Rabi Mahapatra

In this paper, an absorber with multi-band, tunable, high Q, and high sensitivity, based on terahertz periodic two-dimensional patterned graphene surface plasmon resonance (SPR), is proposed. The absorber consists of a bottom metal film separated by a periodically patterned graphene metamaterial structure and a SiO₂ dielectric layer, where the patterned graphene layer is etched by “+” and “L” shapes and circles. It has simple structural features that can greatly simplify the fabrication process. We have analyzed the optical properties of a graphene surface plasmon perfect metamaterial absorber based on graphene in the terahertz region using the finite-difference method in time domain (FDTD). The results show that the absorber device exhibits three perfect absorption peaks in the terahertz bands of f1 = 1.55 THz, f2 = 4.19 THz, and f3 = 6.92 THz, with absorption rates as high as 98.70%, 99.63%, and 99.42%, respectively. By discussing the effects of parameters such as the geometrical dimensions of patterned graphene metamaterial structure “+” width W1, “L” width W2, circular width R, and the thickness of the dielectric layer on the absorption performance of absorber, as well as investigating the chemical potential and relaxation time of patterned-layer graphene material, it was found that the amplitude of the absorption peaks and the frequency of resonance of absorber devices can be dynamically adjusted. Finally, we simulated the spectra as the surrounding refractive index n varied to better evaluate the sensing performance of the structure, yielding structural sensitivities up to 382 GHz/RIU. Based on this study, we find that the results of our research will open new doors for the use of multi-band, tunable, polarization-independent metamaterial absorbers that are insensitive to large-angle oblique incidence.

In this paper, two parallel graphene strip structures are adopted to achieve tunable plasma-induced transparency (PIT) sensors in the terahertz band. Both graphene bands act as bright modes, and a PIT window appears due to the weak hybridization between them. A Lorentzian oscillation coupling model is fitted to the simulation results of the proposed structure by the finite-difference time-domain (FDTD) method and is in good agreement with the simulation results. The performance of the PIT system can be controlled by tuning the geometrical parameters of the structure. In addition, the resonant frequency of the PIT window can be dynamically adjusted by changing the chemical potential and carrier mobility of the graphene strips. When the chemical potential of graphene increases from 0.2 eV to 1 eV, the amplitude modulation depth of the PIT window (2.832 THz, 3.684 THz, and 4.386 THz) can reach 92.39%, 96.14%, and 90.4%, respectively. Furthermore, due to its dispersion characteristics, the realized PIT window has a sensitive response to the surrounding medium, and the sensitivity can be as high as 1.25 THz/RIU. This PIT effect-based graphene microstructure has important implications for the future design of terahertz modulators, optical switches, and ultrasensitive sensors.

In this paper, a plasmonic electro-optical encoder based on graphene at THz frequency is proposed. The surface plasmon polaritons (SPPs) in the graphene–insulator–metal structure are excited by an incident TM wave with a wavelength of 9.3 μm. Graphene plasma waveguides have extremely high confinement, relatively low losses, and high tunability. The switching mechanism is based on the application of an external voltage to locally change the chemical potential of the graphene for encoding. Setting the chemical potential to 1 eV allows SPPs to propagate while lowering the chemical potential to 0.1 eV prevents the SPPs from propagating. A 4 × 2 encoder with a minimum encoding extinction ratio (ER) of 37 dB, a maximum modulation depth (MD) of 99.99%, and a structure area of 0.8 μm² is proposed based on the design rules and simulations using the finite-difference time-domain (FDTD) method. In terms of the obtained results, the proposed structure can be used in optical integrated circuits.

To achieve high accuracy navigation for deep space probes, the DOR measurement relies on the corrections from the TDD estimation of reference weak Quasar signal, which is guided by the forecasting data compensation. However, disturbed by the interference, the forecasting data sometimes is unavailable, which directly leads to invalid compensation and Quasar TDD estimation failure. In this paper, an intelligent 2-D chart method with auto-detection is proposed for DOR measurement without forecasting data support. Consisted of 2-D correlation chart generation, two stages auto-detection and rough estimations iteration refining, the proposed scheme realizes the auto-detection and TDD blind estimations of weak Quasar signal. The validation experiment is established with developed RS-SDR systems, which are installed in ground stations for space DOR navigation and positioning. Results show that the presented 2-D correlation chart extends the system weak Quasar signal detection ability with signal SNR can reach -23dB. The overall accuracy of designed two stages auto-detection for Quasar correlation spectral line search is 85.17%. The proposed scheme improves the utilization rate of Quasar data collected by the system by 24.01%. And the average P-F curve fitting residue error which reflects the TDD estimation accuracy is 0.1931/rad. It proves an equivalent precision to accurate forecasting data processing, which meets the DOR measurement accuracy requirements.

Support Vector Machine (SVM) is a supervised machine learning model for classification tasks. Training SVM on a large number of data samples is challenging due to the high computational cost and memory requirement. Hence, model training is supported on a high-performance server which typically runs a sequential training algorithm on centralized data. However, as we move towards massive workloads, it will be impossible to store all the data in a centralized manner and expect such sequential training algorithms to scale on traditional processors. Moreover, with the growing demands of real-time machine learning for edge analytics, it is imperative to devise an efficient training framework with relatively cheaper computations and limited memory. Therefore, we propose and implement a first-of-its-kind system of multiple FPGAs as a distributed computing framework comprising up to eight FPGA units on Amazon F1 instances with negligible communication overhead to fully parallelize, accelerate, and scale the SVM training on decentralized data. Each FPGA unit has a pipelined SVM training IP logic core operating at 125 MHz with a power dissipation of 39 Watts for accelerating its allocated computations in the overall training process. We evaluate and compare the performance of the proposed system on five real SVM benchmarks.