Software implementation of LDPC decoders has been an active area of development for the last 10 years. Researchers have focused on implementing the computationally expensive algorithm on both GPPs ...and GPUs. A major leap in performance was reported in the groundbreaking paper by Bertrand le Gal 2. This paper builds on the work in 2 by considering the scaling of that implementation on modern many-core processors. We look at the performance of LDPC code specified in the DVB-S2 standard. The large block size of the DVB-S2 code makes the memory architecture of the processor just as important as the clock rate and instruction set. We present results for two generations of Intel Xeons, an Intel Phi (KNL), the recently released AMD EPYC. The key finding is that performance scaling is limited by the amount of available cache memory rather than the number of cores. We also find that heavily multi-threaded, but deterministic software architecture benefits from explicit allocation of threads to cores vs. allowing the operating system to manage threading. The maximum throughput of 1 Gbps was achieved on a mid-range AMD server - issuing a new era of all-software receivers for very high rate waveforms. We also present the performance of the algorithm ported to a low-power ARM processor and compare that to a low-end Intel Core.
Southern Mongolia is part of the Central Asian Orogenic Belt, the origin and evolution of which is not fully known and is often debated. It is composed of several east–west trending ...lithostratigraphic domains that are attributed to an assemblage of accreted terranes or tectonic zones. This is in contrast to Central Mongolia, which is dominated by a cratonic block in the Hangai region. Terranes are typically bounded by suture zones that are expected to be deep-reaching, but may be difficult to identify based on observable surface fault traces alone. Thus, attempts to match lithostratigraphic domains to surface faulting have revealed some disagreements in the positions of suspected terranes. Furthermore, the subsurface structure of this region remains relatively unknown. Therefore, high-resolution geophysical data are required to determine the locations of terrane boundaries. Magnetotelluric data and telluric-only data were acquired across Southern Mongolia on a profile along a longitude of approximately 100.5° E. The profile extends ~ 350 km from the Hangai Mountains, across the Gobi–Altai Mountains, to the China–Mongolia border. The data were used to generate an electrical resistivity model of the crust and upper mantle, presented here, that can contribute to the understanding of the structure of this region, and of the evolution of the Central Asian Orogenic Belt. The resistivity model shows a generally resistive upper crust (0–20 km) with several anomalously conductive features that are believed to indicate suture zones and the boundaries of tectonic zones. Moreover, their spatial distribution is coincident with known surface fault segments and active seismicity. The lower crust (30–45 km) becomes generally less resistive, but contains an anomalously conductive feature below the Gobi–Altai zone. This potentially agrees with studies that have argued for an allochthonous lower crust below this region that has been relaminated and metamorphosed. Furthermore, there is a large contrast in the electrical properties between identified tectonic zones, due to their unique tectonic histories. Although penetration to greater depths is limited, the magnetotelluric data indicate a thick lithosphere below Southern Mongolia, in contrast to the previously reported thin lithosphere below Central Mongolia.
The Hangai Dome, Mongolia, is an unusual high-elevation, intra-continental plateau characterized by dispersed, low-volume, intraplate volcanism. Its subsurface structure and its origin remains ...unexplained, due in part to a lack of high-resolution geophysical data. Magnetotelluric data along a ∼610 km profile crossing the Hangai Dome were used to generate electrical resistivity models of the crust and upper mantle. The crust is found to be unexpectedly heterogeneous. The upper crust is highly resistive but contains several features interpreted as ancient fluid pathways and fault zones, including the South Hangai fault system and ophiolite belt that is revealed to be a major crustal boundary. South of the Hangai Dome a clear transition in crustal properties is observed which reflects the rheological differences across accreted terranes. The lower crust contains discrete zones of low-resistivity material that indicate the presence of fluids and a weakened lower crust. The upper mantle contains a large low-resistivity zone that is consistent with the presence of partial melt within an asthenospheric upwelling, believed to be driving intraplate volcanism and supporting uplift.
•A ∼610 km long electrical resistivity model across the Hangai Dome is presented.•The observed South Hangai fault system shows a clear transition across terranes.•Lower-crustal low-resistivity indicates the presence of fluid and a weakened layer.•A large, low-resistivity zone detected in the upper mantle requires partial melt.•Uplift and volcanism attributed to partial melting in an asthenospheric upwelling.
Geomagnetic storms, which are governed by the plasma magnetohydrodynamics of the solar‐interplanetary‐magnetosphere system, entail a formidable challenge for physical forward modeling. Yet, the ...abundance of high‐quality observational data has been amenable to the application of data‐hungry neural networks to geomagnetic storm forecasting. Almost all applications of neural networks to storm forecasting have utilized solar wind observations from the Earth‐Sun first Lagrangian point (L1) or closer and generated deterministic output without uncertainty estimates. Furthermore, forecasting work has focused on indices that are also sensitive to induced internal magnetic fields, complicating the forecasting problem with another layer of non‐linearity. We address these points, presenting neural networks trained on observations from both the solar disk and the L1 point. Our architecture generates reliable probabilistic forecasts over Est, the external component of the disturbance storm time index, showing that neural networks can gauge confidence in their output.
Plain Language Summary
Geomagnetic storms are capable of damaging infrastructures like power grids and communication lines, motivating our need to forecast them. Solar phenomena produce geomagnetic storms, which occur when these phenomena reach Earth as bursts of the solar wind. Decades of satellite observations of both the solar wind near the Earth and of the Sun itself are promising for forecasting geomagnetic storms with algorithms known as neural networks. Several neural network architectures have been applied to geomagnetic storm forecasting, but their full potential remains unexplored. First, all existing neural networks have used measurements of the solar wind one hour upstream of the Earth or closer. While these observations are critical for understanding geomagnetic storm progression, from them it is nearly impossible to forecast more than an hour in advance. We include observations of the Sun itself, which reach Earth much faster than the solar wind, thereby including information for forecasting further in advance. Second, all existing neural networks have generated forecasts without uncertainty estimates, meaning that end‐users (such as utilities or telecommunications companies) know little about forecast confidence. We present an architecture that generates estimates of uncertainty, and our results demonstrate that neural networks learn how confident to be in their forecasts.
Key points
We present a neural network architecture that utilizes both observations from the L1 point and solar disk, improving forecast reliability
Our neural network architecture learns reliable estimates of uncertainty in multiple‐hour ahead forecasts
Instead of the conventional disturbance storm time (Dst) index we forecast the external component of geomagnetic storms, Est
Full-duplex communications has become a hot area of research. A full-duplex can potentially double the amount of available spectrum, has a positive impact on the design of the protocol stack, and ...simplifies the design of acquisition and tracking. Most, if not all, of the recent research in this area has been focused on implementing FD for commercial terrestrial networks such as WiFi and cellular. The links in these networks are typically limited by multi-user interference and fading. To the best of the author's knowledge, very little work has been done in applying FD to noise-limited systems. A noise limited system, for the purpose of this paper, is one where additive white noise (AWGN) is the primary channel distortion. Typical examples include satellite communications (both ground to satellite and satellite to satellite) and point-to-point microwave relay links 13. Data throughput in these systems is limited by the classical Shannon capacity - fixed available spectrum and fixed available transmit power. Both of these limitations can be due to either technical or regulatory considerations. These systems are usually 'simpler' than typical multi-user networks such as WiFi and cellular. The point-to-point communications and lack of mobility motivate the use of highly directional, high gain antennas. These antennas can be either fixed (e.g. dish) or electronically steered beams (e.g. phased array). In either case, MIMO techniques are usually not applicable since the channel is trivially static (i.e. line of sight). Powerful error correction codes allow operation close to the Shannon limit. The receiver noise figure is the single most important factor in determining the link throughput. This paper presents a detailed implementation of a full-duplex receiver front end for a noise-limited system. The design trade-offs are driven by the desire to minimize the overall receiver noise figure.
A convolutional interleaver is a data reordering operation used to distribute bursts of errors and improve the performance of forward error correction algorithms in the presence of dropout events. ...This type of interleaver is used in several commercial and military standards as it offers the same performance as a traditional rectangular interleaver but with half the memory requirement. This paper presents a very novel algorithm and architecture for convolutional interleaving implemented on modern general-purpose processors (CPUs). The naïve implementation of a convolutional interleaver does not map well to modern multi-processor CPUs due to its non-sequential memory access pattern and inherently serial processing sequence. Non-sequential memory access is very inefficient in general-purpose processors as the high-speed, low-latency cache memory in modern CPU architectures assume some level of data locality. The sparse memory access pattern of a naïve convolutional interleaver implementation has almost no data locality, causing the CPU to frequently stall as it waits on data from the high-latency external memory. In addition, these memory bottlenecks make attempts to multithread the software implementation pointless, as modern general-purpose processor cores share memory resources. The key breakthroughs described in the paper address both these challenges - the algorithm is modified to interleave buffers of bits in a 'cache friendly' order rather than the input order, leading to significantly higher single-core performance. Resolving the memory access bottleneck also enables an efficient parallelization approach that can take advantage of multiple cores with shared memory, pushing performance even higher. Whereas previous convolutional interleaver software implementations ran into memory bottlenecks around 0.5GSps, the multithreaded version of the presented algorithm can be consistently benchmarked at above 8 GSps on a mid-range server CPU.
The initial step is to determine if the interval of time for the Emergency Department registration (door) to the completion of an Electrocardiogram (EKG) differ for men and women. ...does the ...difference in the interval of time have a deleterious effect on the overall heart function?