There is no escaping from hardware, if fact without hardware there is nothing for the software that is so much in fashion. What good is the most sophisticated of AI without somewhere to house it? Yeh I know some advocate storing AI software in another dimension where 'solids' (Deep Space Nine) do not exist. Back in reality let's cover some hardware for transmitters.
In the distant origens of broadcasting, transmitters of all kinds used a valve be it simple triode, klystron, magnetron, TWT (users were called twitters, remember). A 50KW transmitter was a single valve in a massive cabinet, massive PSU, massive fan, highly inefficient and short life time.
In these times it is all solid state GAsFets, SiFets, HEMTFets, SiCFets, GaNFets with a sprinkling of Bipolar, what ever. In all cases small, efficient with a big problem of high power dissapation at a point. This means the power generation has to be distributed into small units (eg: a 20KW FM I designed used 24 SiFets). This means combiners and by analogue splitters have to be used to distibute the modulation between an array of amplifiers and re-combine the output powers afterwards.
So in this section we look at some Combiners.
In the original sense microwaves refered to frequencies above 500MHz. Here I use it in the sense or distributed elements. In times when MICS operate at 35GHz, these are lumped elements used in a distributed environment. So here is the crux of the matter. Is the component in question larger that 0.1 wavelength, if so use 'microwave' techniques, if not used 'normal' lumped techniques.
An example is a combiner using microstrip, is rather large but very cheap to make, manufacture consistent, does not require tuning versa a lumped element version requiring specially wound inductors, production formers, production tuning ie: costly. but sometimes space constraints take preference, so we look at both choices.