At first glance it seems to be an ordinary inverter. Upon closer inspection, the primary wires to the battery and the battery itself would need to supply a rather high amperage at 12 volts to carry that secondary load. Then we see the battery disconnected. This could lead to at least 3 conclusions we can discuss later, but first, the interesting circuit.
One of the conclusions is that a power gain happens somewhere along the way.
In the first schematic we see a ramp generator producing a linearly varying magnetic field in Tr2. The ramp amplitude is sensed (at the secondary) to reset the ramp generator. This varying magnetic bias is applied to one of the the tower coil windings. This ensures that the correct NMR diameter for the gain medium will be hit.
As the LF ramp increases linearly from zero to max, the Larmor frequency (HF) necessary to precess the nuclei, also increases linearly. If the HF is held constant, then the LF ramp will eventually create a correct magnetic field for the HF to become the Larmor frequency of the gain medium.
At the same time, the cyclotron diameter decreases as the LF ramp increases from zero to max.
Notice the ramp generator is synced to the HF chip driver. I'm guessing for each HF pulse, there is a corresponding ramp as there are no dividers shown.
This will establish a phase lock between the LF ramp and HF oscillator. The HF burst becomes the Larmor frequency for the gain medium, only at a certain phase shift with the LF ramp waveform. This shift should be adjustable.
Resonance of C3 provides the 90 degree phase shift required* between the two signals.
Not necessarily 90º phase shift.
90º shift is only required for creating rotating magnetic fields.
Notice (hand drawn schematic) that he has some pulsating DC fed into the coil from the bridge instead of a separate generator for varying the magnetic bias.
I like to keep the magnetic bias constant and sweep the HF according to a ramp FM envelope.
However, the magnetic bias does not have to be constant. It can be variable and the HF can be fixed. This is an inferior method because it leads to the undesirable and usual Lenz/Faraday transformer action, between the LF primary and other windings.
In the latter case, the bias winding should be supplied by some form of DC+AC (e.g. trapezoidal ramp) or only by AC combined with permanent magnet "DC" bias.
It is wasteful to vary the magnetic bias of the gain medium from 0 to max, if it is known that the correct bias for the HF to become the Larmor frequency is much above zero.
To my eye, the metal tube has the dull appearance of a zinc coating rather than a shiny aluminum appearance.
Many materials with non-zero nuclear spins are suitable. Copper, Iron and Zinc are one of them. Brass is an alloy of Copper and Zinc.
Iron and Zinc is require 20x more precise Larmor frequency match than Copper, thus Fe and Zn are more finicky as gain media.