Efficient Deep Learning-Based Detection Scheme for MIMO Communication Systems.

Area of Science:

• Wireless Communication
• Signal Processing
• Machine Learning

Background:

• Multiple-Input Multiple-Output (MIMO) is crucial for next-gen wireless systems.
• Detector complexity in MIMO systems increases with antenna count, posing challenges for massive MIMO.
• Balancing detection complexity and bit error rate (BER) is essential.

Purpose of the Study:

• To develop an efficient deep learning (DL)-based signal detection strategy for MIMO systems.
• To introduce novel DL schemes that improve upon conventional methods in terms of complexity and performance.
• To evaluate the trade-off between detection complexity and BER performance.

Main Methods:

• A DL-based signal detection strategy incorporating a novel input signal labeling preprocessing stage.
• Proposal and evaluation of two new schemes: OH per antenna (OHA) and direct symbol encoding (DSE).
• Comparison of proposed schemes against conventional one-hot (OH) and maximum likelihood (ML) methods.

Main Results:

• The OHA and DSE schemes achieved an F1-score of 0.97 in classification performance.
• Both proposed schemes exhibit lower computational complexity than conventional OH and ML schemes.
• BER performance losses for OHA and DSE are less than 1 dB and 2 dB, respectively, compared to the OH scheme.

Conclusions:

• The proposed DL-based strategy effectively reduces detection complexity in MIMO systems.
• The OHA and DSE schemes offer a favorable trade-off between BER performance and computational cost.
• This approach is suitable for adaptive wireless systems with limited computational resources.