PCB layout problems. 1

[morizabal]

Hello Everyone,

I am fairly new to circuit layout. I've designed a simple strain gage amplifier with a 0 to 10 vdc output. The circuit operates as design with one problem. When put my hand near the sensor wires I get a shift in voltage.

1. What causes this?

2. How can I fix it?

3. What kind I do for the next revision to prevent it?

4. Any resources I can look up?

Circuit is a two layer, two sided SMT with poured ground plane on the bottom side.

Any help would be greatly appreciated.

Mario

 

[buzzp]

I assume the output is where the shift in voltage is occuring. But most likely it is capacitively coupling problems via your hand on the low voltage part of your circuit. A few basic questions: were you in a room with fluorescent lighting? Are the traces and wires short in length, wires twisted to the sensor, and are you using a differential amplifier to 'buff' the signal from the sensor to cancel out common mode noise? Just some things to think about. Omega engineering has some information that you may find useful. A search engine such as google.com will provide a lot of useful information. PCB trade magazines also keep a archive of useful articles, etc,.

 

[melone]

Ground pour the top side as well!!! It won't hurt to have the extra grounding on top! Also, do you have large value resistors (in the Megaohm range)??? If so, when you capacitively couple to the PCB, you will be creating LP filters and start attenuating your signals.

Just a thought!

 

[buzzp]

I am not sure I would have two ground planes. I have heard good and bad about doing this. Maybe you have a case where it would help. But chances are it won't help a thing unless the board is fairly small. Two ground planes on a two consecutive planes can interact with each other through the capacitor that is formed between the two. You may look at the offset voltage charactersitics of the op amp. How much does the voltage shift?

 

[melone]

To truly elimate any problems, simply go to a 6 layer PCB with a dedicated ground and power plane. Then by ground filling the top and bottom layers and routing as many singals as possible on remaining 2 inner layers, then you shield all of your traces from the outside world. However, it sounds like you are stuck with 2 layers. Hopefully, by creating as much ground on both layers (I am assuming that you keep the grounds tied together well [multiple vias per section]), then you should be able to help squelch some of your problems.

 

[buzzp]

If it is going to be in an enclosure then you could get some spray coat to spray inside of the enclosure to help this situtation or spray it with conformal coating. This coating should increase the dielectric of the supposed capacitor and this may eliminate the problem. I should say that everything I have said is purely speculating without knowing all of the details. Your problem could be from a number of different things.
If you make a 6 layer board you better be prepared to shell out some righteous bucks and do not be surprised if you have additional problems. Have you ever tried to diagnose or troubleshoot a 6 layer board for EMI problems? Its a nightmare.
Good luck.

 

[melone]

I have a few in production and they were a drastic improvement with regards to a 4 layer version. However, you are correct with regards to troubleshooting RF and cost. Playing with this stuff is like trying to compress an inflated balloon. You can squeeze in 2 sides but that causes the remaining sides to pop out. RF is as much art as science.

I don't disagree with anything that BUZZP has said, because he is correct. However, all of these suggestions whould be taken with a grain of salt because they might not work in you application! If you are operating in the GHz range, than 2 layers can NEVER work. You need to have all of the ground isolation that you can get!!!! But, if you are running a simple 555 timer that oscilates at 2 Hz, 6 layers is over-kill. I guess, what I am trying to say, is that we will all try to give you suggestions on what could "fix" your problem. Don't be alarmed if some of our suggestions seem to contradict. We all have prejudices based on our past experience.

Good Luck and keep us posted!!!

 

[melone]

One other thing, are your sensor wires a differential pair? if so, are they routed on the PCB next to each other? Are the track lengths the same? etc..

Just a thought!

 

[morizabal]

Thank You For All your input. The problem is that I need to find a quick solution to this particular lot of 250 units. My boss wanted me to design and produce without first prototyping and testing. I managed to put a capacitor across the positive output and negative output of the strain gage and it greatly reduced the effect. I want to get pointers on how to improve the layout of the next revision.

In response to the first reply.

Yes, I am in a room with Flourescent lighting.
The wire to the sensor is a 6 conductor shielded cable, (about 2.5 feet)
I am using a differential instrumentation amplifier (Burr-Brown INA118)
Output is tied to a voltage divider for trimming, and then tied to a unity gain buffer.

For my next revision I will probably go with a 4 layer board. 2 power planes, and 2 trace layers (top/bottom)

What things can I do to improve my circuit.

Basically I have a 12 volt regulator and a charge pump to provide a -12 volts. These voltages power my op-amps and instrumentation amplifier. A 10 volt precision source provides a clean excitation to the strain gage sensor. Two pots adjust zero and span.

Thanks, Mario

 

[buzzp]

There are several things you can do. Rather then mention them all here go to one of the several web pages that describe basic layout guidelines. Try searching for PCB basics and this should turn up several good hits.

 

[electricuwe]

morizabal,

to my mind you are on the right path with your bugfix descirbed in your post dated Apr 2nd. After reading your initial post I guessed that your circuit picks up some Hf and rectifies it (e.g. in the input circuit of the amplifier). That adding the capacitors was helpful shows that this guess might be true.

To my mind your application should work well with a double sided board with a ground plane on one side, if the ground plane is only interrupted for short jumps or small components. Anything else would only add unnecessary cost.

Additionaly try to separate the charge pump from the rest of your design as good as possible and use additional LC-filters for your power supply to the amplifier.

Try to limit bandwith to the frequency range you can expect in a signal coming from a strain gauge by a passive filter before feeding the signal to the amplifier.

The ground connections might als be important: avoid to run currents from the shield through the ground plane of your circuit.

A good resource on the web can be found in application notes from analog devices:

http://www.analog.com

 

[buzzp]

Electricwue brings up an important point, filtering. This should be an integral part of your design. The fact that the problem only occurs when your hand approaches the board (how far away is your hand when this occurs and is it flat over the entire board?) would indicate a board layout origin. What kind of capacitors do you have for bypass caps on the supply to the inst. amp and buffer? You may consider using a ceramic and a tantalum cap across these right near or on the chip as a short term solution. This may be difficult in that you have smt to work with.
I know of a colleague in ran into something similar with a microcontroller resetting when his hand was within an inch or so. As a short term solution, I suggested a tantalum cap across the supply on the micro and the problem went away. They had already ordered 500 through hole boards so the kluge was easier than smt. I think your boss messed up by not ordering a proto, especially with smt for the reason mentioned above. I am sure your boss was trying to save a buck at your expense. As a quick fix, try kluging a tant 10uf cap right on the supply to the chip with minimal solder lengths to the ground plane. Drill a new hole in the board if you have too but make sure the runs are short. Good luck

 

[morizabal]

Thanks for the help,

I managed to put a ceramic capacitor across the output leads and this took care of the problem.
The change in output occured when I got about 2 inches from the cable, when I grabbed the cable the change was more significant. The circuit I design was based on an existing amplifier. One potential problem is that I have pin jumpers that allows for polarity reversal. Meaning the signal from the wheatstone bridge will have through traverse through the jumpers until finally reaching the instrumentation amplifier. On my next revision I will be putting the ins. amp. closer to the signal, including a capacitor and making a 4 layer board. Basically keeping the voltage supply/regulators on one side and amps on the other. Would this be a correct strategy. Also I did include a LP filter on the output of the ins. amp.

Mario

 

[melone]

Not neccessarily. Be careful to properly decouple all of your circuits. It would be a shame to segment up your PCB, and just run "hot" traces (power and/or ground) between segments and nullify all of your efforts. Also, make your ground plane SOLID. Many people will try and convince you that segmenting your ground plane will help contain your noise, but I can say from experience, if this practice is not done correctly, it will only hurt you. The easiest way to nullify a segmented ground plane is to use a microprocessor that recieves both analog and digital signals. This will create nasty ground loops by having a segmented ground plane.

Good luck and keep us posted!!!