50 ohm termination

You can get clean wideband transmission with any impedance at the source and a 50 ohm termination at the end. Voltages will be the same all along the line. Assuming theoretical lossless trans lines of course.

You can also source terminate, namely drive a 50 ohm line from a 50 ohm source. The signal at the end of the line, unterminated, will be identical to the source internal voltage. But intermediate points will be awful.

For extreme cases, terminate both ends at 50 ohms. That will minimize reflections caused by imperfect connectors or terminators. It wastes a lot of power.

You can Spice all this.

Depends.

In general, when you just have one source and one load, with no taps along the way, the best approach is to series-terminate the source.

That is, you put a 50Ω resistor in series with the output connector, and let the load be essentially an open circuit.

The load sees the full signal amplitude, and there’s no huge power dissipation.

Anyplace in between, the waveform is more complicated.

Cheers

Phil Hobbs

Thank you for the advice. In my case I have a 10Mhz signal with very sharp transitions (500ps, 5V) and wanted to make sure I did things properly.

Doing things properly may not be too difficult. Light travels at about one foot (30cm) per nanosecond. Longer electromagnetic waves travels at about two thirds of the pace in coax insulation, so 500psec is about 10cm.

You don't have have to make you terminating resistors all that compact to have them behave predictably with those sorts of relatively slow edges.

Faster stuff can call for symetrical arrangements of compact surface mount resistors and that can get messy.

Double termination - at both source and destination - will give you the nicest looking waveforms, and since you would only see 2.5V at the destination the 125mW you'd have to dissipate at the source and the destination shouldn't be hard to cope with.

Interesting. What's generating the 5v signal? Lots of AC and Tiny Logic chips are that fast, but might strain to drive 50 ohms. We use several tiny triple buffers in parallel sometimes.

Regular thick-film surface-mount resistors are fine as terminators at

LVDS line receivers are great at the receive end.

I used a THS3111CD. Split up my project into a timing and driver board, so i have some 50 ohm BNC cables between and wanted to be extra safe about reflections at these higher frequencies.

Hmm. 5V in 500 ps is pretty good going for a part with 1300 V/us slew. ;)

Cheers

Phil Hobbs

I'm an expert at misreading the divisions on my scope ill have you know :)

I had a really hard time finding a good line driver. I might look into this chip.

Thank you!

Is the signal some analog thing, or a 10 MHz clock? The THS is an opamp, but they can make good cable drivers too, even for clocks.

Lately I'm enamored of BUF602, a unity-gain 1 GHz beast.

Is your signal analog or digital?

digital

I use Tiny Logic triple buffers as line drivers, with all three sections in parallel, and then sometimes two or three chips.

NL37WZ16US costs 10 cents.

Discrete surface mount devices can be pretty compact, and there are some fast integrated circuit devices designed to drive them.

Some of the ECL-to-TTL level shifters did generate a very fast full 0V to 5V swing. I got stuck with up-dating a very fast TTL-based timing circuit in the early 1990's, and used a bit of ECLinPS ECL to get rid of the usual TTL faults, and used 100k ECL-to-TTL converters to push out the TTL house-keeping signals.

They were a lot better than the original TTL signals

It involved adding -4.5V rail to drive the ECL, but with surface mount parts we could squeeze the additional stuff onto same sized printed circuit board that the original system had used.

fyi, im an amateur and may have made a mistake in my design, reason for

From what everyone is saying I dont even need to do that and can get a reliable termination just by using one 50 ohm resistor and avoid dealing with the voltage divider side effect.

the hard part is coming up...soldering these tiny smd components...i bought a microscope and a little platform to hold the boards. going to try hot air soldering as i havent shelled out for a reflow oven yet.

the logic chips im using in my project are 74VHC series.

I can sympathise with the difficulty of coming to terms with soldering smd chips. Around 1989 I got stuck with introducing smd parts to Cambridge Instruments in the UK, because the GaAs chips that I needed to use only came in surface mount packages. We bought a fairly expensive Groatmore hot-air reflow machine that would reflow individual packages.

When I wanted to use similar parts - Motorola ECLinPS devices - at Nijmegen University in the Netherlands, nearly ten years later, they just bought a much cheaper and smaller work station (but didn't let me use it). At Haffmans BV in the Netherlands around 2002 I just used a fine tipped soldering iron under a cheap binocular microscope.

It was fiddly work, but perfectly practical.

Very fiddly, my hands shake too much for stuff this fine. I have a driver and a mosfet:

BD2311NVX-LBE2 - driver GAN190-650FBE - mosfet

That driver is so tiny I'm wondering if the 10Mhz switching speeds are worth it! Researching some strange field interactions so I needed something that could switch a decent voltage (300V) at repetition rates close to 10Mhz. Odd requirements but until I can narrow down parameters I need to sweep up the frequency range as far as I can.

Thanks for all of your help.

When I was blanking and unblanking an an electron beam in a electron microscope, the fast version - which turned the beam on for just 0.5nsec with 100psec transition times - only generated +/-7.5V.

The slow version - in an electron beam microfabricator - had to swing from 60V to OV at up to 10 MHz, because we couldn't get the blanking plates all that close to the beam in the electron beam microfabrictor (which was a million dollar machine, and we only sold a handful per year

- ten per year in the good year).

On 2025-03-19 10:41, john larkin wrote:> On Wed, 19 Mar 2025 06:23:25

-0400, Toaster snipped-for-privacy@dne3.net wrote: >

Just rereading this. John, the prop delay spreads in the datasheet are all over the place--almost a factor of 2 from typical to max over temperature.

I'd expect the three sections to match OK, but paralleling packages seems quite a lot sportier. How well does that work in production?

Cheers

Phil Hobbs