Patent Application
20250113215
The subject matter described herein relates to devices, systems, and methods for optimizing wireless networks and more particularly to building a signal propagation model using a trained machine learning model.
Wi-Fi is a family of wireless network protocols based on the Institute of Electrical and Electronics Engineers (IEEE) 802.11 family of standards, which are commonly used for local area networking of devices and Internet access, allowing nearby digital devices to exchange data via radio waves. Wi-Fi networks are some of the most widely used computer networks in the world, used globally in home and small office networks to link devices together and to connect them to the Internet via a wireless router. Wi-Fi networks often use wireless access points in public places like coffee shops, hotels, libraries, and airports to provide visitors with Internet connectivity for their mobile devices.
The devices, systems, and methods described herein are directed to using data associated with a wireless network operational area to train a machine learning model, where the data includes one or more signal characteristic measurement values. The trained machine learning model is used to build a signal propagation model for the wireless network operational area. In some examples, the signal propagation model may be used to generate a network performance visualization or a heatmap of the wireless network operational area. In further examples, the system may generate one or more recommendations to optimize one or more performance indicators of the wireless network.
Wi-Fi radio bands have relatively high absorption and work best for line-of-sight use. Many common obstructions such as walls, floors, ceilings, pillars, doors, electronic devices, and appliances may interfere with signal propagation through a network operational area, which reduces the effective range of the Wi-Fi network. For example, a Wi-Fi access point range is about 20 m (66 ft) indoors, while the access point range may be as high as 150 m (490 ft) outdoors.
Since Wi-Fi networks are very widely deployed in different environments, one of the most important use cases is the optimization of the performance and capacity of the existing networks. In order to optimize network performance, some existing solutions require detailed measurements of the network operational area, calibration of wall attenuation values, special network planning expertise, and a large amount of manual work to properly analyze, evaluate, and configure existing networks.
For example, some existing solutions require manual configuration of the floor plan in the network operational area. The first step of this type of manual optimization is to define the physical environment. This requires that the user uploads the floor plan and manually inserts any physical obstacles, such as walls, doors, windows, and large furniture into the floor plan. Some of the information, such as wall positions, may be read from digital floor plans, but some information must be inserted manually. In the event that digital floor plans are not available, all obstacles are required to be manually inserted into the floor plan, which is time consuming, inefficient, and lends itself to errors if the user enters incorrect information regarding the obstacles in the network operational area.
The challenges of properly entering the floor plan are compounded when the network operational area involves multiple floors within a building. For example, in multi-floor buildings there are separate floor plans for each floor. Thus, additional manual work is required to correctly position floor plans for each of the different floors on top of each other and with the correct orientation. This problem can be exacerbated when the floor plan images available to a user have scaling or orientation that is not the same for each floor.
Another significant challenge to optimizing Wi-Fi networks with existing solutions is the difficulty in properly calibrating the radio frequency (RF) environment. For example, modeling existing networks with current solutions is labor intensive and time consuming. Accurate manual network planning takes a long time because it is not possible to know how structures, such as walls, doors, shelves, and furniture, will attenuate RF signals without actually measuring the attenuation. In order to determine an accurate attenuation value, each of the structures in the network operational area should be separately measured. Once the attenuation values are known, the user needs to configure the corresponding value to the network plan in the planning tool. However, even if all the structures are separately measured, the propagation calculation would still rely on the theoretical free space path loss (FSPL) model, which is a good estimation in open spaces, but it does not account for the attenuation caused by obstacles or building structures.
Besides the foregoing challenges, it can be difficult to manually define the requirement area, according to existing network planning solutions. In this regard, the goal of defining the requirement area is that coverage will be planned for the defined requirement area. However, in some cases, the requirement area is not defined at all, or survey measurements are not done for the entire requirement area. Both of these cases lead to problems in evaluating network performance since there is a mismatch between the intended requirement area and the planned network.
Existing solutions also require a high level of domain area expertise. For example, the RF environment within the network operational area can frequently change and, therefore, needs to be frequently measured and optimized. Such optimization of the existing Wi-Fi network requires a high level of domain area expertise. Users that are maintaining existing networks are not typically RF experts, which means that they need to find expensive external help when the network is not performing properly.
Additionally, the required quality of the survey performed in the network operational area is not clearly defined and can vary depending on the needs of the owner/operator of the wireless network. Good practice is that there should be survey measurements in every room/floor of the network operational area. In existing solutions, survey measurements are accurate in the proximity of the survey path, but the accuracy of the survey measurements decreases as the distance from the survey path increases.
Even if all the foregoing issues are addressed, it is not easy to propose a new network configuration, which may require moving one or more access points to a new location or predicting signal propagation characteristics for one or more newly added access points.
The devices, systems, and methods described herein are directed to using data associated with a wireless network operational area to train a machine learning model, where the data includes one or more signal characteristic measurement values. The trained machine learning model is used to build a signal propagation model for the wireless network operational area. In some examples, the signal propagation model may be used to generate a network performance visualization or a heatmap of the wireless network operational area. In further examples, the system may generate one or more recommendations to optimize one or more performance indicators of the wireless network.
Although the different examples of devices, systems, and methods for optimizing wireless networks may be described herein separately, any of the features of any of the examples may be added to, omitted from, or combined with any other example. Similarly, any of the features of any of the examples may be performed in parallel or performed in a different manner/order than that described or shown herein.
Wireless network optimization system 100 includes communication interface 108, controller 110, display 112, and transmitter 114 of local computing device 102. In operation, local computing device 102 receives data from measurement device 104 via communication link 106. Communication interface 108 enables communication between measurement device 104 and local computing device 102. In the example shown in
In the example shown in
Measurement device 104 processes the received Wi-Fi signals to obtain data associated with the wireless network operational area. In some examples, the data includes one or more signal characteristic measurement values. For example, Received Signal Strength Indicator (RSSI) measurement values may be included in the data. Of course, any other suitable signal characteristics may be measured and included in the data associated with the wireless network operational area. For example, the signal characteristic measurement values may include other measurements of signal strength and/or quality. In some examples, the signal strength measurement may include a measurement of the multi-path properties and/or the phase of the received Wi-Fi signals. The data received from measurement device 104 may also include the location of wireless network access points and RF measurements taken at different locations in the wireless network operational area along a survey path, which is a route taken by a user through the wireless network operational area while carrying/moving measurement device 104 in order to obtain the data.
Measurement device 104 provides the data associated with the wireless network operational area to local computing device 102 via communication link 106. Local computing device 102 uses controller 110 to train a machine learning model, based at least partially on the data received from measurement device 104. Different machine learning methods, such as a gradient boosted random forest, could be used to train the machine learning model. In this manner, the machine learning model learns the different properties of the wireless network operational area (e.g., walls, other obstacles, overall building structure, and how signals propagate in that environment, including signal reflections and diffractions). In some examples, additional information could be used to train the machine learning model (e.g., the location of walls or other obstacles, transmit power settings of one or more of the access points, and the distance between an access point and the measurement device when measurements were taken during the survey).
Once the machine learning model is trained, controller 110 uses the trained machine learning model to build a signal propagation model for the wireless network operational area. The signal propagation model can estimate the signal propagation between any points in the wireless network operational area. This capability allows the system to simulate the effect of changing various network configurations (e.g., channel configurations, channel widths, or adjusting transmission power or disabling radios of one or more access points) or physical changes (e.g., adding a new access point or moving an existing access point to any other location in the wireless network operational area). Thus, the system can also utilize the signal propagation model to perform Leave-One-Radio-Out (LORO) testing to evaluate the effect of either removing a particular access point from the wireless network operational area or disabling the radio of that particular access point.
Controller 110 includes any combination of hardware, software, and/or firmware for executing the functions described herein. An example of a suitable controller 110 includes software code running on a microprocessor or processor arrangement connected to memory.
In some examples, controller 110 also generates one or more network performance visualizations, based on the signal propagation model. In other examples, controller 110 renders, based on the signal propagation model, a heatmap of the wireless network operational area. In these examples, the rendered heatmap is based on values of one or more performance indicators determined by the signal propagation model. In some examples, different heatmaps may be rendered for different Key Performance Indicators (KPIs) and utilized to propose recommendations for network optimization. These network optimization recommendations may include (1) coverage optimization by adjusting transmit powers, proposing new access points or new locations for existing access points, and/or (2) capacity optimization by reducing channel interference (e.g., by implementing a new channel plan or disabling interfering access points), proposing new access points, or proposing one or more other types of antennas to be used by one or more of the access points.
In any of these examples, display 112 may be used to display the performance visualizations and/or the heatmap to a user. In this manner, the user can easily see where the network operational area has varying degrees of signal strength/coverage.
In further examples, controller 110 generates one or more recommendations to optimize one or more performance indicators of the wireless network. In some examples, a performance indicator may be a signal-to-noise ratio (SNR). Of course, any other suitable performance indicator may be utilized, in other examples.
In still further examples, controller 110 estimates a location of one or more access points in the wireless network operational area, based at least partially on one or more Received Signal Strength Indicator (RSSI) measurements included in the data received from measurement device 104. In these examples, controller 110 trains the machine learning model, based at least partially on the estimated location of the one or more access points.
In some examples, controller 110 trains the machine learning model, based at least partially on a transmit power of one or more access points in the wireless network operational area. In some examples, an access point may indicate its transmission power in its beacon frames. In other examples, the transmission power may be estimated by measuring the received Wi-Fi signals.
In other examples, controller 110 trains the machine learning model, based at least partially on a distance of measurement device 104 from one or more access points in the wireless network operational area when the data is obtained. In some examples, a user may provide, to the system, the distance/location of the measurement device relative to one or more access points while receiving the Wi-Fi signals. In other examples, the measurement device is configured to automatically detect the distance/location of the measurement device relative to one or more access points while receiving the Wi-Fi signals. In these examples, the distance/location of the measurement device relative to the access point is automatically calculated once the location of the access point has been estimated.
In some examples, the signal propagation model covers a requirement area larger than a survey path along which the data is obtained. For example, although only measurements from the survey path are taken, training the machine learning model extends the propagation calculation to cover the full wireless network operational area beyond the survey path and allows generation of a heatmap for the full area, even in cases when the entire wireless network operational area is not fully covered by the survey.
In some examples, controller 110 may determine the boundaries of the requirement area, based at least partially on a convex hull of the data. In this regard, a convex hull of a set of points is understood to mean the smallest convex polygon that encloses all of the points in the set. Convex means that the polygon has no corner that is bent inwards. As used herein, a convex hull of the data refers to a convex hull of (1) all points at which signal measurements were taken during a survey of the wireless network operational area, (2) known or estimated access point locations, or (3) a combination of measurement locations and access point locations. Automatically defining the requirement area in this manner ensures that the requirement area corresponds with the survey measurements, and network performance evaluation can be done correctly.
In some examples, controller 110 may automatically generate a floor map of the wireless network operational area, based on the data received from measurement device 104. In other examples, a user may have the option of entering an existing floor map into local computing device 102, which will automatically detect walls from the entered floor map images. The wall information is then used by controller 110 to train the machine learning model.
In some examples, controller 110 determines, based at least partially on one or more Received Signal Strength Indicator (RSSI) measurements included in the data, a recommended arrangement and/or alignment of a plurality of floors of the wireless network operational area relative to each other. Thus, based on signal strength measurements, floor maps can be automatically arranged in correct order from the bottom floor to the top floor (e.g., in a z-coordinate) of the wireless network operational area. In some examples, the floor map alignment may also include aligning the floors in the x-y plane. In some examples, floor map alignment could also be automated with machine learning techniques (e.g., with image registration to find the optimal transformation (e.g., affine, rotate, shift, zoom etc.)) to maximize a certain metric (e.g., Intersection over Union, dice score, etc.) and align floor maps correctly on top of each other.
In some examples, the wireless network optimization system further comprises a display to display the recommended arrangement of the plurality of floors to a user for approval. Thus, even though automatic floor alignment would be possible in most cases, it would be advantageous to allow a user to approve, modify, or reject the recommended alignment/arrangement of floor maps.
In some examples, controller 110 estimates a quality of the data, based at least partially on a number and a quality of survey measurements taken when performing a survey of the wireless network operational area. Estimating the quality and the amount of survey measurements permits an estimate of the accuracy of the trained machine learning model.
In the example shown in
The foregoing examples utilize machine learning techniques to automatically build a model of the wireless network operational area based on the survey measurements received from a measurement device. By training the machine learning model and utilizing the signal propagation model as described above, the system can generate automated network optimization recommendations to improve existing network performance, capacity, and coverage. Moreover, the foregoing examples eliminate the need for manual definition of the physical environment, eliminate the definition of attenuation values and wall calibration, eliminate manual planning and optimization, do not require special network planning expertise, and provide higher accuracy than currently used FSPL-based propagation models.
Remote computing device 202 receives the data via communication interface 206 and provides the data to controller 208, which trains the machine learning model and builds the signal propagation model, as described in connection with controller 110 of
Clearly, other examples and modifications of the foregoing will occur readily to those of ordinary skill in the art in view of these teachings. The above description is illustrative and not restrictive. The examples described herein are only to be limited by the following claims, which include all such examples and modifications when viewed in conjunction with the above specification and accompanying drawings. The scope of the foregoing should, therefore, be determined not with reference to the above description alone, but instead should be determined with reference to the appended claims along with their full scope of equivalents.
