hybrid cnn-lstm channel estimation for 5g massive mimo systems using sparse pilot reconstruction and time-varying 3gpp channels
Journal of Intelligent Decision Making and Information Science · published · 1 min
a hybrid cnn-lstm estimator reconstructs 5g massive mimo channels from sparse pilots with lower error, lower pilot overhead, and real-time inference latency.
This paper proposes a two-stage hybrid CNN-LSTM estimator for 5G massive MIMO-OFDM channel estimation from sparse pilot grids.
Least-squares pilot estimates are expanded with two-dimensional bilinear interpolation, then refined by a time-distributed CNN and LSTM sequence model trained on time-varying 3GPP TR 38.901 channels with Doppler effects up to 120 km/h.
On a 4x4 MIMO-OFDM setup with 624 subcarriers, the method reduces normalized mean-squared error by up to 88%, roughly halves bit-error rate at mid-range SNRs, and improves spectral efficiency versus conventional interpolation while using half the dense-pilot overhead.
The results point to CNN-LSTM processing as a practical path for accurate, low-overhead, sub-millisecond channel estimation in latency-sensitive 5G deployments.