Neural-Network Constellation Mapping for Phase-Noise-Robust Terminals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing modulation methods such as QAM and APSK suffer significant performance degradation due to phase noise and power amplifier (PA) saturation points, and existing adaptive modulation and coding (AMC) do not adequately account for these non-linear factors.
Innovation Solution
A terminal equipped with a control unit that pre-trains a neural network to map bit sequences to indices and generate modulated symbols, considering phase noise and PA saturation points, and a transmitting unit that uses these symbols for communication.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If higher frequency and wider bandwidth are used for transmission, then data transmission speed is improved, but phase noise increases
Solution Approach 1:
The neural network is pre-trained offline with phase noise variance as input to learn optimal constellation configurations that are robust to phase noise. This preliminary training allows the system to adapt to phase noise conditions without real-time computation delays, enabling high-speed transmission while maintaining robustness against phase noise degradation.
Solution Approach 2:
The system dynamically adjusts constellation parameters (such as point distribution and spacing) based on phase noise variance levels. By changing these parameters according to the measured or estimated phase noise conditions, the system optimizes transmission performance across different frequency and bandwidth configurations, mitigating the harmful effects of phase noise while maintaining high data rates.
2Reliability
If high-power amplifier is used to gain higher signal-to-noise ratio, then signal quality is improved, but non-linear distortion increases
Solution Approach 1:
The neural network learns to adjust constellation parameters to account for power amplifier non-linearities and saturation effects. By modifying the constellation geometry and point spacing based on PA characteristics, the system maintains high signal-to-noise ratio while compensating for non-linear distortion, thereby improving overall transmission reliability without requiring linearization of the power amplifier.
Solution Approach 2:
The system transforms the harmful non-linear distortion into a manageable parameter by training the neural network to recognize and adapt to PA saturation characteristics. The pre-trained network converts the potentially harmful non-linear effects into useful information for optimizing constellation selection, turning the PA's non-linear behavior into a predictable factor that can be compensated for through intelligent modulation design.
3Ease of manufacture
If existing modulation methods are used, then implementation simplicity is maintained, but performance degradation due to phase noise and PA saturation occurs
Solution Approach 1:
The complex task of optimizing modulation schemes for different phase noise and PA saturation conditions is performed in advance through offline neural network training. This preliminary action creates a lookup table of optimal constellation configurations that can be quickly selected during transmission without complex real-time calculations, maintaining implementation simplicity while achieving superior performance robustness against phase noise and PA non-linearities.
Solution Approach 2:
The pre-trained neural network serves as an intermediary between the channel conditions (phase noise variance, PA saturation point) and the modulation scheme selection. Instead of directly implementing complex adaptive modulation algorithms, the system uses the neural network as a mediator that translates channel parameters into optimal constellation choices, simplifying the implementation while improving reliability through intelligent adaptation to adverse conditions.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A terminal includes: a control unit configured to: map bit sequences to indices; output one-hot encoder vectors associated with the indices; and pre-train a neural network (NN) to output modulated symbols in response to the one-hot encoder vectors, a variance of phase noise, and a power amplifier (PA) saturation point, as inputs; and a transmitting unit configured to carry out a transmission to a base station by using modulated symbols that are selected based on a constellation generated by the pre-trained NN.