Research on M-sequence Estimation Based on CNN

Research on M-sequence Estimation Based on CNN

A method based on convolutional neural networks (CNN) for estimation of sequences is proposed. First of all, a sequence estimation model based on CNN is established, the sequences to be estimated are directly used as the inputs of the model. The method effectiveness is verified by estimating m-sequences. Second, by combining FastICA algorithm and the proposed CNN estimation method, a new spreading sequences estimation method named FastICA-CNN for long code direct sequence code division multiple access (LC-DS-CDMA) signals is studied. The FastICA algorithm is used for blind separation of each user’s spreading sequences, the amplitude fuzziness is eliminated by delay multiplication, and the CNN estimation method is used for spreading sequences estimation. The results show that, when the number of users is 2 and signal-to-noise ratio (SNR) is higher than −16 dB, the estimation accuracy of m-sequences in LC-DS-CDMA by FastICA-CNN method tends to be 1; the time required for model training and sequences estimation is 84.71s and 375.4s, respectively. Compared to the methods in references, FastICA-CNN method requires shorter simulation time and has higher estimation accuracy.