Blind equalization of multilevel signals via support vector regression

Support Vector Machine-Based Blind Equalization for High-Order QAM With Short Data Length

“…Two types of channels are used in the simulation as shown in Table II. The first one is a fixed channel commonly used for testing [17] [18], and the second one is a random two-path fading channel with complex channel gains parameters h 0 and h 1 . The magnitudes of h 0 and h 1 are Rayleigh distributed with power difference of 3dB and their phases are uniformly distributed.…”

“…The number of equalizer taps is L w = 21. The central tap of equalizer is initialized to 1+j √ 2 [18] and 1, for CH1 and CH2 respectively, and the remaining equalizer taps are initialized to 0.…”

“…The ISI threshold is set to -25dB (the BER is about 0.01 at such ISI). If convergence is achieved, the ISI will be counted for residual ISI, otherwise, it will be ignored [18]. The results are obtained by averaging 500 independent experiments.…”

“…The results are obtained by averaging 500 independent experiments. The algorithms to benchmark with are the SVR-MMA(4,2), which is the most recent and promising blind batch equalizer [18], and a "naive SVR+NA-SBD", which straightforwardly combines SVR with NA-SBD [11]. From Figs.…”

“…The works in [15]- [17] are based on CMA, which suffers phase ambiguity. Recently, in [18], [19], the error function of MMA is formulated within the SVR framework to resolve the phase ambiguity issue.…”