Oscilloscope 50Ω termination

How good does it have to be?

When measuring any really high-frequency signal, it should be properly terminated. In our world, the de-facto termination is 50Ω resistive, but most oscilloscopes (at least ones I can afford) have an input impedance typically around 1 MΩ and 15 pF. That will cause reflections back into the cable, playing hell with the signal itself, mucking up amplitudes, and possibly causing the signal source to misbehave. Which is why BNC feedthrough terminators exist: plug this into your scope, and plug your signal cable into this. Now the load impedance your signal source sees should be very close to the desired 50Ω.

As always, there are good feedthrough terminators, and there are cheap feedthrough terminators. Which should you choose? How good does a feedthrough terminator need to be? Like, for example, would a cheap BNC tee and a cheap BNC terminator suffice? Like the ones used up to the 1990s in coaxial 10base2 Ethernet networks? Like the ones I gathered by the dozens as my university was upgrading its networks from coax to CAT5 10baseT and 100baseTx? Would those work? I decided to test.

	These are the BNC parts I tested. One standard BNC tee (and I don't expect there to be too much variation from part to part here) and one standard BNC 50Ω terminator (which may be worse—I've actually bought a few brand new, only to find them open circuits instead of 50Ω). Unrelated story: One of those open-circuit terminators happened when I critically needed it on a 10base2 network, and I did not have a spare on hand. I did, however, have a BNC–RCA adapter, an RCA plug, and two 100Ω resistors to solder to it. Worked like a charm!  :)
	I also used a 10 cm or so coax cable to connect these parts to my RigExpert AA-1000 antenna analyzer and to my Siglent SDS1104X-E oscilloscope. These ridiculously short (but sometimes useful) BNC cables are also a relic of the 10base2 era. Whereas nowadays you have the RJ45 socket in your office wall, feeding into the CAT5/6/whatever going to the nearest Ethernet switch, back in the day you'd have two BNC sockets in the wall. One would lead via RG58 cable towards the nearest Ethernet hub, switch or router, the other would lead via RG58 towards the next computer in your network segment. (For the young folk: That's how 10base2 networks were built, in a chain going from one computer to another, each computer connected to the chain through a BNC tee! And there were 50Ω terminators at both ends of the chain. You see how that works from the signal's point of view?) If you removed your computer from the network, you'd unplug its two BNC cables from the BNC wall sockets, and connect the two sockets together with one of these tiny cables to keep the network in one piece.
	I first tested the BNC terminator alone by connecting it directly to the analyzer, with just an N to BNC adapter between them. This SWR curve from zero to 1000 MHz shows how good (or bad) the terminator itself is. With an SWR just short of 1.5 at 1 GHz (at far right) it's actually not too shabby! At 200 MHz (the bandwidth of an SDS1104X-E if it is hacked to open up its full bandwidth, turning it into an SDS1204X-E in everything except the print on the front face) the SWR is below 1.1!
	I then attached the terminator to the BNC tee, and that (using the short BNC cable) to the analyzer, as in the above photo. This SWR curve is from that combination. Not really too different from the terminator alone, so it would seem the tee and the cable are fine. Interestingly, the very lowest-frequency part of the curve is now very close to 1:1. Losses would explain that, of course. Even with the losses, the curve then begins to rise more steeply, so apparently the cheap cable and/or tee do have some effect, but only at significantly high frequencies (above several hundred MHz). And even a good tee will act as a stub at high enough frequency, so this is no surprise really. But I'd have to get a calibration kit (with the necessary open, short and good 50Ω terminations) in order to dig deeper into this. But see how the SWR is pretty close to 1:1 all the way up to almost 300 MHz? (The center frequency being 500 MHz.)
	Now add the oscilloscope. Finally, I attached the "foot" of the tee to my oscilloscope. It was, of course, switched on. Also, since a click is audible between some voltage ranges, indicating attenuators being switched in or out of the circuit, I measured the SWR at each attenuator setting. This one is with the voltage range of 2–10 V/div (at 1× probe setting). As you can see, this curve is way steeper than either one above! The SWR reaches 1.5 already at 200 MHz, and becomes a rather awful 4 by the time 1 GHz is reached (although such frequencies will hardly be of concern, unless you can afford a gigahertz bandwidth oscilloscope—in which case you can afford a more expensive feedthrough terminator as well!).
	This one is with the voltage range of 0.2–1 V/div. It's almost identical to the one above. I had to go back and check my settings, but yes, this is actually correct. I didn't just take a photo twice of the same SWR curve.  :)
	And this one is with the 0.5–100 mV/div range! Obviously removing the attenuators entirely from the signal chain caused a much bigger difference than just changing the amount of attenuation. And it looks like there's some kind of resonance just below 500 MHz. But that frequency is already a tall order for this 100 (or 200) MHz scope. What's actually relevant is the increase in SWR to about 2 at 200 MHz.

So what does all that mean?

In my opinion, it means the cheap 10base2 tees and terminators are just fine for use at least up to 200 MHz! The SWR was extremely well behaved in that frequency range... until the oscilloscope itself was added to the circuit! That means the measurements may actually be degraded somewhat, but mainly due to the oscilloscope, not the cheap tee or terminator! So it seems to me that there's no point in investing in hugely expensive inline terminators, if the oscilloscope itself messes up the load impedance anyway. Now, if the oscilloscope itself had selectable 50Ω and "high-Z" input impedances (as some expensive oscilloscopes do), those might work much better, as the 50Ω option might be designed to compensate its input capacitance. (I say "might" because I don't know if that is the case. I've never had the opportunity to play with such a fancy scope.) But as long as the scope's input is very much reactive, even the best feedthrough terminator won't work better than cheap, repurposed 10base2 hardware!

If you do need better performance, and if you're not measureing ultimately low voltage signals, you could get a high quality attenuator instead. Feed the signal of interest through that attenuator to the tee/terminator/oscilloscope combination, and the SWR will immediately improve. And in fact, a 20 dB attenuator (which corresponds to the 1:10 amplitude ratio of your typical oscilloscope probe) should work perfectly well as a feedthrough terminator as it is—without any final termination following it. With the other end left open, it will present a 51Ω impedance to the source (calculate it yourself if in doubt), which means an SWR of just 1.01:1! And an attenuator will find plenty of other uses as well!

For what it's worth, I did some spot measurements of the complex impedance at a couple of individual frequencies. Yes, it does seem to become inductive instead of capacitive at higher frequencies!

Frequency:	1 MHz	10 MHz	50 MHz	100 MHz	200 MHz
2–10 V/div:	56.7−j0.4 Ω	55.9−j4.1 Ω	45.7−j8.6 Ω	36.4−j3.2 Ω	36.7+j19.2 Ω
0.2–1 V/div:	56.8−j0.4 Ω	55.8−j4.1 Ω	45.7−j8.6 Ω	36.4−j3.1 Ω	36.6+j19.4 Ω
0.5–100 mV/div:	56.8−j0.4 Ω	56.2−j4.0 Ω	44.3−j12.1 Ω	30.1−j3.1 Ω	36.5+j34.8 Ω

So there. I can stop browsing eBay for inline terminators, because they certainly won't be an improvement over what I already have. (And knowing typical eBay quality, they'd probably end up worse instead.)