The rapid advancement in large foundation models is propelling the paradigm shifts across various industries. One significant change is that agents, instead of traditional machines or humans, will be the primary participants in the future production process, which consequently requires a novel AI-native communication system tailored for agent communications. Integrating the ability of large language models (LLMs) with task-oriented semantic communication is a potential approach. However, the output of existing LLM is human language, which is highly constrained and sub-optimal for agent-type communication. In this paper, we innovatively propose a task-oriented agent communication system. Specifically, we leverage the original LLM to learn a specialized machine language represented by token embeddings. Simultaneously, a multi-modal LLM is trained to comprehend the application task and to extract essential implicit information from multi-modal inputs, subsequently expressing it using machine language tokens. This representation is significantly more efficient for transmission over the air interface. Furthermore, to reduce transmission overhead, we introduce a joint token and channel coding (JTCC) scheme that compresses the token sequence by exploiting its sparsity while enhancing robustness against channel noise. Extensive experiments demonstrate that our approach reduces transmission overhead for downstream tasks while enhancing accuracy relative to the SOTA methods.

Existing works on machine learning (ML)-empowered wireless communication primarily focus on monolithic scenarios and single tasks. However, with the blooming growth of communication task classes coupled with various task requirements in future 6G systems, this working pattern is obviously unsustainable. Therefore, identifying a groundbreaking paradigm that enables a universal model to solve multiple tasks in the physical layer within diverse scenarios is crucial for future system evolution. This paper aims to fundamentally address the curse of ML model generalization across diverse scenarios and tasks by unleashing multi-modal feature integration capabilities in future systems. Given the universality of electromagnetic propagation theory, the communication process is determined by the scattering environment, which can be more comprehensively characterized by cross-modal perception, thus providing sufficient information for all communication tasks across varied environments. This fact motivates us to propose a transformative two-stage multi-modal pre-training and downstream task adaptation paradigm. In the pre-training stage, we introduce a multi-modal two-tower model and a corresponding contrastive learning method to integrate the explicit description of the scattering environment and implicit channel state information (CSI) into a universal representation, which encapsulates rich high-level knowledge and can be leveraged for all downstream tasks in different scenarios. Additionally, we present two specially designed model structures to enhance the interaction of communication modalities. In the second stage, based on the frozen pre-trained model, we propose a direct method and a pluggable method for flexible and low-cost task adaptation. Experimental results demonstrate that our proposed approach significantly outperforms benchmarks in both task performance and tuning parameter size for exemplary sub-tasks in unseen scenarios.

Recently, integrated sensing and communications (ISAC) design has attracted great attention for B5G/6G networks. Existing ISAC studies are mainly focused on monostatic sensing with a full-duplex radio. However, full-duplex radio requires complicated self-interference cancellations and many current devices are half-duplex radios. Therefore, it is still interesting to investigate passive bistatic sensing with half-duplex radios. In this paper, integrating passive bistatic sensing into mmWave B5G/6G networks is investigated. The public reference signals are leveraged to extract the frequency-domain channel state information (CSI) for passive sensing. To address the problems of sampling-timing-offset and random phase error, a line-of-sight (LoS) path aided calibration mechanism is first developed. To achieve accurate localization for multiple targets, a novel super-resolution channel impulse response (CIR) based mechanism is then developed. The super-resolution CIR is achieved from CSI through a joint design of spatial-smoothing multiple signal classification (MUSIC) algorithm and template-based tap estimation. With the super-resolution CIR, multiple targets can be first distinguished in CIR taps and then localized by further estimating their angle of arrivals (AoAs). The proposed design is implemented and validated on a prototype system at 28 GHz with 500 MHz bandwidth in indoor environments. Experimental results show that our design can achieve accurate single-target and multi-target localization and tracking. The localization error is 26 cm at 80thpercentile for single-person and 29 cm on average for multi-person setups. The average error to the planned path after the moving-averaged filter is 11 cm and 19 cm for single-person and two-person tracking, respectively.

In this paper, we create a scalable user activity detection (UAD) scheme for massive access. The proposed solution provides a holistic design from signature sequence design and its transmission method at device side to the reception method at base station side. We design a set of constant-modulus signature sequences for accurate, fast and scalable UAD. The designed sequences are the nonorthogonal complex sinusoid sequence. Different devices can be discriminated by assigning distinct digital frequency, which allows for unique association between user and signature sequence and collision avoidance. Taking advantage of the precoding strategies based on phase compensation or pre-equalization, the UAD problem can be casted as a linear inverse problem of nonnegative underdetermined system. At the heart of which, a sparse nonnegative vector can be solved by a nonnegative least square method, the support of which preserves the activity information exactly. In this way, base station can readily identify any possible combination of active users out of the entire set as long as the sequence length more than the number of actual active users, no matter how the total number of devices scales up.

A majority of network-level utility optimization problem can be formulated as a nonconvex Weighted Sum-Rate Maximization (WSRM) problem with respect to power optimization, which plays central role on the ultimate performance of a network. However, achieving global WSRM through distributed approach has been a long standing open problem in interference-coupled networks. This paper creates a distributed global power optimization algorithm for WSRM, which arguably represents the first effort to realize distributed WSRM by jointly exploiting the both virtues of monotonic optimization and nonlinear Perron-Frobenius theory. Benefiting from the nonlinear Perron-Frobenius theory, we derive a distributed projection algorithm at the heart of outer polyblock approximation method developed from monotonic optimization. Such a distributed projection allows each link (user) to generate a consensus about shrinking polyblock relying on nothing more than local measurements of SINR, which will greatly facilitates practical implementation. The shrinking polyblock approximates to the noncovex feasible region with respect to link rate, then we can arrive at the global optimum of WSRM by searching over the vertex set of the convergent polyblock instead of original region. Additionally and remarkably, the well-designed projection algorithm exhibits geometrical convergence rate, which will significantly expedite WSRM by reducing the computation complexity.

Virtual basestation (vBS) live migration is an enabling technology for reducing the power consumption of the baseband unit (BBU) pool of the cloud radio access network (Cloud RAN). Our previous work shows that using the existing virtual machine live migration solutions for vBS migration will introduce a service interruption time of several seconds, which is unacceptable for real-time services carried by basestations. In this paper we propose and develop a seamless and lossless mechanism for vBS live migration. The experiment results show that the proposed solution for vBS live migration introduces a service interruption time of tens of milliseconds and ensures no loss of user data during vBS migration, which significantly outperforms the commercial virtual machine live migration solutions.

The performance of orthogonal space-time block code (OSTBC) transmission in dual-hop amplify-and-forward (AF) multiple-input multiple-output (MIMO) relaying networks over spatially correlated Nakagami-m fading channel is investigated, where the source, relay, and destination terminals are all equipped with multiple antennas. We provide the compact closed-form expression for cumulative distribution function (CDF) of the end-to-end SNR and then exact analytical expressions are derived for the outage probability (OP), probability density function (PDF), and moment generating function (MGF) of the end-to-end SNR based on its CDF. Further, we present the asymptotic expression for CDF in the high SNR regime, which gives an insight of the system performance and the achievable diversity order. The analytical expressions are validated by Monte-Carlo simulations. It's shown that the correlation is detrimental to the MIMO relaying performance.

The joint design of the multi-stream in MIMO relay sigle/multiple user system is investigated. Under the inter-stream interference, a constrained optimization problem with respect to the unknown relay precoder matrix is formulated with a goal of minimizing the total relay transmit power subject to quality of service (QoS) requirement on each data stream. As the original opitimization problem is non-convex, the matrix expression of the cost function and constrained requirement is converting to a scalar optimization problem and the optimal relay precoder and receive filter are obtained. The optimal design of the multiple user relay system under QoS requirement is also provided. By using a reasonable approximation and converting to the generalized Rayleigh form for the cost function, a suboptimal relay precoder matrix for QoS optimization with relay transmit power constraint is obtained. Simulation results demonstrate that for single user model, the minimize relay power will increase with Signal to Interference plus Noise Ratio(SINR) constraint value becoming larger, and for multiple users model, more users in our algorithm will make the MSE gain more obvious as increasing transmitting SNR.

In this paper, the channel estimation related issues in FBMC systems are discussed. A single-symbol preamble structure is proposed and the corresponding channel estimation methods are provided for filter bank multicarrier (FBMC) systems. The simulation results show that compared with the existing methods, the proposed scheme achieves impressive performance gains (up to 4.0dB) with much smaller preamble resource overhead. This is obtained through effective channel filtering for the intrinsic imaginary interference and additive noise exploiting the sparse property of the channel response in delay domain. With the proposed solution, the FBMC system can even save half resources for pilots compared with OFDM systems.

Interference nulling (IN) is a linear precoding technique for downlink multiuser multiple-input multiple-output (MU-MIMO) systems. With perfect channel knowledge at the transmitter, IN eliminates other users' interference at each cell, thus improving the sum rate of the users in the system. To evaluate its suitability in a multicell setup, we have implemented the real-time system based on IN as well as on single-base-station closed-loop (SB-CL) precoding meeting the 3rd generation partnership project (3GPP) long term evolution (LTE)-Advanced system profile over a 20 MHz bandwidth. Our field trial in an indoor deployment of two operating base stations (BSs) and two mobile stations (MSs) shows that (1) if both MSs are located near the cell edge, IN outperforms SB-CL in terms of signal-to-interference-plus-noise-ratio (SINR) and sum throughput; (2) if two MSs are close to their respective serving BSs, IN is inferior to SB-CL. Thus, IN is more favorable for improving the transmission rate and quality of services for users near the cell-edge.

It is well known that coordinated multi-cell transmission in downlink (DL) is a promising technique to significantly increase the system throughput and cell-edge user throughput. In this paper, we propose a scalable two-stage channel feedback mechanism not only to support downlink single-cell multi-user MIMO (MU-MIMO), but also to efficiently support multi-cell coordination. This scheme yields remarkable system performance improvements with considerable uplink (UL) feedback overhead, and also has good backward compatibility for single-cell transmission scheme.

Multi-cell MIMO technique theoretically promises large improvement in spectral efficiency and fairness in cellular networks and is regarded as the key enabling technique for solving the troublesome issue of inter-cell interference. To evaluate its practical performance in the cellular network, we have implemented the real-time downlink multi-cell MIMO trial system operating over bandwidth of 20 MHz. The initial field trial indeed shows a significant spectral efficiency gain of downlink multi-cell MIMO over the traditional single-cell transmissions in an indoor deployment of two operating base stations and two terminals. Particularly, several principal operation modes of multi-cell MIMO are evaluated and compared in the field trial measurements, including coherent joint transmission (JT), non-coherent JT and interference nulling. The field trial results agree quite well with expectations from theory.

A multi-cell hybrid channel state information (CSI) feedback scheme was proposed, which has an ideal nested hierarchical structure to support not only the advanced multi-cell MIMO precoding, but also the traditional single-cell single-user and multi-user precoding transmissions. A quantization scheme for the multi-cell hybrid CSI feedback was provided, using codebook based vector-quantization for the per-cell CSI information part and scalar-quantization for the inter-cell CSI information part. The hierarchical structure of the multi-cell hybrid CSI feedback simplifies greatly the feedback signaling design, and implicitly supports the dynamic switch between different operation modes. Simulation results show that the proposed CSI feedback solution can enable efficient DL single-cell and multi-cell precoding transmissions.

MIMO-OFDM is widely considered as a promising framework of the future broadband mobile communication systems due to its prominent advantages such as high spectral efficiency, low-complexity equalization for multi-path fading, and sufficient exploiting of the spatial degree of freedom. However, an inherent drawback of OFDM, i.e., high sensitivity to frequency offset, limits the application scenarios of MIMO- OFDM significantly. This paper proposes an enhanced MIMO- OFDM scheme, termed double-polarized SFBC-OFDM (DP-SFBC-OFDM), which can improve the robustness to frequency offset with only very low complexity increase at the receiver. Simulation results show that even for very large frequency offset, the system only has limited performance degradation (e.g., normalized frequency offset of 0.4 only lead to performance loss of smaller than 1 dB for 64QAM for the proposed scheme while ordinary transmission scheme totally collapse in this case).

This paper proposes a hierarchical feedback technique to solve the problem of massive feedback overhead in multiuser (MU-) multiple-input-multiple-output (MIMO) transmission, which is a serious obstacle to the application of MU-MIMO in real systems, especially with a joint operation of multiuser scheduling and preceding. Motivated by the observations that the multiuser scheduling phase dominates the feedback overhead and it is less sensitive to the accuracy of channel state information (CSI) than the preceding phase, the proposed technique adopts a two-stage feedback method to reduce the feedback overhead in the scheduling phase, meanwhile guarantee good performance of the preceding phase. In the scheduling phase, long-term CSI, such as mean and covariance channel matrixes, are used to reduce the overhead. In the preceding phase, on the other hand, more accurate short-term CSI is still used, since preceding is more sensitive to the CSI accuracy and also has less feedback overhead as only users selected in the scheduling phase are required to feed back their CSI. An angular information feedback technique is introduced to provide the long-term CSI for the transmitter in the scheduling phase, along with a low-complexity scheduling technique based on angular information. Simulation results have shown that the proposed hierarchical feedback technique can significantly reduce the feedback overhead of MU-MIMO transmission, while still achieve most of the performance gain.

This paper proposes a two-dimensional switching diversity (2DSD) technique to enhance the collaborative spatial multiplexing (CSM) scheme for uplink wireless access. By the 2DSD, a single data stream is transmitted over multiple antennas and the mobile station (MS) alternates the transmit antennas from one time-frequency resource unit (RU) to another or from one group of RUs to another. That is, the RU-to-Antenna mapping pattern can change in both time and frequency domains. The new scheme has several advantages. First, by sharing one pilot pattern among multiple transmit antennas belonging to one MS, it is backward compatible with the legacy CSM specified in the WiMAX Forum mobile system profile Release 1.0 where two users each with one transmit antenna are scheduled in the same RUs to communicate with a common base station. Second, it saves the pilot overhead and allows as many MSs as possible to be involved in the CSM. Third, it can work in an open-loop manner and does not require additional signaling overhead. Finally, it achieves the power gain and diversity gain to improve the performance, as verified by computer simulations.

This paper addresses the joint design of multiuser precoding and scheduling for downlink multiuser (MU-) multiple-input-multiple-output (MIMO) systems with imperfect channel state information at the transmitter (CSIT). The major issue is to control the increased co-channel-interference (CCI) caused by CSIT error among users sharing the same time-frequency resource. We propose an advanced joint multiuser precoding and scheduling technique, which uses the statistical information of the CSIT error to predict the relationship among CCI, precoding matrixes and scheduling results, and utilizes such information to improve the precoding and scheduling algorithms for better CCI control. Simulation results have shown that considerable capacity improvement can be achieved by the proposed technique in the case of imperfect CSIT.

Imperfect channel state information (CSI) at the transmitter is one of the major challenges that the downlink multiuser multiple-input-multiple-output (MU-MIMO) technology faces in real applications. It will lead to increased co- channel interference (CCI) among users and deteriorate the capacity of MU-MIMO seriously, especially at high signal-to- noise-ratios (SNRs). In this paper, we propose an advanced MU-MIMO scheduling technique to improve the capacity of MU-MIMO in the case of imperfect CSI at the transmitter (CSIT). The rational behind the proposed scheme is the observation that the CCI level increases with the number of users simultaneously supported in MU-MIMO transmission. Thus, decreasing the number of simultaneous users can reduce CCI and sometimes even increase the sum capacity. Motivated by this observation, we design a CCI-feedback-aided scheduling technique, which adaptively adjusts the number of simultaneous users based on the CCI measurement at the user side. Simulation results have shown that considerable capacity gain can be achieved by the proposed scheduling technique at the cost of only one bit additional feedback per user and marginal complexity increase at both the transmitter and user side.

Exponential effective SNR mapping (EESM) based link error prediction is widely used for adaptive MIMO-OFDM in system evaluations to simplify simulation complexity. In MIMO system, EESM prediction method is more sophisticated due to specific MIMO detection techniques and spatial selectivity of multiple antennas. Especially for maximum likelihood (ML) detection it is hard to determine post-detected SNR due to non-linearity of ML detection, so penalty method based on MMSE detection has been proposed for EESM mapping. Simulations show EESM based effective SNR for ML receiver has obtained high fitting precision compared with AWGN performance in QPSK or 16QAM modulation, convolution coding with rate 1/2 in 2 times 2 MIMO-OFDM system. This method can be naturally generalized into other modulation and coding rates, higher level of MIMO modes, and other fading channels with frequency selectivity, time selectivity or spatial selectivity.

Orthogonal frequency division multiplexing (OFDM) is a popular method for high data rate wireless transmission. Its capability of turning a frequency-selective channel into a parallel collection of frequency flat subchannels has attracted huge interest from the research community as well as the industry. Multiple input and multiple output (MIMO) configuration is also a hot topic for the merit of an increased diversity gain and/or a higher system capacity. In this contribution, a novel timing synchronization method is proposed for the distributed MIMO-OFDM configuration. The performance of this proposed scheme is evaluated by simulations. The simulation results show that the proposed method offers an accurate estimation on the different time delay caused by the distributed transmitters of the MIMO-OFDM system

In practical OFDM systems, some virtual subcarriers are reserved for easing the requirements on the filter. In this case, the channel impulse response (CIR) estimated by DFT approaches appears a spectral leakage, which results in an error floor for the MSE performance. This paper proposes two enhanced DFT interpolation-based channel estimation methods, termed as EDFTI-CE-A and EDFTI-CE-B, based on the robust Wiener filtering, which can eliminate most of the leakage effect owning to the absence of pilot symbols in the virtual subcarriers and further suppress the effect of additive noise. The analysis and simulation results show that the proposed EDFTI-CE-A can almost eliminate the MSE floor of conventional DFTI-CE with small complexity increase and the proposed EDFTI-CE-B can achieve the same performance as linear MMSE channel estimator but with much lower computational complexity

Generalizing a conventional phase-shifted pilot (PSP) sequences and discrete Fourier transform (DFT) based channel estimator to MIMO-OFDM systems with virtual subcarriers, this paper presents two enhanced channel estimators applying Wiener interpolation and smoothing. Simulation results show that the proposed schemes can almost eliminate the mean square error floor of the conventional PSP/DFT-based channel estimator and improve the channel estimation accuracy further in some cases. Additionally, the application of a robust channel correlation matrix estimator avoids the complicated measurement of channel statistical properties and thus makes the proposed schemes more feasible in practical systems