Dynamic quantized signature vector selection for a cloud radio access network
A communication system is disclosed. The communication system includes a plurality of radio points, each configured to exchange radio frequency (RF) signals with a plurality of wireless devices at a site. The communication system also includes a baseband controller communicatively coupled to the plurality of radio points. The communication system also includes a machine learning computing system communicatively coupled to the baseband controller. The machine learning computing system is configured to determine an expected average throughput associated with each of a plurality of global quantized signature vectors (QSVs), using a Q-function approximation, based on a current state of the communication system. The communication system is also configured to select a global QSV associated with a highest expected average throughput.
1. A communication system, comprising:
a plurality of radio points, each configured to exchange radio frequency (RF) signals with a plurality of wireless devices;
a baseband controller communicatively coupled to the plurality of radio points; and
a machine learning computing system communicatively coupled to the baseband controller, wherein the machine learning computing system is configured to:
determine an expected average throughput associated with each of a plurality of global quantized signature vectors (QSVs), using a Q-function approximation, based on a current state of the communication system;
select a global QSV associated with a highest expected average throughput; and
update the Q-function approximation based on a next state of the communication system and an actual throughput resulting from application of the selected global QSV.
2. The communication system of claim 1 , wherein the baseband controller is configured to determine the current state of the communication system.
3. The communication system of claim 1 , wherein the baseband controller is configured to determine the current state of the communication system based on one or more of the following:
a location of at least one of the wireless devices;
a signature vector for at least one of the wireless devices;
a Quality of Service (QoS) priority of at least one of the wireless devices;
a backlog associated with at least one of the wireless devices;
one or more channel conditions in the communication system; and
a Block Error Rate (BLER) of at least one of the wireless devices.
4. The communication system of claim 1 ,
wherein the baseband controller is configured to determine, for each wireless device, a QSV based on a signature vector for the wireless device; and
wherein the machine learning computing system is further configured to determine the global QSV based on at least some of the QSVs for the wireless devices.
5. The communication system of claim 4 , wherein each QSV is a vector that comprises an element for each of the radio points, wherein each element has a value of 1 for any of the radio points that are communicating with the respective wireless device and a value of 0 for any of the radio points that are not communicating with the respective wireless device.
6. The communication system of claim 1 , wherein each signature vector for a wireless device is determined by the baseband controller based on at least one signal metric, wherein the at least one signal metric is determined at an RP based on a wireless signal received at the RP from the wireless device.
7. The communication system of claim 1 , wherein the machine learning computing system is further configured to apply the selected global QSV by, for each of at least some of the wireless devices, assigning or re-assigning at least one of the radio points to communicate with the respective wireless device.
8. The communication system of claim 1 , wherein the machine learning computing system is further configured to:
determine an expected average throughput associated with each of a plurality of next QSVs, using the Q-function approximation, based on the next state of the communication system; and
select a next global QSV associated with the highest expected average throughput.
9. The communication system of claim 8 , wherein the machine learning computing system is further configured to:
apply the selected next global QSV by assigning or re-assigning at least one of the radio points to communicate with at least one of the wireless devices; and
update the Q-function approximation.
10. The communication system of claim 1 , wherein the expected average throughput is estimated using Q-learning to account for the current state and at least one future state of the communication system.
11. A method for dynamic quantized signature vector (QSV) selection in a communication system, the communication system comprising a baseband controller and a plurality of radio points, wherein each radio point is configured to exchange radio frequency (RF) signals with a plurality of wireless devices, the method comprising:
determining an expected average throughput associated with each of a plurality of global QSVs, using a Q-function approximation, based on a current state of the communication system;
selecting a global QSV associated with a highest expected average throughput; and
updating the Q-function approximation based on a next state of the communication system and an actual throughput resulting from application of the selected global QSV.
12. The method of claim 11 , further comprising determining the current state of the communication system.
13. The method of claim 11 , wherein the current state of the communication system is based on one or more of the following:
a location of at least one of the wireless devices;
a signature vector for at least one of the wireless devices;
a Quality of Service (QoS) priority of at least one of the wireless devices;
a backlog associated with at least one of the wireless devices;
one or more channel conditions in the communication system; and
a Block Error Rate (BLER) of at least one of the wireless devices.
14. The method of claim 11 , further comprising:
determining, for each wireless device, a QSV based on a signature vector for the wireless device; and
determining the global QSV based on at least some of the QSVs for the wireless devices.
15. The method of claim 14 , wherein each QSV is a vector that comprises an element for each of the radio points, wherein each element has a value of 1 for any of the radio points that are communicating with the respective wireless device and a value of 0 for any of the radio points that are not communicating with the respective wireless device.
16. The method of claim 11 , wherein each signature vector for a wireless device is determined by the baseband controller based on at least one signal metric, wherein the at least one signal metric is determined at an RP based on a wireless signal received at the RP from the wireless device.
17. The method of claim 11 , further comprising applying the selected global QSV by, for each of at least some of the wireless devices, assigning or re-assigning at least one of the radio points to communicate with the respective wireless device.
18. The method of claim 11 , further comprising:
determining an expected average throughput associated with each of a plurality of next QSVs, using the Q-function approximation, based on the next state of the communication system; and
selecting a next global QSV associated with the highest expected average throughput.
19. The method of claim 18 , further comprising:
applying the selected next global QSV by assigning or re-assigning at least one of the radio points to communicate with at least one of the wireless devices; and
updating the Q-function approximation.