, sometimes.... e.g. He-Ne lasers exhibit baseband spurs of ~0.1% that sweep back and forth between about 10 kHz and 1 MHz. They're caused by fourth-order mixing products--one optical longitudinal mode beating with a 3rd order intermodulation product of two others. The frequency offset is only a few parts in 10**9, caused by dispersion in the plasma and small amounts of mirror misalignment. When the offset gets too small, the modes jump into lock and the spur goes away for awhile.

I also once had an argon laser whose residual AM noise would go up by 20 dB if you blocked half the beam--there was a spatial side-mode that wasn't actually oscillating, but had enough power in it to cause the beam to wobble back and forth.

John L. has been giving us VCSEL horror stories lately too.

Cheers,

Phil Hobbs

Hello Joerg,

I agree with most if not all of what you say in this message.

Of course you have to. There are various ways to do it: large enough board, more GND layers, thicker copper for the GND plane, perhaps some others. Another option, wherever applicable, is splitting the ground plane thus make its resistance in the area of interest low enough for the currents which flow through it. It is probably the least universal and the trickiest to do of all, but it can be done and it has been done successfully.

This is true, however the ADC is by far not the most noise susceptible part (typ. several V full scale input range), and the ground plane is not split exactly beneath it (that's where the two planes are connected).

Again, this is highly case specific and if one does not feel he knows where to split the grounds - if at all - the better option would indeed be not to do it.

"Follow the currents".... :-)

Dimiter

------------------------------------------------------ Dimiter Popoff Transgalactic Instruments

------------------------------------------------------

Joerg wrote:

In message , dated Tue,

8 Aug 2006, Joerg writes

There used to be chips, like TBA810, with heat-dissipating fins emerging from the package (not glued on to it). Those would make low-inductance lead-outs.

--
OOO - Own Opinions Only. Try www.jmwa.demon.co.uk and www.isce.org.uk
2006 is YMMVI- Your mileage may vary immensely.

John Woodgate, J M Woodgate and Associates, Rayleigh, Essex UK

In message , dated Tue, 8 Aug 2006, John Larkin writes

A popular misconception. 3+ decades of bandwidth, linearity error 10^-3 or better, signal-to-noise ratio approaching the theoretical limit.

--
OOO - Own Opinions Only. Try www.jmwa.demon.co.uk and www.isce.org.uk
2006 is YMMVI- Your mileage may vary immensely.

John Woodgate, J M Woodgate and Associates, Rayleigh, Essex UK

Nah! That stuff is for "Audio Weenies"!

--
Service to my country? Been there, Done that, and I\'ve got my DD214 to
prove it.
Member of DAV #85.

Michael A. Terrell
Central Florida

In message , dated Tue, 8 Aug 2006, Richard Henry writes

Neither. The reason you get two methods is that neither is fundamental. This happens in other situations, too.

The fundamental requirement is to keep the digital currents, including ground currents, out of the analogue circuits and vice versa. If you do that, ***there is no current in the ground interface between the analogue and digital circuits***. So it doesn't matter whether they are joined just at one place or along a line. If there is no current in that bit of copper, it need not be there.

Joining at one point MAY mean that it's easier to be sure you have the currents segregated. But it may not work; you can goof elsewhere, such as at power supply connections. Equally, you can get into trouble with a continuous ground-plane, because you can't see where the currents are going if they have two dimensions to play around in.

--
OOO - Own Opinions Only. Try www.jmwa.demon.co.uk and www.isce.org.uk
2006 is YMMVI- Your mileage may vary immensely.

John Woodgate, J M Woodgate and Associates, Rayleigh, Essex UK

I do in fact sometimes link the 2 grounds with say 10 ohms since, as you say, there should be no net current flowing there.

This is also a good place to ensure you get a nice solid ground connection from anywhere else. Like from chassis for example if there are no other connections of that type elsewhere.

Graham

And cryogenically-conditioned power cords and $100 amplifier vibration isolators and so forth...

I applied for work at one of the local audiophool companies. I was really hoping to get an interview so I could see what they thought of themselves from the inside. But I didn't get a call. Maybe my resume has too much engineering in it.

pins !

The silly little chip-scale packages, with the power pad on the bottom, have super-low impedances to ground.

We're using this as a laser modulator. 15 dB gain and 20 GHz bandwidth is pretty extreme, but it works fine on an FR-4 board. The power pad is the only ground pin.

John