ML PRACH Receiver Retraining for Congestion-Aware Preamble Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing random access (RA) process in LTE and 5G cellular networks faces challenges such as frequent collisions, low preamble-detection probability, and reduced bandwidth due to PRACH congestion, particularly in urban areas with high UE density, leading to inefficient use of uplink resources and increased access delays.
Innovation Solution
A machine learning (ML) based algorithm with tunable parameters is deployed in the base station receiver to optimize preamble detection for contention-based and contention-free RA, using a training dataset collected from both controlled and normal preamble data, triggered by UE reports of poor PRACH performance, reducing collisions and false alarms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a threshold based approach is used for preamble detection, then the detection process is simple, but the detection accuracy decreases when the number of UEs increases
Solution Approach 1:
The patent changes the detection parameters from fixed thresholds to dynamic parameters adapted to current channel conditions and traffic load. The base station adjusts detection parameters based on the number of active UEs, channel quality indicators, and historical collision data, allowing the system to maintain high detection accuracy across varying network conditions without requiring complex real-time processing.
2Productivity
If more preambles are allocated for CBRA, then more UEs can access simultaneously, but the available preambles for CFRA decrease
Solution Approach 1:
The patent implements dynamic preamble allocation where the base station continuously monitors PRACH collision rates and CFRA request patterns, then adjusts the split between CBRA and CFRA preambles in real-time. When collision rates increase, the system automatically reallocates preambles from CBRA to CFRA, and vice versa, ensuring optimal balance between simultaneous access capacity and contention-free access reliability.
3Productivity
If PRACH resources are increased to handle more UEs, then access capacity improves, but uplink bandwidth for data transmission decreases
Solution Approach 1:
The patent changes the time-frequency resource parameters of PRACH dynamically based on network load. During periods of high random access demand, the system expands PRACH resources in the time domain by adding more random access opportunities. During low-demand periods, it contracts PRACH resources and releases the freed uplink bandwidth for data transmission, achieving temporal multiplexing between access and data functions.
4Device complexity
If a single threshold is used for all preambles, then the configuration is simple, but false alarm probability increases
Solution Approach 1:
The patent segments the single threshold parameter into multiple preamble-specific thresholds organized in groups. Preambles are divided into several groups, each with its own optimized threshold based on the specific interference patterns and collision probabilities characteristic of that group. This segmentation reduces false alarms while keeping the configuration manageable through automated group formation algorithms.
Data Source
AI summary
A receiver of a Base Station (BS) uses a Machine Learning (ML) engine for random access preamble detection. The ML engine is trained from time to time wherein one of the triggers for retraining is a plurality of User Equipment (UE) sending a message to the base station, according to an aspect of this invention, that is indicative of poor preamble detection performance, upon which the BS collects controlled preamble detection dataset by selecting and engaging a group of so-called trainer-UEs. The retraining is performed by using both aforementioned controlled as well as normal preamble detection operations.