Coherence measurement system based on Self-Mixing(SM) interferometry with a Terahertz Quantum Cascade Laser(THz-QCL) has been demonstrated to be a highly sensitive sensor. Compared with traditional laser Doppler velocimetry, it has the advantages of self-collimation and eliminating the need for a separate photodetector and mixer, which makes the system more compact; moreover, the exploitation of terahertz source is expected to improve the upper limit of velocity measurement. In this paper, the nonlinear dynamic response to the motion of two independent targets is investigated, which manifests itself in the SM interferometric signal. Through the spectrum analysis of the SM signal, velocities and directions of the two targets are obtained. The research results demonstrate the feasibility of using THz-QCL SM interferometric system to sense the motion of targets, and provide a novel technique for laser Doppler velocity measurement of multiple targets with high speed.

Terahertz technology has important application prospects in the fields of physics, biology, medical diagnosis, non-destructive testing, and wireless communication. As an excellent candidate for both terahertz oscillator and detector, Resonant Tunneling Diode(RTD) has received increased attention in recent years. RTD is a semiconductor device based on the quantum tunneling effect, which simultaneously provides a negative differential resistance and nonlinear current-voltage characteristics. As for detector, it has distinct features of room-temperature operation, small size, easy integration, and high sensitivity, which makes it particularly suitable for applications scenarios such as short-range, ultra-high-speed terahertz wireless communication and the internet of things. This review focuses on recent progresses and applications of terahertz RTD detectors, and aims to provide further perspective on it.

The traditional terahertz detector can only obtain the amplitude of signal. An orthogonal heterodyne mixing structure is presented, which can obtain the amplitude, phase and polarization information at the same time, and effectively improves the sensitivity and information of the detector. Based on 40 nm Complementary Metal Oxide Semiconductor(CMOS) process, the detector connects the inductors in series between the switch stage and the transconductance stage of the traditional Gilbert double balanced mixer structure. The cascode IF(Intermediate Frequency) amplifiers are cascaded to further improve the response voltage of the detector. After simulation and optimization, the voltage responsivity of 1 GHz IF signal is 1 100 kV/W and the noise equivalent power is 26.8 fW/Hz1/2 under the condition of -50 dBm RF(Radio Frequency) power and 0 dBm LO(Local Oscillation) power. The output waveform shows good orthogonality. Meanwhile, a 1: 8 power divider is designed to distribute the LO power. The insertion loss of the power divider is about 5 dB at 150 GHz. The amplitude difference of four-differential-output is 0.8~1.2 dB, and the phase difference is 0.4°~1.7°.

Abstract：The detection of marine targets is an important part of radar signal processing, and has important application value in military and civilian fields. For the radar signal processing of marine targets, the presence of sea clutter has a great influence on the performance of the detection algorithm. Traditional radar signal processing methods are mostly based on statistical theory, and the detection performance under complex environmental conditions and diverse target characteristics may decrease significantly. In recent years, deep learning technology has developed rapidly, providing technical support for reliable target detection methods. This paper summarizes the development of target detection algorithms and deep learning methods in recent years. Starting from the data structure and dimensions of radar signals, using deep learning theory, the paper proposes a multi ?channel fusion flowchart based on two ?dimensional images, three ?dimensional frames processing, and multi ?dimension information fusion intelligent processing framework. Taking the detection of marine targets in navigation radar images as an example, a Precise ROI ?Faster R?CNN radar image detection algorithm is proposed. It is compared with the classic method of target detection through the training and testing of the constructed navigation radar dataset. It is indicated that the proposed method is an effective technical approach with higher detection accuracy and better generalization ability compared with Classical False Alarm Rate(CFAR) and Faster R? CNN methods. It can be an effective solution for intelligent navigation and target detection of marine radars.

In order to accurately determine the position of dangerous chemical gas leak, satellite positioning method is adopted according to the four-receive one-transmit inversion relation between the sensor and the sound source, without regarding the speed of the satellite and the difference between sending and receiving clock. The nonlinear overdetermined equations of sound source location are obtained, and an optimization algorithm combining Quantum Genetic Algorithm(QGA) and Fmincon constraints is proposed to solve the equations, which effectively solves the problem that other methods are easy to fall into local convergence. Experimental results show that compared with Least Squares(LS), Particle Swarm Optimization(PSO) and QGA, the improved method has high convergence performance, high positioning efficiency and high positioning accuracy. Under the same condition, the method improves significantly in terms of important performance evaluation indexes such as Average Root Mean Square Error(ARMSE), Relative Position Error(RPE) and Bessel's Mean Square Error(BMSE).

Aiming at the sensitivity of dual-UAV passive radar to the initial value of Direct Position Determination(DPD) of broadband radiator targets, a direct passive location algorithm is proposed based on Niche Topology Improved Vector Particle Swarm Optimization(NTVIPSO). The proposed algorithm combines the direct method and the indirect method, relying on the indirect method to construct the constraint conditions, and using the niche vector particle swarm to solve the exact position. The simulation results show that the proposed particle swarm algorithm is superior to the vector particle swarm algorithm and the neighborhood topology particle swarm algorithm. The accuracy of the proposed direct passive determination algorithm under the constraint of scale difference in the presence of disturbance is lower than the direct method but better than the indirect method, and the computational complexity is lower than the direct method but higher than the indirect method.

For the two?dimensional Planar Frequency Diverse Array(PFDA) radar, in the process of suppressing the interference which is close to the target in angle dimension and separable in range dimension, the classic beamforming algorithm has the problem of mainlobe distortion when the steering vector is mismatched. Based on the cross ?subarray two ?dimensional planar frequency diverse array with sine offset, this paper uses the variable load constraint with the fastest descent linear constraint minimum variance criterion to solve the correction weight vector when the steering vector is mismatched. When the estimated steering vector is 2°, it can still achieve effective correction and shape?preserving of the mainlobe when the pointing error is high. The simulation verifies the effectiveness of the method in this paper.

For Visual-Inertial Odometry(VIO), complex scenes, such as visual occlusion and moving objects may bring about abnormal visual measurements, which leads to dramatically drop of the positioning accuracy. To this end, a new method is proposed for detecting and processing abnormal visual measurements in VIO. Firstly, by selecting the detection index, setting the prior threshold and designing the detection classifier, the classification and detection of abnormal visual measurement are realized. After that, a multi-sensor fusion strategy and an adaptive error weighting algorithm are proposed, and these algorithms timely eliminate the influence of abnormal visual measurement which is inconsistent with the actual motion. Finally, the detection and processing algorithm of abnormal visual measurement are integrated into Open Keyframe-based Visual-Inertial SLAM(OKVIS), and the Error Detection and Solution of Visual-Inertial Odometry(EDS-VIO) framework is proposed. Evaluation results on the complex scene simulated dataset show that, compared with OKVIS, EDS-VIO has achieved better performance on the dataset, the average location error has been reduced from 1.045 m to 0.437 m. EDS-VIO improves the positioning accuracy and robustness of the VIO in complex scenes.

An One-Wire-Interface(OWI) based on Advanced Peripheral Bus(APB) is designed. APB bus can configure the data and read the status information of the interface, with strong flexibility and monitoring ability. The interface realizes half-duplex two-way communication between master and one or more slave devices through a common data line. Compared with other communication interfaces, OWI can save I/O port resources and reduce hardware costs. Based on the one wire transmission protocol, the data transmission sequence and state machine are designed. Through Register Transfer Level(RTL) simulation and Field Programmable Gate Array(FPGA) verification, the results show that the data can be stably and correctly transmitted through OWI and the data transmission rate can reach 100 kHz.

A novel quad-band hybrid miniaturized antenna based on Asymmetric Coplanar Waveguide(ACPW) feeders Composite Left and Right Hand Transmission Line(CLRH-TL) is proposed. The antenna is composed of ACPW connected with ground loading open resonant ring gap and monopole antenna terminal loading CLRH-TL unit. The conventional monopole antenna is mixed with the metamaterial zero-order resonator antenna. The measured results show that the impedance bandwidths of the antenna with the antenna return loss above 10 dB are 1.55~1.65 GHz, 2.4~3.7 GHz, 5.15~5.36 GHz and 5.71~5.89 GHz respectively, which have good radiation characteristics in the passbands. The antenna structure is simple, compact and easy to integrate. It can be used in GPS, WLAN and WiMAX wireless communication systems.

Aiming at the problems of beam scanning reflectarray antennas such as gain reduction and pattern deterioration during large-angle scanning, a method is proposed to improve the scanning performance of the reflectarray by defocusing the feed. Without loss of generality, a reflectarray antenna that works in the X-band is designed with a square patch structure. The effects of longitudinal defocusing at 15°, 30°, and 45°scanning angles and lateral defocusing at 10°, 20°, and 30°scanning angles on the performance of beam scanning reflectarray antennas are studied. The measured results show that when the scanning angle is small, the lateral defocus causes the beam direction to change, and the beamwidth does not change much. When reducing the size of the longitudinal defocus of the feed, the sidelobe level is reduced, and the gain at the center frequency is improved. The larger the scanning angle is, the more obvious the effect becomes. This is beneficial to improve the large-angle scanning capability of the beam scanning reflective array antenna.

In this paper, the development status of space high-precision Carbon Fiber Reinforced Plastics(CFRP) antenna reflector is given, fundamental problems of high-precision reflector such as material selection, structural design, thermal stability analysis and curing are summarized, and the research progress is primarily discussed focused on the fields of materials, space structure, thermal deformation and process. Moreover, some scientific problems about the high-precision reflector need to be further studied and prospected in aerospace engineering fields.

Traditional U-Net semantic segmentation model has a wide range of applications in the field of biomedical image, but the accuracy of the model is limited by the single-scale and the loss of information caused by up-sampling and down-sampling. To solve the above-mentioned problem, this paper proposes a high level and low level information-rich multi-scale biomedical image segmentation algorithm based on U-Net Encoder-Decoder structure and atrous separable convolution. The proposed algorithm consists of feature extraction network and multi-scale semantic segmentation prediction network. The feature extraction network uses atrous separable convolution and similar residual blocks to replace the up-sampling, down-sampling and convolution blocks in the original U-Net respectively, which maximizes the retention of information while increasing the perception field of vision. Channel attention mechanism is proposed to highlight the expression of core features and suppress the expression of unrelated region of background. The convolutional features with image-level features encoding global context are data mined at multiple scales to boost the performance further. Simulation experiments are conducted on collected embryos and DRIVE datasets, the experience results show that the proposed method has higher accuracy and robustness compared with U-Net and its derivative model.

In order to deal with the Software Defined Network(SDN) load balancing problem, the traditional ant colony algorithm is improved and a joint algorithm is proposed by combining with the server load balancing algorithm. In the proposed method, the weighted least connection scheduling algorithm(server side) is employed, and the server with the least load is selected; the Improved Ant Colony algorithm(Im-ACO) is adopted to find the optimal path to the selected server. Theoretical analysis and experimental results show that the proposed joint algorithm significantly improves the network performance with higher network throughput and lower packet loss rate.

The Monte Carlo(MC) integrators are the most commonly used approaches for numerically approximating radiant light energy propagating in virtual scenes. Distributing the Monte Carlo errors as blue-noise in the screen space significantly improves the visual fidelity. However, it is difficult to achieve such a goal efficiently for complex scenes with arbitrary sample counts and integration dimensionalities. A simple yet effective approach is proposed to deal with this problem. Firstly, bluenoise masks are adopted to scramble the quasi-random sequences, such that the rendering errors have a blue-noise spectrum at low sampling counts. Then, a filtering method is introduced, and the consequent rendering errors maintain the blue-noise spectrum at higher sample counts. Compared with several related works, the proposed method can efficiently obtain more stable blue noise results in different virtual scenes. The proposed method not only produces stable blue-noise errors but also has high computational efficiency, which is suitable for modern Monte Carlo-based renderers.

A continuous parallel collection system for multiple nuclear radiation data types with Multiple Serial ports Mapped to TCP ports(MSMTPS) is designed and implemented in order to deal with the problems of compatible parallel collection of multiple data types, lack of access serial port resources for the computer, and high bit error rate of collected data transmission. The ethernet is adopted to realize the interaction between nuclear radiation data acquisition system and computer. The system is composed of STM32F103 processor and W5500 ethernet chip, with multiple serial ports capable of interacting with ethernet ports simultaneously. By allocating queue buffers for serial ports that receive different radiation data, mapping multiple serial ports to the same TCP port, and establishing a connection with the host computer based on the TCP port, the fast bidirectional transmission between multiple serial ports and a single network port is realized. In the test, three serial ports are utilized to send 498 bytes, 499 bytes and 499 bytes of data at 0.5 s, 1 s, and 5 s intervals for 10 min. The results show that the system is stable and error-free, which meets the design requirements.

Radiation tolerant Micro Electro Mechanical System(MEMS) accelerometers are widely used in aerospace vehicles and nuclear industry. The reported researches are mostly concentrated on radiation hardness of MEMS devices' detecting circuits, while the radiation degradation mechanism of MEMS sensing part is still not well understood. Since MEMS device's micro-sensing part and detecting circuit must operate together, it is challenging to accurately distinguish the attributed degradation ratio of micro-sensing part when operating in the whole device, and acquire its radiation degradation mechanism by using micro and nano measuring techniques. To settle this issue, a radiation degradation measuring technique regarding to MEMS discrete parts is developed, and the radiation degradation attributed to MEMS sensing part is successfully extracted. Then, specifically fabricated MEMS contrast chips and electrical measuring techniques are utilized to study its radiation degradation mechanism. It is found that the radiation degradation attributed to MEMS sensing part is resulting from charging effect of the interface thin oxide layer, rather than charging effect of the parasite capacitor or changes of different resistor parameters. These trapped charges in the interface layer will produce an external electrostatic force on the movable proof mass, and then induce the output drift of MEMS accelerometer. In conclusion, this research experimentally infers a clear understanding of the degradation mechanism of MEMS sensing part under ionizing radiation environment, and it shows useful guidance on the future radiation hardness design of MEMS devices.