To address the growing demand for ultra-high-capacity communications, including uplink, in Beyond 5G and 6G systems, a virtualized terminal architecture utilizing terahertz and millimeter-wave relay transmission has been proposed. The system consists of a user equipment (UE) and multiple wearable relay devices (RDs), which function as distributed virtual antennas. This configuration allows the aggregation of transmit power across devices, enabling spatial multiplexing over several hundred meters, even under wideband MIMO conditions. Although previous studies have evaluated key components and interference effects among RDs, most evaluations were limited to physical-layer measurements using measurement instruments in short-range scenarios. In this study, we developed a prototype system that integrates an ultra-wideband baseband unit, terahertz RF front-end, and dual-polarization multi-beam antenna. We conducted comprehensive and practical end-to-end (E2E) evaluations under realistic transmission distances and MIMO configurations. The system achieved up to 38 Gbps physical-layer throughput using 4×4 MIMO over a long-range relay corresponding to approximately 200 m. Furthermore, real-time wireless transmission of uncompressed 4K video, which is a representative application, was successfully demonstrated, achieving stable playback with an actual throughput of 5.6 Gbps. These results confirm the effectiveness of the virtualized terminal as a promising solution for high-capacity uplink in future wireless networks.

This paper presents a blockage prediction system based on artificial intelligence/machine learning (AI/ML) for millimeter wave (mmWave) communications, with a particular focus on vehicular scenarios. Although numerous related works on ML-based blockage prediction, these works focus on simple scenarios, where the trajectory of the moving object is either fixed or deterministic. In order to make the blockage prediction system more practical, this paper assumes a nondeterministic situation such as both communication terminals and blockers are moving. Furthermore, we have formulated a practical AI/ML-based blockage prediction procedure that is operable in 3GPP compliant systems with taking its beam measurement reports and frame structure. The proposed prediction method uses the measurement report for beam management as input of a neural network. The computer simulation is conducted to show prediction performance of proposed method.

This paper proposes an efficient spatial beam pre-diction method for millimeter-wave communication in multi-Transmission Reception Point (TRP) environments. We developed an AI-based framework leveraging beam measurements from multiple TRPs for beam management in 3GPP systems. We compare three architectural models: Single-TRP Input Single-TRP Output (STISTO), Multi-TRP Input Single-TRP Output (MTISTO), and Multi-TRP Input Multi-TRP Output (MTIMTO). The MTISTO approach achieves superior performance with an average 3.78% improvement in beam selection precision across all TRPs. Using ray-tracing simulations in realistic environments, our approach demonstrates significant beam prediction performance, corresponding to a 1.24 dB loss in received signal power compared to full beam search, while reducing measurement overhead by 93.75%. Additionally, MTISTO improves received signal power by an average of 0.37 dB (and up to 0.43 dB in challenging scenarios) compared to STISTO. This approach particularly excels in boundary regions between TRPs and in areas with building shadowing, demonstrating the feasibility of efficient spatial beam coordination for 5G-Advanced and beyond.

This paper proposes a low-complexity digital post-distortion (DPoD) technique using a frequency-domain attention-based neural network (FDA-NN) for compensating nonlinear distortions in power amplifiers supporting extremely wideband signals in the terahertz (THz) band. The proposed method efficiently captures the frequency-dependent memory effects of the power amplifier by applying a discrete Fourier transform (DFT) to the input signal and generating an attention signal that weights the frequency components according to their relevance for nonlinear distortion compensation. The performance of the proposed method is evaluated through experiments using a wideband power amplifier operating in the THz band and an orthogonal frequency-division multiplexing (OFDM) signal with a bandwidth of 4.8 GHz. The results demonstrate that the FDA-NN achieves comparable performance to more complex models, such as deep neural networks (DNN) and long short-term memory (LSTM) networks, while reducing the computational complexity by 48.9% and 78.7% in terms of floating-point operations (FLOPs), respectively. The proposed FDA-NN presents a promising solution for efficient nonlinear distortion compensation in future extremely wideband THz communication systems.

In this paper, we propose low-complexity digital predistortion (DPD) schemes based on FastGRNN to compensate for the nonlinearity of RF power amplifiers. Conventionally, high-precision recurrent neural network (RNN) models, such as long short-term memory (LSTM) and gated recurrent unit (GRU), have been used to model the behavior of amplifiers, and their excellent compensation performance has been shown in terms of error vector magnitude (EVM) and adjacent channel power ratio (ACPR) has been demonstrated. However, their complex structures result in high computational complexity. To solve this issue, the proposed method is designed to significantly reduce the complexity without significant performance degradation by appropriately applying the FastGRNN models to the DPD. Complexity analysis and experiments using a power amplifier in the 2.0 GHz frequency band showed that the proposed method achieved comparable EVM performance to LSTM with 29.2% floating point operations (FLOPs) and 27.1% trainable parameters.

In Japan, the 5th generation mobile communication system (5G) became commercially available in 2020. The millimeter wave bands such as 28GHz is being used to achieve the peak rate of 10 [Gbps] or higher targeted for 5G. In the late 2020s, low latency and high-capacity data transmission over both the up and down links will become important. This is because 5G will be utilized in the late 2020s, and telemedicine and teleoperation using 4K/8K and other high-definition video transmission will become widespread. In this study, we propose a relaying method that converts frequency multiplexing into spatial multiplexing during relaying, with the goal of achieving low latency and high capacity relaying communications. The user equipment, base stations, and relay stations have different conditions in terms of transmission power and number of antennas. Therefore, the proposed method achieves high capacity by frequency multiplexing in the link where the number of antennas is limited. In addition, the proposed method uses spatial multiplexing to achieve high capacity while suppressing the increase in resource usage in the link where multiple antennas are available. The 39 GHz band, which has more frequency resources than the existing 5G bands, is used for the evaluation in the link of frequency multiplexing. Then, the 28 GHz band, which is used commercially in 5G, is used for the evaluation in the link of spatial multiplexing. For low latency relaying, analog circuits are used during the relaying process to convert between frequency-multiplexed and space-multiplexed signals without modulation and demodulation, while maintaining the number of multiplexes. In this paper, simulation evaluations show that the proposed method improves the communication distance where the throughput exceeds 4 [Gbps] to 6.5 times that of 39 GHz band 5G communications without relaying, and to 1.3 times that of RF repeaters in conventional relaying methods that use the 39 GHz band both before and after relaying, indicating that the uplink communication distance can be extended.

Digital predistortion (DPD) using neural network(NN) has attracted attention as a promising technique for compensating complex nonlinear distortions caused by wideband radio frequency power amplifiers. However, NN-DPD is difficult to implement in hardware due to the large number of multiplications. In this paper, we propose an NN-DPD using DeepShift that replaces neural network multiplications with bitwise shift and sign flipping during both training and inference. First, when DeepShift was applied, we showed that applying residual learning to DPD can reduce performance degradation. Next, we examined pre-trained models using floating-point baseline models and scratch-trained models and found that the pre-trained models perform better in exchange for requiring pretraining. The scratch model, on the other hand, has the advantage of improving the training computational complexity. In an actual experiment of passing a signal with DPD applied through a power amplifier at 2.14 GHz, the error vector magnitude was slightly degraded from 1.31 % trained by the floating-point model to 1.57 % with the pre-trained model and 1.90 % with the scratch model.

The bit reliability based (BRB) and weighted bit reliability based (wBRB) algorithms are non-binary low-density parity-check (LDPC) code decoding algorithms with an excellent tradeoff between computational complexity and performance. However, the performance of these algorithms needs further improvement. We apply deep learning to these algorithms. Weights are assigned to each edge of the Tanner graphs of the non-binary LDPC codes in the proposed algorithms. We demonstrate the effectiveness of applying deep learning to the BRB and wBRB algorithms in terms of implementation and performance. The proposed algorithms achieve an approximately 0.3 dB higher bit error rate performance than the original algorithms in the high SNR region. The increase in computational complexity and memory consumption does not significantly change the implementation of the algorithms.