I recently noticed that the SX running at either 4Mhz or 50Mhz is affected by nearby sparks.
When this happens, the SX hangs or jumps to a random part of the program.

This spark can be from a simple switching on or off of a nearby electrical appliance.
At first I thought that the spark was transmitted through the same mains 240V supply through the adapter into the SX,
then yesterday I did another test with the SX board running from a 12V battery.

I found that the SX is still afffected by the nearby sparks even when it is battery powered !

Has anybody had the same problem?
Any filtering or shielding suggestions?

A while ago, there was a longer thread in this forum dealing with EMI problems, see http://forums.parallax.com/forums/default.aspx?f=7. The methods that help to keep radiation from the SX to the outside low are also good to keep external noise going into the SX low.

I'm using the SX 28 in applications where DC motors (24V, 18Amps) and various releays are turned on and off without trouble. The SX 28 is sitting on a separate small multi-layer board with two inner layers, one connected to Vdd, and the other to Vss. I have measured a capacity of about 1 nF between the Vdd an Vss terminals on an empty PCB. This capacity is caused by the two inner supply layers, and it is almost an ideal capacitor as it has ony very little parasitic series inductance. This board passed EMI approval with no problems, i.e. the radiated RF signals were below the limits, and the board could stand external bursts and spikes without abnormal operation.

Once having made this experience, I now design all my PCBs for commercial applications with two inner supply layers. Today, such boards are not much more expensive than double-sided ones.

This is obviously surpious RFI.  Sparks can be pretty broad in the band of frequencies they deal with.  Automotive ignition can really jam radios when not properly shielded.

If it is coming through the power supply cord, use a ferrrite bead on the cord to trap it.  There is a bit of possiblity that the 12 volt power cord might act as an antenna.  The same is true of any long wires in or out.

Guenther's suggestion is good, but if you already have a board in use or development - you might just want to shield it.  Possibly sandwitch the board between two blank copper clad printed circuit boards with stand offs and have those boards appropriately grounded.  If that isn't enough, wrap it all in aluminum foil to create a complete Farraday Cage.

Honestly, I haven't found the need to try it, but it is worth a try.  If you don't want to deal with the copper clad boards, throw it all in a plastic food box [similar to one for refrigeration of left overs] and cover that with the foil.  Ground the foil.  The plastic will protect the circuit from the foil and possible shorts.

When I did the tests with one of my units, I also tried various methods of shielding w/o too much success. There are always leads going in and out. For an effective shielding, you would need a hermetically sealed metal enclosure with all in/outgoing leads passing through filters. Even the smallest opeing in the shield can act as an antenna either to transmit noise, or to receive it. Nearby sparks often produce electro-magnetic fields, like a radio transmitter. Copper or aluminum only provide an electro-static shielding but no magnetic shield, so they don't protect any lead or component on the SX board from "receiving" electro-magnetic noise. In case of sparks, these are broadband beasts of noise.

The better protection against RFI is to avoid it at the source. If there is any electric device that generates sparks when turned on or off, try to attach a filter or "snubber" to that device. Ok, in case you can't modify the source of RFI, the only way is to protect the "receiver". As Kramer mentioned, power supply leads are usually pretty long, and therefore are nice antennas. Ferrite beads around the leads (as close as possible to the supplied device) sometimes help.

You mentioned that you tried to power the SX from a battery, and nearby sparks still affect the SX. Again, the lead to the battery could act as an antenna, or any other lead connected to the SX.

As usual, discussing RFI comes close to reading a crystal ball - there are so many possible sources of trouble that often a lot of try-and-error is required.

In order to avoid that the SX goes into some unpredictable state after a spark, you might consiter activating the watchdog. This - at least - should reset the SX, giving it a "clean start".

How about having a 5.3V zener diode across the 5V power lines and also across any other leading wires
going into the SX board,  will it help?

What about having snubber RC pairs across high voltage switches (like telephone lines)?
What would be the RC pair values?

I also found that using mechanical relays to connect and disconnect telephone lines (48V) can also produce RFI that can affect the normal execution of the SX.....

It is not the relay coil b'cos when I pull out the RJ11 cable ( phone line ) and test the same relay switching many times,
the SX is not affected.

William,
Zener diodes are way too slow to suppress signals like the ones generated by sparks. Even fast Schottky diodes would not help much. Sometimes, I use such diodes though to protect the SX and other components against over-voltage on the supply.

Yes, I had mentioned the snubber networks in my last post. Such a combination of a resistor and a capacitor in series across the high voltage switch helps to reduce noise at the source, which is always better. I'm using one in an SX application that controls solid state relays. I have 22 nF + 47 Ohm across the relay's output pins. This value is recommended by the relay manufacturer but they say, this is just a value to start with as it depends on the specific type of load which values provide an optimum, so you might need to do some experimenting. Please note that the capacitor must stand a high voltage. For 117 V AC, it should be rated at least for 350V DC.

Sounds strange to me that switching the 48V phone line also causes trouble. Are there any longer leads connected to the SX setup going in parallel with the phone line, or a mains line?

Thank you for your advise Guenther.

The "sparks" ( i didn't really see the sparks, I just assume it happens on the relay contacts ) from switching the ordinary telephone line can affect the SX's execution about 10% of the time, even when there are no parallel cables with the phone line.

That means if you switch on and off the phone line every 1 second, you might see a problem every 20 seconds.
For most ordinary use, the phone line is not switched that frequently so I guess a lot of people didn't notice it as a problem.

If I use a 22nF + 47 ohm snubber across the switch, that snubber will allow "crosstalk" to creep into the line when its supposed to be switched off by the relay !  So I guess that wouldn't do.

Of course if I use a solid state relay from Clare, there is no problem from the phone lines. No need snubbers.
Then the only problem left is the mains power switching "sparks".

Note that what you're suffering is a bit extreme. The SX is a small lump of silicon, and its program is on the same die - there are no buses traipsing around a PCB, picking up interference.
How is your board built? What's attached to the MCLR pin?
Come to think of it, are you taking interrupts off pins, which are being triggered by the spark?

Hi Steve,
I think not that extreme.
My board is double sided about 3" x 3" inch only.  The only "antenna" must be the 12v adapter wire ( about 3 feet long ).
The MCLR is tied to 5V via 10K resistor ( as usual ).
From my tests, I found that many different SX boards  also have the same problem with sparks.

The test is simple.  Plug in the adapter which powers up the SX board.  Then on the same mains extension socket, plug in a 100w reading lamp.
Play with the mains socket switch ( which is just next to the SX 12V adapter ) by switching the lamp on and off every 2 seconds.
If your SX runs on watchdog, you should be able to see it restart in less than 20 seconds.  ( beep the buzzer when restarting to make it easier ).
If it doesn't restart or act errratic, try placing the board nearer to the lamp.  : )
I think most SX users don't test it like this, so they assume it is immune to sparks.
Why don't you test it with your board and let us know the results.

> Why don't you test it with your board and let us know the results.

I run my boards in cars' engine bays (notoriously hostile EMC environment) and one's a power control for my TIG welder (100 amp squarewave AC, with inch-long sparks for ignition), and they don't crash. However, my power supplies are probably a bit of overkill - works for me, but I'd worry about making consumer products thas cost as much.
Similarly, we've got piles of test gear running on SX in factories where things are perpetually being plugged & unplugged. Again, no crashes. I can't help but think that you've got power supply or reset problems - no other pins should be able to cause effects like these, and, with the best will in the world, I don't believe you're inducing very many (mili)volts across the die from EMC.
Is there any way you can watch the power rail with a 'scope?

Hi William,
A while back I installed express PCB, and took a look at your board layout for the SX keyphone.  I am thinking that this is the board you are having problems with.  I intended to post a reply with a comment about the layout, but other things took precedence.  Seeing this thread reminded me about that, and I finally posted that comment.

In two words:  bypass caps.

You don't have any on the SX, assuming you are using the same layout as the key phone board.  Likewise, you only hooked up *one* of the pairs of supply leads on the SX-52.

Bypass caps really are not an optional thing with a chip like the SX which has extremely fast swithcing going on.  The bypass caps keep the supply from drooping when the SX has an instantaneous draw in current.

The multiple power and ground pins are likewise not optional, the SX is a power hog.  It has to be to run so fast.  Without the extra leads to reduce the supply impedance to the chip, it is starving for clean power and ground.

I suggest you jumper the other power leads to appropriate points, and solder some small bypass caps across the adjacent power and ground pins as well.   That might not be practical unless you can get some .03" x .06" wide termination surface mount caps, or .02" x .01" size normal termination types.

I think if you do that, you will find that the SX is quite robust and immune to most electrical events in the neighborhood.

Hi William,
I can only agree with Michael - bypass caps are of absolute importance - I assumed that you had some installed, so I did not ask, and I did not realize that you had posted your PCB design before.

I can only recommend to go back to this thread: http://forums.parallax.com/forums/default.aspx?f=7&m=75182 where several aspects of EMI have been discussed.

As mentioned before, for commercial applications, I have completely switched to four-layer boards with two inner supply layers. These make the best bypass cap you can imagine, and the two additional layers only add marginal production costs. When using SX48/52 devices - as Michael said - all power and ground pins must be connected. When you have a look at the Parallax SX 48/52 Proto Boards, you will notice that they have added bypass caps to each power pin - I can only recommend this.

Hi Guys,
Ok I agree that my single 470nF bypass cap is nearer to the regulator than the SX.
Is 470nF multilayer better than 100nF ceramic as a bypass capacitor?
I will try to put it nearer in the next board iteration.

However I would like to dispute the notion that all the 4 Gnd and all the 4 Vdd pins on the SX48/52 needs to be connected.
Is there any mention of such requirements in the official SX Datasheet?
I also don't agree that you need 4 bypass caps for the SX48/52....   sounds like an overkill.

I believe the QFP design with a pair of Gnd-Vdd pins on each side is to allow easier PCB routing.
If we were to connect all the Gnd and all the Vdd pins together, there would be no space left to route the other pins !

>I also don't agree that you need 4 bypass caps for the SX48/52.... sounds like an overkill.

Err, your board reboots unpredictably... I can't quite understand your logic here.

And yes, you absolutely do need to feed power and ground into all the pins.
If you've got access to a decent oscilloscope, you might like to have a look at the voltage across one of your 'unused' pairs of power/gnd pins. I bet you see horrendous noise - which is what that side of the chip is trying to use as a power supply.
Yes, routing power and ground to all the pins does take effort. However, it's not optional, as you've discovered.

If you think about it, the chip can't generate electricity from nowhere - if there's (say) half an ohm resistance (and inductance) on the power suply pin, if only 8 outputs switch simultaneously, delivering 30mA transiently into a capacitive load (a very likely scenario), then the power pin has to supply 8*30mA through half an ohm. That's 120mV drop, and is pretty much best case. The transient currents will actually be rather more than 30mA, and this is just a rough calculation of the power requirements from switching 8 pins - there's a lot more going on than that.

Hi William,
in general, multilayer caps are the better choice because of less parasitic series inductance. To avoid additional parasitic series inductance, it is necessary to keep the leads as short as possible (SMTs are best as they don't have leads), and to place the cap as close as possible to the supply pins as longer PCB traces would add inductance again. You can't say as a rule of thumb that a higher cap (say 470 nF) is better than a 100 nF) cap - sometimes, smaller values even do better.

I can only agree with Michael that connecting all supply pins helps to reduce the supply impedance, and four filter caps, each of them close to a pair of supply pins are by no means an overkill. In case you did not, please check the other thread (EMI Failure) I had mentioned in my previous message. There you find a lot more information about this topic, so it is not necessary repeating it here.

> Err, your board reboots unpredictably... I can't quite understand your logic here.

It doesn't reboot unpredictable, only 10% likelyhood if there is a switching spark really nearby.  : )
Ok you guys win.
My last question :-
Why the SX48/52 datasheet makes no mention about the need to connect all VDD and GND pins?
There is no mention of 4 bypass caps requirements or their recommended values also...

The use of bypass caps is beyond the scope of the SX datasheets, they are concerned about the operation of the chip itself, not the sucessful design of circuits using them. (As an aside do the datasheets even mention the importance of locating the resonator close to the SX? I cant check the sheets at this moment).

As a matter of fact, UBICOM has released a special application note delaling with EMI in November 2000. Although only dealing with the "small" SXes only, it contains some important information. As I'm nor sure if this dicument is publicly available in the moment, I simply attach it to this post.

William,
I partly agree with Paul. He is right, in general the datasheet describes the specifications of a component and not necessarily its applications. On the other hand, a special note in the datasheet that due to their high speeds, SX controllers require specific attention concerning EMI, and supply bypassing would be appropriate, or at least a reference to the EMI app note.

I know that Parallax is in a state of revising all SX datasheets, so there now is a good chance for adding such information. Rocklin - did you copy? :-) .

Somehow, I feel like a missonary, preaching the four-layer PCB design. With the two inner supply layers, you would not have a problem, connecting all Vdd/Vss pins of the SX48/52, still leaving you enough real estate for other signal traces.

I have designed a general-purpose board for the SX28 that I call the "SX-Multi Board". Although the two inner layers look like Swiss Cheese due to the many plated-through holes for the breadboarding sections, the two inner layers build up a capacity of about 1 nF. This sounds like a pretty small value but as this is an almost ideal bypass capacitor it is more effective than a discrete bypass capacitor placed somewhere on the board. A good (bad?) example where the bypass capacitor should _NOT_ sit is the SX-Tech board. It is placed "miles away" from the Vdd and Vss pins at the left small side of the ZIF socket. Would not be a problem placing it much closer to pins 2 and 4 for better filtering. OK, the SX-Tech board is an experimental tool, not designed for commercial applications. I bet - it would fail any EMI test.

Hi, I just want to say that I defer to Guenther on this.  
My tin foil and box idea is obviously archaic and maybe useless.  The more I read and the more I ponder, a good four layer board with appropriate by-pass capacitors seems necessary.

Originally, I had been hoping to make my own one and two layer boards for microprocessors, but now I see such as merely an educational stage.  [Drilling all those holes by hand is a bit of chore too.]
Guenther's Multi board is quite attractive and a less expensive alternative.

Guether, have you investigated why 4 layer exhibits better EMI properties than 2 layer? Is it because flood fills frequently used for power and ground planes on 2 layer boards create resonant cavities for EMI to be amplified, or the board material dampens EMI because the power planes are internal? Not important, just curious.

Bear in mind that the original poster did almost everything possible to make his board as crash-prone as possible, and the board still mostly worked unless he provoked it with nearby sparks. A 2-layer board, done right, while it will be a pain to get through CE EMC compliance testing, should be fine as a development platform. You get a _lot_ of help from the SX, since all its resources are on-chip. Just feed it power, clocks and reset according to the datasheet, and if you're unsure at all, post your design for a quick review.

Metal boxes are still useful, but it's _always_ better to design the board right in the first place. Cheaper, too.

Since I have not seen any pricing on Guenther's multi board, maybe "less expensive" is a little premature statement to make. I like his board, it looks like their was a lot of thought put into the job of designing his board. Remember, the multi board has to cross the atlantic ocean and all of the US to get to a distributer, that is not going to be a cheap ride. Besides Parallax is going to be selling the SX28 proto board for $9.95, so you know that Guenther's multi board kit is not going to be less than that.

What would you be willing to pay for the board. Me, as cheap as I am, I think $19.95 USD as a complete kit, and I mean a complete kit, would be reasonable.

Hi William, interestingly sparks have even been used in place of ultrasonic transducers to achieve much higher ultrasonic frequencies than typical 40-80KHz. That is, they have good transmission qualities. The fact that it still occured in your battery operated circuit supports this. Sounds like you have inadvertantly come across a great jamming tool. Good metal shielding should solve this as well as a bandpass filter if necessary.

Guenther really put a lot of effort and thought into the original thread about EMI.  I was pretty much the 'student' and he the 'teacher' on that one.  Anyone that doesn't understand his shorter reply herein, should look it up.

Guenther stated that most, if not all the problem, originates on the EMI in the power lines.  Apparently reflections from the oscillator and the reflections from switching  logic at regular periods of time cause the problem.  
Regarding Paul Baker's question -- I would guess the reason for 4 ply being more effective must always come down to creating a tuned circuit via a large power filter capacitor which implies an impedance effect and a resonant effect.  The capacitance just knocks the tops off the high frequency resonance.

Additionally, there is might be 'a halving' of EMI created by the internal floods create a 'semi-shield' for the individual leads. The Vss and Vdd EMI are 180 degrees out of phase, so not much goes beyond them if the surfaces are nearly equal and nearby.  This would be partially similar to eliminating cross talk in ribbon cable by having alternating wires grounded.

I guess we have an entirely new electronic concept - 'the development Autonomy of the digital information from the power source that it depends on.'  That is one that the chip manufactures ponder as a speed barrier.

And, I have been as 'guilty' as William of doing everything I can to make a susceptible EMI board.  In the beginning, you just think that the routing software can 'intelligently' layout a board, but there is  really an art to a genuine production run, stable product.

I suspect it is nearly all dampening, not amplfying.

It would be wonderful if the manufacturer could include a set of small by-pass capacitors within the silicon or at least the package as a 'first line of defence.  It would make a more marketable product as it wouldn't be so tricky to product a board.