This paper presents a neural approach for the parameter estimation of adaptive sparse system identification based on a MCA EXIN (minor component analysis by excitatory and inhibitory learning) linear neuron for TLS (total least squares) problem. We incorporate a log-sum (or Reweighted l1) penalty into the cost function of TLS EXIN (total least squares by excitatory and inhibitory learning) in order to identify the sparse system. The given computer simulations illustrate that the neural approach considerably outperforms the existing TLS EXIN method as well as the LMS-type adaptive methods in the sparse system.