Active phased array radar (APAR) services play a key role in national strategic security equipment and directly support essential tasks such as the national strategic missile defense, over-the-horizon detection, anti-stealth detection, and remote guided attack. APAR technology was used during war in the 1960s and became very popular worldwide because of the emergence of urgent military needs. This technology has essentially affected the world military. Compared with traditional single pulse and pulse Doppler radar technology, APAR has helped to advance radar technologies and had a profound and wide influence on the development of radar. Each antenna element in an APAR is connected with corresponding transmission/reception (T/R) modules. By controlling the phase shifter to change the phase distribution on the antenna aperture, the antenna can be electronically scanned without any mechanical rotations while still covering the whole airspace. Therefore, compared with the disadvantages of traditional mechanical scanning radar, such as a large scanning inertia, limited data rates, a small number of information channels, and difficulty meeting the requirements of adaptation and multifunctionality, APAR has unparalleled advantages such as flexible and non-inertial scanning completed within microseconds, multifunctionality, high reliability, large data rates, low radar cross-sections of radar reflection, strong adaptability and less affected by electromagnetic interference. With the significant demand of modern national defense, radar equipment is continuously being developed with over-the-horizon techniques, accurate detections, strong stealth capabilities, etc. APAR research is advancing toward high-frequency bands, high gains, high pointing accuracies and low sidelobe levels. The high electromagnetic performance of antennas is achieved through strict design parameter requirements such as rigidity, quality of being lightweight and high efficiency heat dissipation of structures. Moreover, various parameters of antennas show high-dimensional and multi-field coupling relationships. It is more sensitive to interference from harsh battlefield environments, which deteriorates the electric performance of antennas and reduces the detection power, guidance accuracy and battlefield survival ability of radars. APAR, known as the eye of the three armies, is a piece of typical equipment involving interdisciplinary disciplines. Its structure, thermal and electromagnetic interactions and mutual restriction coupling relationship are defined as the structural-electromagnetic-thermal (SET) coupling problem of APARs. The main coupling problems include the following. (1) Feed errors affect the antenna electromagnetic performance, including the amplitude and phase errors of the feed network of APARs; failure of the radiation element, temperature drifts of the thermal sensitive electronic components (such as phase shifter in T/R module) and mutual coupling of the antenna elements will cause the amplitude and phase errors of the feed current, which will lead to the deterioration of antenna electromagnetic performance. (2) The structure errors affect the antenna electromagnetic performance, and there are random errors in the manufacturing and assembly of APARs. Vibrations, shocks and thermal power consumptions during service cause deformations of the antenna array, which eventually results in position offsets of the radiation element and changes of the electromagnetic amplitude and phase distribution on the antenna array. Moreover, this results in the change of the transmitted beam and finally, seriously affects the antenna electrical performance. (3) Thermal issues affect the antenna electromagnetic performance, and thousands of T/R modules, which consume large amounts of thermal power, are installed on APARs. On the one hand, it will cause thermal deformation of the antenna array structure; on the other hand, it will also cause performance degradation of the device and finally lead to the deterioration of antenna electromagnetic performance. (4) Changes in the coupling ofstructural, thermal performance and electromagnetic performance will cause deterioration. In APARs, the electromagnetic amplitude and phase of the antenna array will distribute correspondingly under different duty cycle operating modes, which results in a change in the thermal power consumption and temperature distribution. Thus, the thermal deformation of the antenna array structure is affected. Therefore, the structural-electromagnetic-thermal coupling problem of APARs has become a bottleneck issue that hinders their steady development and further enhancements of their performance. In this paper, the development of APARs on different ground-based, shipborne, airborne, missile-borne and spaceborne platforms has been sorted, and the structural characteristics of APARs on each weapon platform have been analyzed. The influence of service loads on APARs in different battlefield environments of land, ocean, air and space has been summarized. Then, the mechanism analysis and modeling method of SET coupling of APARs affected by antenna structure errors, high-temperature ablations of the radome, feed errors of T/R modules and failures of antenna elements are discussed. Moreover, the application of SET coupling technologies in the fields of APAR manufacturing accuracy, high-efficiency heat dissipation, lightweight integrated optimization, sparse array design, etc., as well as the key guarantee technologies in service environments such as APAR condition monitoring, displacement field reconstruction, electrical performance compensation, are summarized. Finally, future research on SET coupling technologies and their application prospects in different research fields are discussed.

Terahertz(THz) technology has shown great research potential and application value in biosensing, microscopic imaging, wireless communication, nondestructive testing and other fields. However, one of the bottlenecks restricting the rapid development of THz technology is the lack of high efficiency, high integration, low power consumption and easily modulated THz emitters. The THz radiation generated by spintronic terahertz emitter(STE) based on the ultrafast spin to charge conversion in ferromagnetic/non-magnetic heterostructure has the advantages of wide frequency band, low cost, easy fabrication and high field intensity, which has attracted the interest of many scholars at home and abroad in recent years. In this paper, the THz emission mechanism based on ultrafast spin dynamics are briefly introduced. And the feasible schemes to regulate and optimize its efficiency are further analyzed and reviewed. Furthermore, the new emission mechanisms and application prospects of STE are analyzed. New emission mechanisms include the emission of spintronic THz with different chirality and polarization states, and the exploration of STE based on new-type materials. And application prospects include spintronic THz spectroscopy, spintronic THz imaging and spintronic THz emission-regulation integration technology. Finally, the paper is summarized, and the future improvement, exploration and application of ultrafast spintronic terahertz emission sources, as well as the microscopic mechanism of spintronics driven by terahertz electromagnetic pulse are prospected.

Hyperspectral sensing technology can acquire spectral, spatial, radiation and other information synchronously, which provides the multi-scale, multi-angle, and multi-dimensional features of land covers. However, there are significant challenges in hyperspectral information extraction, e.g., spectral uncertainty, insufficient utilization of spatial information, and incomplete representation of collaborative information, resulting in poor information extraction and scene interpretation. The applications of hyperspectral interpretation, e.g., earth observation, requires multi-domain information extraction theories and methods to breakthrough these problems. In this survey, we firstly present the existing methods for hyperspectral information extraction and their main problems, and then introduce the fractional information extraction theory and methods of hyperspectral image, which consists of spectral dimension, spatial-spectral dimension, and collaborative dimension. Then, the main theories and applications are introduced, including spectral information adjustment, spatial-spectral information enhancement, and information fusion and transferring of multisource remote sensing data. For spectral dimension, the spectral uncertainty phenomenon makes it difficult to distinguish small targets from complex backgrounds. Focusing on this problem, the fractional-domain spectral information extraction method can improve the performance of hyperspectral anomaly detection. For spatial dimension, the complex spatial distribution of hyperspectral scenes and the lack of labeled samples make the scene interpretation challenging. Focusing on this problem, the fractional-domain spatial-spectral feature extraction methods can effectively generate more discriminative training features and improve the diversity of training sets, which contribute to handling small sample size problems. For the collaborative dimension, the fractional-domain multi-source feature extraction and fusion method can realize the joint use of multi-source and multi-domain features, and achieve high-precision classification. Finally, this survey points out the challenges and development trends of fractional information extraction theory for hyperspectral images. To breakthrough the limitations of hyperspectral data, e.g., high-dimension and low-resolution, it is important to improve the data quality at data- and feature-level. To solve the problem of unavailable training samples, transfer learning techniques are in need to fully exploit the spectral, spatial and collaborative information of the massive unlabeled data in hyperspectral remote sensing images. Targeting the global-scale earth observation by remote sensing, focusing on the condition when some modalities are missing, researches on domain generation and cross-scene classification are in need.

In recent years, artificial intelligence technology led by machine learning algorithms has been widely used in many fields, such as computer vision, natural language processing, speech recognition, etc. A variety of machine learning models have greatly facilitated people's lives. The workflow of a machine learning model consists of three stages. First, the model receives the raw data which is collected or generated by the developers as the model input and preprocesses the data through preprocessing algorithms, such as data augmentation and feature extraction. Subsequently, the model defines the architecture of neurons or layers in the model and constructs a computational graph through operators(e.g., convolution and pooling). Finally, the model calls the machine learning framework function to implement the operators and calculates the prediction result of the input data according to the weights of model neurons. In this process, slight fluctuations in the output of individual neurons in the model may lead to an entirely different model output, which can bring huge security risks. However, due to the insufficient understanding of the inherent vulnerability of machine learning models and their black box characteristic behaviors, it is difficult for researchers to identify or locate these potential security risks in advance. This brings many risks and hidden dangers to personal property safety and even national security. There is great significance to studying the testing and repairing methods for machine learning model security, which can help deeply understand the internal risks and vulnerabilities of models, comprehensively guarantee the security of machine learning systems, and widely apply artificial intelligence technology. The existing testing research for the machine learning model security has mainly focused on the correctness, robustness, and other testing properties of the model, and this research has achieved certain results. This paper intends to start from different security testing attributes, introduces the existing machine learning model security testing and repair technology in detail, summarizes and analyzes the deficiencies in the existing research, and discusses the technical progress and challenges of machine learning model security testing and repairing, providing guidance and reference for the safe application of the model. In this paper, we first introduce the structural composition and main testing properties of the machine learning model security. Afterwards, we systematically summarize and analyze the existing work from the three components of the machine learning model—data, algorithm, and implementation, and six model security-related testing properties-correctness, robustness, fairness, efficiency, interpretability, and privacy. We also discuss the effectiveness and limitations of the existing testing and repairing methods. Finally, we discuss several technical challenges and potential development directions of the testing and repairing methods for machine learning model security in the future.

Traditional camera accumulates photons during an exposure time window to generate a still image or a video in the form of image sequences, which lose the temporal process of the photons flow, leads to an irreconcilable dilemma between high-dynamic and high-speed imaging. Based on the fact that the pixels of the photoelectric sensor are independent, a new continuous photographing principle is proposed: each pixel converts the received photon flow into an electronic flow independently, continuously measures the photoelectric current and converts it into a digital flow, and represents continuously at the pixel level. Then the sequence array ranked according to the pixel layout is a continuous representation of the photons shotting on the sensing plane. The image of any moment can be obtained by intercepting the state of the sequence array, so as to realize continuous imaging.

Furthermore, the spiking continuous photographing principle that modulates the photon flow into a spike sequence is proposed: for each pixel, accumulate charge from the reset state, generate a spike as a flag once the specified threshold is reached, reset and repeat. The duration that a spike takes to be fired is called its spiking width, which is inversely proportional to the light intensity during this period. Based on this, the light intensity during this period can be estimated. The spikes sequence with natural temporal order is a digital representation of the photon flow process. The array of the spike sequences according to the spatial layout of pixels is called “viform”, which contains rich spatial and temporal information of the light process. An image at any moment can be calculated from viform, thereby ultrahigh-speed, high-dynamic and non-blurred continuous imaging is achieved, and the dilemma between high-dynamic and high-speed imaging of exposure imaging is solved completely.

The unique parameter of the spiking continuous photographing principle is the accumulation threshold Q, which corresponds to the number of photons/electrons required to fire a spike. The unique variable is the spike accumulation time τ,which is the easiest to measure accurately and can be concisely represented by natural temporal order. The light intensity at any moment is Q/τ, which is determined by the spike width τ, which range is (0, ∞). Thus, theoretically, arbitrary light intensity could by be represented, and lead to infinite dynamic range imaging. In the physical implementation, the strongest light could be represented depending on the shortest readout time τ? of the circuit, and the weakest light is the dark current, which can be accumulated as a spike within duration τ?. τ?/τ?makes up the imaging dynamic range. With conventional photoelectric devices and circuits, ultrahigh-dynamic imaging with 160 dB or even 180 dB can be achieved. When the imaging time sensitivity is required to be less than τ?, ultrahigh-dynamic imaging can be achieved by sub-threshold quantization of the accumulated voltage.

Photon and photoelectron flow are discrete statistical process that follows a Poisson distribution and is often a “segmental linear” process light intensity changes suddenly while the intensity is stable between successive mutations. A method so-called first spike encoding is proposed: only the moment of the first spike firing and its spike width are output when the intensity changes, and no spike is output thereafter to indicate repetition. This is the optimal lossless compression method of the spike sequence, where the dynamic range and time sensitivity can be significantly improved by improving the time measurement accuracy with almost no increase in representation data volume.

Using mature CMOS photonic devices and standard processes, two spike continuous photographing chips and spike cameras with spatial resolutions of 0.1 million and 1 million pixels have been developed, respectively. The chips use 40 000 Hz synchronous spike output, with a minimum spike width of 25 μs. Practical tests have verified the feasibility of the spike continuous photographing principle and its ability to achieve ultrahigh-speed, high-dynamic, and blur-free imaging.

Viform, the spike continuous photographing representation and the replacement of image and video, captures the temporal and spatial information of the photon flow, will fundamentally reshapes computer vision and visual information processing technology and industries. An open source algorithm framework, SpikeCV, achieves high-speed target detection, tracking, and recognition system that is faster than the human eye by thousands of times, with low computational complexity. As the symmetrical process of spike continuous photographing, spike continuous displaying modulates the ultrahigh-speed spike sequence into extremely high-speed photon flow, enabling high-speed display similar to natural light through glass, solving the motion blur and visual fatigue dizziness caused by low frame rate in traditional display systems. In addition, the combination of continuous photographing and continuous displaying can achieve a single-way transparent glass-like display and ultrahigh-speed light communication without medium.

With the national digital economy strategy to accelerate the implementation of construction, computing infrastructure has become an important engine for the development of the digital economy. The new generation network technology has changed from information data communication to intelligent processing of information data, and the integration of ubiquitous computing, storage and transmission resources has gradually formed the computing networking. As a new type of network architecture first proposed by my country, computing networking is an important basis for promoting the development of my country's information industry and supporting the development strategy of “network strengthening and digital China” in our country's “14th Five-Year Plan” development plan. In this context, computing networking is proposed to build the deep integration of network system and computing system. On the one hand, it improves network service quality, resource scheduling, and service function orchestration capabilities through computing, and realizes intelligent and efficient network computing services; on the other hand, convergence realizes cloud-centric computing resource operation, and uses the network to promote efficient scheduling of computing. However, the research on computing networking is still in its infancy, and no consensus has been reached on architecture, standards, and technologies. The design of relevant architecture and standards relies on traditional network technology, and there is a lack of building a unified computing networking standard system. The research faces many new needs and new challenges, such as “where to compute”, “what to compute” and “how to compute”. In this paper, based on the previous works of identifier networking and smart identifier networking, we propose a novel network architecture with its associated key mechanisms, namely computing integration networking (CIN) in view of the two inevitable trends of deep integration of heterogeneous networks and intelligent network innovation. By multi-network fusion networking and high-efficiency compatibility, multi-dimensional unified identification and intelligent analysis and mapping, on-demand networking and computing-network collaborative transmission, collaborative computing and optimization of computing-network integration, CIN constructs a new network system theory with deep integration of computing networking. It aims to provide diversified computing services for different industries and users on demand. CIN breaks through the development perspective of network private network, combines the development trend of network integration and intelligence, researches and builds a new theoretical system and construction, breaks through key core technologies, and realizes core technologies that are independently controllable and compatible to replace existing networks and functions. The original construction goal of leading international performance meets the major strategic needs of the country and the industry, promotes the development of the national digital economy, and lays a theoretical foundation for the overall layout of the national integrated computing.

Massive multiple input multiple output(MIMO) is regarded as the crucial enabling technology of the fifth generation(5G) mobile communication. Reliable, feasible, and reconfigurable indoor multi-dimensional performance testing and verification are significant to boost basic theoretical research, key technology breakthrough, and commercial deployment for the massive MIMO system. To implement the technique validation, scenario completeness test, and defect assessment for the massive MIMO device, developing reliable and realistic indoor emulated performance testing methodology is essential. The key challenge is how to reconstruct the rich and complete target propagation environment for the device under test(DUT). Based on early research, i.e., efficient virtual probe position determination method and closed-form probe weighting strategy, which realize accurate multi-dimensional emulation for various channel scenarios, this paper proposes a multi-dimensional channel emulation methodology for non-stationary wireless propagation with reconfigurable over the air(OTA) setup. The proposed channel emulation mechanism effectively tackles the challenge of the conventional OTA techniques, i.e., multi-probe anechoic chamber(MPAC), radiated two stage(RTS), and reverberation chamber(RC), to test the realistic performance of the beam-steerable massive MIMO device typically operating in dynamic channel. By employing the phase shifters with various bit numbers, this paper investigates the emulation results of the reconstructed channel spatial profiles in terms of maximum spatial correlation, weightes root mean square correlation error, power angular spectrum(PAS) similarity, and fixes beam power loss. According to the complete scenario and multi-dimensional channel emulation in this paper, it is concluded that the 4 bit phase shifter can cost-effectively provide an accurate phase matrix for indoor realistic propagation channel reconstruction. It also provides the theoretical basis and technical approach for quantifying the impact of phase shifter errors on the virtual OTA method. This paper introduces the phase shifter resolution factor, which has not been reported so far, to the phase-matrix-based virtual OTA test setup for the performance testing of the massive MIMO wireless devices in the realistic complex electromagnetic propagation environment with the reconstructed communication link. Specifically, it mimics the propagation of the waves radiated from probe antennas by adjusting the phase responses among the DUT elements in the phase matrix to enable the emulation of the target channel scenario where the key performance indicators of the adaptive massive MIMO DUT are validated. In addition, the 3 bit phase shifter could also provide satisfying emulation performance using sufficient virtual probes, leading to meaningful guidance for the phase shifter selection and low-complexity phase matrix construction. On the other hand, one can select the 5 bit phase shifter for safety redundancy to ensure emulation accuracy. Furthermore, this paper emulates a more realistic dynamic channel environment based on the ray tracing model to further validate the effectiveness of the derived conclusion on the non-standard measured channel scenario. The accurate ray-tracing channel emulation on multiple positions experimentally demonstrates the applicability and feasibility of the obtained conclusion.

Satellite communication is developing towards the direction of multi-frequency and large bandwidth transmission, multi-granularity flexible switching and forwarding, broadband flexible space high-speed networking, which puts forward higher requirements for the processing and switching capacity, and high-speed transmission capacity of satellite communication payloads. Traditional satellite communication systems usually use microwave technology to process and forward signals on the satellite. There are electronic bottlenecks in processing speed and transmission bandwidth, which makes it difficult to achieve multi-frequency, large bandwidth, multi-granularity, multi-channel data transmission and high-speed, high-capacity inter-satellite data interaction under the premise of taking into account the load weight, volume and power consumption, so as to meet the needs of future satellite communications. Microwave photonics combines microwave and photon technologies, and has the characteristics of wide working frequency band, large instantaneous bandwidth, no electromagnetic interference, flexible access, small size, light weight, etc. Satellite communication payloads based on microwave photonics can overcome the electronic bottleneck of traditional microwave technology by optical means, and greatly improve the transmission and processing performance of multi-frequency and large bandwidth communication signals of satellite communication systems. It provides a new idea for the design of satellite communication payload. Aiming at the limitations of satellite communication based on traditional microwave technology, this paper explores the new microwave photonic satellite communication payload architecture in the future, proposes the composition and implementation scheme of microwave photonic communication payload system, and focuses on the module composition and functional structure of broadband optical-electro/electro-optical array conversion module, large instantaneous bandwidth microwave photonics channelization unit, and multi-scale microwave photonics flexible switching module. On this basis, the key technologies such as broadband low stray microwave photonics frequency conversion, microwave photonics intensive channelization and optical switching matrix are further studied, and the corresponding solutions are proposed. At the same time, in order to reduce the system volume, weight, power consumption and improve the system stability, the referential ideas of the chip and integration technologies of the system are explored. Then, in view of the potential important role of satellite communication payloads of microwave photonic payloads in future satellite communications and space information networks, this paper analyzes and prospects the application scenarios of satellite communications based on microwave photonics payloads, and proposes three typical data transmission modes, namely local data processing and forwarding, remote data transmission and forwarding, and collaborative processing within distributed satellite clusters, which support Q/V, Ka, Ku and other multi-frequency, multi-bandwidth, multi-channel multi-service microwave signal reception and transmission, and high-speed, high-capacity, long-distance laser link data interaction. Finally, the development route and key problems to be solved of satellite communication payload technology based on microwave photonics are summarized and prospected, which provides important theoretical reference and key technical support for the design and application of future multi-frequency integrated satellite communication payloads.

Missile-borne radar is an important sensor for precise guidance. In the terminal guidance stage, mono-pulse mode is usually used to track the target accurately. Centroid jamming of corner reflector is an important passive jamming method for ships to counter anti-ship missiles. In the radar tracking stage, the corner reflectors released by the ship are in the same resolution unit with ship, and the mono-pulse measurement angle points to their centroid. Since the radar echo of the corner reflector is usually stronger than that of ship, the mono-pulse measurement angle is decoded, and the tracking point of the terminal guidance radar is biased to the corner reflector. With the rapid maneuvering of the ship, the distance between them gradually becomes larger, which will cause the “loss track” of ship. The anti-centroid jamming is essentially an estimation problem of unresolved multiple targets, which is usually divided into two steps: one is to determine whether the target is a mixture of target and corner reflector after detection; the other is to accurately measure the target parameters from the coupling echo signal. In the aspect of the existence detection of centroid jamming, there are mainly waveform method and statistic method. The former detects jamming by analyzing the characteristic difference of target echo, such as signal box shape analysis, wavelet transform, etc. The empirical judgment of this method is limited in practical performance. The latter regards the mixture as multiple targets, establishes the statistical model of mono-pulse ratio, and realizes the jamming existence judgment through the likelihood ratio detection. However, the models are complicated to deduce, and only apply to the case of single jamming. As for the target parameter measurement, jamming suppression methods which suppress the jamming through polarization filtering improve the mono-pulse measurement accuracy. However, prior information such as the jamming polarization characteristics or the amplitude ratio between the jamming and the target is required. Based on the mono-pulse statistical model, statistics are constructed to achieve multi targets angle estimation, such as moment estimation, maximum likelihood estimation, etc. This method is limited in practical application because it only applies to a single jamming and depends on a correct statistical model, also requires a large number of pulses to accurately estimate the parameters of the statistical model. The ship target contains abundant scattering mechanism, but the scattering mechanism of the corner reflector array is relatively single, and the polarization information of the two groups is different, which can be used to detect and suppress the corner reflector jamming. However, the existing polarization mono-pulse angle measurement methods adopt the multi-channel method to process the polarization information, and the scattering characteristic difference between the targets is not fully utilized, so the efficiency is limited in practical application. In the study of radar polarization, the synthetic echoes of adjacent targets under specific polarization states are very different. The accurate angle measurement of ship target can be achieved by adjusting the polarization states to suppress the corner reflector through the polarization modulation processing. Based on the theory of polarization modulation super-resolution, this paper analyzes the statistical distribution of characteristic beam and mono-pulse angle estimation for polarization array radar under the condition of centroid jamming of corner reflectors. Moreover, the principle of “ship-reflector” mixture detection and centroid jamming suppression is explained. On this basis, combined with the polarization scattering characteristic of corner reflector and ship, the corner reflector centroid jamming signal model is built. Then, the algorithm of jamming presence detection and angle estimation for ships is proposed and anti-angle induced deflection is realized. The simulation results show that the angle estimation accuracy of the proposed angle estimation method can achieve 0.1 times beamwidth, in addition, is valid with -\mathrm{20} dB polarization isolation. The rate of successful anti-centroid jamming is more than 80%.

Under the stage of terminal guidance, radar seeker can be easily deflated by corner reflector, which will lead to the sharp decline of the precision attack efficiency on the sea. It is urgent to improve the capability of anti corner reflector jamming. The new inflatable corner reflector works in wide frequency band with large radar cross section. Proper assignment of corner reflector array can simulate false targets similar with ships and form diluted jamming. Therefore, corner reflector array should be distinguished from ships. The common countermeasure is to recognize corner reflectors with HRRP and polarization decomposition. However, the former is constrained by spatial geometry, while the latter is affected by the accuracy of polarization measurement, both of which have limited anti-jamming performance. Physically, ships are continuous rigid bodies containing complex structures, and corner reflectors arrays are discrete points with uniform structures. The polarization characteristics of their echoes has significant difference. In this paper, based on the principle of polarization modulation technology, the signal models of corner reflector arrays and ships are established. With modulating the transmitting and receiving polarization, the coherent superposition effect among scatters is changed. The polarization-range 2D image is established and the correlation characteristic parameters are extracted to characteristic target’s difference. Finally, combined with support vector machine, a corner reflector array recognition method is proposed. The results of simulation experiments show that proposed method has stable recognition performance and can effectively combat the diluting jamming. Compared with polarization decomposition method, the recognition accuracy of proposed algorithm improves 7.5% under the low SNR condition, while improves 27.3% when the cross-polarization isolation is higher than -25 dB.

Benefited from the capability of spatial multiplexing, ultra-massive multiple-input multiple-output(MIMO) is one of the key techniques in the forthcoming 6G communications to provide high-speed data services and global massive connectivity. Traditional MIMO technique is realized by large-scale phased-arrays with high-resolution phase shifters. However, the high power consumption and hardware cost of phase-shifting circuits hinder the implementation of ultra-massive phased arrays in practice, thus limiting the deployment and development of ultra-massive MIMO. In this article, a new paradigm named holographic radio is considered for ultra-massive MIMO, where numerous tiny and inexpensive antenna elements are integrated into a compact space to realize high directive gain with low hardware cost, such that the electromagnetic waves can be flexibly regulated and the wireless communication performance can be effectively enhanced. We propose a practical way to enable holographic radio by a novel metasurface-based antenna called reconfigurable holographic surface(RHS). Specifically, RHSs are composed of numerous densely packed tunable metamaterial elements with low power consumption and low hardware cost. The feeds of the RHS are integrated with the meta-surface to generate electromagnetic waves propagating along the meta-surface and exciting the RHS elements one by one. Based on the holographic interference principle, each RHS element can control the radiation amplitude of the incident electromagnetic waves to construct a holographic pattern on the meta-surface, thus realizing holographic beamforming. Based on the working principle of RHSs, we introduce a novel multiple access technique called holographic-pattern division multiple access(HDMA). We develop the principle for HDMA with the main idea of mapping the intended signals for receivers to a superposed holographic pattern constructed by the RHS. A holographic beamforming optimization scheme is also developed to maximize energy efficiency of RHS-aided multi-user broadcast systems. To further verify the effectiveness of HDMA, we implement a prototype of the two-dimensional RHS and build an RHS-aided communication platform. Based on the HDMA scheme, the communication platform is capable of supporting real-time transmission of high-definition video for multiple users. Experimental results also show that the RHS has great potential to achieve high directive gain with simple wiring layout and low power consumption, thereby substantiating the feasibility of the RHS-enabled holographic radio. Moreover, future research directions and the corresponding key challenges for the RHS-enabled holographic radio are also discussed.