You need to test, we're here to help.

You need to test, we're here to help.

19 January 2018

Bandwidth vs. Current Load in Power-Rail Measurements

Connecting a 6" length of coaxial cable between a low-impedance DUT and a 1-MΩ produces ringing artifacts on your signal acquisition
Figure 1: Connecting a 6" length of coaxial cable between
a low-impedance power rail and a 1-MΩ input impedance
produces reflections and ringing artifacts
on your signal acquisition
Among the various challenges we've discussed in measuring noise on power rails are RF pickup and signal-to-noise ratio (SNR). Here's another: how do you achieve high bandwidth in your measurements while also minimizing current load on your DUT? Given that your DUT is a power rail, you really don't want to draw too much current from it. But these two measurement criteria are at loggerheads with each other. It's a quandary, and it has to do with the fundamental nature of signals on interconnects.

Let's say you have a coax cable on your probe. Your oscilloscope's input impedance is 1 MΩ, and you're probing a power rail with low impedance. If any transient should be launched from that rail into your probe, it encounters the 1-MΩ input impedance and reflects back, setting off a cycle of ringing (Figure 1).

How much ringing you'll see depends on the length of your coax cable vs. the bandwidth of your oscilloscope. If you wanted to push the ringing frequency above and beyond the bandwidth limits of your, say, 1-GHz oscilloscope, you'd need that coax cable to be so short that it would be rather impractical. It'd need to be less than 3" long. Any longer than that and you're going to see artifacts of that ringing on your display if you're using the instrument's full bandwidth.

So for practicality's sake, you're going to have a longer run of coax. And whenever you have an impedance mismatch between the oscilloscope's 1-MΩ input impedance and the impedance of the power-rail DUT, you will also have reflections and consequent ringing. As a result, the highest bandwidth you can measure to without artifacts is going to perhaps be lower than you might like.

How can we get around this ringing problem? Simple: We use a 50-Ω input termination on the oscilloscope. Such terminations are designed for just this purpose, to terminate reflections in the cable.

But here lies the quandary. If you use a 50-Ω input termination on the oscilloscope, that comprises a 50-Ω load on the power rail. It's going to limit the voltage amplitude you can observe in a power rail, because connecting that oscilloscope's 50-Ω input termination to the power rail is going to draw about 100 mA from the rail, and that's the power-dissipation capability of the 50-Ω resistor inside the oscilloscope.

Plan B, then, is to use a 10X attenuating probe. It's got a 1-MΩ input to the oscilloscope, so it won't load down the rail. But we also know that the 10X probe is going to rob you of 20 dB of SNR. It's not uncommon for people to use a 450-Ω series resistor at the probe tip to make a "roll-your-own" 10X probe. The load sees 500 Ω, the coax still has its 50-Ω termination, so they're both happy. But again, we've introduced 10X attenuation and have sacrificed SNR on the altar of impedance matching.

Thus, the bottom line here is that using a coax-tipped probe gives us the ability to measure high bandwidths, but to do so, it wants to see a 50-Ω load. But in turn, that loads down the power rail and essentially bars us from probing a power rail carrying more than 5 V. Sometimes, you just have to observe speed limits when you drive, too, and that's life.

Previous posts in this series:

Understand RF Pickup When Measuring Power Rails
How 10X Attenuating Probes Kill Signal-To-Noise Ratio

1 comment:

Unknown said...

If you don't need DC coupling, use a series capacitor to AC couple the signal riding on the power rails. You can buy high bandwidth AC coupling capacitors. Then you can terminate the 50 Ohm coax cable in 50 Ohms without burning out your termination resistor. Keep your connections to the power rail as short as possible.

Post a Comment