WO2021016078A1

WO2021016078A1 - Predictive weather-aware communication network management

Info

Publication number: WO2021016078A1
Application number: PCT/US2020/042524
Authority: WO
Inventors: Jonathan OSTROMETZKY; Gil Zussman; Hagit Messer-Yaron; Dror JACOBY
Original assignee: Ramot at Tel Aviv University Ltd; Columbia University in the City of New York
Current assignee: Ramot at Tel Aviv University Ltd; Columbia University in the City of New York
Priority date: 2019-07-19
Filing date: 2020-07-17
Publication date: 2021-01-28
Anticipated expiration: 2022-01-19
Also published as: US12556937B2; EP4000225A4; EP4000225A1; US20220110012A1

Abstract

Disclosed are methods, systems, devices, and other implementations, including a method that includes obtaining measurement data from at least part of a wireless communication network during a first interval of time, with the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network. The method further includes determining, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time, and managing resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval.

Description

PREDICTIVE WEATHER-AWARE COMMUNICATION NETWORK MANAGEMENT

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to, and the benefit of, U.S. Provisional Application No. 62/876,337, entitled“PREDICTIVE COMMUNICATION NETWORK MANAGEMENT BASED ON WEATHER” and filed July 19, 2019 the content of which is incorporated herein by reference in its entirety.

STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH

[0002] This invention was made with government support under Grant No. CNS- 1910757 by the National Science Foundation (NSF). The government has certain rights in the invention.

BACKGROUND

[0003] Currently, 5G, the next generation of broadband communication networks, is being developed in order to accommodate faster speeds, lower latencies, and larger capacity capabilities. To accomplish this, higher radio frequencies, such as the millimeter- wave

(mmWave) band (28-300GHz) are being developed for use. However, the current challenges with mmWave include higher path loss, reduced scattering, increased noise, and increased blockage potential. mmWave technology is extremely susceptible to moisture in the atmosphere, resulting in rain-induced attenuation and network blockage.

[0004] Cellular providers have traditionally relied on a vast number of Commercial Microwave Links (CMLs) in significant parts of their backhaul networks. These networks of CMLs usually operate at frequencies of 6-40 GHz which are around the K-band frequency range, connect base stations, and cover entire countries. Recently, the use of CMLs in millimeter-wave (mmWave) frequencies of 60-90 GHz, which are part of the E-band range, has gained popularity in smart cities. In smart-cities various applications require ultra high bandwidth and low latency (e.g., HD/4K video camera connectivity, public WiFi backhaul, public safety network, and smart grid) and many locations are beyond the reach of the fiber plant. Newer CMLs currently operate at frequencies of 6 GHz to 90 GHz, and frequencies above 100 GHz will become available in the future.

[0005] One problem associated with such communication networks is their susceptibility to atmospheric phenomena, which attenuate the transmitted radio signals, with precipitation (i.e., rain) being the dominant factor. Although higher frequencies can support higher throughput, as the frequency increases, so does the weather-induced attenuation. On the other hand, supported smart-city mission-critical applications (e.g., traffic cameras) require robust solutions and cannot withstand significant outages.

SUMMARY OF THE INVENTION

[0006] The present disclosure describes solutions for controlling the wireless network in a predictive manner. By using measurements performed by wireless nodes to obtain, for example, standard CMLs’ Tx and Rx measurements, the intensity and location of the attenuation-induced phenomena can be estimated. This information may be implemented via data-driven machine learning tools to extrapolate the near-future properties of the attenuation-caused factors in order to reconstruct the attenuation affecting the CMLs in the vicinity of the detected interference. Based on the attenuation prediction techniques, using both classical models such as the ARIMA (autoregressive integrated moving average) process, as well as machine learning processes, a set of prediction-based network control and management procedures can be implemented. These cross-layered processes span the physical, link, and the network layers, and are configured to react in a predictive manner, in real-time (or near real-time), prior to imminent onset of, for example, rain-related network disturbances. The presented approaches would be used to activate link-level solutions (such as the ATPC and the AMC techniques / algorithms), as well as to dynamically reshape the network topology dynamically, in environmentally-aware management schemes. This, in turn, can increase the future networks Quality of Service (QoS) dramatically.

[0007] Current and future mission-critical applications cannot sustain even instantaneous performance degradation, and therefore, predictive and dynamic networkwide processes are required. The approaches described herein achieve mechanisms to predict future network-wide link states from CML-obtained measurements, which can be used in relation to smart cities connectivity and 5G networks. The approaches described herein focus on backhaul and fronthaul networks which are currently transitioning to E-band (60 - 90 GHz) links, as well as new technology that are very sensitive to rain events. Unlike legacy (4G) cellular networks where local physical layer adaptation has been sufficient, in emerging smart city and 5G networks (that will require low latency and high bandwidth), link and network layer adaptations will be important. Thus, the approaches described herein provide for: 1) the establishment of a paradigm that uses the WCNs self-extracted attenuation measurements as sufficient data

(although that data may be supplemented to achieve higher accuracy model) for the prediction of the link-channels states throughout the network, backed by the known relationships between, for example, the weather and the signal attenuation, and 2) presentation of a weather-sensitive cross layered control processes / algorithms that can optimize network performance by controlling resources such as power, modulation, coding, channel allocation, and channel near real-time switching (that is, the propose approach can satisfy QoS requirements in response to predicted changes in network conditions, taking advantage of WCNs capabilities such as PAAM hardware use and IAB concepts). The approaches described herein are demonstrated in a typical scenario based on actual city-scale real-world WCNs measurements.

[0008] The approaches described herein can be used to realize stand-alone techniques that use past physical measurements available in the network management system, in order to optimize future performance of a multi-links (or multi nodes) network. Such optimization can support applications that require a fixed-level QoS (that is, one that does not change even if the environment conditions change). The approaches described herein realize the stand-alone solutions in the sense that no external information is needed (although such external information could be used to supplement the physical measurement data). The approaches and solutions described herein can use measurements from several links to infer about the future state of other links in the same area (and thus, it is not necessary to have access to measurements from all links). The approaches and solutions described herein can be implemented through a

combination of one or more of the various techniques / processes described in greater detail below.

[0009] Accordingly, in some variations, a method is provided that includes obtaining

measurement data from at least part of a wireless communication network during a first interval of time, with the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network. The method further includes determining, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time, and managing resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval.

[0010] Embodiments of the method may include at least some of the features described in the present disclosure, including one or more of the following features.

[0011] Obtaining the measurement data may include measuring signal attenuation levels, over the first interval of time, at one or more of wireless links of the wireless communication network.

[0012] Measuring the signal attenuation levels may include measuring the signal attenuation level at regulars or irregular time periods within the first interval of time, with the signal attenuation level for a wireless node, m, at a time t within the first interval being computed as a difference between a transmitted signal power level (TSL) for a signal transmitted from a remote wireless node, n_j, and a received signal power level (RSL) for the signal received at the wireless node m.

[0013] Determining the predictive data may include determining the predictive data based on measured signal attenuation levels at the one or more wireless links, and further based on characteristics associated with two or more wireless nodes of the wireless communication network. The characteristics may include, for each of the two or more wireless nodes, available meta data, including one or more of, for example, geographic location, a frequency at which signals are transmitted, polarization attributes of the signal, coding technique applied to the signal, modulation technique applied to the signal, and/or adaptability of the two or more wireless nodes to adjust transmission power and/or adjust signal modulation.

[0014] Determining the predictive data may include determining the predictive data using a learning engine, trained using attenuation time- series data collected from multiple wireless links defined between one or more pairs of wireless nodes of the wireless communication network during dry and wet periods, applied to the measurement data obtained for the wireless communication network. [0015] Determining the predictive data may include determining the predictive data representative of future attenuation behavior for wireless links defined between wireless nodes of the wireless communication network for H subsequent time instances following the end of the first interval.

[0016] Determining the predictive data using the learning engine may include deriving, using a Long Short-Term Memory (LSTM) encoder network, a fixed-length hidden state vector for a time instance in the first interval based on the measurement data obtained for the wireless communication network, and on a previous state vector computed for a previous set of measurement data, the LSTM encoder implementing a non-linear activation function to compute the hidden state vector. Determining the predictive data using the learning engine may further include applying an LSTM decoder to the hidden state vector to determine the predictive data representative of future attenuation behavior for the wireless links of the wireless communication network for the H subsequent time instances.

[0017] Managing resources of the communication network may include determining status of operation data for at least some wireless links defined for one or more pairs of the wireless nodes of the communication network, the status of operation data representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval.

[0018] The method may further include controlling operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available. Controlling the operational characteristics may include one or more of, for example, increasing transmission strength of the first wireless node, and/or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node.

[0019] The method may further include identifying a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance, and identifying an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance. The alternative link may be selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

[0020] Managing resources of the wireless communication network may include deriving throughput values between pairs of wireless nodes of the wireless communication network, to optimize the overall throughput at the future time instance of data traffic to and from one or more sink nodes of the wireless communication network, subject to predicted capacities of the wireless links of the communication network computed according to the predictive data of the future performance of the wireless communication network at the one or more future time instances following the end of the first interval, and configuring one or more of the wireless nodes based on the derived throughput values.

[0021] Deriving the throughput values between the pairs of wireless nodes of the wireless communication network may include deriving the throughput values between the pairs of wireless nodes of the wireless communication network subject to further constraints relating to pre-determined characteristics of the wireless communication network.

[0022] The pre-determined characteristics of the communication network may include one or more of, for example, known capacities of the wireless links, and/or known data traffic demand.

[0023] Managing the resources of the communication network based, at least in part, on the determined predictive data may include one or more of, for example: a) configuring one or more wireless nodes of the wireless communication network based on the predictive data, subject to the requirement that at least one identified wireless node maintains a specified quality-of- service (QoS) level, and/or b) switching between different channels configured for one or more wireless links based on the determined predictive data and on priority of services provided at one or more wireless nodes for which the one or more wireless links are established.

[0024] Determining the predictive data may include determining ambient conditions affecting the future performance of the communication network.

[0025] Determining ambient condition may include determining predicted weather patterns affecting the performance of the wireless communication network, including estimating spatial- temporal rain attenuation patterns predicted to impact the communication network. [0026] The measurements indicative of potential disturbances to operation of the wireless communication network may be indicative of weather-related disturbances.

[0027] In some variations, a system is provided that includes a communication module to obtain measurement data from wireless nodes, forming at least part of a wireless communication network, during a first interval of time, the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network, and a processor-based resource management controller coupled to the communication module. The controller is configured to determine, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time, and manage resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval.

[0028] Embodiments of the system may include at least some of the features described in the present disclosure, including at least some of the features described above in relation to the method, as well as one or more of the following features.

[0029] The system may further include the wireless nodes forming the at least part of the wireless communication network.

[0030] The controller configured to determine the predictive data may be configured to determine the predictive data based on measured signal attenuation levels at the one or more wireless links, and further based on characteristics associated with two or more of the wireless nodes of the wireless communication network, the characteristics including, for each of the two or more wireless nodes, available meta data, including one or more of, for example, geographic location, a frequency at which signals are transmitted, polarization attributes of the signal, coding technique applied to the signal, modulation technique applied to the signal, and/or adaptability of the two or more wireless nodes to adjust transmission power and/or adjust signal modulation.

[0031] The controller configured to determine the predictive data may be configured to determine the predictive data using a learning engine, trained using attenuation time-series data collected from multiple wireless links defined between one or more pairs of wireless nodes of the wireless communication network during dry and wet periods, applied to the measurement data. The predictive data may be representative of future attenuation behavior for wireless links defined between wireless nodes of the wireless communication network for H subsequent time instances following the end of the first interval.

[0032] The controller configured to manage resources of the communication network may be configured to determine status of operation data for at least some wireless links defined for one or more pairs of the wireless nodes of the wireless communication network, the status of operation data representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval. The controller may additionally be configured to perform one or more of, for example: i) control operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available, with the controller configured to control the operational characteristics being configured to cause one or more of, for example, increasing transmission strength of the first wireless node, and/or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node, and/or ii) identify a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance, and identify an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance, the alternative link selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

[0033] The controller configured to manage resources of the wireless communication network may be configured to derive throughput values between pairs of wireless nodes of the wireless communication network, to optimize the overall throughput at a future time instance of data traffic to and from one or more sink nodes of the wireless communication network, subject to predicted capacities of the wireless links of the communication network computed according to the predictive data of the future performance of the wireless communication network at the one or more future time instances following the end of the first interval. The controller may further be configured to configure one or more of the wireless nodes based on the derived throughput values.

[0034] In some variations, non-transitory computer readable media is provided, that includes computer instructions executable on a processor-based device to obtain measurement data from at least part of a wireless communication network during a first interval of time, with the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network, determine, based on the obtained

measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time, and manage resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval.

[0035] Embodiments of the computer readable media may include at least some of the features described in the present disclosure, including at least some of the features described above in relation to the method and to the system.

[0036] Other features and advantages of the invention are apparent from the following description, and from the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[0037] These and other aspects will now be described in detail with reference to the following drawings.

[0038] FIG. 1 is a diagram of an example network that can be controllably and dynamically configured to mitigate predicted network performance degradation.

[0039] FIG. 2 is a map of an actual CMLs network that is a part of an actual cellular network in Gothenborg, Sweden.

[0040] FIG. 3 is network connectivity diagram corresponding to the CML network of FIG. 2. [0041] FIG. 4 is a schematic diagram of an LSTM encoder-decoder architecture to determine multi-step predictive performance data for a wireless communication network for multiple future time instances.

[0042] FIG. 5 is a graph illustrating a sliding-window concept for preparing a multivariate time series input sequences for a supervised learning task.

[0043] FIG. 6 is a diagram showing a predicted changing status of operation for five wireless links.

[0044] FIG. 7 includes diagrams of an example of a network-level rerouting approach, based on the status of operation classification illustrated in FIG. 6.

[0045] FIG. 8 is a diagram of an example network on which a dynamic switching process is performed.

[0046] FIG. 9 is a diagram illustrating a single-step solution for a maximum concurrent flow problem (MCFP) for a wireless communication network.

[0047] FIG. 10 is a flowchart of an example procedure to manage wireless resources of a weather-impacted wireless communication network.

[0048] FIG. 11 are scatter plots providing cross validation for actual and predicted values for minimal and maximal prediction times for three selected network edges (links).

[0049] FIG. 12 is a graph showing averaged RMSE_h values for each of three tested events, for tested edges, as a function of prediction time D7]_¾.

[0050] FIG. 13 are graphs showing the distributions of residuals during a rain event for different values of h.

[0051] FIG. 14 are graphs showing critical values as a function of g, for three tested events and for different prediction times.

[0052] Like reference symbols in the various drawings indicate like elements.

DESCRIPTION [0053] Disclosed are systems, methods, and other implementations for predictive management of backhaul wireless communication network infrastructure (e.g., via mmWave networks for 5G networks, 4G networks, and all other types of wireless networks). The systems and other implementations described herein use prediction models to determine future potential attenuations (caused by weather conditions, or by other factors) at various nodes, and respond accordingly using adaptive nodes (whose communications characteristics, including transmission power, coding schemes, etc., can be adjustably controlled), adaptive arrays, link rerouting, and network management controls to compensate for predicted attenuation / path loss. The implementations described herein are configured to incorporate weather-information obtained from the networks’ own measurements of link characteristics (e.g., signal strength). This information is used to estimate performance degradation of microwave links between at individual nodes of the network, and to manage network use (traffic) accordingly. The measured information can also be used to estimate, extrapolate, and predict the rain-field in a high-tempo- spatial resolution, in near real-time, while using the cellular providers in-house management- tools measurements. These predictions, information regarding the current link states, information regarding the current and predicted traffic statistics, and traffic classes can serve as inputs for a weather-aware (e.g., rain-aware) dynamic network adjustment procedures. The network management procedures described herein can be based on a combination of: (i) using the in- house available networks’ measurements, (ii) predictions regarding future channel states, and/or (iii) real-time link-level and network-level predictive adaptations.

[0054] In some examples, by monitoring the Tx and Rx signal-level at instant n of the available links, the signal attenuation at the links, as well as near-by links, at future time instances (greater than the instance n ) due to weather-related phenomena (or other environmental or human factors) can be predicted. Such prediction data is, in turn, used to predict the future state of the wireless network, and to prepare the network for imminent disturbances, ahead of time. The

implementations described herein are configured to controllably adjust the network’s behavior / performance based on predicted / expected environmentally-related (e.g., weather-induced) attenuation, using a combination of local link solutions, such as automatic transmit power control and adaptive modulation coding, as well as dynamic channel switching, if required, to maintain optimal throughput possible. The approaches described herein provide a higher Quality of Service for the network, which is crucial for mission-critical future applications (e.g., smart cities critical infrastructure and autonomous car support). As will be discussed in greater detail below, experimentations and testing conducted for the approaches described herein, using a set of simulated scenarios which are based on real-world data collected by a city-scale 4G cellular network, show that the present approach can be implemented as a predictive network

management system in real time.

[0055] Thus, with reference to FIG. 1, an example network 100 that can be controllably configured to mitigate predicted performance degradation (e.g., increased expected signal attenuation in wireless links) as a result of environmental factors (e.g., weather related factors, human factors, etc.) is shown. As illustrated, the network 100, includes multiple wireless nodes 1 lOa-n deployed across a geographic area that can be of any size (city-sized area, county-sized area, a large rural region, etc.) and as such the various wireless nodes may be affected by different ambient conditions (e.g., different weather conditions). For illustration purposes only, the wireless nodes 110c and l lOd are shown as being located in areas currently experiencing rain condition (with the device 110c located in an area experiencing more intense rain than the area where the wireless node 1 lOd is located), the wireless nodes 110a, 110b, and 1 lOe are located in areas that are currently partly overcast, while the wireless node 110h (n represents the total number of devices within the network 100 that can be controllably configured, and can be any integer number) is depicted as being located in area currently experiencing sunny conditions. Among other factors, precipitation (rain or snow) at a particular geographic area can impact the quality of transmission (causing signal degradation), but other ambient factors (temperature, barometric pressure, and so on) can also impact the quality of wireless transmission. As will be discussed in greater detail below, weather conditions can be used to model predicted wireless transmission behavior, and thus predict the network behavior at future points of time. While actual weather data may be used to control / manage the network (make preemptive adjustments to transmission power or throughput at individual nodes, or preemptively re-route various links), current attenuation behavior measured for currently operating links (and their variations from the baseline points) can be used instead of, or in addition to, actual weather data, to derive predictive data representative of the future behavior of wireless links. The prediction of wireless links’ behavior can be short range, e.g., for time instances in the near future, or longer range, e.g., over several time instances spanning a relatively long period of time such as 1 hour, or 1 day. [0056] The wireless nodes 1 lOa-n may be any type of wireless device. In the example embodiments described herein. The wireless nodes establish commercial microwave links (CMLs) as part of the cellular backhaul underpinning the network. Such CMLs typically operate at frequencies of 6 - 40 GHz which are around the K-band frequency range, connect base stations, and cover entire countries. The wireless nodes may, in some embodiments, implement back-haul links at frequencies of 60 - 90 GHz, which are part of the E-band range (such links have gained popularity in smart cities). Other types of backhaul communication solutions can also be used. The wireless nodes 1 lOa-n may include radio access network interfacing circuitry (to implement localized transmissions from the nodes to user equipment (UE) devices in the cellular area covered by the particular node according to any type of communication protocol or technology, including 2G, 3G, 4G, 5G, and various other communication technologies in licensed and unlicensed bands. For example, any of the wireless nodes 1 lOa-n may be configured to establish WLAN-type communication channels with a UE, a WWAN-type connection, or a PWAN-type connection (e.g., a short-range personal wireless access network).

A WWAN may be a Code Division Multiple Access (CDMA) network, a Time Division

Multiple Access (TDMA) network, a Frequency Division Multiple Access (FDMA) network, an Orthogonal Frequency Division Multiple Access (OFDMA) network, a Single-Carrier Frequency Division Multiple Access (SC-FDMA) network, a WiMax (IEEE 802.16), or any other 3GPP or IEEE standards (implemented over licensed and unlicensed frequency bands).

[0057] As further illustrated in FIG. 1, one or more of the various wireless nodes (and in some embodiments, all the wireless nodes) may establish a communication channel with a server 120 configured to control / manage the communication resources of the network 100. As will be discussed in greater detail below, the server 120 (e.g., a processor-based computing device) may be configured to obtain measurement data (via a communication module, schematically shown as the module 122, which may be a wireless transceiver and/or a network transceiver, e.g., to interface with wired network connections, electrically coupled to processor-based computing device) during a first interval of time. For example, the server 120 may obtain samples, from one or more of the wireless nodes 1 lOa-n, at a plurality of time interval, spaced at regular or irregular sub-intervals in the first interval. The server 120 may obtain samples representative of measurement data of signal levels at the nodes defining the end-points of a communication link. For example, the measured attenuation level for a wireless node, m, at a time t within the first interval is computed as a difference between a transmitted signal power level (TSL) for a signal transmitted from a remote wireless node, n_j, and a received signal power level (RSL) for the transmitted signal at the wireless node m. As noted, weather conditions, for example the occurrence of rain at some of the wireless nodes, will affect the channel quality, resulting in greater attenuation of a signal transmitted over a channel between the node 110c (shown transmitting signals at a power level of TSL) and the node 1 lOd (shown receiving the signal with a power level of RSL) than would be the case under normal circumstance (e.g., at some pre determined baseline of ambient conditions). In contrast, attenuation levels (or performance degradation) may be closer to normal / expected level for signals transmitted over a channel established between nodes 110a and 110b (or even for a signal transmitted between node 110b and node 110c, where only part of the path will be subject to inclement weather conditions). In some embodiments, the communication module 122 may be configured to measure signal strength (e.g., RSL) of signals received from other nodes (e.g., any of the nodes l lOa-n), and/or transmit signals whose strength can be measured or set. The server 120 can then use locally measured signals (the server 120 and/or its associated communication module 122 may be implemented at one or more of the wireless nodes 1 lOa-n) to determine predictive data. In some embodiments, the predictive data may be generated based entirely on measurements data measured by the server 120 (for signals transmitted from one or more remote wireless nodes). Thus, in some situations, the onboard hardware for the server 120 (including the module 122) may be able to measure data with sufficient temporal resolution to determine predictive data for future performance of at least part of the network 100 in order to manage the network’s resources.

[0058] Having obtained the measurement data, the server 120 is configured to determine, based on the obtained network measurement data, predictive data representative of future performance of the communication network comprising the wireless nodes at one or more future time instances subsequent to an end of the first interval of time. The predictive data may be determined using, for example, a learning engine, trained using attenuation time-series data collected from multiple wireless links defined between one or more pairs of the wireless nodes of the communication network during dry and wet periods, applied to the measurement data obtained for the communication network. In some embodiments, the predicted data may be derived using other types of optimization processes (e.g., ARIMA schemes), or filtering techniques. Such predictive data may thus represent future attenuation behavior for wireless links defined between the one or more pairs of the wireless nodes of the communication network for H subsequent time instances following the end of the first interval. An example of a learning engine (as will further be discussed below) is a Long Short-Term Memory (LSTM) encoder network that computes a fixed-length hidden state vector for a time instance in the first interval based on the measurement data obtained for the communication network, and on a previous state vector computed for a previous set of performance data, with the LSTM encoder implementing a non-linear activation function to compute the hidden state vector. Such a learning engine also includes an LSTM decoder that is applied to the hidden state vector to determine the predictive data representative of future attenuation behavior for the wireless links of the communication network for the H subsequent time instances.

[0059] Based on the predictive data, the server 120 is configured to manage the resources of the communication network for some period of time following the end of the first time interval during which measurements representative of the performance (e.g., attenuation levels) of the network 110 were taken. The management of the communication network 100 may include determining status of operation data for at least some wireless links defined for one or more pairs of the wireless nodes of the wireless communication network 100, with the status of operation data representative of likely degrees of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval. For example, the server 120 (or some other controlling circuitry) may control operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available. Such controlling may include increasing transmission strength of the first wireless node, and/or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node.

Alternatively or additionally, the status of operation data for the various wireless nodes (at one or more future time instances) determined by the server 120 (based on the predictive data) may be used to identify a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable at a subsequent time instance or to be operating at a lower (reduced) capacity, and to identify an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or operating at a reduced capacity at the subsequent time instance. The alternative link selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at a reduced capacity.

[0060] In some embodiments, the server 120 may be configured to implement an optimal network topology switching procedure. In such embodiments, the server 120 may be configured to derive throughput values for wireless links between pairs of wireless nodes of the

communication network, to optimize overall throughput of data traffic to a sink node of the communication network at a future time instance, subject to constraints relating to pre determined characteristics of the communication network (e.g., known capacities of the wireless links, or known data traffic demand), and to configure one or more of the wireless nodes based on the derived throughput values. In deriving the throughput values, the server may derive throughput values between the pairs of wireless nodes of the communication network, to optimize the overall throughput at a future time instance of data traffic to a sink node of the communication network, subject to predicted capacities of the wireless links of the

communication network computed according to the predictive data of the future performance of communication network at the one or more future time instances following the end of the first interval. In some additional examples, the server 120 configured to manage the resources of the communication network based, at least in part, on the determined predictive data is adapted to configure one or more of the wireless nodes of the communication network based on the predictive data subject to the requirement that at least one identified wireless node maintains a specified quality-of-service (QoS) level. Alternatively or additionally, the server may be adapted to switch between different channels configured for one or more wireless links based on the determined predictive data and on priority of services provided at one or more wireless nodes for which the one or more wireless links are established.

[0061] While the server 120 is depicted in FIG. 1 as a stand-alone station or node, the server 120 can be part of any of the various wireless nodes (e.g., it may be housed with the circuitry constituting the wireless node 110h, which is shown as a node with a direct connection to a network 130 (which may be a packet-based network, a telephony network, or any other type of network to communicate signals over long distances). In some examples, the server 120 may be a distributed system in which the various operations are performed at multiple computing / controller devices (with such multiple device housed at one or more wireless nodes 1 lOa-n, or at some other location not specifically illustrated in FIG. 1). The server 120 may be configured to communicate with the wireless nodes either directly (through dedicated separate links to each of the wireless nodes HOa-n), or indirectly through connections passing through intermediate points. Communications links between the server 120 and any of the nodes or the network illustrated in FIG. 1 may be realized as wired or wireless links. The server 120 may thus be configured to establish communication links with one or more of the nodes, to communicate data and/or control signals to those nodes (e.g., control signals to control communication

characteristics such as transmission power, or various other communication parameters affecting operation of the nodes), and receive data and/or control signals from those nodes (including measurement data representative of attenuation levels for wireless links between nodes). The server 120 may also be configured to receive data from other sources (such as a server maintaining current weather information) that it may use when deriving predictive data about the future performance of the communication network 110.

[0062] More specificity and details regarding the operations of the network 100 will now be described.

[0063] As noted, CMLs suffer from induced channel attenuation due to the susceptibility of the microwave signal to weather phenomena. The expected signal attenuation as a function of the frequency due to different atmospheric and weather phenomena is described by the International Telecommunication Union (ITU), with the dominant factor of the CML channel attenuation being rain. The rain induced attenuation can be modeled by a standard Power-Law relationship, which relates the rain rate at time t, with the instantaneous induced channel attenuation (in dB) at that time. This simplified power-law relationship can be used to approximate the rain-induced attenuation for Line-of-Sight (LOS) microwave and mmWave links during the design phase of the cellular networks based on historic statistical meteorological data and the ITU

recommendations (e.g., withstand expected attenuation caused by rain-intensity lower or equal to the 99.99th percentile). However, during strong rainfall and extreme weather events, the weather-induced attenuation will be sufficiently high for some of the CML’ s to become unavailable. These scenarios are becoming more common, as part of the global warming trend results in an increase of extreme weather events. Furthermore, for shorter links in the higher frequencies (E-band) (which are generally located in urban areas), the power law approximation for rain induced attenuation is much less accurate. Moreover, in such scenarios, the dependence of the CML signal level on weather is more complex.

[0064] The model of attenuation in a CML as a function of rainfall is an important tool for the design of legacy backhaul networks. Additionally, standard CMLs that can be viewed as opportunistic sensors for rain monitoring purposes. For example, CMLs can be used to detect and classify rainy periods, distinguishing them from dry periods, and to measure the rainfall accurately in various scenarios. Movement of rain can be predicted and extrapolated using, for example, the advection-diffusion model, optical flow, or tools such as Kalman filters.

[0065] In the approaches and solutions described herein, automated techniques for wireless communication networks management are implemented. Such approaches include wireless communication networks (WCNs) network management systems (NMS) to implement a set of automated processes that, as part of the network management schemes, are designed to adapt the network operations, in real-time, against sudden disturbances and drops in the CMLs signal strength. Two examples of NMSs that deal with the CMLs signal drops using two

local/physical, link-level techniques, include:

1) Automated power transmit control (ATPC) system: This system can increase the transmitted power, Tx, of the CML signal as needed (up-to a pre-defined limit), once the received signal, Rx, drops below a certain predefined threshold.

2) Adaptive modulation and coding (AMC) system: This system automatically switches the modulation of the CML signal based on channel conditions.

[0066] The ATPC and the AMC systems can be implemented as stand-alone processes or as a combined solution. It is worth noting that while that the ATPC will not harm the CML designed throughput, implementation of a different modulation by the AMC in order to maintain a given bit error rate (BER), might reduce the CML throughput (an example would be to drop the channel modulation to a lower QAM scheme). Both the ATPC and the AMC are local-level algorithms / processes, which means that they act on independently on every CML whose hardware support it, based on the specific conditions of the specific CML channel. Both the ATPC and the AMC algorithms can cope with short-term mild to moderate channel fading events. However, more significant or severe fading, which occurs during strong rain events, can result in significant CML throughput decrease, and in extreme cases, the CML might completely fail and become unavailable. Such decreases may not be fully mitigated via local adaptations - once a CML fails, rerouting switching at the network layer could be realized. Some 4G WCNs can be configured to adopt a mesh topology, as are future smart cities and 5G WCNs. Thus, with the support of controllers (e.g., software-defined-systems (SDN)), network-wide rerouting can be realized. Furthermore, specialized hardware, such as the novel phased array antenna module (PAAM), concepts such as the microwave and mmWave integrated access and backhauling (IAB) in 5G introduce new capabilities of real-time ad-hoc CMLs realization. Such approaches of both local physical/link layer-based adaptations and network layer rerouting/switching are based on monitoring the network, and can be implemented herein in a preemptive manner, before performance degradation reaches a point that disrupts the wireless communication network in an unacceptable manner.

[0067] The technology described herein thus provides a weather- sensitive prediction approach for rain attenuation in microwave and millimeter wave (mmWave) networks. The implemented model uses established commercial microwave links to gather data on rainfall monitoring and weather patterns. Machine learning and the cross-layered processes then integrate this data to predict, for example, spatial-temporal rain attenuation patterns. The technology described herein allows providers to predict weather patterns that may interfere with network performance, and preemptively adjust and reroute their systems.

[0068] To help illustrate the approaches described herein for implementing a new predictive network management solution, reference is made to FIG. 2, which shows a map of an actual CMLs network (illustrated as lines, such as example lines 210 and 212, indicating the microwave links between various deployed wireless nodes) which is a part of an actual 4G cellular network 200 in Gothenborg, Sweden. This 4G formation was adapted to create a high connectivity network 300, schematically shown in FIG. 3. As can be seen, the parameters (e.g., location of the nodes, size, etc.) of the high connectivity network 300 resemble the original 4G network 200 of FIG. 2. FIG. 3 shows, for example, the links 310 and 312, which correspond to the links 210 and 212 illustrated in FIG. 2 (not all the links depicted in FIG. 3 are marked in the map-based link layout of FIG. 2). The network 300 represents the available CML network links of FIG. 2 as a mesh-like directional graph. In FIG. 3, the node 13 is the sink node, i.e., a node that provides a connection to a long-distance wired network, such as a packet-based network or a telephony network, and through which data traffic (between the wireless nodes of the network 300 and the wired network) passes. As shown in FIG. 3, the edges 4 5, 4 6, and 10 13 include dual CMLs, and are thus treated as an edge with twice the nominal single CML capacity. For the network 300, the available TSL and RSL measurements collected by the CMLs hardware in this network is used to determine predictive data to dynamically control the network 300. The experimental dataset used is based on actual CMLs measurements and meta-data collected by Ericsson AB. The available data that was included during evaluation and testing of the present implementations described herein includes, apart from the time series of the Tx and Rx per CMLs, also the meta-data of the network, which provides the CMLs’ physical features, such as length, location (longitude and latitude), frequency, and polarization.

[0069] The data set used to derive the predicted data that was tested in the evaluations of the implementations described herein includes records of the transmitted signal power level (TSL) and received signal power level (RSL) measurements (in dB), samples at a time of AT = 10 seconds. The total attenuation level

(in dB) for the time step t in each CML can be computed according to:

where tv, is the transmitter node at one end of a CML link, and n, is the receiver. Other formulations to represent the attenuation level experienced on links, or otherwise to represent the performance (or performance degradation) of the network, may be used.

[0070] Based on attenuation levels computed (e.g., according to the attenuation formulation provided above), prediction data representative of the network performance can be generated.

For example, in some embodiments, a rain-induced attenuation prediction using machine learning approaches may be implemented to derive the predictive data. By using the attenuation time-series data collected from multiple wireless links during dry and wet periods, an artificial neural network (ANN), such as one based on a recurrent neural network (RNN) model, can be trained to predict the sequence of future signals attenuation for each CML given the previous values measured by the network.

[0071] Thus, given the observations (yi, vi, ..., y_t), the goal is to determine (learn) a function /(·) which will output multi- step-ahead predictions for the next H expected attenuation values in each CML (y_t+ 1, ..., _/+//), namely,

[0072] In the above formulation X, represents all the available observations until time t. h l (1, ..., H), and 6) represents arbitrary noise at time step t.

[0073] The learning model implemented may be one employing an Encoder-Decoder model, which is designed as sequence-to- sequence model. A model to solve a multi-step time series forecasting problems can be referred as sequence-to- sequence learning task, where a sequence of values for a range of future interval is predicted.

[0074] In one example embodiment of an RNN-based implementation, a multivariate time series framework for multistep prediction is realized based on an LSTM encoder-decoder architecture. FIG. 4 is a schematic diagram of an LSTM encoder-decoder architecture 400 to determine multi- step predictive performance data for a wireless communication network for multiple future time instances. The architecture 400 uses two LSTM networks, 410 and 420, to generate a future sequence of values.

[0075] Let x_t =

e R^M be the vector attenuation values measured in time step t

(representing the attenuation values measured for each of the CML links of a network, such as the networks 100-300 of FIGS. 1-3). The encoder 410 reads the input sequences

X_t - (x_l,x₁,...,x_t ), and produces an encoded version of the inputs, as a fixed-length vector, through the hidden state vector of the LSTM cell (hi). The hidden state is updated in each time step by: h_t = f_h(h_t_ i, C,)

[0076] The function/_/, is a non-linear activation function, that computes the current hidden state from the inputs and the previous hidden state. At The next LSTM network, the decoder 420, uses this state vector representation to derive the predicted values at future time steps, from the encoded space. A fully connected (FC) layer interprets the decoder’s output sequence and final output layer to predict the H time steps, X_t - (/₊₁,/₊₂,...,/_+ii ) .

[0077] During training, the h norm between the network predictions in each time step and the observed attenuation values Y_t, is used as the objective function:

[0078] The model is trained so as to minimize the objective function over all the available training dataset. In some embodiments, a sliding window scheme may be used to split the training data to sub-sequences of inputs and outputs to modify the multi-step prediction problem into a supervised learning task. Consider an input window size of G, the input to the model in each time step is composed of the T samples of each link. For an output window of size H, the target sequence of the RNN is the next H observations in each link. In FIG. 5, a graph 500 illustrates the concept of the sliding- window used for preparing a multivariate time series input sequences for supervised learning task, where the y-axis is the total attenuation measured in each link. Each time series represents the total attenuation measured in a given CML. Each CML has a different base-line attenuation level, which corresponds to its physical properties such as the length and frequency. A min-max scaling can be used to normalize each feature (i.e., each time series of attenuation) as follows:

[0079] Although the above example embodiment generates the predictive data via an LSTM encoder-decoder model of an RNN-based learning engines, other types / configurations of artificial neural networks may be used in place of the embodiment described herein. Other types of learning engines that may be used to generate predictive performance data include

convolutional neural networks (CNN), and feed-forward neural networks. Feed-forward networks include one or more layers of nodes (“neurons” or“learning elements”) with connections to one or more portions of the input data. In a feedforward network, the

connectivity of the inputs and layers of nodes is such that input data and intermediate data propagate in a forward direction towards the network’s output. Unlike an RNN configuration, there are typically no feedback loops or cycles in the configuration / structure of the feed forward network. Convolutional layers allow a network to efficiently learn features by applying the same learned transformation(s) to subsections of the data. Neural networks can be implemented on any computing platform, including computing platforms that include one or more microprocessors, microcontrollers, and/or digital signal processors that provide processing functionality, as well as other computation and control functionality. The computing platform can include one or more CPU’s, one or more graphics processing units (GPU’s, such as NVIDIA GPU’s, which can be programmed according to, for example, a CUD A C platform), and may also include special purpose logic circuitry, e.g., an FPGA (field programmable gate array), an ASIC (application-specific integrated circuit), a DSP processor, an accelerated processing unit (APU), an application processor, customized dedicated circuity, etc., to implement, at least in part, the processes and functionality for the neural networks, processes, and methods described herein. The computing platforms used to implement the neural networks typically also include memory for storing data and software instructions for executing programmed functionality within the device. Generally speaking, a computer accessible storage medium may include any non-transitory storage media accessible by a computer during use to provide instructions and/or data to the computer. For example, a computer accessible storage medium may include storage media such as magnetic or optical disks and semiconductor (solid-state) memories, DRAM, SRAM, etc.

[0080] The various learning processes implemented through use of the neural networks described herein may be configured or programmed using, for example, TensorFlow (an open- source software library used for machine learning applications such as neural networks). Other programming platforms that can be employed include keras (an open-source neural network library) building blocks, NumPy (an open-source programming library useful for realizing modules to process arrays) building blocks, etc.

[0081] As noted, once the predictive data is generated (e.g., using a learning engine, based on measurement data collected by wireless nodes of the wireless communication network), the wireless network’s resources (e.g., operation, configuration, and routing information for the wireless nodes of the network) can be managed. In some embodiments, status of operation data (representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval) is determined. For example, predicted future attenuation can be used to assign, for each time-step n and for each CML of a given wireless communication network, the following operation status classes:

• Blue: The CML operates normally and supports 100% of the designed throughput. • Green: The CML signal is attenuated to some degree, but, the available ATPC system is capable of increasing the transmitted signal such that the CML still supports 100% of its designed throughput.

• Yellow: The CML signal is strongly attenuated. The CML still supports data

transmission, but only due to the AMC system. Thus, the CML is to operate at a lower throughput than designed.

• Red: The CML is unavailable, due to severe attenuation of the signal. Local link-level solutions (e.g., ATPC and AMC) are unable to counter-effect the rain-induced attenuation destructive effects on the CML.

[0082] An example classification process is provided below.

// Input M is the number of CMLs

// Input Si is the status of the /^th CML (on/off)

// Input ATPCi is the ATPC status of the i^th CML (on/off)

11 Input AMCi is the AMC status of the i^th CML (on/off)

11 Output Li is the state of the 1^th CML

[0083] Other classification schemes to represent the status of operation of links and/or nodes, along with resource management schemes that are based on other classification schemes, may be used instead of, or in addition to, the above-described 4-class (color designated) scheme.

[0084] The specific threshold for each CML, in which its status of operation changes between each of the classes, and in which the local level control processes / algorithms operate (e.g., the ATPC strength, or the AMC current QAM scheme) may be predefined based on the links’ hardware profile, and may implement an hysteresis-like range for the assignment of the different status of operation, in order stabilize the algorithm and prevent rapid fluctuations between different AMC QAM schemes. FIG. 6 is an example diagram 600 showing the predicted changing status of operation for five (M= 5) wireless links, as may have been determined based on predicted performance data derived (e.g., using a learning machine, or via some optimization or filtering procedure) from measurement data provided by the wireless nodes of the

communication network (such as the communication network shown in FIGS. 1-3). Thus, in the example embodiments in which resource management is based on a classification mechanism / scheme such as the one described above and depicted in FIG. 6, resource management may include controlling operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available, with controlling the operational characteristics comprising one or more of, for example, increasing transmission strength of the first wireless node, or reducing throughput for the first wireless node according to adjustments of signal coding and/or modulation techniques used by the first wireless node. [0085] Instead of, or in addition to performing local resource management through control of the operational characteristics of individual wireless nodes, network-level control may also be performed. Consider again the above color-based classification process in which at each time- step, n, the classification process produces for each of the CMLs an operational status for time- steps i Î {n, n + 1, ..., n + j) , where j is the maximal future CMLs performance (e.g., attenuation) prediction time-step. Based on this information, 5G IAB rerouting algorithms, implemented for future time- steps as well, may be used to mitigate network data flow issues when one or more of the wireless links are predicted to be not available or to have reduced-capacity (e.g., links with yellow or red status levels). With reference to FIG. 7, diagrams of an example network-level predictive weather-aware rerouting approach, based on the color-coded status of operation classification described herein, are shown. Diagram 700 illustrates a network of 5 CMLs at t=0. At that time instance, all the CMLs are operational at different states as indicated by their colors. Next, and with reference to diagram 710, the prediction process returns the network state at the next time step: t=l. As can be seen, at that point, L3 is predicted to be unavailable. Therefore, a rerouting to bypass L3 is initiated prior to its disconnection. In this example, there are two possible routes: L2— >L1 (routing option (a) depicted by diagram 730), or L5— >L4 (routing option (b) depicted by diagram 740). No single route is currently (at t=l) preferable to the other. On the other hand, the predictive process is not confined to a single time-step prediction, and can produce a prediction for t=2 (as shown in diagram 720) as well. However, the error of such prediction increases with each future time-step, and, as will be discussed in greater detail below, the accuracy of the prediction as a function of the time can impact the algorithm implementation. In the network state prediction of t=2, it can be seen that L2 is predicted to move to a Yellow state. Therefore, this predictive information will result in the selection of routing option (b), to minimize future throughput reduction. It is to be noted that if the future throughput demand is assumed to be known (or can be estimated to a high degree of accuracy), it is possible to expand the implementation of the proposed network level control process to the more general case in which it operates on the entire range of the CMLs (e.g., allowing switching of every link, including Green and Blue ones), in order to provide an optimal flow solution for the given demands. Thus, in embodiments in which a network level resource management is implemented, the resource management may include identifying a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance, and identifying an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance. The alternative link may be selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

[0086] If it is assumed that the throughput is not sequential - that is, it can be split between more than a single stream of data, it is possible to use all the operational links in a given time-step, including yellow links, and to use the available capacity of each. In a general case situation, in which it is assumed that the demands are known, and that data streams can be split between multiple links, a dynamic switching algorithm, e.g., one based on an adaptation of the maximum concurrent flow problem (MCFP) may be used. FIG. 8 shows a diagram of an example network 800, with respect to which a dynamic switching process is performed, as more particularly described below.

[0087] A network (graph) is defined by G = (V, E), with V vertexes (nodes),V = { v , v ,..., v ] and {E} arcs (edges) (as shown in FIG. 8). Define the requested demand from each of the nodes, v : "i {1, 2,..., N} to the Sink node VN, by D_iN (the sink node may, in some embodiments, be the node that is directly, e.g., through a wired connection, linked to a long-haul network, such a telephony or a packet network; a wireless network, such as the networks described herein, may have more than one sink node). Let be the flow of commodity between the nodes i and N, flowing on the edge e_(k,l). Let c_(k,l) be the (maximum) capacity of the edge e_(k,l)· The concurrent flow throughput factor is defined by 0 £ z £ 1.

[0088] The goal of the dynamic switching process is to maximize z, subject to the following constraints: for every node l 1 < l < N , and for every node i 1 £ i £ N :

N

å f ⁽ i ^, N ⁾

e₍ k l ₎ ^£ c ₍ k _, l ₎ , ^" e ₍ k _, l ₎ ^Î { E } [0089] The solution is the spe

of f ^{(i , N )}

(_{k , l )} , and the achievable maximum value of z.

[0090] If a solution can be found in which z = 1, then the network is able to provide the full required throughput (note that the maximum z which is physically viable is 1). On the other hand, if the solution is such that z < 1, then the full throughput required by the nodes cannot be provided. In that case, since the optimization problem which yielded the maximum value of z was solved, the determined non-zero set of f ⁽ i ^, N ⁾

(_{k , l )} is deemed to be the optimal set (with respect to the MCFP).

[0091] In order to solve the MCFP, standard linear programming-based tools may be used. To illustrate, considered an actual patch taken from a 4G cellular network infrastructure (that patch is located in Gothenborg, Sweden, illustrated in FIG.2, and has an area of roughly 10x10 km², which is a nominal area in which the predictive attenuation process yields accurate predictions). The network topology is similar to the topology depicted in FIG.3. A resultant single-step solution for the MCFP process for the network tested is illustrated for a network diagram 900 of FIG.9. The results provided in FIG.9 were generated through a linear programming procedure, taking into account known demands and link capacities. For the purposes of deriving the solution, the following randomly generated commodity demands between each of the nodes to the sink (node 13) were used: (1) 0.1691, (2) 0.2122, (3) 0.1938, (4) 0.2285, (5) 0.2280, (6) 0.1773, (7) 0.1777, (8) 0.2302, (9) 0.2458, (10) 0.2376, (11) 0.1858, and (12) 0.2001. In this scenario, the capacities of the edges were as follows: e0102– 0.707, e0213– 0.707, e0304– 1, e0405– 2, e₀₄₀₆– 2, e₀₅₀₆– 0.5, e₀₅₀₇– 0.5. e₀₆₀₇– 0.354, e₀₆₁₃– 0.25, e₀₇₀₈– 0.707, e₀₇₀₉– 0.5, e₀₈₁₃– 0.707, e0913– 0.25, e1013– 1, e1110– 0.707, e1113– 0.707, e1213– 0.707. The solution of the MCFP linear programming algorithm was found to be z = 0.81. In FIG.9, the values written on each of the edges, and the respective shades, represent the percentage of the available capacity used by the edge. The shade bar indicates the capacity used: 0®1: 0% ® 100% used.

[0092] Looking at FIG.9, it can be seen that the solution, although optimal for a single iteration (due to the linear programming properties), may not be optimal in the sense that there are some routes in the network that can increase the transmitted throughput. That is, the determined value z solved one particular flow problem. However, in reality, we can allow for nodes that are connected to the sink via un-congested edges to further increase their throughput, even-though it breaches the inherited MCFP fairness criteria. This can be accomplished by running the same linear programming process in iterations, while locking the congested commodities. For example, in the presented example shown in FIG. 9, the first run produces z = 0.81. However, that the branches of nodes 3, 4, 5, 6, 7, 8, 9 is the limiting factor, while nodes 1 and 2, and 10,

11, 12, which lie on two different branches, can increase their transmitted throughput. Thus, in this example scenario, the routes to the sink

, and

create branches (i.e., dedicated routes) that are not connected directly to the main branch that includes all the other nodes (e.g., the nodes 3,4,5,....). Therefore, the value of z = 0.8148 is set via the capping of the throughput that can be transmitted from the main branch (in this scenario, edges 8-13 and 9-13 would likely be saturated). An iterative approach can be used to continue and look at other areas of the graph (such as , to increase the z value for those parts.

[0093] The above example provides a solution to the switching problem based on the MCFP, in a given snapshot, using known commodity demands and the capacities of specific edges (links). However, the goal is to create a predictive switching scheme, which will prepare the network based on the predicted attenuation of the various CMLs. Thus, the following example Predictive Network Switching process (provided below), which uses the outputs of the status of operation color-based classification, relating it into the MCFP, is proposed:

// Input K is the future time-steps to be considered in the prediction

// Input Ec is an E x K matrix which holds the capacity for each of the edges {E} for each of the time-steps n, n + l, n + K

// Input k is an array which the commodity-demand generated from each of the vertices to be transmitted to the sink

// Inputs Lⁿ, Lⁿ⁺¹,..., L^n+k are the state of the i^th edge (i.e., CML), which is the output of the status of operation classification algorithm / process, for time-steps n, n + 1, ..., n + K

II Output z is the concurrent flow throughput factor // Output Cc is an V x E matrix that holds the commodity portion of each of the nodes { V-} V through each of the edges [E]

11 a is an array of length E

[0094] Basically, the status of operation classification process solves the MCFP for the given network for the current time step, to be implemented in the network management scheme. However, if any of the CMLs are marked as Red or Yellow in the future K time-steps, then, the process replaces the current capacity of the CMLs with the minimum capacity that is detected within the future K time-steps. Thus, the network current switching scheme is made to be predictive in the sense that it considers the future loss of capacities. This approach is

conservative in the sense that it considers the worst-case scenario of the CMLs status, and thus provided a congestion-free solution (not considering any capacity -prediction errors). On the other hand, this solution gives a sub-optimal throughput. Nevertheless, given that the capacities values fluctuate slower than the prediction resolution, the proposed process’ performance is very close to the optimal case. Other predictive switching schemes / processes may be used instead of, or in addition to, the above Predictive Network Switching process.

[0095] In some embodiments, the predictive data derived based on the measurement data may be used to manage channel switching and prioritization. For example, during a case of extremely narrow MHN-based bandwidth (e.g., during an extreme storm event), the provider must ensure a minimum and redundant bandwidth to specific users, while sacrificing others. Thus, it is important to address premium services and end-user bandwidth prioritization. For example, in case of imminent network throughput drop, data-flow from specific end-users, such as cloud- assisted cars, should receive higher rerouting priorities. Such resource allocation may be based, for example, on the available dynamic rerouting scheme described herein, in combination with optimal ATPC and AMC schemes.

[0096] In some embodiments, channel switching capabilities may be implemented for CMLs.

For example, processes in which two CMLs channels, one in the K-band and one in the E-band may be deployed on the same physical link. Such embodiments have the advantage of the higher throughput of the E-band, but, in cases where the higher E-band sensitivity to the environment threatens the link operation, the channel will switch to the K-band frequency. Such a switch will create a drop in the available throughput, but in turn will keep the CML alive. Thus, processes may be implemented to perform predictive channel allocation, in combination with ATPC and AMC schemes, and dynamic rerouting. Once the channel throughput drops, prioritization of services should be considered.

[0097] With reference next to FIG. 10, a flowchart of an example procedure 1000 to manage wireless resources of a weather-impacted wireless communication network (such as the networks 100, 200, or 300 shown in FIGS. 1, 2, or 3, respectively) is shown. The procedure 100 (which may be executed on a resource manager controller implemented, for example, on the server 120 of FIG. 1) includes obtaining 1010 measurement data from at least part of a wireless communication network during a first interval of time, with the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network. The measurements indicative of potential disturbances to operation of the wireless communication network may be indicative of weather-related disturbances, or of any other type of disturbance (be it human created, or a natural occurring disturbance). In some examples, obtaining measurement data may include measuring signal attenuation levels, over the first interval of time, at one or more of wireless links of the wireless communication network. Other types of measurement data may also be obtained, including measured environmental data from the local nodes of the network, or measurement data that was determined remotely (e.g., data collected by a centralized global server to monitor local and remote weather conditions). Measuring the signal attenuation levels may include measuring the signal attenuation level at regulars or irregular time periods within the first interval of time, with the signal attenuation level for a wireless node, m, at a time t within the first interval being computed as a difference between a transmitted signal power level (TSL) for a signal transmitted from a remote wireless node, tij, and a received signal power level (RSL) for the transmitted signal at the wireless node .

[0098] With continued reference to FIG. 10, the procedure 1000 further includes determining 1020, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time. In some examples, the predictive data may be determined based on, in addition to the measurement data (e.g., the signal attenuation levels, at the one or more wireless links), characteristics associated with two or more wireless nodes of the wireless communication network, with the characteristics including, for each of the two or more wireless nodes, available meta data such as, for example, geographic location, a frequency at which signals are transmitted, polarization attributes of the signal, coding technique applied to the signal, modulation technique applied to the signal, and/or adaptability of the two or more wireless nodes to adjust transmission power and/or adjust signal modulation. In some embodiments, determining the predictive data may include determining the predictive data using a learning engine, trained using attenuation time-series data collected from multiple wireless links defined between one or more pairs of wireless nodes of the wireless communication network during dry and wet periods, applied to the measurement data obtained for the wireless communication network. In such embodiments, determining the predictive data may include determining the predictive data representative of future attenuation behavior for wireless links defined between wireless nodes of the wireless communication network for H subsequent time instances following the end of the first interval. Determining the predictive data using the learning engine may include deriving, using a Long Short-Term Memory (LSTM) encoder network, a fixed-length hidden state vector for a time instance in the first interval based on the measurement data obtained for the wireless communication network, and on a previous state vector computed for a previous set of measurement data, the LSTM encoder implementing a non-linear activation function to compute the hidden state vector, and applying an LSTM decoder to the hidden state vector to determine the predictive data representative of future attenuation behavior for the wireless links of the wireless communication network for the H subsequent time instances. As noted, other types and/or learning engine configurations may be used for generating the predictive data. Furthermore, the determination of predictive data may also be performed through various optimization and/or filtering procedures (e.g., Kalman filter implementations) .

[0099] In some embodiments, determining the predictive data may include determining ambient conditions affecting the future performance of the communication network. Determining ambient condition may include determining predicted weather patterns affecting the performance of the wireless communication network, including estimating spatial-temporal rain attenuation patterns predicted to impact the communication network.

[00100] The procedure 1000 additionally includes managing 1030 resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval. For example, managing resources of the communication network may include determining status of operation data (e.g., the color coded classification scheme described in relation to FIG. 6) for at least some wireless links defined for one or more pairs of the wireless nodes of the communication network, with the status of operation data being representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval. In such embodiments, the procedure may further include controlling operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available, with controlling the operational characteristics including one or more of, for example, increasing transmission strength of the first wireless node, and/or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node.

[00101] In some embodiments, the resource management operations may include a link rerouting scheme (network level resource management) that includes identifying a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance, and identifying an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance, with the alternative link selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

[00102] In some alternative examples, resource management may be performed according to a dynamic network topology switching scheme. In such examples, managing resources of the wireless communication network may include deriving throughput values between pairs of wireless nodes of the wireless communication network, to optimize the overall throughput at the future time instance of data traffic to and from one or more sink nodes of the wireless communication network, subject to predicted capacities of the wireless links of the

communication network computed according to the predictive data of the future performance of the wireless communication network at the one or more future time instances following the end of the first interval, and configuring one or more of the wireless nodes based on the derived throughput values. Deriving the throughput values between the pairs of wireless nodes of the wireless communication network may include deriving the throughput values between the pairs of wireless nodes of the wireless communication network subject to further constraints relating to pre-determined characteristics of the wireless communication network. The pre-determined characteristics of the communication network may include one or more of, for example, known capacities of the wireless links, and/or known data traffic demand. In some embodiments, managing the resources of the communication network based, at least in part, on the determined predictive data may include one or more of, for example, a) configuring one or more wireless nodes of the wireless communication network based on the predictive data, subject to the requirement that at least one identified wireless node maintains a specified quality-of- service (QoS) level, and/or b) switching between different channels (e.g., K-band channel or E-band channel) configured for one or more wireless links based on the determined predictive data and on priority of services (e.g., autonomous car services are examples of high priority services) provided at one or more wireless nodes for which the one or more wireless links are established.

[00103] To test and evaluate the performance of some of the implementations described herein, several studies, simulations, and experiments were conducted. More particularly, a practical demonstration of the systems and methods described herein was performed on a network of CMLs in Sweden. For the experiment, a set of available CMLs links were picked in an area of 10x10 km², corresponding to estimated spatial correlations of rain fields in the region. The data set included 17 operating links, with length variations between 600 meters to 6 km, as illustrated in FIG. 2. The links operated in the K-Band frequency range, with frequencies of 20 GHz to 40 GHz.

[00104] The dataset included a total of 135K samples for each CML, with total train- validation-test split of 70-15-15. The validation data was used to provide an evaluation of the model fit to tune model parameters. The input window size was determined to be T = 9, by conducting a grid search over the values 7e { 1, 3, 6, 9, 12, 15}. The learning engine was implemented according to an encoder-decoder architecture, such as the one shown in FIG. 4, where both the LSTM layers had one hidden layer, and 128 units each. A stochastic gradient descent procedure with an Adam optimization algorithm was used, with the model having a batch size of 150. All the parameters were selected to minimize the loss over the validation data.

[00105] The systems and methods described herein were tested over three different rain events, as detailed below, to evaluate performance of the approaches for deriving prediction of future values in CML networks.

Table 1

[00106] The average event-induced attenuation values were calculated by first reducing the baseline attenuation (i.e., the reference level attenuation) each link. Then, the averaged attenuation values were calculated over all the CMLs signals. The baseline attenuation levels were calculated by taking the median during the dry period on the same day, per CML.

[00107] FIG. 11 includes scatter plots 1100, 1110, and 1120, providing cross validation for the actual and predicted values for the minimal and maximal prediction times (i.e., DT_h =

{ 10; 60} seconds), for three selected edges (all measurements are given in dB). The predicted response was calculated with the LSTM encoder-decoder model after estimating the unknown model parameters from the training data. The straight line through the scatter of points is the line representing a zero-prediction error.

[00108] Two performance tests were used to evaluate the encoder-decoder model’s results over the test data, to emphasize different properties. The performance was evaluated in terms of accuracy, with an RMSE error metric, and the probability of significant errors occurring. First, the Root-Mean-Square Errors (RMSE) is calculated to measure the differences between values predicted by a model or an estimator and the values observed. For the h time step, the RMSE_h is the error metric for prediction time D7},= 10 h seconds.

[00109] are the observed and predicted attenuation values of the n^th link,

respectively, where the predicted value was generated h steps ahead. Ei represents the set of samples of the i^th event.

[00110] FIG. 12 is a graph showing the averaged RMSE_h values (in dB) for each of the tested events (I, II, and III), over the tested edges, as a function of the prediction time DT_h. The values for each edge is first computed. Then, the RMSE values, for a time step h, for

the various edges included in the graph are averaged. The result provides the average error (in dB) obtained by the algorithm when predicting the future attenuation value for all the CML network according to:

In the above equation, N is the number of one-directional links in the network.

[00111] It is desirable to determine the probability of getting an extreme error in the derived predictions. To that end, the absolute error for the predictions may be defined as the absolute value between the measured and predicted value, as follows:

where and A_l+h are the measured and predicted attenuation. Based on the empirical error statistics, the confidence measure in the forecast can be derived as follows:

[00112] The parameter p_e is defined as the empirical probability that the absolute error in the prediction of h steps ahead, _<¾, is smaller than some critical value x_crit. A confidence interval is set to test the model, by the proportion of the absolute errors in which the values are less than X_crit · The critical value is a function of the desired confidence level, the parameter g.

[00113] FIG. 13 includes graphs 1300, 1310 and 1320, showing the distributions of the residuals during rain event I (in Table 1) for values of h = 1, 3, 6, respectively. The vertical line represents the x_crit value obtained for a g value of 0.95. The residuals histogram implies that the residuals in each time step are approximately normally distributed, and the prediction’s bias is centered around zero. The variance value grew as the prediction time increases, as the RMSE test implied. Moreover, the symmetric distribution of the residuals justifies the use of the absolute error.

[00114] FIG. 14 includes graphs 1400, 1410, and 1420, showing critical values as a function of g, for the three tested events and for different prediction times. As shown, the critical values become larger as the prediction time increase, for lower values of g. For a fixed g value, it is expected that the received prediction is different than the actual observation by less than x_c (dB) with the confidence of 1-g.

[00115] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly or conventionally understood. As used herein, the articles“a” and “an” refer to one or to more than one (i.e., to at least one) of the grammatical object of the article. By way of example,“an element” means one element or more than one element. “About” and/or “approximately” as used herein when referring to a measurable value such as an amount, a temporal duration, and the like, encompasses variations of ±20% or ±10%, ±5%, or ±0.1% from the specified value, as such variations are appropriate in the context of the systems, devices, circuits, methods, and other implementations described herein.“Substantially” as used herein when referring to a measurable value such as an amount, a temporal duration, a physical attribute (such as frequency), and the like, also encompasses variations of ±20% or ±10%, ±5%, or ±0.1% from the specified value, as such variations are appropriate in the context of the systems, devices, circuits, methods, and other implementations described herein.

[00116] As used herein, including in the claims,“or” as used in a list of items prefaced by “at least one of’ or“one or more of’ indicates a disjunctive list such that, for example, a list of “at least one of A, B, or C” means A or B or C or AB or AC or BC or ABC (i.e., A and B and C), or combinations with more than one feature (e.g., AA, AAB, ABBC, etc.). Also, as used herein, unless otherwise stated, a statement that a function or operation is“based on” an item or condition means that the function or operation is based on the stated item or condition and may be based on one or more items and/or conditions in addition to the stated item or condition.

[00117] Although particular embodiments have been disclosed herein in detail, this has been done by way of example for purposes of illustration only, and is not intended to be limiting with respect to the scope of the appended claims, which follow. Features of the disclosed embodiments can be combined, rearranged, etc., within the scope of the invention to produce more embodiments. Some other aspects, advantages, and modifications are considered to be within the scope of the claims provided below. The claims presented are representative of at least some of the embodiments and features disclosed herein. Other unclaimed embodiments and features are also contemplated.

Claims

WHAT IS CLAIMED IS:

1. A method comprising:

obtaining measurement data from at least part of a wireless communication network during a first interval of time, wherein the measurement data includes measurements indicating whether there are potential disturbances to operation of the wireless communication network; determining, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time; and

managing resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless

communication network at the one or more future time instances subsequent to the end of the first interval.

2. The method of claim 1, wherein obtaining the measurement data comprises:

measuring signal attenuation levels, over the first interval of time, at one or more of wireless links of the wireless communication network.

3. The method of claim 2, wherein measuring the signal attenuation levels comprises: measuring the signal attenuation level at regulars or irregular time periods within the first interval of time, wherein the signal attenuation level for a wireless node, m, at a time t within the first interval is computed as a difference between a transmitted signal power level (TSL) for a signal transmitted from a remote wireless node, n_j, and a received signal power level (RSL) for the signal received at the wireless node m.

4. The method of claim 1, wherein determining the predictive data comprises:

determining the predictive data based on measured signal attenuation levels at the one or more wireless links, and further based on characteristics associated with two or more wireless nodes of the wireless communication network, the characteristics including, for each of the two or more wireless nodes, available meta data, including one or more of: geographic location, a frequency at which signals are transmitted, polarization attributes of the signal, coding technique applied to the signal, modulation technique applied to the signal, or adaptability of the two or more wireless nodes to adjust transmission power and/or adjust signal modulation.

5. The method of claim 1, wherein determining the predictive data comprises:

determining the predictive data using a learning engine, trained using attenuation time- series data collected from multiple wireless links defined between one or more pairs of wireless nodes of the wireless communication network during dry and wet periods, applied to the measurement data obtained for the wireless communication network.

6. The method of claim 5, wherein determining the predictive data comprises:

determining the predictive data representative of future attenuation behavior for wireless links defined between wireless nodes of the wireless communication network for H subsequent time instances following the end of the first interval.

7. The method of claim 6, wherein determining the predictive data using the learning engine comprises:

deriving, using a Long Short-Term Memory (LSTM) encoder network, a fixed-length hidden state vector for a time instance in the first interval based on the measurement data obtained for the wireless communication network, and on a previous state vector computed for a previous set of measurement data, the LSTM encoder implementing a non-linear activation function to compute the hidden state vector; and

applying an LSTM decoder to the hidden state vector to determine the predictive data representative of future attenuation behavior for the wireless links of the wireless communication network for the H subsequent time instances.

8. The method of claim 1, wherein managing resources of the communication network comprises:

determining status of operation data for at least some wireless links defined for one or more pairs of the wireless nodes of the communication network, the status of operation data representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval.

9. The method of claim 8, further comprising:

controlling operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available, wherein controlling the operational characteristics comprises one or more of: increasing transmission strength of the first wireless node, or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node.

10. The method of claim 8, further comprising:

identifying a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance; and

identifying an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance, the alternative link selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

11. The method of claim 1, wherein managing resources of the wireless communication network comprises:

deriving throughput values between pairs of wireless nodes of the wireless

communication network, to optimize the overall throughput at the future time instance of data traffic to and from one or more sink nodes of the wireless communication network, subject to predicted capacities of the wireless links of the communication network computed according to the predictive data of the future performance of the wireless communication network at the one or more future time instances following the end of the first interval; and

configuring one or more of the wireless nodes based on the derived throughput values.

12. The method of claim 11, wherein deriving the throughput values between the pairs of wireless nodes of the wireless communication network comprises:

deriving the throughput values between the pairs of wireless nodes of the wireless communication network subject to further constraints relating to pre-determined characteristics of the wireless communication network.

13. The method of claim 11, wherein the pre-determined characteristics of the communication network comprises one or more of: known capacities of the wireless links, or known data traffic demand.

14. The method of claim 1, wherein managing the resources of the communication network based, at least in part, on the determined predictive data comprises one or more of: a) configuring one or more wireless nodes of the wireless communication network based on the predictive data, subject to the requirement that at least one identified wireless node maintains a specified quality-of- service (QoS) level; or

b) switching between different channels configured for one or more wireless links based on the determined predictive data and on priority of services provided at one or more wireless nodes for which the one or more wireless links are established.

15. The method of claim 1, wherein determining the predictive data comprises:

determining ambient conditions affecting the future performance of the communication network.

16. The method of claim 15, wherein determining ambient condition comprises:

determining predicted weather patterns affecting the performance of the wireless communication network, including estimating spatial-temporal rain attenuation patterns predicted to impact the communication network.

17. The method of claim 1, wherein the measurements indicative of potential

disturbances to operation of the wireless communication network are indicative of weather- related disturbances.

18. A system comprising:

a communication module to obtain measurement data from wireless nodes, forming at least part of a wireless communication network, during a first interval of time, the measurement data including measurements indicating whether there are potential disturbances to operation of the wireless communication network; and

a processor-based resource management controller, coupled to the communication module, configured to:

determine, based on the obtained measurement data, predictive data representative of future performance of the wireless communication network at one or more future time instances subsequent to an end of the first interval of time; and

manage resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless communication network at the one or more future time instances subsequent to the end of the first interval.

19. The system of claim 18, further comprising:

the wireless nodes forming the at least part of the wireless communication network.

20. The system of claim 18, wherein the controller configured to determine the predictive data is configured to:

determine the predictive data based on measured signal attenuation levels at the one or more wireless links, and further based on characteristics associated with two or more of the wireless nodes of the wireless communication network, the characteristics including, for each of the two or more wireless nodes, available meta data, including one or more of: geographic location, a frequency at which signals are transmitted, polarization attributes of the signal, coding technique applied to the signal, modulation technique applied to the signal, or adaptability of the two or more wireless nodes to adjust transmission power and/or adjust signal modulation.

21. The system of claim 18, wherein the controller configured to determine the predictive data is configured to: determine the predictive data using a learning engine, trained using attenuation time- series data collected from multiple wireless links defined between one or more pairs of wireless nodes of the wireless communication network during dry and wet periods, applied to the measurement data, wherein the predictive data is representative of future attenuation behavior for wireless links defined between wireless nodes of the wireless communication network for H subsequent time instances following the end of the first interval.

22. The system of claim 18, wherein the controller configured to manage resources of the communication network is configured to:

determine status of operation data for at least some wireless links defined for one or more pairs of the wireless nodes of the wireless communication network, the status of operation data representative of likely degree of attenuation for the at least some wireless links at the one or more future time instances subsequent to the end of the first interval; and

perform one or more of:

i) control operational characteristics of a first wireless node, transmitting signals through a first wireless link, based on the status of operation data predicting at least some attenuation for the first wireless link but indicating the first wireless link remains available, wherein the controller configured to control the operational characteristics is configured to cause one or more of: increasing transmission strength of the first wireless node, or reducing throughput for the first wireless node according through adjustments of signal coding and/or modulation techniques used by the first wireless node; or

ii) identify a first wireless link, from the at least some wireless links, predicted, based on the determined status of operation data, to be unavailable or to be operating at a lower capacity at a subsequent time instance, and identify an alternative link for data transmission to re-route data from the first wireless link identified to be unavailable or to be operating at the lower capacity at the subsequent time instance, the alternative link selected from one or more other wireless links predicted, based on the determined status of operation data, to be available at a later time instance to the subsequent time instance at which the identified first wireless link is predicted to be unavailable or operating at the lower capacity.

23. The system of claim 18, wherein the controller configured to manage resources of the wireless communication network is configured to:

derive throughput values between pairs of wireless nodes of the wireless communication network, to optimize the overall throughput at a future time instance of data traffic to and from one or more sink nodes of the wireless communication network, subject to predicted capacities of the wireless links of the communication network computed according to the predictive data of the future performance of the wireless communication network at the one or more future time instances following the end of the first interval; and

configure one or more of the wireless nodes based on the derived throughput values.

24. Non-transitory computer readable media comprising computer instructions executable on a processor-based device to:

obtain measurement data from at least part of a wireless communication network during a first interval of time, wherein the measurement data includes measurements indicating whether there are potential disturbances to operation of the wireless communication network;

manage resources of the wireless communication network based, at least in part, on the determined predictive data representative of the future performance of the wireless