	US 12,287,419 B1
	UWB NLOS signal recognition method based on the first path of CIR
Qianqian Cai, Guangzhou (CN); Haoqiang Ou, Guangzhou (CN); Junwei Li, Guangzhou (CN); Xiaopei Chen, Guangzhou (CN); and Wentao Zhong, Guangzhou (CN)
Assigned to Guangdong University of Technology, Guangzhou (CN)
Filed by Guangdong University of Technology, Guangzhou (CN)
Filed on Jun. 25, 2024, as Appl. No. 18/754,063.
Claims priority of application No. 202311291191.7 (CN), filed on Oct. 8, 2023.
Int. Cl. G01S 5/02 (2010.01); G01S 5/14 (2006.01); G06N 3/08 (2023.01); G06N 20/10 (2019.01)

CPC G01S 5/0218 (2020.05) [G01S 5/14 (2013.01); G06N 3/08 (2013.01); G06N 20/10 (2019.01)]

11 Claims

1. An UWB NLOS signal recognition method based on a first path of CIR, including:

constructing UWB (Ultra-Wideband) ranging system, wherein the UWB ranging system includes the tags as well as a base station, in an LOS (Line-of-Sight) case there is no obstacle between the base station and the tag, and in the NLOS (Non-Line-of-Sight) case there is an obstacle between the base station and the tag;

controlling the base station to communicate with the tags, processing raw CIR (Channel Impulse Response) data obtained from each communication, and then constructing and labeling data samples, after that, utilizing these data samples to respectively construct raw CIR datasets for both LOS and NLOS scenarios;

performing peak filtering on a CIR waveform of each data sample in the raw CIR datasets to identify a first path peak of the data sample;

extracting valid data in each data sample as new data samples, and utilizing one-hot encoding to generate data labels for new data samples based on the first path peak;

utilizing new data samples and their corresponding data labels from the raw CIR datasets in both scenarios to construct a training dataset, establishing a machine learning model and training the machine learning model by using the training dataset;

saving the machine learning model, and recognizing unknown CIR signals by using the machine learning model;

wherein a step of performing an extreme point filtering process on the CIR waveforms of the data samples in each CIR raw data set includes:

for LOS and NLOS conditions, channel impulse response c(t) in each data sample from the CIR datasets can be constructed as follows:

wherein K is a total number of the multipath, t is a time parameter, τ_kis a time delay of a kth path, a_kis a amplitude of the kth path, δ(⋅) is an impulse response,

using a following formula to perform the extreme point filtering process,

a gradient ∇_kbetween the kth and k−1th path component amplitude is expressed as

where k=2, 3, . . . , N and N is a total number of sample points,

a_kis regarded as a local peak if ∇_kand ∇_k-1satisfy a following formula

∇_k×∇_k-1<0 (3)

retaining the local peaks ax to complete the peak filtering process of the CIR waveform, with a filtered CIR denoted as c*(t)

wherein a step of identifying the first path peak of the data sample includes:

for c*(t), traverse each sample point ax, where i=1, 2, . . . , M, and M is a number of sample points in c*(t), formula (4) is used to calculate a mean amplitude of the sample points {a₁*, a₂*, . . . , a_i*}, and determine whether the sample point a_i* meets a condition of formula (5), identifying the sample point a_i* as a first path peak point,

where a_j* is a jth sample point, μ_i* is a mean amplitude of {a₁*, a₂*, . . . , a_i*},

if a_i* is the maximum value among {a₁*, a₂*, . . . , a_i*}, and is greater than 4 times a mean amplitude, then a_i* is considered an outlier of {a₁*, a₂*, . . . , a_i*}, the first path peak point in that data sample

extracting the valid data from each data sample as a new data sample based on the first path peak point, including:

for each data sample in the raw CIR dataset, M sample points (starting from the identified first path peak point) are extracted from the channel impulse response c(t) of the N sample points in that data sample to form the new data sample c′(t).