An optimized deep learning model for a highly accurate DOA and channel estimation for massive MIMO systems

Massive multiple-input multiple-output (MA-MIMO) has been hailed as an auspicious technology for the future generation of wireless communications because it can considerably increase the capacity of the communication network. However, using the maximum likelihood (ML) direction-of-arrival (DOA) estimate method is severely constrained in actual systems because of the computationally expensive multi-dimensional searching procedure. This paper proposes a novel approach to estimate DOA and channels by incorporating deep learning into the MA-MIMO system. Here, a deep belief network (DBN) is used to learn both the spatial structures in the angle domain and the statistics of the wireless channel through both online and offline learning procedures. Also, a bald eagle search (BES) Optimization is used along with DBN to attain high precision through optimal training. The proposed model can estimate the channel based on the predicted DOA and the complex gain. According to numerical results, the suggested method performs significantly better than state-of-the-art methods, particularly in tough conditions like low signal-to-noise ratio (SNR) and a finite number of snapshots. The proposed DBN-BES technique accomplishes less root mean square error (RMSE) as 0.01 for SNR of 5 dB in elevation calculation and 0.02 for SNR of 5 dB in azimuth calculation. Also, the proposed algorithm greatly reduces computational complexity.