DIY Reference 10 MHz Distribution Amplifier

I wanted to get a distribution amplifier for my 10 MHz lab reference signal from my GPSDO to feed my equipment, like my timer/frequency counter, signal generator, and other stuff. I also needed a way to distribute a 1 PPS signal from my Trimble GPSDO to my NTP servers and other clocks and monitoring systems. The Trimble only got two 10 MHz outputs and one 1 PPS signal output, so I needed to add more, especially for experimentation.

I saw this video by Gerry Sweeney where he modifies an Extron 300MX video distributions amplifier, adding a rubidium reference standard and was intrigued. But I changed the modifications to fit my needs and here is the result.

I got hold of an Extron ADA 4 300MX HV, a model where there are only four outputs per inputs, instead of six for the RGB and sync inputs. As you can see in the picture, there are plugs in the extra two outputs per row. But it was easy to upgrade to the more prominent model by desoldering a termination resistor and solder in a new 50 Ω BNC connector. But my first order of business was to try to lower the noise of the switching power supply inside the unit, feeding the Texas Instrument CLC409[PDF] opamps used to distribute the RGB channels.

I could have made it easier for myself by replacing the switching power supply with a linear one. But I was curious about how much work it would be to try to filter out the switching noise, and my experience working with switching power supplies are very limited. So it was an excellent excuse to read up on how they work and how to get the noise level down.

Switching Noise

The power supply on the backplane has an LT1376 1.5A, 500kHz Step-Down Switching Regulator. But the electrolytic capacitors and the inductor are on the same side as the connectors, which makes space a problem when they designed the power supply. There’s no room for large capacitors. There’s also one 7805 and 7905 DC voltage regulator for the +-5 volt rails, where they’ve put a ferrite bead on the negative 7905 input to keep the noise down. And there’s always more noise on the negative rail.

First order of business was to find a good point to do the noise measurements. I wanted the test point as close as possible to the op-amps, so I made small coils of wire that I soldered to the ground side of one of the decoupling SMD capacitors, so I didn’t need to use a long ground lead on the oscilloscope probe. It acts like an antenna and picks up a lot of noise, making noise measurement impossible.

Then it’s just a matter of sticking the oscillator probe in the coil and put the measuring pin at the voltage rail side of the SMD capacitor.

Even after passing through the 7805/7905 DC regulators, the rail was noisy. They have a decent PSRR, but we’re talking about switching noise at 500 kHz, so a lot of it goes through the DC converter. There is decoupling very near the op-amps, so some of the noise gets filtered out, but I added low ESR catalytic capacitors (Panasonic FM-Series) on the back side of the board, where there’s lots of space.

I also added some storage buffers on each rail by soldering 10 µF tantalum capacitors on both the positive and negative rails on each op-amp. Naturally, I put one of the tantalum capacitors the wrong way, so I had a beautiful explosion when I turned on the power. Holy Moly, how tantalums goes up in magic smoke! It fried the CLC409 in the process. So I had to use my SMD desoldering heat gun to remove the faulty op amp and start to find a replacement. The LMH6702 op-amp has replaced the CLC409 as it is now obsolete. An improved version with better specs, so I ordered two and replaced both op amps on the upper row to get six outputs with the “nicer” ones.

1 PPS Distribution

My first thought was to use the sync circuit already on the board, just like Garry Sweeney did in his video, but the output looked horrible! First a lot of switching of the signal back and forth, but also a lot of pre and post ringing. So I decided to rebuild the 1 PPS distribution myself. All I wanted was input with hysteresis and no ringing if possible. So I used a 74AC14 Hex Inverter with Schmitt Trigger Input [PDF] which has hysteresis on the input signal and makes an output signal with speedy rise time without ringing. I then took that inverted signal and fed it into a 74AC04 Hex Inverter and connected two outputs with a 100 Ω resistor on each output for more drive capability. My initial plan was to use four outputs with 200 Ω resistors, but two seems to be enough.

Naturally, there will be some propagation delay, but by using the 74AC line, which is much quicker than the regular HC series of IC, you get a faster circuit. The delay is higher than I would have liked, but it should come down when I make a PCB using SMD components and not the hodgepodge I’m using now. The outputs of the board, mounted inside the box are connected with long wires going directly to the outputs. So I can shorten the cables considerably when I get the PCBs by installing it directly on the main PCB.

The output signal looks nicer on the scope than the signal coming out from the Trimble GPSDO. The upper, yellow line is the output from the buffer; the bottom blue signal is from the GPSDO. No more ringing.

I now feed the GPSDO through a box that widens the pulse width of the PPS signal from 10 µS to around 50% duty cycle.

Future Modifications

I’m waiting for some ferrite beads that fit on the legs of the 7805. So one on the input and one on the output. I have also built an LC filter that I’ll put in front of the DC regulators to minimize the switching noise. There is an excellent article by the late Jim Williams called Minimizing Switching Regulator Residue in Linear Regulator Outputs [PDF] where different ways of minimizing switching noise are discussed and measured. A great read.

Output Bandwidth

I’m not going to use the amplifier for anything higher than 10 MHz, but I did measure the bandwidth as far as I could go with my equipment, which is 120 MHz. I fed the input with a sine wave with a 1-volt peak to peak with 50 Ω termination with my Siglent SDG 2042X and measured the signal on my Rigol 1054Z. I increased the frequency, waiting for the signal to go 3dB down, measured at 0.707 volts on the oscilloscope. The signal generator can output sine waves to 120 MHz, and the scope has a bandwidth of 100 MHz, but it works up to 120 MHz (I’ve hacked the scope and the signal generator). The lowest measured value was around 0.900 Volts, so the bandwidth is a lot higher than 120 MHz.

And at 100 MHz

There’s plenty of bandwidth if I need it. That is not surprising when checking the data sheet for the LMH6702; it had 720 MHz bandwidth at 2 Vpp. The old CLC409 op-amps have a 350 MHz bandwidth at 2 Vpp. But op amps with this kind of specs are notoriety fickle, so they oscillate easily. Decoupling is paramount.

So far I’m happy with the results. I want to get the noise floor down on the positive and negative rails as far as possible. Not because I need to, but because It’s fun to improve and measure. All the equipment locks to the lab reference signal without problems, and my Siglent signal generator outputs nicer sine waves when fed through the amplifier at higher frequencies than when using the internal oscillator. And it’s bang on when it comes to accuracy.