Abstract
The low-cost ‘THz Torch’ wireless link technology is still in its infancy. Until very recently, inherent limitations with available hardware has resulted in a modest operational figure of merit performance (Range times Bit Rate). However, a breakthrough was reported here by the authors, with the introduction of ‘Cognitive Demodulation’. This bypassed the thermal time constant constraints normally associated with both the thermal emitter and sensor; allowing step-change increases in both Range and Bit Rate with direct electronic modulation. This paper concentrates on advancements to the bit error rate (BER) performance. Here, separate techniques are introduced to the demodulation software that, when combined, result in enhanced Cognitive Demodulation. A factor of more than 100 improvement in BER was demonstrated within the laboratory and approximately a 60-fold improvement in a non-laboratory environment; both at the maximum Range and Bit Rate of 2 m and 125 bps, respectively, demonstrated recently. Moreover, demodulation speed is increased by almost a factor of 10,000; allowing for real-time demodulation while easing future computational hardware requirements. In addition to these software advancements, the paper demonstrates important improvements in hardware that has brought the technology out of the laboratory, with field trials being performed within an office corridor.
Highlights
Introduction of thermal emitter calibrationThis section introduces a thermal emitter calibration algorithm to complete the model of the entire system
Where, We(t) is the thermal emitter’s input power, as a function of time t; Ce is the heat capacity of the thermal emitter; T(t) is the transient temperature that varies between the lower extreme limit of the ambient temperature T0 and its steady-state temperature; ke is the thermal conductivity; σe is the Stefan–Boltzmann constant; ε is the emissivity of the emitter; and Ae is the heated surface area of the thin-film thermal emitter
The recurrent neural network (RNN)[20] is a family of neural networks having a special structure that is tailored for time series analysis
Summary
Introduction of thermal emitter calibrationThis section introduces a thermal emitter calibration algorithm to complete the model of the entire system. Atmospheric attenuation and pyroelectric sensor are all adequately characterized, the mean-absolute-error (MAE) between the expected (predicted by the model) and received signals, corresponding to a predefined training sequence ( vtrain[t]and v[t] , respectively), is a non-analytical function of the emitter parameter values:. To address all these shortcomings, we introduce a neural network (NN) based demodulator, which combines thermodynamics-based modelling with artificial intelligence. Unlike template-matching based algorithms, which employ a MLE rule, we exploit a Maximum A Posterior (MAP) rule This directly approximates the posterior probability p(si = S1|v) using a neural network.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
