Patent Grant
11695202
This application claims priority to PCT Application No. PCT/EP2018/069078, filed on Jul. 13, 2018, which is incorporated herein by reference in its entirety.
Example embodiments relate to methods, systems and computer programs for controlling one or more radiating elements, for example the radiating elements of one or more antennas which may form part of an antenna array.
An antenna may comprise one or more radiating elements which are fed with the same radio frequency signal. An antenna array may comprise multiple such antennas with individual radio frequency chains. Antenna arrays are used for various purposes, for example for beamforming in telecommunications networks. Radiating elements may also be used for receiving radio frequency energy.
A first aspect may provide an apparatus, comprising: means for receiving a performance metric for an antenna array comprised of a plurality of radiating elements, the performance metric being based on performance data associated with the antenna array, the antenna array having a radiating configuration represented by configuration parameters; means for updating the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; means for re-configuring the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
The re-configuring means may be configured to change the physical geometry by moving one or more radiating elements based on the re-configured configuration.
The updating means may be further configured to iteratively update the configuration parameters dependent on received performance metrics resulting from the updated configuration parameters, until a stop condition is reached.
The stop condition may correspond to a fixed number of update iterations.
The stop condition may correspond to the performance metric reaching the target value. The stop condition may correspond to a fixed number of update iterations for which the performance metric does not move towards the target value.
The updating means may comprise a machine learning model configured to estimate, based on stored training data, updated configuration parameters likely to move the performance metric towards the target value. The machine learning model may be trained with data derived from initial connections quality and subsequently updating the model based on the results caused by the updated configuration parameters. The machine learning model may be trained with initial configuration parameters and subsequently updated based on the results of re-configurations caused by the updated configuration parameters. The machine learning model may be trained from a digital sibling.
The performance metric may be received from a computing means configured to receive performance data from a physical node of at least part of a communications network with which the antenna array forms part.
The computing means may be associated with a base station of at least part of a cellular communications network with which the antenna array forms part.
The performance metric may be received from a simulation means which simulates a physical node of at least part of a communications network with which the antenna array forms part.
The simulated physical node may be associated a base station of at least part of a cellular communications network with which the antenna array forms part.
The computing or simulation means may comprise a part of the apparatus.
The received performance metric may represent the measured or simulated value of one or more of data throughput, call drop rate, outage probability and energy consumption associated with the antenna array.
The performance metric may represent said measured or simulated values for each of a different number of user equipment within a cell of a cellular communications system associated with the antenna array.
The configuration parameters may comprise one or more of, for a jth radiating element of an ith antenna of the antenna array:
i, j⊂
3
and the means for re-configuring the radiating configuration of the antenna array may be configured to update one or more of said configuration parameters to effect a corresponding change at one or more of the radiating elements of the antenna array.
The means for re-configuring the radiating configuration of the antenna array may be configured to perform the re-configuring in substantially real-time or near real-time based on the last-received value of the performance metric.
Another aspect may provide a method, comprising: receiving a performance metric for an antenna array comprised of a plurality of radiating elements, the antenna array having a radiating configuration represented by configuration parameters; updating the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; and re-configuring the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
The re-configuring may change the physical geometry by moving one or more radiating elements based on the re-configured configuration.
The method may further comprise iteratively updating the configuration parameters dependent on received performance metrics resulting from the updated configuration parameters, until a stop condition is reached.
The stop condition may correspond to a fixed number of update iterations. The stop condition may correspond to the performance metric reaching the target value. The stop condition may correspond to a fixed number of update iterations for which the performance metric does not move towards the target value.
The updating may comprise using a machine learning model configured to estimate, based on stored training data, updated configuration parameters likely to move the performance metric towards a target value.
The method may further comprise training the machine learning model with data derived from initial connections quality and subsequently updated based on the results caused by the updated configuration parameters. The method may further comprise training the machine learning model with initial configuration parameters and subsequently updating based on the results of re-configurations caused by the updated configuration parameters.
The machine learning model may be trained with data from a digital sibling.
The performance metric may be received from a computing means configured to receive performance data from a physical node of at least part of a communications network with which the antenna array forms part.
The computing means may be associated with a base station of at least part of a cellular communications network with which the antenna array forms part.
The performance metric may be received from a simulation means which simulates a physical node of at least part of a communications network with which the antenna array forms part.
The simulated physical node may be associated a base station of at least part of a cellular communications network with which the antenna array forms part.
The computing or simulation means may comprise a part of the apparatus.
The received performance metric may represent the measured or simulated value of one or more of data throughput, call drop rate, outage probability and energy consumption associated with the antenna array.
The performance metric may represent said measured or simulated values for each of a different number of user equipment within a cell of a cellular communications system associated with the antenna array.
The configuration parameters may comprise one or more of, for a jth radiating element of an ith antenna of the antenna array:
i, j⊂
3
and the re-configuring of the radiating configuration of the antenna array may comprise updating one or more of said configuration parameters to effect a corresponding change at one or more of the radiating elements of the antenna array.
The re-configuring of the radiating configuration of the antenna array may comprise performing the re-configuring in substantially real-time or near real-time based on the last-received value of the performance metric.
Another aspect may provide an apparatus comprising at least one, at least one memory directly connected to the at least one processor, the at least one memory including computer program code, and the at least one processor, with the at least one memory and the computer program code being arranged to perform the method of: receiving a performance metric for an antenna array comprised of a plurality of radiating elements, the antenna array having a radiating configuration represented by configuration parameters; updating the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; and re-configuring the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
The at least one processor, with the at least one memory and the computer code are further arranged to perform the method of any above definition connected with the above aspect.
Another aspect may provide a computer program product comprising a set of instructions which, when executed on an apparatus, are configured to cause the apparatus to carry out the method of: receiving a performance metric for an antenna array comprised of a plurality of radiating elements, the antenna array having a radiating configuration represented by configuration parameters; updating the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; and re-configuring the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
Another aspect provides a non-transitory computer readable medium comprising program instructions stored thereon for performing a method, comprising: receiving a performance metric for an antenna array comprised of a plurality of radiating elements, the antenna array having a radiating configuration represented by configuration parameters; updating the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; and re-configuring the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
Another aspect may provide an apparatus comprising: at least one processor; and at least one memory including computer program code which, when executed by the at least one processor, causes the apparatus: to receive a performance metric for an antenna array comprised of a plurality of radiating elements, the antenna array having a radiating configuration represented by configuration parameters; to update the configuration parameters dependent on the received performance metric by means of estimating new configuration parameters for moving the performance metric towards a target value; and to re-configure the radiating configuration of the antenna array based on the updated configuration parameters such that the physical geometry of the antenna array is changed.
Reference to a “means” above may refer to any of hardware, software, electrical or electronic circuitry, or any combination thereof, configured to perform the stated functions.
Example embodiments will now be described by way of non-limiting example, with reference to the accompanying drawings, in which:
Example embodiments relate to controlling one or more radiating elements, for example of one or more radio frequency antennas. Example embodiments may relate to an apparatus, method and/or computer program for controlling one or more antenna radiating elements (hereafter “radiating elements”) of one or more antennas of an antenna array; an antenna array is a physical apparatus which comprises a plurality of distinct antennas. The antennas of the array may work together, effectively as a single antenna, to transmit or receive radio frequency waves. Antenna arrays are used in various fields, for example in cellular communications.
Example embodiments may relate to adaptively controlling the radiating elements of one or more antennas. The adaptive control may be performed in real-time or near real-time to provide on-the-fly modifications to a current antenna array configuration based on, for example, current or very recent conditions, such as network conditions, the number of user equipment in one or more cells associated with the antenna array, and so on. These examples are not to be considered limiting on the scope of the present disclosure and are given merely by way of example. The antenna array configuration may refer to characteristics of individual antennas, and/or characteristics of individual radiating elements thereof, in terms of their phase, gain, and positional characteristics, for example. Modifications to the antenna array configuration may be for the purpose of modifying radiation or receiving characteristics of one or more radiating elements. All or a subset of the radiating elements may be re-configured in this way.
Referring to
Some characteristics of an antenna array are antenna spacing and total size (relative to the wavelength) which is known as the aperture. The size may determine the directivity of the array, that is its ability to focus radiated energy towards certain directions. The number of antennas may determine the radiated/received energy.
Referring to
In some embodiments, other characteristics of a given radiating element 20 may comprise one or more of position, azimuth angle and elevation angle. For this purpose, mechanical positioning means may be associated with each radiating w element 20. An example mechanical positioning means may comprise a positional actuator 24 (“PA”). A plurality of such actuators 24 may be used, for example linear and/or rotational positional actuators, for modifying one or more of the position, azimuth angle and elevation angles.
Embodiments herein may provide a controller 26 for controlling the configuration of one or more the radiating elements 20 by means of controlling one or more respective phase shifter and amplifier elements 22 and/or positional actuators 24. In
In overview, example embodiments may involve the controller 26 (or another processing device associated with the controller) receiving a performance metric L(Ω) for an antenna array to comprised of a plurality of radiating elements 20. The performance metric L(Ω) may be based on performance data associated with the antenna array to, the antenna array having a radiating configuration represented by configuration parameters. In some embodiments, the configuration parameters may correspond to one or more of the above characteristics of given radiating elements 20. For example, the jth radiating element of the ith antenna of the antenna array to may the following configuration parameters:
i, j⊂
3
The values of the configuration parameters may reflect the settings of individual radiating elements 20 in correspondence to the configuration.
The configuration parameters may be combined into a parameter vector:
ωi,j=[xT,ϕi,j,θi,j,βi,j,αi,j]T∈R7.
The antenna array 10 may have M antennas, where the ith antenna has Ni radiating elements. The set of all configuration parameters of the antenna array 10 may be denoted by
Ω={ωi,j:i=1, . . . ,M:j=1, . . . ,Ni}.
Embodiments may also comprise updating the configuration parameters Ω dependent on the received performance metric L(Ω), and re-configuring the radiating configuration of the antenna array 10 based on the updated configuration parameters {tilde over (Ω)}1. The values of the updated configuration parameters {tilde over (Ω)}1 therefore reflect updated settings to be applied to the individual radiating elements 20. This may be achieved for example by one or more of increasing or decreasing the phase shift ai,j, the signal gain ai,j, and modifying via the positional actuators 24 the position xi,j, azimuth angle ϕi,j and elevation angle θi,j of the individual radiating elements 20.
The performance metric L(Ω) may be an arbitrary metric. For example, the performance metric L(Ω) may be a value representative of data throughput, e.g. cell sum throughput, 5th percentile throughput, median throughput or geometric throughput. For example, the performance metric L(Ω) may be a value representative of one or more of call drop rate, outage probability, energy consumption and so on. One or more of the above examples may be the basis of computing a performance metric L(Ω), which is a value indicative of performance of the one or more performance characteristics.
By modifying one or more configuration parameters, the performance metric L(Ω) is likely to change. Therefore, embodiments may involve modifying the one or more configuration parameters Ω with the aim of moving the performance metric L(Ω) towards a target. The target may represent an optimal condition, e.g. solving the optimisation problem:
arg max L(Ω) (1)
but some other target, usually in the direction of improving technical performance, may be used.
In some embodiments, it is proposed to use reinforcement learning (RL) and/or supervised learning (SL) methods to adapt the positions, and/or other controllable characteristics, of the radiating elements 20 with the aim of improving, if not maximising, the performance metric L(Ω). In this way, embodiments enable changing of the geometry an antenna array 10, on-the-fly, such that it can be automatically improved or optimized for any radio environment, without human interaction. The main benefits are improved performance, as well as reduced capital and ongoing expenditure.
The performance metric L(Ω) may itself either be computed from measurements based on real data associated with the antenna array 10, or from simulations. The performance metric L(Ω) may also be available in explicit mathematical form.
A controllable antenna array according to example embodiments may combine two or more of the
There will now be described examples of how control is achieved.
In overview, control may be performed using hardware, software or a combination thereof.
The learning algorithm may be configured to produce an antenna array configuration {tilde over (Ω)}t which is subsequently applied to the antenna array 62 through the antenna array controller 70.
The base station may be configured to measure the performance metric L({tilde over (Ω)}t), for example the throughput achieved by the user equipments 66 within its cell, which is subsequently forwarded to the learning algorithm 68, which in turn is configured to produce an updated antenna configuration and so on which re-configures the characteristics of the antenna array 62.
A more specific, but still general learning algorithm which works for any performance metric L(Ω) will now be described, relating to the high-level view provided in
In a first operation 901, we let t=0 and choose a set of initial parameters Ωt.
In another operation 902 we compute a perturbed configuration {tilde over (Ω)}t. For example, for i=1, . . . , M, j=1, . . . , Ni, a random vector εi,j may be drawn from some distribution p(ε). For example, p(ε) could be a multivariate normal distribution with zero mean and a fixed covariance matrix σ2I, i.e., p(ε)=N(0, σ2I). We could also have a different distribution pi,j (ε) for each radiating element of the antenna array 62.
In another operation 903, the antenna array 62 may be configured according to:
{tilde over (ω)}ti,j=ωti,j+εi,j,i=1, . . . ,M,j=1, . . . ,Ni. (2)
This defines the parameters {tilde over (Ω)}t.
In another operation 904, the performance metric L({tilde over (Ω)}t) is computed, e.g. measured.
In another operation 905, it is determined if a stop condition is reached. If not, another operation 906 may comprise computing parameter updates according to:
ωi,jt+1=ωi,jt+ηL({tilde over (Ω)}t)∇ω
for some learning rate η>0. For p(ε)=N(0, σ−2I), this boils down to:
In the same or a different operation, the value oft is updated to +1, and the process may return to operation 902.
The initial antenna array configuration Ωt may be obtained from simulations. In this case, the performance metric L(Ω) is not measured from a real-world system, e.g. data received from the antenna array 62, but simulated on a computer. Having good initialization is helpful to speed-up the learning process. However, it is only optional and not required for realizing embodiments herein.
In some embodiments, if the updated parameter vectors ωt+1 or {tilde over (ω)}t+1 are determined to comprise non-feasible values, for example if they correspond to a radiating element position that is not physically possible, the parameter vectors may be projected back to a set of feasible parameter vectors. For example, entries of the vector may be clipped to certain minimum and/or maximum values.
Alternatively, the learning rate q may be adjusted such that i, j remains feasible.
In some embodiments, operations 903 and 904 may be run multiple times to compute performance metrics for different perturbations of the same antenna configuration. In this case, the gradient computed in operation 906 may be averaged over these values.
In operation 906, the learning-rate η may be computed and adapted by any stochastic gradient descent (SGD) algorithm, e.g., ADAM, RMSProp, Momentum, to give some examples.
The stop criterion in operation 905 may take multiple forms. For example, the process may stop after a fixed number of training iterations, or stop when L({tilde over (Ω)}t) has not improved during a fixed number of iterations, or stop when L({tilde over (Ω)}t) has reached a desired value. In some embodiments, there may be no stop operation 905, i.e. the training may run indefinitely.
The second term of equation (3) is also known as the policy gradient. Various methods to improve the convergence speed of this algorithm exist in the (deep) learning literature, see, e.g. R. S. Sutton, D. McAllester, S. Singh, and Y. Mansour, Policy gradient methods for reinforcement learning with function approximation, Proceedings of the 12th International Conference on Neural Information Processing Systems, NIPS, pages 1057-1063, Cambridge, Mass., USA, 1999. MIT Press.
A learning algorithm for an differentiable performance metric L(Ω) will now be described.
In some cases, the gradient ∇ωi,j L(Ω) may be explicitly calculated. This may be the case when L(Ω) is either computed analytically/numerically or through simulations.
Referring to the flow diagram of
In a first operation 1001, we let t=0 and choose a set of initial parameters Ωt.
In another operation 1002, we configure the antenna array 62 according to Ωt.
In another operation 1003, we computer the performance metric L(Ωt).
In another operation 1004, it is determined if a stop condition is reached. If not, another operation 1005 may comprise computing parameter updates according to:
ωi,jt+1=ωi,jt+η∇ω
for a learning rate η>0.
In the same or a different operation, the value oft is updated to and the process may return to operation 1002.
Similar to some other embodiments, if the updated parameter vectors ωt+1 or {tilde over (ω)}t+1 are determined to comprise non-feasible values, for example if they correspond to a radiating element position that is not physically possible, the parameter vectors may be projected back to a set of feasible parameter vectors. For example, entries of the vector may be clipped to certain minimum and/or maximum values.
The learning-rate η may be computed and adapted by any stochastic gradient descent (SGD) algorithm, e.g., ADAM, RMSProp, Momentum, to give some examples.
In some embodiments, operations 1002 and 1003 may be run multiple times to compute performance metrics for different perturbations of the same antenna configuration. In this case, the gradient computed in operation 1006 may be averaged over these values.
In some embodiments, the above-mentioned initial configuration of the antenna array 62 can be obtained by simulation. That is, the algorithms described above may be used to find parameters Ω through simulations which are then used as initial parameters Ω° for the algorithm running on a real system. This is schematically shown in
So far, the antenna configuration parameters are optimized with respect to a given performance metric L(Ωt). However, it may be possible that there are other parameters s∈RP which should be explicitly taken into account. These parameters may comprise, for example, the number of user equipments 66 in a cell, the time of the day, weather conditions, etc.
In some examples, a goal is to find a mapping Ω=ƒ(s) from these additional parameters to the optimal antenna array configuration that maximizes the performance metric
L(Ω,s).
In this case, we may assume that the function ƒ has tunable parameters μ, i.e., ƒ(s)=ƒμ(s), and we would like to solve the optimization problem:
arg max L(ƒμ(s),s). (6)
μ
We may obtain the following algorithm for non-differentiable performance metrics L(ƒμ(s), s). A flow diagram showing processing operations in this case is shown in
In a first operation 1101, we let t=0 and choose a set of initial parameters μt. We also compute Ωt=ƒμt(s).
In another operation 1102, we compute a perturbed configuration {tilde over (Ω)}t. This may comprise taking the parameters s and computing Ωt=ƒμt(s). For i=1, . . . , M, j=1, . . . , Ni, we may draw a random vector εi,j from some distribution p(ε). For example, p(ε) could be a multivariate normal distribution with zero mean and fixed covariance matrix σ2I, i.e., p(ε)=N(0, σ2I). We may also have a different distribution pi,j(ε) for each radiating element of the antenna array.
In another operation 1103, we may configure the antenna array 62 according to:
{tilde over (ω)}i,jt=ωi,jt+εi,j,i=1, . . . ,M,j=1, . . . Ni (7)
This defines the parameters {tilde over (Ω)}t.
In another operation 1104, the performance metric L({tilde over (Ω)}t) is computed, e.g. measured.
In another operation 1105, it is determined if a stop condition is reached. If not, another operation 1106 may comprise computing parameter updates according to:
μt+1=μt+ηL({tilde over (Ω)}t,s)∇μ
for a learning rate η>0.
In the same or a different operation, the value oft is updated to t+1, and the process may return to operation 1102.
In some embodiments, operations 1103 and 1104 may be run multiple times to compute performance metrics for different perturbations of the same antenna configuration. In this case, the gradient computed in operation 1106 may be averaged over these values.
For differentiable performance metrics L(ƒμ(s), s) we propose the following method, the processing operations of which are shown in
In a first operation 1201, we let t=0 and choose a set of initial parameters μt. We also compute Ωt=ƒμt(s).
In another operation 1202, we compute a perturbed configuration {tilde over (Ω)}t. This may comprise taking the parameters sand computing Ωt=ƒμt(s).
In another operation 1203, we may configure the antenna array 62 according to Ωt=ƒμt(s).
In another operation 1204, we may compute performance metric L(Ωt,s).
In another operation 1205, it is determined if a stop condition is reached. If not, another operation 1206 may comprise computing parameter updates according to:
μt+1=μt+η∇μ
for a learning rate η>0.
In the same or a different operation, the value oft is updated to t+1, and the process may return to operation 1202.
In some embodiments, operations 1203 and 1204 may be run multiple times to compute performance metrics for different perturbations of the same antenna configuration. In this case, the gradient computed in operation 1206 may be averaged over these values.
In some embodiments, the mapping ƒμ(s) may be performed using a neural network (NN) and μ may denote its trainable parameters, e.g., its weights and/or biases.
In some embodiments, if the updated parameter vectors ωt+1 or {tilde over (ω)}t+1 are determined to comprise non-feasible values, for example if they correspond to a radiating element position that is not physically possible, the parameter vectors may be projected back to a set of feasible parameter vectors. For example, entries of the vector may be clipped to certain minimum and/or maximum values. Alternatively, the learning rate η can be adjusted such that ωt+1 always remains feasible. Another way is to “punish” the learning system may be by feedback of a predefined negative loss when a non-feasible setting is chosen, so that by minimizing the loss, the systems “learns” to avoid such configurations.
The learning algorithm may employ a machine learning model, which may be initialised with data derived from initial connections quality, e.g. signal quality, and subsequently updated based on the results caused by the updated configuration parameters. The machine learning model in other embodiments may be initialised with initial configuration parameters.
The memory may be volatile or non-volatile. It may be e.g. a RAM, SRAM, a flash memory, a FPGA block ram, a DCD, a CD, a USB stick, and a blue ray disk.
If not otherwise stated or otherwise made clear from the context, the statement that two entities are different means that they perform different functions. It does not necessarily mean that they are based on different hardware. That is, each of the entities described in the present description may be based on a different hardware, or some or all of the entities may be based on the same hardware. It does not necessarily mean that they are based on different software. That is, each of the entities described in the present description may be based on different software, or some or all of the entities may be based on the same software. Each of the entities described in the present description may be embodied in the cloud.
Implementations of any of the above described blocks, apparatuses, systems, techniques or methods include, as non-limiting examples, implementations as hardware, software, firmware, special purpose circuits or logic, general purpose hardware or controller or other computing devices, or some combination thereof. Some embodiments may be implemented in the cloud.
It is to be understood that what is described above is what is presently considered the preferred embodiments. However, it should be noted that the description of the preferred embodiments is given by way of example only and that various modifications may be made without departing from the scope as defined by the appended claims.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2018/069078 | 7/13/2018 | WO |
| Publishing Document | Publishing Date | Country | Kind |
|---|---|---|---|
| WO2020/011368 | 1/16/2020 | WO | A |
| Number | Name | Date | Kind |
|---|---|---|---|
| 5969689 | Martek | Oct 1999 | A |
| 8643559 | Deng | Feb 2014 | B2 |
| 10079431 | Deng | Sep 2018 | B2 |
| 20130143592 | Brisebois et al. | Jun 2013 | A1 |
| 20130331079 | Racz et al. | Dec 2013 | A1 |
| 20160162783 | Tan et al. | Jun 2016 | A1 |
| 20160165462 | Tan | Jun 2016 | A1 |
| 20160165472 | Gopalakrishnan et al. | Jun 2016 | A1 |
| 20180175498 | Kurniawan | Jun 2018 | A1 |
| Entry |
|---|
| Sutton et al., “Policy Gradient Methods for Reinforcement Learning with Function Approximation”, Proceedings of the 12th International Conference on Neural Information Processing Systems, Nov. 1999, pp. 1057-1063. |
| Dandanov et al., “Dynamic Self-optimization of the Antenna Tilt for Best Trade-off Between Coverage and Capacity n Mobile Networks”, Wireless Personal Communications, vol. 92, Oct. 27, 2016, 28 pages. |
| Moghaddam et al., “Joint Tilt Angle Adaptation and Beamforming in Multicell Multiuser Cellular Networks”, arXiv, Jan. 13, 2017, pp. 1-20. |
| Pedras et al., “Antenna Tilt Optimization Using a Novel QoE Model Based on 3G Radio Measurements”, The 20th International Symposium on Wireless Personal Multimedia Communications (WPMC2017), Dec. 17-20, 2017, pp. 124-130. |
| Guo et al., “Spectral- and Energy-efficient Antenna Tilting in a Hetnet Using Reinforcement Learning”, IEEE Wireless Communications and Networking Conference (WCNC), Apr. 7-10, 2013, pp. 1-6. |
| Christodoulou et al., “The Use of Machine Learning in Smart Antennas”, IEEE Antennas and Propagation Society Symposium, Jun. 20-25, 2004, pp. 321-324. |
| International Search Report and Written Opinion received for corresponding Patent Cooperation Treaty Application No. PCT/EP2018/069078, dated Mar. 29, 2019, 14 pages. |
| Zhang et al., “Machine Learning for Predictive On-Demand Deployment of UAVs for Wireless Communications”, arXiv, Apr. 30, 2018, 6 pages. |
| Office action received for corresponding European Patent Application No. 18746851.7, dated Jul. 27, 2022, 6 pages. |
| Number | Date | Country | |
|---|---|---|---|
| 20210328341 A1 | Oct 2021 | US |
