Adapted Compressed Sensing for IoT Applications

Compressed Sensing (CS) is a technique for the acquisition of signals using a number of measurements that is potentially much smaller than the number of samples at the Nyquist rate. It can be seen as an extremely simple encoding stage allowing a low cost compression that perfectly fits applications in which data storage and/or transmission costs are an issue while there are also stringent limits on local computation capabilities, a recurrent scenario in IoT applications. The optimization of such systems forces a definite departure from classical “universal” CS. Rather, spectral-like information on the signal to acquire can be exploited in a well-defined design flow that maximizes compression performance while keeping the computational complexity at very low level. The talk describes the basics of CS and introduces the ideas behind such a design flow. It then shows how to apply the method to the acquisition of few real-world signals mapping both compression and complexity reduction performances into the energy required by the task of locally acquiring a physical quantity that must be transmitted to a non-local hub. A pointer to some fully developed software allowing experiments and design of these adapted CS solutions will be given.