DeepWaveform Beam Inference for mmWave Latency Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current beam management techniques in mmWave networks require pilot sequences and beam sweeping, leading to inefficiencies and increased latency, especially in ad hoc networks where links are volatile and short.
Innovation Solution
The DeepBeam framework uses waveform-level deep learning to infer the angle of arrival and transmission beam of incoming transmissions without pilot sequences or explicit coordination between the transmitter and receiver, leveraging a trained classification module to determine waveform characteristics and associate quality metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pilot-based exhaustive beam sweep is used for initial access, then beam management can be performed with coordination between TX and RX, but the procedure becomes very inefficient taking up to 164 milliseconds to complete
Solution Approach 1:
The receiver performs beam management autonomously by passively analyzing incoming waveforms from data transmissions without requiring active coordination or pilot signals from the transmitter. The receiver uses its own processing capabilities to infer beam information and select optimal beam pairs independently
Solution Approach 2:
The patent replaces the traditional mechanical beam sweeping process with a signal processing approach using deep learning neural networks. Instead of physically sweeping beams to find the best alignment, the system uses waveform analysis and machine learning to infer beam characteristics and make intelligent beam selection
2Reliability
If traditional beam sweeping with SSBs is used, then beam alignment can be established through coordination, but the complexity becomes quadratic in the number of beams
Solution Approach 1:
The receiver independently performs beam management by passively analyzing waveforms from data transmissions. It uses neural networks to infer beam information without requiring the transmitter to actively coordinate or sweep beams, reducing the overall system complexity
Solution Approach 2:
The patent replaces the quadratic complexity mechanical beam sweeping process with a signal processing approach. The receiver uses waveform analysis and deep learning to directly infer beam characteristics, eliminating the need for exhaustive testing of all beam combinations
3Reliability
If coordination between TX and RX is required for beam management, then beam selection can be optimized, but the effective channel utilization is reduced
Solution Approach 1:
The receiver performs beam management autonomously without requiring transmitter coordination. It passively analyzes incoming data transmissions and uses neural networks to infer optimal beam pairs, allowing the channel to be used more efficiently without waiting for coordinated beam setup procedures
Data Source
AI summary
A system and method for beam management in a wireless network are provided. A learning module having a trained classification module processes received I/Q input samples to determine transmitted beam information of incoming RF transmissions. The learning module includes a beam inference engine to determine waveforms waveform characteristics of incoming RF transmissions beams, and an angle of arrival engine operative to determine an angle of arrival of the incoming RF transmissions beams on an antenna array. An incoming RF transmission beam and angle of arrival are selected based on the determined waveforms for beam management operations.