Electronic Military & Defense Annual Resource

5th Edition

Electronic Military & Defense magazine was developed for engineers, program managers, project managers, and those involved in the design and development of electronic and electro-optic systems for military, defense, and aerospace applications.

Issue link: https://electronicsmilitarydefense.epubxp.com/i/546858

Contents of this Issue

Navigation

Page 27 of 51

send information back, via RF, with fewer bandwidth require- ments or lag. An encryption may be added on the remote unit to ensure higher security at the expense of speed. Also, some image-processing applications need to annotate the video stream. Again, this inserts some lag, but it has realistic military applications, such as watermarking. Internet Of Things The Internet of Things (IoT) is attracting powerful com- mercial forces and is expanding remote sensor options for military repurposing. Within the IoT there are opposing forces: The remote sensor/camera wants to be cheap and disposable, which makes a case for centralized processing, but there are strong pressures for distributed processing. Hackable networks call for encryption. Bluetooth or other limited bandwidth infrastructures call for preprocessing to limit the amount of data transmitted. Array Processing Multisensor grids require high bandwidth to capture 100 percent of available information and centralized processing to reassemble that data into a usable data set. In the case of land-based systems, such as the Square Kilometer Array telescope, the centralized processing has to reconcile the cap- tured data with the earth's rotation while keeping focused on a remote object — nontrivial and power-hungry processing that lends itself to centralization. 3D Imaging Reconstruction 3D imaging reconstruction is both math- and power- hungry. For example, con- sider tomographic recon- struction, using either a moving sensor or discrete sensors generating multi- ple images of one object. (Think of the movie "The Matrix," when the camera pans around the hero mid- kick, suspended in air.) The reconstruction task is math-heavy and can be a power hog. It's largely repetitive, lending itself to massive parallelization in FPGA hardware. Neural Networks This is an emerging area in which the argument for pre-pro- cessing is uploading only the most useful data from the sen- sors (source: i-Abra). In this model, the sensor/camera has the capability to sort out information at the point of capture and fil- ter out all but the useful bits. This allows higher bandwidth for critical data, even to the point of rapid recognition, which can increase focus or bandwidth for an obtained target on the fly. The downside is that 100 percent of the original data never is transmitted and is therefore lost. The upside is that more of the system resources are deployed "looking" for exactly what you want. For instance, the cognitive neural network (CNN) might alter bandwidth on the fly, based on machine- to-machine interface (e.g., if artificial intelligence detects something of interest, it switches to higher resolution on that particular portion of the image), creating a better recording for human intelligence. Input resolution can be changed on the fly, as well, and still be processed through the neural network. Spectral Imaging A specialized variant of machine imaging, spectral imaging allocates more processing logic to a limited spectrum. In des- ert or space imaging, identifying an anomaly of interest does not require a full spectrum of colors. Efficacy actually can be enhanced by dedicating more FPGA resources to a nar- row spectrum and then dividing that limited range into finer increments. Companies like Resonon deploy this method for food sorting, but we believe it has military applications (e.g., identifying beige trucks in a beige desert). We've worked on algorithms that dedicate the largest part of the logic resources to detecting deltas and applying mod- erate frame rate capture to a limited-chromatic field to detect and identify changes that could represent threats. The key cri- terion is change, and the frame rate can be resolved around the maximum speed that an object of interest might achieve. Thus, logic resources can be largely dedicated to resolution and measurement of differential between frames (e.g., move- ment). Spectrum is relegated to a distant third priority. Spectral Analysis Spectral analysis using the Fast Fourier Transform (FFT) algo- rithm can be implemented to analyze the frequency content of a signal where an N-point FFT takes N clock cycles to compute. For example, a 1024-point FFT can be computed in 10.24 microseconds when the FPGA clock is 100 MHz. UAV Vision UAV vision actually is one of the more balanced applications of FPGAs. There needs to be a high enough frame rate to allow sufficient resolution when the UAV is flying at a par- ticular speed. Resolution may be adjusted if the terrain is less of an issue, as one might encounter with a mostly flat desert. Processing may be the "hog" in this application, particularly with stereoscopic vision, wherein much of the logic is used to reconcile or reassemble a single vision. Technology 28 Figure 2: The locus of processing can be driven by application type and/or cost tradeoffs. Figure 3: Tomography reconstructs data from multiple sensors into a single image. Electronic Military & Defense Annual Resource, 5th Edition

Articles in this issue

Archives of this issue

view archives of Electronic Military & Defense Annual Resource - 5th Edition