There are vector phase shifter/modulator devices on the market. Basically, you generate a fixed shift (ideally around 90 degrees, but not critical) and then proportionally add/substract the original and fixed-shift phases to produce any desired output phase.
Getting best performance (especially if the fixed shift is not 90 degrees) would require using DACs to generate the control voltages for each phase from a lookup table based on measurements previously taken at a particular frequency (or automatically over a range of frequencies, perhaps using a GPIB-connected vector network analyzer)
RE-EDIT: Since you have a solution at low frequencies, one option is to just mix each of these up to 500 MHz using the same local oscillator and pass the outputs through matched band filters. This is slightly more practical if you use arbitrary generators capable of outputting a higher frequency, say 100 MHz, as then the filter requirements are looser. Ultimately this is sort of reshuffling of the same idea - it's still multiplication, but your control inputs are moderate frequencies instead of DC voltages and it moves the shifting requirement through the multiplier to where it is easy, at the cost of requiring some filters on the other side. And there's even a form where you replace the filters with a quadrature (2-phase) local oscillator and image reject mixers.
ADDITIONAL IDEA:
A pair of lower frequency linked generators could be used as references for PLL synthesizers multiplying to the desired frequency range. Changing the phase of the low frequency signal will result in a phase change of the high frequency one, of magnitude multiplied by the multiplication factor (think of the phase change as a time delay, with which the higher frequency signal must also align). The catch is that extremely fine control of phase would be necessary at the low frequency to get moderate resolution control at the higher one. For example, if you have a 20 MHz signal synthesized at 100 MSPS, a delay of one sample is 5 entire periods of a 500 MHz product! As a result, this would require a DDS with many bits of residual phase - that is to say, less significant bits of the phase accumulator that accumulate internally, and only eventually roll over into the bits that are of a high enough order to feed into the lookup table that generates sine samples. Any decent DDS has some of these; in this case you'd need an extreme. The idea probably works best when the DDS frequency is as high as practical - ie a few hundred MHz clocked at a gigasample (which is something you've been able to buy as an IC from Analog Devices etc for a few years now) and the PLL multiplication ratio is fairly low.
Most of these ideas seek to use a greater quantity of relatively inexpensive (per unit) active circuitry and even software to limit the requirement of expensive per-unit-adjusted precision passive elements. Unfortunately, most require pairs of filters or a shift network with performance that is either similar, or been per-unit characterized so that its imperfections can pre-compensated in the control settings used. The method using two low frequency DDS's and image reject mixers goes closest to avoiding this, but it needs near perfectly orthogonal quadrature LOs at fixed frequencies for each band - for example 400 MHz to mix to the 500 MHz band. It may be possible to create two phases by digitally dividing from a higher frequency, otherwise there would be a shift element that would need to be aligned for each band of interest. The soundcard-as-HF-exciter ham software-defined-radio people have done some looking at precompensating the synthesized signals to compensate for imperfect IQ LOs and mixers which could be looked into, but since the idea is to have perfect cancellation of the image frequency (vs filter it out) this is pretty critical.
Simple answer if $$ available
If budget permits, Agilent and presumably others offer dual channel and synchronizable arbitrary waveform generators that will do 500 MHz either directly or via an IQ mixer. As an off the shelf solution, this would be closest to extending what you are able to do at 20 MHz to the 550 MHz need. Such equipment is rented as if not more often than purchased outright.
If you were expecting a 1-5V output and you accidentally purchased a current output model isolator, simply connect a high-precision 250 ohm resistor across the output terminals (or even better, across the input terminals of whatever is connected to your signal isolator). Resistors of this type are commonly available for exactly this purpose.
Ohm's law:
V = I * R
V = [4mA - 20mA] * (250 ohms)
V = 1V - 5V
While it may seem like a pain, a 4-20mA signal is preferable to a voltage output in many applications:
- It allows the detection of a broken sensor wire (it goes to 0mA) in addition to sensing the process variable (sensor output).
- Some sensors can be 'loop powered' - they require less than 4mA to operate, and can therefore be used as a two-wire device without need for a separate power supply wire. The savings in wiring cost alone can be significant.
- A current signal eliminates error due to resistive losses in the conductors between sensor and readout/ADC. (Though it trades this for errors related to the 250 ohm resistor)
- Noise immunity is substantially better due to the inherent common mode rejection. This is a huge win in industrial settings when tend to be very electrically noisy.
Best Answer
You could use RF transformers - each primary connects to each input and the secondaries are all wired in series to produce A+B+C+D as an overall output. What do you next depends on you - you could terminate the combined secondary with 200 ohms and use this into whatever it needs to feed or, you could use a fifth transformer to convert the 4 series secondaries down to 50 ohm impedance.
Other problems (and only you know whether this is a real problem you might face) is that terminating the 4 series secondaries will produce cross-talk from A to B, C and D and any combination thereof so you might want to terminate each secondary in a 50 ohm but, if you draw any current from the "combined secondary transformer" this will also produce crosstalk.
This, of course may not be an issue to you. If it is then a small "high-impedance" buffer amp would "isolate" the secondaries reasonably well.