Harnessing predictive models of durations of channel availability for enhanced opportunistic allocation of radio spectrum

• US 8,254,393 B2
• Filed: 06/29/2007
• Issued: 08/28/2012
• Est. Priority Date: 06/29/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Horvitz, Eric J.
• Primary Examiner(s): Hailu, Kibrom T
• Assistant Examiner(s): Beyen, Zewdu A
• Application Number:
  US11/771,349
• Publication Number:
  US 20090003201A1
• Time in Patent Office: 1,887 Days
• Field of Search: 705/80, 370/348, 370/395.41
• US Class Current: 370/395.41
• CPC Class Codes: H04L 41/0896 Bandwidth or capacity management, i.e. automatically increasing or decreasing capacities flow or congestion control using dynamic resource allocation, e.g. in-call renegotiation, H04L47/76
• CPC Class Codes: H04L 41/147 for predicting network behaviour

Abstract

A proactive adaptive radio methodology for the opportunistic allocation of radio spectrum is described. The methods can be used to allocate radio spectrum resources by employing machine learning to learn models, via accruing data over time, that have the ability to predict the context-sensitive durations of the availability of channels. The predictive models are combined with decision-theoretic cost-benefit analyses to minimize disruptions of service or quality that can be associated with reactive allocation policies. Rather than reacting to losses of channel, the proactive policies seek switches in advance of the loss of a channel. Beyond determining durations of availability for one or more frequency bands statistical machine learning also be employed to generate price predictions in order to facilitate a sale or rental of the available frequencies, and these predictions can be employed in the switching analyses The methods can be employed in non-cooperating distributed models of allocation, in centralized allocation approaches, and in hybrid spectrum allocation scenarios.

Claims

1. A system that optimizes usage of resources in a communication network comprising:

a bandwidth sensing component that;

learns from monitoring communication channels over time and identifies one or more communication channels of unidentified availability to use in an opportunistic manner based at least upon a prediction of duration of availability of the bandwidth; and

infers future availability of the one or more identified channels of unidentified availability, a duration of future availability of the one or more identified channels of unidentified availability, or both, by utilizing statistical machine learning and a set of stored event data comprising mean inter-availability intervals or bursts of usage from the monitoring of the communication channels over time; and

a bandwidth allocation component that adaptively determines time allocations and price predictions for the one or more identified communication channels of unidentified availability, and allocates the one or more identified channels of unidentified availability based on frequency demands, the bandwidth allocation component comprising;

a preference model that;

guides decisions regarding proactive switching of channels versus an imposed reactive switch;

maximizes duration of usage of a first channel by monitoring time of first channel usage and switching to a second channel pre-selected from the one or more communication channels of unidentified availability based on a decision-theoretic employing probability distribution based on likelihoods, costs, and benefits of different switching times; and

proactively switches channels prior to an imposed reactive switch upon rising probability of being forced into the reactive switch.

2. The system of claim 1, the bandwidth sensing component predicts the duration of availability in a cellular telephone network, by estimating time until at least one of an arrival of a new call or a call handed off from another cell which would occupy the one or more identified channels of unidentified availability.

3. The system of claim 1, the bandwidth sensing component further comprises a statistical classifier in order to sense the availability and duration of the one or more identified channels of unidentified availability.

4. The system of claim 3, the classifier is explicitly trained based at least on historical data wherein the availability and duration of availability of the one or more identified channels of unidentified availability are considered.

5. The system of claim 3, the classifier is implicitly trained based at least on current inferences regarding the availability and duration of availability of the one or more identified channels of unidentified availability.

6. The system of claim 1, the bandwidth allocation component employs at least frequency demand to facilitate determination of the allocations.

7. The system of claim 1, the bandwidth allocation component further predicts the prices based on a geographic location that generates the frequency demand.

8. The system of claim 1, the bandwidth allocation component further comprises a classifier to at least one of allocate the one or more identified channels of unidentified availability or generate the price predictions.

9. The system of claim 8, the classifier is explicitly trained based at least on historical data wherein the allocations and the price predictions of the one or more identified channels of unidentified availability were correctly inferred.

10. The system of claim 8, the classifier is implicitly trained based at least on current inferences regarding the allocations and the price predictions of the one or more identified channels of unidentified availability.

11. The system of claim 1, the bandwidth allocation component allocates the one or more identified channels of unidentified availability in a cellular communication network based at least upon co-channel interference.

12. The system of claim 1, wherein the statistical machine learning comprises neural networks, expert systems, Bayesian belief networks, support vector machines (SVMs), Hidden Markov Models (HMMs), fuzzy logic, and/or data fusion.

13. The system of claim 1, wherein the bandwidth sensing component further builds one or more predictive models to yield one or more statistical classifiers to infer the future availability of the one or more identified channels of unidentified availability, the duration of future availability of the one or more free channels of unidentified availability, or both.

14. One or more computer storage media not consisting of a signal per se, storing computer-executable instructions for optimizing usage of communication network resources, the instructions, when executed, configure a processor to perform acts comprising:

monitoring frequency bands in a network;

storing a set of frequency band event data comprising mean inter-availability intervals or bursts of usage from the monitoring of the frequency bands;

building, using the set of frequency band event data and statistical machine learning, predictive models to predict duration of future availability of the one or more channels;

identifying one or more of the one or more predicted channels that is likely to be free in the future based at least upon detected bandwidth usage and the predictive models;

adaptively determining allocations for the one or more channels that are likely to be free in the future;

adaptively determining price predictions for the one or more channels that are likely to be free in the future; and

maximizing duration of usage for the one or more channels that are likely to be free in the future based on a decision-theoretic employing probability distribution based on likelihoods, costs, and benefits of different switching times.

15. The one or more computer storage media not consisting of a signal per se of claim 14, wherein the identifying identifies the one or more channels that are likely to be free in the future by employing automated decision making harnessing statistical models of channel availability.

16. A system that predicts channel availability across radio frequency channels allocated to respective service providers by comprising:

a bandwidth sensing component that;

monitors a plurality of radio frequency channels allocated to respective service providers;

stores radio frequency channel event data comprising mean inter-availability intervals or bursts of usage from the monitoring of the plurality of radio frequency channels allocated to respective service providers;

builds, using the stored radio frequency channel event data and statistical machine learning, predictive models to predict a point in time, and a duration of time, when each of the plurality of radio frequency channels allocated to respective service providers is likely to be available in the future;

identifies one or more of the plurality of radio frequency channels allocated to respective service providers to be used in the future by any of the respective service providers based at least upon the predictive models; and

a bandwidth allocation component that;

determines one or more of the identified one or more radio frequency channels allocated to respective service providers to be used by other service providers based at least upon the predicted points in time, durations of time, and price prediction based on a decision-theoretic employing probability distribution based on likelihoods, costs, and benefits of different switching times of the identified one or more radio frequency channels; and

allocates the determined one or more radio frequency channels to be used by the other service providers based on frequency demands of the plurality of radio frequency channels allocated to respective service providers.