Learning-based spectrum occupancy prediction exploiting multi-dimensional correlation

Patent metadata

Field	Value
Publication number	US-12526644-B2
Application number	US-202118249789-A
Country	US
Kind code	B2
Filing date	Oct 12, 2021
Priority date	Oct 23, 2020
Publication date	Jan 13, 2026
Grant date	Jan 13, 2026

Abstract

The present disclosure relates to a spectrum occupancy prediction, which employs a first trainable module and a second trainable module. The first and the second trainable modules predict spectrum occupancy at a respective first and second communication devices based on past occupancies and/or occupancies in adjacent subband(s). The prediction from the first trainable module and the second trainable module is input to a trainable output (third) module, which then provides the spectrum occupancy prediction.

First claim

The invention claimed is:

1 . A method for prediction of spectrum occupancy comprising:
predicting, by applying a first trainable model, a first likelihood of spectrum occupancy in a wireless bandwidth based on past spectrum occupancy related to a first communication device, selecting a second communication device based on its relative position to the first communication device, predicting, by applying a second trainable model, a second likelihood of spectrum occupancy in the wireless bandwidth based on past spectrum occupancy related to the second communication device, predicting, by applying a trainable output model, spectrum occupancy in the wireless bandwidth based on the first likelihood and the second likelihood, and performing resource management for the first communication device and the second communication device according to the spectrum occupancy predicted by the trainable output model.

2 . The method according to claim 1 , wherein the predicting of the first likelihood and/or the predicting of the second likelihood for a first subband of the wireless bandwidth is further based on frequency occupancy in a second subband of the wireless bandwidth different from the first subband.

3 . The method according to claim 1 , further comprising inputting to each of the first trainable model and the second trainable model a spectrum occupancy vector for a predetermined time instant, in which:
each vector coordinate is associated with a subband of the wireless bandwidth, and a value of said vector coordinate indicates whether the associated subband was occupied at the predetermined time instant.

4 . The method according to claim 1 , wherein the first communication device is a base station, the second communication device is a base station, the prediction by the output trainable model relates to a third communication device which is a user equipment; and
the method further comprises selecting of the first communication device and the second communication device based on their relative position to the third communication device.

5 . The method according to claim 4 , wherein the relative position between a base station and a user equipment is determined based on at least one of:
a number of handovers between the base station and the user equipment; channel quality between the base station and the user equipment; and/or spatial distance between the base station and the user equipment.

6 . The method according to claim 1 , wherein the first communication device is a user equipment, the second communication device is a user equipment, and the prediction by the trainable output model relates to the first communication device.

7 . The method according to claim 4 , further comprising performing resource management for the third communication device according to the spectrum occupancy predicted by the trainable output model.

8 . The method according to claim 1 , further comprising N steps of predicting, wherein:
each q-th step, q being one to N, comprises applying an q-th trainable model to predict a q-th likelihood of spectrum occupancy in the wireless bandwidth based on past spectrum occupancy related to a q-th communication device, N being an integer larger than 2; the predicting, by applying the trainable output model, spectrum occupancy in the wireless bandwidth is based on each q-th likelihood.

9 . The method according to claim 1 , wherein the first trainable model, the second trainable model, and/or the trainable output model comprise at least one of convolutional neural network and/or long-short term memory.

10 . The method according to claim 1 , wherein
the past spectrum occupancy related to the first communication device and/or the past spectrum occupancy related to the second communication device are obtained comprising measuring power spectral density or received signal strength indicator values.

11 . A method for training of a predictor for prediction of spectrum occupancy, the model comprising N trainable models and a trainable output model, wherein the method comprises:
obtaining, for each q being 1 to N with N being an integer larger than 1, a q-th pair of spectrum occupancy likelihood in a wireless bandwidth related to a q-th communication device and a ground-truth spectrum occupancy in the wireless bandwidth, wherein one or more of the N communication devices are selected based on their respective relative positions; training each q-th trainable model among N trainable models with the q-th pair; and training the trainable output model with the outputs of the N trainable models and the ground-truth spectrum occupancy.

12 . The method according to claim 11 , wherein a q-th spectrum occupancy in the wireless bandwidth is obtained by detecting the spectrum occupancy by a q-th communication device.

13 . The method according to claim 11 , wherein the training of the q-th trainable model is performed as a multi-task training for a plurality of subbands at the same time.

14 . A device for prediction of spectrum occupancy comprising:
a first trainable model for predicting a first likelihood of spectrum occupancy in a wireless bandwidth based on past spectrum occupancy related to a first communication device, a second trainable model for predicting a second likelihood of spectrum occupancy in the wireless bandwidth based on past spectrum occupancy related to a second communication device, wherein the second communication device is selected based on its relative position to the first communication device, a trainable output model for predicting spectrum occupancy in the wireless bandwidth based on the first likelihood and the second likelihood, wherein the device for prediction of spectrum occupancy is further configured to perform resource management for the first communication device and the second communication device according to the spectrum occupancy predicted by the trained output model.

15 . The device for prediction of spectrum occupancy according to claim 14 ,
wherein the first trainable model and/or the second trainable model and/or the trainable output model are embodied on one or more FPGAs.

Assignees

• Vestel Elektronik Sanayi Ve Ticaret As

Inventors

• Aygül Mehmet Ali
• Nazzal Mahmoud
• Arslan Hüseyin

Classifications

• H04W16/22 Primary

  Traffic simulation tools or models · CPC title

• H04L41/16

  using machine learning or artificial intelligence · CPC title

• H04B17/3912

  Simulation models, e.g. distribution of spectral power density or received signal strength indicator [RSSI] for a given geographic region · CPC title

• H04W4/023

  using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds · CPC title

• G06N20/00

  Machine learning · CPC title