A Look-Up Table Assisted BiLSTM Neural Network Based Digital Predistorter for Wireless Communication Infrastructure.

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• Neural networks offer superior performance in digital predistortion due to their ability to model nonlinear systems.
• Existing predistortion techniques struggle to efficiently compensate for both static and dynamic nonlinearities in power amplifiers (PAs).

Purpose of the Study:

• To propose and validate a novel hybrid predistorter, the look-up table assisted bidirectional long-short term memory (LUT-A-BiNN) neural network.
• To leverage a cascaded architecture where a look-up table handles static distortions and a BiLSTM network addresses dynamic distortions.

Main Methods:

• A hybrid predistorter architecture was designed, combining a look-up table (LUT) with a bidirectional long-short term memory (BiLSTM) neural network.
• The LUT-A-BiNN model was experimentally validated using 5G test signals to assess its predistortion performance.
• Performance was compared against a single-box BiLSTM neural network predistorter.

Main Results:

• The proposed LUT-A-BiNN predistorter achieved a 5 dB improvement in adjacent channel leakage ratio (ACLR) compared to the standalone BiLSTM predistorter.
• The hybrid approach effectively separated and compensated for static and dynamic nonlinear distortions.
• The signal-agnostic performance characteristic of the BiLSTM was maintained.

Conclusions:

• The LUT-A-BiNN model presents a significant advancement in digital predistortion for nonlinear systems, particularly PAs.
• This hybrid approach offers a more effective solution for compensating complex distortions, leading to improved spectral efficiency.
• The validated performance demonstrates the practical applicability of this novel predistorter in modern communication systems like 5G.