The spectrum above was taken right at the 10 volt secondary of the filament transformer, ground tied to centre tap. A measurement done afterwards from leg to leg was of course 6dB higher overall but otherwise the same. How much of it appears on the plate is naturally dependent on tube and circuit but it's unquestionably on the cathode. As to the cause, I suspect it's the transformer!
Here's why. I benched a very low distortion (0.02%) parafeed headphone amp using a toroidal 240 volt to 20 volt power transformer. The max plate swing into the primary is about 100 volts p-p. Curious how an EI transfomer would perform as an OPT I substituted the same Hammond filament transformer seen here, which is rated 120 volt to 10 volt. Since the voltage ratios are the same the headphone amp plate load should have been roughly equivalent. The distortion shot through the roof and displayed a very complex set of harmonics. The result got me curious about the Hammond in a filament application where it's getting near double the primary voltage and delivering real current. I'll leave it to the black art 'magnetmaticians' to comment on cause but I suspect the core material, i.e. Hammond strikes again.
The standard logic dictates that EI power transformers are preferable in audio partially due to higher internal losses acting to filter power line junk. Toroids are so effcieint and wide band everything passes straight through. As I so often find, the picture is much more complex when a meter enters it. This was intended to be a 'budget' 813SE but now I'm torn between Tentalab CCS filament supplies using these transformers or finding toroid 10 VAC trannies.
Majestic: The software is WaveSpectra 1.31 -
http://www.ne.jp/asahi/fa/efu/soft/ws/ws.html - which was just recently updated to 1.40. In combination with Sinegen -
http://e-cat.nm.ru/sinegen/ - it makes a great freeware PC suite.