In long-haul unrepeated optical transmission systems, the different input distributions employing constellation shaping, have led to increased achievable information rate (AIR) at the expense of reduced signal-to-noise ratio (SNR). In this paper, at optimal launch power, enhanced Gaussian noise (EGN) model based probabilistic shaping (PS) is implemented to inhibit the fiber non-linearities. The method exploits the inherent system and input conditions to result into maximum mutual information (MI) without using digital signal processors. Using sequential quadratic programming (SQP), a robust 16-QAM coherent optical system is implemented in back-to-back configuration employing probabilistic shaping. The shaping is optimized for minimum values of the objective function, second and fourth order moments of input. The higher order noise coefficients are thereby reduced, resulting in minimal non-linear interference noise (NLIN) at increased values of SNR. The optimized EGN model returns values of 1.41 and 2.24 for μ^4 and μ^6 respectively, having highest SNR value of 14.52 dBm and the least SNR penalty of 0.15 dBm towards the uniform probability distribution.