Requirements for IoT Sensors Using Embedded Compressed Sensing Encoders with Deterministic Sensing Matrices

We consider a WSN monitoring environmental signals. Sensor nodes compress data employing the Compressive Sensing (CS) framework, exploring signals sparsity to reduce the number of transmissions. We propose the use of the Karhunen–Loève Transform (KLT) as sparsifying basis and design maximally incoherent deterministic sensing matrices. Real–life signals are used in simulation, and their CS measurements are quantized before transmission. The rate–distortion performance obtained after the reconstruction of monitored data is evaluated. Important requirements for using this framework in an IoT scenario are also investigated, such as the response time (latency) of the WSN, the impact of packet loss in the reconstruction of the sensed signals and the energy consumption of sensor nodes to transmit coded measurements. Simulation results show that the KLT-based deterministic sensing matrices overcome both Noiselets and DCT–based deterministic ones, and the proposed CS coding scheme is robust against packets loss.