Method, system and computer program for radio emitter location determination

Abstract

The invention relates to a method (100) for radio emitter location determination. The method comprises selecting (S100) a set of potential radio emitter locations. The method also comprises, for each potential radio emitter location, obtaining S200 topographical image pixel data relating to a region covered by the radio emitter if placed at the potential radio emitter location. The method further comprises, for each pixel of the obtained topographical image pixel data, determining (S300) a value relating to radio signal attenuation. The method additionally comprises determining (S400) a radio emitter location from the selected set of potential radio emitter locations based on the determined values of the pixels of the obtained topographical image pixel data. The invention also relates to a system, a computer program product comprising instructions and a computer-readable recording medium storing instructions for implementing the method (100) for radio emitter location determination.

Description

[0001]METHOD, SYSTEM AND COMPUTER PROGRAM FOR RADIO EMITTER LOCATION DETERMINATION TECHNICAL FIELD The present disclosure relates to methods, systems and computer programs for radio emitter location determination. BACKGROUND ART Current methods for radio emitter location determination relies on mapping out a region with practical measurements and / or relying on simulations; either pure empirical models or empirical- analytical models. The so-called Longley-Rice model, a radio propagation model for predicting the attenuation of radio signals for a telecommunication link in the frequency range of 20 MHz to 20 GHz, is one of the most widely used models for exploring the choice of radio emitter location on radio signal attenuation. Other popular choices include the Irregular Terrain Model, ITM, and the more elaborate Irregular Terrain with Obstructions Model, ITWOM, the latter combining empirical data with Beer’s law and Snell’s law. Though practical measurements and the aforementioned models can provide a good basis for optimal locations for radio emitters, the time it takes to arrive at the optimal radio emitter location can be very long; too long if real-time decisions are needed. There is a need in the art for methods of radio emitter location determination that enable faster decisions without loss of accuracy. SUMMARY OF THE INVENTION An objective of the present disclosure is to provide methods, systems and computer programs for radio emitter location determination that enable faster predictions with respect to the state of the art while also providing comparable or improved prediction accuracy. It is also an objective of the present disclosure to provide methods and systems for radio emitter location determination that are able to incrementally improve its ability for accurate predictions as new data becomes available, either as real-world observations or from highly accurate but slower computer simulations. The present disclosure relates to a method for radio emitter location determination. The method comprises selecting a set of potential radio emitter locations. The method further comprises, for each potential radio emitter location, obtaining topographical image pixel data relating to a region covered by the radio emitter if placed at the potential radio emitter location. The method also comprises, for each pixel of the obtained topographical image pixel data, determining a value relating to radio signal attenuation. The method additionally comprises determining a radio emitter location from the selected set of potential radio emitter locations based on the determined values of the pixels of the obtained topographical image pixel data. By determining a value relating to radio signal attenuation for each pixel of the obtained topographical image pixel data, a radio signal attenuation profile for the radio emitter if placed at a specific location can be estimated with a high degree of confidence. Thus, a technical effect of the disclosed method is that the method enables a robust way of determining radio emitter location with respect to a desired criterion for radio emitter performance. Specifically, the method enables taking directionality and topographical information, such as vegetation and / or elevation, into account when determining the radio emitter location. In particular, the resulting radio signal attenuation map resulting from determining a value relating to radio signal attenuation for each pixel of topographical image data, can be used to e.g. assess both how efficiently the radio emitter would cover a certain region and how exposed the radio emitter might be to detection from a potential adversary trying to locate the radio emitter. According to some aspects, the step of determining a value relating to radio signal attenuation comprises, for each pixel of the obtained topographical image pixel data, determining, using a neural network, a class membership from a set of predetermined classes relating to radio signal attenuation, and the step of determining the radio emitter location is further based on the on the determined class memberships of the pixels of the obtained topographical image pixel data. Using a neural network to determine the value relating to radio signal attenuation has two technical effects and advantages; first, neural networks can operate in millisecond regimes, thereby enabling real-time determination of radio emitter location. Secondly, neural networks can improve their performance over time; as more predictions are made and compared to real- world data or high-accuracy simulations, the parameters of the neural network can be updated to improve performance. Thus, a method for radio emitter location determination that can improve performance over time is provided. According to some aspects, the step of determining a radio emitter location further comprises, for each potential radio emitter location, determining a value relating to an optimal radio emitter location based on the determined values relating to radio signal attenuation. Additionally, the step of determining the radio emitter location is further based on the determined values relating to an optimal radio emitter location. Determining a value relating to an optimal radio emitter location enables comparing the potential radio emitter location along a dimension represented by the determined value. By determining a value relating to an optimal radio emitter location, the step of determining a radio emitter location can be automated. Additionally, depending on how the value is determined, the determination may be performed in real-time. Thus, the method may be used in real-time applications where e.g. path planning is based on radio signal attenuation; path planning maximizing radio signal coverage and / or stealth is enabled. According to some aspects, the topographical image pixel data comprises height and / or terrain data relating to the region covered by the radio emitter if placed at the potential radio emitter location. Height and terrain, such as vegetation and buildings, can have a huge impact on how radio signals propagate and thus what the resulting radio signal attenuation will look like. By comprising height and / or terrain data into the topographical image pixel data, the height and / or terrain data can act synergistically with a neural network, such as a UNet, when predicting radio signal attenuation. Neural network architectures that look at pixel information on multiple scales, such as UNets, can thus integrate height and / or terrain data across the region of interest in a fast and flexible manner, thereby providing highly accurate predictions of radio signal attenuation. According to some aspects, the radio emitter is a radar emitter configured to emit radar signals. The present disclosure also relates to a system for radio emitter location determination, the system comprising computational circuitry configured to carry out the method as described above and below. The present disclosure further relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method as described above and below. The present disclosure additionally relates to a computer-readable recording medium storing instructions which, when executed by a computer, cause the computer to carry out the method as described above and below. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 illustrates aspects of the proposed method for radio emitter location determination; and Figure 2 illustrates aspects of the proposed system for radio emitter location determination. DETAILED DESCRIPTION Figure 1 illustrates a method 100 for radio emitter location determination according to some aspects of the present disclosure. The method comprises selecting S100 a set of potential radio emitter locations. According to some aspects, the set of potential radio emitter locations are selected S100 by sampling random locations within a predetermined region. According to some aspects, the set of potential radio emitter locations are selected S100 by user input. This allows the selection to be based on human experience and domain knowledge. If performed in relation to a system for radio emitter location determination, as described above and below, the step of selecting S100 is guided by user input, e.g. via a user interface or as provided computer program code. If the radio emitter strength is known or approximated, an effective radio emitter transmission range can be determined. The transmission range in combination with transmission directions will thus relate to a region covered by the radio emitter if placed at a potential radio emitter location. If the radio transmitter is configured to transmit radio signals with equal strength in all directions, the region covered by the radio emitter if placed at the potential radio emitter location will have a circular shape with the center at the potential radio emitter location. According to some aspects, the step of selecting S100 a set of potential radio emitter locations is based on the region covered by the radio emitter if placed at the potential radio emitter location. According to some aspects, the set of potential radio emitter locations comprises a grid of potential radio emitter locations. According to some aspects, a packing algorithm, for instance a circle packing algorithm, is used to select S100 the set of potential radio emitter locations. By selecting S100 the set of potential radio emitter locations based on a predetermined grid-like pattern, good radio signal coverage exploration can be ensured without spending time or much computational resources on potential radio emitter location selection, which may be critical for real-time applications. The method further comprises, for each potential radio emitter location, obtaining S200 topographical image pixel data relating to a region covered by the radio emitter if placed at the potential radio emitter location. According to some aspects, the topographical image pixel data comprises height and / or terrain data relating to the region covered by the radio emitter if placed at the potential radio emitter location. Obtaining topographical data in the form of image pixel data enables the use of image processing methods, such as the use of convolutional neural networks, CNNs, that enable rapid model inference, typically in the milli-second regime. The use of image data thus allows neural network models to be trained on real-world data or model predictions from slower models, such as those of the prior art. Height data in the form of pixels allows a model to predict radio signal propagation obstacles as well as scattering. Terrain data, such as vegetation and housing, can be of great importance when predicting radio signal attenuation. Thus, by including height and / or terrain data, higher accuracy of predicting radio signal attenuation and scattering can be achieved. The method also comprises, for each pixel of the obtained topographical image pixel data, determining S300 a value relating to radio signal attenuation. According to some aspects, the determination S300 is performed using a neural network. The neural network may be any neural network suitable for processing image pixel data, such as a convolutional neural network, CNN, a graph convolutional neural network, GCNN, or an attention-based neural network, for instance a transformer-based neural network. According to some aspects, the step of determining S300 a value relating to radio signal attenuation comprises, for each pixel of the obtained topographical image pixel data, determining S310, using a neural network, a class membership from a set of predetermined classes relating to radio signal attenuation. Determining S310, using a neural network, a class membership from a set of predetermined classes, also known as semantic segmentation, has a set of advantages when applied to radio signal attenuation prediction. Using semantic segmentation enables outlier prevention, e.g. when classifying pixel membership to a set of radio signal attenuation intervals, as described further below. Thus, the use of semantic segmentation enables more stability in training the neural network and more robust inferences. A further advantage is that the inferred class data from the neural network can be made very compact. For instance, class membership for a set of sixteen classes only requires four bits per pixel, compared to a typical eight-bit to thirty-two-bit representation for the red, green and blue channels common to regular image data. The reduced data further facilitates real-time application of the disclosed method. According to some further aspects, the set of predetermined classes comprises a set of classes relating to radio signal attenuation intervals. For instance, the set of classes relating to radio signal attenuation intervals may comprise sixteen classes, each class covering a 10 dBm, decibel-milliWatts, decremental interval, ranging from attenuation of 0 dBm to – 150 dBm. The method additionally comprises determining S400 a radio emitter location from the selected set of potential radio emitter locations based on the determined values of the pixels of the obtained topographical image pixel data. The determined values of the pixels can be further processed before the radio emitter location is determined S400. Since the determined values of the pixels can be represented as pixel data, any method that is suitable for making decisions based on pixel data can be used to determine S400 the radio emitter location. According to some aspects, the determination can be made manually by a user, e.g. via a user interface. According to some aspects, the step of determining S400 a radio emitter location is performed by a neural network. It may be convenient to rank the potential radio emitter locations along some metric to facilitate the determination S400 of the radio emitter location. According to some aspects, the method comprises performing a summation based on the determined S300 values relating to radio signal attenuation. If the values relating to radio signal attenuation directly reflects an estimated measure of radio signal attenuation, adding all the values relating to radio signal attenuation thus provides a total measure of how well a region around a potential radio emitter location is covered by the radio emitter; the lower the total measure, the better the radio signal coverage. Any summation may be performed as weighted summations in order to reflect priorities in radio emitter location. According to some aspects, the step of determining S400 the radio emitter location is performed by a user. This allows the determination to be based on human experience and domain knowledge. If performed in relation to a system for radio emitter location determination, as described above and below, the step of determining S400 the radio emitter location is guided by user input, e.g. via a user interface or as provided computer program code. According to some aspects, the radio emitter is a radar emitter configured to emit radar signals. In the context of military applications, e.g. when the radio emitter comprises a radar signal emitter and the radio signals relate to radar signals, examples of priorities may include surveilling a range of radio signal directions and / or positioning the radio emitter as close as possible or as far away as possible from a region of interest. Another example of priority is to reduce the probability of the radio signal being detected. Thus, the weighted summation may be performed to minimize the radio signal attenuation within a range of desired radio signal directions. The weighted summation may be performed to choose a location within a desired distance with respect to a region of interest. The weighted summation may be performed to estimate a probability of the radio signal being detected. The estimation of the probability of the radio signal being detected at one particular radio emitter location may be based on radio signal attenuation estimations for a set of other potential locations. The priorities may themselves be weighted amongst each other, e.g. by performing, for each potential radio emitter location and each priority, a first set of weighted summations and performing a second set of weighted summations based on said first set of weighted summations. In summary, according to some aspects, the step of determining S400 a radio emitter location further comprises, for each potential radio emitter location, determining S410 a value relating to an optimal radio emitter location based on the determined S300 values relating to radio signal attenuation, and wherein the step of determining S400 the radio emitter location is further based on the determined S410 values relating to an optimal radio emitter location. According to some aspects, the step of determining S400 the radio emitter location is further based on the on the determined class memberships of the pixels of the obtained topographical image pixel data. By using semantic segmentation as a basis for determining the radio emitter location, a number of technical effects and advantages are provided. Semantic segmentation prevents the potential for outliers. Semantic segmentation further allows confidence estimates for each value relating to radio signal attenuation, and consequently enables a confidence estimate for the whole region effectively covered by the radio emitter if placed at the respective potential radio emitter location. Thus, semantic segmentation enables confidence estimates for each potential radio emitter location. According to some aspects, the step of determining S400 the radio emitter location further comprises, for each potential radio emitter location, determining a confidence estimate relating to the determined S300 values relating to radio signal attenuation, and wherein the step of determining S400 the radio emitter location is further based on the determined confidence estimates. Semantic segmentation further provides a degree of robustness, since each class of the semantic segmentation may absorb some of the variation in radio signal attenuation estimation that is typically associated with any model. Additionally, semantic segmentation enables a data efficient representation of radio signal attenuation data. For instance, with respect to the example above, with radio signal attenuation being mapped to sixteen different signal attenuation intervals, each value relating to radio signal attenuation can be represented by only four bits. This is in contrast to image pixel data which is typically represented by eight to thirty-two bits, in one to three channels. The reduction in the amount of data needed to represent radio signal attenuation can significantly improve real-time performance in a real-time scenario, both in terms of processing time as well as amount of data that may be communicated between actors operating in the real-time scenario. Figure 2 illustrates aspects of the proposed system 265 for radio emitter location determination. The system is configured to select S100 a set of potential radio emitter locations 220a, 220b. According to some aspects the system selection is guided by user input, e.g. via a user interface or as provided computer program code. Consider a region of interest 210 for which one wants to place a radio emitter at some location with respect to the region of interest that is optimal with respect to a specific task. According to some aspects, the system 265 is configured to receive geographical data comprising the region of interest 210, and wherein the system 265 is further configured to receive user instructions on how to select S100 the set of potential radio emitter location 220a, 220b. The user instructions may comprise the set of potential radio emitter locations to be selected. According to some aspects, the user instructions comprises instructions, e.g. as computer program code, which, when executed, causes the system to select S100 a set of potential radio emitter locations 220a, 220b based on any of the methods discussed in relation to Figure 1, above. In the illustrated example, the potential radio emitter locations 220a, 220b are all placed in a grid and within the region of interest 210; this is merely for illustrative purposes. There are no limitations on the placement of the selected potential radio emitter locations. One or more of the potential radio emitter locations 220a, 220b may be selected to lie outside the region of interest 210. The various ways of selecting potential radio emitter locations discussed in relation to Figure 1 applies here as well. The system 265 is further configured to, for each potential radio emitter location, obtaining S200 topographical image pixel data relating to a region 230a, 230b covered by the radio emitter if placed at the potential radio emitter location 220a, 220b. According to some preferred aspects, the obtained topographical image pixel data covers a square or rectangular region 240a, 240b comprising the topographical image pixel data relating to the region 230a, 230b covered by the radio emitter if placed at the potential radio emitter location 220a, 220b. This facilitates processing the topographical image pixel data 230a, 230b by a convolutional neural network. A further advantage is that the topographical image data is often extracted or sampled from a large data set of topographical data that is configured to have its data sampled in rows and columns for maximum efficiency. Thus, the choice to obtain S200 the topographical image pixel data as comprised in a square or rectangular image act synergistically to both enable rapid data extraction while also enabling rapid data processing, by a neural network, of the obtained topographical image pixel data. According to some aspects, the topographical image pixel data comprises height and / or terrain data relating to the region covered by the radio emitter if placed at the potential radio emitter location. According to some aspects, the radio emitter is a radar emitter configured to emit radar signals. According to some aspects, the topographical image pixel data is provided by a user, e.g. via an interface 263 of the system 265. According to some aspects, the system comprises a persistent memory 264, herein also referred to as a computer-readable recording medium, i.e. a memory storage that retains data after power provided to the persistent memory has been shut off, such as a hard drive, configured to store the topographical image data. According to some aspects, the system is configured to obtaining S200 topographical image pixel data from an external data base comprising the topographical image pixel data. The system 265 is also configured to, for each pixel of the obtained topographical image pixel data, determine S300 a value relating to radio signal attenuation. According to some aspects, the system comprises a neural network. According to some preferred aspects, the neural network comprises a UNet. The UNet 270 comprises a downsampling path 271 and an upsampling path 272. Downsampling and upsampling refer to sampling processes that changes the spatial dimensions of image data. For instance, downsampling may be performed by interpolation methods, such as bilinear, trilinear or nearest neighbor interpolation, pooling methods, such as min, max or average pooling, or convolutions with strides greater than one pixel. Likewise, upsampling may be performed by interpolation methods, bilinear, trilinear or nearest neighbor interpolation, or transposed convolution. The downsampling path 271 comprises a first set of convolution layers 250a-d configured to perform a convolution on a set of first feature maps 240a-b, 255a-c and output a second set of feature maps 255a-d. The downsampling path 271 is further configured to successively downsample 255a’-c’ the second set of feature maps 255a-c of all but one 250d of the first set of convolution layers 250a-d and provide the downsampled feature map 255a’-c’ as an input, i.e. a first feature map, to a downstream convolution layer 250b-d. According to some aspects, the downsampling path 271 is further configured to process at least one feature map of the second set of feature maps by a second set of convolution layers 250e-g. The UNet is configured to transfer the second set of feature maps 255a-d to the upsampling path 272. The upsampling path 272 comprises a third set of convolution layers 260a-d. Each convolution layer of the third convolution layer is configured to receive a feature map having the same spatial dimensions as a corresponding feature map from the second set of feature maps 255a-d. The upsampling path 272 is further configured to generate concatenated feature maps 255e, 255g, 255i, by concatenating feature maps from the second set of feature maps 255a-c and the upsampled feature maps output from the third set of convolution layers that have matching spatial dimensions. One of the main technical effects and advantages is the ability of the UNet to integrate information across multiple spatial scales. By identifying relevant information in the topographical image pixel data at each spatial scale, physical phenomenon relating to topography across the region 230a, 230b covered by the radio emitter if placed at the potential radio emitter location 220a, 220b can be captured effectively. The feature maps 255j output by the convolution layer 260a of the third set of convolution layers having the largest spatial dimensions comprises the determined S300 values relating to radio signal attenuation 230b’. Apart from the architecture, the neural network obtains its functionality from a set of parameters, often called weights and biases. Training refers to adjusting said parameters based on a target behavior. In the present disclosure, the target behavior comprises making accurate inferences, i.e. given an input 240b, generating an output 255j that comprises inferred data 230b’ that approximate desired target data 234, as described above and below. During training the determined S300 values relating to radio signal attenuation is compared 280 to corresponding target values 234. The target values 234 could come from experimental data and / or simulations, such as the models of the prior art. According to some aspects, when determining S300 a value relating to radio signal attenuation, the system 265 is further configured to, for each pixel of the obtained topographical image pixel data, determining S310, using the neural network, a class membership 232a-d from a set of predetermined classes relating to radio signal attenuation, and to base the determination of the radio emitter location on the determined class memberships 232-ab of the pixels of the obtained topographical image pixel data. During training, the predicted class memberships 232a-b are compared to ground truth class memberships 232c-d. Another technical effect and advantage of using a UNet is that a UNet enables the use of a pretrained neural network, e.g. a neural network trained to extract features from images for classification purposes, such as ResNet-34, for the downsampling path, and rely on so-called transfer learning to train the rest of the UNet. Transfer learning means that one uses parts of a first neural network that has been trained for a first task as part of a second neural network that is then trained for a second task. Transfer learning can significantly speed up the training process and may reduce the amount of training data needed to reach a desired performance of the neural network. The latter is particularly important, since the target data used for supervised training may require time-consuming and costly experimental data collection and / or simulations; by using a neural network architecture that enables transfer learning, significant reductions in time and cost to configure the system 265 may be achieved. The system 265 is additionally configured to determine S400 a radio emitter location from the selected set of potential radio emitter locations based on the determined values of the pixels of the obtained topographical image pixel data. According to some aspects, the system is configured to provide the determined S300 values relating to radio signal attenuation to a user, e.g. via a user interface, such as a graphical user interface. According to some further aspects, the system is further configured to receive input from the user, the input comprising instructions which, when executed by the system, causes the system to determine S400 the radio emitter location from the selected set of potential radio emitter locations. According to some aspects, the system is configured to, for each potential radio emitter location, determine S410 a value relating to an optimal radio emitter location based on the determined S300 values relating to radio signal attenuation, and wherein the system is further configured to base the determination S400 of the radio emitter location on the determined S410 values relating to an optimal radio emitter location. According to some aspects, the system 265 comprises computational circuitry configured to perform the above-mentioned steps. According to some further aspects, the computational circuitry comprises a neuromorphic chip configured to perform at least one of the steps performed by the neural network 270. Neuromorphic chips are typically very energy efficient and allow the network function it is configured to perform to be updated with further training. According to some aspects, the system 265 comprises a computer-readable recording medium storing instructions, the instructions comprising the parameters of the neural network 270 and / or computer program code which, when executed, causes the system 265 to carry out any of the above-mentioned method steps. According to some aspects, the computer-readable recording medium comprises a hard drive. According to some aspects, the system 265 comprises a central processing unit 266, CPU, configured to execute computer program code, which, when executed, causes the system 265 to carry out any of the above-mentioned steps. According to some aspects, the system 265 comprises a graphics processing unit 267, GPU. The GPU enables the system 265 to carry out most or all of the operations of the neural network in parallel, thereby greatly improving the speed at which the system 265 is able to carry out the disclosed method. According to some aspects, the system also comprises a random-access memory 268, RAM. The RAM 268 enables the system 265 to load computer program code for the CPU 266, as well as facilitating data transfer between the CPU 266 and the GPU 267. The present disclosure also relates to a computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method as disclosed above and below. The present disclosure further relates to a computer-readable recording medium storing instructions which, when executed by a computer, cause the computer to carry out the steps of the method as disclosed above and below.

Claims

CLAIMS 1. A method (100, 200) for radio emitter location determination, the method comprising - selecting (S100) a set of potential radio emitter locations, - for each potential radio emitter location, obtaining (S200) topographical image pixel data relating to a region covered by the radio emitter if placed at the potential radio emitter location, - for each pixel of the obtained topographical image pixel data, determining (S300) a value relating to radio signal attenuation, - determining (S400) a radio emitter location from the selected set of potential radio emitter locations based on the determined values of the pixels of the obtained topographical image pixel data, wherein the step of determining (S300) a value relating to radio signal attenuation comprises - for each pixel of the obtained topographical image pixel data, determining (S310), using a neural network, a class membership from a set of predetermined classes relating to radio signal attenuation, and wherein the step of determining (S400) the radio emitter location is further based on the on the determined class memberships of the pixels of the obtained topographical image pixel data.

2. The method according to claim 1, wherein the step of determining (S400) a radio emitter location further comprises - for each potential radio emitter location, determining (S410) a value relating to an optimal radio emitter location based on the determined (S300) values relating to radio signal attenuation, and wherein the step of determining (S400) the radio emitter location is further based on the determined (S410) values relating to an optimal radio emitter location.

3. The method according to claim 1 or 2, wherein the topographical image pixel data comprises height and / or terrain data relating to the region covered by the radio emitter if placed at the potential radio emitter location.

4. The method according to any of the preceding claims, wherein the radio emitter is a radar emitter configured to emit radar signals.

5. A system (265) for radio emitter location determination, the system comprising computational circuitry configured to carry out the method according to any of claims 1- 4.

6. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any of claims 1-4.

7. A computer-readable recording medium storing instructions which, when executed by a computer, cause the computer to carry out the method according to any of claims 1-4.

Images