Diagnostic-aware polar coding and adaptive hybrid automatic repeat request with reinforcement learning for hemodynamic telemonitoring of PTSD in 5G/IoT networks

Wearable telemonitoring of post-traumatic stress disorder (PTSD) relies on heart rate variability derived from the sphygmogram, where the diagnostic conclusion rests on the temporal positions of a few fiducial points, a distinction to which conventional channel metrics are blind. The paper introduces a diagnostic weight of a sparse representation coefficient, obtained from the sensitivity of the fiducial points to a perturbation of that coefficient, and shows that the energy ordering and the diagnostic ordering of the discrete cosine transform (DCT) are opposite. A diagnostic integrity criterion, measured as the deviation of the SDNN index, replaces the bit error rate (BER) as the optimization target. The reliability gradation that a polar code creates among its information subchannels carries the most diagnostically significant bits on the most reliable subchannels, which yields unequal error protection (UEP) at no cost in redundancy or code rate. The retransmission decision follows an estimate of diagnostic risk from soft decoder outputs, with the threshold adapted by a reinforcement learning (RL) agent. Simulation over a multipath non-line-of-sight (NLOS) channel gives a margin of 1.5 dB at equal diagnostic fidelity, a near-threefold reduction of dicrotic notch detection failures at equal block error rate (BLER), and a quarter fewer retransmissions.