Design and implementation of auto encoder based bio medical signal transmission to optimize power using convolution neural network

Abstract

Real-time biomedical signal transmission requires IoTs and cloud infrastructure. In this work, we investigate feasible lossy compression approaches that leverage the temporal and spatial dynamics of the signal along with current algorithms based on Compressive Sensing (CS) that use signal correlation in space and time. These techniques are altered so they may be applied efficiently to a distributed WSN. To achieve this, we proposed Convolution Neural Network (CNN) based Optimized Bio-Signals Compression using Auto-Encoder (BCAE), which integrates auto-encoder and feature selection. Instead of using the entire signal as an input, we encode the main part of the signal and send it to the desired location. Reconstruction decrypts without signal loss. Realistic aggregation and data collection procedures can improve data reconstruction accuracy. We compare various techniques' reconstruction error vs. energy requirements. The simulation results reveal that packet loss is 40% and data reconstruction error is 5%. Data forwarding time is lowered by 16.36%, while network energy usage is cut by 23.59%. The proposed method outperforms with existing techniques and the results are validated using MATLAB.

Keywords

• Auto-encoder
• Feature selection
• Biomedical signals
• CNN
• WSN
• IoT