Field

The present invention relates to a method, a system and a computer program product related to wireless communication network antennas. More particularly, the invention relates to a method, a system, and a computer program product for extracting antenna radiation pattern of an antenna in its normal operation environment using sparse field measurements.

Background

Mobile radio networks, such as 3G and 5G cellular networks, consists of many different network elements. Air interface between a User Equipment (UE) and the mobile radio network is provided by radiating antennas. In modern mobile radio networks, antennas are typically directional antennas, but in practice the antennas can be of any type, from omnidirectional antennas to massive multiple-input-multiple-output (MIMO) active antennas.

Radiation pattern of an antenna is always carefully designed and verified by antenna manufactures so that in the cellular network the capacity and quality are maximally utilized.

Calculated patterns are ideal, and the actual radiation pattern can be verified in a radio frequency (RF) laboratory environment. However, there are many factors which cause variation on the actual radiation pattern of the antenna in its operating environment. For example, manufacturing, manual handling during logistics and installation, aging and other environmental threats change the radiation pattern. In the actual operation environment, there may also be physical obstacles in near-field proximity of the antenna, which cause distortion of the antenna radiation pattern.

Therefore, a method to determine radiation pattern of the antenna in its actual operating environment is needed.

Description of the related art

Patent application WO2013123496 Al discloses a method for antenna pattern estimation that extrapolates patterns from a small number of far- field measurements. A recursive method is used for obtaining the final estimate that is close enough to the actual radiation pattern. A priori knowledge of the antenna's design is used to solve the inverse source problem.

Patent application US2020003817 AA is one of many publications which disclose an UAV-based system for measurement and characterization of antenna in its normal operation environment.

Patent application CN111523568 A discloses using deep neural network and radiation data compensation for antenna array fault diagnosis. Radiation data is obtained at a plurality of measurement points in far-field region using various predetermined fault scenes. Then the network is trained to recognize these faults. Method reduces amount of data required to be collected in the training stage.

Although various methods have been proposed for determining antenna radiation pattern, these typically require information on design of the antenna and/or number of measurements performed in the far-field of the antenna and/or feeding the antenna with specific feeding schemes. Thus, a method is needed, which enables determining radiation pattern of the antenna in its normal operating environment. Summary

An object is to provide a method and apparatus so as to solve the problem of determining radiation pattern of an antenna in its operating environment. The objects of the present invention are achieved with a method according to the claim 1. The objects of the present invention are further achieved with computer program product according to the claim 11.

The preferred embodiments of the invention are disclosed in the dependent claims.

According to a first aspect, a method for extracting antenna radiation pattern of an antenna under investigation is provided. The method comprises obtaining field measurement data by field measurements of received far-field signals radiated by the antenna under investigation in its operating environment. The field measurement data represents a plurality of measurement points, corresponding a portion of cells of a grid covering a geographical area served by the antenna under investigation. The method comprises using a deep neural network (DNN) for training a DNN model. The DNN model is taught, using the field measurement data as a target, to determine a pathloss of the signal between the antenna under investigation and each respective measurement point, wherein the DNN model determines pathloss on basis of at least the field measurement data, geographical location of the antenna under investigation, azimuth direction from each respective measurement point to the antenna or from the antenna to each respective measurement point, and geographical data of the operating environment. The method further comprises replacing the geographical data with a flat environment model for generating, by the trained DNN model, a two-dimensional radiation pattern of the antenna under investigation, wherein the trained DNN model determines pathloss at each cell of the grid using the flat environment model. The information representing the two-dimensional radiation pattern may be stored in a memory and/or displayed on a display.

According to a second aspect, the geographical data of the operating environment comprises building information, vegetation type and height information and topographical information, and wherein the flat environment model lacks such building information, vegetation type and height information and topographical information of the operating environment.

According to a third aspect, the field measurement data comprises information on geographical location of each measurement point, and at least one characteristic of at least one signal received wirelessly at the respective measurement point from at least one antenna by a mobile measurement apparatus, wherein the at least one antenna comprises the antenna under investigation. The mobile measurement apparatus performing the field measurements has been configured to be locked on a specific frequency, to be locked on a specific physical cell id (PCI) of a cellular network, or to scan a plurality of signals in a scanning mode. The DNN model is provided with field measurement data concerning signals received from the antenna under investigation, but not concerning signals received from any other antennas.

According to a fourth aspect, the field measurement data comprises geographical location of the respective measurement point, and at least one of a Reference Signal Received Power (RSRP), a Synchronization Signal Reference Signal Received Power (SS-RSRP) and a received signal level (Rx-level) and optionally a Received Signal Strength Indicator (RSSI), at each respective measurement point. According to a fifth aspect, the antenna under investigation is a directional antenna, and the field measurement data comprises data concerning measurement points within the approximate main lobe area of the antenna under investigation, or the antenna under investigation is an omnidirectional antenna, and the field measurement data comprises data concerning measurement points at any direction about the antenna under investigation.

According to a sixth aspect, the method further comprises generating a coverage map of the antenna under investigation in its operation environment using the trained DNN model for determining pathlosses at each cell of the grid.

According to a seventh aspect, the DNN model comprises a plurality of geographical data layers. The method comprises, for each geographical data layer, generating a plurality of data arrays, each comprising data pixels concerning projections of lines between the antenna under investigation and respective measurement points on the respective geographical data layer, data concerning geographical coordinates of the antenna, azimuth direction from the antenna to the respective measurement points or from the respective measurement points to the antenna, and geographical coordinates of the measurement points, and training the DNN model using said plurality of data arrays corresponding to each of the plurality of geographical data layers.

According to an eighth aspect, radio propagation model used by the DNN model is further enhanced by providing each of the plurality of data arrays with additional radio environment data. The additional radio environment data concerns objects near each respective measurement point and said concatenating additional radio environment data with the respective data array concerning the respective measurement point. The additional radio environment data comprises height information of surrounding buildings, natural barriers and/or reflective surfaces near each respective measurement point.

According to a ninth aspect, the DNN model is taught in batches, using the supervised learning method and using said plurality of data arrays as input and said measurement points as target.

According to a tenth aspect, in each cell in the grid represents one or more data pixels.

According to another aspect, a computer program product is provided, comprising computer executable code which, when performed by a computing device or system, performs the method according to any of the above aspects.

According to a further aspect, a non-transitory computer readable medium is provided, comprising computer executable instructions which, when executed by a computing device or system, causes the computing device or system to perform a method according to any of the first to tenth aspect.

The present invention is based on the idea of teaching a deep neural network (DNN) with sparse field measurements to develop a DNN model of the environment of the antenna, and to utilize this DNN model to determine a two-dimensional radiation pattern of the antenna by replacing geographical data from the DNN model with a flat environment.

The present invention has the advantage that it enables determining the two-dimensional antenna radiation pattern with sufficient accuracy to detect faults in the antennas due to various reasons, including but not limited to mechanical faults and aging, as well as unexpected distortions in the radiation pattern caused by physical obstacles in the near-field of the antenna. The invention requires only a limited number of field measurements, thus saving resources and cost.

Brief description of the drawings

In the following the invention will be described in greater detail, in connection with preferred embodiments, with reference to the attached drawings, in which

Figure 1 shows a map of a geographical area.

Figure 2 shows a coverage map of an antenna.

Figure 3 shows a radiation pattern of the antenna.

Figure 4 shows a map of another geographical area.

Figure 5 shows a coverage map of an antenna.

Figure 6 shows a radiation pattern of the antenna.

Figure 7 illustrates main steps of a method.

Detailed description

The figure 1 shows a map (10) of a geographical area. This example illustrates a hilly rural area with fields, forests, a lake and some houses. Geographical positions of two base stations (100a, 100b) are marked on the map. Each base station (100a, 100b) comprises one or more antennas. The antennas may be part of a 3G, 4G or 5G cellular network infrastructure, providing wireless network connectivity to mobile phones in the area.

The map (10) shows some roads, along which a mobile measurement apparatus is transported on a vehicle. During this transport, the mobile measurement apparatus performs field measurements, detecting far-field radio signals emitted by at least one antenna and received by the mobile measurement apparatus. The mobile measurement apparatus performs field measurements of received radio signals in a plurality of measurement points (120) marked as black dots on the map. The measurement points (120) are sparse in comparison to the whole assumed coverage area of an antenna under investigation (101).

During the field measurement, the mobile measurement apparatus preferably stores in memory the field measurement data concerning the received radio signal as well as respective geographical location of each measurement point. The mobile measurement apparatus may be enabled to send obtained field measurement data wirelessly over wireless data connections.

Figure 1 also shows an arrow (102), which illustrates actual physical direction to which this specific, directional antenna under investigation (101) points. This information was used as reference while testing the method, but it is not part of the measurement data.

For distinguishing received radio signals among all possible radio signals received not only from the antenna under investigation (101), but also other antennas of the same (100a) or other base stations (100b) in the area, the mobile measurement apparatus is preferably, as known in the art, capable of recognizing signals from different antennas and, if so needed, recording information on each measurement point that enables distinguishing between signals received from different antennas. The mobile measurement device may be locked on a specific frequency so that it only records data regarding radio signals with that specific frequency, and/or the mobile measurement device may be locked on a specific physical cell id (PCI) of a cellular network to record data concerning that PCI only. The mobile measurement device may also scan a plurality of signals in a scanning mode, while storing information on the respective PCI or frequency with each piece of field measurement data stored at each measurement point. This way, data obtained by the field measurement may be utilized later for investigating more than one antenna. As a result, data concerning measurement points regarding the antenna under investigation may be distinguished among all field measurement data obtained by the mobile measurement apparatus.

In a typical setting, the field measurement data comprises at least one of a Reference Signal Received Power (RSRP) of LTE, a Synchronization Signal Reference Signal Received Power (SS-RSRP) of 5G, a received signal level (Rx-level) of GSM and 3G, and additionally a Received Signal Strength Indicator (RSSI) at each respective measurement point, but other characteristics may be stored as well as known in the art.

For determining antenna radiation pattern, information regarding strength of the signal received by the mobile measurement apparatus in measurement points is important.

Stored field measurement data is used by Deep Neural Network (DNN) for performing supervised learning. In this process, the DNN is provided with field measurement data on measurement points concerning the antenna under investigation (101). The DNN is also provided with information regarding geographical location of the antenna under investigation, azimuth direction from respective measurement points to the antenna or vice versa, and geographical data of the operating environment of the antenna under investigation. With azimuth direction we refer to an angle direction in the horizontal plane as known in the art, equivalent to angle directions read from a compass. Geographical data of the operating environment provided for the DNN preferably comprises building information, vegetation type and height information and topographical information, representing the forms and features of land surfaces on the area, all of which are known to have effects on propagation of radio signals. Typically, geographical data is represented as a plurality of layers, each layer comprising specific type of information. For example, geographical data of any selected area may comprise layers for orthoimagery, topographical information (elevation above sea level), transportation (roads, railroads etc.), geographic names, vegetation (i.e. average height of vegetation, type of vegetation), hydrography, boundaries, structures (buildings, masts, bridges etc.) and land cover. Not all these layers are relevant for radio propagation, but only those layers which comprise relevant geographic data for developing the DNN model are used in this method.

The DNN performs supervised learning using field measurement data as target and information on geographical location of the antenna under investigation, azimuth direction from each respective measurement point to the antenna or vice versa, and geographical data of the operating environment to develop a DNN model for radio propagation of signals emitted by the antenna under investigation that matches with the target information provided by the measurement data points. In a typical setting, the resulting DNN model comprises a grid of cells with equal size that covers geographical area around the antenna under investigation. Each measurement point belongs to a cell in the grid. The purpose of the DNN in this application is to approximate the function of radio propagation for the transmitting antenna in its normal operating environment.

The trained DNN model may then be used for generating a coverage map (20), as shown in the figure 2, for the antenna under investigation (101). The DNN model utilizes geographical data of the operating environment in cells of the grid not provided with field measurement results to determine received signal strength from the antenna under investigation. The coverage map shows determined received signal strength in each cell of the grid, which represents a set size geographical area. Actual calculations of the DNN model may be made on grid with smaller cell size than what is shown in the coverage map. For example, the coverage map (20) shown in the figure 2 shows cells, that represent 10x10 pixels (data points) of the actual DNN model. For illustration purposes, the coverage map typically represents a plurality of ranges of received signal strength as different colors shown as an overlay on a map. In this example, different colors represent Reference Signal Received Power (RSRP) of LTE1800. Likewise, any other suitable characteristics of the received signal may be measured and presented on the map.

Above illustrated method of generating a coverage map is a known application of supervised DNN learning.

The inventor has surprisingly found out, that the DNN model trained for developing the coverage map can further be used for determining an image of actual, two-dimensional radiation pattern of the antenna under investigation in its normal operating environment. This can be achieved by removing geographical data from the DNN model by replacing it with a flat environment model. The flat environment model lacks any information on building information, vegetation type and height information and topographical information. In the specific exemplary implementation, only a single piece of height information, namely altitude from the sea level at the foot of the mast on which the antenna under investigation is installed was used as the flat environment model, while any other information on the environment was removed. The figure 3 shows an example of a two-dimensional radiation pattern (300) of the antenna obtained on basis of the same DNN model that was used for generating the coverage map of the Figure 3. It should be highlighted, that this result has been achieved without providing the DNN model with any information regarding the type or construction of the antenna under investigation, but the resulting radiation pattern is easily recognizable as an azimuth, in other words a horizontal radiation pattern of a typical flat panel antenna used in cellular radio systems.

The surprising finding in this two-dimensional horizontal radiation pattern is, that the direction of the main lobe, towards west (left in this figure), is not same as the actual, physical orientation of the antenna, indicated by the white arrow, would suggest. This kind of deviation may occur for example due to obstacles in the near field of the antenna in its operation environment, which cause distortion of the radiation pattern, or due to mechanical or electrical problems or misconfiguration of the antenna itself and/or its electrical or physical down tilt.

Figure 4 shows a map (10) of another exemplary geographical area, in this case an urban area with more dense population and buildings. For increased cellular network coverage, there are three base stations (100a, 100b, 100c) in this area, each having at least one antenna. In an actual field measurement, the mobile measurement apparatus was transported within the area served by the antenna under investigation (101) at base station 100a, performing measurement at sparse measurement points (120) illustrates with black dots on the maps. The DNN model was trained using the field measurement data obtained.

The figure 5 shows a coverage map (20) of the antenna under investigation (101) generated using the DNN model trained using the field measurement data. In addition to marking geographical location of the bases station 101a and/or the antenna under investigation (101), an arrow (102) pointing out from the antenna (101) is marked on the coverage map, which illustrates actual physical direction to which this specific, directional antenna under investigation (101) points. This information was not provided to the DNN model at any phase but was used for testing the method and its capabilities.

The figure 6 shows a modified coverage map (30) with the two- dimensional horizontal antenna radiation pattern (300) generated for the antenna under investigation (101) by replacing the geographical data with the flat environment model. The two-dimensional antenna radiation pattern (300) is shown as light pattern on the otherwise dark modified coverage map (30). The radiation pattern (300) can be recognized as a typical radiation pattern of a directional antenna (101), with a strong main lobe towards right (east), with two side lobes approximately to up (north) and down (south) and a back lobe to the left (west). The surprising finding in this result is, however, that the direction of the main lobe is not same as the actual, physical orientation of the antenna, indicated by the white arrow, would suggest. This kind of deviation may occur for example due to obstacles in the near field of the antenna in its operation environment, which cause distortion of the radiation pattern. For example, such nearfield obstacle may be a guy wire of the mast. It may have been initially intended, that the main lobe of the antenna under investigation (101) in this example should be directed more towards to south-east rather than right towards east, but in practice this was not achieved due to the obstacle in the near field of the installed antenna. Upon detecting the mismatch between the actual installation and the intended result, a corrective action may be initiated. Instead or in addition to obstacles in the near field, deviation of the antenna radiation pattern from the intended may be caused by some incidents during transporting and/or installing the antenna, for example misconfiguration of tilt of the antenna, or due to aging.

Effects of the near-field obstacles cannot be fully predetermined during design and manufacturing of the antenna nor during designing the network and installing the antenna, but these may cause significant deviation of the achieved cellular network coverage by the antenna. The invented method enables detecting deviation of the actual radiation pattern from that intended using sparse field measurements and the DNN model and provides an effective way to recognize need for corrective measures. The invented method may also be used to detect other types of problems in the antenna in its normal operation environment, such as misconfiguration of the electrical or physical down tilt. Remotely operated tilt actuators may also be jammed or broken, and thus can be identified as misconfigured down tilt.

The method may be applied also to multibeam and active antennas used in 5G, so that envelope of the radiating beams can be detected. Multibeam passive antennas have typically large number of jumper cables and there is a high risk of connecting the cables in wrong ports. The resulting envelope of radiating beams will be different from what was intended and with this invention this type of installation mistakes can be detected.

Figure 7 illustrates main steps of a method according to the embodiments of the invention.

In the step 71, field measurement data is obtained. Field measurements are used for collecting data regarding received far-field signals radiated by an antenna under investigation in its operating environment. Field measurement data represents a plurality of measurement points. The measurement points correspond to a portion of cells of a grid covering a geographical area served by the antenna under investigation. In comparison to the entire number of cells in the grid, measurements points represent a minor portion of the cells. For example, measurement points that represent less than 5%, even less than 2% of all cells in the grid may be enough for generating a decent horizontal radiation pattern of the antenna under investigation using the DNN model. Increasing the share of the measured cells naturally improves then accuracy of the DNN model to be trained. Preferably, measurement points cover the area under study with reasonably even distribution.

In the step 72, supervised learning is by the deep neural network (DNN) for training a DNN model. The DNN model is taught, using the field measurement data as a target, to determine a pathloss of the signal between the antenna under investigation and each respective measurement point. The DNN model determines a model to calculate the pathloss on basis of at least the field measurement data, geographical location of the antenna under investigation, azimuth direction from respective measurement points to the antenna or from the antenna to the respective measurement points, and geographical data of the operating environment so that the pathloss calculated using the DNN model corresponds to the pathloss detected in the field measurements.

In the step 73, a coverage map of the antenna under investigation in its operation environment may be generated using the trained DNN model. The trained DNN model enables determining pathloss at each cell of the grid, including both the cells corresponding to field measurement points and also cells not covered by the field measurements. Generation of the coverage map is an optional step. If purpose is just to determine the radiation pattern of the antenna, this step may be omitted.

In the step 74, geographical data used by the DNN model is replaced with a flat environment model.

In the step 75, and a two-dimensional radiation pattern of the antenna under investigation is generated using the same, trained DNN model. Effect of using the flat environment model is that the trained DNN model now generates a two-dimensional radiation pattern of the antenna under investigation, since no environmental factors affect to the generated two- dimensional radiation pattern, which can be considered as a kind of modified coverage map.

The information on the two-dimensional radiation pattern may then be stored a memory and/or the two-dimensional radiation pattern of the antenna under investigation may be displayed on a display.

As disclosed above, geographical data is typically defined as a plurality of layers, each representing one type of information. For example, a map application can easily define, which layers are to be presented. The DNN model trained for generating antenna coverage maps and antenna radiation patterns comprises a plurality of geographical data layers. A skilled person in the art, recognizes that various possibilities exist how the DNN model is built to apply input data for desired results.

In following, more details of an exemplary, non-limiting practical implementation for performing supervised learning by the DNN to develop the DNN model is disclosed. Let us have k geographical data layers in the DNN model.

For each measurement point, and for each geographical data layer used by the DNN model, a line is drawn from antenna to measurement point, and the line is projected on the respective data layer. Data pixels along the line are extracted to define a variable size of one-dimensional array A.

The k arrays A for each measurement point may then be standardized into a predefined sized array B, and all k pieces of arrays B are concatenated to form one long one-dimensional array C.

Further, three other one-dimensional arrays, representing geographical coordinates of the antenna under investigation, azimuth direction from the antenna to the measurement point or vice versa and coordinates of the measurement point are concatenated to the one-dimensional array C to form a one-dimensional, fixed size array X. This exemplary method uses the main propagating component for creation of the array X.

The DNN model, which is based on a radio propagation model, is then trained using a supervised learning mode. Learning is performed in batches using an array [X x k] as input to the DNN and measurement points as the target. The DNN thus trains the DNN model so that any input among those in the [X x k] array results to values of the measurement points.

For improving accuracy of the radio propagation model, the onedimensional array X may further be provided with additional radio environment data. Such additional radio environment data may be obtained by filtering a two-dimensional array comprising data concerning objects near each respective measurement point. The additional radio environment information may comprise height information of surrounding buildings, natural barriers and/or reflective surfaces near the respective measurement point. The two-dimensional array may be filtered around each respective measurement point and flattened so that this data can further be concatenated to the one-dimensional array X. This operation adds reflective and other radio propagation affecting elements to the DNN model instead of mere main propagation component between the antenna and the measurement point.

As known by a person skilled in the art, source data may comprise N- dimensional data, which may be formatted for the neural network using any suitable formatting, depending on the implementation. Thus, instead of using the exemplary one-dimensional array as input for the neural network, any other data formatting may be used without departing from the scope.

It is apparent to a person skilled in the art that as technology advanced, the basic idea of the invention can be implemented in various ways. The invention and its embodiments are therefore not restricted to the above examples, but they may vary within the scope of the claims.