Signal Constellation is just a plot of complex numbers. Now when a phase noise is introduced, it means you are multiplying each of the complex numbers (constellation) by exp(i*phi), (phi is the phase noise). This results in rotation of the constellation space. This implies your decision boundaries also rotate...
--
Now phase noise, to be simply stated is, noise in the phase term. The output of an oscillator if usually Ac*cos(2*pi*fc*t + phi), phi is an arbitrary constant initial phase. The spectrum of this signal is a delta function at ±fc.
Now a phase noise is included in this expression as
Ac*cos(2*pi*fc*t+phi+Nph(t)), where Nph(t) is the phase noise term. The spectrum of this signal is shown to have skirts around a delta function at fc and -fc. ie. you have frequency components at fc ± δf, and -fc ± δf.
The broader these 'skirts', the poorer your oscillator is. The jitter is also expressed to be the manifestation of phase noise.