Table 5.
Comparison with State-of-the-Art Methods.
| Method | Battery Awareness | Key Reported Metric (from source paper) | BLE–ECG– Battery Co-design | Validation Dataset | Ref. |
|---|---|---|---|---|---|
| Fixed BLE Tx (0 dBm) | None | BLE 5.0 default Tx power range: −20 to + 8 dBm; nRF52840 max Tx = + 8 dBm, Rx sensitivity = − 96 dBm (at 1 Mbps). No SoC-adaptive control; BER degrades with distance under fixed power. | No | BLE 5.0 Spec. + nRF52840 datasheet | 3,32,38 |
| Duty-Cycle ON/OFF Adaptation | Threshold (ON/OFF) | [13]: 10.53% battery-life improvement with sleep-mode; 98% ECG classification accuracy on MIT-BIH arrhythmia dataset. [5]: threshold-based sleep-mode achieves 35% energy reduction (from 20 J to 13 J); 98% data accuracy; 97.5% PDR; latency < 92 ms. Neither study evaluates QRS clinical accuracy under degraded BLE channel conditions. | No | MIT-BIH (ECG) + WBAN simulation | 6,14 |
| ECG Compression (WNNM + AMP) | Indirect (rate reduction) | IWNNM-AMP achieves best PRD and RMSE across 6 algorithms; PRD reduced by 0.17–4.56 relative to WNNM-AMP (CR = 5.12, all noise intensities). Signal reconstruction only; no BLE channel or BER model; no QRS detection sensitivity reported. | No | MIT-BIH (clean channel) | 16 |
| Adaptive TX Power (per-channel BLE) | Partial (RSSI/SNR) | Up to 50% energy reduction; no clinical ECG accuracy reported | No | BLE testbed (no ECG) | 15 |
| DNN ECG Denoising (CNN/RNN) | None | VGG: 97.31% accuracy, 94.18% F1; ResNet: 95.07% accuracy, 89.61% F1 across 6 datasets for noise-type classification (BW, EM, MA); no BLE channel or battery model. | No | 6 public datasets | 19 |
| Proposed BAFE (this work) | Full (6-state SoC) | Se = 91.9% mean (MIT-BIH) at 20% SoC (BER = 1.78 × 10⁻²); 8/10 records ≥ 95% ANSI/AAMI EC57; 144.5% max lifetime extension; Pareto frontier fully characterized | Yes | MIT-BIH + PTB-XL | — |
All metrics in the “Key Reported Metric” column are taken directly from the cited source papers. No prior method jointly optimizes BLE transmission power, battery SoC modeling, and clinical QRS detection accuracy — the gap this work addresses. BW baseline wander, EM electrode motion, MA muscle artifacts, PRR packet reception ratio, AMP approximate message passing.