Lots of timing-type instruments, such as frequency counters, AWGs, and oscilloscopes have the option to be tied to an external frequency reference, most commonly 10 MHz. This enables the correlation of measurements from different instruments, as well as vastly improving their accuracy and repeatability.
Impedance and Termination
It is useful at this point to mention impedance. Many single-ended clocked signals are intended to be driven into a 50 Ω load. This means the driving device will have an output impedance of 50 Ω, will need to be connected to a cable with a characteristic impedance of 50 Ω, and the signal must terminate at the receiver with an input impedance of, you guessed it, 50 Ω.
Reflections
Firstly, signal reflections can occur if you fail to maintain this impedance along the signal path, provided the cable run is long enough. This, of course, depends on the rise or fall time of the signal, and this bares a close relationship to the frequency. Generally, the higher the frequency of a signal, the shorter the rise/fall time, and the shorter the cable can be before reflections happen.
Above is a simple lossless transmission line model driven from a 50 Ω source. The driver level is set to 2V, which results in the desired signal level of 1V along the line. The output impedance of the driver and the transmission line characteristic impedance form a resistor divider by a factor of two, hence the reduced signal level. As will be seen, since the load is NOT 50 Ω, but 1 kΩ (like the reference input impedance of an Agilent 53132A) the pulse will be reflected with significant amplitude.
As an example, the Agilent 53132A timer/counter in front of me has a reference input specification of 200mV – 10V at 1kΩ, so will require an external terminating BNC 50Ω resistor, connected via a ‘Y’ adaptor. However, you would have to exceed 10V before damage would occur, and this would be quite unlikely in practice.
Tweaking the simulation to increase the pulse count and density, we can see how the waveform gets mangled by returning reflected pulses.
What you would actually see would be affected by the cable length, the degree of impedance mismatch, and the cable loss. I could have simulated with a model of the RG58 Coax cable commonly available, but that would not help illustrate these points so clearly.
Perhaps I will post about that another time?
Poor termination or cable matching can really ruin your day. Best to just terminate properly and not worry about it.
Signal Level
Secondly, without the correct load impedance, the signal level will be wrong. This can cause the receiving device to be under-driven and fail to see the signal. Worse, if badly over-driven, damage can occur if it the receiver is not adequately protected.
Timing Source
In my lab, I have a GPS receiver that disciplines an internal TCXO reference. In order to maintain a very accurate and stable frequency. This device provides 10 MHz, 1 Hz (aka 1 PPS) outputs as well as generating NMEA-0183 sentences over a serial port. This is a very useful thing to have if you care about timing accuracy.
I need (well, want) a rack mount solution to take this 10 MHz reference clock from my GPS receiver and buffer it out to various instruments that take an external reference input.
Being a hacker and a cheapskate, commercial offerings are prohibitively expensive. Offerings on Aliexpress, like this one, leave me cold.
I read on some random forum recently, about someone modifying a video distribution amplifier (VDA). The idea was simply to swap out the 75 Ω terminating resistors for 50 Ω. These VDAs are sort of similar to timing distributors; they take in one video signal, amplify and drive out onto multiple connectors.
Anyway, I bought a used one from eBay and took a look inside.
I was only interested in the video section on the left, so I ripped it open and investigated further.
Obviously, I just dived straight in and ripped it apart to take out that block of ten blue resistors at the bottom.
The Issues
Then the penny dropped. This is a single amplifier driving all outputs in parallel. That strip of copper on the left is the common side, and each output is purely just connected via its own 75 Ω resistor.
This isn’t really what I had in mind, and not really adequate for even my simple needs.
The biggest issue with this arrangement in my mind, is there is no isolation between outputs, so any noise or reflections on one output channel will propagate to all other outputs.
Output Level
The other issue that comes to mind is that the signal level would be affected. As more and more of the outputs are populated and terminated all those 50 Ω loads will drag the signal level down further.
If I were to use the VDA as it is, and just swap out the 75 Ω resistors for 50 Ω, I would end up with a situation similar to the above; ten loads of 50 Ω being driven by a single source. I chose to model the shared driver with a 10 Ω output impedance. I didn’t measure the VP722, but it doesn’t seem too far off.
If we back off a little, disconnect nine of the cables and just imagine the bottom of the circuit above, with a single transmission line (T1) for now.
First, calculate Vin :
V_{in} = 2\times{{R_5 + R_1}\over {R_{V_1}+R_1+R_5}} = 2\times{{50 + 50}\over {50+50+10}} = 1.82 VThe voltage level at R1 would then be:
V_{R_1} = V_{in} \times {R_1\over {R_1+R_5}} = 1.82 \times {50\over {50 + 50}} = 909 mVSo, all good, the voltage level is close to the desired 1V and could be improved with a better driver.
Now, connect the other nine loads, and, the signal level now would be a fair bit less; you can calculate this easily, as the effective load on the driver would be the parallel combination of all ten coax cable characteristic impedances combined with the load, or (50+50)/10 = 10 Ω
Now, that 2V pulse at the driver output, Vin would actually be:
V_{in} = 2 \times {10\over {R_{V_1} + 10}} = 2 \times {10\over {10 + 10}} = 1 VV_{R_1} = V_{in} \times {R_1\over {R_1+R_5}} = 1 \times {50\over {50 + 50}} = 500 mVThese values can be seen in the simulation results:
To alleviate these concerns, I want a proper driver per channel. I don’t really have any choice but to ditch the PCB and make something specific.
The Plan
I can reuse the case, the nice little toroid transformer. It makes sense to use a linear power supply to keep the power noise to a minimum. Switched-mode supplies can add a lot of unhelpful noise if inappropriately applied to a circuit. The metal case will provide useful shielding to keep external noise out of the box, and that will help to keep external noise away from the timing outputs.
Signals
In reality, we’re not really dealing with square wave signals at all in this application. I merely took a diversion to talk about this in a simplistic way. With frequency standards, we are dealing with a more pure signal – a sine wave.
The square wave would have odd-harmonic frequency components, which all add to the noise we don’t want. A sine wave has a single tone, the fundamental, and it’s only the phase and frequency of this we care about.
In fact, we don’t even care about the phase so much. Correctly speaking, we’re not dealing with a problem of synchronisation (bringing systems into phase alignment) but of syntonisation, which means bringing systems into frequency alignment.
I shall write about phase another day.
The distribution amplifier we desire is just that, an amplifier not a digital buffer. We need to take in a signal and drive it out with isolated outputs with a decent signal level. If this level is too much for a specific device, then it can be attenuated appropriately. We still need proper termination, of course.
For instruments that use this reference signal in a digital manner, they would include input signal conditioning, typically a Schmitt trigger,
Isolation
The next matter is one of isolating the DC paths. A naïve approach would be to simply connect the shield/ground of the output BNC jack straight to the analog ground, for all outputs, and hook the signals straight to the amplifier outputs with a terminating resistor.
This is not necessarily bad and will work. But, it is not a great thing to do.
There are many reasons for this including removing ground loops, isolating power supply noise paths as well as issues of DC level.
Basic electrical safety is also a big consideration. If a power supply fails in a weird way, and shorts line AC to ground in a box somewhere, you don’t want it propagated to every instrument via that common ground connection!
To achieve this, we will specify a requirement for an isolating transformer for each output. This way there is no galvanic connection from the outside world into the box.
Requirements
Simply put, I want:
- 10 off outputs 10 MHz sine wave at minimum 1Vrms into 50 Ω
- 5 off outputs of 1PPS as above (TBD)
- Isolation for all outputs via transformers
- Linear power supply for low-phase noise
- Shielded metal case
Next Steps
So, the next step is to measure up the PCB mounting locations, the connector positions and draw up a board outline, and get cracking.
LTspice Source Code: TransmissionLine multipleTransmissionLines
