Dynamic machine learning decision threshold for resource allocation
Methods and apparatuses are provided for dynamic machine learning decision threshold for resource allocation/de-allocation. In one embodiment, a network node includes processing circuitry configured to cause the network node to dynamically adjust an allocation decision threshold; and determine whether to allocate at least one radio resource based at least in part on the allocation decision threshold. In one embodiment, a network node includes processing circuitry configured to cause the network node to dynamically adjust a de-allocation decision threshold; and determine whether to de-allocate at least one radio resource based at least in part on the de-allocation decision threshold.
1 . A method implemented in a network node, the method comprising:
dynamically adjusting an allocation decision threshold; and
determining whether to allocate at least one radio resource based at least in part on the allocation decision threshold, comprising:
estimating a benefit of allocating the at least one radio resource to a wireless device, wherein estimating the benefit to the wireless device is based at least in part on using a machine learning algorithm;
comparing the estimated benefit to the allocation decision threshold; and
one of allocating and not allocating the at least one radio resource to the wireless device based at least in part on the comparison of the estimated benefit to the allocation decision threshold,
wherein dynamically adjusting the allocation decision threshold comprises:
determining whether there is at least one radio resource that is available for the allocation to a wireless device;
increasing the allocation decision threshold by at least one step up parameter when there is at least one radio resource that is available for the allocation to the wireless device and the estimated benefit to the wireless device is greater than or equal to the allocation decision threshold; and
decreasing the allocation decision threshold by at least one step down parameter when there is an unavailability of at least one radio resource for the allocation to the wireless device,
wherein a size of at least one of the step up parameter and the step down parameter is based at least in part on a target allocation, wherein the size of the step down parameter is described according to a formula:
downStep=AllocTarget/1−allocTarget×upStep
where downStep is the step down parameter, upStep is the step up parameter, and allocTarget is the target allocation.
2 . The method of claim 1 , wherein determining whether to allocate the at least one radio resource further comprises:
determining whether to allocate the at least one radio resource for a sounding reference signal, SRS, based at least in part on the allocation decision threshold.
3 . The method of claim 1 , wherein dynamically adjusting the allocation decision threshold comprises:
dynamically adjusting the allocation decision threshold to achieve the target allocation.
4 . The method of claim 1 , wherein the target allocation comprises a target probability that an event will occur.
5 . The method of claim 4 , wherein the target allocation is based at least in part on:
a target allocation error.
6 . The method of claim 4 , wherein the target probability is a predetermined target probability that the event will occur, the event being one of:
that a radio resource is unavailable for allocation to a wireless device; and
that the radio resource is unavailable for allocation to the wireless device while the estimated benefit to the wireless device is greater than or equal to the allocation decision threshold.
7 . The method of claim 1 , wherein estimating the benefit to the wireless device is further based at least in part on a total of data transmitted in a downlink channel to the wireless device.
8 . A method implemented in a network node, the method comprising:
dynamically adjusting a de-allocation decision threshold; and
determining whether to de-allocate at least one radio resource based at least in part on the de-allocation decision threshold, comprising:
estimating a non-benefit of de-allocating the at least one radio resource to a wireless device, wherein estimating the non-benefit to the wireless device is based at least in part on using a machine learning algorithm;
comparing the estimated non-benefit to the de-allocation decision threshold; and
one of de-allocating and not de-allocating the at least one radio resource to the wireless device based at least in part on the comparison of the estimated non-benefit to the de-allocation decision threshold,
wherein dynamically adjusting the de-allocation decision threshold further comprises:
increasing the de-allocation decision threshold by at least one step up parameter when the one of the de-allocating and not de-allocating is de-allocating; and
decreasing the de-allocation decision threshold by at least one step down parameter when the one of the de-allocating and not de-allocating is not de-allocating,
wherein a size of at least one of the step up parameter and the step down parameter is based at least in part on a target de-allocation, wherein the size of the step down parameter is described according to a formula:
downStepDealloc=deallocTarget/1−deallocTarget×upStepDealloc
where downStepDealloc is the step down parameter, upStepDealloc is the step up parameter, and deallocTarget is the target de-allocation.
9 . The method of claim 8 , wherein determining whether to deallocate the at least one radio resource further comprises:
determining whether to de-allocate the at least one radio resource for a sounding reference signal, SRS, based at least in part on the de-allocation decision threshold.
10 . The method of claim 8 , wherein dynamically adjusting the de-allocation decision threshold comprises:
dynamically adjusting the de-allocation decision threshold to achieve the target de-allocation.
11 . The method of claim 8 , wherein the target de-allocation comprises a target probability that an event will occur.
12 . The method of claim 11 , wherein the target de-allocation is based at least in part on:
a target de-allocation error.
13 . The method of claim 11 , wherein the target probability is a predetermined target probability associated with at least one of:
a probability of de-allocating resources to a wireless device; and
a probability of de-allocating resources to the wireless device and a same wireless device is subsequently allocated resources.
14 . The method of claim 8 , wherein dynamically adjusting the de-allocation decision threshold further comprises:
decreasing the de-allocation decision threshold by at least one step down parameter when the one of the de-allocating and not de-allocating is de-allocating; and
increasing the de-allocation decision threshold by at least one step up parameter and at least one step down parameter when the one of the de-allocating and not de-allocating is not deallocating and is further a re-allocation.
15 . The method of claim 8 , wherein estimating the non-benefit to the wireless device is further based at least in part on a total of data transmitted in a downlink channel to the wireless device.
16 . A network node comprising processing circuitry, the processing circuitry configured to cause the network node to:
dynamically adjust an allocation decision threshold; and
determine whether to allocate at least one radio resource based at least in part on the allocation decision threshold, comprising:
estimating a benefit of allocating the at least one radio resource to a wireless device, wherein estimating the benefit to the wireless device is based at least in part on using a machine learning algorithm;
comparing the estimated benefit to the allocation decision threshold; and
one of allocating and not allocating the at least one radio resource to the wireless device based at least in part on the comparison of the estimated benefit to the allocation decision threshold,
wherein dynamically adjusting the allocation decision threshold comprises:
determining whether there is at least one radio resource that is available for the allocation to a wireless device;
increasing the allocation decision threshold by at least one step up parameter when there is at least one radio resource that is available for the allocation to the wireless device and the estimated benefit to the wireless device is greater than or equal to the allocation decision threshold; and
decreasing the allocation decision threshold by at least one step down parameter when there is an unavailability of at least one radio resource for the allocation to the wireless device,
wherein a size of at least one of the step up parameter and the step down parameter is based at least in part on a target allocation, wherein the size of the step down parameter is described according to a formula:
downStep=allocTarget/1−allocTarget×upStep
where downStep is the step down parameter, upStep is the step up parameter, and allocTarget is the target allocation.
17 . A network node comprising processing circuitry, the processing circuitry configured to cause the network node to:
dynamically adjust a de-allocation decision threshold; and
determine whether to de-allocate at least one radio resource based at least in part on the de-allocation decision threshold, comprising:
estimating a non-benefit of de-allocating the at least one radio resource to a wireless device, wherein estimating the non-benefit to the wireless device is based at least in part on using a machine learning algorithm;
comparing the estimated non-benefit to the de-allocation decision threshold; and
one of de-allocating and not de-allocating the at least one radio resource to the wireless device based at least in part on the comparison of the estimated non-benefit to the de-allocation decision threshold,
wherein dynamically adjusting the de-allocation decision threshold further comprises:
increasing the de-allocation decision threshold by at least one step up parameter when the one of the de-allocating and not de-allocating is de-allocating; and
decreasing the de-allocation decision threshold by at least one step down parameter when the one of the de-allocating and not de-allocating is not de-allocating,
wherein a size of at least one of the step up parameter and the step down parameter is based at least in part on a target de-allocation, wherein the size of the step down parameter is described according to a formula:
downStepDealloc=deallocTarget/1−deallocTarget×upStepDealloc
where downStepDealloc is the step down parameter, upStepDealloc is the step up parameter, and deallocTarget is the target de-allocation.
18 . The method of claim 1 , wherein estimating the benefit to the wireless device is further based at least in part on a total of time that the wireless device has been active in a system of the network node.
19 . The method of claim 1 , wherein estimating the benefit to the wireless device is further based at least in part on a downlink inactivity time.
20 . The method of claim 8 , wherein estimating the non-benefit to the wireless device is further based at least in part on a total of time that the wireless device has been active in a system of the network node.
21 . The method of claim 8 , wherein estimating the non-benefit to the wireless device is further based at least in part on a downlink inactivity time.
22 . The method of claim 14 , wherein the de-allocation decision threshold is derived by adding a hysteresis parameter to a negative value of an allocation decision threshold.
23 . The method of claim 8 , wherein the de-allocation decision threshold is derived by adding a hysteresis parameter to a result of one minus an allocation decision threshold.
24 . The method of claim 4 , wherein the target allocation is based at least in part on a cost associated with allocating the at least one radio resource.
25 . The method of claim 4 , wherein the target allocation is based at least in part on a number of radio resource control, RRC, reconfigurations associated with allocating the at least one radio resource.
26 . The method of claim 11 , wherein the target de-allocation is based at least in part on a cost associated with de-allocating the at least one radio resource.
27 . The method of claim 11 , wherein the target de-allocation is based at least in part on a number of radio resource control, RRC, reconfigurations associated with de-allocating the at least one radio resource.