Hierarchical Deep Channel Coding for Short-Block Low-Latency Decoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current channel coding techniques, such as LDPC and turbo codes, face challenges in meeting the low-latency requirements of 6G cellular data networks for augmented reality and ultra-reliable low-latency communication (URLLC) applications, especially with short block lengths, and are inefficient in handling non-AWGN channels, leading to suboptimal performance and redundant feedback mechanisms.

Innovation Solution

A hierarchical channel coding approach using a concatenation of two autoencoders with feedback from layer-0 to layer-1, incorporating learned recurrent neural network codes and conditional feedback for iterative error refinement, which reduces redundant feedback and improves decoding performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional channel coding techniques (LDPC, turbo codes) are used, then reliability is improved, but latency increases and block error rate performance deteriorates for short block lengths

Engineering Contradiction:
Improveblock error rate performanceVSAvoidlatency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent divides the channel coding task into two hierarchical layers: layer-0 autoencoder for initial encoding/decoding and layer-1 autoencoder for iterative error refinement. This segmentation allows the system to process short block lengths efficiently at layer-0 while providing optional refinement at layer-1, thereby reducing latency compared to traditional single-stage codes while maintaining reliability.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements dynamic feedback mechanisms where the layer-1 autoencoder selectively refines errors based on layer-0 decoding outcomes. The feedback loop adapts the coding strategy based on channel conditions and error patterns, enabling optimal performance for short block lengths without incurring fixed latency overhead of traditional approaches.

Inventive Principle:
Principle #15Dynamics

2Reliability

If traditional channel coding techniques are used, then reliability is improved, but device complexity increases due to redundant feedback mechanisms

Engineering Contradiction:
Improvedecoding performanceVSAvoidfeedback mechanism complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent segments the feedback mechanism into hierarchical layers where layer-0 provides primary encoding and layer-1 provides selective refinement. This segmentation eliminates redundant feedback by only activating layer-1 when layer-0 decoding is insufficient, thereby reducing overall system complexity while maintaining decoding performance.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements intelligent feedback control where the layer-1 autoencoder is activated based on layer-0 decoding performance metrics. This feedback mechanism avoids unnecessary processing by only engaging additional complexity when errors are detected, optimizing the trade-off between reliability and device complexity.

Inventive Principle:
Principle #23Feedback

3Reliability

If hierarchical deep channel coding is used, then block error rate performance is improved for short block lengths, but device complexity increases

Engineering Contradiction:
Improveblock error rate performanceVSAvoidcoding structure complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent divides the coding structure into two manageable layers with distinct functions: layer-0 for primary encoding and layer-1 for error refinement. This segmentation makes the complex hierarchical structure more tractable and implementable while achieving superior block error rate performance for short block lengths compared to traditional single-layer codes.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements a nested hierarchical structure where layer-1 is embedded within the overall coding framework and only activates when needed. This nesting approach allows the system to maintain simple operation for most cases (using only layer-0) while providing enhanced protection when required, balancing complexity and performance.

Inventive Principle:
Principle #7Nested doll (Nesting)

4Reliability

If iterative error refinement with feedback is used, then decoding performance is improved, but loss of time increases due to multiple transmissions

Engineering Contradiction:
Improvedecoding performanceVSAvoidtransmission time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary error correction at layer-0 before potential layer-1 refinement, eliminating many errors in advance. This preliminary action reduces the need for iterative refinement and multiple transmissions, thereby improving decoding performance while minimizing additional transmission time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback control where layer-1 iterative refinement is selectively activated based on layer-0 decoding outcomes. This feedback mechanism avoids unnecessary iterative processing and retransmissions by only engaging additional refinement when errors are detected, optimizing the balance between decoding performance and transmission time.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP4261746A1Hierarchical deep channel coding
Publication Date: 2023.10.18 NOKIA SOLUTIONS & NETWORKS OY
  • EP4261746A1 patent drawingFigure 1
  • EP4261746A1 patent drawingFigure 2
  • EP4261746A1 patent drawingFigure 3

AI summary

Systems, methods, apparatuses, and computer program products for hierarchical coding. One method may include encoding, by a layer-0 autoencoder, bit sequence x into a layer-0 codeword. The method may further include transmitting, by the layer-0 autoencoder, the layer-0 codeword over a channel, and decoding, by the layer-0 decoder, a received layer-0 codeword.