Monday, September 5, 2011

Probe Attenuation - The Overlooked Specification

The other day I plugged my iPod into my car stereo.  The music was soft so I cranked up the stereo volume.  The music was now loud enough, but it sounded terrible and hollow.  Suddenly I realized my iPod volume was too low and I was amplifying a very small signal.  As I turned up my iPod to supply a large signal to my stereo (and simultaneously turned down my stereo volume), I could achieve the same net loudness while getting a fuller sound.  How does this relate to measuring a signal with your oscilloscope?

The oscilloscope industry is abuzz in chatter regarding who has the best digitizer.  One company will tell you that their 8-bit oscilloscope has more effective bits than another company's 8-bit oscilloscope.  Another company introduces a "12-bit oscilloscope" with no indication anywhere of true effective bit performance.  If you've ever attended one of my power supply measurement classes, you would know that hi-res mode on an 8-bit oscilloscope can provide >11 effective bits of resolution, and that anybody who sells you a 12-bit oscilloscope without specifying effective bits over frequency is likely selling you an impressive label more than a useful tool.

So the harried engineer wants to know which scope most accurately measures his signal, and yes, while effective bits is a good metric, there is something even more important than the oscilloscope itself that many overlook.

Before the front-end amplifies your signal, before the track-and-hold slows it down, and before the digitizer samples and stores the signal, the signal must pass through a probe.  And the most important probe number, especially for power work is......


Why does attenuation matter so much?

Imagine you just powered on your real-time oscilloscope and you need to probe your circuit.  You could just use the probe that came with your oscilloscope, a standard 500MHz 10x passive probe.  But this measurement needs to be accurate and you want to be sure you have the proper probe selected.

If you look online, there are lots of tutorials to help you pick a probe.  There's even an "app for that" on your iPhone and iPad.  There are passive probes and active probes, high voltage differential probes and high speed differential probes.  Many oscilloscope users will decide their need (single-ended or differential), their voltage, and their bandwidth, and then pick a probe.

Remember back to my post on the V/div knob.  Each click of that knob adjusts the oscilloscope's front-end gain.  It is amplifying your signal to fill the digitizer, from the top of the screen to the bottom.  Modern oscilloscopes are able to read the probe attenuation value from the probe itself through a probe sense resistor (that springy pin on the BNC end of the probe).  When the screen reads out a certain V/div, it is actually calculating it from the attenuation.

In other words, attach a 10x probe to the oscilloscope.  Most oscilloscopes come with some sort of 10x passive probe.  The industry standard is a 500MHz passive probe, although in the following screenshots I used a Tektronix TPP1000 1GHz passive probe.  Either way, the probe actually divides the signal down by a factor of 10 as it presents it to the oscilloscope.  Then the oscilloscope must reamplify the signal.  In the following screenshot, the oscilloscope reads 200mV/div.  However, the oscilloscope is actually set to 20mV/div, but the 10x probe it reads means the screen is effectively at 200mV/div.

This is also why you cannot see low V/div settings with a probe attached.  If your minimum setting is 10mV/div, then with a 10x probe, your minimum setting is 100mV/div.  Some oscilloscopes have a vertical software zoom on lower V/div settings as some do, but that doesn't actually give you any more information.

In the following screenshot taken on a Tektronix MSO4104B, I want to measure the noise and ripple on the top of my pulse.  By drawing a histogram box (in brown), I can measure peak-to-peak and rms deviation, seeing that the signal is 131mV peak-to-peak and 18mV RMS during the on-period.  If I were to try to go to a small V/div setting, I could reach 10mV on this signal as the MSO4104B has a true 1mV/div setting on it.  10 x 1mV/div = 10mV/div.  The 1mV/div setting is bandwidth limited, but it is not a zoom.

Also note that I made these measurements using "hi-res" mode and running at 500MS/s.  This should provide me about 220MHz and 9 effective bits.

To get a better measurement, I tried a different probe.  This one is also a passive probe and is 500MHz, but is only 2x attenuation, the Tektronix TPP0502.  Remember the problems I had when I tried to amplify the quiet signal from my iPod?  Turning up the signal from my iPod made it sound better on my stereo.  So by not attenuating my signal so much, there is less degradation in the same signal on the screen.  Once again, my settings will give me the same 220MHz and 9 effective bits.  And once again the scope reads 200mV/div.  But in this case, the scope is really set to 100mV/div.  Using the TPP0502, I can go down to 2mV/div with absolutely no software zoom (2 x 1mV/div = 2mV/div).

My peak-to-peak has decreased from 131.4mV to 87.5mV, my RMS has decreased from 18.4mV to 11.5mV.  The trace is visibly less noisy.  I got about a 35% improvement from probe attenuation alone.

Also note that the TPP0502 is 2x, but still a full 500MHz.  Most of the time, 1x and 2x probes are severely bandwidth limited, and provide only about 25MHz of bandwidth.

So when you want to pick a probe, make sure that the attenuation is as small as possible while still capturing your signal of interest.

For instance, suppose I want to probe a 640V signal with a passive probe.  A typical ~1kV passive probe like the Tektronix P5100A, the Agilent 10076A, or the LeCroy PPE series will have a 100x attenuation.  That means that before the oscilloscope has a chance to digitize the signal, it is reduced by a factor of 100x. Suppose I want to measure ripple on my supply?  The ideal probe will not only have the right voltage and right bandwidth, but also the least attenuation.  The best car stereo system in the world can't improve the sound from an iPod at low volume.  Likewise, having more real or effective bits doesn't help if the signal is too attenuated.

Tektronix TPP0850
For these signals, I'd want to find a probe with more than 10x attenuation, but less than 100x.  One example would be the Tektronix TPP0850 with 50x of attenuation, or even the P5120 with 20x of attenuation.  When I talk about high-bandwidth probes, the typical trade-off is between voltage and bandwidth.  For power probing, you also have to consider the tradeoff between attenuation and bandwidth..  The P5120 is 200MHz, but only 20x.  The TPP0850 is 800Mz and 50x.  Either one would do a superior job than a 100x probe in evaluating a power supply ripple.

Tektronix P52xx Series
The same logic applies when looking at high-voltage differential probes.  When working in high-voltage systems, you often do not want the ground of your oscilloscope connected to the ground of your circuit.  It creates a ground loop that can stop your circuit from functioning properly, and the excessive return current into the oscilloscope ground could damage it.  And please never "float" your oscilloscope with a cheater plug that has no ground pin - it can be dangerous and destroy your precious investment.

In these designs, you should not use a passive high voltage probe, but rather an active high-voltage differential probe.  These probes isolate the ground of the oscilloscope away from the ground of the measurement.  Many of these probes have switchable attenuation, so pick the one with switches that closely match your design parameters.  The P5202A is 20x or 200x, the P5205A is 50x or 100x, and the P5210A is 100x or 1000x.

Hopefully that gives you a little better perspective on picking a probe.  You want to be sure you pick the probe with the lowest attenuation possible that does not sacrifice the voltage and bandwidth requirements of your signal.

1 comment:

  1. Keep up the good work It is rare to find a good and interesting article like this one for us engineers.