Existing Visual Question Answering (VQA) models have explored various visual relationships between objects in the image to answer complex questions, which inevitably introduces irrelevant information brought by inaccurate object detection and text grounding. To address the problem, we propose a Question-Driven Graph Fusion Network (QD-GFN). It first models semantic, spatial, and implicit visual relations in images by three graph attention networks, then question information is utilized to guide the aggregation process of the three graphs, further, our QD-GFN adopts an object filtering mechanism to remove question-irrelevant objects contained in the image. Experiment results demonstrate that our QD-GFN outperforms the prior state-of-the-art on both VQA 2.0 and VQA-CP v2 datasets. Further analysis shows that both the novel graph aggregation method and object filtering mechanism play a significant role in improving the performance of the model.

We demonstrate the generation, of a mmWave signal via the injection of an optical frequency comb (OFC) into an integrated tunable dual distributed Bragg reflector (DBR) laser as well as the fiber transmission and the processing of this signal by an optical beamforming network (OBFN). The dual DBR laser is based on a hybrid indium phosphide (InP)polymer photonic integrated circuit (PIC). Two different cases have been examined in which the microwave signal is centered around 39 GHz and 60 GHz respectively, carrying quadrature amplitude modulation (QAM) formats at 0.5 Gbaud. In this proof-of-concept scenario, the OBFN consists of two optical paths, where the relative true time delay is induced by an optical delay line (ODL). Extensive comparison between the back-toback (B2B) case and scenarios with transmission over 25 km of standard single-mode fiber (SSMF) has been made using the error-vector magnitude (EVM) and the bit-error ratio (BER) as evaluation criteria. In all cases, error-free transmission was suggested for all QPSK signals, whereas a worst-case EVM of 11.8% was observed for 16-QAM transmission, successfully showcasing the concept's potential. The generated microwave signal's frequency can be set arbitrarily high, provided that highspeed photodetection equipment is available for the detection and down-conversion of the signal. Extension to higher antenna elements (AEs) numbers is straight-forward, relying only on the number of available photodetectors.

No abstract

Optoelectronic technology is expected to be the cornerstone of sub-THz communication systems, enabling access to and use of the vast frequency resources found in this portion of the spectrum. In this work we demonstrate a photonics-enabled sub-THz wireless link operating in real-time settings, using a PIN-PD-based THz emitter, and a THz receiver based on an ultra-fast photoconductor. The real-time generation and detection of the information signal is performed by an intermediate frequency (IF) unit based on a commercially available mmWave platform, operating at 1.6 GBaud. The evaluation of our setup takes place on two phases. Firstly, a homodyne scenario is demonstrated, where the same pair of lasers is used at the transmitter and receiver side. Secondly, we demonstrate a heterodyne scheme, employing optical phase locking techniques at the receiver. Errorfree operation was achieved in both scenarios at a bit rate of 3.2 Gb/s, over 1 m of free-space with ambient air. The broadband characteristics of our setup were validated, achieving error-free transmission over a 0.22 THz range, spanning from 90 up to 310 GHz. Finally, the stability of our real-time link was successfully demonstrated, showing stable SNR performance at the receiver with adaptive capabilities, over a time period of 5 min and 22 sec.

An all-photonic THz-receiver PIC comprising an on-chip frequency stabilization scheme and a novel InP-based photoconductive antenna is presented. Characterization of the key photonic building blocks shows the functionality of the PIC.