Stable QAM development with less BER on convergence time-compression type Q-learning for mass audio signal transmission
Kazuhide Okada · The Journal of the Acoustical Society of America · 2015
This paper presents method which protects the temporary hang-up on the communication line and sustains the demodulation of the clear audible signal at the receiver, when the mass sound data is sent from the transmitter. QAM is one of the digital modulation technology, mapping the modulating signal not only toward the phase but also toward the amplitude on Q-I constellation, derived from QPSK. This modulation can pack larger data in the fixed period than QPSK. But once the communication path is exposed by stuff jitter or random jitter, the quantization error on its coordinate occurs, which means coordinate axes often subtly rotate with returning to the original position. In order to minimize such quantization errors on demodulation, Q-learning as one of Reinforcement Learning was used in this study. In the design of the feedback system comprehending the agent and the environment, the angle of the reverse vibration of the upper each axis for the quick restoring to the normal quantization becomes an action at. And the reward rt is the relative baud rate, while the status st is BER, as I/F between Agent and the environment. The Quantity of a state action combination Q(st, at) was updated as the index which measures the value of actions, with the Quantity computation steps decreased by TTD (Truncated Temporal Difference) and Log-time overlooking at Q(st, at) in the training process of the experiment. The degree of control on coordinate's axes vibrations triggered by injected jitter was evaluated by the visible decrease in BER.