Traffic interception in fiber optical video-systems

In fiber-optic video systems, as well as in optical communication systems, standard single mode optical fibers (SSMF, standard G.652) are usually used. One of the advantages of these fibers is the ability to use CWDM in a wide spectrum. At the same time, more optimal near the wave-length of 1550 nm are provided by non-zero dispersion fiber (NZDSF, standard G.655) fibers. However, as studies have shown, these optical fibers have an increased sensitivity to bending. This fact can be used to traffic interception. It is shown that fiber-optics systems with SSMF have more protection from traffic interception than systems with NZDSF. To transmit a high-confidentiality video signal, special techniques, such as frequency modulation, can be used, or additional noise signals can be added.

Harnessing Optoelectronic Noises in a Hybrid Photonic Generative Adversarial Network (GAN)

Integrated programmable optoelectronics is emerging as a promising platform of neural network accelerator, which affords efficient in-memory computing and high bandwidth interconnectivity. The analog nature of optical computing and the inherent optoelectronic noises, however, make the systems error-prone in practical implementations such as classification by discriminative neural networks. It is thus imperative to devise strategies to mitigate and, if possible, harness optical and electrical noises in optoelectronic computing systems. Here, we demonstrate a prototypical hybrid photonic generative adversarial network (GAN) that generates handwritten numbers using an optoelectronic core consisting of an array of programable phase-change optical memory cells. We harness optoelectronic noises in the hybrid photonic GAN by realizing an optoelectronic random number generator derived from the amplified spontaneous emission noise, applying noise-aware training by injecting additional noise to the network, and implementing the trained network with resilience to hardware non-idealities. Surprisingly, the hybrid photonic GAN with hardware noises and inaccuracies can generate images of even higher quality than the noiseless software baseline. Our results suggest the resilience and potential of more complex hybrid generative networks based on large-scale, non-ideal optoelectronic hardware. The demonstrated GAN architecture and the proposed noise-aware training approach are generic and thus applicable to various types of optoelectronic neuromorphic computing hardware.

On the tuning of plateaus in atmospheric and oceanic ¹⁴C records to derive calendar chronologies of deep-sea cores and records of ¹⁴C marine reservoir age changes

We assess the methodology of the so-called 14C plateau tuning (PT) technique used to date marine sediment records and determine 14C marine reservoir ages (MRAs) as recently reviewed by Sarnthein et al. (2020). The main identified problems are linked to the assumption of constant MRA during 14C age plateaus; the lack of consideration of foraminifera abundance changes coupled to bioturbation that can create spurious plateaus in marine sediments; the assumption that plateaus have the same shapes and durations in atmospheric and oceanic records; the implication that atmospheric 14C / 12C peaked instantaneously from one plateau to the next; that the 14C plateaus represent 82 % of the total time spent between 14 000 and 29 000 cal yr BP, whereas during the remaining 18 % of the time, the radiocarbon clock was running almost 5 times faster than the radioactive decay; that the sparsity, combined with the level of analytical uncertainties and additional noise, in both atmospheric and marine data do not currently allow one to reliably or robustly identify plateaus (should they exist) beyond 15 000 cal yr BP; and that the determination and identification of plateaus in the deep-sea cores is reliant upon significant changes in sedimentation rate within those marine sediments which are, a priori, unknown and are not verified with an independent method. The concerns we raise are supported and strengthened with carbon cycle box model experiments and statistical simulations of pseudo-atmospheric and pseudo-marine records, allowing us to question the ability to identify and tune 14C age plateaus in the context of noisy and sparse data.

Additional noise reduction with diffracting elements on barriers using numerical and standard calculation methods

The noise reduction of a (low) barrier can be enhanced by using an additional element with quarter-wavelength resonators with varying depths. A so-called Whiswall or WHIStop deflects sound upwards for specific frequencies. Measurements for a 1.1 meter high Whiswall and for a 1.1m barrier are compared in a separate paper. The enhanced barrier effect is measured at a short distance behind the barrier, for several situations. In this paper these measurements are compared with the results of a numerical finite element model (FEM) to validate this model. Next, the noise reduction is calculated at long ranges, up to 600 meters, for different point-to-point scenarios representative for road and rail traffic. A numerical parabolic equation method (PE) is coupled to the FEM model and a representative downwind condition is taken into account. The results at longer distance are used to design an engineering method for the enhanced barrier effect that can be used in standard noise calculation models, such as the Dutch national calculation model (SRM2) or the ISO 9613-2 standard.

On the Intrinsic Differential Privacy of Bagging

Differentially private machine learning trains models while protecting privacy of the sensitive training data. The key to obtain differentially private models is to introduce noise/randomness to the training process. In particular, existing differentially private machine learning methods add noise to the training data, the gradients, the loss function, and/or the model itself. Bagging, a popular ensemble learning framework, randomly creates some subsamples of the training data, trains a base model for each subsample using a base learner, and takes majority vote among the base models when making predictions. Bagging has intrinsic randomness in the training process as it randomly creates subsamples. Our major theoretical results show that such intrinsic randomness already makes Bagging differentially private without the needs of additional noise. Moreover, we prove that if no assumptions about the base learner are made, our derived privacy guarantees are tight. We empirically evaluate Bagging on MNIST and CIFAR10. Our experimental results demonstrate that Bagging achieves significantly higher accuracies than state-of-the-art differentially private machine learning methods with the same privacy budgets.

DCNet: Noise-Robust Convolutional Neural Networks for Degradation Classification on Ancient Documents

Analysis of degraded ancient documents is challenging due to the severity and combination of degradation present in a single image. Ancient documents also suffer from additional noise during the digitalization process, particularly when digitalization is done using low-specification devices and/or under poor illumination conditions. The noises over the degraded ancient documents certainly cause a troublesome document analysis. In this paper, we propose a new noise-robust convolutional neural network (CNN) architecture for degradation classification of noisy ancient documents, which is called a degradation classification network (DCNet). DCNet was constructed based on the ResNet101, MobileNetV2, and ShuffleNet architectures. Furthermore, we propose a new self-transition layer following DCNet. We trained the DCNet using (1) noise-free document images and (2) heavy-noise (zero mean Gaussian noise (ZMGN) and speckle) document images. Then, we tested the resulted models with document images containing different levels of ZMGN and speckle noise. We compared our results to three CNN benchmarking architectures, namely MobileNet, ShuffleNet, and ResNet101. In general, the proposed architecture performed better than MobileNet, ShuffleNet, ResNet101, and conventional machine learning (support vector machine and random forest), particularly for documents with heavy noise.

Impacts of additional noise on the social interactions of a cooperatively breeding fish

Anthropogenic noise is a global pollutant known to affect the behaviour of individual animals in all taxa studied. However, there has been relatively little experimental testing of the effects of additional noise on social interactions between conspecifics, despite these forming a crucial aspect of daily life for most species. Here, we use established paradigms to investigate how white-noise playback affects both group defensive actions against an intruder and associated within-group behaviours in a model fish species, the cooperatively breeding cichlid Neolamprologus pulcher . Additional noise did not alter defensive behaviour, but did result in changes to within-group behaviour. Both dominant and subordinate females, but not the dominant male, exhibited less affiliation and showed a tendency to produce more submissive displays to groupmates when there was additional noise compared with control conditions. Thus, our experimental results indicate the potential for anthropogenic noise to affect social interactions between conspecifics and emphasize the possibility of intraspecific variation in the impacts of this global pollutant.

Design and Optimization of Sensitivity Enhancement Package for MEMS-Based Thermal Acoustic Particle Velocity Sensor

In this paper, small-sized acoustic horns, the sensitivity enhancement package for the MEMS-based thermal acoustic particle velocity sensor, have been designed and optimized. Four kinds of acoustic horns, including tube horn, double cone horn, double paradox horn, and exponential horn, were analyzed through numerical calculation. Considering both the amplification factor and effective length of amplification zone, a small-sized double cone horn with middle tube is designed and further optimized. A three-wire thermal acoustic particle velocity sensor was fabricated and packaged in the 3D printed double cone tube (DCT) horn. Experiment results show that an amplification factor of 6.63 at 600 Hz and 6.93 at 1 kHz was achieved. A good 8-shape directivity pattern was also obtained for the optimized DCT horn with the lateral inhibition ratio of 50.3 dB. No additional noise was introduced, demonstrating the DCT horn’s potential in improving the sensitivity of acoustic particle velocity sensors.

Electromagnetic reducers in electromechanical systems

Currently, due to the rising cost of electricity, low-power wind turbines (1-5 kW) are often used to supply consumers with electricity. In this case, wind turbines are used with both horizontal and vertical axes of rotation, the speed of which at an average wind speed V = 5 ÷ 10 m / s and is quite low, and is approximately n = 100 - 300 rpm. A low-speed electric generator for a wind generator with such a speed of rotation with a direct connection of the wind rotor shaft and the electric generator has a large number of poles and reaches a fairly large size. Therefore, magnifying gears (multiplexers) are often used and can increase the speed of the electric generator several times and, thus, reduce the mass of its active part, because the electromagnetic moment is proportional to the volume of the electric machine. However, manual transmissions are a source of additional noise, require frequent maintenance and reduce the durability of the wind turbine. This article will use permanent magnet reducers for wind turbines, which, unlike mechanical reducers, do not create additional noise, do not require lubrication, their durability is higher, operating costs are also significantly reduced, while the magnetic reducer can be integrated with an electric generator. at a wind rotor power P = 4 kW and speed n = 100-300 rpm, high-speed electric generator and magnetic reducer have approximately 2 times less total weight of magnets and 1.7 times less total weight of active materials (magnetic reducer + electric generator) than a low-speed multipole external generator. The aim of the study is to develop and implement an electromagnetic reducer in electromechanical systems. The basis of such systems are high-coercive magnets. To achieve this goal, the following tasks are set: - literary-patent search on the research topic; - selection of a prototype of a magnetic reducer and calculation of its main parameters; - development of graphical and numerical models to evaluate the effectiveness of the developed prototype; - optimization of the design of the magnetic reducer; - development of a system for converting mechanical energy with low potential into electricity; - prototyping and experimental studies of the system of conversion of mechanical energy with low potential into electrical energy

Error Estimation of Linear Polarization Data from Coronagraphs – Application to STEREO-A/SECCHI-COR1 Observations

We attempt to quantitatively study the uncertainties of the polarized brightness and for the polarization angle which are to be expected for measurements from noisy image detectors in classical coronagraphs. We derive the probability density functions (PDF) which apply to polarization observations with polarization filters at 0∘, 60∘ and 120∘. The noise in the directly observed image intensities is assumed to be normally distributed. We find that for low and medium signal-to-noise ratios the polarized brightness obeys a distribution with a strongly biased mean which, if not taken account of, leads to an overestimation of the polarized brightness and degree of polarization. The PDFs are compared with data from the SECCHI-COR1 coronagraph onboard STEREO-A in order to detect systematic or random perturbations of the polarized brightness and the polarization angle beyond the unavoidable photon and detector hardware noise. This noise is estimated from two successive filter sequences taken in-flight during calm coronal conditions on 18 May 2008 and is expressed in the form of an intensity–variance relation. Two small deviations between the measured distributions and the predicted PDF for the polarization angle were found. The standard deviation of the polarization angle error decreases with increasing signal-to-noise ratio of the polarized brightness. For ratios larger than about 8 this decrease was found not as steep anymore as predicted which could hint to a small additional noise source. Next, we found a systematic constant deviation of the polarization angle by $-1^{\circ }$ − 1 ∘ for all signal-to-noise ratios of the polarized brightness. Besides these small discrepancies, our theoretically derived PDFs agree quite well with the distributions of measured brightnesses in test regions of the images. The PDFs we present here can equally be applied to similarly measured data from other coronagraphs and may help to quantify uncertainty limits of the derived polarization. They can be used for in-flight health checks of an instrument, are useful when separating unpolarized stray light from the polarized K-corona and when comparing the observed polarization data with results from model simulations.