Chen H.,MathWorks Inc. |
Gentile R.,MathWorks Inc.
IEEE International Symposium on Phased Array Systems and Technology | Year: 2017
Phased array radar system development starts with the design and analysis of the antenna array. While the benefits of phased array systems are great, the corresponding complexity grows with the additional transmit and receive channels [1,2,3,4,5,6]. Large antenna arrays can involve complex designs that include subarrays and a variety of configuration options. Having a flexible system level model that can be used to design, measure and assess the performance of an antenna array before it is built is critical to avoid costly errors late in the design cycle. In addition, design analysis and 'what if' exercises can be greatly simplified with the availability of simulation capabilities early in the design cycle. A simulation framework will be described that can be used to analyze beam patterns of the antenna arrays. It can also be used to explore geometry options, element spacing, tapering parameters, array thinning, and steering performance. The framework makes it possible to model failed or imperfect array elements, develop calibration techniques, and assess the impacts of mutual coupling among elements in the antenna array. Additionally, to assess the impacts of phased array design decisions in the context of overall system level performance, a more complete framework for modeling and simulation will also be described by which multi-domain, multi-function radar systems can be designed and analyzed. © 2016 IEEE.
News Article | April 20, 2016
We imaged the primary visual cortex of an awake 9-month-old C57BL/6 male mouse, as described previously10, 13, with a custom-built two-photon microscope12. Using volumetric in vivo two-photon calcium imaging of a genetically encoded calcium indicator (GCaMP3), we measured the time-resolved responses of a population of identified neurons to a wide array of stimuli including drifting gratings (up to 16 directions, 3 spatial, and 2 temporal frequencies). Following 12 days of imaging calcium responses in the same cohort of neurons, we labelled blood vessels with a tail vein injection (rhodamine B-conjugated dextran) and acquired an in vivo fluorescence volume. The animal’s brain was then prepared for large-scale transmission EM as described previously8. 3,700 serial sections (<50 nm thick) were cut and imaged spanning a 450 μm × 450 μm × 150 μm volume at 4 nm × 4 nm × 40 nm per voxel resolution. Sections representing the middle third of the EM volume were aligned and imported into CATMAID16 for distributed, online, manual reconstruction and targeted volumes around identified synapses were exported for volumetric segmentation and PSD analysis. EM reconstructed neurons were identified in the in vivo stack by using the blood vessels as landmarks. Apical dendrites originating from deeper neocortical lamina were similarly identified and corresponded by location and branching geometry of their apical tufts. Permutation tests were used in statistical analyses, unless otherwise noted. All procedures were conducted in accordance with the ethical guidelines of the NIH and approved by the IACUC at Harvard Medical School. For cranial window implant surgery the mouse was anesthetized with isoflurane (1.2–2% in 100% O ). Dexamethasone (3.2 mg per kg body weight, intramuscular) was administered on the day before surgery and atropine (0.2 mg per kg body weight, intraperitoneally) at the beginning of surgery. Using aseptic technique, we secured a headpost in place using cyanoacrylate, dental acrylic, and C&B Metabond (Parkell), and made a 5 mm craniotomy over the left visual cortex (centre: ~2.8 mm lateral, 0.5 mm anterior to lambda) as described previously32. A 5 mm glass cranial window was implanted consisting of an 8 mm coverslip cured to two 5 mm coverslips (Warner #1; total thickness: ~0.5 mm; thickness below skull: ~200 mm) using index-matched adhesive (Norland #71). We secured the window in place using cyanoacrylate and dental acrylic. We habituated the mouse with water scheduling so that water was delivered only during and immediately after head restraint training. We increased the duration of head restraint sessions over the course of 2 weeks, from 3 min to 2 h32. We then performed retinotopic mapping of visual cortical areas using widefield intrinsic autofluorescence imaging, measuring autofluorescence produced by blue excitation (470 nm centre, 40 nm band, Chroma) through a green/red emission filter (longpass, 500 nm cutoff). We collected images using a CCD camera (Sensicam, Cooke, 344 × 260 pixels spanning 4 mm × 3 mm; 2 Hz acquisition rate) through a 5× air objective (0.14 NA, Mitituyo) using ImageJ acquisition software. For retinotopic mapping, stimuli were presented at 2–6 retinotopic positions for 10 s each, with 10 s of mean luminance between trials. GCaMP3 expression was targeted by viral injection. Dexamethasone (3.2 mg per kg body weight, intramuscular) was administered at least 2 h before coverslip removal. The mouse was anesthetized (isoflurane, 1–1.5%) and the cranial window was sterilized with alcohol and the coverslip removed. We then volume injected (50–100 ml min−1, Stoelting) 30–100 nl of a 10:1 mixture of AAV2/1.hSynap.GCaMP3.3.SV4033 (Penn Vector Core) and 1 mM sulforhodamine-101 (Invitrogen) to visualize the injection. Using the blood vessel pattern observed during widefield imaging as a guide, we made an injection in the posterior part of primary visual cortex at a depth of ~250 μm below the pial surface. After injection, a new cranial window was sealed in place and the mouse recovered. A 120 Hz LCD monitor (Samsung 2233RZ, 2200) was calibrated at each temporal frequency using a spectrophotometer (Photoresearch PR-650). We confirmed waveforms were sinusoidal by measuring luminance fluctuations of a full-field sinusoidally modulated stimulus (using a photomultiplier tube, Hamamatsu). The monitor was positioned so that the stimulus patch was 21 cm from the contralateral eye. Local 40° Gabor-like circular patches (sigmoidal 10–90% falloff in 10°) containing either square-wave (for mapping retinotopy with widefield intrinsic autofluorescence and targeting GCaMP3 injections) or sine-wave (for mapping position of receptive fields with two-photon imaging) drifting gratings (80% contrast) were alternated with periods of uniform mean luminance (59 cd m−2). In an effort to increase the population of responsive cells and explore receptive field parameters we presented gratings of varying directions at multiple spatial and temporal frequencies or at different positions in the visual field. We presented either 8 directions at 3 spatial frequencies (0.06, 0.12, and 0.24 cycles per degree (cpd)) and 2 temporal frequencies (2 and 8 Hz), 16 directions at 2 spatial frequencies (0.04 and 0.16 cpd) and 2 temporal frequencies (2 and 8 Hz), 8 directions at 6 positions, or 16 directions at 4 positions (45–115° eccentricity and −5–25° elevation), for a total of 64 stimulus types plus 10% blank trials. Stimuli were centred on the location eliciting maximum calcium responses in the imaged field (monocular cortex), which most effectively drove responses in the population for experiments that did not vary stimulus position. All stimuli in a given protocol were presented in a pseudo-random order (sampling without replacement), and presented 3 times per volume experiment with 2–4 experiments per volume per day. Imaging was performed with a custom-designed two-photon laser-scanning microscope12. Excitation light from a Mai Tai HP DeepSee laser (Spectra-Physics) with dispersion compensation was directed into a modulator (Conoptics) and a beam expander (Edmund Optics). The expanded beam was raster scanned into the brain with a resonant (4 kHz, Electro-Optical Products) and a conventional galvanometer (Galvoline) (240 line frames, bidirectional, 31 Hz) through a 0.8 numerical aperture (NA) 16× objective lens (Nikon). Emitted photons were directed through a green filter (centre: 542 nm; band: 50 nm; Semrock) onto GaAsP photomultipliers (no cooling, 50 μA protection circuit, H7422P-40MOD, Hamamatsu). The photomultiplier signals were amplified (DHPCA-100, Femto), and low-pass filtered (cutoff frequency = ~700 kHz). These and the mirror driver signals were acquired at 3.3 MHz using a multifunction data acquisition board (PCI-6115, National Instruments). Images were reconstructed in MATLAB (MathWorks) and continuously streamed onto a RAID array. Microscope control was also performed in MATLAB using an analogue output board (PCI-6711, National Instruments). The laser’s dispersion compensation was adjusted to maximize collected fluorescence. A piezoelectric objective translator on the microscope enabled imaging multiple 300 × 300 × 100 μm volumes with 8 planes at 4 Hz separated by ~12.5 μm allowing us to capture the response properties of many cells through the depth of L2/3. The imaged field of view was 200–300 μm on a side at resolution of 0.8–1.2 μm per pixel (dwell-time ~2.7 μs). GCaMP3 was excited at 920 nm. Laser power was automatically adjusted as a function of imaging depth at the modulator with power exiting the objective ranging from 30–60 mW. During imaging, the mouse was placed on 6-inch diameter foam ball that could spin noiselessly on an axel (Plasteel). We monitored trackball revolutions using a custom photodetector circuit and recorded eye movements using an IR-CCD camera (Sony xc-ei50; 30 Hz) and infrared illumination (720–2,750 nm bandpass filter, Edmund). Visual stimuli were presented for 4 s with 4 s of mean luminance between trials. Recording sessions were 2–6 h in duration. Use of the genetically encoded calcium indicator GCaMP3, permitted recording from the same neurons over multiple days with the selectivity of calcium signals stable over several days of imaging (Extended Data Fig. 1)32, 34, 35. Within this volume we obtained calcium signals from cell bodies of superficial layer (L2/3) neurons and large calibre apical dendrites that continued beyond the depth of our imaging volume and had branching morphologies consistent with deep layer pyramidal cells. These were likely from L5 neurons because of their large calibre, and because most L6 pyramidal cells do not project their apical dendrites more superficially than L436, 37. The calcium signals from these deep layer apical dendrites stem from either forward-38, 39 or back-propagating action potentials40, are consistent across days (Extended Data Fig. 1) and along the length of the deep layer apical dendritic trunks (Extended Data Fig. 2), and therefore most likely reflect the response properties of the soma. We relocated the cohort of neurons daily by using the vasculature’s negative staining as fiducial landmarks. For the final in vivo imaging session, we injected the tail vein with a fluorescent dye to label blood vessels (rhodamine B isothiocyanate–Dextran (MW ~70k), 5% v/v, Sigma) and acquired a fluorescence stack to correspond the calcium-imaged neurons in vivo with their identities in the EM volume ex vivo8 (see below, and Extended Data Fig. 4). Following in vivo two-photon imaging the animal was perfused transcardially (2% formaldehyde/2.5% glutaraldehyde in 0.1 M cacodylate buffer with 0.04% CaCl ) and the brain was processed for serial-section TEM. 200 μm thick coronal vibratome sections were cut, post-fixed, and en bloc stained with 1% osmium tetroxide/1.5% potassium ferrocyanide followed by 1% uranyl acetate, dehydrated with a graded ethanol series, and embedded in resin (TAAB 812 Epon, Canemco). We located the calcium-imaged region by matching vasculature between in vivo fluorescence and serial thick (1 μm) toluidine blue (EMS) sections cut from an adjacent vibratome sections, then cut ~3,700 serial (<50 nm) sections on an ultramicrotome (Leica UC7) using a 35 degree diamond knife (EMS-Diatome) and manually collected sections on 1 mm × 2 mm dot-notch slot grids (Synaptek) that were coated with a pale gold Pioloform (Ted Pella) support film, carbon coated, and glow-discharged. Following section pickup, we post-stained grids with uranyl acetate and lead citrate. Using the custom-built transmission electron microscope camera array (TEMCA)8 we imaged the ~3,700 section series, targeting a ~450 μm × 450 μm region for each section (Fig. 1c). Acquired at 4 nm per pixel in plane, this amounted to ~100 terabytes of raw data to date comprising 30 million cubic microns of brain and >10 million (4,000 × 2,672 pixel) camera images. Magnification at the scope was 2,000×, accelerating potential was 120 kV, and beam current was ~90 microamperes through a tungsten filament. Images suitable for circuit reconstruction were acquired at a net rate of 5–8 million pixels s−1. Approximately the middle third of the series (sections 2,281–3,154) was aligned using open source software developed at Pittsburgh Supercomputing Center (AlignTK)8 and imported into CATMAID16 for distributed online visualization and segmentation. Within the analysed EM series there were 51 missing sections. Nineteen were single section losses. There were 2 instances each of missing 2, 3, and 4; and 1 instance each of missing 6 or 8 consecutive sections near the series boundaries. Folds, staining artefacts, and sometimes cracks occurred during section processing, but were typically isolated to edges of our large sections and therefore did not usually interfere with manual segmentation. To find the correspondence between the cells imaged in vivo with those in the EM data set, a global 3D affine alignment was used with fiducial landmarks manually specified at successively finer scales of vasculature and then cell bodies to re-locate the calcium-imaged neurons in the EM-imaged volume (Extended Data Fig. 4). Apical dendrites arising from deep layer (putative L5) pyramidal neurons were identified by their characteristic morphology36, 41, 42 (also see below). Their correspondence was facilitated by the unique branching patterns of their apical tufts and those that could not be unambiguously identified were not included in the functional analysis. We first traced the axonal and dendritic arbors of the functionally characterized neurons in the EM data set by manually placing a series of marker points down the midline of each process to generate a skeletonized model of the arbors using CATMAID16 (Figs 1d, 2a, 3a, Extended Data Fig. 6, Supplementary Data 1–3). We identified synapses using classical criteria42. For each synapse on the axon of a functionally characterized cell, dendrites of postsynaptic excitatory neurons were traced either to the boundaries of the aligned volume or centripetally back to the cell body8. We identified deep layer apical dendrites of (putative L5) pyramidal cells by their large calibre, high spine density, and their continuation beyond the bottom border of the EM volume, which spans from the pial surface through L4. For each neuronal target reconstruction included in the analysis, a second independent annotator verified the tracing by working backwards from the most distal end of every process. An additional round of validation was done for each synapse between functionally characterized cells where a third annotator who had not previously traced the pre- or post-synaptic process, independently verified the anatomical connectivity blind to previous tracing work. We began this independent round of validation at each synapse and traced the pre- and postsynaptic processes centripetally. If the initial reconstruction and subsequent verification of the reconstruction diverged, that connection and the segmentation work distal from the point of divergence was excluded from further analysis. EM reconstruction and validation was performed blind to cells’ functional characteristics and targeted cells were initially assigned to individual annotators pseudo-randomly weighted by tracing productivity. We performed targeted volumetric reconstructions of synapses connecting functionally characterized cells by developing tools to interface with CATMAID cutout, locally align, and catalogue volumes of interest based on location (Fig. 4a; for example, 400 pixels × 400 pixels × 41 sections or 3.2 μm × 3.2 μm × 1.64 μm volumes centred on synapses represented by CATMAID connectors). Presynaptic boutons, postsynaptic spines, their parent axons and dendrites, and postsynaptic density (PSD) areas were manually segmented with itk-SNAP (http://www.itksnap.org/). PSD areas were calculated as described previously43 with obliquely cut or en face synapse areas measured using their maximum z-projection. En face or obliquely cut synapses were identified by serial sections that starkly transitioned from a clear presynaptic specialization hosting a vesicle pool, to a distinctly different postsynaptic cell, typically with an intervening section of electron dense area representing the postsynaptic density and/or synaptic cleft (for example, Extended Data Fig. 5). In vivo calcium imaging data was analysed in MATLAB and ImageJ (NIH) as described previously12, 13. To correct for motion along the imaged plane (x–y motion), the stack for each imaging plane was registered to the average field of view using TurboReg44. A 5 pixel border at each edge of the field of view was ignored to eliminate registration artefacts. Masks for analysing fluorescence signal from neurons were manually generated corresponding to cells in the EM volume, registered to the in vivo anatomical fluorescence stack, and to individual physiological imaging planes. Time courses of cells spanning multiple physiological imaging planes were weighted by dwell time in each plane and averaged across planes. Evoked responses for each EM identified cells were measured for each stimulation epoch as the difference in fractional fluorescence (% ΔF/F ) between the 5 s after and the 2.5 s before stimulus onset (pre-stimulus activity), and averaged across stimulus repetitions. We quantified visual responsiveness of each cell by calculating the average Pearson correlation coefficient of the responses to all stimuli across repetitions (average trial-to-trial correlation). We defined the significance of visual responses as the probability (P value) that the observed trial-to-trial correlation is larger than the correlation obtained from a full random permutation of the data for spatial and temporal frequency experiments (P < 0.05) and experiments where stimulus position was varied (P < 0.01). In retinotopic experiments designed to increase the number of characterized neurons, we found cells that did not reliably respond to stimuli ± 20° from the centre of the display. These cells that either had receptive fields smaller than our stimuli or stimuli were positioned at the at the edge of their receptive fields. We considered these cells as potentially driven by stimulus edge effects and therefore excluded such experiments from further analysis. To estimate the preferred orientation, direction, and spatiotemporal frequency, we modelled responses with a combination of a multivariate Gaussian with spatial frequency (x and y, deg), temporal frequency (Hz) and position (x and y, deg) as independent dimensions, a constant gain factor, and a static exponent. We fit the model to data using a large-scale nonlinear optimization algorithm (Trust Region Reflective, MATLAB Optimization Toolbox, MathWorks Inc.), generating multiple fits from randomly selected starting points and selected the best fit (least-square criterion). The quality of model fits was inspected visually for all neurons included in the data set. EM connectivity was analysed using custom written software in MATLAB and Python. Connectivity analysis that did not utilize functional information (Figs 1e and 3, Extended Data Fig. 7) started with the entire population of excitatory neuronal targets in the reconstructed network. Network modularity and neuron connectivity motifs (Fig. 1e and Extended Data Fig. 7) were analysed with code modified from the Brain Connectivity Toolbox45. We used an implementation of the Louvain method17 followed by consensus portioning46 for weighted and directed graphs to detect communities, or interconnected pyramidal neuron targets, from our EM reconstructed network purely by anatomical connectivity. For this analysis we included only the 201 traced neurons having multiple synaptic partners (degree ≥ 2). The number of synapses reconstructed between neurons was used as weights for all analyses. Modularity Q was given by the standard equation: where l is the total number of edges, given by where N is the total set of nodes, a is the (i,j) element of the weighted adjacency matrix, δ(m ,m ) is 1 if i and j are in the same community and 0 otherwise, and are the in and out degrees of the jth and ith nodes respectively, calculated by To generate null models of connectivity matrices for hypothesis testing, we shuffled the reconstructed adjacency conditioned on our sample degree, weight and strength distributions (Extended Data Fig. 7)31, 47. Analysis of connectivity with neuronal function restricted our sample population to those cell pairs where both pre- and post-synaptic cells were functionally characterized. For orientation tuning (Figs 1d, f, 2, 4a–c, Extended Data Figs 5, 6, 8, 9), between 50 neurons, there were 29 connected pairs. On average, we detected 1.3 synapses per connected pair where we measured orientation selectivity for both cells. We varied retinotopic position and spatial and temporal frequencies of the grating stimulus with the goal of improved measurement of orientation preference for more cells. The sensory physiology of a subset of cells were simultaneously recorded across multiple stimulus parameters. These 120 cells were used for signal correlation analysis (Extended Data Fig. 10). Potential synapse length (L ) represents the degree to which pairs of neurons’ axonal and dendritic arbors come sufficiently close to make a synapse (Fig. 2a, c–f, 3b, d, Extended Data Figs 9, 10). For excitatory pyramidal cells, we computed this length of potential synaptic connectivity between all pairs by first resampling the dendritic and axonal arbor skeletons to a maximum segment length of 40 nm (the average thickness of the EM sections) and summing the length of all dendrite segments within a maximum spine length distance of the axon (s = 5 μm: Figs 2, 3 and Extended Data Fig. 10; s = 1 μm: Extended Data Fig. 9). We use s = 5 μm based the longest spine connecting functionally connected neurons (~ 5 μm). Analysis of neurons connected by multiple synapses (Fig. 3) was not restricted to cell pairs where both pre and post-synaptic cells were physiologically characterized. This population included 137 neurons connected by 267 synapses in 115 multi-synapse cell pairs whose axonal and dendritic arbors were traced exhaustively in the aligned volume. As a comparison population, we used 25 unique pairs connected by one synapse from the functionally characterized population described above, because they were also reconstructed throughout the aligned volume. To examine whether poly-synaptic connectivity occurs greater than random, we first computed a population average synapse rate (λ ) normalized by potential synapse length, by dividing the total number of synapses reconstructed from the set of 50 functionally characterized neurons by their total pairwise L . We next compared λ for individual neuron pairs each connected by different numbers of synapses (Fig. 3b). This was used to assess whether multiple synapses occurred more often than predicted from a simple Poisson model. We examined the frequency of clustered vs distant synapses by comparing synapse pairs that were separated by >20 μm or <20 μm. For each synapse from each pair of neurons connected by n synapses, we computed the total L within 20 μm or beyond 20 μm from that synapse. We then took the fraction of the overlap beyond 20 μm: as the expected probability that each of the (n − 1) other synapses will occur >20 μm away. The expected number of distant synapse was taken as (n − 1) times the fraction of overlap beyond 20 μm, which was compared with the actual number of distant synapses observed (Fig. 3d). 3D renderings were generated using Blender (http://www.blender.org/) (Figs 1d, 2a, 3a, Extended Data Fig. 6, Supplementary Data 1–3), Imaris (Bitplane) (Extended Data Fig. 4 and Supplementary Video 1), and itk-SNAP (Fig. 4a). Cytoscape (http://www.cytoscape.org/) was used for network graph layouts (Figs 1f). Statistical methods were not used to predetermine sample sizes. Statistical comparisons between sample distributions were done with Permutation tests (that is, Monte Carlo-based Randomization tests) unless otherwise noted. Permutation tests were ideal as we do not assume the underlying distributions are normal, nor need the observations to be independent. For Permutation tests, we computed the incidence of differences between means or Pearson’s linear correlation coefficient of randomly drawn samples from combined sample distributions exceeding the empirical difference (Figs 2b–d, f, 4c and Extended Data Figs 7b, 9a, b, 10c, d). Cochran-Armitage two-sided tests for trend were used on proportional binned data with linear weights (Fig. 2b, f). Standard errors were calculated from bootstrapped sample distributions. For cumulative distributions (Figs 2c, d, and Extended Data Figs 9a, b, 10c, d), we repeatedly resampled by randomly drawing with replacement from the sample distribution the number of observed values 1,000–10,000 times and extracted the standard deviation at each step of the empirical CDF. For binned data (Fig. 2b, f, and Extended Data Fig. 9d), each resampled distribution was binned and the standard deviation was computed from the resampled probabilities or rates within each bin. Custom code is available upon request.
Ling J.,MathWorks Inc. |
Li J.,University of Florida
IEEE Journal of Oceanic Engineering | Year: 2012
The acoustic communication channel is frequency selective with long memory, leading to severe intersymbol interference (ISI). To mitigate ISI, equalizer becomes an indispensable module in the receiver structure. However, the time-varying nature of the underwater acoustic environment imposes unique challenges to the design of an effective equalizer. First, the equalization process needs to be performed on a block basis and the block length could be short. Second, concerning that the dynamic acoustic medium makes the newly acquired channel information readily outdated, it is desirable that the equalizer performance is robust against the inaccuracy of the channel information when the transmission scheme involves cross-block reference. In this paper, we consider a statistical semiblind equalizer implemented by the Gibbs sampler techniques. The proposed equalizer conducts channel estimation and symbol detection in a joint manner, and it is robust to the accuracy of the channel information. The effectiveness of the proposed semiblind equalizer is demonstrated using both simulated and the 2008 Surface Processes and Communications Experiment (SPACE08, Martha's Vineyard, MA) in-water experimentation examples. © 2011 IEEE.
Tan Y.,MathWorks Inc. |
Venkatesh V.,North Carolina State University |
IEEE Transactions on Parallel and Distributed Systems | Year: 2013
Large-scale hosting infrastructures have become the fundamental platforms for many real-world systems such as cloud computing infrastructures, enterprise data centers, and massive data processing systems. However, it is a challenging task to achieve both scalability and high precision while monitoring a large number of intranode and internode attributes (e.g., CPU usage, free memory, free disk, internode network delay). In this paper, we present the design and implementation of a Resilient self-Compressive Monitoring (RCM) system for large-scale hosting infrastructures. RCM achieves scalable distributed monitoring by performing online data compression to reduce remote data collection cost. RCM provides failure resilience to achieve robust monitoring for dynamic distributed systems where host and network failures are common. We have conducted extensive experiments using a set of real monitoring data from NCSU's virtual computing lab (VCL), PlanetLab, a Google cluster, and real Internet traffic matrices. The experimental results show that RCM can achieve up to 200 percent higher compression ratio and several orders of magnitude less overhead than the existing approaches. © 2013 IEEE.
Ljung L.,Linköping University |
Singh R.,MathWorks Inc.
IFAC Proceedings Volumes (IFAC-PapersOnline) | Year: 2012
Version 8.0 of MATLAB's System Identification toolbox is released with version R2012a of MATLAB in the spring of 2012. This release presents a re-engineered implementation of the code using the new MATLAB object-oriented programming. Two main features are (1) that the toolbox commands and plots are seamlessly integrated with the other MATLAB toolboxes that deal with linear dynamic systems and (2) several new features and model objects. The toolbox now supports multi-input-multi-output (MIMO) systems across all model objects, and more emphasis is placed on continuous-time models. Also a new model object, idtf covers MIMO transfer function models in both continuous and discrete time. © 2012 IFAC.
Jia T.,MathWorks Inc. |
Duel-Hallen A.,North Carolina State University |
Hallen H.,North Carolina State University
IEEE Transactions on Vehicular Technology | Year: 2013
The long-range prediction (LRP) of fading signals enables adaptive transmission methods for rapidly varying mobile radio channels encountered in vehicular communications, but its performance is severely degraded by the additive noise and interference. A data-aided noise reduction (DANR) method is proposed to enhance the accuracy of fading prediction and to improve the spectral efficiency of adaptive modulation systems enabled by the LRP. The DANR includes an adaptive pilot transmission mechanism, robust noise reduction (NR), and decision-directed channel estimation. Due to improved prediction accuracy and low pilot rates, the DANR results in higher spectral efficiency than previously proposed NR techniques, which rely on oversampled pilots. It is also demonstrated that DANR-aided LRP increases the coding gain of adaptive trellis-coded modulation (ATCM). Finally, for low-to-medium signal-to-noise ratio (SNR) values, we show that LRP-enabled adaptive modulation performs better for realistic reflector configurations than for the conventional Jakes model (JM) data set. © 1967-2012 IEEE.
Mahapatra S.,MathWorks Inc.
AIAA Modeling and Simulation Technologies Conference 2011 | Year: 2011
The use of simulation studies to better understand the dynamic behavior of a system under investigation is at the core of verifying your designs early in the development process. Despite the amount of data that such studies produce, a 3D representation of the system creates a more complete understanding of system behavior. This paper describes the use of 3D animation in simulation-centric workflows to augment early verification activities, such as those used in Model-Based Design. The evolution of technology and domain specialization in the simulation and 3D graphics fields presents several challenges for using 3D animation in simulation-centric studies. A set of examples specific to MATLAB and Simulink environment illustrate how to meet these challenges. © 2011 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
Mahapatra S.,MathWorks Inc.
AIAA Modeling and Simulation Technologies Conference 2011 | Year: 2011
In recent times, the aircraft industry has witnessed an upward trend in development programs such as the Joint Strike Fighter (JSF) program requiring the customization of a single design to meet diverse customer requirements dictated by considerations such as application, cost, and operational considerations. Many of these dynamic changes in nature have required design component variations on top of a fixed master design. The concept of modularity applied intelligently to meet such needs has proved to be a cost-effective and efficient paradigm to meet these challenges. A well-designed modular architecture with shared interfaces and a common core enables synthesis of multiple product variants or product derivatives by switching in and out different components with varying attributes or implementations. However, such a paradigm when applied to a development process centered on Model-Based Design would necessitate tool support from design within the software environment to deployment on final hardware. In this paper, we introduce variant semantics and their usage within the graphical modeling environment such as Simulink. We discuss their nature that can be parametric and structural during the modeling phases but also are reflected in the automatically generated code for hardware deployment. Also, we introduce a scripting methodology for efficiently mapping a custom design to a permutation of variants and their subsequent abstraction for ease of understanding. Since several modular designs exist for any given design, we also outline a set of best practices for partitioning the design for scalability and maintainability. Variants present a variety of uses in the context of Model-Based Design workflows. They enable the creation of modular design platforms facilitating reuse and customization. Design exploration where several alternatives exist for a component can now be managed efficiently to simulate every design possibility in a combinatorial fashion for a given test suite. For large-scale problems, these could be distributed on a cluster of multicore computers for overall speedup with our scripting methodology. Alternatively, different test suites could also be mapped for efficiently managing relevant tests for a design. Maintenance activities of existing aircraft may require the upgrade of several components with no deterioration in existing performance requiring the testing of these upgrades in the model. Design elaboration and integration is a challenging activity where low fidelity components are replaced by more specialized ones. Since the order in which these components are integrated influence design quality and subsequent iterations, it is possible to carry out several separate integrations that increase confidence. Based on the evaluation criteria, a subset of these designs could be short listed for rapid prototyping or hardware-in-the-loop testing. With automatic code generation, variant components in the software model are mapped to C function code variants that can be switched by simply modifying the preprocessor definitions. Conversely, if there be hardware variants such as floating or fixed-point microprocessors, they will require the use of variants upstream with different modeling implementations. Using Simulink-based examples, we illustrate the above scenarios and outline strategies on how organizations can leverage these possibilities to reuse while enhancing their existing knowledge to meet the design challenges of the future. © 2011 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.
Bhavsar P.,Clemson University |
Chowdhury M.,Clemson University |
He Y.,MathWorks Inc. |
Rahman M.,Clemson University
Transportation Research Part C: Emerging Technologies | Year: 2014
This paper presents an integrated simulator "CUIntegration" to evaluate routing strategies based on energy and/or traffic measures of effectiveness for any Alternative Fuel Vehicles (AFVs). The CUIntegration can integrate vehicle models of conventional vehicles as well as AFVs developed with MATLAB-Simulink, and a roadway network model developed with traffic microscopic simulation software VISSIM. The architecture of this simulator is discussed in this paper along with a case study in which the simulator was utilized for evaluating a routing strategy for Plug-in Hybrid Electric Vehicles (PHEVs) and Electric Vehicles (EVs). The authors developed a route optimization algorithm to guide an AFV based on that AFV driver's choice, which included; finding a route with minimum (1) travel time, (2) energy consumption or (3) a combination of both. The Application Programming Interface (API) was developed using Visual Basic to simulate the vehicle models/algorithms developed in MATLAB and direct vehicles in a roadway network model developed in VISSIM accordingly. The case study included a section of Interstate 83 in Baltimore, Maryland, which was modeled, calibrated and validated. The authors considered a worst-case scenario with an incident on the main route blocking all lanes for 30. min. The PHEVs and EVs were represented by integrating the MATLAB-Simulink vehicle models with the traffic simulator. The CUIntegration successfully combined vehicle models with a roadway traffic network model to support a routing strategy for PHEVs and EVs. Simulation experiments with CUIntegration revealed that routing of PHEVs resulted in cost savings of about 29% when optimized for the energy consumption, and for the same optimization objective, routing of EVs resulted in about 64% savings. © 2014 Elsevier Ltd.
MathWorks Inc. | Date: 2015-05-27
A method may include causing a first model to be executed. The causing the first model to be executed may be performed by a device. The method may further include causing a second model to be executed to simulate a functionality of the first model. The causing the second model to be executed may be performed by the device. The method may further include interacting with a model element, of the second model, associated with implicitly accessing information regarding a state of the first model. The state may be a representation of the first model at a particular simulation time-step. The interacting with the model may be performed by the device. The method may further include accessing, by the model element, information associated with the state of the first model. The accessing the information may be performed by the device.