Algorithm for mitigation of impact of uplink/downlink beam mismatch
According to an aspect, there is provided an apparatus for the performing the following. The apparatus implements, separately for at least one downlink beam, a reinforcement learning model, where a state defines which of the plurality of uplink beams belong to a priority beam set for uplink reception corresponding to a downlink beam, an action is defined as an addition of a new uplink beam to the priority beam set, a removal of an uplink beam from the priority beam set or doing nothing and a reward is calculated based on a change in uplink signal-to-noise ratio due to an action adjusted with a cost for taking the action. The apparatus calculates iteratively at least one optimal state using at least one reinforcement learning model based on uplink signal-to-noise ratio statistics and on the plurality of optimal downlink beams for transmission to said plurality of terminal devices.
1. An apparatus, comprising:
at least one processor; and
at least one memory storing instructions to be executed by the at least one processor, wherein the at least one memory and the instructions are configured to, with the at least one processor, cause the apparatus at least to perform:
implementing, separately for at least one downlink beam of a plurality of downlink beams of an access node, a reinforcement learning model, wherein a state, an action and a reward of the reinforcement learning model for a downlink beam are defined as follows:
the state defines which of the plurality of uplink beams belong to a priority beam set for uplink reception from one or more source terminal devices for which said downlink beam is an optimal downlink beam,
the action in a given state is defined as an addition of a new uplink beam of the plurality of uplink beams to the priority beam set, a removal of an uplink beam from the priority beam set or doing nothing, and
the reward of taking a given action in a given state is calculated based on a change in uplink signal-to-noise ratio statistics of the one or more source terminal devices due to an action adjusted with a cost for taking the action; and
calculating iteratively at least one optimal state defining at least one priority beam set using at least one reinforcement learning model based on uplink signal-to-noise ratio statistics derived or derivable from uplink reference signal received power statistics for signals measured using a plurality of uplink beams of the access node from a plurality of terminal devices and on a plurality of optimal downlink beams of the access node for transmission to said plurality of terminal devices.
2. The apparatus of claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform, following generation of the at least one priority beam set:
maintaining, in said at least one memory or in an external memory, information on a priority beam set of one or more uplink beams of the access node, wherein the priority beam set is associated with a downlink beam of the access node, the priority beam set being one of said at least one priority beam set;
determining an optimal beam for uplink reception from one or more terminal devices for which said downlink beam is used for transmission by performing the following:
causing performing beam sweeping, at the access node, with the one or more terminal devices using the priority beam set; and
in response to a maximum received power measured for the priority beam set exceeding a first pre-defined power threshold following a completion of the beam sweeping of the priority beam set, selecting an uplink beam corresponding to the maximum received power measured for the priority beam set as the optimal beam.
3. The apparatus of claim 2 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
maintaining, in said at least one memory or in the external memory, information on a secondary beam set of one or more uplink beams of the access node, wherein the secondary beam set is associated with said downlink beam of the access node; and
performing the determining of the optimal beam for uplink reception further by:
in response to the maximum received power measured for the priority beam set failing to exceed the first pre-defined power threshold, causing performing beam sweeping, at the access node, with the one or more terminal devices using the secondary beam set; and
in response to a maximum received power measured for an uplink beam in the secondary beam set exceeding the first pre-defined power threshold, selecting the uplink beam of the secondary beam set corresponding to the maximum received power measured as the optimal beam.
4. The apparatus of claim 3 , wherein the determining of the optimal beam for uplink reception further comprises:
in response to a maximum received power measured for an uplink beam in the priority beam set exceeding a second pre-defined power threshold during the beam sweeping of the priority beam set, stopping the beam sweeping of the priority beam set and selecting said uplink beam of the priority beam set as the optimal beam, wherein the second pre-defined power threshold is higher than the first pre-defined power threshold; and
performing the selecting of the uplink beam corresponding to the maximum received power measured for the priority beam set as the optimal beam in response to the maximum received power measured for the priority beam set failing to exceed the second pre-defined power threshold but exceeding the first pre-defined power threshold following the completion of the beam sweeping of the priority beam set.
5. The apparatus of claim 3 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
in response to detecting, during the beam sweeping of secondary beam set, that the maximum received power measured for the uplink beam in the secondary beam set exceeds the first pre-defined power threshold, stopping the beam sweeping of the secondary beam set and performing the selecting of the uplink beam of the secondary beam set corresponding to the maximum received power measured for the secondary beam set as the optimal beam.
6. The apparatus according to claim 3 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
maintaining, in said at least one memory or the external memory, information on a plurality of priority beam sets of one or more uplink beams and a plurality of secondary beam sets of one or more uplink beams of the access node, wherein the plurality of priority and secondary beam sets are associated with a respective plurality of downlink beams; and
performing the determining of the optimal beam for uplink reception for a plurality of downlink beams based on the plurality of priority and secondary beam sets separately.
7. The apparatus according to claim 2 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
determining whether one or more pre-defined beam sweeping conditions for initiating uplink beam sweeping are satisfied; and
causing the determining of the optimal beam in response to at least one of the one or more pre-defined beam sweeping conditions being satisfied.
8. The apparatus of claim 7 , wherein the one or more pre-defined beam sweeping conditions define a pre-defined schedule for performing beam sweeping, a third pre-defined power threshold for power received using a current uplink beam and/or one or more pre-defined criteria for detecting excessively rapid switching between two uplink beams.
9. The apparatus according to claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
causing transmission of information on said at least one optimized state to the access node.
10. The apparatus according to claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processor, cause the apparatus to perform:
maintaining, in said at least one memory or an external memory, said uplink reference signal received power statistics for signals measured using said plurality of uplink beams of the access node from the plurality of terminal devices or said uplink signal-to-noise ratio statistics derivable from said uplink reference signal received power statistics and information on said plurality of optimal downlink beams of the access node for transmission to said plurality of terminal devices.
11. The apparatus according to claim 1 , wherein, in said at least one reinforcement learning model, an action defining an addition of a new uplink beam of the plurality of uplink beams to the priority beam set is associated with a positive cost, an action defining a removal of an uplink beam from the priority beam set is associated with a negative cost and an action defining doing nothing is associated with a zero cost.
12. The apparatus according to claim 1 , wherein, in said at least one reinforcement learning model, said change in the uplink signal-to-noise ratio statistics is defined as a dB-difference between Ath percentile signal-to-noise ratios after and before taking the action or as a sum or a weighted sum of a plurality of dB-differences between signal-to-noise ratios after and before taking the action calculated for different percentiles, A being a positive real number.
13. The apparatus according to claim 1 , wherein the calculating for a reinforcement learning model of said at least one reinforcement learning model comprises performing:
defining an initial state as a random state or using one or more pre-defined criteria,
repeating the following, starting from the initial state, until the highest reward of a plurality of calculated rewards is zero:
calculating, for a plurality of actions from a state, the plurality of rewards using the reinforcement learning model for a given downlink beam based on uplink signal-to-noise ratio statistics of one or more source terminal devices for which the given downlink beam is the optimal downlink beam derived from said reference signal receiver power statistics and
in response to the highest reward of the plurality of rewards being positive, executing an action of the plurality of actions associated with the highest reward so as to define a new state.
14. The apparatus of claim 13 , wherein the calculating of the plurality of rewards comprises:
determining uplink signal-to-noise ratio statistics for the state and a plurality of new states resulting from performing of the plurality of actions based on uplink reference signal received power statistics for the state and the plurality of new states, wherein the uplink reference signal received power statistics for the state and the plurality of new states are limited to statistics relating to the one or more source terminal devices; and
calculating the plurality of rewards based on a change in the uplink signal-to-noise ratio statistics between the current state and the plurality of new states adjusted with the cost for taking the action.
15. The apparatus according to claim 1 , wherein, in said at least one reinforcement learning model, the state is defined as a binary vector having a length equal to the number of the plurality of uplink beams producible by the access node.
16. The apparatus according to claim 1 , wherein the reward for adding an uplink beam, for removing an uplink beam and for doing nothing is defined as
R
n
=
∑
j
=
1
J
(
δ
j
Δ
SNR
j
[
dB
]
)
-
γ
[
dB
]
,
where
{
γ
>
0
for
adding
an
uplink
beam
γ
<
0
for
removing
an
uplink
beam
γ
=
0
for
doing
nothing
.
and J is a positive integer, δ j are pre-defined positive scaling factors, ΔSNR j define dB-differences of signal-to-noise ratios after and before taking the action for different percentiles of the uplink signal-to-noise ratio statistics and γ is the cost for taking the action.
17. The apparatus of claim 16 , wherein J is equal to 2.
18. A method, comprising:
implementing, separately for at least one downlink beam of a plurality of downlink beams of the access node, a reinforcement learning model, wherein a state, an action and a reward of the reinforcement learning model for a downlink beam are defined as follows:
the state defines which of the plurality of uplink beams belong to a priority beam set for uplink reception from one or more source terminal devices for which said downlink beam is an optimal downlink beam,
the action in a given state is defined as an addition of a new uplink beam of the plurality of uplink beams to the priority beam set, a removal of an uplink beam from the priority beam set or doing nothing, and
the reward of taking a given action in a given state is calculated based on a change in uplink signal-to-noise ratio statistics of the one or more source terminal devices due to an action adjusted with a cost for taking the action; and
calculating iteratively at least one optimal state defining at least one priority beam set using at least one reinforcement learning model based on uplink signal-to-noise ratio statistics derived or derivable from uplink reference signal received power statistics for signals measured using a plurality of uplink beams of the access node from a plurality of terminal devices and on a plurality of optimal downlink beams of the access node for transmission to said plurality of terminal devices.
19. A non-transitory computer readable medium encoded with program instructions that, when executed in hardware, cause the hardware to perform:
implementing, separately for at least one downlink beam of a plurality of downlink beams of an access node, a reinforcement learning model, wherein a state, an action and a reward of the reinforcement learning model for a downlink beam are defined as follows:
the state defines which of the plurality of uplink beams belong to a priority beam set for uplink reception from one or more source terminal devices for which said downlink beam is an optimal downlink beam,
the action in a given state is defined as an addition of a new uplink beam of the plurality of uplink beams to the priority beam set, a removal of an uplink beam from the priority beam set or doing nothing, and
the reward of taking a given action in a given state is calculated based on a change in uplink signal-to-noise ratio statistics of the one or more source terminal devices due to an action adjusted with a cost for taking the action; and
calculating iteratively at least one optimal state defining at least one priority beam set using at least one reinforcement learning model based on uplink signal-to-noise ratio statistics derived or derivable from uplink reference signal received power statistics for signals measured using a plurality of uplink beams of the access node from a plurality of terminal devices and on a plurality of optimal downlink beams of the access node for transmission to said plurality of terminal devices.