Ternary LDPC Error Correction for Arrhythmia Classification in Wireless Wearable Electrocardiogram Sensors

Joint Learning and Channel Coding for Error-Tolerant IoT Systems Based on Machine Learning

“…Error-tolerance is usually provided in ML-based IoT systems for reliable operation in safety-critical applications [7]; otherwise, errors causing data corruption may have catastrophic consequences (e.g., potential life and property loss). Therefore, protection techniques and strategies targeting different parts of the system, such as the model computation [8], data transmission [9], memories [10], ML algorithms [11] and error detection schemes [12], have been pursued in the technical literature.…”

“…In such a scheme, some parity bits are generated in the encoding process (prior to transmission) and then for example padded to the data being protected. If channel errors affect either the original data being protected or the parity bits when transmitted, then the redundancy can be employed to attain an acceptable bit error Joint Learning and Channel Coding for Error-Tolerant IoT Systems based on Machine Learning rate (BER) [13] (or symbol error rate (SER) for ternary data [9]) in the decoding process once the data is received. Therefore, when employing channel coding, additional hardware must be utilized for the encoding/decoding process as well as increasing the number of transmitted parity bits, resulting in an increase of power/energy; this overhead is significant when an ECC is utilized for error correction of multiple bits.…”

“…This poses a challenge for ECC design in IoT systems because conventional coding schemes incur a significant overhead to attain a low error rate for the decoded data. Therefore, improvements in ECC design with low hardware overhead have received significant attention in the technical literature [9], [16].…”

“…For example, ternary low-density parity-check (LDPC) codes [9] have been recently proposed to provide channel protection for an electrocardiogram (ECG) monitoring and arrhythmia classification system [17] (which is introduced in Section II in more detail). LDPC codes are one of the most widely used ECC types and have been used in various communication standards (e.g., 5G new radio [18], IEEE 802.11n for wireless local area network (WLAN) [19], IEEE 802.16e for mobile broadband wireless access system [20], DVB-S2/T2/C2 for digital video broadcasting [21]).…”

“…Therefore, alternative designs, such as that proposed for the ECG system of [17] must be considered to meet the hardware requirements. In this application, the IoT device is a wearable biosensor that converts the input signals to ternary symbol streams for efficient transmission, so a corresponding ternary LDPC code with a low complexity encoder is designed [9]. Compared to traditional LDPC codes with binary bits, this ternary coding scheme provides hardware-efficient encoding and thus it significantly reduces power dissipation.…”