Machine learning based adaptation of QoE control policy
A node of a wireless communication network receives first data indicating a desired quality of experience level for user data traffic of a user of the wireless communication network. Based on a control policy and the desired quality of experience level, the node determines a rule for controlling the user data traffic. Further, the node obtains second data indicating an estimated quality of experience level for the user data traffic subject to control according to the rule. Based on the first data and the second data, the node adapts the control policy, e.g., using a reinforcement learning, RL, mechanism.
1. A method in a first node of controlling user data traffic in a wireless communication network, the method comprising:
receiving from a session management function (SMF) a session modification (SM) request and a desired quality of experience (QoE) level for user data traffic of a user of the wireless communication network, the SM request including a quality enforcement rule (QER);
determining an estimated QoE level for the user data traffic subject to the QER;
adapting the QER based at least in part on a reward obtained from a reinforced learning process applied to the desired QoE level and the estimated QoE level; and
applying the adapted QER to the user data traffic.
2. The method according to claim 1 , comprising:
estimating the QoE level for the user data traffic subject to control according to the QER.
3. The method according to claim 1 , comprising:
receiving data indicating an actual QoE level for the user data traffic; and
adapting a control policy based on the received data.
4. The method according to claim 3 , wherein computation of the reward of the reinforced learning process is further based on received data.
5. The method according to claim 1 , further comprising adapting a control policy based on the reinforced learning process, and wherein computation of a reward of the reinforced learning process is based on first data, computation of a state of the reinforced learning process is based on second data, and a control rule corresponds to an action from an action space of the reinforced learning process.
6. The method according to claim 1 , wherein the desired QoE level is user specific.
7. The method according to claim 1 , wherein the user data traffic is generated by one or more services and the desired QoE level is service specific.
8. The method according to claim 1 , further comprising receiving first data in response to starting of a service generating the user data traffic.
9. The method according to claim 8 , wherein the first data originates from a provider of a service generating the user data traffic.
10. The method according to claim 1 , comprising:
indicating a capability of controlling the user data traffic in accordance with the desired QoE level to at least one further node of the wireless communication network.
11. The method according to claim 1 , comprising:
forwarding the user data traffic to or from a user equipment connected to the wireless communication network.
12. The method according to claim 1 , wherein the first node includes at least one of a General Packet Data Service Gateway Support Node, GGSN, a Packet Data Gateway, PGW, and a User Plane Function, UPF of a 3 rd Generation Partnership Project, 3GPP, technology.
13. A method in a first node of controlling user data traffic in a wireless communication network, the method comprising:
indicating to a second node of the wireless communication system, a session modification (SM) request and a desired quality of experience level for user data traffic of a user of the wireless communication network, the SM request including a quality enforcement rule (QER);
sending an adapted QER to the second node, the adapted QER being based at least in part on a reward obtained at the second node from a reinforced learning process applied to the desired QoE level and the estimated QoE level; and
indicating to the second node an actual QoE level for the user data traffic, the actual QoE level being based at least in part on the adapted QER.
14. The method according to claim 13 , comprising:
monitoring the actual QoE level.
15. The method according to claim 14 , comprising:
monitoring the actual QoE level based on at least one of user data traffic generated by the second node and user data traffic received by the second node.
16. The method according to claim 13 , comprising:
receiving an indication of a capability of the second node to control the user data traffic in accordance with the desired QoE level; and
in response to the indication, providing data to the second node, the data indicating the actual QoE level.
17. The method according to claim 13 , wherein the second node is configured to forward the user plane traffic to or from a user equipment connected to the wireless communication network.
18. The method according to claim 13 , wherein the second node includes at least one of a General Packet Data Service Gateway Support Node, GGSN, a Packet Data Gateway, PGW, and a User Plane Function, UPF, of a 3rd Generation Partnership Project technology.
19. A node for a wireless communication network, the node being configured to:
receive first data indicating a desired quality of experience (QoE) level for user data traffic of a user of the wireless communication network and a quality enforcement rule (QER);
determine an estimated QoE level for the user data traffic subject to the QER; and
adapt the QER based at least in part on a reward obtained from a reinforced learning process applied to the desired QoE level and the estimated QoE level; and
apply the adapted QER to the user data traffic.
20. A first node for a wireless communication network, the node being configured to:
indicate to a second node of the wireless communication network, a session modification (SM) request and a desired quality of experience QoE level for user data traffic of a user of the wireless communication network, the SM request including a quality enforcement rule (QER);
send an adapted QER to the second node, the adapted QER being based at least in part on a reward obtained from a reinforced learning process applied to the desired QoE level and the estimated QoE level; and
indicate to the second node an actual quality of experience QoE level for the user data traffic, the actual QoE level being based at least in part on the adapted QER.