Design of a flexible analog-to-feature converter for intelligent acquisition of biological signals
Analog-to-Feature conversion is an acquisition method thought for IoT devices.
The converter aims at extracting only the useful information for a given task, directly on the analog signal.
A2F converter can offer an accuracy of 97% for abnormal signal detection in ECG signals while greatly reducing the energy consumption.
The aim is to build the first generic A2F converter based on the Non Uniform Wavelet Sampling to acquire a broad range of low frequency signals (ECG, EMG, EEG, speech…).
Wavelet sampling offer several degrees of freedom (time instant, frequency and bandwidth) which permit to have an adaptive feature extraction.
Wavelet frames are highly redundant and very large so feature selection algorithms permits to remove redundancy and evaluate the different solutions to find the best subset of wavelet.
The aim is to find a set of wavelets easy to generate, with high classification performances and a low power consumption. Samples are finally used in a machine-learning algorithm to detect abnormal signals.