Ground and Ground plane on 2 layer board

Using a single point ground that you are using should be fine. Make sure that you try and locate this star ground as close to the exit point on the PCB. If you check out National's website for app notes on the LM2575 you should find some good info on ground plane and proper switcher layout.

Liuc wrote:

Speaking generally and not getting into the specifics of your design:

The purpose of the power (ground) system is to provide a constant voltage supply to your logic circuit. This supply must present the constant voltage in the presence of varying and surging currents caused by the electronic switching. In order to maintain this regulation, the power (ground) system must have a very low impedance across a fairly significant bandwidth. The exact bandwidth that you need to be concerned about depends not so much on the clock rate of your circuit (the 20Mhz and 4MHz that you mention) but the rise and fall times of the digital signals. Providing the constant voltage, low impedance supply is often accomplished through proper use of decoupling capacitors. You didn't mention them in your design, but if this concept is new to you then you should investigate it.

Many, especially novice, designers also forget about the return currents. When they look at a circuit schematic they see a driver connecting to a receiver and the current path between them is obvious. What is not as obvious is that there is also a return current, through the 'ground' path between the devices. The ground path needs to be as low of an impedance as the power path. The only structure that will provide a sufficiently low impedance for the ground is a wide flat structure, or a plane. The reason being that at the frequencies of interest, which as I pointed our earlier are far greater than your clock rate, the impedance of the ground becomes dominated by the inductance and to get a low inductance (impedance) you need a wide flat structure. A ground plane that is full of holes is still infinately better than a serpentine trace.

A copper pour will work for a ground plane. On your board which is two layer, it probably matters little wether the plane is on the top or bottom side of the board. If you are planning on wave soldering the board, you may want to consider putting the ground plane on the 'top' as it is possible for the board to warp when it is exposed to the wave solder, though this is a more minor consideration.

Regarding the point about a single ground connectiont that was mentioned, you want to keep in mind how current will flow. The idea is to avoid multiple ground paths into or out of a circuit or region of the board as these cause ground loops which can cause ground voltage differences in the board and open the window for EMI.

Hi Noway2, first of all thaks to you and to Mr. Wizard for your answers. About what you wrote, you are right: I forgot to write about dec capacitors. I am using a 100nF capacitor (ceramic multilayer) on each Vcc pin of the ICs used in the 2 boards.

I still have a question: how can I estimate correctly the bandwith "to be concerned about" about my digital signals? I mean, the digital signals involved in my boards are:

lines for relay switching -> very slow switching line for inputs -> also there I think the signal is enough slow

serial communications:

rs485, working at 9600bps I2C and 1Wire bus, same speed

so which is the correcnt approach to estimate bandwith for the impedance of the ground (power) system?

thanks again

Liuc

Noway2 ha scritto:

The most useful concept to understand is 'loop area', part of which is the idea of 'return currents', as discussed in another msg this thread. By minimizing loop area, you will be minimizing inductances. Large copper pours, including ground planes, provide one way of minimizing loop areas. However, on a 2-layer board, it can be difficult to avoid cutting up the ground pour, so that there are numerous 'moats' and 'slots'. Study this subject until you understand why such features should be avoided where possible and how you can route traces on the other layer so that return currents are not forced to go around the moats and slots. Keywords for your search are: circuit board, layout, inductance, loop area, moat, slot. There are also many excellent articles and books by authors including: Mark Montrose, Howard Johnson, Sanjaya Maniktala. Paul Mathews

Since your board doesn't have any low-level analog stuff, you may as well just pour solid copper on the bottom side. You'll certainly also need some signals on the bottom, so arrange for them to be generally short so the ground plane isn't terribly chopped up by their clearances, which it sounds like you've done. There's no advantage to deliberately sectioning the ground plane into regions for different functions and it can cause problems. The whole "star grounding" thing is mostly popular superstition.

Pouring topside ground usually doesn't help, as all you wind up with is a lot of slivers of copper not doing much except making solder-bridge shorts and being confusing in general.

I could post a couple of pics to a.b.s.e.

John

To estimate the bandwidth of a digital signal, you can use a concept called the knee frequency. Roughly speaking, the knee frequency is the point (frequency) at which the energy content of the signal starts to drop off significantly. When you use the knee frequency concept, you can't treat it as a hard and fast rule, but it can tell you if you are in the clear, have something to consider, or have an outright problem.

If I remember correctly, the formula for the knee frequency that I use is 0.5 / Trise (in ps), which is the formula I picked up from Dr. Howard Johnson's book. There are probably other definitions. So if your logic signal transisitions in 2ns (2000 ps) the Fknee is .5 / 2000 x10-12 or 250Mhz. This means that in order to avoid distorting the signal edges, which can be critical in digital systems, you need to make sure that your board is capable of handling an effective bandwidth of 250Mhz as there will be significant engery content below this point.

To understand why there is energy content up to this point, think of a Fourier representation of a digital pulse, which consists of a sum of odd harmonics. The higher order harmonics are what gives the pulse its sharp edges and hence the sharper the edges, the greater the significance of the higher order harmonics.

These concepts also get into why when you probe a circuit that you alter its behavior and why your o-scope isn't necessarilly giving you an accurate picture. If you have a scope with a 100Mhz bandwidth you will be missing the higher frequency effects.

I believe Howard uses 0.35/Tr with the stated assumption of a Gaussian frequency response in the measurement system. With contemporary instruments, this is sometimes a little overly pessimistic... see, e.g.,

There aren't really any hard and fast rules - one thing to keep in mind is current paths. Current always flows from the power supply through the load (your circuit) and back. When you're laying out your power supplies and parts, visualize which copper will have the current flowing from and to the supply, and try to minimize the interaction between those current flows for different supplies. I think this is the idea behind the "single point" or "star" ground. If you watch your layout, that shouldn't be necessary. Where is the current flowing is what you need to watch.

Also, I've always said you can't overcapacitate. 10uF or more at the power entry point and a .1 ceramic (or more, depending on power/ground pins) at each chip has always worked for me. :-)

Good Luck! Rich

Here is a link from National Semicondutor's website on layout switcher supplies. Note how single point grounding is used.

Rich Grise wrote:

That's just because it's easier to do the layout. ;-)

Cheers! Rich

Hi again, and thanks to eveybody who answered me: particulary to the posts about signal bandwidth. Now I don't have cleared all my doubts but I have things more clear.

thanks again,

Liuc

Noway2 ha scritto:

I recall having seen this relation in Tektronix application notes relating the bandwidth of a scope to the minimum distinguishable rise time of that scope in the early 70's. The caveat about Gaussian frequency response sounds familiar, too.

John Perry