Automatic modulation classification (AMC) algorithms based on deep learning (DL) have been widely studied in the past decade, showing significant performance advantage compared to traditional ones. However, the existing DL methods generally behave worse in computational complexity. For this, this paper proposes a lightweight convolutional neural network (CNN) for AMC task, where we design a depthwise separable convolution (DSC) residual architecture for feature extraction to prevent the vanishing gradient problem and lighten the computational burden. Besides that, in order to further reduce model complexity, global depthwise convolution (GDWConv) is adopted for feature reconstruction after the last (non-global) convolutional layer. Compared to recent works, the experimental results show that the proposed network can save approximately 70~98% model parameters and 30~99% inference time on two well-known benchmarks.

We propose a novel specific emitter identification (SEI) algorithm based on amplitude probability distribution (APD). By studying the subtle differences in nonlinear distortion of power amplifiers, the proposed algorithm leverages the real signal amplitude interval probability distribution (AIPD) and analytical signal’s constellation circle interval probability distribution (CIPD) to extract the local distortion features. The simulation results on both the simulated signal and actual signal show that compared to the benchmark methods, the proposed algorithm can achieve significant performance gains with relatively low complexity and has high robustness to training set size and prior information of signal to noise ratio (SNR).

Based on the Laurent decomposition (LD) and the Rimoldi decomposition (RD) with the reduced-search technique, we propose a low-complexity continuous phase modulation (CPM) detection algorithm through the operation of the g -th power. In this way, the denominator of the modulation index is effectively equivalent to a smaller value, achieving the usage of much fewer trellis states in the Viterbi algorithm (VA). Finally, simulation results show that, compared to the optimal detection method, the proposed algorithms significantly reduce the complexity with negligible bit error rate (BER) loss.

This letter concerns low complexity receiver design for Wiener phase-noise channels, where the variance of the additive white Gaussian noise (AWGN) is unknown. By representing the system with a factor graph, both the variational message passing and the sum-product message passing are used to achieve efficient joint decoding, phase noise estimation, and AWGN precision (the reciprocal of AWGN variance) estimation. Compared with the state-of-the-art receiver, the proposed receiver is able to achieve the same performance with much lower complexity.

This paper considers a multi-user multiple-input multiple-output (MU-MIMO) system with independent phase noise (PN) at user terminals (UTs) due to non-synchronous noisy local oscillators. The conventional factor graph of MU-MIMO involves an observation factor with multiple high-dimensional variables in the form of multiplication and summation. In this work, the MU-MIMO factor graph is represented by introducing auxiliary variables, which enables the decomposition of the observation factor into several simpler sub-factors and then use different message passing techniques to obtain approximate marginals. Specifically, belief propagation and expectation propagation (BP-EP) are used for the linear factors for random walk process, modulation and coding, while mean field (MF) is used for non-linear factor in exponential form. To reduce the computational complexity, the non-Gaussian MF messages are approximated to be Gaussian by using the second order Taylor expansion at its belief obtained in the previous iteration. The proposed receiver has a quadratic computational complexity per symbol per iteration. Simulation results show that the performance of the proposed receiver is better than existing methods and is close to the matched filter bound (MFB) in terms of the bit error rate (BER).

Considering orthogonal frequency division multiplexing (OFDM) system, we propose a joint channel estimation, noise precision estimation, symbol detection and bit decoding algorithm. In this algorithm, we incorporate the recently proposed vector approximate message passing (VAMP) with the belief propagation and mean field (BP-MF) framework, where BP is applied within subgraph of demodulation and decoding, and MF is used to handle nonlinear observations for estimating unknown noise precision, and VAMP is employed to estimate channels in densely connected factor graph. Furthermore, we use expectation propagation (EP) to demonstrate that the proposed approach converges after only one iteration. Compared to the state-of-the-art algorithm which exploits generalized approximate message passing (GAMP), the proposed scheme achieves the same bit error ratio (BER) performance, but lower per-iteration complexity and faster convergence speed.