Generally, a network administrator designs, constructs, and operates an enterprise network. To manage a network correctly, the network administrator needs to understand its configuration. Since inconsistencies between the network design understood by the administrator and the actual network configuration might arise due to mistakes or errors, a method for automatically detecting such inconsistencies is needed. The following five techniques are necessary for this purpose: (i) a machine-readable notation to represent the network configuration, (ii) a tool to write down the network design in the machine-readable notation defined in (i), (iii) a tool to automatically detect the current network configuration and write it down in the machine-readable notation defined in (i), (iv) a tool to compare the two outputs generated in (ii) and (iii), and (v) a network management framework using machine-readable notation to simplify the network administrator’s tasks. This paper employs the network ontology called Bonsai for (i). Bonsai can represent not only a physical network configuration but also a network configuration with various network virtualization technologies such as VLAN (Virtual Local Area Network) and overlay. This paper proposes three tools, nc-design, nc-detect, and nc-diff, for (ii)-(iv) and confirms that they work as expected in a test network. In addition, this paper proposes a network information sharing framework called KANVAS (Knowledge base system in wide Area Networks with general Versatility, Availability, and Scalability) for (v). Evaluation results in a test network with virtualization technologies show that the proposed network management method can localize a network failure in a practical time.

The MQTT Version 5.0 specification, developed in 2019, introduces shared subscriptions to distribute messages among multiple subscribers, improving the scalability and availability of MQTT-based data infrastructures. However, the current subscriber selection method for shared subscriptions is implementation-dependent, and many vendor products and open-source software only support static load balancing methods. Static load balancing methods cannot adapt to the changes in subscriber data processing performance due to the influence of other workloads, potentially leading to processing delays for the overall workload. In this study, we propose a method to achieve dynamic load balancing for shared subscriptions by managing the state of a group of subscribers connected to the broker in real-time. Our proposed method extends the MQTT Control Packet to enable dynamic load balancing based on the number of pending messages by transmitting the status of subscriber message processing to the broker at any time. The evaluation results demonstrate that the proposed method reduces the latency of the overall workload by 98% compared to the existing static load balancing methods.

In Japan, some core network topologies have been designed using regional railways information, and three models with regional characteristics, such as population distribution, have been published. For the fifth generation (5G) of wireless communication and beyond 5G (B5G), metropolitan area networks (MANs) are expected to be the primary network to satisfy the strict service requirements. To design MAN models with regional information, a MAN model design algorithm with regional railways information and compression algorithm to reduce the size of the designed network were proposed. Two models, Tokyo TMN23 and TMN12, for the 23 wards of Tokyo were designed using these algorithms. Those models have already been utilized in some existing studies. In this study, we designed MAN models for Aichi prefecture, which includes Nagoya city, and Osaka prefecture, which includes Osaka city, to further enhance B5G research in Japan. The model designed for Aichi has 62 nodes and 84 links, whereas the model for Osaka has 66 nodes and 101 links. These models have day and night versions of the population densities. We investigated the characteristics of the two designed MAN models by comparing them with the Tokyo TMN models. Thus, this study developed the first MAN models of Aichi and Osaka prefectures by using railways information. The proposed models are developed by considering regional characteristics and are expected to be widely used for network simulation and analysis in research.

Japan recently suffers from a serious social issue that is becoming difficult to maintain not only logistics services but also regional transportation services of route buses and district railways nationwide due to the progress of labor shortage and regional depopulation caused by declining birthrate and aging population. Automated driving technologies, including Connected and Automated Vehicles (CAVs), are indispensable to realizing sustainable logistics and regional transportation services, and require the reduction of further communication latency in mobile networks that support the technologies. Unfortunately, the current mobile networks, i.e., 4G and 5G specified by 3GPP, suffer from a problem where the communication latency increases with the distance between the serving base stations (BSs) and a mobile core network (CN) node over the transport network (TN). This problem becomes more severe, especially in areas far from the backbone network station accommodating the CN. This is because the User plane (U-plane) packets shall always be passed through the CN in the current mobile networks. In order to solve this problem in the current mobile networks, Segment Routing IPv6 for the Mobile User Plane (SRv6 MUP) that integrates 3GPP-based mobile network with Segment Routing over IPv6 (SRv6), has been proposed within IETF. SRv6 MUP is designed to allow the U-plane packets to bypass the CN and is equipped with the functionality to advertise the IP routing information to TN among BSs and CN in consideration of the user equipment (UE) mobility, including inter-cell handover (HO) between BSs. In the mobile network applied with SRv6 MUP, this design and functionality enable optimizing the U-plane packet route. This achieves not only low-latency edge computing but also low-latency Peer-to-Peer (P2P) communication among UEs since the U-plane packets bypass the CN. This paper first overviews SRv6 MUP technology as well as describes its distinctive mechanisms that support UE mobility, including HO between radio BSs, and that optimize the U-plane packet routes. Next, this paper proposes a novel concept of low-latency Vehicle-to-Network (V2N) and Vehicle-to-Network-to-Everything (V2N2X) utilizing the distinctive mechanisms of SRv6 MUP towards automated driving era. Then, it also presents field trial tests of low latency in-vehicle UE-to-UE communication applying SRv6 MUP under a live commercial 5G StandAlone (SA) network environment assuming application to low-latency Vehicle-to-Network-to-Vehicle (V2N2V), and the results demonstrate the effectiveness of SRv6 MUP in practical network environments.

To realize Society 5.0, the construction of urban digital twins is an urgent task. Among the various concern, human mobility is one of the most critical aspects that has attracted significant attention, and numerous approaches have been proposed so far. For example, the methods using CCTV cameras or LiDAR could perform high quality estimation, but they require the acquisition of sensitive data, e.g., video footage or gait data, and pose challenges of social acceptability due to the perception of surveillance. Therefore, this study aims to develop methods to estimate the number of people (crowdedness level) stay in various public spaces and the movement between places in the city (people flow) by utilizing WiFi and/or BLE (Bluetooth Low Energy) signals emitted by personally owned smartphones. In particular, this paper introduces a method for estimating crowdedness levels and people flow based on BLE advertising packet data (including RSSI and random addresses), and the five practical studies in the real-world settings.

In the Internet of Things (IoT), a type of Low Power Wide Area (LPWA) system, only a limited number of bits can be transmitted in a single packet. To increase the transmission of data bits, packet-level index modulation (PLIM) has been proposed. In PLIM, index modulation is achieved by assigning symbols to a packet index including the channel index and time slot index. However, in IoT, time slot index detection errors can occur owing to clock drift caused by low-precision oscillators in transceivers. To mitigate these errors, periodic time synchronization is typically employed. This study proposes differential PLIM (DPLIM) as an alternative to periodic synchronization. Although DPLIM is constrained by packet loss, it has the potential to reduce the overhead of periodic time synchronization. The proposed method is evaluated using a prototype sensor node designed for real-world deployment and implemented on long range wide area network (LoRaWAN).

With the growth of wireless communication technologies, the number of mobile devices equipped with wireless LAN (WLAN) interfaces have increased. Then, WLAN has become one of important infrastructures in the current access network. In addition, it is anticipated to serve as a key network in next-generation wireless networks. However, it is known that WLAN has various issues such as throughput performance degradation and deployment of AP in appropriate place have emerged. This paper introduces our challenge to address these issues based on mathematical models that describe physical phenomena observed in nature. Expecially, we focus on the access controls and transmission rate prediction techniques. Then, this paper demonstrates some results of our studies.

IP-based video transmission has advanced significantly, leading to widespread adoption for transferring video data in production environments. Meanwhile, HTTP streaming of compressed video remains the dominant method for distributing content to end-users, requiring data format conversions between production and distribution systems. We have been exploring the application of video transmission technologies to video distribution and proposing in-network video processing technologies. In this paper, we propose an overview of video processing technology within a network and two specific configuration technologies: a video switching and compositing method and a delay adjustment method, and evaluate them.

In sixth-generation mobile communications, considerations are being given to using the sub-Terahertz (sub-THz) band to achieve high-data-rate communications exceeding 100 Gbps. To realize such transmission, it is crucial to employ spatial multiplexing through a hybrid MIMO system in addition to leveraging its wide bandwidth. In this paper, after we state a channel model that we used in this study, we describe details about the hybrid MIMO system. Channel estimation error caused by noise degrades MIMO spatial multiplexing performance. To reduce noise, we introduced two countermeasures. The first method combines plural pilot signals in phase, and the second one forcibly replaces a noise portion of the impulse response of the channel with zero. We evaluate the characteristics of MIMO spatial multiplexing in the sub-THz band through computer simulations, assuming office and factory environments. The results indicate that the performance of the spatial multiplexing is much better than that of the single-layer case when the channels are perfectly estimated, and that channel estimation errors caused by noise have a notable impact on the transmission characteristics of MIMO spatial multiplexing. The degradation from the ideal case is about 15 bps/Hz or more depending on the environment. This is due to the discrepancy between the eigenvalues of the actual and estimated channels, which leads to an inadequate power allocation to each layer. However, when the two countermeasures are implemented, the degradation is reduced to about 3 bps/Hz, and MIMO spatial multiplexing achieves a higher channel capacity than single-layer transmission. Additionally, this study reveals that factory environments exhibit a greater spatial multiplexing effect than office environments.

In this paper, we propose a novel transceiver design method for orthogonal frequency division multiple access (OFDMA)-based wireless powered communication networks (WPCNs). In WPCNs, wireless devices harvest energy from radio frequency signals and subsequently use this energy to transmit information signals. To address the inherent doubly near-far problem in WPCNs, we explore the use of distributed antennas. The proposed method optimizes the wireless energy transmission time duration, energy beamforming vector, and subcarrier and power allocation for wireless information transmission through sum-rate maximization via alternating optimization. Simulation results demonstrate that, compared to conventional centralized WPCNs, the proposed distributed WPCN achieves a higher sum rate and effectively mitigates the doubly near-far problem.

In this paper, we propose a multiuser detection (MUD) method using interference suppression filters in an uplink low-density spreading orthogonal frequency division multiplexing (LDS-OFDM) system. The filters are designed without the knowledge of pilot symbols to minimize the multiple access interference (MAI) component for reliable data detection. We analytically demonstrate that MAI can be completely suppressed. Additionally, we show that the resulting composite channel matches a predetermined target channel up to a complex scalar, implying that channel estimation reduces to estimating the scalar. Furthermore, we introduce a simple active user detection method that does not require prior knowledge of the number of active users. The effectiveness of the proposed method is demonstrated through computer simulations.

To achieve the extreme high data rate and extreme coverage extension requirements of sixth generation (6G) wireless communication, high-frequency bands such as millimeter wave (mmW) and sub-terahertz (sub-THz) frequencies, and non-terrestrial networks (NTN), are two of the macro trends of 6G candidate technologies, respectively. However, non linearity of power amplifiers (PA) is a critical challenge for both high-frequency bands and NTN. Therefore, low peak to average power ratio (PAPR) waveform and modulation design becomes one of the most significant research topics for 6G. To maintain fifth generation (5G) New Radio (NR) backward compatibility, many waveform and modulation enhancement schemes based on DFT-s-OFDM have been proposed, including frequency domain spectrum shaping with spectrum extension (FDSS + SE), π/4-QPSK, symbol level interleaving, and unified non-orthogonal waveform (uNOW). This paper firstly provides a general description of the waveform and modulation enhancement schemes based on DFT-s-OFDM. To analyze the factors affecting PAPR, the time domain equivalent shaped-pluses of these enhancement schemes are derived and compared. Based on the analysis, three influencing factors that reduce the PAPR of DFT-s-OFDM are analyzed and summarized. Secondly, a low-complexity scheme, asymmetric DFT-s-OFDM, achieving time-domain compression and expansion (TD-CE) based on asymmetric DFT spreading, is proposed and discussed. Thirdly, the optimal constellation rotations for BPSK and QPSK to further improve the PAPR performance of asymmetric DFT-s-OFDM are proposed. Finally, the PAPR and block error rate (BLER) performance of these 6G candidate enhancement schemes for NTN and mmW scenarios are evaluated and compared based on link-level simulation with non-ideal factors (nonlinear PA and phase noise (PN)), which reveals that asymmetric DFT-s-OFDM with QPSK rotation can achieve the best PAPR and BLER performance among all the 6G candidate enhancement schemes for both NTN and mmW scenarios. In addition to its low-complexity and backward compatibility, asymmetric DFT-s-OFDM with QPSK rotation can be considered a promising candidate low-PAPR waveform and modulation scheme for 6G.

Multibeam satellite communication systems have emerged as a promising solution for enhancing spectrum efficiency and user connectivity in next-generation wireless networks. However, the performance of such systems often degrades due to severe multi-user interference resulting from aggressive frequency reuse and beam overlap. To address this, Symbol-Level Precoding (SLP) has been proposed for its ability to exploit constructive interference and improve signal quality beyond traditional block-level techniques. Despite its advantages, SLP suffers from high computational complexity, limiting its practicality for real-time applications in satellite systems with constrained processing resources. In this paper, we propose a novel Selective Symbol-Level Precoding (SSLP) method that significantly reduces the computational burden while maintaining near-optimal performance. First, we introduce a symbol selection technique that classifies transmitted symbols as either constructive or destructive based on known Channel State Information (CSI) and noise conditions. Then, we selectively apply SLP only to destructive symbols while directly transmitting constructive ones, thus reducing unnecessary optimization. Furthermore, we extend the proposed method to coded systems by dropping destructive symbols during transmission and reconstructing them using channel decoding. Simulation results show that SSLP achieves up to 70% reduction in computational complexity, while incurring a maximum Bit Error Rate (BER) degradation of approximately 2 dB compared to conventional SLP. By dynamically adjusting the number of precoded symbols, SSLP enables a flexible trade-off between performance and complexity, making it well-suited for next-generation multibeam satellite communication systems.

In this study, we propose a localization method for localizing interference sources affecting geostationary satellites, using a downlink signal from a single geostationary satellite. The proposed localization method cancels out the frequency offset caused in satellite transponder by taking double differential in terms of time and using a reference station. Furthermore, we examine the relationship between the localization accuracy and the satellite motion. Results indicate that Circular motion along the east-west and north-south directions within a two-dimensional plane parallel to the Earth’s surface, results in a wider intersection angle of FOA contours due to larger temporal transitions of the FOA contours. This wider intersection angle consequently improves localization accuracy compared to other circular motion directions. For example, using the Ku-band downlink signal from a geostationary satellite positioned at 150°E longitude to locate a Tokyo-based radio source yielded a localization error of approximately 8.4 km. Conversely, circular motion along the north-south and geocenter-exocenter directions within a two-dimensional plane perpendicular to the Earth’s surface, and circular motion along the east-west and geocenter-exocenter directions within a two-dimensional plane perpendicular to the Earth’s surface, resulted in narrower intersection angles of FOA contours due to smaller temporal transitions, thus leading to decreased localization accuracy.

Internet-wide scanning is essential for collecting network information but also serves as the first step of cyberattacks. In contrast to IPv4, the vast and sparsely populated IPv6 address space makes exhaustive scanning impractical. Instead, IPv6 scanners use hitlist and target generation methods to obtain potential targets. This paper presents a comprehensive study of IPv6 scanning activities, focusing on how fresh IPv6 prefixes are discovered and exploited by scanners. To investigate these scanning activities, we deployed multiple darknets and honeynets, combined with four DNS-based address exposure methods, to attract the scanners’ attention. Over an 18-month period, we collected 107.6 million scanning sessions from 23,199 unique source IP addresses across 170 ASes. The exposed addresses were quickly discovered by DNS scanners and added to hitlist, resulting in a substantial number of scans. The majority of the scans target the IPv6 addresses associated with IPv4, while IPv6-only DNS records largely remain hidden. This suggests that using IPv6 only for DNS without associating with IPv4 can help maintain anonymity. In addition, our analysis reveals the discovery process of fresh IPv6 prefixes, from target discovery, probing, hitlist registration to large-scale follow-up scans. The responsive addresses in the honeynet attract significant scans compared to the darknet. We categorize the scanners into three types based on their scan preferences. Finally, we analyze the intentions of scanners based on their scanning scale and behavioral patterns, including reconnaissance scans, volumetric and application-layer attacks. We also compare IPv6 scanning behavior with that observed in an IPv4 darknet. These findings shed light on the IPv6 scanning activities, offering insights to enhance IPv6 network security.

For live streaming, all users require the same content simultaneously. In this situation, multicast communication is expected to be easily realized in information centric network (ICN). Our careful evaluation results show that multicast communications surprisingly cannot be easily realized in Adaptive Bit-Rate (ABR) over ICN as it is. Root cause of this difficulty is that complete synchronization both in ABR bitrate selection behavior and also TCP congestion control behavior is required for realizing multicast communication, and this complete synchronization in both layers is quite difficult. Our stepwise improvements finally reveal that strong cooperation of the ABR algorithm and the TCP congestion control which enforces complete synchronization in these two layers, is significantly important for realizing multicast communications. As the final result of step-wise improvement, in this paper, we propose a novel traffic aggregation scheme, called Forced ABR/Congestion-Control-Algorithm (CCA) initialization (F-ABR-CCA-init) that synchronizes the segment download timing of the streaming users. Our evaluation results by computer simulations show that F-ABR-CCA-init substantially improves QoE by increasing selected bitrate of users owing to multicasting of ICN. Furthermore, we also reveal that F-ABR-CCA-init achieves higher QoE performance under several environments with multiple live streaming users in different bottleneck-bandwidths with different congestion severity.

The encryption of DNS communication for privacy protection has been gaining much attraction in recent years. The IETF has standardized protocols for discovering the encryption method of DNS communication between DNS full-service resolvers and authoritative DNS servers through RFCs 9461 and 9539. However, RFC 9461 has a limitation in that the SVCB records required to discover the encryption protocols supported by authoritative DNS servers must be resolved in plaintext. On the other hand, RFC 9539 generates unnecessary traffic by attempting encrypted communication even with authoritative DNS servers that do not support encryption protocols. Moreover, when both plaintext and encrypted communication are used to maintain compatibility, there is a risk of privacy information leakage. To address this issue, we propose and evaluate a new approach in which the parent authoritative DNS server indicates the encryption protocols supported by the child authoritative DNS server in the NS records during the zone delegation without violating the conventional DNS name resolution. The contributions of this paper can be summarized as i) Design and implement a privacy-preserving full DNS over HTTPS (DoH) architecture. ii) Propose NS record based information sharing to provide a compatibility with existing DNS architecture. iii) The functionality and performance of the proposed method are evaluated using the prototype system. iv) A discussion is provided on the proposed method and further enhancements for DNS privacy protection. Through the evaluations of the prototype system, we confirmed that the proposed architecture outperforms the existing approach using SVCB records to discover the encryption protocols supported by the authoritative DNS servers.

Vehicle edge computing provides abundant computing and storage resources at the network edge, aiming to meet the growing demand of latency-sensitive vehicular applications. Efficiently utilizing these resources is crucial for improving task execution efficiency in vehicular networks. However, existing research often overlooks the limited service capacity of edge servers and the dependencies present in vehicular applications, resulting in suboptimal offloading decisions. Additionally, the high mobility of vehicles introduces further challenges in making optimal offloading decisions. To address these issues, this paper proposes a task offloading algorithm based on Multi-Agent Deep Reinforcement Learning (MADRL) for vehicular applications with dependencies and diverse service requirements. The algorithm comprehensively considers vehicle mobility and the limited service capacity of edge servers to meet heterogeneous Quality of Service (QoS) demands in a heterogeneous vehicular network. In the proposed algorithm, we first decompose dependent task offloading into a sequential offloading process and model this process as a Markov Decision Process (MDP). The Multi-Agent Deep Deterministic Policy Gradient(MADDPG) algorithm is then applied to solve the problem. Simulation results show that the proposed scheme has better performance in reducing average task processing latency.

To reduce network operations workload, there will be strong demand for more efficiency in the inspection and repair of network facilities. Dispatch-less operations supported by remotely controlled drones can significantly improve work efficiency, but current network systems cannot meet the strict latency required for such operations. To achieve end-to-end links that have excellent stability and low-latency, this work proposes a real-time optical path control scheme that supports the a combination of optical and wireless network domains. Evaluations confirm that the proposed scheme can perform optical path control based on information from multiple domains fast enough for remote drone control. Further study will be needed to investigate the effects of coordination between a wireless controller, and to perform system-level evaluations including telemetry and optical switch response time.

Traditional intrusion detection systems (IDSs), leveraging machine learning (ML) algorithms, have improved the detection accuracy of unknown attacks by continuously updating ML models but have underestimated the context switching overhead between kernel and user spaces. To address this issue, existing studies have implemented real-time IDSs using neural networks (NNs) in the kernel space by offloading the quantized models trained with post-training quantization (PTQ) to extended Berkeley Packet Filter (eBPF). However, they cannot fine-tune the model parameters through the additional training because the PTQ applies the quantization to the trained model. In addition, their IDSs are based on supervised learning, which requires a large amount of labeled data. In this paper, we propose a real-time in-kernel IDS leveraging eBPF, unsupervised learning, and quantization-aware training (QAT) to enhance continuous learning. Evaluation results demonstrate that the proposed in-kernel IDS exhibits almost the same detection accuracy as the traditional user-space IDS. From the viewpoint of the packet processing speed, the proposed in-kernel IDS can serve 224 K packets per second while the user-space IDS can only serve 3.2 K packets per second.

Rapid changes in cross-polarization discrimination (XPD) of the Ka-band satellite radio wave signal (19.45 GHz) are observed at 1 or 0.1 sec intervals in 77 thunderstorm events from 1990 to 2006 at Osaka Electro-Communication University. About one third of the rapid changes are found to coincide with the cloud-to-ground lightning strokes on the south side of our earth station at the distance up to 20 km. Also, more than two thirds of them show that cross-polar phases move toward -90 deg, which means that canting angles of ice crystals move toward the horizontal direction. The XPD values and the canting angles of ice crystals associated with cross-polar phases are, for the first time, estimated just around the point of lightning discharges, using the observed XPD and cross-polar phases before and after the rapid changes. The XPD value at the lightning point is considered to become about 10 dB lower than other place before the lightning due to the almost vertical alignment of ice crystals in the thundercloud. In addition, the XPD values and cross-polar phases are directly traced during the rapid changes at the discharging point using those measured at 0.1 sec intervals. It is found that the XPD values due to the alignment of ice crystals almost disappear during about 0.5 sec, yielding the increase in XPD at the lightning point, while the mean canting angle of ice crystals at the lightning point moves from nearly vertical to horizontal direction during this short period. In the case of larger thunderclouds, however, the vertical alignment of ice crystals at the lightning point and their nearly horizontal alignment in other places seem to cause the increase in XPD due to depolarization cancellation before the lightning, and give rise to the decrease of XPD after the lightning.

The effects of satellite diversity on Ka and Ku band rain attenuation are investigated, using radio waves continuously received from JCSAT-1, BS, and CS (CS-3 or N-STAR) at Osaka Electro-Communication University from 1995 to 1998. The frequencies and polarizations of these satellites are converted into the same values in both Ka and Ku bands by a novel frequency scaling method, considering the difference in raindrop size distributions (DSD) for each rainfall event. As for the attenuation of the time percentages of 0.01%, the link unavailability is reduced by the satellite diversity techniques down to about 0.006% between the two paths and about 0.004% among the three paths, respectively, with azimuth angle separations of about 30 to 60°. The azimuth angular dependence of the satellite diversity is further investigated, using the radio wave signal levels obtained from JCSAT-3, Superbird-A, and Superbird-C in addition to BS and CS from 2003 to 2007. In this experiment, the same elevation angle as CS-3 or N-STAR is applied to all domestic satellites, by correcting the difference in the elevation angles together with the frequency scaling to the Ka band. Then, the improvement factor of the satellite diversity is found to be increased up to a factor of 2, as the azimuth angular separation between the propagation paths is increased from 7 to 76°. The improvement factor is, however, slightly decreased, when the elevation angle of the original orbital position is applied to each satellite, due to the deviation from that of CS-3 or N-STAR.

This paper investigates a round-trip delay (RTD) and channel-aware scheduler in wireless communications for feedback control applications. We propose a simple uplink (UL) and downlink (DL) scheduling method that achieves high spectrum efficiency while preventing the RTD from exceeding the allowable maximum delay as much as possible. The proposed method derives the scheduling metric for round-trip communication to minimize the appropriately defined risk of transmission failure based on the expected delay time of candidate terminals. The proposed method is simple and realistic since it reduces the number of parameters to be carefully tuned compared to the conventional α-proportional fair round-trip multiple deadlines (AP-RMD) method. Computer simulation results reveal that the proposed method accommodates more terminals while guaranteeing the maximum allowable RTD under a limited system bandwidth with fewer parameter adjustments than conventional methods, including AP-RMD.

With the widespread application of unmanned aerial vehicle (UAV) swarms in rescue operations, counter-terrorism, reconnaissance and other domains, the demand for effective swarm regulation and countermeasures has become increasingly urgent. Accurate swarm types identification serves as a fundamental prerequisite for achieving effective regulation and countermeasures. However, there has been minimal research on swarm types identification, and existing methods often face challenges such as complex feature extraction and low recognition accuracy. Aiming at the above problems, this paper proposes a method utilizing deep learning techniques to identify UAV swarm types based on micro-Doppler signatures. This approach leverages the unique characteristics of swarm micro-motions to enhance recognition accuracy. First, a radar echo signal model for UAV swarms is established, grounded in the principles of the micro-Doppler effect and electromagnetic scattering. Next, the micro-motion features are derived from the echo signal using the short-time Fourier transform (STFT), thereby constructing the swarm feature dataset. Finally, a ResNet18 network model leveraging micro-Doppler features is developed to achieve accurate swarm types identification. Simulation results indicate that ResNet18 achieves the best performance in UAV swarm types identification compared to widely used target recognition network models such as AlexNet, GoogLeNet and ResNet34. Furthermore, the proposed method requires fewer network parameters and lower computational overhead, while exhibiting strong robustness to noise. Notably, even at a signal-to-noise ratio (SNR) of 0 dB, the recognition accuracy for the six different swarm target types remains above 90%.