A New K Best Sphere Decoder in 16 × 16 MIMO System using Deep Learning Algorithm

In this paper, a new
$\boldsymbol{K}$
best sphere decoder is proposed for a
$16\times 16$
MIMO receiver, in which the radius is determined by a deep learning mechanism using neural networks. The modified
$\boldsymbol{K}$
best sphere decoders are having performances very close to maximum likelihood, with significantly reduced computational complexities as compared to that of the conventional detection techniques. The significant improvement in the achieved results is because of the deep learning algorithm's ability to limit the number of codewords at the receiver. The efficiencies of the resultant deep learning-based sphere decoder are evaluated in terms of spectral efficiency and symbol error rate. The effectiveness of the proposed algorithm is supported through the simulations carried on MATLAB software for
$16\times 16$
MIMO systems, using 16 QAM modulation.