Neural Network Third Layer for Wireless Distortion Compensation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural networks used in wireless communication systems, such as those for Digital Pre-Distortion (DPD), require significant processing resources due to their complex structure, leading to high calculation demands.
Innovation Solution
A parameter determination apparatus and method that adds a third linear layer between existing layers in the neural network, where the third layer includes fewer nodes without non-linear activation functions, and learns weights between this layer and the subsequent layer to select valid connecting paths, reducing the number of processing operations required.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a neural network with complex structure is used for distortion compensation, then distortion compensation performance is improved, but processing amount increases
Solution Approach 1:
The neural network is segmented into multiple layers with different node counts. Specifically, the network includes a first layer with first nodes, a second layer with second nodes, and a third layer with third nodes where the number of third nodes is smaller than the number of second nodes. This segmentation allows the network to process information through multiple stages with varying computational complexity, maintaining distortion compensation performance while reducing overall processing requirements through strategic layer design.
Solution Approach 2:
Different layers of the neural network are assigned different qualities in terms of node count and activation functions. The third layer uses fewer nodes and no non-linear activation functions compared to the second layer, creating local variations in computational intensity. This local quality differentiation allows the network to maintain accuracy where needed while reducing computational load in specific regions, thereby balancing performance and processing amount.
2Adaptability or versatility
If the number of nodes in each layer is increased, then neural network capability is improved, but device complexity increases
Solution Approach 1:
The neural network employs dynamic connection strategies where not all nodes in subsequent layers are connected to all nodes in previous layers. Instead, specific connection patterns are established between layers, with the third layer having fewer connections than the second layer. This dynamic connection approach maintains neural network capability for distortion compensation while reducing structural complexity by avoiding complete connectivity between all layers.
Solution Approach 2:
The network uses partial connectivity rather than complete connectivity between layers. Specifically, the third layer has fewer nodes and fewer connections than would be present in a fully connected network with the same number of layers. This partial action approach maintains sufficient neural network capability for the distortion compensation task while significantly reducing device complexity by omitting unnecessary connections and nodes.
3Adaptability or versatility
If non-linear activation functions are added to each node, then neural network processing capability is improved, but processing amount increases
Solution Approach 1:
Non-linear activation functions are selectively applied only to certain nodes and layers rather than all nodes. Specifically, the third layer nodes do not include non-linear activation functions, while the first and second layers may have activation functions applied to their nodes. This local quality differentiation maintains the necessary processing capability for distortion compensation in critical layers while reducing the overall processing amount by omitting activation functions in layers where they provide minimal benefit.
Data Source
AI summary
A parameter determination apparatus adds a third layer between first and second layers of the neural network. The third layer includes a third node not including a non-linear activation function. Outputs of first nodes of the first layer is inputted to the third node The number of the third node of the third layer is smaller than the number of second nodes of the second layer. The parameter determination apparatus further learns a weight between the third and second layers as a part of the parameters and selects, as a part of the parameters, one valid path used as a valid connecting path in the neural network for each second node from connecting paths that connect the third node and the second nodes on the basis of the learned weight.