Neural Network Detector Selection Without Softmax Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing MIMO detector selection methods in communication systems face challenges due to high hardware complexity, particularly with the implementation of normalization functions like softmax, which increases computational complexity and power consumption.
Innovation Solution
The proposed solution involves performing neural network-based detector selection at inference time without applying a normalization function, such as softmax, thereby reducing hardware complexity. Additionally, conservative detector selection methods are employed to ensure reliable performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a normalization function (softmax) is applied in neural network-based detector selection, then the selection accuracy is improved, but the hardware complexity and power consumption increase
Solution Approach 1:
The patent extracts and removes the normalization function (softmax) from the neural network-based detector selection process. By taking out this computationally intensive component, the system achieves detector selection without the associated hardware complexity and power consumption, while maintaining acceptable selection performance through alternative methods.
Solution Approach 2:
The patent employs simpler, less computationally expensive alternatives to the softmax normalization function. These cheaper computational objects replace the expensive normalization operation, reducing hardware requirements and power consumption while still enabling functional detector selection based on neural network outputs.
2Reliability
If a high complexity detector is used for all REs, then the BLER is minimized, but the power consumption increases
Solution Approach 1:
The patent implements dynamic detector selection where different detectors are chosen for different resource elements based on instantaneous channel conditions. This dynamic approach allows the system to use high complexity detectors only when necessary (poor channel conditions) and switch to lower complexity detectors when channel conditions are good, thereby maintaining reliability while reducing overall power consumption.
Solution Approach 2:
The patent applies local quality by tailoring the detector complexity to specific local channel conditions rather than using a uniform high complexity detector across all resource elements. Each RE receives an appropriately sized detector based on its specific channel state, optimizing the balance between reliability and power consumption at each local position.
3Use of energy by moving object
If the complexity of detector is reduced, then the power consumption is reduced, but the BLER increases
Solution Approach 1:
The patent uses dynamic detector selection to adapt detector complexity to instantaneous channel conditions. The system monitors channel quality and dynamically switches between low complexity and high complexity detectors, ensuring that power consumption is reduced during good channel conditions while maintaining low BLER during poor channel conditions through the use of higher complexity detectors when needed.
Solution Approach 2:
The patent incorporates feedback mechanisms where the system continuously monitors channel conditions and uses this information to select appropriate detector complexity levels. This feedback loop ensures that the detector complexity is always matched to the current channel state, preventing BLER increases while optimizing power consumption based on actual system needs.
Data Source
AI summary
A system and a method are disclosed for selecting a detector using an NN for each RE in a communication system. A method includes receiving, by the electronic device, at an inference time, a signal from a transmitting device; extracting features from the received signal; inputting the extracted features to an NN, which is trained, at least in part, with a normalization function; and selecting, for each RE, a detector from a set of detectors based on non-normalized outputs of the NN.