Channel measurements for a 50G passive optical network (PON) show a significant amount of intersymbol interference (ISI) due to chromatic dispersion (CD) and the use of low-cost, bandwidth-limited 25 Gbaud class components. The channel characteristics at the input of the forward error correction (FEC) decoder are measured for several receiver equalization schemes, and the error correlation at the output of each equalizer is characterized using a sub-class of a Fritchman’s Markov model. This model is used to generate error sequences that match the observed correlation statistics, and these sequences are then used to evaluate the performance of a candidate low-density parity-check (LDPC) code with lifting factor 256 and hard-input decoding. It is shown that for a bandwidth-limited 50G PON system with 83 ps/nm dispersion, the optical power sensitivity penalty (OPSP) due to a reduction in error correction performance caused by correlated errors is 0.3–0.6 dB. Precoding, i.e., differential encoding of the input data and differential decoding, helps to reduce the OPSP for some equalizers, but the penalty increases for other equalizers. The use of a 256-bit segment-interleaver only marginally improves the decoding performance, whereas bit-interleaving across four codewords reduces the penalty to within 0.05 dB relative to the binary symmetric channel.