METHOD AND DEVICE FOR AI-BASED USER DEVICE SPEED ESTIMATION USING UPLINK SRS MEASUREMENTS

Description

Technical Field

[0001] This disclosure relates generally to methods and apparatus for an artificial intelligence (AI)-assisted user equipment (UE) speed estimation based on uplink sounding reference signal (SRS) channel measurement inputs. In particular, methods and apparatus extract the features of either a set of power spectrum density measurements or a set of pre-processed frequency domain real and imaginary portions of UL SRS measurements and feed the features to an AI classifier for UE speed estimation.Background Art

[0002] To meet the demand for wireless data traffic having increased since deployment of 4G communication systems, efforts have been made to develop an improved 5G or pre-5G communication system. Therefore, the 5G or pre-5G communication system is also called a 'Beyond 4G Network' or a 'Post LTE System'.

[0003] The 5G communication system is considered to be implemented in higher frequency (mmWave) bands, e.g., 28GHz or 60GHz bands, so as to accomplish higher data rates. To decrease propagation loss of the radio waves and increase the transmission distance, the beamforming, massive multiple-input multiple-output (MIMO), Full Dimensional MIMO (FD-MIMO), array antenna, an analog beam forming, large scale antenna techniques are discussed in 5G communication systems.

[0004] In addition, in 5G communication systems, development for system network improvement is under way based on advanced small cells, cloud Radio Access Networks (RANs), ultra-dense networks, device-to-device (D2D) communication, wireless backhaul, moving network, cooperative communication, Coordinated Multi-Points (CoMP), reception-end interference cancellation and the like.

[0005] In the 5G system, Hybrid FSK and QAM Modulation (FQAM) and sliding window superposition coding (SWSC) as an advanced coding modulation (ACM), and filter bank multi carrier(FBMC), non-orthogonal multiple access(NOMA), and sparse code multiple access (SCMA) as an advanced access technology have been developed.

[0006] Knowledge of either the terminal speed or the category in which the terminal speed falls is vital for optimizing various radio resource management functions such as handover, mobility load balancing, and transmission scheduling at the network. A simple example could be that for transmission scheduling, the network may use frequency diverse scheduling to high speed users in order to maximize robustness of reception. For low speed users on the other hand, it may be preferable for the network to provide localized frequency scheduling so as to maximize the system spectral efficiency. As part of mobility robustness optimization, the network could adapt its handover parameters differently based on the mobility of the terminal. For mmWave systems, estimating the terminal speed is potentially useful for beam prediction (and channel prediction techniques) considering the correlation in channel properties across the various beams used by the terminals. This can significantly reduce the beam scanning and beam sweeping time typically involved in mmWave systems.

[0007] A publication of US 2004 / 097197 A1 discloses that received signals in a MIMO, SIMO or MISO mobile communication system are estimated and subsequently used to estimate speed of a mobile station.

[0008] A prior art of US 2016 / 080095 A1 recites a method for estimating speed of a user equipment connected to a base station of a wireless network, the method comprising the following steps: performing signal strength measurements of a radio signal transmitted between the user equipment and the base station; performing a spectral analysis of the signal strength measurements; determining the frequency of a local maximum in the power spectrum of the signal strength measurements; estimating, from previously established reference data, the speed of the user equipment that corresponds to the determined frequency, the reference data associating a given user equipment speed with a certain determined frequencyDisclosure of Invention Solution to Problem

[0009] This disclosure provides an artificial intelligence (AI) assisted approach to categorize or determine the speed of a user equipment (UE) based on uplink sounding reference signal (SRS) channel measurement inputs.

[0010] In a first embodiment, an apparatus for performing a wireless communication is provided as defined in the appended claims.

[0011] In a second embodiment, a method for estimating a speed of a mobile client device is provided as per the appended claims.

[0012] Other technical features may be readily apparent to one skilled in the art from the following figures, descriptions, and claims.

[0013] Before undertaking the DETAILED DESCRIPTION below, it may be advantageous to set forth definitions of certain words and phrases used throughout this patent document. The term "couple" and its derivatives refer to any direct or indirect communication between two or more elements, whether or not those elements are in physical contact with one another. The terms "transmit," "receive," and "communicate," as well as derivatives thereof, encompass both direct and indirect communication. The terms "include" and "comprise," as well as derivatives thereof, mean inclusion without limitation. The term "or" is inclusive, meaning and / or. The phrase "associated with," as well as derivatives thereof, means to include, be included within, interconnect with, contain, be contained within, connect to or with, couple to or with, be communicable with, cooperate with, interleave, juxtapose, be proximate to, be bound to or with, have, have a property of, have a relationship to or with, or the like. The term "controller" means any device, system or part thereof that controls at least one operation. Such a controller may be implemented in hardware or a combination of hardware and software and / or firmware. The functionality associated with any particular controller may be centralized or distributed, whether locally or remotely. The phrase "at least one of," when used with a list of items, means that different combinations of one or more of the listed items may be used, and only one item in the list may be needed. For example, "at least one of: A, B, or C" may be understood as "only A", "only B", "only C", "both A and B", "both A and C", "both B and C", "all of A, B, and C", or variations thereof.

[0014] Moreover, various functions described below can be implemented or supported by one or more computer programs, each of which is formed from computer readable program code and embodied in a computer readable medium. The terms "application" and "program" refer to one or more computer programs, software components, sets of instructions, procedures, functions, objects, classes, instances, related data, or a portion thereof adapted for implementation in a suitable computer readable program code. The phrase "computer readable program code" includes any type of computer code, including source code, object code, and executable code. The phrase "computer readable medium" includes any type of medium capable of being accessed by a computer, such as read only memory (ROM), random access memory (RAM), a hard disk drive, a compact disc (CD), a digital video disc (DVD), or any other type of memory. A "non-transitory" computer readable medium excludes wired, wireless, optical, or other communication links that transport transitory electrical or other signals. A non-transitory computer readable medium includes media where data can be permanently stored and media where data can be stored and later overwritten, such as a rewritable optical disc or an erasable memory device.

[0015] Definitions for other certain words and phrases are provided throughout this patent document. Those of ordinary skill in the art should understand that in many if not most instances, such definitions apply to prior as well as future uses of such defined words and phrases.Brief Description of Drawings

[0016] For a more complete understanding of this disclosure and its advantages, reference is now made to the following description, taken in conjunction with the accompanying drawings, in which: FIG. 1 illustrates an example computing system according to various embodiments of the disclosure; FIGS. 2A and 2B and FIGS. 3A and 3B illustrate example devices in a computing system according to various embodiments of the disclosure; FIG. 4 illustrates an exemplary diagram 400 for explaining Doppler Shift in a mobile radio channel according to an embodiment of the disclosure; FIG. 5 illustrates a diagram of a LTE sounding reference signal (SRS) transmitted from a moving terminal according to an embodiment of the disclosure; FIGS. 6A, 6B and 6C illustrate uplink (UL) SRS transmissions according to embodiments of the disclosure; FIG. 7 illustrates a flowchart for training an artificial intelligence (AI) based user equipment (UE) speed classifier for estimating a speed class for a UE according to an embodiment; FIG. 8 illustrates a flowchart for estimating a speed class from power spectral density measurements according to an embodiment of the disclosure; FIG. 9 illustrates an example graph of empirically derived power spectral density (PSD) according to an embodiment of the disclosure; FIG. 10A illustrates an example of wideband SRS transmissions, and FIG. 10B illustrates an example of frequency hopped SRS (FH SRS) transmissions, according to embodiments of the disclosure; FIGS. 11A and 11B illsutrate two methods for correlating across channel impulse responses (CIRs) for the case of FH SRS according to an embodiment of the disclosure; FIG. 11C illustrates an exemplary flowchart for pre-processing frequency domain channel measurements for a deep neural network architecture according to embodiments of the disclosure; FIG. 12 illustrates an example of a sliding-window based mechanism to track a UE speed according to an embodiment of the disclosure; FIG. 13A illustrates an example of a convolutional neural network (CNN) classifier for estimating a UE speed based on Doppler power spectrum measurements, and FIG. 13B illustrates an example of a CNN speed classifier based on raw frequency domain UL SRS measurements, according to embodiments of the disclosure; and FIG. 13C illustrates an exemplary flowchart 1320 for a deep neural network architecture with different sets of neural network layer weights to be applied to each layer of the multiple CNN layers, according to embodiments of the disclosure. Best Mode for Carrying out the Invention

[0017] FIGS. 1 through 13C, discussed below, and the various embodiments used to describe the principles of this disclosure in this patent document are by way of illustration only and should not be construed in any way to limit the scope of the disclosure. Those skilled in the art will understand that the principles of this disclosure may be implemented in any suitably arranged wireless communication system.

[0018] FIG.1 illustrates an example wireless network 100 according to this disclosure. The embodiment of the wireless network 100 shown in FIG.1 is for illustration only. Other embodiments of the wireless network 100 could be used without departing from the scope of this disclosure.

[0019] The wireless network 100 includes an eNodeB (eNB) 101, an eNB 102, and an eNB 103. The eNB 101 communicates with the eNB 102 and the eNB 103. The eNB 101 also communicates with at least one Internet Protocol (IP) network 130, such as the Internet, a proprietary IP network, or other data network.

[0020] Depending on the network type, the term "base station" or "BS" can refer to any component (or collection of components) configured to provide wireless access to a network, such as transmit point (TP), transmit-receive point (TRP), an enhanced base station (eNodeB or eNB or gNB), a macro cell, a femtocell, a wireless fidelity (Wi-Fi) access point (AP) , or other wirelessly enabled devices. Base stations may provide wireless access in accordance with one or more wireless communication protocols, e.g., 5G 3GPP new radio (NR) interface / access, long term evolution (LTE), LTE advanced (LTE-A), high speed packet access (HSPA), Wi-Fi 802.11a / b / g / n / ac, etc. For the sake of convenience, the terms "BS" and "TRP" are used interchangeably in this patent document to refer to network infrastructure components that provide wireless access to remote terminals. Also, depending on the network type, the term "user equipment" or "UE" can refer to any component such as "mobile station," "subscriber station," "remote terminal," "wireless terminal," "receive point," or "user device." For the sake of convenience, the terms "user equipment" and "UE" are used in this patent document to refer to remote wireless equipment that wirelessly accesses a BS, whether the UE is a mobile device (such as a mobile telephone or smartphone) or is normally considered a stationary device (such as a desktop computer or vending machine).

[0021] Some embodiments of the disclosure are described assuming cellular downlink (DL) communications. However, the same / similar principles and related signaling methods and configurations can also be used for a cellular uplink (UL) and a side link (SL)

[0022] Mobile speed classification methods described herein rely on obtaining the speed class through estimating the Doppler spread of the underlying mobile radio channel.

[0023] The eNB 102 provides wireless broadband access to the network 130 for a first plurality of user equipments (UEs) within a coverage area 120 of the eNB 102. The first plurality of UEs includes a UE 111, which may be located in a small business (SB); a UE 112, which may be located in an enterprise (E); a UE 113, which may be located in a Wi-Fi hotspot (HS); a UE 114, which may be located in a first residence (R); a UE 115, which may be located in a second residence (R); and a UE 116, which may be a mobile device (M) like a cell phone, a wireless laptop, a wireless PDA, or the like. The eNB 103 provides wireless broadband access to the network 130 for a second plurality of UEs within a coverage area 125 of the eNB 103. The second plurality of UEs includes the UE 115 and the UE 116. In some embodiments, one or more of the eNBs 101-103 may communicate with each other and with the UEs 111-116 using 5G, long-term evolution (LTE), LTE-A, WiMAX, or other advanced wireless communication techniques.

[0024] Dotted lines show the approximate extents of the coverage areas 120 and 125, which are shown as approximately circular for the purposes of illustration and explanation only. It should be clearly understood that the coverage areas associated with eNBs, such as the coverage areas 120 and 125, may have other shapes, including irregular shapes, depending upon the configuration of the eNBs and variations in the radio environment associated with natural and man-made obstructions.

[0025] As described in more detail below, one or more of BS 101, BS 102 and BS 103 include 2D antenna arrays as described in embodiments of the disclosure. In some embodiments, one or more of BS 101, BS 102 and BS 103 support the codebook design and structure for systems having 2D antenna arrays.

[0026] Although FIG.1 illustrates one example of a wireless network 100, various changes may be made to FIG.1. For example, the wireless network 100 could include any number of eNBs and any number of UEs in any suitable arrangement. Also, the eNB 101 could communicate directly with any number of UEs and provide those UEs with wireless broadband access to the network 130. Similarly, each eNB 102-103 could communicate directly with the network 130 and provide UEs with direct wireless broadband access to the network 130. Further, the eNB 101, 102, and / or 103 could provide access to other or additional external networks, such as external telephone networks or other types of data networks.

[0027] FIGS. 2A and 2B illustrate example wireless transmit and receive paths according to this disclosure. In the following description, a transmit path 200 may be described as being implemented in an eNB (such as eNB 102), while a receive path 250 may be described as being implemented in a UE (such as UE 116). However, it will be understood that the receive path 250 could be implemented in an eNB and that the transmit path 200 could be implemented in a UE. In some embodiments, the receive path 250 is configured to support the codebook design and structure for systems having 2D antenna arrays as described in embodiments of the disclosure.

[0028] The transmit path 200 includes a channel coding and modulation block 205, a serial-to-parallel (S-to-P) block 210, a size N inverse fast fourier transform (IFFT) block 215, a parallel-to-serial (P-to-S) block 220, an add cyclic prefix block 225, and an up-converter (UC) 230. The receive path 250 includes a down-converter (DC) 255, a remove cyclic prefix block 260, a serial-to-parallel (S-to-P) block 265, a size N fast fourier transform (FFT) block 270, a parallel-to-serial (P-to-S) block 275, and a channel decoding and demodulation block 280.

[0029] In the transmit path 200, the channel coding and modulation block 205 receives a set of information bits, applies coding (such as a low-density parity check (LDPC) coding), and modulates the input bits (such as with quadrature phase shift keying (QPSK) or quadrature amplitude modulation (QAM)) to generate a sequence of frequency-domain modulation symbols. The serial-to-parallel block 210 converts (such as de-multiplexes) the serial modulated symbols to parallel data in order to generate N parallel symbol streams, where N is the IFFT / FFT size used in the eNB 102 and the UE 116. The size N IFFT block 215 performs an IFFT operation on the N parallel symbol streams to generate time-domain output signals. The parallel-to-serial block 220 converts (such as multiplexes) the parallel time-domain output symbols from the size N IFFT block 215 in order to generate a serial time-domain signal. The add cyclic prefix block 225 inserts a cyclic prefix to the time-domain signal. The up-converter 230 modulates (such as up-converts) the output of the add cyclic prefix block 225 to an RF frequency for transmission via a wireless channel. The signal may also be filtered at baseband before conversion to the RF frequency.

[0030] A transmitted RF signal from the eNB 102 arrives at the UE 116 after passing through the wireless channel, and reverse operations to those at the eNB 102 are performed at the UE 116. The down-converter 255 down-converts the received signal to a baseband frequency, and the remove cyclic prefix block 260 removes the cyclic prefix to generate a serial time-domain baseband signal. The serial-to-parallel block 265 converts the time-domain baseband signal to parallel time domain signals. The size N FFT block 270 performs an FFT algorithm to generate N parallel frequency-domain signals. The parallel-to-serial block 275 converts the parallel frequency-domain signals to a sequence of modulated data symbols. The channel decoding and demodulation block 280 demodulates and decodes the modulated symbols to recover the original input data stream.

[0031] Each of the eNBs 101-103 may implement a transmit path 200 that is analogous to transmitting in the downlink to UEs 111-116 and may implement a receive path 250 that is analogous to receiving in the uplink from UEs 111-116. Similarly, each of UEs 111-116 may implement a transmit path 200 for transmitting in the uplink to eNBs 101-103 and may implement a receive path 250 for receiving in the downlink from eNBs 101-103.

[0032] Each of the components in FIGS. 2A and 2B can be implemented using only hardware or using a combination of hardware and software / firmware. As a particular example, at least some of the components in FIGS. 2A and 2B may be implemented in software, while other components may be implemented by configurable hardware or a mixture of software and configurable hardware. For instance, the FFT block 270 and the IFFT block 215 may be implemented as configurable software algorithms, where the value of size N may be modified according to the implementation.

[0033] Furthermore, although described as using FFT and IFFT, this is by way of illustration only and should not be construed to limit the scope of this disclosure. Other types of transforms, such as discrete fourier transform (DFT) and inverse discrete fourier transform (IDFT) functions, could be used. It will be appreciated that the value of the variable N may be any integer number (such as 1, 2, 3, 4, or the like) for DFT and IDFT functions, while the value of the variable N may be any integer number that is a power of two (such as 1, 2, 4, 8, 16, or the like) for FFT and IFFT functions.

[0034] Although FIGS. 2A and 2B illustrate examples of wireless transmit and receive paths, various changes may be made to FIGS. 2A and 2B. For example, various components in FIGS. 2A and 2B could be combined, further subdivided, or omitted and additional components could be added according to particular needs. Also, FIGS. 2A and 2B are meant to illustrate examples of the types of transmit and receive paths that could be used in a wireless network. Any other suitable architectures could be used to support wireless communications in a wireless network.

[0035] FIG. 3A illustrates an example UE 116 according to this disclosure. The embodiment of the UE 116 illustrated in FIG. 3A is for illustration only, and the UEs 111-115 of FIG. 1 could have the same or similar configuration. However, UEs come in a wide variety of configurations, and FIG. 3A does not limit the scope of this disclosure to any particular implementation of a UE.

[0036] The UE 116 includes an antenna 305, a radio frequency (RF) transceiver 310, transmit (TX) processing circuitry 315, a microphone 320, and receive (RX) processing circuitry 325. The UE 116 also includes a speaker 330, a main processor 340, an input / output (I / O) interface (IF) 345, a keypad 350, a display 355, and a memory 360. The memory 360 includes a basic operating system (OS) program 361 and one or more applications 362.

[0037] The RF transceiver 310 receives, from the antenna 305, an incoming RF signal transmitted by an eNB of the network 100. The RF transceiver 310 down-converts the incoming RF signal to generate an intermediate frequency (IF) or baseband signal. The IF or baseband signal is sent to the RX processing circuitry 325, which generates a processed baseband signal by filtering, decoding, and / or digitizing the baseband or IF signal. The RX processing circuitry 325 transmits the processed baseband signal to the speaker 330 (such as for voice data) or to the main processor 340 for further processing (such as for web browsing data).

[0038] The TX processing circuitry 315 receives analog or digital voice data from the microphone 320 or other outgoing baseband data (such as web data, e-mail, or interactive video game data) from the main processor 340. The TX processing circuitry 315 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate a processed baseband or IF signal. The RF transceiver 310 receives the outgoing processed baseband or IF signal from the TX processing circuitry 315 and up-converts the baseband or IF signal to an RF signal that is transmitted via the antenna 305.

[0039] The main processor 340 can include one or more processors or other processing devices and execute the basic OS program 361 stored in the memory 360 in order to control the overall operation of the UE 116. For example, the main processor 340 could control the reception of forward channel signals and the transmission of reverse channel signals by the RF transceiver 310, the RX processing circuitry 325, and the TX processing circuitry 315 in accordance with well-known principles. In some embodiments, the main processor 340 includes at least one microprocessor or microcontroller.

[0040] The main processor 340 is also capable of executing other processes and programs resident in the memory 360, such as operations for channel quality measurement and reporting for systems having 2D antenna arrays as described in embodiments of the disclosure. The main processor 340 can move data into or out of the memory 360 as required by an executing process. In some embodiments, the main processor 340 is configured to execute the applications 362 based on the OS program 361 or in response to signals received from eNBs or an operator. The main processor 340 is also coupled to the I / O interface 345, which provides the UE 116 with the ability to connect to other devices such as laptop computers and handheld computers. The I / O interface 345 is the communication path between these accessories and the main controller 340.

[0041] The main processor 340 is also coupled to the keypad 350 and the display unit 355. The operator of the UE 116 can use the keypad 350 to enter data into the UE 116. The display 355 may be a liquid crystal display or other display capable of rendering text and / or at least limited graphics, such as from web sites.

[0042] The memory 360 is coupled to the main processor 340. Part of the memory 360 could include a random access memory (RAM), and another part of the memory 360 could include a Flash memory or other read-only memory (ROM). Although FIG. 3A illustrates one example of UE 116, various changes may be made to FIG. 3A. For example, various components in FIG. 3A could be combined, further subdivided, or omitted and additional components could be added according to particular needs. As a particular example, the main processor 340 could be divided into multiple processors, such as one or more central processing units (CPUs) and one or more graphics processing units (GPUs). Also, while FIG. 3A illustrates the UE 116 configured as a mobile telephone or smartphone, UEs could be configured to operate as other types of mobile or stationary devices.

[0043] FIG. 3B illustrates an example eNB 102 according to this disclosure. The embodiment of the eNB 102 shown in FIG. 3B is for illustration only, and other eNBs of FIG. 1 could have the same or similar configuration. However, eNBs come in a wide variety of configurations, and FIG. 3B does not limit the scope of this disclosure to any particular implementation of an eNB. It is noted that eNB 101 and eNB 103 can include the same or similar structure as eNB 102.

[0044] As shown in FIG. 3B, the eNB 102 includes multiple antennas 370a-370n, multiple RF transceivers 372a-372n, transmit (TX) processing circuitry 374, and receive (RX) processing circuitry 376. In certain embodiments, one or more of the multiple antennas 370a-370n include 2D antenna arrays. The eNB 102 also includes a controller / processor 378, a memory 380, and a backhaul or network interface 382.

[0045] The RF transceivers 372a-372n receive, from the antennas 370a-370n, incoming RF signals, such as signals transmitted by UEs or other eNBs. The RF transceivers 372a-372n down-convert the incoming RF signals to generate IF or baseband signals. The IF or baseband signals are sent to the RX processing circuitry 376, which generates processed baseband signals by filtering, decoding, and / or digitizing the baseband or IF signals. The RX processing circuitry 376 transmits the processed baseband signals to the controller / processor 378 for further processing.

[0046] The TX processing circuitry 374 receives analog or digital data (such as voice data, web data, e-mail, or interactive video game data) from the controller / processor 378. The TX processing circuitry 374 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate processed baseband or IF signals. The RF transceivers 372a-372n receive the outgoing processed baseband or IF signals from the TX processing circuitry 374 and up-converts the baseband or IF signals to RF signals that are transmitted via the antennas 370a-370n.

[0047] The controller / processor 378 can include one or more processors or other processing devices that control the overall operation of the eNB 102. For example, the controller / processor 378 could control the reception of forward channel signals and the transmission of reverse channel signals by the RF transceivers 372a-372n, the RX processing circuitry 376, and the TX processing circuitry 374 in accordance with well-known principles. The controller / processor 378 could support additional functions as well, such as more advanced wireless communication functions. For instance, the controller / processor 378 can perform the blind interference sensing (BIS) process, such as performed by a BIS algorithm, and decodes the received signal subtracted by the interfering signals. Any of a wide variety of other functions could be supported in the eNB 102 by the controller / processor 378. In some embodiments, the controller / processor 378 includes at least one microprocessor or microcontroller.

[0048] The controller / processor 378 is also capable of executing programs and other processes resident in the memory 380, such as a basic OS. The controller / processor 378 is also capable of supporting channel quality measurement and reporting for systems having 2D antenna arrays as described in embodiments of the disclosure. In some embodiments, the controller / processor 378 supports communications between entities, such as web RTC. The controller / processor 378 can move data into or out of the memory 380 as required by an executing process. The memory 380 stores various artificial intelligence (AI) algorithms for estimating a UE speed and a training dataset to train the various AI algorithms.

[0049] The controller / processor 378 is also coupled to the backhaul or network interface 382. The backhaul or network interface 382 allows the eNB 102 to communicate with other devices or systems over a backhaul connection or over a network. The interface 382 could support communications over any suitable wired or wireless connection(s). For example, when the eNB 102 is implemented as part of a cellular communication system (such as one supporting 5G, LTE, or LTE-A), the interface 382 could allow the eNB 102 to communicate with other eNBs over a wired or wireless backhaul connection. When the eNB 102 is implemented as an access point, the interface 382 could allow the eNB 102 to communicate over a wired or wireless local area network or over a wired or wireless connection to a larger network (such as the Internet). The interface 382 includes any suitable structure supporting communications over a wired or wireless connection, such as an Ethernet or RF transceiver.

[0050] As described in more detail below, the transmit and receive paths of the eNB 102 (implemented using the RF transceivers 372a-372n, TX processing circuitry 374, and / or RX processing circuitry 376) support communication with aggregation of FDD cells and TDD cells.

[0051] Although FIG. 3B illustrates one example of an eNB 102, various changes may be made to FIG. 3B. For example, the eNB 102 could include any number of each component shown in FIG. 3. As a particular example, an access point could include a number of interfaces 382, and the controller / processor 378 could support routing functions to route data between different network addresses. As another particular example, while shown as including a single instance of TX processing circuitry 374 and a single instance of RX processing circuitry 376, the eNB 102 could include multiple instances of each (such as one per RF transceiver).

[0052] According to various embodiments, an apparatus of a base station comprises a communication interface configured to receive uplink (UL) sounding reference signals (SRSs) from a terminal, and at least one processor configured to obtain a number of UL SRS measurements from UL SRSs received from the terminal, the number of UL SRS measurements exceeding a threshold value, extract features for estimating a mobility of the terminal from the UL SRS measurements, and determine a category of the terminal based on the extracted features.

[0053] According to various embodiments, the category is determined by using a machine learning (ML) classifier, and the category includes at least one of: an estimate of either a travel speed, or a range of travel speed of the terminal, an estimate of either a travel velocity or a range of travel velocity of the terminal, an estimate of either a doppler frequency or a range of doppler frequency of a dominant radio frequency (RF) propagation path, or an estimate of either a doppler frequency per path or a range of doppler frequency per path on a set of RF propagation paths.

[0054] According to various embodiments, the ML classifier is a deep neural network including at least one of: multiple convolutional neural network (CNN) layers, multiple fully connected (FC) neural network layers, or multiple recurrent neural network layers.

[0055] According to various embodiments, the extracted features include a set of frequency domain doppler power spectral density (PSD) measurements that are estimated by: deriving channel impulse responses (CIRs) on a per transmit and receive antenna pair basis, from one UL SRS measurement or by combining UL SRS measurements, buffering a certain number of CIRs, in a sliding window, wherein each estimated CIR is formed at different SRS reception times, deriving a time-domain auto-correlation function obtained by correlating the buffered CIRs at different SRS reception times, and taking a fast fourier transform (FFT) of the time-domain auto-correlation function. A size of the sliding window is adjustable based on the category to which the terminal belongs to.

[0056] According to various embodiments, if, during each SRS transmission opportunity, the terminal transmits the UL SRSs across an entire bandwidth, the time-domain auto-correlation function is derived by measuring auto-correlation functions at different lags of the buffered CIRs obtained from time-consecutive UL SRS measurements.

[0057] According to various embodiments, if, during each SRS transmission opportunity, the terminal transmits SRSs across a portion on an entire bandwidth in a frequency hopping manner: the time-domain auto-correlation function is obtained by correlating the buffered CIRs derived from uplink SRS measurements on time-adjacent SRS transmissions, or the time domain auto-correlation function is obtained by correlating buffered CIRs derived from uplink SRS measurements extracted from UL SRS hopping transmission occasions occurring on identical frequency sub-bands.

[0058] According to various embodiments, the extracted features are outputs of a linear or non-linear function of real and imaginary portions of channel measurements derived from UL SRS measurements per transmit and receive antenna pair during each UL SRS transmission occasion. The linear or non-linear function comprises a pre-processing process of applying a fusion function of the real and imaginary portions of the channel measurements. The fusion function is configured to output: the real and imaginary portions of the channel measurements in a frequency domain, or a convex combination of weighted real portions and imaginary portions of the channel measurements in one of the frequency domain, a receive (RX) antenna domain, a transmit (TX) antenna domain, a subcarrier domain, or a channel snapshot domain.

[0059] According to various embodiments, each different set of neural network weights corresponds to different estimated signal to noise ratio (SNR) levels. The processor is configured to: estimate an SNR level based on a reference signal received power (RSRP) and a reference signal received quality (RSRQ) reported, or based on channel quality indication (CQI) feedback from the terminal; select a set of neural network layer weights among the set of weights, based on the estimated different SNR level; and load a weighted neural network architecture for determining the category to be used to estimate mobility of the mobile client device, by applying the selected set of neural network weights to each CNN layer and fully connected layer.

[0060] According to various embodiments, if the extracted features are either a set of power spectrum density measurements or a set of pre-processed frequency domain real and imaginary portions of UL SRS measurements. The ML classifier utilizes a deep neural network architecture comprising multiple dimensional CNN and multiple FC layers. The deep neural network architecture comprises a pooling layer placed between a CNN layer and a subsequent CNN layer; a flattening layer placed between a final CNN layer and a first FC layer; and a last FC layer producing an activation output determined using a soft-max output activation wherein a dimension of the activation output equals a number of candidate categories. The deep neural network architecture returns an output that is determined by an index for an argument maximizer of the activation output of the last fully connected layer, and that specifies an estimated category of the mobility of the terminal.

[0061] According to various embodiments, the apparatus further comprises a memory. The memory is configured to store a deep neural network architecture including a number of layers and connectivities between each layer and its subsequent layer for the deep neural network architecture; and different sets of neural network layer weights to be applied to each layer of the multiple CNN layers. The processor is configured to load a set of neural network layer weights based on a certain signal measurement or feedback from the terminal.

[0062] According to various embodiments, the fusion function is at least one of: an inverse tangent function to obtain phases of the real and imaginary portions of the channel measurements; a weighted combination function, w*H re + (1- w)* H im , where w takes values between 0 and 1, and H re and H im respectively indicate each real portion and each imaginary portion of the channel measurements in the frequency domain; or a weighted combination function, w 1 H rx1 + w 2 H rx2 + ..... + w rxN H rxN , where weights w rxi , 1 <= i <= N, are each non-negative, all w rxi ; summing to 1, and H rxi denotes a channel measurement at receive antenna i.

[0063] According to various embodiments, real portions and imaginary portions of channel measurements are concatenated in one of four dimensions to produce a four-dimensional complex tensor including one of: (2N rx )×N tx ×N sub ×M, where real and imaginary portions of channel measurements are concatenated in a receive (RX) antenna domain; N rx ×(2N tx )×N sub ×M, where real and imaginary portions of channel measurements are concatenated in a transmit (TX) antenna domain; N rx ×N tx ×(2N sub )×M, where real and imaginary portions of channel measurements are concatenated in the subcarrier domain; or N rx × N tx ×N sub × (2M) , where real and imaginary portions of channel measurements are concatenated in the channel snapshot domain, where N tx denotes a quantity of TX antennas, N rx denotes a quantity of RX antennas, N sub denotes a quantity of subcarriers used by the UL SRSs, and M denotes a quantity of channel snapshots.

[0064] According to various embodiments, if, during each SRS transmission opportunity, the client device transmits the SRS across an entire bandwidth, channel measurements at a given SRS transmission occasion are derived from corresponding UL SRS measurements.

[0065] According to various embodiments, if, during each SRS transmission opportunity, the client device transmits its SRS across a portion of an entire bandwidth in a frequency hopping manner, channel measurements for the entire bandwidth are derived by concatenation of the channel measurements obtained from UL SRS measurements occurring on consecutive UL SRS hopping transmission occasions.

[0066] FIG. 4 illustrates an exemplary diagram 400 for explaining Doppler Shift in a mobile radio channel according to an embodiment of the disclosure.

[0067] The physical movement of a wireless transmitting (resp. receiving) device relative to the receiver (resp. transmitter) creates the well-known Doppler effect. The Doppler effect manifests itself in a frequency shift, termed Doppler Shift, of the transmitted signal as experienced at the receiver. The Doppler shift f is given by the following expression: f = f d cos θ , f d : = v f c c .

[0068] In the above expression, f d is the maximum Doppler frequency, v denotes the speed of the moving terminal, f c designates the carrier frequency and θ denotes the angle made by the rays relative to a (reference) coordinate system (often denoting the angle between the path direction and the user moving direction). Because of multi-paths, each multipath wave arrives at the receiver from different directions, thus different propagation paths result in different Doppler shifts. Thus, estimating the Doppler shift of a given path (e.g. dominant path) does not explicitly provide the terminal speed information). Considering the azimuthal and elevation direction, the Doppler shift as described above can be generalized via the following expression where v is the velocity vector and r ^ rx , n , m T designates the spherical unit vector with azimuth departure angle ϕ n,m,ZOD and elevation departure angle θ n,m,ZOD , It is evident that different velocity vectors can result in identical Doppler shift (depending on the propagation paths and angles made by the velocity vector with the ray propagation direction). f d , n , m = r ^ rx , n , m T ⋅ v ¯ λ 0 , v ¯ = v . sinθ v cosϕ v sinθ v sinϕ v cosθ v T r ^ rx , n , m T = sinθ n , m , ZOD ϕ n , m , ZOD sinθ n , m , ZOD sinϕ n , m , ZOD cosθ n , m , ZOD

[0069] The aggregation of the received signal from a multitude of propagation paths results in a spectral broadening of the received signal, also known as the Doppler Spread, relative to that sent by the transmitter. The Doppler spread is thus a measure of the increase in bandwidth at the received signal relative to the transmitted signal, caused as a result of the time rate of change of the underlying radio channel.

[0070] Existing speed classification works can be roughly classified into the following four categories. 1) Maximum likelihood methods that rely on periodic channel estimation to infer one or more properties of the channel. 2) Level crossing rate based methods that count the number of times that a process crosses a certain level. 3) Covariance based methods that estimate the second-order statistics of the received signal power. 4) Autocorrelation based methods that estimates the autocorrelation of the estimated channel impulse response (CIR) and infer the Doppler spread / UE speed by looking at the power spectral density.

[0071] Level crossing rate and covariance methods suffer from increased sensitivity with reduced SNR and suffer from higher error when the UE speed is small. Furthermore, threshold-based techniques may work in a certain environment when a threshold is fine-tuned, but the same threshold may fail to work in other settings. Maximum likelihood-based methods typically have high computational complexity and require a priori knowledge of the system SNR / noise variance.

[0072] The disclosure provides a new method relying on an AI based speed classifier. The AI system includes a neural network which processes uplink signal measurements made on an LTE signal such as the sounding reference signal (SRS) and determines the most likely speed class for that user from among N classes number of speed classes.

[0073] FIG. 5 illustrates a diagram 500 of a LTE SRS signal transmitted from a moving terminal according to an embodiment of the disclosure.

[0074] The SRS signal is used to determine the uplink channel quality using which the eNB can decide what uplink resource blocks and modulation and coding scheme should be employed while scheduling that UE. The neural classifier could also reside at the terminal by making use of downlink reference signal measurements (e.g., based on cell specific reference signal (CRS) / channel state information reference symbol (CSI-RS)) wherein the said application can find use in device to device (D2D) systems as well. As seen in performance evaluations under 3GPP channel model, the scheme is able to classify a UE speed with a high degree of accuracy even when the direction of the terminal's motion and the angle of different propagation paths are randomly chosen. The output of the neural network can either be an absolute estimated UE speed or an estimated category in which the UE speed belongs. In another embodiment, the CQI (or modulation and coding scheme (MCS)) reports may also be used as inputs to the classifier in order to identify the UE speed along similar lines as mentioned in this disclosure.

[0075] The scheme described in the disclosure provides an artificial intelligence (AI) assisted speed classification scheme. The AI classifier is a neural network-based classifier. Two different AI classifier architectures have been evaluated. The first architecture is a fully connected neural network, while the second architecture is a convolutional neural network. The disclosure also provides a supervised learning method to train the neural network. The disclosure describes serval input features that are derived from the uplink SRS channel measurements, using which the neural network determines the speed class for that terminal.

[0076] Unlike previous works, the benefits of the embodiments in the disclosure include the ability to simultaneously determine speed categories for multiple users, zero reliance on prior knowledge of system parameters (e.g. noise variance, signal-to-noise ratio (SNR), etc.) a low sensitivity to the level of noise, a reasonable computational complexity, and applicability for both wide-band and frequency hopping SRS measurements. While the approach described in the disclosure assumes that the channel measurements are derived from the SRS measurements, the approach can be extended to estimating a terminal speed based on downlink reference signal measurements as well.

[0077] FIGS. 6A, 6B and 6C illustrate uplink (UL) SRS transmissions according to embodiments of the disclosure. The embodiment shown in FIGS. 6A, 6B and 6C are for illustration only. Other embodiments could be used without departing from the scope of this disclosure.

[0078] The SRS is sent on the last symbol of an uplink subframe. The SRS transmission can occur either in a periodic manner or be triggered aperiodically via a downlink DCI. In case of periodic SRS, the transmission occurs once every P ms , where P ∈ {2,5,10,20,40,80,160, 320} ms is the network configured SRS transmission periodicity as illustrated in FIG. 6A. Different UE are multiplexed to transmit SRS on the same subframe by transmitting using different cyclic shift and frequency comb. At the eNB receiver, the users are separated after taking into account the unique combination of the cyclic shift and comb associated for each user. For coverage limited UE, to ensure that their transmission power density provides reasonable SNR at the receiver, the SRS transmission is constrained over a portion of the entire (or wideband) uplink bandwidth as illustrated in FIG. 6B. In another embodiment, a UE can be configured to transmit SRS via frequency hopping so that by combining the sounding positions across different hops, the eNB receiver can obtain the channel across the entire bandwidth, as illustrated in FIG. 6C.

[0079] FIG. 7 illustrates a flowchart 700 for training an AI based UE speed classifier for estimating a speed class (among M > 1 speed classes) for one or more UE according to an embodiment. The embodiment of the flowchart 700 shown in FIG. 7 is for illustration only. Other embodiments of the flowchart 700 could be used without departing from the scope of this disclosure.

[0080] In step 710, the AI system equipped with an AI based UE speed classifier generates a set of UL SRS measurements. The AI system generates the labeled data i.e., collection of speed classes and associated features for training the neural network classifier. In an embodiment, the UE speed class for training the neural network could be obtained by simultaneously collecting uplink SRS measurements and (time-stamped) positions of the terminal and measuring the absolute rate of change of the terminal position. A high rate of change of terminal position indicates that the label for that UE moves at "a high speed" while low rate of change indicates the label belongs to "a low-speed". Possibly the terminal could obtain its time-stamped location and communicate to the network e.g. via the application layer. In an alternative embodiment, through a drive test, the network can set up various terminals moving at different (known) speeds and collect uplink SRS measurements associated with the uplink SRS transmissions at each speed.

[0081] In step 720, the input features to the AI classifier are derived from the SRS measurements spaced P ms apart where P is the spacing between consecutive SRS transmissions.

[0082] In step 730, The dataset comprising the input features are divided into a training dataset for a training phase and a test dataset for a test phase.

[0083] In step 740, the flowchart is divided into a training stage (steps 740 and 750) and a test stage (steps 760 and 770). The training phase (steps 740 and 750) is used to adapt the weights of the classifier by feeding the neural network with labeled data. Through the training phase, the AI classifier learns to discriminate different speed category by identifying the aspect of the input feature best suited for each category. The training phase includes N train labeled data {(x i , y i ), 0 <= i <= N train -1}, where x i designates the input feature (e.g. a set of uplink channel measurements (or a function of channel measurement) spaced P ms apart) and y i denotes the label (i.e. UE speed class) tagged to that measurement. Note that the training can be performed offline so that once the neural network is trained, it is ready to classify speeds belonging to unlabeled data.

[0084] In step 760, the AI classifier is fed in with features corresponding to unlabeled data and the output of the AI classifier is used to tag the speed class for the unlabeled data.

[0085] In step 770, the AI classifier estimates the speed class (among M > 1 speed classes) for one or more UE based on uplink signal measurements associated with each UE.

[0086] In step 780, the AI classifier ends the training or test.

[0087] The classifier accuracy refers to the percentage of the unlabeled data for which the neural network correctly assigns the speed category.

[0088] Various types of the input features can be extracted from UL SRS measurements for the AI classifier. In a first embodiment, the input feature for the AI classifier equals the power spectral density derived from the fast fourier transform (FFT) of the tap gain auto-correlation function associated with the estimated channel impulse response (CIR). In a second different embodiment, the input feature equals the raw frequency domain / time domain uplink SRS channel coefficients.Power Spectral Density Feature Extraction

[0089] FIG. 8 illustrates a flowchart 800 for estimating a speed class from power spectral density (PSD) measurements according to an embodiment of the disclosure. The embodiment is for illustration only. Other embodiments could be used without departing from the scope of this disclosure.

[0090] The flowchart 800 for extracting the PSD which is used to provide the input feature for the AI classifier is shown in FIG. 8. While the flowchart 800 shows the procedure for a single user, the same procedure can be straight-forwardly extended when multiple UEs are multiplexed to simultaneously sound in the same region of bandwidth. The following paragraphs describe the mathematical procedure to extract the autocorrelation function and the power spectral density from the SRS channel estimate.

[0091] In step 810, the AI system collects and buffers a number of UL SRS measurements derived from UL SRS transmissions of the mobile client device, the number exceeding a threshold. Assume the first SRS transmission occurs at time t = t 0 and assume M separate UL SRS measurements are performed, each measurement separated by the SRS transmission interval P.

[0092] In step 820, the AI system processes SRSs using Zadoff-Chu (ZC) sequence per UE.

[0093] In step 830, the AI system derives CIRs on a per transmit and receive antenna pair basis, from one UL SRS measurement or by combining UL SRS measurements. The AI system also buffers a certain number of CIRs, in a sliding window, wherein each estimated CIR is formed at different SRS reception times. In an embodiment, the AI system extracts L strongest channel coefficient from the CIR for each mobile client device.

[0094] In step 840, for each tap, the AI system extracts autocorrelation functions at different lags. The AI system designates the estimate of the time varying channel gain of tap l at time t= t 0 +mP as { h l [m], 0 <= l <= L-1}. Assuming wide-sense stationarity, the normalized tap gain autocorrelation function R l [n] denotes the correlation of the l th< tap at two different times separated by a lag equaling nP and computed as: R l n = E h l * m h l m + n E h l m 2 , 0 ≤ n ≤ M − 1

[0095] The numerator and denominator can be empirically computed via E h l * m h l m + n ≈ 1 M − n ∑ m = 0 M − n − 1 h l * m h l m + n and E h l m 2 ≈ 1 M ∑ m = 0 M − 1 h l m 2 .

[0096] In step 850, the AI system calculates an average autocorrelation across L'<=L taps. The empirical autocorrelation estimate after averaging across different taps is designated as R ^ n = 1 L ∑ l R l n , 0 ≤ n ≤ M - 1.

[0097] In step 860, the AI system performs FFT to get a Doppler PSD. The Doppler PSD denoted as S(f), f ∈ [-1 / 2P,1 / 2P] equals the discrete fourier transform of R l ^ n . The power spectral density tends to take a U-shaped profile with the width of the profile approximately equaling 2f d .

[0098] In step 870, the AI system inputs the Doppler PSD as features to the neural network to categorize a speed of mobile client device. the category output by the ML classifier includes at least one of an estimate of either a travel speed or a range of travel speed of the mobile client device, an estimate of either a travel velocity or a range of travel velocity of the mobile client device, an estimate of either a Doppler frequency or a range of Doppler frequency of a dominant radio frequency (RF) propagation path, or an estimate of either a Doppler frequency per path or a range of Doppler frequency per path on a set of RF propagation paths.

[0099] The maximum UE speed up to which the PSD method can be used for classification is determined as a function of the SRS transmission interval P ms (in milli-seconds) and the uplink carrier frequency f c,GHz and mathematically given as:- MaxSpeed kmph = 540 P ms × f c , GHz

[0100] Thus, for 5ms sounding interval at 2.1 GHz carrier, the maximum UE speed up to which the classifier works equals approximately 51.4 Kmph. UE speeds beyond 51.4 Kmph are aliased when the Doppler PSD is computed and could be erroneously classified as a low speed UE.

[0101] FIG. 9 illustrates an example graph 900 of empirically derived PSD according to an embodiment of the disclosure. The embodiment shown in FIG. 9 is for illustration only. Other embodiments could be used without departing from the scope of this disclosure.

[0102] The Doppler power spectral density at a terminal speed equaling 19.95 Kmph (f d = 38.8 Hz) is as shown in FIG. 9. It can be seen that the width of the U-shaped profile approximately equals 77.6 Hz. If the receiver has multiple transmit and receive antennas, then the power spectral density can be separately derived for each TX-RX antenna pair and fed as input features into a 1D / 2D convolutional neural network. By training the neural network with labeled power spectral densities, the AI classifier learns to recognize the speed category for that user.

[0103] In case of a frequency flat channel (L = 1, communication bandwidth W is much less than the reciprocal fo the delay spread) and assuming with infinite scattering, from the Clarke's model, it can be shown that the normalized autocorrelation R ^ l n , corresponding to a lag n, reduces to the zeroth order Bessel function of the first order given as J 0 (2nπf d P) where f d is the maximum doppler frequency (as described earlier in document). The power spectral density S(f) reduces to a U-shaped profile given by following equation S f = 2 W f d 1 − fW / f d 2 , − f d W ≤ f ≤ f d W 0 , else

[0104] FIG. 10A illustrates an example of wideband SRS transmissions 1005, and FIG. 10B illustrates an example of frequency hopped SRS (FH SRS) transmissions 1010, according to embodiments of the disclosure. The embodiments shown in FIGS. 10A and 10B are for illustration only. Other embodiments could be used without departing from the scope of this disclosure.

[0105] An additional consideration while deriving the PSD measurements is whether the UE of interest transmits SRS across the entire bandwidth or whether it performs FH SRS transmission.

[0106] In case of wide-band SRS as illustrated in FIG. 10A, the tap gain autocorrelation function (and hence PSD) can be obtained by correlating the CIRs across consecutive UL SRS measurements. The maximum estimable UE speed is then a function of the maximum SRS transmission interval and is given as expressed in the previous page. During each SRS transmission opportunity, the mobile client device transmits the UL SRSs across an entire bandwidth, the time-domain auto-correlation function is derived by measuring auto-correlation functions at different lags of the buffered CIRs obtained from time-consecutive UL SRS measurements. The channel measurements at a given SRS transmission occasion are derived from corresponding UL SRS measurements.

[0107] In case of FH SRS as illustrated in FIG. 10B, since the UE sounds on a different region of spectrum during each hop, the CIR obtained by taking the IDFT over the SRS frequency domain position on a given hop may not accurately represent the true CIR (based on the wideband channel). During each SRS transmission opportunity, the mobile client device transmits SRSs across a portion on an entire bandwidth in a frequency hopping manner. The channel measurements for the entire bandwidth are derived by concatenation of the channel measurements on each bandwidth portion, obtained from UL SRS measurements occurring on consecutive UL SRS hopping transmission occasions. The time-domain auto-correlation function is obtained by correlating the buffered CIRs derived from uplink SRS measurements on time-adjacent SRS transmissions, or the time domain auto-correlation function is obtained by correlating buffered CIRs derived from uplink SRS measurements extracted from UL SRS hopping transmission occasions occurring on identical frequency sub-bands.

[0108] FIGS. 11A and 11B illsutrate two methods for correlating across CIRs for the case of FH SRS according to an embodiment of the disclosure. The embodiments shown in FIGS. 11A and 11B are for illustration only. Other embodiments could be used without departing from the scope of this disclosure.

[0109] There are several methods of correlating the CIRs to compute the autocorrelation function. In the first method 1105, the CIRs are correlated by combining SRS hops on per identical sounding resource element, as illustrated in FIG. 11A. While this method accurately captures the channel variation across SRS on identical resources, the maximum estimable UE speed is limited by how often the UE sounds on identical RE positions, and given as MaxSpeed kmph = 540 P ms × N hops × f c , GHz

[0110] In the second method 1110, the CIRs are correlated by combining adjacent SRS hops even though the hops may correspond to different sounding resource elements, as illustrated in FIG. 11B. This method has the same maximum estimable UE speed as that corresponding to wide-band SRS.

[0111] In the third method, when the UE transmits SRS on a sub-band SB k , the delays and complex channel gains are estimated by correlating the frequency domain channel corresponding to SB k . The frequency domain channel on the other sub-bands SB l (l is different from k) is estimated by reconstructing the channel response from the estimated delays and complex channel gains. After assembling the estimated channel over other sub-bands in addition to the sub-band on which the UE sounds, the entire wide-band SRS measurement is available. Thereafter, the doppler power spectrum is estimated by correlating the IFFT of the reconstructed frequency domain channel across consecutive sounding occasions.In-phase and Quadrature (IQ) Channel Feature Extraction

[0112] In some embodiments, the raw channel measurement can be used to train the AI classifier. The motivation behind this embodiment is that raw channel measurement will include more information related to UE speed. Training with large amount of raw measurements, the AI classifier can automatically learn the features of channel measurements and store these learned features into the weight parameters of the neural network. The raw channel measurements in this embodiment could be measurements on several dimensions, e.g., time dimensional channel snapshots (SRS) with a periodicity of P ms , channel samples on OFDM subcarriers, spatial dimensional channel samples on different antennas, real part of channel and imaginary part of the channel.

[0113] There are several methods to design which part of raw channel measurement will be input to the AI classifier.

[0114] In an embodiment, the extracted features are outputs of a linear or non-linear function of real and imaginary portions of channel measurements derived from UL SRS measurements per transmit and receive antenna pair during each UL SRS transmission occasion, wherein the linear or non-linear function comprises a pre-processing process of applying a fusion function of the real and imaginary portions of the channel measurements, wherein the fusion function is configured to output the real and imaginary portions of the channel measurements in a frequency domain, or a convex combination of weighted real portions and imaginary portions of the channel measurements in one of the frequency domain, a receive (RX) antenna domain, a transmit (TX) antenna domain, a subcarrier domain, or a channel snapshot domain.

[0115] In this embodiment, both real and imaginary part of frequency domain channel measurements will be input to the AI classifier. Denote the channel as H which is a 4 dimensional N rx ×N tx × N sub × M complex tensor, where N rx denotes the number of transmit antennas, N tx denotes the number of receive antennas, N sub denotes the number of subcarriers used by SRS, M denotes the number of channel snapshots. The input to the AI classifier will be the concatenated tensor where real part of the channel measurement and imaginary part of the channel measurement will be concatenated in one dimension. For example, the input to AI classifier is of the dimension of (2N rx )×N tx × N sub × M tensor where real and imaginary part of channel measurements are concatenated in RX antenna dimension, or it can be N rx ×(2N tx )× N sub × M tensor where real and imaginary part of channel measurements are concatenated in tx antenna domain, or it can be N rx ×N tx × (2N sub )× M tensor where real and imaginary part of channel measurements are concatenated in subcarrier domain, or it can be N rx ×N tx × N sub × (2M) tensor where real and imaginary parts of channel measurements are concatenated in channel snapshot domain.

[0116] FIG. 11C illustrates an exemplary flowchart 1120 for pre-processing frequency domain channel measurements for a deep neural network architecture according to embodiments of the disclosure. The embodiment shown in FIG. 11C is for illustration only. Other embodiments of the flowchart 1120 could be used without departing from the scope of this disclosure.

[0117] The base station buffers a number of UL SRS frequency domain measurements derived from different UL SRS transmissions of the mobile client device in step 1122.

[0118] The real part and imaginary part of each frequency domain channel measurement will be first pre-processed by a general fusion function f ri in step 1124, and then input to an AI classifier with input dimension of N rx ×N tx × N sub × M, in step 1126. For example, one fusion function can be the inverse tangent function to get the phase of the channel measurement. Another fusion function can be f ri = w 0 H Re + 1 − w 0 H Im where w 0 ∈ [0,1]. Specifically, if w0=0, then only imaginary part of channel measurements will be input to AI classifier; if w0=1, then only real part of channel measurements will be input to AI classifier. HRe and HIm respectively indicate each real portion and each imaginary portion of the channel measurements in the frequency domain.

[0119] In yet another embodiment, a fusion function is applied to either the RX antenna domain, or applied to TX antenna domain, or to subcarrier domain, or in channel snapshot domain, or a combination of these domains. A fusion function across different domains will further reduce the complexity of the AI classifier. In one example of this method, a fusion function is applied to the RX antenna domain. Denote H 0 Rx< , H 1 Rx< , H 2 Rx< , ..., H Nrx Rx< as the channel measurements on each Rx antenna, the fusion function f Rx can be expressed as f Rx = w 0 × H 0 Rx + w 1 × H 1 Rx + w 2 × H 2 Rx + … + w Nrx × H Nrx Rx , where w 0+ w 1+... + w Nrx =1, w 0 , w 1, ... , w Nrx ∈ [0,1].

[0120] In yet another method of this embodiment, both real and imaginary parts of time domain raw channel measurement will be input to the AI classifier. Denote the channel as H which is a 4-dimensional N rx ×N tx × N taps × M complex tensor, where N taps denotes the number of taps of the estimated CIR. The same concatenating method, fusion method can be used as mentioned in the above embodiments. In general, any function f() may be defined which can operate over the channel measurements and then the resulting output may be input to the neural network. The type of the neural network chosen can be a simple feed forward neural network, or a recurrent neural network or any advanced neural network and can be adapted for the need.

[0121] FIG. 12 illustrates an example of a sliding-window based mechanism to track a UE speed 1200 according to an embodiment of the disclosure.

[0122] The AI system can use a sliding-window 1205 based mechanism to track and identify a UE speed or changes to a UE speed class. The AI system buffers a certain number of CIRs, in a sliding window, wherein each estimated CIR is formed at different SRS reception times, wherein a size of the sliding window is adjustable based on the categories to the speed of a mobile client device belongs to.

[0123] Since the exact time when the UE changes its speed is unknown to the gNB (base station) or to any other terminal (considering the other end terminal with which a link is established is moving). The appropriate window size can need to be figured out in order to obtain the testing data. The size of the window and the moving rate of the window can be chosen appropriately considering the confidence required. In an embodiment, the window size can be proportional to the confidence required.

[0124] The output layer of the classifier provides a probability estimate of the individual class membership probabilities of the terminal speed based on the UL SRS measurements. For example, with M class categories (e.g. for M = 2, the categories for the labeled data during training stage could be Speed <= 25 Kmph and 25 <= Speed <= 51 Kmph), the neural classifier will return as output a vector of M outputs summing to one. In an embodiment, the speed category for the user is determined by the class corresponding to the index of the maximum within the vector. To realize the outputs of the neural classifier o be a valid probability distribution, the output layer can have a "softmax" activation function. Other embodiments for activation functions at the output layer of the neural network are also possible as long as the activation function returns a vector of M outputs summing to one.

[0125] FIG. 13A illustrates an example of a convolutional neural network (CNN) classifier for estimating a UE speed based on Doppler power spectrum measurements, and FIG. 13B illustrates an example of a CNN speed classifier based on raw frequency domain UL SRS measurements, according to embodiments of the disclosure. The embodiment shown in FIGS. 13A and 13B are for illustration only. Other embodiments of the flowchart 700 could be used without departing from the scope of this disclosure.

[0126] The deep neural networks as illustrated in FIGS. 13A and 13B include multiple CNN layers, multiple fully connected (FC) neural network layers, and multiple recurrent neural network layers.

[0127] The convolutional neural network architecture 1305 as shown in FIG. 13A estimates a UE speed based on a set of power spectrum density measurements. The convolutional neural network architecture 1305 comprises two one-dimension (1D) convolutional layers followed by a fully connected neural network architecture. Pooling layers are present between the output of each convolutional layer and the next layer in order to reduce the dimension of the output.

[0128] The deep neural network architecture 1310 as shown in FIG. 13B estimates a UE speed based on a set of pre-processed frequency domain real and imaginary portions of UL SRS measurements. The deep neural network architecture 1310 includes a pooling layer placed between a CNN layer and a subsequent CNN layer, a flattening layer placed between a final CNN layer and a first FC layer, and a last FC layer producing an activation output determined using a softmax output activation wherein a dimension of the activation output equals a number of candidate categories. The deep neural network architecture returns an output that is determined by an index for an argument maximizer of the activation output of the last fully connected layer, and that specifies an estimated category of the mobility of the mobile client device.

[0129] FIG. 13C illustrates a flowchart 1320 for a deep neural network architecture with different sets of neural network layer weights to be applied to each layer of the multiple CNN layers, according to claimed embodiments of the disclosure. The embodiment shown in FIG. 13C is for illustration only. Other embodiments of the flowchart 1320 could be used without departing from the scope of this disclosure.

[0130] The deep neural network architectures 1305 and 1310 can include a number of layers and connectivities between each layer and its subsequent layer for the deep neural network architecture, and different sets of neural network layer weights to be applied to each layer of the multiple CNN layers.

[0131] In step 1322, the deep neural network architecture obtains a signal measurement (e.g., estimate an SNR level) or feedback from the mobile client device.

[0132] In step 1324, the deep neural network architecture selects one set of neural network layer weights, among different sets of neural network layer weights, based on a certain signal measurement or feedback from the mobile client device, and apply the selected set of neural network layer weights to each layer of the multiple CNN layers.

[0133] In step 1326, the deep neural network architecture loads a weighted neural network architecture by applying the selected set of neural network weights to each CNN layer and fully connected layer.

[0134] Each different set of neural network weights corresponds to different estimated signal to noise ratio (SNR) levels. The deep neural network architectures 1305, 1310 estimate an SNR level based on a reference signal received power (RSRP) and a reference signal received quality (RSRQ) reported, or based on channel quality indication (CQI) feedback from the mobile client device, selects a set of neural network layer weights among the set of weights, based on the estimated different SNR level, and loads a weighted neural network architecture for determining the category to be used to estimate mobility of the mobile client device, by applying the selected set of neural network weights to each CNN layer and fully connected layer.

[0135] In some embodiment of this disclosure, the deep neural network architectures 1305, 1310 can adapt to the change of SNR levels. During the training stage, the deep neural network architectures 1305, 1310 can be trained with channel measurements at different SNR levels and stores the neural network weights at each SNR level. During the online testing stage, with different SNR level, deep neural network architectures 1305 and 1310 load different weights to do UE speed estimation / classification. The SNR level can be obtained from RSRP and RSRQ reported from UE. The following is the procedures for UE speed estimation considering different SNR levels: 1) Estimate the SNR level for a UE based on the RSRP and RSRQ information; 2) based on different SNR level, load different neural network weights; 3) obtain the channel measurements and input measurement or pre-processed measurement to the neural network architectures; and 4) get the UE speed estimation / classes.

[0136] In some embodiment of this disclosure, the output layer of the neural network can be modified to a rectified linear unit (ReLU) function. During the supervised learning stage, the true UE speed will be feed to the neural network. After training, the neural network will have the capability to predict the UE speed based on the channel measurements.

[0137] Classifier accuracy has been shown 2 class classifications, although the method can be straight-forwardly extended to more than 2 classes. In the evaluations, it is assumed that the SRS is transmitted once every P ms = 5 ms and the carrier frequency on uplink equals f c,GHz = 2.1 GHz.

[0138] Table 1 shows classifier accuracies for wideband SRS in a non-line-of-sight (NLOS) channel. In Table 1, the classification accuracy is shown for the case of a wide-band SRS transmission assuming a 5 ms SRS transmission interval, NLOS channel (resp. line-of-sight (LOS) channel) and carrier frequency equaling 2.1 GHz. The speed thresholds for the two classes are Speed < 25 Kmph (Class 1) and 25 Kmph < Speed < 51 Kmph (Class 2).

[0139] At SNR = 6 dB, under NLOS channel, the neural classifiers provide nearly 90 % classification accuracy. It can be seen that even when the SNR equals 0 dB, the accuracies achieved by the neural classifiers are at least 85 % for the NLOS scenario. The performance of the neural classifiers under LOS conditions is worse because the dominant LOS path implies that the classifier has more difficulty learning the magnitude of the velocity vector. Nevertheless, the performance of the classifiers is uniformly over 75 %. Table 1Channel ParametersSchemeAccuracySNR= 6 dB(Non-AI baseline) PSD74%Neural PSD88%Raw channel (96 RB)94%Raw channel (6RB)94%SNR= 0 dB(Non-AI baseline) PSD74%Neural PSD85%Raw channel (96 RB)94%Raw channel (6RB)92.5% Table 2 Classification accuracy for Wide Band SRS in Line of Sight Channel Channel ParametersSchemeAccuracyLOS, SNR= 6 dB(Non-AI baseline) PSD-Neural PSD76-78%Raw channel (96 RB)80.5%Raw channel (6RB)83%LOS, SNR= 0 dB(Non-AI baseline) PSD70%Neural PSD73%Raw channel (96 RB)78%Raw channel (6RB)78%

[0140] Table 2 analyzes the classification accuracy with the neural classifiers with lower SRS transmission periodicities. It is seen that good classifier accuracy is achieved as long as the sounding interval is 10 ms or lower.

[0141] Table 3 shows the impacts of a lower SRS transmission periodicity on the classifier accuracy. Table 3System ParametersChannelFeatureClassifier AccuracyP ms = 10 msLOSPSD76%SNR = 6 dBRaw Channel82%NLOSPSD86%Raw Channel95%P ms = 20 msLOSPSD58%SNR = 6 dBRaw Channel52%NLOSPSD69%Raw Channel81%P ms = 40 msLOSPSDTBDSNR = 6 dBRaw Channel54%NLOSPSDTBDRaw Channel59%

[0142] Table 4 below shows classification accuracies for FH SRS by correlating across SRS hops with identical sounding RE positions. Table 4System Parameters Speed Classes Channel Model Classifier Accuracy N hop = 2 hops, P ms = 5 ms, SNR = 6 dBClass 1 : Speed <= 12.8 Kmph3GPP TR 38.901 NLOS model77%Class 2: 12.8 Kmph < Speed < 25.6 Kmph3GPP TR 38.901 LOS model90%

[0143] Table 5 below shows classification accuracies for FH SRS by correlating across adjacent SRS hops. Table 5System Parameters Speed Classes Channel Model Classifier Accuracy N= 2 hops, P ms = 5 ms, SNR = 6 dBClass 1 : Speed <= 25 Kmph3GPP TR 38.901 NLOS model94 %Class 2: 25 Kmph < Speed < 51 Kmph3GPP TR 38.901 LOS model80 %N hop = 4 hops, P ms = 5 ms, SNR = 6 dB3GPP TR 38.901 NLOS model80%3GPP TR 38.901 LOS model76%

[0144] From Table 4 and Table 5, it can be seen that the classification accuracy when the UE transmits FH SRS is quite competitive against the case when the UE transmits SRS in a wide-band manner.

[0145] According to various embodiments, a method for operating a base station in a wireless communication system comprises receiving uplink (UL) sounding reference signals (SRSs) from a terminal; obtain a number of UL SRS measurements from UL SRSs received from the terminal, the number of UL SRS measurements exceeding a threshold value; extracting features for estimating a mobility of the terminal from the UL SRS measurements; and determining a category of the terminal based on the extracted features.

[0146] According to various embodiments, the category is determined by using a machine learning (ML) classifier. The category includes at least one of: an estimate of either a travel speed, or a range of travel speed of the terminal; an estimate of either a travel velocity or a range of travel velocity of the terminal; an estimate of either a doppler frequency or a range of doppler frequency of a dominant radio frequency (RF) propagation path; or an estimate of either a doppler frequency per path or a range of doppler frequency per path on a set of RF propagation paths. The ML classifier is a deep neural network including at least one of: multiple convolutional neural network (CNN) layers; multiple fully connected (FC) neural network layers; or multiple recurrent neural network layers.

[0147] According to various embodiments, the extracted features include a set of frequency domain doppler power spectral density (PSD) measurements that are estimated by: deriving channel impulse responses (CIRs) on a per transmit and receive antenna pair basis, from one UL SRS measurement or by combining UL SRS measurements; buffering a certain number of CIRs, in a sliding window, wherein each estimated CIR is formed at different SRS reception times; deriving a time-domain auto-correlation function obtained by correlating the buffered CIRs at different SRS reception times; and taking a fast fourier transform (FFT) of the time-domain auto-correlation function. A size of the sliding window is adjustable based on the category to the terminal belongs to.

[0148] According to various embodiments, if, during each SRS transmission opportunity, the terminal transmits the UL SRSs across an entire bandwidth, a time-domain auto-correlation function is derived by measuring auto-correlation functions at different lags of the buffered CIRs obtained from time-consecutive UL SRS measurements.

[0149] According to various embodiments, if, during each SRS transmission opportunity, the terminal transmits SRSs across a portion on an entire bandwidth in a frequency hopping manner: a time-domain auto-correlation function is obtained by correlating the buffered CIRs derived from uplink SRS measurements on time-adjacent SRS transmissions; or the time domain auto-correlation function is obtained by correlating buffered CIRs derived from uplink SRS measurements extracted from UL SRS hopping transmission occasions occurring on identical frequency sub-bands.

[0150] According to various embodiments, the extracted features are outputs of a linear or non-linear function of real and imaginary portions of channel measurements derived from UL SRS measurements per transmit and receive antenna pair during each UL SRS transmission occasion. The linear or non-linear function comprises a pre-processing process of applying a fusion function of the real and imaginary portions of the channel measurements. The fusion function is configured to output: the real and imaginary portions of the channel measurements in a frequency domain; or a convex combination of weighted real portions and imaginary portions of the channel measurements in one of the frequency domain, a receive (RX) antenna domain, a transmit (TX) antenna domain, a subcarrier domain, or a channel snapshot domain.

[0151] According to various embodiments, the method further comprises estimating a signal-to-noise ratio (SNR) level based on a reference signal received power (RSRP) and a reference signal received quality (RSRQ) reported, or based on channel quality indication (CQI) feedback from the terminal, wherein each different set of neural network weights corresponds to different estimated SNR levels; selecting a set of neural network layer weights among the set of weights, based on estimated different SNR levels; and loading a weighted neural network architecture for determining the category to be used to estimate mobility of the mobile client device, by applying the selected set of neural network weights to each convolutional neural network (CNN) layer and fully connected layer.

[0152] None of the description in this application should be read as implying that any particular element, step, or function is an essential element that must be included in the claim scope. The scope of patented subject matter is defined only by the claims. Use of any other term, including without limitation "mechanism," "module," "device," "unit," "component," "element," "member," "apparatus," "machine," "system," "processor," or "controller," within a claim is understood by the applicants to refer to structures known to those skilled in the relevant art.

Claims

1. A base station (101, 102) in a wireless communication, the base station comprising: a communication interface configured to receive sounding reference signals, SRSs, from a user equipment, UE, (116); and at least one processor (378) configured to: receive, from the UE (116), the SRSs, obtain SRS measurements from the SRSs received from the UE (116), wherein a number of the obtained SRS measurements exceeds a threshold value, extract, from the SRS measurements, features for estimating a mobility of the UE (116), wherein the features include a set of frequency domain Doppler power spectral density, PSD, measurements, the base station being characterized in that the at least one processor (378) is further configured to: estimate a signal to noise ratio, SNR, level based on a channel quality indication, CQI, feedback from the UE (116), select a set of neural network layer weights among sets of neural network layer weights for a neural network architecture, based on the estimated SNR level, load the weighted neural network architecture for determining a speed category used to estimate the mobility of the UE, by applying the selected set of the neural network layer weights to each convolutional neural network, CNN, layer and fully connected layer included in a deep neural network, determine the speed category in which a speed of the UE (116) belongs based on the extracted features by using the deep neural network, wherein each of the sets of the neural network layer weights for the neural network architecture corresponds to each of estimated SNR levels, and wherein the speed category includes a range of an estimated speed of the UE.

2. The base station (101, 102) of claim 1, wherein the speed category further includes at least one of a range of an estimated doppler frequency of a dominant radio frequency, RF, propagation path, or a range of an estimated doppler frequency per path on a set of RF propagation paths.

3. The base station (101, 102) of claim 1, wherein the frequency domain Doppler PSD measurements are estimated by: deriving channel impulse responses, CIRs, on a per transmit and receive antenna pair basis, from one SRS measurement or by combining the SRS measurements; buffering a certain number of CIRs, in a sliding window, wherein the buffered CIRs are formed at different SRS reception times; deriving a time-domain auto-correlation function obtained by correlating the buffered CIRs; and performing a fast fourier transform, FFT, of the time-domain auto-correlation function, wherein a size of the sliding window is adjustable based on the speed category to which the terminal (116) belongs to.

4. The base station (101, 102) of claim 1, wherein the deep neural network is fed a true speed of the UE while the deep neural network is trained, and wherein the deep neural network is configured to predict a speed of the UE after training.

5. A method for operating a base station (101, 102) in a wireless communication system, the method comprising: receiving, from a user equipment, UE, (116), sounding reference signals, SRSs; obtain SRS measurements from the SRSs received from the UE (116), wherein a number of the obtained SRS measurements exceeds a threshold value; extracting, from the SRS measurements, features for estimating a mobility of the UE (116), wherein the features include a set of frequency domain Doppler power spectral density, PSD, measurements; the method being characterized in that the method further comprises: estimating a signal-to-noise ratio, SNR, level based on a channel quality indication, CQI, feedback from the UE (116); selecting a set of neural network layer weights among sets of neural network layer weights for a neural network architecture, based on the estimated SNR level; loading the weighted neural network architecture for determining a speed category used to estimate the mobility of the UE, by applying the selected set of the neural network layer weights to each convolutional neural network, CNN, layer and fully connected layer included in a deep neural network, and determining the speed category in which a speed of the UE (116) belongs based on the extracted features by using the deep neural network, wherein each of the sets of the neural network layer weights for the neural network architecture corresponds to each of estimated SNR levels, and wherein the speed category includes a range of an estimated speed of the UE.

6. The method of claim 5, wherein the speed category further includes at least one of a range of an estimated doppler frequency of a dominant radio frequency, RF, propagation path, or a range of an estimated doppler frequency per path on a set of RF propagation paths.

7. The method of claim 5, wherein the frequency domain Doppler PSD measurements are estimated by: deriving channel impulse responses, CIRs, on a per transmit and receive antenna pair basis, from one SRS measurement or by combining the SRS measurements; buffering a certain number of CIRs, in a sliding window, wherein the buffered CIRs are formed at different SRS reception times; deriving a time-domain auto-correlation function obtained by correlating the buffered CIRs; and performing a fast fourier transform, FFT, of the time-domain auto-correlation function, wherein a size of the sliding window is adjustable based on the speed category to the UE (116) belongs to.

8. The method of claim 5, wherein the deep neural network is fed a true speed of the UE while the deep neural network is trained, and wherein the deep neural network is configured to predict a speed of the UE after training.