Suppressing switching noise (spikes)

[abfer]

I have some problems with switching noise at hard switching devices. I see the effects of this, for example, at the output of a dc/dc converter, or when i try to measure ac motor current etc. There are spikes on the signals that i measure, at the switching frequency. How can i eliminite these spikes especially on measurements?

 

[Skogsgurra]

I assume that you are discussing some kind of a semiconductor switch and not an electromechanical one. There are lots of information on how to suppress switching transients from the latter. RC snubbers, MOVs, zeners, diode-resistor combinations are the ones used.

With semiconductor switches like IGBTs, thyristors, MOSFETs and so on, there are usually free-wheel diodes built in, either in the device pr parallel to it. There may also be snubbers.

What you can do to eliminate switching noise from a measuring device is to limit band-width. You can usually do it by putting a resistor in series with the signal and then - between resistor and input - a capacitor between signal and signal ground.

Make the R about 1 percent of input impedance to avoid large error contributions (e.g. an oscilloscope input with 1 megohms shall have around 10 kohms). Then make the RC time constant equal to one divided by 2PIxf0. Example: you want 100 kHz band-width and have a 10 kohm resistor, then chose RC = 1/(100000x6.28) = 1.6 microseconds, which makes the capacitor 1.6/10000 = 0.16 nanofarads.

You may also need to use differential input and build a symmetric low pass filter. You can use the same math as above, but you split the resistor into tw equal parts. For better noise reduction, you may need to use an active filter with more poles. But that is not something to go into details on here.



Gunnar Englund

http://www.gke.org

 

[zappedagain]

These spikes are often referred to as 'ringing' or oscillation. From a controls perspective, you have an underdamped system. Gunnar is correct that adding some series resistance will help to dampen out the oscillation.

A long GND lead on your oscilloscope probe can cause you to induce ringing at the scope due to the lead inductance (your scope may show ringing that isn't really there). Whenever I see ringing in the MHz range the first thing I do is to shorten my GND lead to about 0.5 inch (12 mm) or so; a standard 3" GND clip can introduce a lot of ringing. Most of the quality probe manufacturers include a GND-spring-clip just for this.

Also, the spikes are mainly driven by the rise-time and fall-time of the signal, not the signal frequency. If you can 'soften' your hard switch by increasing the transition time that will help too. The price you pay for that is that your switch has to dissipate much more power during the transition.

 

[abfer]

Switching frequency is 10KHz. Hard swtiching device is, igbts that make hard switching, in present case. Spikes or ringings occur at rising or falling edges. I already use low pass filters with a corner frequency of 1KHz at the output of sensors. igbts have internal diodes and capacitor snubbers. Why doesn't low pass filter eliminate this spike? Doesn't it fast enough? I used ordinary capacitors in filter. Do multilayer capacitors help me in this case? Or using mov helps me? If so, Where to use mov? In power stage or in measurement stage? Also movs degrade in time as i know but i don't familiar with them so much. So what about reliability when i use mov? Does it require another co-protection circuit?

 

[Skogsgurra]

If you are using low pass filters with 1 kHz corner frequency and that still doesn't work, then there is something wrong with the setup. You are saying that you have put the filters on the output from the sensors. I think that it would be much more appropriate to have the filters on the sensor input - if they are not LEM or similar current sensors, where no filters can be used.


DO NOT add any filtering or suppressing components to the IGBT circuits. They are usually designed to work with the protection they already have. Adding components can be very bad.

MOVs do not help in this case. Re their degradation, it is true that a MOV that receives energy pulses close to its maximum rating degrades quickly - a few hundred or thuosand pulses may kill it. But if designed with a good safety margin (energywise) it will not degrade at all. There are (used to be) some very good application notes that describe this on the Siemens-Matsushita site.


Gunnar Englund

http://www.gke.org

 

[abfer]

Sensors are lem type. So i can't use filtering at the input as you've also mentioned. Actually output voltages are low. Peak to peak voltage of the output with spike is ~500mV but the sensors have a output of as low as 20mV to 2V. Spike's magnitude is more than the signal in lower currents but its effect disappears as current goes up.

 

[Skogsgurra]

What you describe is what we see in almost all situations where we want to record the current signal directly from the LEM output. It is very much about shielding, grounding in the right place and filtering. You simply have to learn by doing. Remember that you have a very intense HF transmitter (the IGBTs act as transmitters with band-width up to and above 10 MHz) close to where you measure.

The best thing to do is usually to use a differential amplifier with filtering. The next best thing is to make sure (as zapped says) that your ground lead is VERY short and that GND is taken close to where the LEM output is connected to ground. And, of course, pick up the signal close to that point.

Gunnar Englund

http://www.gke.org

 

[abfer]

-Signal output is close to ground.

-Does, putting sensor circuits in a shielded box, help?

-I didn't understand how differential amplifier will help.
Do you assume that ground has noise?

-By the way, i make the measurements by an ossilloscope which has isolated differential probes.

-Do transzorbs help? Do they degrade as MOVs?

 

[Skogsgurra]

As I said, you now are in a phase where learning by doing is the next step.

The answers to your question are:
1 I know.
2 Since they usually are internally shielded - I do not think that it would do any good.
3 Yes. Ground always has lots of noise in switchers and frequency inverters. That noise is producing a normal mode signal if allowed to develop a current in the ground lead. That's why a differential amplifier/probe is good. But it has to be symmetric. Same resistance and capacitance in + and - inputs. Same frequency response in both channels.
4 Good. You should not have these problems then.
5 No, they do not help at all. They are used to absorb transient overvoltages. Not interference on a signal.


Gunnar Englund

http://www.gke.org

 

[abfer]

Thank you for your interest, so much. I further want to say some points. When i use averaging feature of te scope (It takes several samples and calculates the average)the spikes get lost as they are very narrow according to the whole actual signal. Can't we realize this with an analog circuit?(Actually i still don't understand why a filter or a capacitor doesn't absorb this spike even i use mkt or ceramic capacitors.)
When i change probe factor from x10 to x1, which means the gain of x10 of the signal, spikes still occur but their magnitude is small according to the signal. Can we use this property with a similar behaving circuit and does it help us to eliminate this noise?

 

[Skogsgurra]

The averaging feature of the scope is usually exactly what a low pass filter does. (There are two different approaches to averaging, one does take the moving average and the other one simulates an RC filter).

So, the RC filter should be - and probably is - doing the same thing as the average function in the scope.

The reason that it doesn't work quite as expected is usually that there are stray signals getting into the wiring or that there is a normal mode signal in the ground wire.

As I said, you are on your own now and have to find out for yourself. Think about the circuitry not only as a resistor and a capacitor but also lots of parasitic coupling capacitances and inductivities that are spread all over the wiring. As a rule of thumb, you have about 1 microhenry and 100 picofarad for each metre of wire.

Gunnar Englund

http://www.gke.org