Adaptive channel equalization using classification trees

This paper focuses on adaptive equalization of binary signals in a baseband digital telecommunication system. Equalization and detection are considered as a classification problem. Fixed-length sequences of received observations form a multidimensional signal space, which can be partitioned using the proposed classification tree algorithm. Top-down approach is used in tree induction, and splitting is done based on information gain criterion. Overfitting is avoided by utilizing a pruning algorithm. The advantages of this method are its simplicity and straightforwardness and thereby the reduction of computational complexity compared to other well performing equalizers. Experimental results and comparison with a cascade-correlation trained multilayer perceptron neural network equalizer are given.