When not to use a ground plane?

[Neilm]

If you have a coil that leaks magnetic field you may want to remove the plane underneath it so it induces no current. ( here is another common mistake : lets put copper to shield the inductor..... Copper does not shield magnetics ! )

A copper band around a transformer / inductor can act as a shorted turn and will absorb any RF energy that is leaking out. This will reduce the RF and is used to reduce the emissions for EMI tests.

Neil

[free_electron]

ahh. that is something different. you are shorting the field by giving it a closed electrical path. this may work for smps transformers ( there you see that often , but they dont use a hard short... there is some resistance in the path , )

but for a simple ferrite coil like a 1210 package this doesnt work.

[jerry507]

Im not arguing with you, it is an odd conversation. You are just proving your own truth.

Yea he is, it is very very strange

[free_electron]

Im not arguing with you, it is an odd conversation. You are just proving your own truth.

Yea he is, it is very very strange

it's not my truth. this is how it is explained in courses given Howard Johnson ( Black magic for High speed deisgns ) and Montroses' EMc and Signal integrity courses.

[jerry507]

After rereading what you said, I think I see the confusion. You're trying to say that the use of ground planes has nothing to do with the highest fundamental frequency in use, and everything to do with the highest harmonic frequency seen?

[w2aew]

the base frequency of the signal doesnt matter for this.
a 1 hz signal with 1 ps risetime produce an equal amount harmonics  as a 1Mhz ignal with 1pS risteime. the base freuqency isshited, thatis all. the number of harmonics does not change.

therefore : it is purely edge rate that determines how much energy is blasted in the spectrum. the frequency if the edge determines where the harmonics begin.

a 1Mhz 1vpp squre wave with an edge of 100nS does not produce as many harmonics as a 1MHz 1vpp with a 1pS edge rate.
simply becasue you don;t need that many harmonics to get that edge to be so fast...

Not that you need another viewpoint here - but the statement above that says that the number of harmonics doesn't change with the base frequency of a squarewave (assuming equal risetimes) is simply not true.

A square wave is composed of energy at the fundamental frequency, and energy at the ODD harmonic frequencies.  The more high order harmonics included, the faster the edge rate.  Thus, an ideal 10Hz square wave will have energy at 10Hz, 30Hz, 50Hz, 70Hz, etc.  An ideal 1MHz squarewave will have energy at 1MHz, 3MHz, 7MHz, etc.  If both of these squarewaves have equal risetimes, then the harmonic energy of them will roll off at the same high frequency value, but the NUMBER of harmonic components will be dramatically different between the two since the spacing is so different.

[jahonen]

the base frequency of the signal doesnt matter for this.
a 1 hz signal with 1 ps risetime produce an equal amount harmonics  as a 1Mhz ignal with 1pS risteime. the base freuqency isshited, thatis all. the number of harmonics does not change.

therefore : it is purely edge rate that determines how much energy is blasted in the spectrum. the frequency if the edge determines where the harmonics begin.

a 1Mhz 1vpp squre wave with an edge of 100nS does not produce as many harmonics as a 1MHz 1vpp with a 1pS edge rate.
simply becasue you don;t need that many harmonics to get that edge to be so fast...

Not that you need another viewpoint here - but the statement above that says that the number of harmonics doesn't change with the base frequency of a squarewave (assuming equal risetimes) is simply not true.

A square wave is composed of energy at the fundamental frequency, and energy at the ODD harmonic frequencies.  The more high order harmonics included, the faster the edge rate.  Thus, an ideal 10Hz square wave will have energy at 10Hz, 30Hz, 50Hz, 70Hz, etc.  An ideal 1MHz squarewave will have energy at 1MHz, 3MHz, 7MHz, etc.  If both of these squarewaves have equal risetimes, then the harmonic energy of them will roll off at the same high frequency value, but the NUMBER of harmonic components will be dramatically different between the two since the spacing is so different.

This discussion has been off-topiced somewhat but I think it might interest someone anyway...

Purely odd harmonics will only occur if square wave has precisely 50/50 duty cycle. If we change the duty cycle to something else like 1/3 or 2/3, then 3rd harmonic and multiples of will disappear, but we get second harmonic. In any case regardless of the duty cycle, the spectrum envelope obeys sin x/x (sinc) shape. Having 50/50 duty cycle means that nulls of the sinc functions will be at even harmonic points, and odd harmonics are exactly at lobe peak points of the sinc function. At very extremes of the duty cycle, the sinc main lobe gets wide (approaching constant when duty cycle is shrunk into dirac delta impulse), so all harmonics will be of same amplitude.

The edge rate is significant in sense that it introduces second inflection point on the spectrum envelope. For ideal theoretical squarewave, harmonics always go down 1/f fashion, i.e. 6 dB/octave ad infinitum. Having a finite edge rate means that when certain frequency is reached, the roll-off increases to 1/f^2, i.e. 12 dB/octave. That particular frequency is also considered to be the critical high limit for EMC.

Edge rate is significant parameter in sense that signals which have steep edge rate, must be cared for regardless of their repetition frequency. For single-transition edge-sensitive signals, corruption of edge will ruin the signal even if there is no frequency. Now this means that if we get a single edge transition cleanly from the source to the load, it means absolutely nothing how many of these transitions are sent per time unit, they will always get transmitted cleanly (assuming that the edge does not slow down along the way). We can even put several transitions simultaneously on the line without any problems.

Thus I understand free_electron's point of view perfectly and I share his opinion. Purely thinking as harmonics, for example one might think that having 10 harmonics of fundamental is enough to transmit the signal. This is not the case if we have very extreme duty cycle. That can be misleading if we are operating at odd parameters. Thinking via edge rate, doesn't fail even on extreme cases.

Regards,
Janne

[free_electron]

And there you go. Janne explained it in more detail. The problem is that signals in real life are never 50% duty cycle... So if you focus at frequency you can only handle 1% of the problems. Databuses and serial transport systems like PCIx , SATA , HDMI, are very critical. When it comes to signal integrity there is only one thing that matters : getting the edge across undistorted.

When it comes to EMC the problem is 'trapping' all the energy so it gets absorbed in the receiving end and does not reflect back onto the line and transmitter.if it reflects back it will radiate ( there goes you compliance testing...) and it will intermodulate...( there goes your signal integrity... )...

EMC and signal integrity are two completely different things, but one can cause the other... And solving one can cure the other, or it can harm the other.

Fact of the matter is that fast processors have slew rate control for both reasons (integrity and emc).

Series termination and co trolled current sources on the output. Some transceiver cells are even adaptive. When tansitioning they turn on two parallel current sources. When a certain voltage is reached one is turned off , when final voltage is reached the second one is turned off and only a 'maintenance' source is remaining. Fpgas have the cells programmable and you set them up by selecting the transceiver standerd you need.

I have seen asics that have auto-tune. At powerup they transmit a sequence of pulses and they reprogram the current sources to get the right shape. Used for very fast transceivers to overcome hassles on the board.

[Baliszoft]

I wasnt telling you that rise/fall times have nothing to do with this. I was just stating, that rise and fall times affect (control) the spectrum of the signal (go and read all this from the beginning). But you do not want to be aware of this, you just stick to that the fundamental frequency does not count (which is true unless the frequency is too high), and that edge rise/fall times are only important. It is also true (no doubt), but they are just an "indirect" parameter - as they control the harmonic content of the signal. It is a simplified explanation/solution of the actual problem for those who are thinking in the time domain.
The harmonic content is causing most of the electromagnetic radiation!

Setting a relatively slow rise/fall time is just one (undoubtly easy and effective) way of controlling EMI, but it is not the only way. Where it is not possible, f.eg a low pass filter can be used. It will have the same result.

it's not my truth. this is how it is explained in courses given Howard Johnson ( Black magic for High speed deisgns ) and Montroses' EMc and Signal integrity courses.

You have learned the lesson well, you just do not understand it.

[free_electron]

I wasnt telling you that rise/fall times have nothing to do with this. I was just stating, that rise and fall times affect (control) the spectrum of the signal (go and read all this from the beginning). But you do not want to be aware of this, you just stick to that the fundamental frequency does not count (which is true unless the frequency is too high), and that edge rise/fall times are only important. It is also true (no doubt), but they are just an "indirect" parameter - as
The harmonic content is causing most of the electromagnetic radiation!
You have learned the lesson well, you just do not understand it.

Yes, rise and fall times affect the spectrum. You got that right.

But you can not look at the frequency of a signal. You want to rip a signal apart into a base frequency and harmonics and then strip off the harmonics with a filter. There is two problems with that :

1) It only works for simple repetitive signals. Have a complex waveform like a modulated signal and you just made it fubar...Especially when complex modulations are used like qam.
2) any form of filtering affects signal integrity! Can't do that because you distort the signal !   Signal integrity is lost. If you start filtering the harmonics you will affect rise and fall time and this can , and will, screw up things.

For a lot of signals there is no base frequency. There is only a base frequency if a signal is repetitive in nature.
Good luck finding a fundamental for a databus or address line, or a modern switching regulator like a constant on-time or a hysteretic regulator... There is none. It dances all over the place. So your filtering sceme doesnt work. Slap filters on a databus and you will change the edges causing data corruption because sometimes you do not meet the required level in time when the memory is latching in.

The only solution is to control dv/dt at the source, or impedance match everything so there is no loss of energy along the path. Loss of energy radiates...

In order to do this work you need to know dv/dt (edge rate) and you need to impedance match.
Frequency is! For the most part, irrelevant.

[Baliszoft]

Yes, electron. You are right. I am wrong.

[free_electron]

you're just being sarcastic now...

it is not a matter of being right or wrong. it's a matter of learning a technique that is applicable when doing a board.

[jerry507]

Why are you focusing on the fundamental as if it's some fixed number? Any realistic signal like you describe can be easily thought of as a band + harmonics of that band. It's exactly the same concept as a single fundamental frequency + harmonics. If you don't have that connection, than it's difficult to turn an edge rate into frequencies you can actually manage. If you just want to manually back down your edge rates until you hit EMI targets, that's fine. But if you want to actually DESIGN to a target, you need to have a better understanding of the spectral content of the signal you're working on, even if it's non-repetitive.

Even if you back this discussion away from the edge rate problem (which actually has nothing to do with copper pours directly), I don't actually know of many situations where pours are inappropriate. As free_electron stated, there are cases where they add stray capacitance, but I'd still say that it's better to have the pour there and manage those effects where necessary. But in a lot of cases, pours aren't going to net you a whole lot. In fact, nice tight layouts with well thought out current paths are probably just flat out better than copper pours in any case.

Unless they're heat sinks.

[free_electron]

on the subject of copper pours : those can be actually very ineffective and cause of trouble... as frequency increases the return path is flowing right beneath the coupled source trace. ( for low frequencies the electrons kinda flow wherever they like in the ground plane , for high frequencies they follow the shape of the trace above...)

here is a potential problem on a multilayer board :

top layer trace in horizontal direction
inner one is a ground plane
inner two is anotehr signal plane where there is again a trace in horizontal direction..

due to the tight coupling there now is intermodulation betwene the two signals... it happens because of the ground plane !
so , for signals contaiting high frequencies : DO NOT run them on top of each other shielde by a groundplane. you still need to cross at 90 degree angles. simply because the return current will cause intermodulation in the ground ( ground bounce )

i have sene designs where a bus transceive was routed with half the traces above the ground plane , half below the ground plane... and all the return currents nicely intermodulated in the plane itself.. causing havoc.


you always need to analyze your current loops. currents should never share a common pathway ( sometimes you cannot avoid , and sometimes you can ). that is the principle between having 'analog and 'digital' grounds. you keep the noise in the digital ground. digital circuits have really good noise margin ( don't overdo it though )
the analog ground is anothe rmatter then. and you can only use a piece of copper as a shield if there is NO current flowing in it.

other fun tidbits : planes canradiate tremendously. example : 4 layer board : inner two layers are a power and a ground.
parts are concentrated in the center. with large 'open area's' towards the edge of the board.. this is effectively a dipole.... two pieces of copper sticking out with noewhere for the current to return... radiation galore !
so , if you have large open spaces on a board : create a return path : slap a few capacitors close to the edge so any ac current (dc current does not radiate) is shunted through the caps and the dipole is effetively shorted.

we had this problem in a Co rack. one of the boards was a diagnostic card. since all boards are the same size this board was mainly empty apart from a littel lump of parts( some cpu and assorted parts ) close to the back end card connector ( these boards slide in large 24 inch telecoms racks. i was working on ADSL central office boards at the time )

only a few signals travel from the far end to the front. some led's and an rs232 connector.
this board was radiating as hell... we thought its the rs232 , its the leds... tried everything. ferrite beads , whatever.
until we simply placed a few 10nF caps at the front of the board. that solved the problem. the planes were simply long antennas radiating .
So her : remove the plane under the void area's or shunt it with a cap.

[Omicron]

At the risk of making matters worse let me try to give another point of view.

Firstly there seems to be confusion about dV/dt and rise-fall times. These things are not the same! dV/dt, in general, is useless as an indicator for high frequency signal content because dV/dt is scaled with amplitude. I can create any kind of dV/dt I want using a 1 KHz sine wave, simply by increasing or lowering the amplitude. What matters in terms of the spectral content is rise and fall times, which are independent of amplitude. If you limit yourself to digital signals that have a fixed set of levels then you can arguably talk about dV/dt or edge rate and use it interchangeably with rise- and fall time, because in that case you obviously have just one fixed signal amplitude as a reference.

In the "black magic" book that free refers to it is shown that for digital signals the spectral bandwidth of the signal is related to the rise- and fall times. The smaller the rise or fall times the larger the bandwidth of the signal. The fact that you can translate rise and fall times into a bandwidth shows that there is both a time domain explanation and a frequency domain explanation and those two are perfectly equivalent. So arguing that you need to look at rise times or arguing that you need to look at the spectrum is moot, just take whichever of the two is more convenient in a particular situation. When dealing with digital signals looking at the rise times is more convenient because you can easily find these in data sheets etc. When you are dealing with complex comms signals it may perhaps be easier to use the spectrum (not my cup of tea so I wouldn't really know).

Well, my 2 cents anyway ;-)

[jahonen]

Well, data/address buses as well as serial links behave essentially as random noise sources as far as spectral content is concerned. So if we have a 100 Mbit/s random data flowing in a simple digital signal. What kind of spectrum this kind of signal has? It turns out that the spectrum is initially flat, and drops to null around data rate, 100 MHz (excluding a non-ideal clock feedthrough which should not occur in theory, however even small imperfection, like asymmetry, on the signal gives rise to clock feedthrough). Then we have again a continuous lump until 200 MHz where we have another null and so on. So voltage spectrum is essentially a continuous (measurement made by me some years ago, using Cyclone II EP2C8):



And regarding to emissions, it really is the current what causes the radiation (acceleration/deceleration of an electron causes electrons to emit photons, which are in this case at radio frequency) so one must think how to make a low-pass filter for a current source, not voltage source for the filter to be effective. That was one of eureka-moments when I figured that out. Current spectrum is very different and has spiked form in the time domain, and it usually has little resemblance to voltage spectrum. This is kind of bad thing since it can lead to very wrong conclusions. Too bad almost all tools are voltage oriented and there are few small wide bandwidth probes to visualize the current spectrum of a signal.

I find it interesting that crosstalk could happen across ground plane. I have been in belief that skin effect prevents the current flowing very deep in the copper. For fastest transitions, the current flows only on few µm deep layer so we have plenty of margin in typical 35/17 µm thick copper inside of the PCB. Nevertheless, I haven't personally ever seen crosstalk problems happening via groundplane, but of course I can't deny that it couldn't happen. I have been evaluating Sonnet professional lately, maybe I'll give it a go and see what it says...

Regards,
Janne