A fast comparator could look at a sine wave and a DC level to make PWM at its output. Even a triangle if linearity is important, which is isn't to transmit a 1-bit digital state.
A differential-input gate, like a 10EP part, is a decent comparator, cheaper and simpler than an ADCMP thing.
How about a 4 - 20 ma current loop? The transmission line is shielded if you are going to carry 2 GHz signals. You can terminate in 50 ohms to match the cable impedance. Then all you need is a wide bandwidth current source which should be easy. The output voltage is simply E = I * R.
You can beat the bandwidth of PWM due to the filtering required at the receiver. Even extensive filtering leaves some ripple, which reduces the accuracy of data transmission.
With controlled current, you don't have to worry about voltage drops along the transmission line. You can set a current value that lasts for years, even with multiple powerups. Unlike PWM, the overall accuracy is determined by relatively few components.
The signal on an AC-coupled channel always averages zero. If you poke a PWM square wave through it, you get an asymmetric amplitude square wave with zero DC average.
Telecom-type optical signals are usually transmitted by switching a laser on/off, following the digital input. The transmitters are usually AC coupled, so need continuous, roughly 50% digital activity to work right.
That is typically received by a photodiode, an AC-coupled AGC amplifier, and a zero crossing detector, to typically make an AC-coupled differential CML output.
Ethernet type signals are transformer coupled on both ends.
That gets interesting to push a DC coupled signal through, theoretically and experimentally.
I'm interested in inherently AC-coupled channels, namely telecom type fiber links or ethernet-style transformer-couples twisted pairs.
We already sell fast DC coupled fiber links, laser on/off things. But the receive threshold has to be manually set and the effective threshold varies with path attenuation. Somebody tie-wraps a fiber down, or adds a bulkhead feedthrough, and the threshold is trashed.
If you want a continuous signal but only have AC channels, consider a sine/cosine oscillator, or multiphase oscillator, where the sum of the squares of the signals is constant (good for gain control) but the frequency can mismatch against a local oscillator (giving you an FM decoded value, or phase-shift value if that's preferable).
Thermistor sensing of a sine and cosine pair of heaters is a fairly easy way to do gain control.
I can speak to how optical Ethernet ensures that their signals can be transformer-coupled, regardless of what data is sent.
The basic trick is that the code space of the line symbols is large enough that there are at minimum two line symbols that can be used to represent each possible data symbol (or control symbol). The line symbols are sent by OOK of the laser, it taking multiple on-off flashes to encode each line symbol. The transmitter keeps a running sum of the ones and zeros sent, and chooses which of the pair of line symbols to use depending on the running sum - if the sum is growing in one direction, the chosen symbols will drive the sum back towards zero average.
There is a founding article that I read twenty years ago on the design of such codes. If I recall, this was invented at IBM. While I may have a copy of that article, I'll probably never find it now.
Optical ethernet is almost always SFP modules. Their i/o is capacitive coupled differential CML, no transformers.
That's 8b10b or longer versions. The idea is to keep the 1/0 balance exact long-term.
I want to send PWM, which doesn't, which is why it's interesting.
A typical FPGA serdes engine natively does 8b10b and longer balanced codes. We have used the serdes engines to make PWM and programmable pulses too. You have to get inside the serdes blocks and work near the end.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.