A Hybrid Biosignal Compression Model for Healthcare Sensor Networks

Recent development in wearable sensor technology helps to collect biological signals at a low cost. Collecting and analyzing different biomarkers are anticipated to improve the preventative health care system through customized medical applications. The wearable sensors are battery-operated and based on technology with restricted resources, and they must use simple approaches to handle storage and energy properly. To achieve this goal, apply a lossy predictive coding-based method to compress signals at the sensors to reduce the energy needed to transmit data, minimize the storage space required, and extend battery life. This paper proposes a combination of Long-Short-Term-Memory(LSTM) and XGBoost-based hybrid model to address the challenge of sparse signal reconstruction in terms of multiple sampling vectors under compressed sensing, based on the assumption that the signal vectors are jointly correlated. The Proposed model achieves better compression efficiency than the baseline models considered for comparison and minimizes the energy consumption and storage space required. The performance results show that the proposed model extends the lifetime of the sensors and HSN.