Sparse I/Q-joint DNN nonlinear equalization based on progressive pruning for a photonics-aided 256-QAM MMW...

Area of Science:

• Photonics
• Telecommunications Engineering
• Artificial Intelligence

Background:

• Nonlinearity in photonics-assisted millimeter-wave (MMW) systems poses challenges for high-order modulation formats like 256-quadrature-amplitude-modulation (QAM).
• Traditional equalizers, such as Volterra nonlinear equalizers, can be computationally intensive and less effective with complex modulation schemes.

Purpose of the Study:

• To propose and experimentally demonstrate an efficient nonlinear equalizer using a pruning I/Q-joint deep neural network (DNN).
• To mitigate nonlinearity in MMW systems, enabling higher data throughput and reduced complexity.

Main Methods:

• Development of a pruning I/Q-joint deep neural network (DNN) architecture.
• Experimental demonstration in a photonics-assisted MMW system transmitting a 256-QAM signal.
• Comparison of the proposed DNN equalizer with traditional Volterra nonlinear equalizers and I/Q dual DNNs.

Main Results:

• The pruning I/Q-joint DNN achieved a net throughput of 66.67 Gbps for 256-QAM E-band MMW transmission.
• Computational overhead was compressed by 32% compared to a non-pruned DNN.
• The proposed equalizer exhibited 20.21% less complexity than traditional Volterra nonlinear equalizers.
• A 16% pruning ratio improvement was observed compared to I/Q dual DNNs, highlighting the effectiveness of deciphering the I/Q relationship.

Conclusions:

• The pruning I/Q-joint DNN is an efficient and effective solution for nonlinearity mitigation in high-order QAM MMW systems.
• This approach offers significant advantages in terms of computational complexity reduction and improved performance.
• The method holds promise for future advancements in high-speed wireless communication systems.