**1. Introduction**

Biometrics is a technology that is widely used as means of access control in different application domains, from smartphones and automobiles, to healthcare, e-commerce, security and the military. The main idea behind it is to perform the operation of matching a given value of a biomarker with a reference value that represents an individual [1,2].

Typically, biomarkers are some physiological and/or behavioural attributes that are unique for each human being. The most recent biometric systems rely on various biomarkers, but the most commonly used ones are based on fingerprint scanning, face and/or voice recognition, iris scanning, hand geometry, finger vein, etc.

The main requirements that a biomarker should fulfill are the following [1]:


There are a few additional requirements that are considered a big plus:


As was shown in [3–6], the electrocardiogram signal (ECG) is a very promising biometric marker. Historically, ECG was mostly used for medical (diagnostic) purposes, but recent progress in the fields of consumer electronics and information technologies has already enabled it applications in biometric systems [7–9].

On the other hand, ECG-based identification systems are still not quite widespread in commercial and government services, and many of them are provided as research prototypes or very new commercial products that have just appeared on the market [10,11].

Given this situation, some questions remain open, for example ECG signal reliability and reproducibility over time, its behavior in real-world applications, the potential impact of the measurement process, hardware (sensors) configuration and matching algorithms on the identification performance, etc.

Consequently, the main aim of this paper is to answer some of these questions and provide some estimates and insights on how robust and reliable the ECG biometric markers are. In order to accomplish this task and figure out some common trends and limitations for the ECG-based identification, we carried out some experiments for a broad range of system configurations (various datasets, matching algorithms, records length, lead type, sampling rate, ADC resolution, etc.).
