Towards Flexible and Low-Power Wireless Smart Sensors: Reconfigurable Analog-to-Feature Conversion for Healthcare Applications

Abstract

Analog-to-feature (A2F) conversion based on non-uniform wavelet sampling (NUWS) has demonstrated the ability to reduce energy consumption in wireless sensors while employed for electrocardiogram (ECG) anomaly detection. The technique involves extracting only relevant features for a given task directly from analog signals and conducting classification in the digital domain. Building on this approach, we extended the application of the proposed generic A2F converter to address a human activity recognition (HAR) task. The performed simulations include the training and evaluation of neural network (NN) classifiers built for each application. The corresponding results enabled the definition of valuable features and the hardware specifications for the ongoing complete circuit design. One of the principal elements constituting the developed converter, the integrator brought from the state-of-the-art design, was modified and simulated at the circuit level to meet our requirements. The revised value of its power consumption served to estimate the energy spent by the communication chain with the A2F converter. It consumes at least 20 and 5 times less than the chain employing the Nyquist approach in arrhythmia detection and HAR tasks, respectively. This fact highlights the potential of A2F conversion with NUWS in achieving flexible and energy-efficient sensor systems for diverse applications.

Keywords: Gm-C integrator; analog-to-feature converter; arrhythmia detection; feature selection; human activity recognition; low power; non-uniform wavelet sampling; wireless smart sensors.

Figure 1

Architecture of the acquisition system with the reconfigurable A2F converter based on the NUWS feature extraction.

Figure 2

Examples of wavelets: (a) two non-overlapping and (b) two overlapping Haar wavelets; (c) the real part of the Gabor wavelet.

Figure 3

Overall methodology applied throughout the study.

Figure 4

Accuracy of binary arrhythmia detection performed by different classifiers without the SFS feature selection.

Figure 5

Impact of basic SFS algorithm on binary arrhythmia detection performed by NN classifiers (built with TensorFlow): (a) feature selection only with IG; (b) feature selection with IG + basic SFS.

Figure 6

Metrics of binary arrhythmia detection performed by NN classifiers with adapted SFS.

Figure 7

Multiclass HAR performance with basic SFS algorithm: (a) accuracy achieved by NN classifiers of different structures (*—classifier with default initializers); (b) confusion matrix of the preferable dictionary-classifier configuration.

Figure 8

Accuracy of the multiclass HAR performed by NN classifiers with the adapted SFS algorithm and features generated from (a) acceleration and angular velocity signals; (b) acceleration signals.

Figure 9

Hardware implementation of the feature extraction chain.

Figure 10

Impact of the quantification level on the accuracy of the multiclass HAR performed by NN classifiers with adapted SFS algorithm ($nEx{t}_{max}=8$).

Figure 11

G_m-C integrator schematic based on the design in [68].

Figure 12

Results of DC simulations in the open-loop configuration to define (a) $V_{t1}$ and (b) $V_c$.

Figure 13

Results of AC simulations in the integrator configuration with a common-mode voltage $V_{CM}=0.9 \mathrm{V}$ to define $V_{t2}$ and $V_{gc}$ voltages: (a) cut-off frequency $F_c$; (b) DC gain error $G{E}_{DC}$.

Figure 14

Impact of common-mode voltage $V_{CM}$: (a) $v_{in}$–$V_{out}$ characteristic; (b) cut-off frequency $F_c$ and DC gain error $G{E}_{DC}$.

Figure 15

Impact of biasing voltages $V_{b1}$ and $V_{b2}$.

Figure 16

Metrics of binary arrhythmia detection performed by NN classifiers with optimized SFS.

Figure 17

Energy required to process 10 s of (a) ECG and (b) inertial signals with different wireless sensor approaches.

Figure 18

Energy required to process 10 s of (a) ECG and (b) inertial signals with different A2F approaches.