We know that power spectral density (PSD) can be defined for any wide-sense stationary (WSS) signal as the Fourier Transform of its autocorrelation function, i.e. if for a signal $x(t)$: $$E\{x(t)\}=\eta\quad ,\quad\text{constant}\\E\{x(t+\tau)x^*(t)\}=R(\tau)$$then the PSD can be defined as the FT of $R(\tau)$. So here is my question:
How can one define a similar PSD for non-WSS signal?
Thanks in advance!