LightAMC: Lightweight Automatic Modulation Classification via Deep Learning and Compressive Sensing

Abstract

Automatic modulation classification (AMC) is an promising technology for non-cooperative communication systems in both military and civilian scenarios. Recently, deep learning (DL) based AMC methods have been proposed with outstanding performances. However, both high computing cost and large model sizes are the biggest hinders for deployment of the conventional DL based methods, particularly in the application of internet-of-things (IoT) networks and unmanned aerial vehicle (UAV)-aided systems. In this correspondence, a novel DL based lightweight AMC (LightAMC) method is proposed with smaller model sizes and faster computational speed. We first introduce a scaling factor for each neuron in convolutional neural network (CNN) and enforce scaling factors sparsity via compressive sensing. It can give an assist to screen out redundant neurons and then these neurons are pruned. Experimental results show that the proposed LightAMC method can effectively reduce model sizes and accelerate computation with the slight performance loss.

Reinforcement Learning Based Cooperative Coded Caching Under Dynamic Popularities in Ultra-Dense Networks

“…1 − ε solve a(t) = arg max aQ (x(t), a) based on (12) and the procedure 1), 2), 3) w.p. ε choose an action randomly that meets the constraints (MDS Coding): MBS codes the contents with a c (t) = 0 and a c (t) ≥ a c (t−1) based on MDS coding during offpeak hours 5: (Coded Packets Delivery): MBS sends p different packets, each containing (a c (t) − a c (t − 1))B bits of the cth coded content, to p SBSs correspondingly for caching during off-peak hours 6: (Cooperative Transmission): User g ∈ G s selects d SBSs to form S d g and requests contents during peak hours of time slot t + 1 (Estimation of Cooperative Caching Action): Estimate a caching action under the state x(t + 1) by solving a(t + 1) = arg max a Q x(t + 1), a based on (12) and the procedure 1), 2), 3) The agent determines the caching strategy based on RL with SBS cooperation while the fragments stored at the SBSs are uncoded. Specifically, each SBS randomly stores the fragments with the corresponding size according to the action made by the agent.…”

Data-Driven Deep Learning to Design Pilot and Channel Estimator for Massive MIMO

“…Recently, deep learning (DL) has shown its great potential to revolutionize communication systems by applying deep neural network (DNN) to various communication and signal processing problems [9]- [11], which include modulation recognition [12], [13], signal detection [14], CSI feedback [15], and channel estimation [16]- [18], network routing and traffic control [19]- [21], et al Specifically, in [15], a novel CSI sensing and recovery mechanism, called CsiNet, was developed to recover CSI with improved reconstruction quality and reduced feedback overhead, which was closely related to the autoencoder in DL. In [16], a DL-based channel estimation and direction-of-arrival (DOA) estimation solution was proposed for massive MIMO systems, where the DNN was exploited to learn the statistical characteristics of wireless channels and the spatial structure in the angle domain.…”

The Performance Evaluation of Big Data-Driven Modulation Classification in Complex Environment

“…At present, an algorithm based on DL method is used for signal classification, and each dataset under each signal-to-noise ratio is sent to a DL network to train a model, and a plurality of models are trained in a plurality of signal-to-noise ratios. The resulting classification results will be more convincing if the datasets at all signal-to-noise ratios are fed to the deep learning network at the same time [15]- [17].…”