CA2977300A1 - System and methods for channel modeling/estimation in a wireless communication network - Google Patents
System and methods for channel modeling/estimation in a wireless communication network Download PDFInfo
- Publication number
- CA2977300A1 CA2977300A1 CA2977300A CA2977300A CA2977300A1 CA 2977300 A1 CA2977300 A1 CA 2977300A1 CA 2977300 A CA2977300 A CA 2977300A CA 2977300 A CA2977300 A CA 2977300A CA 2977300 A1 CA2977300 A1 CA 2977300A1
- Authority
- CA
- Canada
- Prior art keywords
- model
- channel
- trained
- continuous
- frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000004891 communication Methods 0.000 title abstract description 9
- 238000010801 machine learning Methods 0.000 claims abstract description 8
- 230000004044 response Effects 0.000 claims description 5
- 238000013461 design Methods 0.000 abstract description 5
- 238000004088 simulation Methods 0.000 abstract description 4
- 230000006399 behavior Effects 0.000 description 4
- 230000005540 biological transmission Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 208000009119 Giant Axonal Neuropathy Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000008713 feedback mechanism Effects 0.000 description 1
- 201000003382 giant axonal neuropathy 1 Diseases 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3912—Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region
Landscapes
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Electromagnetism (AREA)
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
A system and method for communication in a wireless communication network is provided. This method proposes to apply machine learning methods including but not limited to generative models in some parts of the communication link which needs to deal with the wireless channel. Several specific cases also presented where generative models are used for channel modeling and simulation, channel estimation at the receiver, and knowledge of channel information at the transmitter and the feedback design.
Description
System and methods for channel modeling/estimation in a wireless communication network Abstract A system and method for communication in a wireless communication network is provided. This method proposes to apply machine learning methods including but not limited to generative models in some parts of the communication link which needs to deal with the wireless channel.
Several specific cases also presented where generative models are used for channel modeling and simulation, channel estimation at the receiver, and knowledge of channel information at the transmitter and the feedback design.
Description When data is transmitted wirelessly from a source to a destination, it gets affected by many factors including the conditions/state of the environment, including channel (the link between the source and destination) state and existence of other possible transmitters in the network.
Having good understanding of the channel state realization or even only its behavior can be useful in many areas of data transmission. For example, a) having good models for the channel, it can help us to generate many channel samples which are good representative of the actual channels and can be used for simulation purposes which can then be used for design of good transmission schemes.
b) if the channel state can be acquired at the destination (completely, or even partly) it can use that information to better detect the signal, c) if the channel state is known at the transmitter (completely, or even partly), it can use that information to construct a signal which is more robust to channel impairment.
In this disclosure we propose that machine learning schemes and in particular but not limited to the generative models can be beneficial in several aspects of channel modeling/estimation including but not limited to the above mentioned areas.
Before going to the details we first make a quick note that there are two main methods/goals for data modeling, one that we want to build a model that by looking at the data it can tell something about the data, for example which group it belongs to or what the value of a particular output would be. We call them discriminative models. The other group of models are called generative models and they try to generate data samples similar to the data that the actual system generates.
The measure of similarity is different based on the application, for example could be human perception or Euclidian distance.
Input data, could be one or Discriminative . 0 Could have none, one or multiple metrics each could multiple quantitative of have one or multiple features Models ___ 1, qualitative outputs Input parameter, could be Generative ___ 10. One or multiple outputs none, one or multiple following the same pattern of parameters each may have models the actual data ___________________________________________ 10.
one or multiple features The idea of this disclosure is to use different machine learning algorithm and in particular but not limited to generative models in wireless commutation including but not limited to when channel information is needed.
In the following, we bring some specific examples while we point out these are just a few examples and many other application can be thought of in this regards.
1- Use of the generative models for generation of samples that has high correlation with the actual channel realizations.
In many steps of design and performance evaluation of wireless communication systems and schemes, it is needed that we have access to many channel realizations to perform simulations for example. To simplify accessing to such channel realizations, researchers have already proposed some models to produce samples which have similar behaviors to the actual channel realizations (e.g. statistical methods and/or ray based modeling schemes). The proposed idea here is to first use some of the real life measurement to train a generative models (like a variational auto-encoder, Generative adversarial network (GANs), ....) and then use these models to generate samples of the channel (in time-frequency region) which is similar to the actual channel samples. Note that this technique of generating simulated data using the generative models is not limited to channel modeling and can be extended to any system that we want to have many of its samples but we can only afford simulating it. Note that the generative models could be also conditional on some parameters meaning that by setting some input parameters the models gives back a set of data that are related to a particular property.
For example, by changing a parameter from "a" to "b" the type of channel realization that the models generate becomes more resembles to channels samples for a link in which the user speed is "v_a" and "v_b" respectively.
Several specific cases also presented where generative models are used for channel modeling and simulation, channel estimation at the receiver, and knowledge of channel information at the transmitter and the feedback design.
Description When data is transmitted wirelessly from a source to a destination, it gets affected by many factors including the conditions/state of the environment, including channel (the link between the source and destination) state and existence of other possible transmitters in the network.
Having good understanding of the channel state realization or even only its behavior can be useful in many areas of data transmission. For example, a) having good models for the channel, it can help us to generate many channel samples which are good representative of the actual channels and can be used for simulation purposes which can then be used for design of good transmission schemes.
b) if the channel state can be acquired at the destination (completely, or even partly) it can use that information to better detect the signal, c) if the channel state is known at the transmitter (completely, or even partly), it can use that information to construct a signal which is more robust to channel impairment.
In this disclosure we propose that machine learning schemes and in particular but not limited to the generative models can be beneficial in several aspects of channel modeling/estimation including but not limited to the above mentioned areas.
Before going to the details we first make a quick note that there are two main methods/goals for data modeling, one that we want to build a model that by looking at the data it can tell something about the data, for example which group it belongs to or what the value of a particular output would be. We call them discriminative models. The other group of models are called generative models and they try to generate data samples similar to the data that the actual system generates.
The measure of similarity is different based on the application, for example could be human perception or Euclidian distance.
Input data, could be one or Discriminative . 0 Could have none, one or multiple metrics each could multiple quantitative of have one or multiple features Models ___ 1, qualitative outputs Input parameter, could be Generative ___ 10. One or multiple outputs none, one or multiple following the same pattern of parameters each may have models the actual data ___________________________________________ 10.
one or multiple features The idea of this disclosure is to use different machine learning algorithm and in particular but not limited to generative models in wireless commutation including but not limited to when channel information is needed.
In the following, we bring some specific examples while we point out these are just a few examples and many other application can be thought of in this regards.
1- Use of the generative models for generation of samples that has high correlation with the actual channel realizations.
In many steps of design and performance evaluation of wireless communication systems and schemes, it is needed that we have access to many channel realizations to perform simulations for example. To simplify accessing to such channel realizations, researchers have already proposed some models to produce samples which have similar behaviors to the actual channel realizations (e.g. statistical methods and/or ray based modeling schemes). The proposed idea here is to first use some of the real life measurement to train a generative models (like a variational auto-encoder, Generative adversarial network (GANs), ....) and then use these models to generate samples of the channel (in time-frequency region) which is similar to the actual channel samples. Note that this technique of generating simulated data using the generative models is not limited to channel modeling and can be extended to any system that we want to have many of its samples but we can only afford simulating it. Note that the generative models could be also conditional on some parameters meaning that by setting some input parameters the models gives back a set of data that are related to a particular property.
For example, by changing a parameter from "a" to "b" the type of channel realization that the models generate becomes more resembles to channels samples for a link in which the user speed is "v_a" and "v_b" respectively.
2- Another application could be on channel estimation. Currently the main methods of channel estimations are blind estimation and pilot based estimation. In the first method, the receiver estimate the channel without/ or with minimal help of the transmitter. In the pilot method the transmitter sends some preset data at some time slots/ subcarriers and the receiver looks into those pilot locations and since it does know what has been transmitted it can find out the channel state by comparing the received signal and the transmitted one. As these pilot are not covering the whole range of the time/frequency after estimating the channel at the pilot locations, the receiver uses some methods to interpolate/extrapolate the estimated channel form the pilot locations to all other places on the time/frequency grid. The way that machine learning could help is if we somehow learn about the way that channel is behaving in that environment and then construct a model for that settings, we can then use that model alongside the blind estimation or pilot based estimation to get to a more accurate state of the network.
Several ways exist to learn about the channel behavior in that environment.
One method could be that at the beginning (when we do not have the model) we only use common channel estimation methods (like blind and pilot based schemes) to estimate the channel for a range of frequency and time to have a complete channel measurement. We continue this process until we collect many samples of the channel. Using these data we then construct our channel model.
Also we can find a generative model for such input data which is able to generate some channel response for that range of time-frequency. Using this model we are able to find what are some possible channel responses in that range of time-frequency given only a portion of that frequency (it basically tries to generate samples of the channel in that time/frequency range which have good correlation with the given part of the resource grid). So, the output of such prediction (from some known part of the time/freq grid to others ¨for example where the pilots have been transmitted to other locations) can also be used for reducing the number of pilots needed for transmission as the shortage of pilots can be compensated by the extra information that we will get from the generative model. It is also possible to first (when the generative model is not that accurate) to send many pilots and when the model gets better reduce the number of pilots as the extra information from the generative model is enough to determine the channel with even fewer pilots.
An alternative method also is to generate a conditional generative models of the channel and then instead of sending pilots to estimate the channel directly, send the required input of the generative model to the receiver so it can construct the channel. Or maybe combine this method with the pilot method or send some signals that the receiver can find out the required inputs to the conditional generative model and then use the model to generate the channel which resembles the actual realized channel.
Several ways exist to learn about the channel behavior in that environment.
One method could be that at the beginning (when we do not have the model) we only use common channel estimation methods (like blind and pilot based schemes) to estimate the channel for a range of frequency and time to have a complete channel measurement. We continue this process until we collect many samples of the channel. Using these data we then construct our channel model.
Also we can find a generative model for such input data which is able to generate some channel response for that range of time-frequency. Using this model we are able to find what are some possible channel responses in that range of time-frequency given only a portion of that frequency (it basically tries to generate samples of the channel in that time/frequency range which have good correlation with the given part of the resource grid). So, the output of such prediction (from some known part of the time/freq grid to others ¨for example where the pilots have been transmitted to other locations) can also be used for reducing the number of pilots needed for transmission as the shortage of pilots can be compensated by the extra information that we will get from the generative model. It is also possible to first (when the generative model is not that accurate) to send many pilots and when the model gets better reduce the number of pilots as the extra information from the generative model is enough to determine the channel with even fewer pilots.
An alternative method also is to generate a conditional generative models of the channel and then instead of sending pilots to estimate the channel directly, send the required input of the generative model to the receiver so it can construct the channel. Or maybe combine this method with the pilot method or send some signals that the receiver can find out the required inputs to the conditional generative model and then use the model to generate the channel which resembles the actual realized channel.
3- Another application could be on the way that the transmitter find out information about the channel state. One common way for gaining such knowledge is that the receiver measure the channel using an approach and then feedback the information to the transmitter. The accuracy of the feedback information depends on the gradually of the feedback information (what is the time and frequency spacing between the feed backed information) and also the quantization method and the number of bits that can be feedback.
We propose that machine learning approaches can be used in feedback mechanism as well. For example one possible way is to use generative models and conditional generative models.
Meaning that first build a generative model for the channel. For example, a model that by getting a few parameter generate a sample of possible channel (in time and frequency) that have similar behavior to the actual channel realization. Having such channel, the receiver first measure the channel and using that it find out what information is needed as the input of the generative model such that it generate similar channel as what it has measured. Then it feedback these parameters to the transmitter and transmitter uses these information as the input of its generative model. It then generates a channel sample that should have similar properties of the actual channel and the transmitter might be able to use the generated channel as if it has access to the actual channel realization.
Application of this method in massive MIMO scenario is also very promising as instead of sending feedback for lots of time/freq resources the feedback is only a few parameters that are needed for the generative model to generate good sample of the channel.
At the end, we iterate that these are just a few examples of how machine learning and generative models can be used in performance improvement, analysis and design of wireless link and there are many more example that can follow up.
We propose that machine learning approaches can be used in feedback mechanism as well. For example one possible way is to use generative models and conditional generative models.
Meaning that first build a generative model for the channel. For example, a model that by getting a few parameter generate a sample of possible channel (in time and frequency) that have similar behavior to the actual channel realization. Having such channel, the receiver first measure the channel and using that it find out what information is needed as the input of the generative model such that it generate similar channel as what it has measured. Then it feedback these parameters to the transmitter and transmitter uses these information as the input of its generative model. It then generates a channel sample that should have similar properties of the actual channel and the transmitter might be able to use the generated channel as if it has access to the actual channel realization.
Application of this method in massive MIMO scenario is also very promising as instead of sending feedback for lots of time/freq resources the feedback is only a few parameters that are needed for the generative model to generate good sample of the channel.
At the end, we iterate that these are just a few examples of how machine learning and generative models can be used in performance improvement, analysis and design of wireless link and there are many more example that can follow up.
Claims (19)
1. A method for application of machine learning methods for modeling/estimation of at least some features of wireless channel, the method comprising Receiving a set of one or more wireless channel realization(s) (partial or complete channel or some features of it in a domain), Training a model that captures channel statistics in a domain Using the trained model to generate or predict or estimate or reconstruct the wireless channel or a set of its features in a domain
2. A method of claim 1, wherein received wireless channel realizations comprises receiving set of channel response as a form of complex numbers over a block of time-frequency
3. A method of claim 2, wherein the model is a generative model including but not limited to variational autoencoder or generative adversarial network (GAN)
4. A method of claim 3, wherein the model is trained to capture statistics of the wireless channel
A method of claim 4, wherein the model has some other inputs act as tuning parameters.
6. A method of claim 2, wherein the model is used to generate samples with similar statistics of a channel response
7. A method of claim 2, wherein the received data is something other than wireless channel realizations and is another data that we have many samples and we want to model it but not able to model it mathematically.
8. A method of claim 2, wherein the model is trained to construct a continuous/not-continuous portion of block of time-frequency from another given continuous/not-continuous portion of time-frequency block
9. A method of claim 8, wherein it is used for finding the smallest set of inputs that is needed to get to a desirable output accuracy.
10. A method of claim 2, wherein the model is trained to construct (estimate) the block of continuous/not-continuous time-frequency from a given continuous/not-continuous portion of time-frequency block.
11. A method of claim 10, wherein the prior channel reconstruction schemes like MMSE
interpolating can be used alongside the trained model to get to a more accurate result
interpolating can be used alongside the trained model to get to a more accurate result
12. A method of claim 10, wherein it is used for finding the smallest set of inputs block that is needed to get to a desirable output accuracy.
13. A method of claim 8, where a generative model is trained and is used to generate outputs with similar statistics of the actual channel realizations. Then use this generative models to make some sample which match to the known portion of the channel realizations and use the results to estimate the unknown part.
14. A method of claim 2, wherein the model is trained to construct a continuous/not-continuous portion of block of time-frequency from a set of features that it has received.
15. A method of claim 2, wherein the model is trained to construct a continuous/not-continuous portion of block of time-frequency from a set of features that it has received and continuous/not-continuous portion of time-frequency block.
16. A method of claim 2, wherein the model is trained to generate/reconstruct the complete or portion of channel response using some features received.
17. A method of claim 2, wherein the model is trained to determine some features which are good for the model in claim 16 to have more accurate results.
18. A method of claim 17, wherein the resulted features is used as the input of model trained in 16.
19. A method of claim 17, wherein the number of antenna between the transmitter and the receiver is various sizes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA2977300A CA2977300A1 (en) | 2017-08-25 | 2017-08-25 | System and methods for channel modeling/estimation in a wireless communication network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA2977300A CA2977300A1 (en) | 2017-08-25 | 2017-08-25 | System and methods for channel modeling/estimation in a wireless communication network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2977300A1 true CA2977300A1 (en) | 2019-02-25 |
Family
ID=65519439
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA2977300A Abandoned CA2977300A1 (en) | 2017-08-25 | 2017-08-25 | System and methods for channel modeling/estimation in a wireless communication network |
Country Status (1)
| Country | Link |
|---|---|
| CA (1) | CA2977300A1 (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111988156A (en) * | 2019-05-22 | 2020-11-24 | 华为技术服务有限公司 | Method for creating network simulation platform, network simulation method and corresponding device |
| US20220174634A1 (en) * | 2018-03-02 | 2022-06-02 | DeepSig Inc. | Learning communication systems using channel approximation |
-
2017
- 2017-08-25 CA CA2977300A patent/CA2977300A1/en not_active Abandoned
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220174634A1 (en) * | 2018-03-02 | 2022-06-02 | DeepSig Inc. | Learning communication systems using channel approximation |
| US11991658B2 (en) * | 2018-03-02 | 2024-05-21 | DeepSig Inc. | Learning communication systems using channel approximation |
| CN111988156A (en) * | 2019-05-22 | 2020-11-24 | 华为技术服务有限公司 | Method for creating network simulation platform, network simulation method and corresponding device |
| WO2020233708A1 (en) * | 2019-05-22 | 2020-11-26 | 华为技术有限公司 | Method for creating network simulation platform, and network simulation method and corresponding device |
| CN111988156B (en) * | 2019-05-22 | 2022-04-05 | 华为技术服务有限公司 | Method for creating network simulation platform, network simulation method and corresponding device |
| US11856424B2 (en) | 2019-05-22 | 2023-12-26 | Huawei Technologies Co., Ltd. | Method for creating network simulation platform, network simulation method, and corresponding apparatuses |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US12021572B2 (en) | Systems and methods for wireless signal configuration by a neural network | |
| Kasparick et al. | Kernel-based adaptive online reconstruction of coverage maps with side information | |
| Wallace et al. | Time-varying MIMO channels: measurement, analysis, and modeling | |
| CN115053147A (en) | Detecting a location of motion using wireless signals and a topology of wireless connections | |
| CN105656819A (en) | Self-adaptive channel estimation method based on compressed sensing and large-scale MIMO | |
| CN107563437B (en) | Ultra-wideband non-line-of-sight identification method based on random forest | |
| Büyükçorak et al. | Lognormal mixture shadowing | |
| US20220383118A1 (en) | Generating variable communication channel responses using machine learning networks | |
| Qin et al. | A generalized semantic communication system: From sources to channels | |
| KR20180080122A (en) | Apparatus for wireless transmission | |
| Seyedsalehi et al. | Propagation channel modeling by deep learning techniques | |
| CA2977300A1 (en) | System and methods for channel modeling/estimation in a wireless communication network | |
| CN111510848A (en) | Terminal position prediction method, device, medium and equipment | |
| CN117997380A (en) | Performance test method and device for dual-mode communication, electronic equipment and medium | |
| CN111435926B (en) | MIMO system channel prediction method, device, medium and equipment | |
| Naderi et al. | The design of measurement-based underwater acoustic channel simulators using the INLSA algorithm | |
| US9429612B2 (en) | Methods and apparatuses for generating real-environment noise using statistical approach | |
| WO2023121659A1 (en) | Communication device and method for performing communication signal processing | |
| CN104980238B (en) | A kind of cooperative frequency spectrum sensing method sparse based on group | |
| US11507803B2 (en) | System for generating synthetic digital data for data multiplication | |
| WO2023041202A1 (en) | Improved pilot assisted radio propagation channel estimation based on machine learning | |
| CN103731218B (en) | Method for error modeling on basis of TDD cellular network transmitting terminal channel state information | |
| CN104486041B (en) | A kind of transmission mode collocation method and device | |
| Zentarra et al. | From Channel Measurement to Training Data for PHY Layer AI Applications | |
| Yang et al. | Deep-learning-based bouncing-order prediction for propagation channel characterization using graph modeling |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| FZDE | Discontinued |
Effective date: 20230227 |
|
| FZDE | Discontinued |
Effective date: 20230227 |