Signal Ground VS Chassis Ground - Should they be coupled?

[JDW]

Here's a more specific case that perhaps some of you could offer your detailed thoughts about...

I would like to present the case of a motherboard taken from a 1987 Apple Macintosh SE computer.  The chassis of that computer is metal, and is earth ground when a 3-prong power cord is used.  There's a metal bar atop the motherboard which makes contact with the metal chassis, to provide chassis ground to the connector portion of the motherboard.  The motherboard gets it's primary ground via the main wiring harness that attaches between the motherboard and the analog board (which drives the CRT and which leads to the power supply). 



I have confirmed from observing the motherboard Gerber files ground plane is divided into two parts, and the parts are tied together by a single trace on the top side of the board.  That single trace is shown in red in the photo above and the Gerber file below.



I'm not 100% sure why it is designed like that, unless it anticipates a dead short to ground in the chassis or external connectors, such that the trace shown in red above would act as a fuse and blow (split apart) when a short occurs. 

Any thoughts?

[David Hess]

I'm not 100% sure why it is designed like that, unless it anticipates a dead short to ground in the chassis or external connectors, such that the trace shown in red above would act as a fuse and blow (split apart) when a short occurs.

Apple was not the only one to use a single trace like that, and I have seen such traces blown.  Sometimes those traces are intended as fuses, but another reason to do that is to prevent ground currents between the external chassis connections to stop the chassis from radiating EMI.  A single connection to the external chassis will prevent this, but then what do you do when you have multiple shielded connectors connecting to the chassis?

[T3sl4co1l]

They made it worse, but it didn't much matter because the enclosure is metallized IIRC.

The effect is that the main area of the board, pushes itself around by the reaction force from all those damned traces crossing the gap.

EMI is higher inside the enclosure.  But only within logic levels give or take, so extreme coupling would be needed to corrupt logic levels elsewhere in the circuit.  It has relative immunity from itself, which is fine -- it's functional.

This would be a very bad idea if there were connections on the front, or extender cords to external drives (through the front slot/bays), etc.  Such connections would need significant mitigations applied (heavy filtering, tons of data line CMCs, isolation).  But such cases are not, I think, part of any normal build of that machine, so the fact that the enclosure is shielded, should be enough to deal with EMC.

As for the connector side of the board: the slot does avoid PS currents flowing across it.  It obviously doesn't prevent signal currents running around it (along the slot, necking down through the one lone linking trace), and it degrades the signal quality of the traces crossing it (which I guess the big DIP there is a SCSI controller? ref: https://upload.wikimedia.org/wikipedia/commons/6/65/Apple_Macintosh_SE_Main_PCB.jpg ).

Probably it doesn't matter much, because with so many chips, and trace length -- and ribbon cables and so on -- it may never pass EMC with a plastic (unshielded) enclosure, even with perfect layout.  So there's not much if any cost to pay for the poor layout, as there's no cost-saving alternative.  Perhaps one might argue for a 2-layer board (assuming that's 4 there), but with the PLCC socket, and wide buses, that's probably infeasible, even if the layout could be done while respecting signal quality (which it definitely can't, not in 2 layers with component density like that!).

On a side note, I wonder how CRTs were for EMC, in general.  TVs were certainly common enough with plastic (or wood, hah) enclosures, and most monitors as well I think.  Granted, a lot of monitors have metal shields or heatsinks in place, dealing with a lot of that -- intentionally or otherwise -- at the board/assembly level already.

Which begs the question, why not shield the motherboard?  Perhaps they didn't have low enough EMI between cables, so still needed the enclosure shielded.  Perhaps a shield would be too bothersome given the number of connections through it?  Counterpoint: I have a late model (contemporary, heh) XT clone with a shielded motherboard, with plenty of holes in said shield for cables, expansion slots, etc., and a molded plastic enclosure.

Tim

[JDW]

The effect is that the main area of the board, pushes itself around by the reaction force from all those damned traces crossing the gap.

Can you explain what you mean?  For example, what would occur had they not divided up the motherboard ground plane into two parts (with that single trace joining them) and instead just kept the entire ground plane as a single piece?

Which begs the question, why not shield the motherboard? 

The motherboard is shielded with a flexible metal shield which covers the entire bottom, left, right and back (connector) sides. (Naturally, the bottom is covered with insulating material so the solder side cannot short against it.)



Secondary case shielding was accomplished with a metallic spray coating on the inside of the plastic case.



Lastly, how do I specify a pixel width for the inline images in my posts?  These full width images are annoying.

[David Hess]

The effect is that the main area of the board, pushes itself around by the reaction force from all those damned traces crossing the gap.

Can you explain what you mean?  For example, what would occur had they not divided up the motherboard ground plane into two parts (with that single trace joining them) and instead just kept the entire ground plane as a single piece?

It means the return currents for the traces which cross the gap have to go all the way around to the trace, which increases the loop area and emissions.  A ground plane allows the return currents to closely follow the signal trace, minimizing loop area and emissions to minimum.

Was it worth increasing board emissions to lower emissions from the chassis?  I have to assume so.  Maybe they took other steps which are not so apparent, like limiting the bandwidth of the signals or using differential signaling.

[T3sl4co1l]

Ah OK, shield over it besides. Hmm, maybe it was worse than I expected!

Can you explain what you mean?  For example, what would occur had they not divided up the motherboard ground plane into two parts (with that single trace joining them) and instead just kept the entire ground plane as a single piece?

Like so:



Without the slot, Vcm would be, single, maybe 10s of mV.  With slot, it can be 100s; maybe even several volts if a lot of the bus changes all at once.  I'd be willing to bet they could've saved that internal shield, or, perhaps with adding a few more ground points (midway and/or around the board edges) to the enclosure to reduce the Vcm and ground-return impedance for the cables.  Maybe some ferrite beads on the cables to help that out a bit, and likewise better grounding (to enclosure) for the various peripherals (drives, CRT board, PSU).

Or maybe not; I saw apparently there's an upgrade module for as much as 50MHz? That must be within the card though... little on the main board can run that fast, surely?  (Not much logic on there; I see a couple 74LS, and a couple 74F. Hard to say.)  That'd all be pretty rough going, at 5V supply, 3.3V at best -- it's a lot of voltage to launch down a ~100 ohm trace, tens of millions of times per second, and indeed a lot of power as well, which is why as clock rates have gone up, signal levels have gone down.

I'm not sure exactly how sharp the MC68k, or other hardware, drove their pins; not sure offhand if that's full CMOS (say 74HC ish) or TTL-ish (NMOS, HMOS) levels, in terms of both speed and voltage range.  [HMOS, was an Intel thing; I don't know what Motorola did at the time.]

Anyway, I'm sure it's enough trace length that -- not so much signals coming from the CPU end towards the SCSI chip, because that's just a small capacitive load -- but the return path, SCSI driving the bus, is definitely a concern.  I guess I should've drawn the voltage source on the other side, but whatever it is, that's the loop that's the problem.

Lastly, how do I specify a pixel width for the inline images in my posts?  These full width images are annoying.

Like HTML: put a [img width=foo] in there.

Tim

[T3sl4co1l]

Was it worth increasing board emissions to lower emissions from the chassis?  I have to assume so.  Maybe they took other steps which are not so apparent, like limiting the bandwidth of the signals or using differential signaling.

They definitely did, like, there's a whack of AM26LS30/32 for RS-422 (that'd be AppleTalk I think?), plus SCSI is... well usually differential, I guess I don't know if it was here, I could look at a few more datasheets to tell...

And they knew there would be problems.  The narrow DIPs there, resistor packs.  With "FILTER" written by them... though plain resistors aren't much for filtering.  (If they were ferrite arrays, that might be something, but I'm pretty sure Bourns never made one of those, at least not in the common yellow or blue plastic like that? 115-0002 is just resistors, in any case.)  There's a handful of caps and such, but not nearly enough for all the signals going around, I think; the shielded D-subs (and, the others are DIN connectors, right?) are definitely mandatory for meeting EMC with devices attached.

Granted, that's maybe not such a smoking gun, as D-subs and DINs have always been very popular.

A cost-reduction argument for unshielded differential cables might be made; AppleTalk could be similar to Ethernet in that regard, but with DC coupling and ordinary UART signaling (or, whatever else they wanted to use!).  And it could still stomp all over the PC's pitiful RS-232 serial; assuming they can get data to it fast enough, of course (buffering and/or DMA would probably be mandatory).

...Which, oh neat, the SCC does a bit of link layer protocol actually, but apparently doesn't do almost any buffering, so that kinda sucks for sustained data rates.

Not sure what's the display on this thing, I mean it's internal obviously, but does that just go up the main power connector and harnessing links that to the CRT board?  I suppose it's just a bitstream anyway, that thing was monochrome.  A schmitt receiver on the far end will clean up whatever ringing/droop the harness incurs.  Only a... what, 10ish MHz pixel clock? Oh, or maybe that 15.6672MHz can.

Tim

[JDW]

...the return currents for the traces which cross the gap have to go all the way around to the trace, which increases the loop area and emissions.  A ground plane allows the return currents to closely follow the signal trace, minimizing loop area and emissions to minimum.

Was it worth increasing board emissions to lower emissions from the chassis?

I am aware about loop area and emissions (EMI considerations), but it seems you are saying that the deliberate design decision by Apple engineers to break the motherboard ground plane into 2 pieces -- 2 pieces connected to each other only by that single trace -- "lowers emissions from the chassis" by way of the very slight resistance and parasitic inductance inherent to that single trace, allowing that single trace to somewhat act like a very low end ferrite bead, filtering HF noise coming into the motherboard from the chassis (or connectors).

Without the slot, Vcm would be, single, maybe 10s of mV.  With slot, it can be 100s; maybe even several volts if a lot of the bus changes all at once. 

Right, which is why I don't really understand why the broken up ground plane is needed at all.  Just keep the entire ground plane ground, with no breaks, and add ferrite beads, etc. at the noise points, which would be where the metal bracket solders to the motherboard, and of course the connectors, which are all on the same edge of that motherboard.

I saw apparently there's an upgrade module for as much as 50MHz?

Not 50MHz for the Macintosh SE.  The SE (8MHz 68000 CPU), like the superior SE/30 (16MHz 68030 CPU), has a PDS slot for accelerators, but the accelerators for the SE are a bit more low end than the SE/30.  There are 25MHz 68030 PDS accelerators out there, but the 50MHz 030 or 40MHz 040 accelerators are reserved for the SE/30.  The design of the SE/30 motherboard is similar insofar as there is a break as shown below, but the broken parts are connected across different planes via a few thru-hole IC pins from what I can see in the Gerbers.

Like HTML: put a [img width=foo] in there.

To reduce inline image width, per your suggestion, I tried adding "[img width="500px";]" & [img width=500px]" to no avail.  Any further specifics would be greatly appreciated.

SCSI is... well usually differential, I guess I don't know if it was here, I could look at a few more datasheets to tell...

It is plain old SCSI-2 on the SE and SE/30 computers.

The narrow DIPs there, resistor packs.  With "FILTER" written by them... though plain resistors aren't much for filtering.

The 4 "FILTER" chips are BOURNS RC networks, part number 4120R-601-250/201. They are becoming harder to find, so someone has made a recreation here.

Not sure what's the display on this thing...

9" B&W CRT displaying 512x342 pixels.

[T3sl4co1l]

I am aware about loop area and emissions (EMI considerations), but it seems you are saying that the deliberate design decision by Apple engineers to break the motherboard ground plane into 2 pieces -- 2 pieces connected to each other only by that single trace -- "lowers emissions from the chassis" by way of the very slight resistance and parasitic inductance inherent to that single trace, allowing that single trace to somewhat act like a very low end ferrite bead, filtering HF noise coming into the motherboard from the chassis (or connectors).

The one they they probably do get, is less current across (laterally) the connector area -- though I'm still not very confident in this because of the sheer number of traces crossing the gap.  To the extent that the outer metalwork and metallization are imperfect, this could help to reduce the voltage drop between them, as seen from the outside (where it matters, to EMC) -- it just seems a very cut-off-your-nose-to-spite-your-face kind of move.

Right, which is why I don't really understand why the broken up ground plane is needed at all.  Just keep the entire ground plane ground, with no breaks, and add ferrite beads, etc. at the noise points, which would be where the metal bracket solders to the motherboard, and of course the connectors, which are all on the same edge of that motherboard.

Maybe they read one of those trade mags back in the day, discussing slotted grounds, and got way too excited about the trick.

I see it all the time; far more often than is deserved.  99.5% of the time you think about slotting a plane -- no, it's actually worse that way, don't do it.

It's those tiny few cases remaining, where it does help out.  But knowing which ones, requires a lot of experience.

No idea who built this board, how much experience they had -- maybe it was intentional and expertly done, maybe it was ignorant, we can't tell at this point.  (Unless you get lucky and they're on Twitter and watching, or something. )

To reduce inline image width, per your suggestion, I tried adding "[img width="500px";]http://" & [img width=500px]" to no avail.  Any further specifics would be greatly appreciated.

...Where at?  I don't see an edit...

Like this (quote my post to see the code):







Also, fortunately the forum shrinks images to fit width, so those of us using modestly sized browser windows don't have to strain our eyes to look at things.  Also, you could consider sparing a few megabits on your viewers by reducing the images in the first place.

It is plain old SCSI-2 on the SE and SE/30 computers.

Ah, so parallel it is (I guess? Wikipedia still isn't terribly specific, at a glance..), and that explains the lack of terminators.

The 4 "FILTER" chips are BOURNS RC networks, part number 4120R-601-250/201. They are becoming harder to find, so someone has made a recreation here.

Ah, cool!  So they do get some attenuation there, that helps a lot.

9" B&W CRT displaying 512x342 pixels.

So without porches, and at 60Hz, 10.5MHz pixel clock; shouldn't be much higher than that, actually, I would be shocked if they have much overscan... maybe they put a 2/3 PLL in that system chip or something?  Which I guess is handling graphics too, and, hmm would it have onboard VRAM for that then?  At least being monochrome it doesn't need a whole lot I guess.

PCs were much simpler in that respect; graphics was either a pile of chips around a CRTC (6845 usually), or some ASIC doing the same things (probably most EGA+?). The attached VRAM of which is very obvious.   (My XT clone indeed has 128k Paradise EGA; what luxury! Too bad the ISA bus is so dreadfully slow, along with the 8086, to do very much with it... )

Tim

[mansaxel]

To reduce inline image width, per your suggestion, I tried adding "[img width="500px";]" & [img width=500px]" to no avail.  Any further specifics would be greatly appreciated.

The proper syntax is:

Code: [Select]

[img width=300 height=300]picture.jpeg[/img]

Sizing arguments are without units (pixels are assumed), and are optional.

[David Hess]

...the return currents for the traces which cross the gap have to go all the way around to the trace, which increases the loop area and emissions.  A ground plane allows the return currents to closely follow the signal trace, minimizing loop area and emissions to minimum.

Was it worth increasing board emissions to lower emissions from the chassis?

I am aware about loop area and emissions (EMI considerations), but it seems you are saying that the deliberate design decision by Apple engineers to break the motherboard ground plane into 2 pieces -- 2 pieces connected to each other only by that single trace -- "lowers emissions from the chassis" by way of the very slight resistance and parasitic inductance inherent to that single trace, allowing that single trace to somewhat act like a very low end ferrite bead, filtering HF noise coming into the motherboard from the chassis (or connectors).

It is not the motherboard ground plane necessarily, but the multiple connections to the chassis from the motherboard ground plane.  Without a single point ground, return currents through the motherboard ground plane would cause return currents through the external chassis between the shielded connectors, which would then cause the chassis to radiate EMI.  When cables are connected, the return currents would travel through their shields as well.

Alternatives include using only one external shielded connection so there is only one point connected to the chassis, or isolating the external shields from the chassis which has its own problems.  A secure seamless connection between the connectors and chassis to keep the currents on the inside of the chassis would be ideal but is difficult to achieve; this usually means finger stock and this solution can be found on many PCs to varying levels of thoroughness.

[wizard69]

Ground loops suck and yes there is a possibility here of a problem.   Beyond ground loops if the "grounds" are not at the same potential you have  real posibility of frying cabling and possibly other hardware.

As for your post you do not mention what sort of transducers you are using.   It makes a difference.    For example thermoucouples can come in variants that have the junction grounded and those that are ungrounded.    As somebody has already mentioned there are so many factors here that no one can provide an absolutely correct answer.    Personally I'd ground the supply at one point on the chassis and work from there.

i am thinking out loud here... if the signal ground is tied to chassis ground, and i connect it to another mains grounded system, that can case a long loop between grounds that can have voltage difference both in noise and mains. this would make for a noise source... or am i thinking completely wrong?

You are not wrong but if you are talking a large installation where ground potential can vary you might end up with excessive current on your ground.   In extreme cases you could burn up your cables.   Supposedly these sorts of industrial systems have equal potential grounding installations but that never seems to be perfect in practice.   So in a modern facility you would have less of a chance of excessive ground currents.   If you are limited to a lab or small installation, ground currents should be less of a problem.

Noise on the other hand is funny.   You can follow recommendations precisely and have noise problems.   Ideally you would have the ability to lift the ground at one end of the sensor cable if needed.

if this is the case, then id think keeping the signal ground isolated from mains/chassis might make sense?