US 12,739,017 B2Granted

Cell shaping with reinforced learning

AssigneeTelefonaktiebolaget LM Ericsson (publ)

InventorsGuoqiang Lu, Xixian Chen, Haomin Li, Edward Mah

Life of this patent

Estimating the term…

Loading patent…

Abstract

A method and network node for cell shaping with reinforced learning are disclosed. In some embodiments, for each of a plurality of phase offset trial values: a first reward in response to first trial value applied to antenna elements having a first polarization is determined. A second reward in response to a second trial value applied to antenna elements having a second polarization is determined. A first phase offset apply to the antenna element having the first polarization is determined based at least in part on the plurality of first rewards and on a probable reward in response to the first phase offset. A second phase offset to be applied to antenna elements having the second polarization is determined based at least in part on the plurality of second rewards and on a probable reward in response to the second phase offset.

Claims18

Dependent claims

A method for determining precoder weights to be applied to antenna elements of an array of antenna elements in a network node, the method comprising:

for each of a plurality of phase offset trial values:
determining a first reward in response to first phase offset trial value applied to each of at least one antenna element having a first polarization;
and
determining a second reward in response to a second phase offset trial value applied to each of at least one antenna element having a second polarization;
determining a subsequent first phase offset to be applied to each of the at least one antenna element having the first polarization based at least in part on the plurality of first rewards and based at least in part on a probable reward in response to the subsequent first phase offset;
determining a subsequent second phase offset to be applied to each of the at least one antenna element having the second polarization based at least in part on the plurality of second rewards and based at least in part on a probable reward in response to the subsequent second phase offset;
determining a subsequent phase offset being based at least in part on a Thompson sampling algorithm, the Thompson sampling algorithm configured to model the probable reward based at least in part on a Gaussian distribution;
and
at least one of the first reward and the second reward being based at least in part on a feedback signal from each of at least one wireless device (WD).

6 more elements

A network node configured to determine precoder weights to be applied to antenna elements of an array of antenna elements, the network node comprising processing circuitry configured to:

for each of a plurality of phase offset trial values:
determine a first reward in response to first phase offset trial value applied to each of at least one antenna element having a first polarization;
and
determine a second reward in response to a second phase offset trial value applied to each of at least one antenna element having a second polarization;
determine a subsequent first phase offset to be applied to each of the at least one antenna element having the first polarization based at least in part on the plurality of first rewards and based at least in part on a probable reward in response to the subsequent first phase offset;
and
determine a subsequent second phase offset to be applied to each of the at least one antenna element having the second polarization based at least in part on the plurality of second rewards and based at least in part on a probable reward in response to the subsequent second phase offset, determining a subsequent phase offset being based at least in part on a Thompson sampling algorithm, the Thompson sampling algorithm configured to model the probable reward based at least in part on a Gaussian distribution, and at least one of the first reward and the second reward being based at least in part on a feedback signal from each of at least one wireless device (WD).

4 more elements

Dependencies are inferred from the claim text.