Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Results from the HX-POL simulations. The panels show the Fig. 7. distributions of azimuthal scattering angles for events with a >5 keV Si detector hit and a >50 keV CZT detector hit (upper panel), and for photons with a >5 keV hit in at least two Si detectors (lower panel) for 100% polarized X-ray beams (solid lines) and an un-polarized beams (dashed lines). A source with the flux from the Crab Nebula was assumed and an ON-observation time of 3 hrs. The absolute numbers of counts correspond to 3 hour observations of an X-ray source with the flux and spectrum of that of the Crab Nebula. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Figures 7 and 8 show results for 100% polarized and un- polarized X-ray beams (no backgrounds). The absolute number of events correspond to a 3 hr ON-observation of a source with the flux and spectrum of that of the Crab Nebula [27] (the emission from the Crab Nebula is of course not polarized to 100%). The upper panel of Fig. 7 shows the distribution of azimuthal scattering angles for “Si-CZT” events where one Si detector records a hit (most likely from a Compton interaction) and one CZT detector pixel records a hit (most likely from a photo-effect interaction). The lower panel of Fig. 7 shows the same for “Si-Si” events with hits in two or more Si detectors. The analysis of Si-Si events discards events in which the two highest energy depositions are less than than 3.2 pixel pitches away from each other when projected onto the plane of the detectors. This cut leads to a rather even distribution of detected azimuthal scattering angles. The events that do not pass this cut also hold information and could be used in a more sophisticated analysis. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 8. The panel shows the energy distribution of photons producing Si-CZT events (solid line) and Si-Si events (dashed line). |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | where Rsrc is the total source counting rate, Rbg is the total background counting rate and T is the ON integration time (assumed to equal the OFF integration time). |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Table I lists the values Rsrc, Rbg, and MDP for the Si- CZT events, the Si-Si events, and for both types of events. We assume interspersed ON-OFF observations with a ON and OFF integration time of 3 hrs each. The background was adopted from the balloon results reported by Sch¨onfelder et al. [29]. The prediction of the sensitivity of the Si-CZT events should be robust. The sensitivity of the Si-Si events will be poorer than stated in Table I, as we did not attempt to model neutrons. Elastically scattering neutrons can mimic Compton- events. Figure 8 shows the distribution of the energy of the Si-CZT events and the Si-Si events. Most Si-CZT events have energies between 60 keV to 400 keV. Most Si-Si events have energies between 50 keV and 300 keV. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Fig. 9. The graph shows the results of a simulated observation of the Crab Nebula (3 hours ON, 3 hours OFF). The polarization degree can be measured in several independent energy bins. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | of 3 gr cm−2. The graph shows that HX-POL will be able to measure the polarization of the Crab Nebula in several independent energy bins. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | Additional observation targets for a one-day balloon-flight are black hole binaries and accreting neutron stars. The binary black holes Cygnus X-1 and GRS 1915+105 are prime targets because of their X-ray brightness. Cygnus X-1 is an exceptional source as it shows persistent flaring activity. In the HEAO-4 survey it was the brightest > 80 keV source [30]. The source GRS 1915+105 is well known for frequent flaring epochs with a stunning diversity of flux and spectral patterns [31]. Binary black holes exhibit different emission states (thermal, hard, and steep power law) that depend on the structure of the accretion disk and the accretion disk corona, and on the relative importance of a relativistic outflow (jet) [32]. Synchrotron emission and Compton scatterings could both lead to polarization degrees that are observable with HX-POL. Bright accreting neutron stars are Hercules X-1 and Cen X-3. Owing to the emission mechanism and to the propagation of the X-rays through the highly magnetized plasma surrounding the neutron stars, the X-rays from these objects might exhibit polarization degrees of several 10% (e.g. [33], [4]). |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | energy cut-off [34], [35]. This effect cannot be observed in terrestrial laboratories and would provide a powerful test of QED in extreme conditions. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | A longer balloon flight would have a realistic chance to measure the polarization of the hard X-rays from extragalac- tic sources, e.g. from “extreme synchrotron blazars”, mass- accreting supermassive black holes with highly relativistic jets pointing at the observer [36]. Their synchrotron emission extends into the very hard X-ray regime. For example, for Mrk 501, the BeppoSAX mission measured a spectral energy distribution with a “low-energy” component peaking in the >100 keV energy range [37]. The high-energy component, presumably inverse Compton emission, peaks in the GeV/TeV energy range [38]. Extreme synchrotron blazars show long lasting periods (3 months or more) of intense flaring activity. A positive detection of a high polarization degree would show that the magnetic fields in AGN jets are well ordered in the “blazar zone” ∼1 pc away from the black hole, and would add an important constraint to models of jet formation. |
Processed_HX-POL_-_A_Balloon-Borne_Hard_X-Ray_Polarimeter.txt | This paper describes the concept of a balloon-borne hard X-ray polarimeter. The detector module combines a low-Z Compton scatterer (Si) with a high-Z photoeffect absorber (CZT). The detector combination achieves excellent polariza- tion sensitivity in the 50 keV to 400 keV energy band. Based on 3+3 hrs of ON-OFF observations during a one-day balloon flight, the polarization of strong Crab-like sources can be measured for polarization degrees well below 10%. On a 15- 30 day balloon flight, polarization degrees of a few percent can be measured. The 50-400 keV energy band is well suited for a balloon-borne experiment: atmospheric absorption hinders observations below 20 keV and observations above 400 keV have to deal with very low photon fluxes and require relatively large experiments and/or long integration times, both better suited for space borne experiments. One of the strengths of the Si/CZT detector combination is the excellent energy resolution that makes it possible to measure the energy dependence of the polarization properties. Compared to experiments that use Si detectors only, the Si/CZT combination avoids the neutron backgrounds that plague experiments exclusively made of low- Z detectors. Compared to experiments that use CZT only, the Si/CZT combination achieves a lower energy threshold. In CZT detectors, photoeffect interactions dominate over Comp- ton interactions up to energies of ∼250 keV; furthermore, <100 keV photons are absorbed quickly in CZT, making it necessary to use CZT detectors with very small pixel pitches. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | We explore the possibilities of measuring the longitudinal profile of individual air showers beyond 𝑋max when using very dense radio arrays such as SKA. The low-frequency part of the Square Kilometre Array, to be built in Australia, features an enormous antenna density of about 50,000 antennas in the inner core region of radius 500 m, with a frequency band from 50 to 350 MHz. From CoREAS simulations, a SKA-Low antenna model plus noise contributions, and adapted LOFAR analysis scripts, we obtain a resolution in the shower maximum 𝑋max and energy that is considerably better than at LOFAR. Already from this setup, we show that at least one additional parameter of the longitudinal profile can be measured. This would improve mass composition analysis by measuring an additional composition-dependent quantity. Moreover, it would offer an opportunity to discriminate between the different predictions of hadronic interaction models, hence contributing to hadronic physics at energy levels beyond man-made accelerators. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | Figure 1: An example of a longitudinal particle distribution at 𝑋max = 650 g/cm2, showing the effect of varying 𝐿 (left) and 𝑅 (right). The variations of 𝐿 and 𝑅 have been exaggerated compared to typical natural values. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | The low-frequency part of the Square Kilometre Array [1] to be built in Australia will feature an unprecedented number and density of antennas. Whereas LOFAR [2] has about 300 usable antennas in a diameter of 320 m, SKA will have about 50, 000 in a radius of 500 m, and extends the frequency band to 350 MHz. Hence, it is expected to open up new possibilities in measuring air showers into fine detail. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | LOFAR has been successful in measuring 𝑋max to an accuracy suitable for cosmic-ray mass composition studies [3, 4]. The longitudinal evolution of the number of particles in an air shower, reaching a maximum at 𝑋max, has two extra parameters containing information about mass com- position; moreover, their expected value and distribution varies with hadronic interaction model [5]. Hence, being able to measure these would offer an opportunity to improve mass composition measurements, and to distinguish between hadronic interaction models at high energies. Here we report a first analysis to reconstruct longitudinal profile parameters. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | with 𝑁 denoting the number of particles, and 𝑋 the traversed depth in the atmosphere in g/cm2. In this distribution, 𝑋max is the maximum or mode, 𝐿 is related to the variance (second moment), and 𝑅 to the skewness (third moment). The effect of varying 𝐿 and 𝑅 is demonstrated in Fig. 1. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | The present analysis focuses on two parts: comparing radio footprints directly as a function of 𝐿 and 𝑅, and doing a more complete reconstruction of 𝑋max, 𝐿, and 𝑅 using simulated measurements with realistic sky noise. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | Our simulations have antennas put on a radial grid or ‘star shape’, with radial density higher in the first 100 m from the core, then declining outward. For a reconstruction from a dense layout of antennas, an interpolation is needed. We have developed an interpolation method that improves over the radial basis function method [10] used earlier, which does not perform as well on footprints of higher-frequency signals. It makes use of the natural radial structure of the footprint, and of the (mainly) cos(𝜙)-dependence of the pulse energy from geomagnetic and charge-excess origin combined. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | For interpolation in the radial direction, a standard one-dimensional interpolation is used on the angular Fourier components 𝑐𝑘 (𝑟) and 𝑠𝑘 (𝑟) separately. A cubic spline interpolation was found useful here. After this, Eq. 2 can be used at any position within the radial boundaries of the simulated antennas. Using this method, simulations having on the order of 200 antennas would suffice for producing pulse energy footprints to an accuracy as desired for application to SKA measurements. A dedicated publication on this method and its precision is forthcoming. |
Processed_Prospects_for_measuring_the_longitudinal_particle_.txt | with the best-fit energy scale factor 𝐴, and 𝑓𝑖 the pulse energies in shower 𝑖. The MSE is the ‘noiseless’ equivalent of the 𝜒2 used when fitting simulations to realistic data with noise. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Abstract—Excessive caregiver workload in hospital nurses has been implicated in poorer patient care and increased worker burnout. Measurement of this workload in the Intensive Care Unit (ICU) is often done using the Nursing Activities Score (NAS), but this is usually recorded manually and sporadically. Previous work has made use of Ambient Intelligence (AmI) by using computer vision to passively derive caregiver-patient interaction times to monitor staff workload. In this letter, we propose using a Multiscale Vision Transformer (MViT) to passively predict the NAS from low-resolution thermal videos recorded in an ICU. 458 videos were obtained from an ICU in Melbourne, Australia and used to train a MViTv2 model using an indirect prediction and a direct prediction method. The indirect method predicted 1 of 8 potentially identifiable NAS activities from the video before inferring the NAS. The direct method predicted the NAS score immediately from the video. The indirect method yielded an average 5-fold accuracy of 57.21%, an area under the receiver operating characteristic curve (ROC AUC) of 0.865, a F1 score of 0.570 and a mean squared error (MSE) of 28.16. The direct method yielded a MSE of 18.16. We also showed that the MViTv2 outperforms similar models such as R(2+1)D and ResNet50-LSTM under identical settings. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | This study shows the feasibility of using a MViTv2 to passively predict the NAS in an ICU and monitor staff workload automatically. Our results above also show an increased accuracy in predicting NAS directly versus predicting NAS indirectly. We hope that our study can provide a direction for future work and further improve the accuracy of passive NAS monitoring. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | The Intensive Care Unit (ICU) is a complex environment where critically ill patients require constant monitoring and care provided by many caregivers in different roles [1, 2]. Caregiver workload, particularly in the ICU setting, is increasingly becoming recognized as an issue affecting patient safety, caregiver health and health provider costs [2]. The complexities of caregiver workload are influenced by various factors such as patient acuity, task complexity, staffing numbers and human factors impacting stress and burnout [2- 4]. In a systematic review by Pappa et al., at least 1 in 5 Health Care Workers (HCWs) reported symptoms of depression and anxiety, while 38.9% had insomnia [5]. In another systematic review, al Falasi et al. found that acute post-traumatic stress disorder (PTSD) symptoms were observed in 3.4% to 71.5% of HCWs [6]. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Given the contribution of workload burden to burnout, one proposed solution is to increase its monitoring to allow for more even distribution. Currently, the main methods used to calculate nursing workload are the Therapeutic Intervention Scoring System-28 (TISS- 28), the Nine Equivalents of Manpower Score (NEMS) and the Nursing Activities Score (NAS) [7-10]. When compared with TISS- 28 and NEMS in the ICU setting, the NAS achieved better scores and is most accurate at measuring nursing workload in the ICU [11-13]. Yet, like all other scoring systems, the NAS is recorded manually and typically only once every eight hours. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | transparent, ubiquitous and intelligent [15]. For instance, using cameras to continuously monitor vital signs of patients without any patient contact [16]. In addition, AmI has been used to observe other subjective patient states such as delirium or pain [17, 18]. More recently, AmI was used to monitor physical interaction times between healthcare staff and patients, with the goal of measuring staff workload to reduce overwork [18, 19]. There is potential for AmI to be used to continuously monitor metrics that are typically manually scored, such as the NAS. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | In this letter, we propose two Multiscale Vision Transformer (MViT) methods for the automatic calculation of the NAS in the ICU using low resolution infrared video feeds. The first indirect method predicts the NAS activity and infers the NAS from it. The second direct method predicts the NAS immediately without interpreting individual actions. Using 458 of our self-annotated ICU videos, we also compare the MViT to traditional models such as R(2+1)D and ResNet50-LSTM under identical settings. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Figure 1 provides an overview of our research workflow. From a comprehensive dataset of thermal videos captured in the ICU, 882 videos were chosen at random. Two independent clinicians then manually labelled each video with 1 of 23 possible NAS activities observed in the footage. Following this labelling process, the videos underwent preprocessing to prepare them for model training. In the final step, the trained model was used to predict the NAS. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Fig. 1. Research workflow for the prediction of NAS from an ICU video using a custom model (i.e. MViTv2, R(2+1)D or ResNet50- LSTM). |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Videos were captured using a Thermal Experts TE-Q1 narrow- angle camera (Daejon, Korea) mounted on the ceiling at a 30° downward angle, 2.2m away from the patient. The ward temperature was maintained at 22°C. All videos were captured at a resolution of 384 x 288 pixels and at 6 frames per second. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Each video was watched by 2 ICU registrars, who then labelled the 23 NAS activities as occurring (True) or not occurring (False) in the video. The registrars were blinded to the patient details, and the infrared video allowed patient privacy to be maintained. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | The videos had between 676 to 820 frames (mean = 761.59) and were 112 to 137 seconds long. In the 882 videos, 14 out of 23 NAS activities were observed. The remaining 9 NAS activities were not observed in any video. The ‘Before’ column of Table 1 shows the 14 NAS activities with their number of occurrences observed before preprocessing. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Given that a video may be labelled True for 1 or more NAS activity, this was a multilabel classification problem with 14 classes. To reduce the complexity of the problem, we simplified the problem into a single label classification problem by disregarding labels with less than 50 occurrences and only using videos with exactly 1 occurrence. After preprocessing, we are left with 458 videos from 8 classes (Table 1). |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | We propose the use of the Multiscale Vision Transformer v2 (MViTv2) model for NAS prediction. Proposed by Yanghao et al. in 2022, MViTv2 is a state-of-the-art model that achieved high accuracies on video classification tasks using the benchmark datasets Kinetics-400, Kinetics-600, Kinetics-700 and Something-Something- v2 [20]. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | MViTv2 is an improved version of a multiscale vision transformer (MViT), whose main idea is to combine Vision Transformers (ViT) with multiscale processing [20, 21]. Firstly, a ViT divides an image into a sequence of patches and applies a transformer model to it [21, 22]. Next, multiscale processing is done by increasing the number of channels while reducing spatial resolution [21]. This is equivalent to the convolutional and pooling layers in a Convolutional Neural Network (CNN) [23]. The idea is that a lower resolution reduces computational power while allowing the model to capture higher- level features [21, 23]. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | In our experiments, we used 2 methods to predict the NAS: 1) indirectly using predicted activity to infer the NAS; and 2) directly predicting the NAS. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | For the indirect method, we used a MViTv2 to predict the NAS activity, then inferred the NAS using the Average NAS values from Table 1. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | We used Pytorch 2.0’s implementation of MViTv2 pretrained on the Kinetics-400 dataset [26]. To customise it for our use, we added a Rectified Linear Unit (ReLU) activation function, then a final dense layer mapping the original 400 possible outputs to 8 possible outputs. The 8 possible outputs represent the prediction of the 8 possible NAS activities. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | During all training phases, we used 5-fold cross validation with 30 epochs in each fold and a batch size of 3. All models were trained on a Titan V GPU using Adam optimizer with learning rate = 0.00003. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | To evaluate the performance of our models, we calculated the accuracy (1), the area under the receiver operating characteristic curve (ROC AUC), the macro-averaged F1 score and the mean squared error (MSE) (2) for the indirect method, and the MSE for the direct method. The ROC AUC is the average AUC of the 8 curves given by plotting True Positive Rate (TPR) against False Positive Rate (FPR) for each class. The macro-averaged F1 score is the average F1 score across all classes, where each F1 score is calculated using (3). To infer predicted and true NAS in (2), the Average NAS of the activity is used (Table 1). |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Since pretraining was done using 16 frames for input, our videos were converted into 16 frames to use the pretrained weights. As our shortest video was 676 frames long, the spacing between the 16 frames was made as large as possible (42 frames) to ensure that these 16 frames are representative of the video. Hence for each video, the central 672 frames were taken and 16 frames with frameskip of 42 were taken from the 672 frames. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Table 2: Accuracy, ROC AUC, macro-averaged F1 score and MSE of NAS prediction using MViTv2, R(2+1)D and ResNet50-LSTM models. Results shown are the average across 5 folds. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | For the direct method, we used the MViTv2 to predict the NAS score directly. Similar to the indirect method, we used Pytorch 2.0’s implementation of MViTv2 pretrained on the Kinetics-400 dataset [26]. We added a ReLU activation function, then a final dense layer mapping the original 400 possible outputs to 1 possible output. The 1 output represents the NAS. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | In the same way as the indirect method, we selected 16 frames from each video as input and the MViTv2 will output the predicted NAS. The loss function used was PyTorch 2.0’s MSELoss with reduction set to ‘sum’. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Finally, we compared both methods for 3D-CNN and CNN-RNN model types. We used the R(2+1)D and ResNet50-LSTM models which are generic examples of each type respectively. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | a The direct method does not have an accuracy, ROC AUC or F1 score as the model outputs a numerical score. Only MSE can be calculated which tells us the closeness of the predicted score to the true score. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | In our R(2+1)D model, we used Pytorch 2.0’s implementation of R(2+1)D pretrained on the Kinetics-400 dataset [26]. For input, videos were converted into 16 frames in an identical way as above. Using our R(2+1)D, the NAS was predicted both directly and indirectly. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | The MViTv2 achieved the best 5-fold results across all metrics, performing better than the R(2+1)D and ResNet50-LSTM models (Table 2). MViTv2 achieved an accuracy of 57.21%, ROC AUC score of 0.865, F1 score of 0.570 and MSE of 28.16 for indirect prediction and MSE of 18.16 for direct prediction. In comparison, the R(2+1)D was second-best with an accuracy of 56.12%, ROC AUC of 0.849, F1 score of 0.550 and MSE of 28.70 for indirect prediction and MSE of 18.45 for direct prediction. Finally, ResNet50-LSTM came in last with an accuracy of 51.72%, ROC AUC of 0.816, F1 score of 0.498 and MSE of 34.01 for indirect prediction and MSE of 26.63 for direct prediction. This suggests that the MViTv2 is a more effective model for NAS prediction compared to the R(2+1)D and ResNet50-LSTM. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Comparing indirect and direct predictions of NAS, we see that the direct prediction is more precise since its MSE is consistently lower. By bypassing the need for predicting NAS activity, direct prediction is more streamlined and allows the model to predict any NAS value, giving it higher precision. In contrast, the indirect prediction has to decide on the NAS activity in the video, and can only output 8 specific NAS values corresponding to the 8 possible NAS activities. This makes indirect prediction less precise and increases its MSE. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | In this letter, we demonstrate 3 video classification models for predicting the NAS, propose a novel approach to passively and automatically predict caregiver workload, and ultimately show superior performance of MViTv2. This approach of workload monitoring has the added benefit of anonymizing staff and patients through the use of low-resolution thermal videos. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | We also experimented with a novel method of directly predicting the NAS. By not recording specific activity but only deriving a score, no personal information is saved, maintaining privacy while still providing the ability to continuously and anonymously monitor workload. |
Processed_Use_of_a_Multiscale_Vision_Transformer_to_predict_.txt | Other future directions include predicting the NAS for videos with more than 1 activity, and predicting other NAS activities or specific NAS activity subcategories. All these will help to increase the scope and accuracy of NAS prediction. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | form the O(a)-improvement of both setups to 1-loop order in perturbation theory, and we describe a way of minimising higher order cutoff effects by a redefinition of the renormalised coupling. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | c(cid:13) Copyright owned by the author(s) under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike Licence. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | Strongly interacting gauge theories other than QCD have been the focus of several studies in the recent past (see [1] for a review). These can be constructed by coupling Yang-Mills fields to N f fermions transforming under higher dimensional representations R of the gauge group SU (N). For certain values of N and N f they are expected to show conformal or quasi conformal behaviour over some range of scales. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | Schrödinger Functional (SF) finite volume renormalisation schemes provide a useful tool for studying non-perturbatively the scaling properties of the coupling. However, for the lattices acces- sible in numerical simulations these schemes are known to display lattice artefacts arising from the bulk and from the temporal boundaries, which may hide the universal continuum properties of the theory under consideration. Special care has to be taken in removing them such that continuum extrapolations can be done in a controlled and reliable way. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | In this work we examine the cutoff effects in the coupling at 1-loop order in perturbation theory, where the continuum limit is known. We will consider the standard SF setup with Wilson quarks [2, 3], and the chirally rotated SF (c SF) which implements the mechanism of automatic O(a) improvement [4]. Even after removing the O(a) effects in the coupling, the remaining lattice artefacts are still large [5, 6, 7] for the typical lattice sizes used in simulations and further strategies have to be developed to minimise them. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | In the text we will briefly review some of the basic definitions of the SF coupling and the step scaling function (SSF). We then study choices for the fermionic boundary conditions in the spatial directions which improve performance in numerical simulations. Finally, we propose a strategy to reduce the cutoff effects of the coupling in SU (3) with a possible application in SU (2). |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | and use them as a measure for the size of the cutoff effects of a specific regularisation. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | Table 1: Phases of the boundary fields for SU(3) [8] and SU(2) [2] used in this work. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | periodic. Fermion fields satisfy temporal boundary conditions with projectors ([3, 4]). The spatial boundary conditions for the fermions are discussed in section 3. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | In the c SF setup the bulk is automatically free from O(a) effects once the coefficient zf of a dimension 3 boundary counterterm has been tuned [4]. O(a) effects coming from the boundaries are removed by fixing 2 coefficients ct and ds. As for the SF setup, 1-loop improvement in the coupling is achieved by setting the coefficients to their tree level values zf = 1, d(0) s = 0.5 and c(0) t = 1, and by setting only c(1) to the correct 1-loop value. The bare mass is set to m0 = 0 as in the SF. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | The value of q can be chosen following the guideline principle established in [9]. In numerical the simulations the efficiency of the inversion algorithms depends on the condition number, i.e. ratio l max/l min between the largest and smallest eigenvalues of the squared fermion matrix.While l min strongly depends on q , l max is rather insensitive to it. Hence, by chooding the angle q such that l min in the BF is as large as possible the condition number is minimised, optimising the per- formance of the simulations in the perturbative regime. The lowest eigenvalue as a a function of q is shown in figure 1. The profile l min(q ) depends on the chosen BF and the fermion representation considered. Near optimal values of q for the groups and representations studied in this work can be found in table 2 and will be kept for the rest of the study. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | Figure 1: Lowest eigenvalue l min (in units of L− for the fundamental and symmetric representations of SU(3). The vertical discontinuous line denotes the selected value of q . The plots for the representations of SU(2) are not shown. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | Figure 2: Cutoff effects for the fermionic part of the SSF for SU(3) in the fundamental, symmetric and adjoint representations. Both setups (SF and c SF) are shown before and after O(a) improvement. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | for reducing the higher order cutoff effects, in which the BF is left intact and the coupling constant is modified. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | the direction in the algebra su(3) given by ( √3 2 seen by replacing h √3 l 8 2 this change does not affect the BF at all. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | A similar definition can be given for SU (3) by considering all 8 generators of the algebra. Whether these observables will help or not in reducing the cutoff effects in SU (2) is currently under inves- tigation. |
Processed_Lattice_artefacts_in_the_Schrödinger_Functional_co.txt | We have shown a way of controlling the 1-loop cutoff effects in the SSF for the representations of SU (3) without modifying the BF. The results of this perturbative calculation are encouraging, but it remains to be checked whether the cutoff effects at higher orders in perturbation theory can be removed in a similar fashion. Moreover, experience should be accumulated with the non- perturbative behaviour of v for representations other than the fundamental. The original choice of n = 0 for SU (3) was taken in order to obtain the best signal to noise ratio for the coupling in the pure gauge theory [8], so for the theories we are considering one probably has to find a balance between the reduction of cutoff effects and the quality of the signal. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Abstract—We report characterization results for an engi- neering prototype of a next-generation low-frequency radio astronomy array. This prototype, which we refer to as the Aperture Array Verification System 0.5 (AAVS0.5), is a sparse pseudo-random array of 16 log-periodic antennas designed for 70–450 MHz. It is co-located with the Murchison Widefield Array (MWA) at the Murchison Radioastronomy Observatory (MRO) near the Australian Square Kilometre Array (SKA) core site. We characterize the AAVS0.5 using two methods: in-situ radio interferometry with astronomical sources and an engineering approach based on detailed full-wave simulation. In- situ measurement of the small prototype array is challenging due to the dominance of the Galactic noise and the relatively weaker calibration sources easily accessible in the southern sky. The MWA, with its 128 “tiles” and up to 3 km baselines, enabled in-situ measurement via radio interferometry. We present array sensitivity and beam pattern characterization results and compare to detailed full-wave simulation. We discuss areas where differences between the two methods exist and offer possibilities for improvement. Our work demonstrates the value of the dual astronomy–simulation approach in upcoming SKA design work. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 1. The AAVS0.5 is an array of 16 dual-polarized SKALA antennas (inset) pseudo-randomly placed in an 8 m diameter circle. In current imple- mentation, the antennas are mounted over soil without a groundplane. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | its response to celestial radio emission. For antenna array characterization, bright compact cosmic radio sources, with some caveats, are convenient tools, as their positions on the celestial sphere are well known and they emit broadband noise [7], enabling wide band measurement of the antenna and receiver under test (AUT) [7], [8]. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | In the current pre-construction phase of the Square Kilome- tre Array (SKA) radio telescope [1]–[3], verification systems are essential in demonstrating that candidate designs meet functionality, cost target and site requirements. To demon- strate potential telescope and sub-system designs, a series of Aperture1 Array Verification Systems (AAVSs) are being constructed. The AAVS0 [4], [5] was the first prototype constructed in Lord’s Bridge (UK). The AAVS0.5, shown in Fig. 1, is the first verification system in Australia2 [6]. In the near future, the AAVS0.5 will be succeeded by the AAVS1, a larger prototype3 having at least one phased array of 256 antennas corresponding to a full SKA low-frequency aperture array station. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | However, at low frequencies such sources are not numerous in the southern sky. Furthermore, the Galactic noise, which is a greatly extended rather than compact source, is dom- inant [9]–[11]. As we will discuss in detail, we overcome these challenges through low-frequency radio interferometry, which for the AAVS0.5 is performed in conjunction with the Murchison Widefield Array (MWA) [12], [13]. The MWA is one of several operating low-frequency radio interferometers, including LOFAR [14] and PAPER [15]. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | This paper is organized as follows. Sec. II provides a description of the AAVS0.5 array. Full-wave simulation of the array is discussed in Sec. III. Sec. IV reviews the method of characterization via radio interferometry. Sec. V reports measurement results and comparison to full-wave simulation. Concluding remarks are given in Sec. VI. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Fig. 2 shows the AAVS0.5 location, which is within the extent of the MWA “tiles” (each of which is an sub-array of 4×4 bow-tie antennas). This arrangement offers up to 127 MWA tiles at the sufficiently large distances to the AAVS0.5 that are necessary for its characterization (further discussion in Sec. IV). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | The AAVS0.5 consists of 16 dual-polarized log-periodic “SKALAs” capable of operation from 50 to 650 MHz [16]. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | c(cid:13)2015 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Fig. 3 illustrates the pseudo-random positions of the SKALAs located within an 8 m diameter circle. The sparse pseudo- random placement, as opposed to regular spacing, has been selected to prevent the appearance of grating lobes [17]–[20] and scan blindess [4], [21], [22] over the 50 to 350 MHz SKA bandwidth (7 : 1) [23], [24]. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 3. The AAVS0.5 as-built points. In the field, y-axis is aligned to North. The as-built points are the locations of the bases of the SKALAs established through professional differential GPS survey post construction. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | The AAVS0.5 relies on the MWA infrastructure for power, RF signal chain and data processing back-end [12]. The key components of the signal path are shown in Fig. 4 and described below. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Low-noise amplifier4 (LNA): nominal noise temperature and gain at 180 MHz are 33 K and 43 dB respectively. The LNA is mounted at the feed point of each SKALA and powered via the coax from the MWA beamformer (B/F). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Modified MWA B/F: modified to accept the LNA, whose gain is higher than a standard MWA LNA. Beamforming is accomplished through 32 analog delay steps, 435 ps each [12], allowing the AAVS0.5 to scan to zenith angles (ZA) of ∼ 30◦. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | The digitized data are sent to the MWA correlator such that the AAVS0.5 can be used with normal MWA observing [28] (not shown). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 4. The AAVS0.5 RF system block diagram for one polarization of the SKALA. ASC stands for “analog signal conditioning”. RG-223 cables are phased matched solid polyethylene (PE) 50 Ω coaxial cables. RG-6 cable is 75 Ω PE coaxial cable provides power and communications to the MWA B/F. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Fig. 5 reports the measured receiver noise temperature of the AAVS0.5 system via the RF signal path as described above. The measurement was performed on-site using a hot/cold noise source connected to a LNA and then to the rest of the system via a 16-to-1 power divider (de-embedded in post-processing). The outliers between 250–300 MHz are likely due to satellite RFI. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | An important motivation of the AAVS0.5 characterization is to verify the accuracy of the type of tools available in the design of low-frequency aperture arrays. As the AAVS0.5 does not have a ground plane in this implementation, it is likely to be closely coupled to the environment. As a result, full- wave simulations have become essential in accounting for the effects of soil and cables in the design process. Confirming the accuracy these models is of great interest. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Fig. 6 depicts a typical AAVS0.5 model in a commercial Method-of-Moments code, FEKO5. The antennas are modeled as a perfect electric conductor (PEC) and the antenna ports are loaded with the measured LNA impedance reported in Fig. 7. The LNA is a balanced (input) to single-ended (output) device with high common mode rejection ratio (CMRR >∼ 35 dB) in the frequency range of interest. The coaxial cables are simulated as 15 m long wires, 3 mm in diameter. Given the high CMRR, the cables are terminated with open circuits near the feed points as shown in the inset in Fig. 6. The simulated soil properties are based on measured soil samples from the MRO, which were characterized experimentally as a function of moisture level, listed in Tab. I at spot frequencies. With the above parameters as input, the beam is simulated for the required pointing direction, where each antenna is given equal amplitude and a relative phase corresponding to the switched- in time delay. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 6. A screen capture of the 16-element AAVS0.5 antenna array simulated in FEKO. The antenna bases are placed on semi-infinite soil (not shown). The 15 m coaxial cables are modeled as L-shaped wires seen in the figure. The vertical part of the L-shape extends to just below the feed point shown in the inset; the horizontal part is at ground level. Each antenna feed point is loaded with measured LNA impedance. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Table I MEASURED RELATIVE PERMITTIVITY (cid:15)r AND CONDUCTIVITY σ (S/M) OF MRO SOIL FOR DIFFERENT MOISTURE LEVELS. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | where Dmax is the maximum array directivity for a given pointing direction, and η is the radiation efficiency, which includes radiation losses due to the soil as the array is simulated on a semi-infinite lossy dielectric medium in FEKO (Fig. 6). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | where D(θ, φ) array directivity pattern (e.g., D(θ, φ)/Dmax in Fig. 8) in spherical coordinates7, Tsky(θ, φ) is the sky under observation (e.g., Fig. 9) and dΩ = sin θdθdφ is the infinitesimal element of solid angle. The first two terms in (2) constitute the total noise temperature seen at the receiver’s input due to the antenna; this is referred to as total antenna temperature. Note that Tsky increases exponentially with wavelength such that it is the dominant noise source at low <(∼ 200 MHz) frequencies [9]–[11]. For our simulated A/T, we use sky models based on standard all-sky maps, being relatively low angular resolution observations of the sky at radio frequencies [29], [30] to calculate Ta in (3). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 8. The FEKO-simulated array normalized beam pattern D(θ, φ)/Dmaxfor the AAVS0.5 on soil with 2% moisture at 160 MHz (Dmax = 18 dB) for the X (East-West) polarization, pointing towards Az=0◦, ZA=14.6◦. The AAVS0.5 half-power beamwidth is ∼ 14◦ at this frequency. Radiation efficiency η = 0.83. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 9. Sky hemisphere Tsky(θ, φ) at 160 MHz [30] over the MRO site on 18 January 2015, 1:42 local time (9.27 h local sidereal time). The spatially extended Galactic plane is seen from top-left to bottom-right. Azimuth angle is clockwise from North towards East. The white plus indicates the location of the calibrator source Hydra A. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Parameters such as array sensitivity and beam pattern are direction-specific performance measures. Direct sensitivity measurement usually involves pointing on and off a bright, compact astronomical source to measure array response in a particular direction [7], [10]. This approach is practicable for an AUT with sufficient sensitivity to detect that object in the presence of system noise. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | where k is the Boltzmann constant (using k = 1380 leads to SEFD in unit of Jy). |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Due to its small size relative to operating wavelength and the high sky noise at 160 MHz, the AAVS0.5 is insufficiently sensitive to discern the difference in total power (on/off source) with integration times of the order of seconds. As an illustration, consider the simulated sensitivity of the AAVS0.5 of ∼ 0.06 m2/K. Using (4), the SEFD of the AAVS0.5 is approximately 46 kJy, which is much greater than easily accessible8, “bright” (∼0.5 kJy) point-like sources in the southern sky [31]. However, radio interferometry enables us to detect these compact sources and use them for the AAVS0.5 characterization. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | where Vij(ˆx) is the visibility measurement between antennas i and j, I(ˆkx) is the flux density of the sky noise incident on the array, ˆkx is the normalized free-space wavenumber in the x-direction and ˆx = x/λ. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Fig. 10 illustrates the spatial filtering property of an inter- ferometer in one dimension described by (5). For simplicity, consider only the real part of (5). The black curve in top graph depicts a bright extended source illustrative of Galactic noise as a function of direction. The area under that curve is 40. The black curve in the bottom graph illustrates a compact source with spatial extent << 1/ˆx, with integrated value 0.5. The red curve in the top graph represents the integrand in (5) for ˆx = 40 for the extended source; it is a product of the black curve with cos(2πˆkx40). Note that the area under that red curve reduces to 0.16. In contrast, the areas under the curves remain virtually unchanged for the compact source (bottom graph), showing that the interferometer ˆx = 40 is sensitive to the desired compact source but insensitive to the undesired bright extended source. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 11. Amplitude (Jy) of calibrated MWA visibility measurements as a function of baseline length in wavelengths, projected towards Hydra A. These data are for a 2-minute snapshot observation of Hydra A with the MWA and the AAVS0.5 on 18 January 2015. Only a single 40 kHz, X- polarization channel at 160 MHz is shown. The black curve is a 10th- order polynomial fit of the amplitude, and the magenta curves show the 1σ flux density standard deviation of the mean of each calibrated visibility measurement (see Sec. IV-C). The horizontal dashed line shows the expected measurement if the noise contribution was solely from Hydra A as an idealized point source (310 Jy at 160 MHz). The vertical dashed line marks the 30λ baseline cut-off. |
Processed_Characterization_of_a_Low-Frequency_Radio_Astronom.txt | Figure 10. A numerical example of (5), showing the spatial filtering property of an interferometer in 1-D for an extended source (top) and compact source (bottom) for ˆx = 0 and 40. The legends show the area under each curve. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.