Automatic fine-grained radio map construction and adaption
View Patent ↗An automatic wireless fine-grained ratio map construction and adaptation system may include a Gaussian process regression (GPR) model constructed with real wireless received signal strength (RSS) measurements collected in a free space to provide coarse RSS estimation in a constrained space, and a generative adversarial network (GAN) to provide fine-grained RSS estimation in the constrained space by using an output of GPR as an input for a generator of GAN, modeling the irregular RSS distributions in complex indoor environments. The system may generate realistic RSS data in the constrained space that has not been manually site-surveyed.
1. An automatic wireless fine-grained ratio map construction and adaptation system, comprising:
a sensor configured to collect real wireless received signal strength (RSS) measurements in a free space;
a processor configured to construct a Gaussian process regression (GPR) model based at least in part on the real wireless RSS measurements collected by the sensor in the free space to provide coarse RSS estimation in a constrained space; and
a generator of a generative adversarial network (GAN) configured to provide fine-grained RSS estimation in the constrained space by using an output of the GPR as an input for the generator of the GAN.
2. The system of claim 1 , wherein the system is further configured to model the irregular RSS distributions in complex indoor environments.
3. The system of claim 2 , wherein the system is further configured to generate realistic RSS data in the constrained space that has not been manually site surveyed.
4. The system of claim 1 , wherein the environment can be classified as the free space where the sensor can access freely.
5. The system of claim 1 , wherein the environment can be classified as the constrained space where the sensor cannot access easily or where the noise of its measurement is high.
6. The system of claim 1 , wherein the processor is configured to construct the GPR model by using the RSS measurements at calibration points as well as their two-dimensional coordinates collected by the sensor in the free space to capture anomalous RSS variations on a rough level.
7. The system of claim 6 , wherein the processor is further configured to construct the GPR model by providing a coarse RSS estimation at the calibration points and adopting them as the input for the generator of the GAN instead of random noise.
8. The system of claim 7 , wherein the nonlinear relations between the spatial and radio space captured by the GPR can be inherited by the GAN.
9. The system of claim 1 , wherein the generative adversarial network (GAN) is configured to reveal the irregular RSS distribution in an environment that is not explored by the GPR.
10. The system of claim 1 , wherein the generative adversarial network (GAN) is configured to synthesize realistic RSS data to fool a discriminator of the GAN while the discriminator of the GAN tries to distinguish whether the data is real or fake.
11. The system of claim 1 , wherein providing the fine-grained RSS estimation comprises leveraging the GPR model to generate coarse estimations and use them as the input for the generator of the GAN.
12. The system of claim 11 , wherein the output of the generator of the GAN is the fine-grained RSS estimation.
13. A method, comprising:
a sensor of an automatic wireless fine-grained ratio map construction and adaptation system collecting real wireless received signal strength (RSS) measurements in a free space;
a processor of the automatic wireless fine-grained ratio map construction and adaptation system constructing a Gaussian process regression (GPR) model based at least in part on the real wireless RSS measurements collected by the sensor in the free space to provide coarse RSS estimation in a constrained space; and
a generator of a generative adversarial network (GAN) of the automatic wireless fine-grained ratio map construction and adaptation system providing fine-grained RSS estimation in the constrained space by using an output of the GPR as an input for the generator of the GAN.
14. The method of claim 13 , further comprising modeling the irregular RSS distributions in complex indoor environments.
15. The method of claim 14 , further comprising generating realistic RSS data in the constrained space that has not been manually site surveyed.
16. The method of claim 13 , wherein the processor constructing the GPR model includes the processor using the RSS measurements at calibration points as well as their two-dimensional coordinates collected by the sensor in the free space to capture anomalous RSS variations on a rough level.
17. The method of claim 16 , further comprising the processor constructing the GPR model by providing a coarse RSS estimation at the calibration points and adopting them as the input for the generator of the GAN instead of random noise.
18. The method of claim 13 , further comprising the GAN revealing the irregular RSS distribution in an environment that is not explored by the GPR.
19. The method of claim 13 , further comprising the GAN synthesizing realistic RSS data to fool a discriminator of the GAN while the discriminator of the GAN tries to distinguish whether the data is real or fake.
20. The method of claim 13 , wherein the generator providing the fine-grained RSS estimation comprises the generator leveraging the GPR model to generate coarse estimations and use them as the input for the generator of the GAN.