This paper proposes a method of increasing the efficiency of the CATV plant in the upstream direction. The general approach is to reduce the convergence time of the equalizer to allow a shorter training sequence. Specifically, the upstream channel is modeled as a postdetection filter. A short unique word is inserted near the beginning of the preamble to estimate the coefficients of the postdetection filter. The postdetection filter is then crudely inverted to estimate a few key tap weights of the equalizer. The equalizer is initialized with these estimated tap weights before adaptive updating is turned on. The method is evaluated for the echo-laden DOCSIS CATV upstream channel with a signal-to-noise ratio (SNR) of 25 dB. Plots of the probability of the modulation error ratio exceeding 19- and 22-dB values versus the length of the training sequence show that the performance is comparable to an unseeded equalizer with a training sequence that is about 50 symbols longer.