[Nota: the bandwidth makes me think about field seismic signal, which has literature on 50 Hz and 60 Hz power line removal]
The question revolves on how much your signal differs from a sine at 50 Hz.
On the one hand, imagine that your signal is a pure 50 Hz frequency. It would be in indistinguishable from the power line disturbance. On the other hand, if it is very random (and quite independen), you could try to identifies the power line, and subtract it.
Between the two extremes, there are many techniques from non-parametric methods to parametric modeling. If you care for spectral methods, spectrum interpolation is described in Section 2.3 "Signal Processing: Power Line Noise Removal" from
Reducing power line noise in EEG and MEG data via spectrum with spectrum interpolation code:
If power line noise interference is considered as an additional
component – with a peak at 50 Hz – superimposed on the continuous
power spectrum curve of MEG/EEG data, it can be removed by
interpolating the curve of the power spectrum at the respective
Another option is to exploit the potential of signal separation of transformation, like short-term Fourier transforms, and exploit similar masking or interpolation in the time-frequency or time-scale domain.