Table 3.
Comparison of the hardware setup proposed with analogous state-of-the-art designs and methods.
| Feature/Design | This work | Burello et al.63 | Feng et al.64 | Van Helleputte et al.65 | Yoo et al.66 |
|---|---|---|---|---|---|
| Analog Headstage | 8 ch + 32 filters | No | No | 3 ch ECG + ETI | 8 ch + 8 filters |
| Encoding | Asynchronous delta-encoding | Synchronous local binary pattern | External ADC @173.61S/s/@256S/s | Sigma-delta ADC @500S/s | Multiplexed SAR ADC @32KS/s |
| Application | HFO detection | Seizure detection | Seizure detection | Personal health monitoring | Seizure classification |
| Power consumption (headstage+processor) | 58.4 uW + 555.6 uW | 64 mW* | 45 mW*** | 183 uW + 191 uW** | 66 uW + 2.03 uJ / classification |
| Platform | Custom analog + DYNAP | Nvidia Tegra X2 | Altera Cyclone II FPGA | Custom analog+ ARM CortexM0 | Custom |
| Learning/Processing | SNN, randomized dataset | Hyperdimensional vectors, specific dataset | SVM, both randomized and specific datasets | ICA, PCA, CWT and feature extraction | SVM, specific dataset |
* Power for processing 24 channels.
** Power measured for R-peak detection with motion artefact reduction.
*** Not multichannel.