Table 1 Features and challenges of existing deep learning-based massive MIMO detection.

Yu et al.²⁰

AMIC-Net

It uses backwards propagation to learn input sequences more effectively

It highly reduces the gradient vanishing

It struggles to manage the complex situation of massive MIMO

Mahmoud et al.²¹

MSDS

It highly minimizes the high computational complexity needed for the detection

It reduces the energy required for the channel inference process

It does not detect non-coherent signals effectively

It does not provide better accuracy because of the absence of encoding techniques

Wei et al.²²

LcgNet

It highly reduces the cost required for the memory

It eliminates the complexity of detecting signals

It requires more training data

Operational costs are high

Nigatu and Zhang²³

DNN

It effectively handles the penalty parameters to enhance the performance of the network

It does not need technical computation to update the variables

Computational complexity is high

Xu and Du²⁴

MLNet

It greatly reduces the computations required for the detection

It has less complexity

It lacks in latency

It suffers from packet loss

Tan et al.²⁵

CHEMP

It provides high robustness

It requires less training time

It is highly expensive

It is not suitable for software implementation

Liao et al.²⁶

DNN

It effectively manages the connection between each layer

It is difficult to process the data with a complex size

Nguyen et al.²⁷

DNN and OAMP-Net

It provides high efficiency

It could be reused for multiple detections

It needs more amount of data

It does not consider the necessary information during training