IP Library › Granted Patent US 12,273,154
Granted Patent B1
US 12,273,154 · App. 18/917,819 · Granted Apr 8, 2025

Radio signal prediction in an environment

Inventors: Csaba Mate Jozsa (Budapest, HU); Gábor Sörös (Budapest, HU); Lóránt Farkas (Budapest, HU)
Assignee: Nokia Solutions and Networks Oy
H04B17/3913H04B17/3912
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,273,154
App. No.
18/917,819
Granted
Apr 8, 2025
Kind
B1
Abstract

Disclosed is a method comprising: creating a first structure graph whose nodes represent points of a first structure map, the nodes being associated with state vectors obtained using feature vectors of the points of the first structure map; creating a first RF graph having nodes representing points of a respective first RF map; creating a first input graph from the first structure graph and the first RF graph; updating state vectors of the nodes of the first input graph; creating an output graph whose nodes represent target spatial points of the environment; connecting the output graph with at least the first input graph; updating the state vectors of the output graph; inputting the state vectors of the output graph to a trained machine learning model to obtain a prediction of a signal propagation characteristic at the target points.

Claims (104)

1. An apparatus comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus at least to perform:

receiving a first structure map comprising spatial points representing a structure of an environment and feature vectors descriptive of the points;

receiving at least a first radio frequency, RF, map for respective at least one RF signal source, the first RF map comprising spatial points representing locations of the environment where measurements of RF signals from a respective RF signal source are performed, the first RF map further comprising feature vectors descriptive of the RF measurements;

creating using a first structure edge connecting rule a first structure graph, whose nodes represent the points of the first structure map, the nodes being associated with state vectors obtained using the feature vectors of the points of the first structure map;

creating using a first RF edge connecting rule, for the first RF map, a first RF graph, the first RF graph having nodes representing the points of the respective first RF map, wherein the nodes of the first RF graph are associated with state vectors obtained using the feature vectors of the first RF map;

creating a first input graph from the first structure graph and the first RF graph by using a second edge connecting rule;

updating state vectors of the nodes of the first input graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

creating an output graph whose nodes represent target spatial points of the environment, the nodes of the output graph being associated with state vectors;

connecting the output graph with at least the first input graph using an output edge connecting rule, wherein the first input graph comprises the nodes with the updated state vectors;

updating the state vectors of the output graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

inputting the state vectors of the output graph to a trained machine learning model to obtain a prediction of a signal propagation characteristic at the target points.

2. The apparatus of claim 1 , wherein the instructions, when executed by the at least one processor, further cause the apparatus at least to perform:

receiving an acquired initial structure map comprising initial spatial points representing the structure of the environment and initial feature vectors descriptive of the initial points;

receiving at least one acquired initial RF map, the initial RF map comprising initial points representing locations of the environment where measurements of RF signals from a respective RF signal source are performed, the initial RF map further comprising initial feature vectors descriptive of the RF measurements;

determining a first set of structure feature centers representing the initial points of the initial structure map;

determining a first set of RF feature centers representing the initial points of the initial RF map;

clustering the initial points of the initial structure map around the first set of structure feature centers, thereby obtaining a first set of structure clusters;

clustering the initial points of the initial RF map around the first set of RF feature centers of the initial RF map, thereby obtaining a first set of RF clusters;

performing a combination of the initial feature vectors of the first set of structure clusters to obtain a first set of structure convolutional feature vectors representing the first set of structure feature centers;

performing a combination of the initial feature vectors of the first set of RF clusters to obtain a first set of RF convolutional feature vectors representing the first set of RF feature centers;

providing the points of the first structure map as the first set of structure feature centers, and the feature vectors of the first structure map as the first set of structure convolutional feature vectors;

providing the points of the first RF map as the first set of RF feature centers, and the feature vectors of the first RF map as the first set of RF convolutional feature vectors.

3. The apparatus of claim 2 , wherein the instructions, when executed by the at least one processor, further cause the apparatus at least to perform:

determining the structure feature center by at least: combining the points that represent a respective area of the environment to obtain the structure feature center; and

determining the RF feature center by at least: combining the points that represent a respective area of the environment to obtain the RF feature center.

4. The apparatus of claim 2 , wherein the acquired initial structure map being a three-dimensional, 3D, point cloud.

5. The apparatus of claim 2 , the combination of feature vectors being performed by a trained first encoder, herein referred to as first local encoder, the first local encoder being configured to process each cluster of points of the clusters and to output a convolutional feature vector.

6. The apparatus of claim 2 , wherein the state vector is updated using a first encoder, referred to as first global encoder, the first global encoder is configured to receive as input the state vector of the specific node and the state vectors of the intra-graph and inter-graph neighborhood nodes of the specific node, and to provide the updated state vector of the specific node.

7. The apparatus of claim 6 , wherein the first global encoder is a trained graph neural network updating state vectors of the first input graph.

8. The apparatus of claim 6 , wherein the trained machine learning model being a decoder, wherein the first local encoder, the first global encoder and the decoder are jointly trained to determine their learnable parameters such that a deviation between the predicted signal propagation characteristics and known signal propagation characteristics is optimized.

9. The apparatus of claim 2 , wherein the instructions, when executed by the at least one processor, further cause the apparatus at least to perform:

determining a second set of structure feature centers representing the points of the first structure map;

determining a second set of RF feature centers representing the points of the first RF map;

clustering the points of the first structure map around the second set of structure feature centers, thereby obtaining a second set of structure clusters;

clustering the points of the first RF map around the second set of RF feature centers of the first RF map, thereby obtaining a second set of RF clusters;

performing a combination of the feature vectors of the second set of structure clusters to obtain a second set of structure convolutional feature vectors representing the second set of structure feature centers;

performing a combination of the feature vectors of the second set of RF clusters to obtain a second set of RF convolutional feature vectors representing the second set of RF feature centers;

providing the points of a second structure map as the second set of structure feature centers, and the feature vectors of the second structure map as the second set of structure convolutional feature vectors;

providing the points of a second RF map as the second set of RF feature centers, and the feature vectors of the second RF map as the second set of RF convolutional feature vectors.

10. The apparatus of claim 9 , wherein the instructions, when executed by the at least one processor, further cause the apparatus at least to perform:

creating using a second structure edge connecting rule a second structure graph whose nodes represent the points of the second structure map, the nodes being associated with state vectors obtained using the feature vectors of the points of the second structure map;

creating using a second RF edge connecting rule a second RF graph, the second RF graph having nodes representing the points of the second RF map, wherein the nodes of the second RF graph are associated with state vectors obtained using the feature vectors of the second RF map;

creating a second input graph from the second structure graph and the second RF graph by using the second edge connecting rule;

updating state vectors of the nodes of the second input graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

wherein nodes of the output graph are connected further using the second input graph, wherein the second structure map comprises a number of points smaller than or equal to the number of points of the first structure map, wherein the number of points of the second RF map is smaller than or equal to the number of points of the first RF map.

11. The apparatus of claim 10 ,

wherein the inter-graph neighborhood nodes of the node of the first structure graph comprise nodes of at least one of: the first RF graph, the second structure graph or the second RF graph;

wherein the inter-graph neighborhood nodes of the node of the first RF graph comprise nodes of at least one of: the first RF graph different from the specific first RF graph, the second structure graph, the second RF graph, or the first structure graph.

12. The apparatus of claim 10 , wherein the state vector of a specific node is updated using a second encoder, referred to as second global encoder, the second global encoder is configured to receive as input the state vector of the specific node and the state vectors of the neighborhood nodes of the specific node, and to predict the state vector of the specific node.

13. The apparatus of claim 1 , wherein a first edge connecting rule is the first RF edge connecting rule or the first structure edge connecting rule, the first edge connecting rule being the output edge connecting rule or being different from the output edge connecting rule, the first edge connecting rule being the second edge connecting rule or being different from the second edge connecting rule.

14. A method comprising:

receiving a first structure map comprising spatial points representing a structure of an environment and feature vectors descriptive of the points;

receiving at least a first radio frequency, RF, map for respective at least one RF signal source, the first RF map comprising spatial points representing locations of the environment where measurements of RF signals from a respective RF signal source are performed, the first RF map further comprising feature vectors descriptive of the RF measurements;

creating using a first structure edge connecting rule a first structure graph whose nodes represent the points of the first structure map, the nodes being associated with state vectors obtained using the feature vectors of the points of the first structure map;

creating using a first RF edge connecting rule, for the first RF map, a first RF graph, the first RF graph having nodes representing the points of the respective first RF map, wherein the nodes of the first RF graph are associated with state vectors obtained using the feature vectors of the first RF map;

creating a first input graph from the first structure graph and the first RF graph by using a second edge connecting rule;

updating state vectors of the nodes of the first input graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

creating an output graph whose nodes represent target spatial points of the environment, the nodes of the output graph being associated with state vectors;

connecting the output graph with at least the first input graph using an output edge connecting rule, wherein the first input graph comprises the nodes with the updated state vectors;

updating the state vectors of the output graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

inputting the state vectors of the output graph to a trained machine learning model to obtain a prediction of a signal propagation characteristic at the target points.

15. The method of claim 14 , further comprising:

receiving an acquired initial structure map comprising initial spatial points representing the structure of the environment and initial feature vectors descriptive of the initial points;

receiving at least one acquired initial RF map, the initial RF map comprising initial points representing locations of the environment where measurements of RF signals from a respective RF signal source are performed, the initial RF map further comprising initial feature vectors descriptive of the RF measurements;

determining a first set of structure feature centers representing the initial points of the initial structure map;

determining a first set of RF feature centers representing the initial points of the initial RF map;

clustering the initial points of the initial structure map around the first set of structure feature centers, thereby obtaining a first set of structure clusters;

clustering the initial points of the initial RF map around the first set of RF feature centers of the initial RF map, thereby obtaining a first set of RF clusters;

performing a combination of the initial feature vectors of the first set of structure clusters to obtain a first set of structure convolutional feature vectors representing the first set of structure feature centers;

performing a combination of the initial feature vectors of the first set of RF clusters to obtain a first set of RF convolutional feature vectors representing the first set of RF feature centers;

providing the points of the first structure map as the first set of structure feature centers, and the feature vectors of the first structure map as the first set of structure convolutional feature vectors;

providing the points of the first RF map as the first set of RF feature centers, and the feature vectors of the first RF map as the first set of RF convolutional feature vectors.

16. The method of claim 15 , further comprising:

determining the structure feature center by at least: combining the points that represent a respective area of the environment to obtain the structure feature center; and

determining the RF feature center by at least: combining the points that represent a respective area of the environment to obtain the RF feature center.

17. The method of claim 16 , further comprising:

determining a second set of structure feature centers representing the points of the first structure map;

determining a second set of RF feature centers representing the points of the first RF map;

clustering the points of the first structure map around the second set of structure feature centers, thereby obtaining a second set of structure clusters;

clustering the points of the first RF map around the second set of RF feature centers of the first RF map, thereby obtaining a second set of RF clusters;

performing a combination of the feature vectors of the second set of structure clusters to obtain a second set of structure convolutional feature vectors representing the second set of structure feature centers;

performing a combination of the feature vectors of the second set of RF clusters to obtain a second set of RF convolutional feature vectors representing the second set of RF feature centers;

providing the points of a second structure map as the second set of structure feature centers, and the feature vectors of the second structure map as the second set of structure convolutional feature vectors;

providing the points of a second RF map as the second set of RF feature centers, and the feature vectors of the second RF map as the second set of RF convolutional feature vectors.

18. The method of claim 17 , further comprising:

creating using a second structure edge connecting rule a second structure graph whose nodes represent the points of the second structure map, the nodes being associated with state vectors obtained using the feature vectors of the points of the second structure map;

creating using a second RF edge connecting rule a second RF graph, the second RF graph having nodes representing the points of the second RF map, wherein the nodes of the second RF graph are associated with state vectors obtained using the feature vectors of the second RF map;

creating a second input graph from the second structure graph and the second RF graph by using the second edge connecting rule;

updating state vectors of the nodes of the second input graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

wherein the output graph is connected further using the second input graph, wherein the second structure map comprises a number of points smaller than or equal to the number of points of the first structure map, wherein the number of points of the second RF map is smaller than or equal to the number of points of the first RF map.

19. The method of claim 18 ,

wherein the inter-graph neighborhood nodes of the node of the first structure graph comprise nodes of at least one of: the first RF graph, the second structure graph or the second RF graph;

wherein the inter-graph neighborhood nodes of the node of the first RF graph comprise nodes of at least one of: the first RF graph different from the specific first RF graph, the second structure graph, the second RF graph, or the first structure graph.

20. A non-transitory computer readable medium comprising program instructions that, when executed by an apparatus, cause the apparatus to perform at least the following:

receiving a first structure map comprising spatial points representing a structure of an environment and feature vectors descriptive of the points;

receiving at least a first radio frequency, RF, map for respective at least one RF signal source, the first RF map comprising spatial points representing locations of the environment where measurements of RF signals from a respective RF signal source are performed, the first RF map further comprising feature vectors descriptive of the RF measurements;

creating using a first structure edge connecting rule a first structure graph whose nodes represent the points of the first structure map, the nodes being associated with state vectors obtained using the feature vectors of the points of the first structure map;

creating using a first RF edge connecting rule, for the first RF map, a first RF graph, the first RF graph having nodes representing the points of the respective first RF map, wherein the nodes of the first RF graph are associated with state vectors obtained using the feature vectors of the first RF map;

creating a first input graph from the first structure graph and the first RF graph by using a second edge connecting rule;

updating state vectors of the nodes of the first input graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

creating an output graph whose nodes represent target spatial points of the environment, the nodes of the output graph being associated with state vectors;

connecting the output graph with at least the first input graph using an output edge connecting rule, wherein the first input graph comprises the nodes with the updated state vectors;

updating the state vectors of the output graph, the updating of the state vector being performed using state vectors of intra-graph neighborhood nodes of the respective node and inter-graph neighborhood nodes of the node;

inputting the state vectors of the output graph to a trained machine learning model to obtain a prediction of a signal propagation characteristic at the target points.

Assignments (6)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: MATE JOZSA, CSABA; SÖRÖS, GÁBOR; FARKAS, LÓRÁNT
To: NOKIA SOLUTIONS AND NETWORKS KFT.
Reel/Frame 069748/0656 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: NOKIA SOLUTIONS AND NETWORKS KFT.
To: NOKIA SOLUTIONS AND NETWORKS OY
Reel/Frame 069748/0660 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: LAPPALAINEN, ANDREW
To: NOKIA OF AMERICA CORPORATION
Reel/Frame 069748/0664 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: ANTERO KESKITALO, ILKKA
To: NOKIA SOLUTIONS AND NETWORKS OY
Reel/Frame 069748/0667 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: NOKIA OF AMERICA CORPORATION
To: NOKIA TECHNOLOGIES OY
Reel/Frame 069748/0670 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Jan 6, 2025
From: NOKIA SOLUTIONS AND NETWORKS OY
To: NOKIA TECHNOLOGIES OY
Reel/Frame 069748/0674 →
Priority Claims (1)
FI 20236154 · Oct 17, 2023 · national
References Cited (40)
US 10834532B2 · Fuerst et al. · 2020 [cited by applicant]
US 11425635B2 · Zeng et al. · 2022 [cited by applicant]
US 20170064515A1 · Heikkila · 2017 [cited by examiner]
CN 110213003A · 2019 [cited by applicant]
CN 113329437A · 2021 [cited by applicant]
CN 114205846A · 2022 [cited by applicant]
WO 2019081781A1 · 2019 [cited by applicant]
WO 2019109780A1 · 2019 [cited by applicant]
WO 2020125349A1 · 2020 [cited by applicant]
Anusuya et al., “Wireless Channel Models for Indoor Environments”, Defence Science Journal, vol. 58, No. 06, Nov. 2008, pp. 771-777. [cited by applicant]
Ostlin et al., “Macrocell Path-Loss Prediction Using Artificial Neural Networks”, IEEE Transactions on Vehicular Technology, vol. 59, No. 06, Jul. 2010, pp. 2735-2747. [cited by applicant]
Ayadi et al., “A UHF Path Loss Model Using Learning Machine for Heterogeneous Networks”, IEEE Transactions on Antennas and Propagation, vol. 65, No. 07, Jul. 2017, pp. 3675-3683. [cited by applicant]
Liao et al., “The Rayleigh Fading Channel Prediction via Deep Learning”, Wireless Communications and Mobile Computing, vol. 2018, Article 6497340, Jul. 25, 2018, pp. 1-11. [cited by applicant]
Zhang et al., “Path Loss Prediction Based on Machine Learning: Principle, Method, and Data Expansion”, Applied Sciences, vol. 09, No. 09, May 9, 2019, pp. 1-18. [cited by applicant]
Koivumaki et al., “Impacts of Point Cloud Modeling on the Accuracy of Ray-Based Multipath Propagation Simulations”, IEEE Transactions on Antennas and Propagation, vol. 69, No. 08, Aug. 2021, pp. 4737-4747. [cited by applicant]
Wisanmongkol et al., “An ensemble approach to deep-learning-based wireless indoor localization”, IET Wireless Sensor Systems, vol. 12, No. 02, Apr. 2022, pp. 33-55. [cited by applicant]
Zhang et al., “Air-to-Air Path Loss Prediction Based on Machine Learning Methods in Urban Environments”, Wireless Communications and Mobile Computing, vol. 2018, Article 8489326, Jun. 13, 2018, pp. 1-9. [cited by applicant]
Aldossari et al., “Predicting the Path Loss of Wireless Channel Models Using Machine Learning Techniques in MmWave Urban Communications”, 22nd International Symposium on Wireless Personal Multimedia Communications (WPMC… [cited by applicant]
Hashemifar et al., “Augmenting visual SLAM with Wi-Fi sensing for indoor applications”, Autonomous Robots, vol. 43, Jul. 18, 2019, pp. 2245-2260. [cited by applicant]
Ayyalasomayajula et al., “Deep Learning based Wireless Localization for Indoor Navigation”, Proceedings of the 26th Annual International Conference on Mobile Computing and Networking, Sep. 21-25, 2020, 14 pages. [cited by applicant]
Jakes et al., “Microwave Mobile Communications”, Wiley-IEEE, 1994, 78 pages. [cited by applicant]
Roche et al., “2.5D extensions of the Frequency Domain ParFlow Algorithm for Simulating 802.11b/g Radio Coverage in multifloored buildings”, IEEE Vehicular Technology Conference, Sep. 25-28, 2006, 5 pages. [cited by applicant]
Friis, “A Note on a Simple Transmission Formula”, Proceedings of the I.R.E. and Waves and Electrons, May 1946, 254-256. [cited by applicant]
Sinkhorn et al., “Concerning Nonnegative Matrices and Doubly Stochastic Matrices”, Pacific Journal of Mathematics, vol. 21, No. 02, Dec. 1967, 9 pages. [cited by applicant]
Besl et al., “A Method for Registration of 3-D Shapes”, Pacific Journal of Mathematics, Proceedings vol. 1611, Sensor Fusion IV: Control Paradigms and Data Structures (ROBOTICS), 1991, pp. 586-606. [cited by applicant]
Ioffe et al., “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift”, Proceedings of the 32 nd International Conference on Machine Learning, vol. 37, Jul. 7-9, 2015, 9 pages. [cited by applicant]
Ba et al., “Layer Normalization”, arXiv, Jul. 21, 2016, pp. 1-14. [cited by applicant]
Krizhevsky et al., “ImageNet Classification with Deep Convolutional Neural Networks”, Advances in Neural Information Processing Systems 25 (NIPS), Dec. 3-6, 2012, pp. 1-9. [cited by applicant]
Vaswani et al., “Attention Is All You Need”, arXiv, Aug. 2, 2023, pp. 1-15. [cited by applicant]
Wu et al., “Group Normalization”, Proceedings of the European Conference on Computer Vision (ECCV), Sep. 8-14, 2018, pp. 1-17. [cited by applicant]
Sarlin et al., “SuperGlue: Learning Feature Matching with Graph Neural Networks”, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 13-19, 2020, pp. 4938-4947. [cited by applicant]
Thomas et al., “KPConv: Flexible and Deformable Convolution for Point Clouds”, Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), Oct. 27-Nov. 2, 2019, pp. 6411-6420. [cited by applicant]
Qin et al., “Geometric Transformer for Fast and Robust Point Cloud Registration”, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 19-24, 2022, pp. 11143-11152. [cited by applicant]
Seidel et al., “914 MHz Path Loss Prediction Models for Indoor Wireless Communications in Multifloored Buildings”, IEEE Transactions on Antennas and Propagation, vol. 40, No. 02, Feb. 1992, pp. 207-217. [cited by applicant]
Cavalcanti et al., “A Hybrid Path Loss Prediction Model based on Artificial Neural Networks using Empirical Models for LTE And LTE-A at 800 MHz and 2600 MHz”, Journal of Microwaves, Optoelectronics and Electromagnetic A… [cited by applicant]
Torun et al., “Spatial Prediction of Channel Signal Strength Map Using Deep Fully Convolutional Neural Network”, 56th Asilomar Conference on Signals, Systems, and Computers, Oct. 31-Nov. 2, 2022, pp. 553-558. [cited by applicant]
Office action received for corresponding Finnish Patent Application No. 20236154, dated Mar. 13, 2024, 10 pages. [cited by applicant]
Liu et al., “A Generalizable Indoor Propagation Model Based on Graph Neural Networks”, IEEE Transactions on Antennas and Propagation, vol. 71, No. 07, Jul. 2023, pp. 6098-6110. [cited by applicant]
Chen et al., “A Graph Neural Network Based Radio Map Construction Method for Urban Environment”, IEEE Communications Letters vol. 27, No. 05, May 2023, pp. 1327-1331. [cited by applicant]
Office action received for corresponding Finnish Patent Application No. 20236154, dated Jun. 14, 2024, 4 pages. [cited by applicant]