Rate splitting multiple access(RSMA)has shown great potentials for the next generation communication systems.In this work,we consider a two-user system in hybrid satellite terrestrial network(HSTN)where one of them is...Rate splitting multiple access(RSMA)has shown great potentials for the next generation communication systems.In this work,we consider a two-user system in hybrid satellite terrestrial network(HSTN)where one of them is heavily shadowed and the other uses cooperative RSMA to improve the transmission quality.The non-convex weighted sum rate(WSR)problem formulated based on this model is usually optimized by computational burdened weighted minimum mean square error(WMMSE)algorithm.We propose to apply deep unfolding to solve the optimization problem,which maps WMMSE iterations into a layer-wise network and could achieve better performance within limited iterations.We also incorporate momentum accelerated projection gradient descent(PGD)algorithm to circumvent the complicated operations in WMMSE that are not amenable for unfolding and mapping.The momentum and step size in deep unfolding network are selected as trainable parameters for training.As shown in the simulation results,deep unfolding scheme has WSR and convergence speed advantages over original WMMSE algorithm.展开更多
Existing deep unfolding methods unroll an optimization algorithm with a fixed number of steps,and utilize convolutional neural networks(CNNs)to learn data-driven priors.However,their performance is limited for two mai...Existing deep unfolding methods unroll an optimization algorithm with a fixed number of steps,and utilize convolutional neural networks(CNNs)to learn data-driven priors.However,their performance is limited for two main reasons.Firstly,priors learned in deep feature space need to be converted to the image space at each iteration step,which limits the depth of CNNs and prevents CNNs from exploiting contextual information.Secondly,existing methods only learn deep priors at the single full-resolution scale,so ignore the benefits of multi-scale context in dealing with high level noise.To address these issues,we explicitly consider the image denoising process in the deep feature space and propose the deep unfolding multi-scale regularizer network(DUMRN)for image denoising.The core of DUMRN is the feature-based denoising module(FDM)that directly removes noise in the deep feature space.In each FDM,we construct a multi-scale regularizer block to learn deep prior information from multi-resolution features.We build the DUMRN by stacking a sequence of FDMs and train it in an end-to-end manner.Experimental results on synthetic and real-world benchmarks demonstrate that DUMRN performs favorably compared to state-of-theart methods.展开更多
基金sponsored by National Natural Science Foundation of China (No. 61871422, No.62027801)
文摘Rate splitting multiple access(RSMA)has shown great potentials for the next generation communication systems.In this work,we consider a two-user system in hybrid satellite terrestrial network(HSTN)where one of them is heavily shadowed and the other uses cooperative RSMA to improve the transmission quality.The non-convex weighted sum rate(WSR)problem formulated based on this model is usually optimized by computational burdened weighted minimum mean square error(WMMSE)algorithm.We propose to apply deep unfolding to solve the optimization problem,which maps WMMSE iterations into a layer-wise network and could achieve better performance within limited iterations.We also incorporate momentum accelerated projection gradient descent(PGD)algorithm to circumvent the complicated operations in WMMSE that are not amenable for unfolding and mapping.The momentum and step size in deep unfolding network are selected as trainable parameters for training.As shown in the simulation results,deep unfolding scheme has WSR and convergence speed advantages over original WMMSE algorithm.
基金partially supported by the National Key R&D Program of China(No.2020YFA0714101)the National Nature Science Foundation of China(Nos.61872162,62102414,62172415,and 52175493).
文摘Existing deep unfolding methods unroll an optimization algorithm with a fixed number of steps,and utilize convolutional neural networks(CNNs)to learn data-driven priors.However,their performance is limited for two main reasons.Firstly,priors learned in deep feature space need to be converted to the image space at each iteration step,which limits the depth of CNNs and prevents CNNs from exploiting contextual information.Secondly,existing methods only learn deep priors at the single full-resolution scale,so ignore the benefits of multi-scale context in dealing with high level noise.To address these issues,we explicitly consider the image denoising process in the deep feature space and propose the deep unfolding multi-scale regularizer network(DUMRN)for image denoising.The core of DUMRN is the feature-based denoising module(FDM)that directly removes noise in the deep feature space.In each FDM,we construct a multi-scale regularizer block to learn deep prior information from multi-resolution features.We build the DUMRN by stacking a sequence of FDMs and train it in an end-to-end manner.Experimental results on synthetic and real-world benchmarks demonstrate that DUMRN performs favorably compared to state-of-theart methods.