An Adaptive Compressive Sampling Reconstruction Method for the Internet of Things

Although compressive sampling (CS) is an efficient data compression approach for the Internet of Things (IoT), this work shows that the effectiveness of CS is highly dependent on the reconstruction techniques employed, and that the underlying shape of a signal also affects reconstruction accuracy. Hence, choosing a suitable CS reconstruction technique is crucial to maximizing the advantages of CS. We propose an adaptive method that continuously and automatically identifies the most appropriate reconstruction techniques based on the shape of the signal and a CS dictionary. The method then uses the Kolmogorov-Smirnov (KS) test to continuously monitor whether the signal changes its shape. When a signal change is detected, the method uses the Least Absolute Shrinkage and Selection Operator (LASSO) to reconstruct the signal initially and employs the Longest Common Subsequence (LCSS) to identify the next best CS reconstruction techniques from the CS dictionary. The proposed method is tested using real vibration IoT data, and the results are promising. Experimental results demonstrate the method's effectiveness, showing: (i) a 5.39%-22.08% increase in reconstruction accuracy compared to randomly applying CS methods, (ii) a 1.64× speed improvement over Basis Pursuit (BP), and (iii) an additional 1.5% accuracy gain when employing a signal-truncation approach.