In wireless systems aided by reconfigurable intelligent surfaces (RISs), channel state information plays a pivotal role in achieving the performance gain of RISs. Mobility renders accurate channel estimation (CE) more challenging due to the Doppler e
ffect. In this letter, we propose two practical wideband CE schemes incorporating Doppler shift adjustment (DSA) for multi-path and single-path propagation environments, respectively, for RIS-assisted communication with passive reflecting elements. For the multi-path scenario, ordinary CE is first executed assuming quasi-static channels, followed by DSA realized via joint RIS reflection pattern selection and transformations between frequency and time domains. The proposed CE necessitates only one more symbol incurring negligible extra overhead compared with the number of symbols required for the original CE. For the single-path case which is especially applicable to millimeter-wave and terahertz systems, a novel low-complexity CE method is devised capitalizing on the form of the array factors at the RIS. Simulation results demonstrate that the proposed algorithms yield high CE accuracy and achievable rate with low complexity, and outperform representative benchmark schemes.
Mobile devices are becoming an important carrier for deep learning tasks, as they are being equipped with powerful, high-end mobile CPUs and GPUs. However, it is still a challenging task to execute 3D Convolutional Neural Networks (CNNs) targeting fo
r real-time performance, besides high inference accuracy. The reason is more complex model structure and higher model dimensionality overwhelm the available computation/storage resources on mobile devices. A natural way may be turning to deep learning weight pruning techniques. However, the direct generalization of existing 2D CNN weight pruning methods to 3D CNNs is not ideal for fully exploiting mobile parallelism while achieving high inference accuracy. This paper proposes RT3D, a model compression and mobile acceleration framework for 3D CNNs, seamlessly integrating neural network weight pruning and compiler code generation techniques. We propose and investigate two structured sparsity schemes i.e., the vanilla structured sparsity and kernel group structured (KGS) sparsity that are mobile acceleration friendly. The vanilla sparsity removes whole kernel groups, while KGS sparsity is a more fine-grained structured sparsity that enjoys higher flexibility while exploiting full on-device parallelism. We propose a reweighted regularization pruning algorithm to achieve the proposed sparsity schemes. The inference time speedup due to sparsity is approaching the pruning rate of the whole model FLOPs (floating point operations). RT3D demonstrates up to 29.1$times$ speedup in end-to-end inference time comparing with current mobile frameworks supporting 3D CNNs, with moderate 1%-1.5% accuracy loss. The end-to-end inference time for 16 video frames could be within 150 ms, when executing representative C3D and R(2+1)D models on a cellphone. For the first time, real-time execution of 3D CNNs is achieved on off-the-shelf mobiles.
This paper compares three candidate large-scale propagation path loss models for use over the entire microwave and millimeter-wave (mmWave) radio spectrum: the alpha-beta-gamma (ABG) model, the close-in (CI) free space reference distance model, and t
he CI model with a frequency-weighted path loss exponent (CIF). Each of these models have been recently studied for use in standards bodies such as 3GPP, and for use in the design of fifth generation (5G) wireless systems in urban macrocell, urban microcell, and indoor office and shopping mall scenarios. Here we compare the accuracy and sensitivity of these models using measured data from 30 propagation measurement datasets from 2 GHz to 73 GHz over distances ranging from 4 m to 1238 m. A series of sensitivity analyses of the three models show that the physically-based two-parameter CI model and three-parameter CIF model offer computational simplicity, have very similar goodness of fit (i.e., the shadow fading standard deviation), exhibit more stable model parameter behavior across frequencies and distances, and yield smaller prediction error in sensitivity testing across distances and frequencies, when compared to the four-parameter ABG model. Results show the CI model with a 1 m close-in reference distance is suitable for outdoor environments, while the CIF model is more appropriate for indoor modeling. The CI and CIF models are easily implemented in existing 3GPP models by making a very subtle modification -- by replacing a floating non-physically based constant with a frequency-dependent constant that represents free space path loss in the first meter of propagation.
