Radio Transmitter Machine Learning Waveform Pre-distortion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep learning-based wireless communication systems face challenges with non-linear power amplification, leading to in-band distortion and out-of-band emissions, which hinder detection accuracy and cause interference.
Innovation Solution
A radio transmitter is designed with machine learning models to mitigate distortion by mapping bits into a symbol grid, modulating into a time-domain waveform, and power amplifying it, while using neural networks to make the waveform resistant to nonlinear distortion and control out-of-band emissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If power amplification is applied to enhance signal strength, then transmission power is improved, but in-band distortion and out-of-band emissions increase
Solution Approach 1:
The patent applies preliminary action by using a machine learning model to pre-process and clean the time-domain waveform before it enters the power amplifier. The model predicts and removes potential distortions in advance, so that when the signal passes through the non-linear power amplifier, the harmful effects are minimized. This is evident in the processing chain where the learned waveform goes through a machine learning model (cleaning stage) before power amplification.
Solution Approach 2:
The machine learning model serves as an intermediary between the waveform generation and power amplification stages. It acts as a mediator that processes the time-domain waveform, removing artifacts and preparing it for power amplification in a way that minimizes distortion. The model receives the initial waveform and outputs a cleaned version that is more suitable for subsequent power amplification.
2Manufacturing precision
If machine learning models are used to clean waveforms, then distortion is reduced, but device complexity increases
Solution Approach 1:
The patent replaces traditional mechanical/filter-based distortion correction methods with a machine learning-based approach. Instead of using complex analog filters or multiple hardware stages to clean the waveform, the system uses a neural network model that has learned optimal cleaning patterns during training. This substitution reduces hardware complexity while achieving comparable or superior waveform cleaning performance.
Solution Approach 2:
The machine learning model is trained using specific parameters and loss functions that optimize waveform cleaning. During training, the model adjusts its internal parameters based on feedback from power amplifier characteristics and desired output waveforms. This parameter optimization allows the model to achieve effective distortion removal without requiring complex hardware configurations, as the complexity is encoded in the model's learned parameters rather than physical components.
Data Source
AI summary
According to an example embodiment, a radio transmitter includes at least one processor and at least one memory including computer program code. The at least one memory and the computer program code may be configured to, with the at least one processor, cause the radio transmitter to obtain bits to be transmitted; map the bits into a symbol grid in time-frequency domain; modulate the symbol grid into a first time-domain waveform; input the first time-domain waveform into a machine learning model, producing a second time-domain waveform; power amplify the second time-domain waveform, producing an amplified time-domain waveform; and transmit the amplified time-domain waveform. A radio transmitter, a method and a computer program product are disclosed.