Participants
To ensure sufficient statistical power for detecting meaningful effects in our analysis, a minimum of 30 participants was determined through power analysis. Consequently, a total of 116 candidates were initially recruited from the Special Warfare Command of the Republic of Korea Army non-commissioned officer training course. During the recruitment process, the individuals were carefully selected to ensure that they had no diagnosed physical or psychiatric disorders. After the initial recruitment, 12 participants were discharged based on aptitude test results, and 4 were subsequently discharged owing to training-related injuries. To maintain the integrity of the data set, an additional 20 cases were excluded from the analysis. This exclusion was due to contamination by artifacts or mismatched EEG and ECG data sets during the data analysis process. The remaining candidates were then randomly assigned to two groups: MT (n = 42) and non-MT (n = 38). The the experimental protocol approved the ethical guidelines of the Declaration of Helsinki and was approved by the Institutional Review Board of the Korea University (KUIRB-2023-0058-01).
Military training (MilT)
All participants completed a 5-week program encompassing both MilT and MT. The MilT course consisted of five weeks of basic MilT followed by three weeks of advanced PLF training. Basic training included firearm training and long-distance marching, which induced considerable physical and mental stress. The subsequent advanced training course presented a more challenging environment compared to basic training. To minimize potential injuries and other issues associated with the rigorous nature of the training, more stringent control and supervision were implemented in the advanced training phase. In the sixth week, all individuals participated in PLF training using a mock-tower, and their training performance was evaluated and scored. To reduce the potential for experimental bias, the instructors who performed the MilT were unaware of whether the trainees had received MT training. The PLF training scores were based on an evaluation of focus and posture during three critical phases: on the mock tower, during the descent, and after landing. A maximum score of 56 points was assigned according to predefined criteria.
Mindfulness training (MT)
The MT program was performed for seven weeks, drawing inspiration from the Jon Kabat-Zinn program [28]. MT comprised two distinct components: a 60-min session led by an expert certified in the international mindfulness-based stress reduction program at the Mindfulness Center in Brown (121 South Main Street, Providence, Rhode Island 02903, USA) and a 20-min session designed for daily individual practice. The weekly 60-min sessions were structured in three segments. The first segment provided a 5-min overview of the MT process. The subsequent 30-min segment incorporated meditation techniques focusing on mindfulness, attention, and abdominal breathing. During the MT sessions, participants were instructed to sit comfortably in a quiet, noise-free environment. The sessions concluded with a 25-min guided interview led by the experts, focusing on techniques for managing negative emotions. For daily individual MT practice, participants engaged in a 20-min session each evening before sleep, using the same approach used in the second part of the weekly sessions. This individual practice was self-paced by the MT-trained participants. To minimize experimental bias, the MT and non-MT groups were separated before the start of MT training. Additionally, strict controls were applied to ensure that the groups were unaware of the purpose of the MT training. As a result, during training, the MT and non-MT groups were kept physically separate.
Data acquisition
Experimental setup and measurement conditions
On the same day of weeks 1, 3, and 5, all participants underwent a series of physiological and psychological measurements, including PSS, ECG, and EEG (Fig. 1). PSS was administered first, followed by ECG and EEG recordings. These recordings were performed in a quiet, dimly lit room while participants were seated in a relaxed, resting state to ensure data consistency. However, owing to environmental constraints, physiological measurements were not feasible during the outdoor PLF training sessions; only performance scores were recorded. This approach allowed for standardized conditions for physiological measurements while accommodating the logistical demands of the MilT environment.
Perceived stress scale 10 (PSS-10)
The PSS-10, which is widely used in social psychological research, was used to assess the degree of self-perceived stress in participants [29]. The PSS-10 is a 10-item scale used to assess the degree to which stress is perceived in the environment. The items provide information on how unpredictable, uncontrollable, or overloaded respondents’ lives were in the past month. The PSS-10 has a range of 0–40 points, with higher scores generally considered indicative of high levels of perceived stress. The scale has a high internal consistency and retest reliability.
Electrocardiography (ECG) acquisition
ECG (Polar H10, Polar Electro, Kempele, Finland) was recorded simultaneously with EEG for 12 minutes to investigate the quantitative effect of MT on cardiac response. The apparatus was fixed to the celiac plexus using a chest strap.
Electroencephalography (EEG) acquisition
Resting-state EEG was acquired in a dark, quiet room, with the participants sitting in a comfortable position with their eyes closed. Brain activity was recorded for 12 min. The recordings were made using the MINDD SCAN (YBRAIN, Gyeonggi-do, Republic of Korea) device, which uses 17 semi-dry channels (F3, F4, F7, F8, C3, C4, T7, T8, P3, P4, P7, P8, O1, O2, Fz, Cz, and Pz; reference: left ear lobe) based on the 10–20 international system, with a sampling rate of 500 Hz.

Illustration of the study design depicting the schedule of Military training and Mindfulness Training
Data processing and analysis
Preprocessing of physiologic signal
Data processing and analysis were performed using MATLAB 2023a (MathWorks, Natick, MA, USA) with the EEGLAB toolbox. Both EEG and ECG signals were preprocessed using finite impulse response (FIR) filters, with a 1–40 Hz band-pass filter applied to remove low- and high-frequency noise. EEG data were then converted to the frequency domain using a fast Fourier transform to calculate the power spectral density (PSD). The analysis focused on power within the alpha band (8–12 Hz), which is commonly associated with emotional regulation. ECG data were examined by extracting RR intervals as the primary measure for HRV. Artifacts such as ectopic beats, which were not removed during initial filtering, were manually identified and excluded to maintain data quality. To further ensure reliability, multiple researchers independently reviewed the ECG recordings to verify data accuracy.
Heart rate variability (HRV)
The temporal-domain analysis is derived from the NN interval, representing the time between consecutive R-peaks in a continuous ECG signal. In this study, we used several key metrics: the average R-R interval, the standard deviation of the NN interval (SDNN), the root mean square of successive differences in R-R intervals (RMSSD), and the percentage of normal R-R intervals differing by more than 20 ms (pNN20). Frequency-domain HRV parameters were estimated from the PSD of the NN interval series. The analysis focused on three main frequency components: low frequency (LF) from 0.04 to 0.15 Hz and high frequency (HF) from 0.15 to 0.4 Hz.
Alpha asymmetry on frontal cortex
To perform an EEG analysis, a common average reference was applied in order to remove the influence of neural activity that may occur in the reference on the left ear. To calculate alpha asymmetry (AA) in the frontal cortex, the power spectral density (PSD) was estimated using a fast Fourier transform with Hamming window and the alpha-band PSD was calculated using an average PSD of 8–12 Hz. To calculate the laterality of the frontal lobe, the electrode pair was matched with the F7 and F8 electrodes in the frontolateral area, and the F3 and F4 electrodes in the frontomedial area. The equation for alpha asymmetry on frontal cortex is as follows:
$$\beginaligned AA = \ln \left( Right\ channel\right) -\ln (Left\ channel) \endaligned$$
(1)
A positive AA indicates that the power of the right hemisphere is greater than that of the left hemisphere, whereas a negative alpha asymmetry on frontal cortex indicates that the alpha activation of the left hemisphere is higher.
Statistical analysis
Repeated measures analysis of variance (ANOVA) was used to compare the means of the PSS, AA in the frontal lobe (including frontolateral and frontomedial areas), and HRV across the different weeks. Tukey’s post-hoc test was then applied to perform pairwise comparisons, identifying weeks that differed significantly from each other. The significance value was adjusted using the stricter Bonferroni correction (\(\alpha\) = 0.017). An independent t-test was performed to examine differences between the MT and non-MT groups. Prior to statistical analysis, the Kolmogorov-Smirnov test was used to assess the normality of the data. To evaluate the association between PLF training performance and the presence of MT training, PLF training scores were analyzed. Owing to the lack of a clear gold standard for dividing PLF training scores into “good” and “bad” performers, we used an exploratory univariate logistic regression model. This model categorized performers based on the median score, resulting in a binary outcome variable representing performance level. The independent variables included a binary variable indicating the presence of MT training and continuous variables representing HRV parameters (indicating ANS balance) and frontal lobe asymmetry (representing emotional regulation). To examine the mediating effect of ANS balance and emotional regulation on the relationship between MT and performance, we performed mediation analysis following Baron and Kenny’s approach [30], with a Sobel test to confirm the significance of the mediating effect. Indicators from week 5 and PLF training results were used in the analysis. Bootstrapping with 1,000 resamples was applied to estimate population parameters through replacement random resampling. All statistical analyses were performed using SPSS version 25.
link