Abstract: An 83.2- km, 20-Gb/s QPSK DFT-S-OFDM PON based on superimposed training (ST), super-Nyquist image induced aliasing, and direct detection (DD) is first experimentally demonstrated using superimposed training-aided channel estimation (ST-CE) and equalization. Instead of training-aided channel estimation (TA-CE), the linear impairments of the channel using superimposed training without any bandwidth loss is equalized. The idea of ST-CE is that the training sequences (TS) are arithmetically added to the information symbols instead of being placed in separate exclusive time slots. Accurate CE is obtained by using the first-order statistics of the received signals. To further increase the reachable distance, the receiver bandwidth is slightly increased and super-Nyquist image induced aliasing to compensate the dispersion induced power fading is used. Results show that the superimposed training channel estimation (ST-CE) achieve similar estimation performance as that of the TA-CE method in a long reach IM/DD DFT-S-OFDM passive optical network (PON) system, without compromising its spectral efficiency. This method can save bandwidth, and improve the utilization of the spectrum.