Robust TOA-estimation using convolutional neural networks (or other function approximations) on randomized channel models
Methods and systems related to neural networks or other function approximators operate for training a neural network is provided, or another function approximator, for inferring a predetermined time of arrival of a predetermined transmitted signal on the basis of channel-impulse-responses, CIRs, of transmitted signals between a mobile antenna and a fixed antenna, the method having: obtaining a channel impulse response condition characteristic, CIRCC, descriptive of channel impulse responses of transmitted signals associated with mobile antenna positions within a reach of the fixed antenna; generating, by simulation, a training set of simulated CIRs which are associated with different times of arrival in one or more simulated scenes, and which fit to the CIRCC; training the neural network, or other function approximator, using the simulated CIRs and the different associated times of arrivals to obtain a parametrization of the neural network, or other function approximator, associated with the CIRCC.
1 . A method for classifying a mobile antenna, exchanging predetermined transmitted signals with a fixed antenna, on the basis of a predetermined channel impulse response, CIR, from which a time-of-arrival measurement is to be performed, the method comprising:
using a neural network or a function approximator, which comprises a multidimensional polynomial function that learns a fitting function, for inferring from the CIRs of the predetermined transmitted signals, times of arrival of the predetermined transmitted signals exchanged between the mobile antenna and the fixed antenna,
determining, based on the times of arrival, or channel impulse response condition characteristic, CIRCC, which is descriptive of channel impulse responses of the transmitted signals exchanged between the mobile antenna and the fixed antenna, a movement profile of the mobile antenna,
analyzing the movement profile of the mobile antenna in terms of plausibility,
in case of the movement profile of the mobile antenna being plausible, classifying the mobile antenna to a first antenna group,
in case of the movement profile of the mobile antenna not being plausible, classifying the mobile antenna to a second antenna group.
2 . The method of claim 1 , further comprising, in case of classifying the mobile antenna to a second antenna group, to perform a further check for analyzing statistical data regarding measurements taken from the transmissions sent by the classified mobile antenna.
3 . The method of claim 2 , wherein the statistical data analyzed at the further check comprise the CIRCC of the signal from which the time of arrival has been acquired.
4 . The method of claim 2 , wherein the method comprises, if the further check reveals that the current neural network, or the function approximator, is incorrect, then using this conclusion for retraining the neural network, or the function approximator.
5 . The method of claim 1 , wherein the time of arrival measurement and times of arrival of the predetermined transmitted signals are relative times of arrival.
6 . The method of claim 1 , involving a plurality of receiving antennas.
7 . The method of claim 1 , wherein the neural network, or the function approximator, is trained using a method for training the neural network, or the function approximator, for inferring a predetermined time of arrival of a predetermined transmitted signal on the basis of channel-impulse-responses, CIRs, of transmitted signals between a mobile antenna and a fixed antenna, the method comprising:
acquiring a channel impulse response condition characteristic, CIRCC, which is descriptive of channel impulse responses of transmitted signals associated with mobile antenna positions within a reach of the fixed antenna;
generating, by simulation, a training set of simulated CIRs which are associated with different times of arrival in one or more simulated scenes, and which fit to the CIRCC; and
training the neural network, or the function approximator, using the simulated CIRs and the different associated times of arrivals so as to acquire a parametrization of the neural network, or the function approximator, associated with the CIRCC.
8 . The method of claim 1 , wherein the neural network infers time of arrivals by performing a grid search on an input CIR by using weights acquired during the neural network training.
9 . The method of claim 1 , wherein the neural network performs a dropout by randomly discarding inferred values.
10 . The method of claim 1 , wherein the neural network performs a linear regression.
11 . The method of claim 1 , the neural network using a 1-dimensional kernel.
12 . The method of claim 1 , wherein the CIRCC comprises, for at least one spatial position, a large-scale parameter, LSP.
13 . The method of claim 1 , wherein the above-mentioned CIRCC comprises, for at least one spatial position, a Rician K-factor, KF.
14 . The method of claim 1 , wherein the above-mentioned CIRCC comprises, for at least one spatial position, a delay spread, DS.
15 . The method of claim 1 , wherein the above-mentioned CIRCC comprises a value roughly indicating how far multipath components are delayed.
16 . The method of claim 1 , wherein the above-mentioned CIRCC comprises, for at least one spatial position, a parameter expressed as a probability density function, PDF.
17 . The method of claim 1 , wherein the neural network is a supervised network.
18 . The method of claim 1 , wherein the neural network is an unsupervised network.
19 . The method of claim 1 , further comprising selecting, using a present CIRCC, out of a plurality of parametrizations associated with CIRCCs, a predetermined parametrization fitting to the present CIRCC, so that the neural network or the function approximator, so that the neural network is parametrized using the predetermined parametrization.
20 . The method of claim 19 , wherein selecting is performed during inference operations.
21 . The method of claim 1 , further comprising:
checking whether the CIRCC fits to a predetermined present CIRCC with which a predetermined parametrization of the neural network, or the function approximator, is parametrized, and
as long as the check reveals that the CIRCC fits to the predetermined present CIRCC, inferring, using the neural network or the function approximator, parametrized using the predetermined parametrization, the times of arrival, and
as soon as the present CIRCC no longer fits to the predetermined present CIRCC, initiate a compensation step in which the neural network, or the function approximator is retrained and/or reparametrized according to another parametrization.
22 . The method of claim 1 , further comprising checking whether the CIRCC is similar to a previously determined CIRCC, and,
in case it is determined that the CIRCC fits to the previously determined CIRCC, the neural network or the function approximator is maintained,
otherwise, in case it is determined that the CIRCC does not fit to the previously determined CIRCC, the neural network or the function approximator is maintained, the neural network or the function approximator is changed or retrained.
23 . The method of claim 1 , further comprising, in case of the mobile antenna being classified in the second group, retraining of the neural network or the function approximator.
24 . The method of claim 1 , wherein the method comprises analysing the CIRCC and, in case it is determined that a new neural network or the function approximator, or parametrization of the neural network or the function approximator, is needed, training a new neural network or function approximator.
25 . A non-transitory digital storage medium having stored thereon a computer program for performing a method for classifying a mobile antenna, exchanging predetermined transmitted signals with a fixed antenna, on the basis of a predetermined channel impulse response, CIR, from which a time-of-arrival measurement is to be performed, the method comprising:
using a neural network or a function approximator, which comprises a multidimensional polynomial function that learns a fitting function, for inferring from the CIRs of the predetermined transmitted signals, times of arrival of the predetermined transmitted signals exchanged between the mobile antenna and the fixed antenna,
determining, based on the times of arrival, or channel impulse response condition characteristic, CIRCC, which is descriptive of channel impulse responses of the transmitted signals exchanged between the mobile antenna and the fixed antenna, a movement profile of the mobile antenna,
analyzing the movement profile of the mobile antenna in terms of plausibility,
in case of the movement profile of the mobile antenna being plausible, classifying the mobile antenna to a first antenna group,
in case of the movement profile of the mobile antenna not being plausible, classifying the mobile antenna to a second antenna group,
when the computer program is run by a computer.
26 . A user equipment, UE, for inferring a predetermined time of arrival of a predetermined transmitted signal on the basis of channel-impulse-responses, CIRs, of transmitted signals between the UE and a fixed antenna, the UE comprising a mobile antenna and being configured for:
intermittently acquiring a present channel impulse response condition characteristic, CIRCC, which is descriptive of CIRs of transmitted signals associated with UE positions within a reach of the fixed antenna;
checking whether the present CIRCC fits to a predetermined present CIRCC with which a predetermined parametrization of a neural network, or a function approximator which comprises a multidimensional polynomial function that learns a fitting function, is associated, and
as long as the check reveals that the present CIRCC fits to the predetermined present CIRCC, inferring, using the neural network or the function approximator, parametrized using the predetermined parametrization, the predetermined time of arrival, and
as soon as the present CIRCC no longer fits to the predetermined present CIRCC, cease the use of the neural network, or the function approximator, parametrized using the predetermined parametrization, for inferring the predetermined time of arrival, and/or initiate a compensation step in which the neural network, or the function approximator is analyzed and/or retrained.
27 . The UE of claim 26 , further comprising a sensor providing information suggesting a change of environment, the information suggesting the change of environment triggering a new session of intermittently acquiring a present CIRCC.
28 . The UE of claim 27 , wherein the sensor is an inertial sensor.
29 . The UE of claim 27 , wherein the sensor is global positioning navigation sensor.
30 . The UE of claim 26 , wherein the present CIRCC comprises, for at least one spatial position, a large-scale parameter, LSP.
31 . The UE of claim 26 , wherein the present CIRCC comprises, for at least one spatial position, a Rician K-factor, KF.
32 . The UE of claim 26 , wherein the present CIRCC comprises, for at least one spatial position, a delay spread, DS.
33 . The UE of claim 26 , wherein the present CIRCC comprises a value roughly indicating how far multipath components are delayed.
34 . The UE of claim 26 , wherein the present CIRCC comprises, for at least one spatial position, a parameter expressed as a probability density function, PDF.
35 . A system for classifying a mobile antenna, exchanging predetermined transmitted signals with a fixed antenna, on the basis of a predetermined channel-impulse-response, CIR, from which a time-of-arrival measurement is to be performed, the system comprising the fixed antenna and being configured for:
using a neural network or a function approximator, which comprises a multidimensional polynomial function that learns a fitting function, for inferring, from the CIRs of the predetermined transmitted signals, times of arrival of the predetermined transmitted signals exchanged between the mobile antenna and the fixed antenna,
determining, based on the time of arrival measurements, or channel impulse response condition characteristic, CIRCC, which is descriptive of channel impulse responses of the transmitted signals exchanged between the mobile antenna and the fixed antenna, a movement profile of the mobile antenna,
analyzing the movement profile of the mobile antenna in terms of plausibility,
in case of the movement profile of the mobile antenna being plausible, classifying the mobile antenna to a first antenna group,
in case of the movement profile of the mobile antenna not being plausible, classifying the mobile antenna to a second antenna group.