In order achieve beyond 400 Gb/s serial optical transmission using commercially available equipment, in this work, we propose the use of channel capacity achieving modulation formats. We present a method to determine the optimum signal constellation for an arbitrary dispersion map. This method can be described by the following steps. (i) Determine the probability density functions (pdfs) of the fiber-optics channel, which can be approximated by complex Gaussian function or estimated by evaluating histograms. (ii) Use the resulting pdfs to determine the optimum input distribution to achieve the channel capacity. A split-step method is presented to perform the optimization over the probabilities and mass points of input distributions. (iii) After quantizing the optimum input distribution the optimum signal constellation can be determined. We study two types of dispersion maps currently in use, which are suitable for metro and mediumhaul applications. For each of them we determine the optimum signal constellation. The numerical results demonstrate the efficiency of this method and show that the optimized signal constellation can increase the transmission distance at most by three spans against its counterpart.