An optimum multilayer perceptron neural receiver for signal detection

Area of Science:

• Signal processing
• Machine learning
• Information theory

Background:

• The optimum likelihood-ratio receiver is a standard for signal detection.
• Previous models assumed uniform signal amplitudes across receiver inputs.
• Generalizing this receiver is crucial for diverse signal environments.

Purpose of the Study:

• To generalize the M-input optimum likelihood-ratio receiver for unequal signal amplitudes.
• To identify an equivalent optimal neural network (neural receiver) for signal detection.
• To evaluate the performance and robustness of the proposed neural receiver.

Main Methods:

• Generalization of the likelihood-ratio receiver framework.
• Development of an equivalent multilayer perceptron neural network architecture.
• Monte Carlo simulations to assess detection performance.

Main Results:

• An equivalent optimal neural receiver was identified for M-dimensional signals corrupted by Gaussian noise.
• The neural receiver's detection capability proved insensitive to training set size or level.
• Performance was validated against analytical results and simulations.

Conclusions:

• The proposed neural receiver offers a robust solution for signal detection with varying amplitude inputs.
• This approach extends the applicability of neural networks in complex signal processing scenarios.
• The findings highlight the potential of neural networks in overcoming limitations of traditional receivers.