To better understand the origin, we demodulated the signal of the power-line sniffer (Em_IPSCB_50Hz) with a pure 50 Hz sine using matlab. Fig 1 shows the 'phase noise' of the 50 Hz line, plotted for a time that the sideband frequency was either low or high. The result looks like a square wave with varying period and an amplitude of one millirad. One explanation might be a heavy electrical load that is switched on and off with a frequency of a few Hz.
We were first suspecting the charging of the battery of the UPS, but no correlation was found with the load level. Then, with some luck, we found that the frequency of the sidebands shows perfect correlation with one signal of the air-conditioning of the MC building (IMAC_MC2_Loop_1_ActiveOut), see fig 2. According to Pasquale, this signal monitors the control signal sent to one of the heaters. This building is heated by an electrical resistance, which is a load of some 10 kW that is pulse-width modulated using a triac. Note that all the other buildings are heated with hot water.
As a deditated test, we changed the temperature set-point of the MC building for a few minutes, which caused the heater to switch on. There is a clear effect on the sidebands, see fig 3.
