The acquisition of accurate channel state information is critical for enhancing the transmission quality and energy efficiency of reconfigurable intelligent surface-aided millimeter wave systems with multiple-input multiple-output antenna arrays at the transceiver ends. Due to the cascaded doubly sparse channel property in angular and delay domains, it is difficult to acquire nonzero angular gains sampled by overcomplete dictionaries with low pilot overhead. In this work, a channel-estimation method based on the sparse Kalman filter (SKF) framework is proposed, in which a state-space model is established for a sparse complex angular gain vector in both static and dynamic scenarios, and a zero-padding block-diagonal pilot pattern is designed. A linear l ₁ -norm subdifferential pseudo-measurement equation is deduced. The on-grid angular gain vector, along with the corresponding support, is recovered by the sequential processing of multiple pilot tone observations. To enhance the l ₁ -norm convergence and thus reduce the pilot overhead, an accelerated algorithm is developed, which incorporates an exponential factor to accelerate the gradient descent procedure. The simulation results demonstrate that the method can capture the angular support using a few pilot blocks. The normalized mean square error performance for the angular gain vector was better than that for the orthogonal matching pursuit. The accelerated SKF algorithm achieves faster convergence and higher accuracy compared with the ordinary SKF algorithm, and it is easy to implement online.

In communication and sensing integrations, an intelligent reflective surface (IRS) will provide an auxiliary wireless link to enhance coverage and information transmission quality. For IRS‐aided millimetre wave multi‐input multi‐output systems, the instantaneous cascaded channel is highly dimensional, resulting in the difficulty of acquiring channel state information. This letter proposed a tensor‐based Kalman filtering method. Under a parallel factor decomposition condition, a distributed sub‐filter algorithm is established. This method benefits from low‐dimensional tensor modes and known IRS phase shifts. It can track the channel with a few pilot overheads.

To promote the performance of ComNet and OAMPNET for CP-free OFDM, improved ComNet based on expectation propagation (IComNet-EP) is proposed in this paper. The IComNet-EP receiver includes channel estimation network (CENET), expectation propagation equalization (EPE) block and demodulation module. The channel state information is estimated by a model-driven neural network, CE-NET. Due to inter-symbol interference (ISI), the EPE block removes the ISI by buffering the previous OFDM symbol first and then detects the signal by expectation propagation algorithm. The IComNet-EP has comparative complexity with the OAMP-NET and ComNet but has better robustness for the detection part. The simulation results show that the proposed scheme offers a much lower bit-error rate than competitive algorithms.

Unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) has emerged as a practical means of providing computation and communication support for geographically dispersed Internet of Things (IoT) terminals. However, the broadcast nature of wireless links makes offloading data vulnerable to cooperative eavesdropping. Moreover, the limited onboard energy of UAVs and the latency-sensitive characteristics of MEC services lead to a challenging trade-off between the secrecy rate, energy consumption, and latency. To address this issue, we formulate a utility maximization problem to jointly optimize UAV trajectory and task-offloading decisions in UAV-assisted MEC systems against multiple eavesdroppers. Due to the strong coupling among optimization variables and the non-convexity of the problem, an enhanced twin-delayed deep deterministic policy gradient (TD3) framework integrating Hindsight Experience Replay (HER) and Prioritized Experience Replay (PER) is proposed to improve convergence efficiency and learning stability. Furthermore, a system utility-driven reward function is designed to balance the secrecy rate, energy consumption, and processing latency under different application requirements. The simulation results demonstrate that the proposed approach consistently outperforms DDQN, DDPG, and conventional TD3 in terms of system utility, secrecy performance, convergence speed, and adaptability to different scenarios.