Blind Equalization Based on Data Reuse and Reliability Labeling

In the absence of a training sequence, blind equalization algorithm usually takes thousands or even tens of thousands symbols to converge, which is especially evident when the channel is severely distorted. When the short data is required to achieve equalization, the current blind equalization algorithms will be difficult to converge. This letter presents a short data reuse blind equalization algorithm based on reliability labeling. If the output symbol corresponding to the same input is labeled twice with reliability during the process of the current and last data reuse, the algorithm will switches to the decision-directed model automatically. At the same time the letter gives a simple way for model switching by a look-up table. Simulations show that the proposed algorithm performs better compared with Modified Constant Modulus Algorithm with long data.