A method of recovering a most likely value of each symbol transmitted through an optical communications network using a high speed optical signal. A stream of multi-bit digital samples of the optical signal is processed to generate a respective multi-bit estimate X′(n) of each transmitted symbol. A first function is applied to each symbol estimate X′(n) to generate a corresponding soft decision value {tilde over (X)}(n). Each soft decision value {tilde over (X)}(n) is processed to generate a corresponding hard decision value. {circumflex over (X)}(n) having an ideal amplitude and phase. A plurality of successive soft decision values and hard decision values are processed to determine a second function, which is applied to each soft decision value {tilde over (X)}(n) to generate a most likely symbol value {circumflex over ({circumflex over (X)}(n).