Using again an alkaline electrolyte cell, I tried retrieving the average strength of the signal over time at a frequency where the difference between the on/off state was high. I used a standard RTL-SDR USB adapter (radio receiver), a ready-made utility (rtl_tcp from the 'osmocom' package) to retrieve the samples and custom-made code in Python to process them.
Some related links:
The average strength is calculated roughly every second with samples retrieved at a rate of 256 kHz. The peak value over those samples is also retrieved. Below is the graph from a brief period of testing under various conditions. The low flat areas are during off conditions.
It appears that many dynamically changing variables affect it, and the EMI might not necessarily always increase when increasing applied voltage. The onset of plasma appearance can be stronger in magnitude, to only drop substantially when a visible plasma appears (left portion of the graph).
After a plasma is visible and voltage high, The signal at times slowly increases under stationary input conditions, presumably from the electrolyte heating up and its concentration increasing near the cathode. Sometimes instead it suddenly drops in response to spontaneous changes in the reaction (middle portion of the graph).
The signal can also be affected by walking in the same room, which presumably causes the generated emissions to bounce differently to the antenna, which seemed unusual for a ~82 MHz frequency. The line-of-sight path between the antenna and the cell is unimpeded (not shown here).
Both with an acidic and alkaline electrolyte it does not seem that a large cathode helps, even if the reaction looks stronger and power put into the cell substantially increases (from prior testing—not shown here).
I also tried retrieving data at a high rate (about 2MHz) and as I pointed out in a different thread, the noise appears to have a 1/f character. With a logarithmic scale for both axes in the frequency spectrum, a more or less straight line is obtained.
What this implies exactly, I'm not sure but it's probably not too unexpected. More information here: 1/f noise - Scholarpedia.