Apparatus, method and computer program

Abstract

There is provided an apparatus for a receiver, the apparatus comprising means for receiving an input signal based on a frequency domain signal, means for performing de-precoding on the received input signal, and means for providing the de-precoded signal as an input for a machine learning model, wherein the output of the machine learning model is a likelihood ratio.

Description

Technical Field

[0001] The present application relates to a method, apparatus, system and computer program, and in particular but not limited to a deep learning receiver for discrete Fourier transform (DFT) spread orthogonal frequency division multiple access (OFDM) signals. Background Art

[0002] A communication system may be viewed as a facility for enabling communication sessions between two or more entities (such as user terminals, base stations, and / or other nodes) by providing a carrier wave between the various entities involved in the communication path. For example, a communication system may be provided with the aid of a communication network and one or more compatible communication devices. A communication session may include, for example, data communications such as voice, video, electronic mail (email), text messages, multimedia, and / or content data for bearer communications. Non-limiting examples of services provided include two-way or multi-way calls, data communications or multimedia services, and access to data network systems such as the Internet.

[0003] In a wireless communication system, at least a portion of a communication session between at least two stations occurs over a wireless link. Examples of wireless systems include public land mobile networks (PLMNs), satellite-based communication systems, and different wireless local area networks, such as wireless local area networks (WLANs). Some wireless systems can be divided into cells and are therefore often referred to as cellular systems.

[0004] A user can access a communication system by means of an appropriate communication device or terminal. The user's communication device may be referred to as a user equipment (UE) or user device. The communication device is provided with appropriate signal receiving and transmitting devices, which are used to achieve communication, such as access to a communication network or direct communication with other users. The communication device can access a carrier provided by a station (e.g., a base station of a cell) and transmit and / or receive communications on the carrier.

[0005] Communication systems and associated equipment typically operate according to a given standard or specification that specifies what the various entities associated with the system are allowed to do and how it should be achieved. Communication protocols and / or parameters that should be used for the connection are also typically defined. An example of a communication system is UTRAN (3G radio). Other examples of communication systems are the Long Term Evolution (LTE) of the Universal Mobile Telecommunications System (UMTS) radio access technology and the so-called 5G or New Radio (NR) networks. NR is being standardized by the Third Generation Partnership Project (3GPP). Summary of the invention

[0006] In a first aspect, a device for a receiver is provided, the device comprising a device for receiving an input signal based on a frequency domain signal, a device for performing de-precoding on the received input signal, and a device for using the de-precoded signal as an input to a machine learning model, wherein the output of the machine learning model is a likelihood ratio.

[0007] The means for performing de-precoding may include an inverse discrete Fourier transform (iDFT) de-precoding function.

[0008] The apparatus may include means for performing the de-precoding on a data branch of the input signal.

[0009] The data branch may include frequency-time grid symbol N f ×N t , where N f Including the number of subcarriers, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0010] The apparatus may include means for concatenating the de-precoded signal with an original channel estimation branch of the frequency domain signal and providing the concatenated signal as an input to the machine learning model.

[0011] The apparatus may include means for providing the input signal to another machine learning model for processing the input signal prior to de-precoding, and means for performing de-precoding on an output of the another machine learning model.

[0012] The output of the other machine learning model may be 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0013] The output of the another machine learning model may also include a constellation monitoring output.

[0014] The device may include means for causing a trigger to be sent to a device for training the machine learning model and another machine learning model (if any), and means for receiving a confirmation from the device in response to the request.

[0015] The device may include means for providing training configuration information to the device, means for receiving training data from the device in response to the configuration information, and means for providing the training data to an entity for training the machine learning model and another machine learning model (if any).

[0016] The apparatus may include means for receiving deployment configuration information from a transmitter of the input signal, and means for determining whether to deploy a machine learning model and another machine learning model (if any) based on the deployment configuration information.

[0017] The deployment configuration information may include at least one of an indication to perform the de-precoding on the frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

[0018] The apparatus may include means for determining not to deploy the machine learning model and the other machine learning model, if any, based on the fallback indicator.

[0019] In a second aspect, a method for a receiver is provided, the method comprising: receiving an input signal based on a frequency domain signal, performing de-precoding on the received input signal, and providing the de-precoded signal as an input to a machine learning model, wherein the output of the machine learning model is a likelihood ratio.

[0020] Performing de-precoding may include an inverse discrete Fourier transform (iDFT) de-precoding function.

[0021] The method may comprise performing the de-precoding on a data branch of the input signal.

[0022] The data branch may include frequency-time grid symbol N f ×N t , where N f Including the number of subcarriers, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0023] The method may include cascading the de-precoded signal with an original channel estimation branch of the frequency domain signal, and using the cascaded signal as an input to the machine learning model.

[0024] The method may include providing the input signal to another machine learning model for processing the input signal before de-precoding, and performing de-precoding on the output of the another machine learning model.

[0025] The output of the other machine learning model can be 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0026] The output of the another machine learning model may also include a constellation monitoring output.

[0027] The method may include causing a trigger to be sent to a device for training the machine learning model and another machine learning model (if present), and receiving a confirmation from the device in response to the request.

[0028] The method may include providing training configuration information to the device, receiving training data from the device in response to the configuration information, and providing the training data to an entity for training the machine learning model and another machine learning model (if present).

[0029] The method may include receiving deployment configuration information from a transmitter of the input signal and determining whether to deploy the machine learning model and another machine learning model (if any) based on the deployment configuration information.

[0030] The deployment configuration information may include at least one of an indication to perform the de-precoding on the frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

[0031] The method may include determining not to deploy the machine learning model and the other machine learning model (if any) based on the fallback indicator.

[0032] In a third aspect, a device is provided, comprising: at least one processor and at least one memory comprising computer program code, wherein the at least one memory and the computer program code are configured to utilize the at least one processor together to cause the device to, at least at a receiver: receive an input signal based on a frequency domain signal, perform de-precoding on the received input signal, and provide the de-precoded signal as an input to a machine learning model, wherein the output of the machine learning module is a likelihood ratio.

[0033] The apparatus may include an inverse discrete Fourier transform (iDFT) de-precoding function.

[0034] The apparatus may be caused to perform the de-precoding on a data branch of the input signal.

[0035] The data branch may include frequency-time grid symbol N f ×N t , where N f Including the number of subcarriers, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0036] The device may be enabled to cascade the de-precoded signal with an original channel estimation branch of the frequency domain signal, and use the cascaded signal as input to the machine learning model.

[0037] The device may provide the input signal to another machine learning model for processing the input signal before de-precoding, and perform de-precoding on the output of the other machine learning model.

[0038] The output of the other machine learning model may be 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0039] The output of the another machine learning model may also include a constellation monitoring output.

[0040] The device can be caused to cause a trigger to be sent to a device for training a machine learning model and another machine learning model (if any), and to receive a confirmation from the device in response to the request.

[0041] The device may be caused to provide training configuration information to the device, receive training data from the device in response to the configuration information, and provide the training data to an entity used to train the machine learning model and the other machine learning model (if any).

[0042] The device may be enabled to receive deployment configuration information from a transmitter of the input signal and determine whether to deploy the machine learning model and the other machine learning model (if any) based on the deployment configuration information.

[0043] The deployment configuration information may include at least one of an indication to perform the de-precoding on the frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

[0044] The device may be caused to determine not to deploy the machine learning model and the other machine learning model (if any) based on the fallback indicator.

[0045] In a fourth aspect, a computer-readable medium is provided, comprising program instructions for causing a device to perform at least the following operations at a receiver: receive an input signal based on a frequency domain signal, perform de-precoding on the received input signal, and provide the de-precoded signal as an input to a machine learning model, wherein the output of the machine learning model is a likelihood ratio.

[0046] Performing de-precoding may include an inverse discrete Fourier transform (iDFT) de-precoding function.

[0047] The apparatus may be caused to perform the de-precoding on a data branch of the input signal.

[0048] The data branch may include frequency-time grid symbol N f ×N t , where N f Including the number of subcarriers, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0049] The device may be enabled to cascade the de-precoded signal with an original channel estimation branch of the frequency domain signal, and use the cascaded signal as input to the machine learning model.

[0050] The device may be configured to provide the input signal to another machine learning model for processing the input signal before de-precoding, and to perform de-precoding on the output of the other machine learning model.

[0051] The output of the other machine learning model may be 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0052] The output of the another machine learning model may also include a constellation monitoring output.

[0053] The device may be caused to cause a trigger to be sent to a device for training a machine learning model and another machine learning model (if any), and to receive a confirmation from the device in response to the request.

[0054] The device may be caused to perform: providing training configuration information to the device, receiving training data from the device in response to the configuration information, and providing the training data to an entity used to train the machine learning model and another machine learning model (if any).

[0055] The device may be caused to perform: receiving deployment configuration information from a transmitter of the input signal, and determining whether to deploy the machine learning model and the other machine learning model (if any) based on the deployment configuration information.

[0056] The deployment configuration information may include at least one of an indication to perform the de-precoding on the frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

[0057] The device may be caused to perform an operation of determining not to deploy the machine learning model and the other machine learning model (if any) based on the fallback indicator.

[0058] In a fifth aspect, a non-transitory computer-readable medium is provided, comprising program instructions for causing an apparatus to perform a method according to at least the third or fourth aspect.

[0059] In the above, many different embodiments have been described. It should be understood that further embodiments can be provided by combining any two or more of the above embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Embodiments will now be described, by way of example only, with reference to the accompanying drawings, in which:

[0061] Figure 1 shows a schematic diagram of an example 5GS communication system;

[0062] Figure 2 A schematic diagram of an example mobile communication device is shown;

[0063] Figure 3 A schematic diagram of an example control arrangement is shown;

[0064] Figure 4 A block diagram showing a conventional UL receiver architecture with potential DeepRx replacement;

[0065] Figure 5 A block diagram showing the DeepRx architecture for OFDM systems;

[0066] Figure 6 The transmitter (Tx) and receiver (Rx) signal processing chains for OFDM and DFTS-OFDM systems are shown;

[0067] Figure 7 shows a flow chart of a method according to an example embodiment;

[0068] Figure 8 shows a block diagram of the structure of precoding DeepRx according to an example embodiment;

[0069] Fig. 9 shows a detailed block diagram of the structure of precoding DeepRx according to an example embodiment;

[0070] Fig.10 shows a block diagram of a de-precoding embedded hybrid DeepRx according to an example embodiment;

[0071] Fig.11 shows a detailed block diagram of the architecture of the De-precoding Embedded Hybrid DeepRx according to an example embodiment;

[0072] Fig.12 shows a signaling diagram according to an example embodiment;

[0073] Fig.13 shows a signaling diagram according to an example embodiment;

[0074] Fig.14 shows a block diagram of a schedule according to an example embodiment;

[0075] Fig.15 shows a block diagram of a schedule according to an example embodiment;

[0076] Fig.16 shows uncoded bit error rates (BER) for a conventional receiver and receivers according to first and second example embodiments;

[0077] Fig.17 shows the coding BER for a conventional receiver and a receiver according to the first and second exemplary embodiments;

[0078] Fig.18 shows the BER in training for a conventional receiver and a receiver according to the first and second embodiments;

[0079] Fig.19 A constellation output according to an example embodiment is shown. DETAILED DESCRIPTION

[0080] Before explaining each example in detail, refer to Figures 1 to 3 Certain general principles of wireless communication systems and mobile communication devices are briefly explained to aid understanding of the technology behind the described examples.

[0081] An example of a suitable communication system is a 5G system (5GS). The network architecture in 5GS may be similar to that of advanced LTE. The base stations of the NR system may be referred to as next generation Node Bs (gNBs). Changes in the network architecture may depend on the need to support various radio technologies and more sophisticated QoS support, as well as some on-demand requirements, such as QoS levels to support QoE from the user's perspective. In addition, network-aware services and applications, as well as service- and application-aware networks, may bring changes to the architecture. These are related to information-centric networks (ICNs) and user-centric content delivery networks (UC-CDN) approaches. NR may use multiple-input multiple-output (MIMO) antennas, more base stations or nodes than LTE (the so-called small cell concept), including macro sites that operate in cooperation with smaller base stations, and may also employ various radio technologies to obtain better coverage and enhanced data rates.

[0082] 5G networks may utilize network function virtualization (NFV), which is a network architecture concept that proposes virtualizing network node functions into "building blocks" or entities that can be operationally connected or linked together to provide services. Virtualized network functions (VNFs) may include one or more virtual machines that use standard or common types of servers rather than custom hardware to run computer program code. Cloud computing or data storage may also be utilized. In radio communications, this may mean performing node operations at least in part in a server, host, or node that is operably coupled to a remote radio head. Node operations may also be distributed among multiple servers, nodes, or hosts. It should also be understood that the distribution of labor between core network operations and base station operations may be different from LTE or even non-existent.

[0083] Figure 1 A schematic diagram of a 5G system (5GS) 100 is shown. The 5GS may include a user equipment (UE) 102 (which may also be referred to as a communication device or terminal), a 5G radio access network (5GRAN) 104, a 5G core network (5GCN) 106, one or more application functions (AFs) 108, and one or more data networks (DNs) 110.

[0084] An example 5G core network (CN) includes functional entities. The 5G CN 106 may include one or more access and mobility management functions (AMFs) 112, one or more session management functions (SMFs) 114, an authentication server function (AUSF) 116, a unified data management (UDM) 118, one or more user plane functions (UPFs) 120, a unified data repository (UDR) 122, and / or a network exposure function (NEF) 124. The UPF is controlled by the SMF (session management function) which receives policies from the PCF (policy control function).

[0085] The CN is connected to the terminal device via a radio access network (RAN). The 5G RAN may include one or more gNodeB distributed unit functions connected to one or more gNodeB (GNB) centralized unit functions. The RAN may include one or more access nodes.

[0086] The UPF (User Plane Function) whose role is called PSA (Protocol Data Unit (PDU) Session Anchor) may be responsible for forwarding frames to and from the UE between the DN (Data Network) and the tunnel established over 5G towards the UE exchanging traffic with the DN.

[0087] Now refer to Figure 2 describing in more detail possible mobile communication devices, Figure 2A schematic partial cutaway view of a communication device 200 is shown. Such a communication device is generally referred to as a user equipment (UE) or terminal. Suitable mobile communication devices may be provided by any device capable of sending and receiving radio signals. Non-limiting examples include a mobile station (MS) or mobile device, such as a mobile phone or so-called "smart phone", a computer provided with a wireless interface card or other wireless interface facilities (such as a USB dongle), a personal data assistant (PDA) or a tablet computer provided with a wireless communication function, a voice over IP (VoIP) phone, a portable computer, a desktop computer, an image capture terminal device such as a digital camera, a game terminal device, a music storage and playback device, a vehicle-mounted wireless terminal device, a wireless endpoint, a mobile station, a laptop embedded device (LEE), a laptop mounted device (LME), an intelligent device, a wireless customer premises equipment (CPE), or any combination of these devices, etc. The mobile communication device may provide, for example, data communications for communications such as carrying voice, email (email), text messages, multimedia, etc. Therefore, users may obtain and provide a variety of services via their communication devices. Non-limiting examples of these services include two-way or multi-way calls, data communications or multimedia services, or simply access to a data communications network system such as the Internet. Broadcast or multicast data may also be provided to the user. Non-limiting examples of content include downloads, television and radio programs, videos, advertisements, various alerts and other information.

[0088] The mobile device is typically provided with at least one data processing entity 201, at least one memory 202 and possibly other components 203 for software and hardware assistance in performing the tasks it is designed to perform, including controlling access to and communications with access systems and other communication devices. Data processing, storage and other related control devices may be provided on appropriate circuit boards and / or chipsets. This feature is indicated by reference numeral 204. The user may control the operation of the mobile device by means of a suitable user interface such as a keyboard 205, voice commands, a touch-sensitive screen or touchpad, a combination thereof, etc. A display 208, a speaker and a microphone may also be provided. In addition, the mobile communication device may include appropriate connectors (wired or wireless) to other devices and / or for connecting external accessories (e.g., a hands-free device).

[0089] The mobile device 200 may receive signals over the air or radio interface 20 via appropriate means for receiving signals, and may transmit signals via appropriate means for transmitting radio signals. Figure 2 In the mobile device, the transceiver arrangement is schematically represented by a block 206. The transceiver device 206 may be provided, for example, by means of a radio component and an associated antenna arrangement. The antenna arrangement may be arranged internally or externally to the mobile device.

[0090] Figure 3An example of a control device 300 for a communication system is shown, which is for example coupled to and / or used to control a station of an access system such as a RAN node, such as a base station, eNB or gNB, a relay node or a core network node such as an MME or S-GW or P-GW, or a core network function such as an AMF / SMF, or a server or a host. The method can be implemented in a single control device or across more than one control device. The control device can be integrated with a node or module of a core network or RAN or located outside it. In some embodiments, the base station includes a separate control device unit or module. In other embodiments, the control device can be another network element, such as a radio network controller or a spectrum controller. In some embodiments, each base station can have such a control device and a control device provided in a radio network controller. The control device 300 can be arranged to provide control of communications in a service area of ​​the system. The control device 300 includes at least one memory 301, at least one data processing unit 302, 303 and an input / output interface 304. Via the interface, the control device can be coupled to a receiver and a transmitter of a base station. The receiver and / or transmitter can be implemented as a radio front end or a remote radio head.

[0091] The technology, called DeepRx (Deep Learning Receiver), aims to enhance the capabilities of the physical layer (PHY) using customized AI / ML techniques. DeepRx studies based on both simulation and real-world data have shown that under the same radio frequency (RF) channel conditions, the performance gain compared to conventional methods in terms of reducing transmission errors is non-negligible. At the same time, superiority has also been shown in terms of robustness against different channel conditions and Doppler effects.

[0092] Figure 4 A conventional uplink (UL) receiver architecture is depicted with multiple digital signal processing (DSP) blocks 402, which include channel estimation, equalization, and demodulation. DeepRx 403 can be used to replace DSP blocks, such as Figure 4 In other words, the demapped frequency-time domain data in the user plane, and key messages from the scheduler 401 such as the modulation and coding scheme (MCS), physical resource block (PRB), and rank in the radio resource management (RRM) plane will become the input of DeepRx 403. The output of DeepRx 403 can be fed to the traditional de-precoding function module.

[0093] Figure 5An example of the DeepRx architecture for an Orthogonal Frequency Division Multiple Access (OFDM) system is shown. The architecture includes a deep convolutional neural network (CNN) 503 structured receiver designed for OFDM signal detection. DeepRx 503 takes as input the frequency-time domain resource elements of the received data (after Fast Fourier Transform (FFT)) and the corresponding raw channel estimate, and outputs the bit likelihood ratios (LLRs) for decoding. By exploiting the frequency-time domain correlation and the distribution pattern of the transmitted codewords, DeepRx 503 can surpass conventional receiver schemes, such as those based on conventional LMMSE receivers.

[0094] In uplink transmission, discrete Fourier transform (DFT) spread OFDM (DFTS-OFDM) has been adopted to improve the power amplifier (PA) efficiency on the device side. It is designed for coverage-challenged situations and supports flexible discrete Fourier transform (DFT) size in the implementation. DFTS-OFDM must be supported in the device, and so the network can configure DFT precoding for a specific device if necessary.

[0095] Figure 6 The transmitter (Tx) and receiver (Rx) signal processing chains for an OFDM system and a DFTS-OFDM system are shown. The DFTS-OFDM Tx signal processing chain includes DFT precoding, and the DFTS-OFDM Rx signal processing chain includes inverse DFT (iDFT) deprecoding.

[0096] As described in the 3GPP specifications, the UE determines whether transform (DFT) precoding is applied, and the UE shall treat transform (DFT) precoding as "enabled" or "disabled" according to the parameters configured by higher layers.

[0097] Therefore, compatibility with DFTS-OFDM systems is another requirement for DeepRx implementation, especially when it is deployed at the base station (BS) side.

[0098] DeepRx has shown success on single-input single-output (SISO) or multiple-input multiple-output (MIMO) OFDM systems. However, when moving to single-carrier OFDM scenarios (e.g., DFTS-OFDM), non-negligible performance degradation or difficulty in DeepRx training has been observed.

[0099] In other words, the current DeepRx is not compatible with the DFTS-OFDM format, which, as mentioned above, is a must-have feature for 5G and above devices. The incompatibility of DeepRx with the DFTS-OFDM system may be due to the fact that the CNN-structured DeepRx may not be able to learn to mimic the iDFT de-precoding operation that spans the entire OFDM frame ( Figure 6 Therefore, a customized system design is required to make DeepRx compatible with the DFTS-OFDM waveform.

[0100] Figure 7 A flow chart of a method according to an example embodiment is shown. The method may be performed at a receiver.

[0101] In S1 , the method comprises receiving an input signal based on a frequency domain signal.

[0102] In S2, the method comprises performing de-precoding on a received input signal.

[0103] In S3, the method includes providing the de-precoded signal as an input to a machine learning model, wherein an output of the machine learning model is a likelihood ratio.

[0104] The frequency domain signal may be a single carrier frequency domain signal or a DFTS-OFDM signal.

[0105] Signal deprecoding is the inverse operation of signal precoding. If signal precoding is achieved with the help of DFT precoding, then iDFT deprecoding will be the inverse operation. Deprecoding can be performed by the iDFT deprecoding function.

[0106] The method can provide two solutions to enhance the compatibility of DeepRx and DFTS-OFDM signal detection.

[0107] In a first example embodiment, the input signal is a frequency domain signal. The method may include performing de-precoding on a data branch of the input signal. The data branch may include frequency-time grid symbol N f ×N t , where N f Including the number of subcarriers, and N t Including the number of orthogonal frequency division multiple access (OFDM) codewords. The method may include cascading the de-precoded signal with the original channel estimation branch of the frequency domain signal, and using the cascaded signal as an input to the machine learning model. In this example embodiment, the machine learning model may be referred to as DeepRx.

[0108] In the first option, where de-precoding is performed on the frequency domain signal, a pre-performed de-precoding function is introduced on the received data branch before being concatenated with the original channel estimation branch, which is then fed to the conventional DeepRx.

[0109] Figure 8 An example structure according to the first option is shown. The input of DeepRx 803 is kept the same as Figure 5 802, i.e., the data branch and the original estimation branch, where the data branch is the frequency-time resource element after FFT, and the original estimation branch is the original estimated channel response based on sparse pilots. An iDFT de-precoding function 802 is introduced to be applied to the received data branch, which acts as a pre-de-precoding operation before the data is concatenated with the lower branch (the original channel estimation branch) and then fed to DeepRx 803.

[0110] Fig. 9 Shows Figure 8 A more detailed structure of the example shown in .

[0111] Taking 1T1R transmission as an example, the input data includes branch 1 and branch 2, which include the received codewords after FFT, and FFT converts them into N-sized codewords. f ×N t The frequency-time grid symbol, where N f refers to the number of subcarriers, and N t Refers to the number of OFDM symbols in a time slot (defined as 14 according to 3GPP). Branches 1 and 2 are the real and imaginary parts of the received symbol, respectively. Resource elements (REs) occupied by DMRS are set to "0".

[0112] The input data also includes branches 3 and 4, which include the original channel estimation branches, where non-demodulation reference signal (DMRS) resource elements (REs) are set to "0". A branch size is also N f ×N t . Branches 3 and 4 are also the real and imaginary parts respectively.

[0113] All branches are cascaded before being fed to the neural network (NN) 803.

[0114] The output data includes dimensions N mod ×N f ×N t bit LLR, where "N mod ” represents the number of bits per RE and depends on the modulation scheme.

[0115] In the training phase, the output will be used to calculate the loss function with the transmitted bits for model training. In the inference phase (application), the output will be fed to the decoder.

[0116] De-precoding is applied only to the received symbols, which are then separated into real and imaginary parts. De-precoding is a fixed operation following the normal operation in the frequency domain.

[0117] exist Fig. 9 In the example structure shown in , NN structure 803 (also referred to as DeepRx) includes 2 convolutional layers and 11 ResNet blocks in between. Any other suitable NN structure may be used.

[0118] In a second example embodiment, the method may include providing the input signal to another machine learning model prior to de-precoding to process the input signal and perform de-precoding on the output of the another machine learning model. In this embodiment, the input signal is a processed frequency domain signal (i.e., a frequency domain signal that has been processed by another machine learning model).

[0119] In this example embodiment, the machine learning model may be referred to as PostDeepRx, and another machine learning model may also be referred to as DeepRx.

[0120] In a second example embodiment, a cascaded design may be provided with a DeepRx-Deprecoding-PostDeepRx structure, where the functional block of the deprecoding is embedded between DeepRx and PostDeepRx whose parameters are jointly learned and optimized.

[0121] In this example embodiment, the output of another machine learning model is 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

[0122] Outputs of another machine learning model may also include constellation monitoring outputs.

[0123] The constellation monitoring output can be provided for further analysis. If the input and output of the iDFT de-precoding block are well designed, the output of the iDFT de-precoding has semantic meaning. Through further expert or machine-type analysis, the hardware and channel conditions and even DeepRx itself can be understood. In other words, this is a step towards making the DeepRx receiver easily interpretable and therefore more trustworthy.

[0124] Fig.10The overall structure of an example according to the second option is shown. The de-precoding function block 1002 is embedded between DeepRx 1003b and PostDeepRx 1003a. The data format of the output of DeepRx 1003b and the input size of PostDeepRx 1003a is designed to be 2×N f ×N t , where N f and Nt denote the iDFT size and the number of OFDM symbols, respectively. Dimension 2 refers to the real and imaginary parts of the equalized signal.

[0125] Fig.11 Shows Fig.10 A more detailed structure of the overall structure in .

[0126] As described with reference to the first exemplary embodiment, the input data in this exemplary embodiment comprises a data branch and an original estimation branch. However, in this example, no de-precoding is applied to the received symbols.

[0127] The output data of the machine learning model 1003a includes a dimension of N mod ×N f ×N t ” bit LLR, and the output of another machine learning model (1003b) includes a dimension of 2×N f ×N t Post-equalization constellation-like mode.

[0128] In the training phase, the bit LLR outputs are used to compute the loss function with the transmitted bits used for computational model training. In the inference phase (application), the bit LLR outputs will be fed to the decoder and the constellation-like patterns can be recorded for further analysis.

[0129] In a second example embodiment, a de-precoding function 1002 is embedded between NNs, as described below.

[0130] Embedded de-precoding is a fixed operation and is not learnable in back-propagation. PreDeepRx and PostDeepRx can be trained together.

[0131] In a second exemplary embodiment, DeepRx includes two separate NNs, such as Fig.10 As shown in .

[0132] The first NN is called DeepRx NN 1003b, which includes 2 convolutional layers and 11 ResNet blocks. The number of cores of the output convolutional layer is set to 2, and the output data dimension is 2×N f ×N t .

[0133] The second NN is called PostDeepRx 1003a, which includes 2 convolutional layers and 1 ResNet in between. The number of cores in the input convolutional layer is 2, and the input data dimension is 2×N f ×N t The number of cores in the output convolutional layer depends on the modulation scheme.

[0134] Any other suitable NN structures may be used instead of the examples described with reference to the drawings.

[0135] The output of DeepRx 1003b and the input of PostDeepRx 1003a (2×N f ×N t ) design can have two advantages.

[0136] One is to force the output of DeepRx 1003b to comply with iDFT de-precoding, whose output can be sequentially coupled to PostDeepRx 1003a. Thus, DeepRx 1003b opens a "window" for constellation monitoring at the output of iDFT de-precoding block 1002. The intermediate constellation derived at the second output branch can be perceived by human or machine type analysis for channel condition (e.g., nonlinearity, Doppler) monitoring.

[0137] Another is to perform PostDeepRx to learn nonlinear demodulation detection and let DeepRx act as a joint nonlinear channel estimator and equalizer. Therefore, to some extent, it allows the fine design of DeepRx and PostDeepRx based on the intermediate results of the learned constellation pattern.

[0138] In both the first and second exemplary embodiments described above, binary cross entropy may be used as the loss function. The formula is:

[0139]

[0140] where y i represents the binary bit 0 / 1, and P(y i ) is the corresponding probability calculated by feeding the output LLR into the sigmoid function. N is the total number of bits after channel coding.

[0141] During training, non-data RE locations are ignored in back-propagation.

[0142] The de-precoding block is fixed and is not learned when training the two ML models. The machine learning models can be trained in online mode or offline mode. Offline training can be performed on other devices where the model is not deployed.

[0143] Online training requires signaling and actions between the transmitter and receiver.

[0144] The equipment used to train the model may include a receiver (performs dataset construction, data processing, model training, updating, deployment), a transmitter (prepare training data and send it over the air), and functional units in other platforms that can perform all related processes.

[0145] The method may include causing a trigger to be sent to a device (e.g., a transmitter such as a user device) for training a machine learning model (and, if present, another machine learning model) and receiving an acknowledgement from the device in response to the request.

[0146] The method may include providing training configuration information (such as, but not limited to, carrier frequency, subcarrier number, MCS, rank, transform precoding enable indication (DFT precoding), sequence random seed, and number of transmission time intervals (TTIs)) to the device, receiving training data from the device in response to the configuration information, and providing the training data to an entity for training the machine learning model (and, if present, another machine learning model). The entity for training the machine learning model and the other machine learning model may be located at a receiver, another base station such as a gNB), or a cloud. The entity for training the machine learning model and the other machine learning model may be a local data pool and computing unit (LDP&CU).

[0147] Fig.12 shows an example signaling diagram for training a machine learning model in online mode.

[0148] In a first step 1201, a receiver (eg, a BS for uplink transmission) triggers online training or fine-tuning.

[0149] In a second step 1202, the transmitter responds with an acknowledgement of the training request.

[0150] In the third step 1203, the receiver sends a training configuration, which may include carrier frequency, subcarrier number, MCS, rank, transform precoding enable indication (DFT precoding), sequence random seed, number of TTIs required for training, etc.

[0151] In a fourth step 1204, the transmitter sends a training sequence over the air. The receiver should know the training sequence to calculate the loss.

[0152] In a fifth step 1205, the receiver sends the data to the local data pool and computing unit (LDP&CU).

[0153] In a sixth step 1206, the model is trained in the computational unit using the training data in the LDP.

[0154] In a seventh step 1207, the CU sends the trained model to the receiver user plane.

[0155] In an eighth step 1208, the receiver deploys the model.

[0156] Model updating and fine-tuning can be considered as part of the online training process, which can be triggered by events periodically and aperiodically. The steps of model updating and fine-tuning are as described above. Fig.12 described.

[0157] The receiver side may determine to deploy the machine learning model. For example, the method may include receiving deployment configuration information from a transmitter of an input signal, and determining whether to deploy the machine learning model (and if present, another machine learning model) based on the deployment configuration information.

[0158] In an example embodiment, in order to apply a machine learning model to wireless transmission, the transmitter notifies the receiver of the corresponding configuration. Before using DeepRx for signal processing, the receiver should determine to activate the corresponding functional block (such as a de-precoding block) and select the corresponding neural network (NN) based on the configuration.

[0159] The deployment configuration information may include at least one of an indication to perform de-precoding on a frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

[0160] Fig.13 shows the signaling diagram for the DeepRx example deployment.

[0161] In step 1, 1301, the transmitter sends a configuration including transform precoding activation information, subcarrier number, rank, MCS, etc. to the receiver.

[0162] In step 2, 1302, the receiver schedules whether DeepRx should be used.

[0163] If no, go back to the normal method.

[0164] If yes, go to step 3.

[0165] In step 3, 1303, based on the configuration, the receiver determines to enable the de-precoding function block and selects a network for the current rank, MCS, etc.

[0166] In step 4, 1304, the transmitter sends user data to the receiver.

[0167] In step 5, 1305, the receiver processes the data using DeepRx.

[0168] In order to schedule DeepRx, the RRM scheduler may inform the user plane of the receiver of information for DeepRx, which includes DFT precoding enabling information, PRB number, rank number, MCS, backoff indicator, etc.

[0169] The DeepRx model scheduling for the first and second options are as follows: Fig.14 and 15 As shown in . The scheduler 1401 notifies the de-precoding block 1402 in two proposed schemes. Fig.14 and 15 As shown in , the user plane signal processing block can be scheduled by the RRM plane scheduler 1401 using the following information, including: transform precoding (TP) enable information to activate the flexible de-precoding function block; subcarrier number or PRB number for flexible de-precoding function (iDFT size); rank number (currently uplink DFTS-OFDM transmission in 3GPP only supports rank 1); information on at least one of the MCSs to determine the output core dimension of DeepRx and a fallback indicator to inform the receiver whether it should fall back to the conventional method or apply DeepRx.

[0170] In the first step, the RRM scheduler 1401 informs DeepRx whether DFT precoding is enabled. In the second step, if DFT precoding is enabled, the deprecoding block 1402 of DeepRx is activated, and if not, the deprecoding function block 1402 is deactivated.

[0171] The scheduler may also schedule model switching between OFDM and DFTS-OFDM in corresponding scenarios, and model fallback to a conventional receiver. The method may include determining not to deploy a machine learning model (and if present, not to deploy another machine learning model) based on a fallback indicator. For example, in the case where the performance of DeepRx may deteriorate, such as continuous errors in transmission, etc., the receiver side may trigger a fallback to a conventional receiver.

[0172] Falling back to a regular receiver model can also trigger model training (as mentioned above Fig.12 updating or fine-tuning as described).

[0173] The receiver determines whether to fall back to the normal method.

[0174] If the receiver decides to fall back to the normal method, the RRM plane scheduler notifies the user plane to fall back to the normal method, and the receiver applies the normal method. If not, the RRM plane scheduler notifies the user plane to apply DeepRx, and the receiver applies DeepRx (i.e. deploys the machine learning model).

[0175] Both proposed schemes were tested in real-life proof-of-concept scenarios and the results showed the benefits of the invented receiver architecture in real-world situations. Based on these proof-of-concept evaluations, the second option achieved faster convergence in training.

[0176] The proposed DeepRx scheme is evaluated using data from a testbed for wireless transmission. Hardware nonlinearities are present and DFT precoding is enabled during the data collection phase to reduce the PAPR in transmission. The Tx signal propagates in the air and is captured by the receiver at a distance of about 4 feet.

[0177] The Tx / Rx antenna number is 1. Two DMRS symbols are used to estimate the channel. The transmit modulation scheme is 64QAM. Table 1 shows the list of configurations under evaluation.

[0178] Table 1

[0179]

[0180]

[0181] The proposed schemes are tested with DFT precoding enabled and compared with a conventional LMMSE receiver in a 25RB (300 subcarriers) configuration with coding rate 873 / 1024. There are 4000 subframes of data for training and 1000 subframes for inference.

[0182] Figures 16 to 18 The validation results from the test data are shown (to compare the performance on non-pretrained data). The uncoded and coded BER curves obtained by evaluation are shown in Figure 2. Fig.16 and Fig.17 The top curve shows the BER using a conventional LMMSE receiver. The following curves show the performance using Pre-iDFT de-precoded DeepRx, and the bottom curve is obtained from iDFT de-precoded Embedded Hybrid DeepRx.

[0183] The x-axis is E RE / N definition, where E RE is the denoised signal power estimate, and N is the noise estimate. During the training phase, only E RE The data was taken from an ERE / N value of 23.15 dB (the last ERE / N value in the figure), while in the test phase, data from more ERE / Ns were used for inference.

[0184] It can be clearly observed that at an uncoded BER level of 10%, the first option DeepRx shows a gain of 0.6 dB over the practical LMMSE receiver, while the second option DeepRx outperforms the conventional LMMSE receiver by about 1.2 dB. RE / N, the uncoded BER is reduced to 1 / 10. In the coded BER curve, the first option shows a ~1dB gain over the LMMSE possible, while the second option DeepRx can achieve a gain of >2dB at a 100% coded BER level. RE / N, error-free performance is achieved, while the LMMSE performance remains at the level of 1% to 2%.

[0185] like Fig.18 As shown in the figure, the second option also has faster convergence during training. It can be seen that the second option converges faster and has lower loss / BER levels than the first option, while using the previous DeepRx, the NN is not improved by training.

[0186] like Fig.19 The constellation-like pattern 1900 shown in FIG. 1 can be recorded from the second output after the embedded de-precoding layer from the iDFT embedded hybrid DeepRx solution. It can be seen that the modulation scheme used is 64QAM, but with some rotation. Fig.19 The expanded shape marked by the black outline 1901 reflects the nonlinearity caused by the test bench hardware power amplifier. Such a constellation-like pattern 1900 can reflect the possible impairments in the transmission process. This also helps to understand the underlying mechanism of DeepRx.

[0187] refer to Figures 7 to 15 The method and apparatus described enhance the compatibility of DeepRx with DFTS-OFDM systems. For DFTS-OFDM, much higher gains than conventional LMMSE receivers can be achieved, for example, a 1% BER using a conventional LMMSE receiver can be reduced to error-free by the proposed DeepRx method.

[0188] The second example embodiment may provide a second constellation-like output that may reveal nonlinearities in hardware etc. and make DeepRx easier to interpret.

[0189] An apparatus for a receiver may include a device for receiving an input signal based on a frequency domain signal, a device for performing precoding on the received input signal, and a device for providing a de-precoded signal as an input to a machine learning model, wherein the output of the machine learning model is a likelihood ratio.

[0190] It should be understood that these devices may include or be coupled to other units or modules, such as radio components or radio heads for transmission and / or reception. Although these devices are described as one entity, different modules and memories may be implemented in one or more physical or logical entities.

[0191] It should be noted that although some embodiments have been described with respect to 5G networks, similar principles may also be applied to other networks and communication systems. Thus, although certain embodiments are described above by way of example with reference to certain example architectures for wireless networks, technologies and standards, the embodiments may be applied to any other suitable form of communication system other than that shown and described herein.

[0192] It is also noted herein that while the above describes various exemplifying embodiments, there are numerous variations and modifications which may be made to the disclosed solution without departing from the scope of the present invention.

[0193] In general, various embodiments may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects of the disclosure may be implemented in hardware, while other aspects may be implemented in firmware or software, which may be executed by a controller, microprocessor, or other computing device, but the disclosure is not limited thereto. Although various aspects of the disclosure may be shown and described as block diagrams, flow charts, or using some other graphical representation, it is well understood that, as non-limiting examples, the blocks, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0194] As used in this application, the term "circuitry" may refer to one or more or all of the following:

[0195] (a) hardware circuit implementations only (such as implementations in analog and / or digital circuits only) and

[0196] (b) a combination of hardware circuitry and software, such as (where applicable):

[0197] (i) a combination of analog and / or digital hardware circuitry and software / firmware, and

[0198] (ii) any portion of a hardware processor and software (including a digital signal processor), software and memory that work together to enable a device (such as a mobile phone or server) to perform various functions, and

[0199] (c) Hardware circuits and / or processors, such as a microprocessor or portion of a microprocessor, that require software (e.g., firmware) to operate, but where the software may not be present, the software may not be required for operation.

[0200] This definition of circuitry applies to all uses of the term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers implementations of only a hardware circuit or processor (or multiple processors) or a portion of a hardware circuit or processor and its (or their) accompanying software and / or firmware. The term circuitry also covers, for example and if applicable to a particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in a server, cellular network device, or other computing or network device.

[0201] The various embodiments of the present disclosure may be implemented by computer software that can be executed by a data processor of a mobile device (such as in a processor entity), or by hardware, or by a combination of software and hardware. Computer software or programs, also referred to as program products, include software routines, applets and / or macros that can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer executable components that are configured to execute the various embodiments when the program is run. One or more computer executable components may be at least one software code or a portion thereof.

[0202] Also in this regard, it should be noted that any blocks of the logic flow as shown in the figures may represent program steps, or interconnected logic circuits, blocks and functions, or a combination of program steps and logic circuits, blocks and functions. The software may be stored on physical media such as memory chips or memory blocks implemented within a processor, on magnetic media such as hard disks or floppy disks, and on optical media such as, for example, DVDs and their data variants, CDs. Physical media are non-transitory media.

[0203] The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology such as semiconductor-based memory devices, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The data processor may be of any type suitable for the local technical environment and may include, as non-limiting examples, one or more of a general purpose computer, a special purpose computer, a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an FPGA, a gate level circuit, and a processor based on a multi-core processor architecture.

[0204] Embodiments of the present disclosure may be implemented in various components such as integrated circuit modules. The design of integrated circuits is generally a highly automated process. Complex and powerful software tools are available to convert logic level designs into semiconductor circuit designs for etching and formation on semiconductor substrates.

[0205] The independent claims set out the scope of protection sought by various embodiments of the present disclosure. Embodiments and features described in this specification that do not belong to the scope of the independent claims (if any) are to be construed as examples that help understand the various embodiments of the present disclosure.

[0206] The above description has provided a complete and informative description of exemplary embodiments of the present disclosure by way of non-limiting examples. However, various modifications and adaptations may be apparent to those skilled in the art in view of the above description when read in conjunction with the accompanying drawings and the appended claims. However, all such and similar modifications of the teachings of the present disclosure will still fall within the scope of the present invention as defined in the appended claims. In fact, there are also embodiments that include a combination of one or more embodiments with any other embodiments discussed previously.

Claims

1. A device for a receiver, the device comprising: means for receiving an input signal based on a frequency domain signal; means for performing de-precoding on said received input signal; as well as Means for providing the de-precoded signal as input for a machine learning model, wherein the output of the machine learning module is a likelihood ratio.

2. The device according to claim 1, wherein: The means for performing de-precoding comprises an inverse discrete Fourier transform (iDFT) de-precoding function.

3. An apparatus according to claim 1 or 2, comprising means for performing said de-precoding on a data branch of said input signal.

4. The device according to claim 3, wherein: The data branch includes frequency-time grid symbols N f ×N t , where N f Including the number of subcarriers, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

5. The device according to claim 3 or 4 comprises a device for cascading the de-precoded signal with the original channel estimation branch of the frequency domain signal, and using the cascaded signal as an input of the machine learning model.

6. The device according to claim 1 or 2, wherein: The device includes a device for providing the input signal to another machine learning model for processing the input information before de-precoding, and a device for performing de-precoding on the output of the other machine learning model.

7. The device according to claim 6, wherein: The output of the other machine learning model is 2×N f ×N t , where N f Including iDFT size, and N t Includes the number of Orthogonal Frequency Division Multiple Access (OFDM) symbols.

8. The device according to claim 7, wherein: The output of the another machine learning model also includes a constellation monitoring output.

9. The device according to any one of claims 1 to 8, comprising: means for causing a trigger to be sent to a device for training the machine learning model and, if present, another machine learning model; as well as means for receiving an acknowledgement from the device in response to the request.

10. The device according to claim 9, comprising means for providing training configuration information to said device; means for receiving training data from the device in response to the configuration information; as well as Means for providing the training data to an entity used to train the machine learning model and, if present, to train another machine learning model.

11. An apparatus according to any one of claims 1 to 10, comprising a device for receiving deployment configuration information from a transmitter of the input signal; and a device for determining whether to deploy the machine learning model and whether to deploy the other machine learning model if any based on the deployment configuration information.

12. The device according to claim 11, wherein The deployment configuration information includes at least one of an indication of performing the de-precoding on the frequency domain signal, subcarrier number information, a modulation and coding scheme, a rank, and a backoff indicator.

13. The apparatus of claim 12, comprising means for determining not to deploy the machine learning model based on the fallback indicator, and for not deploying the other machine learning model if present.

14. A method comprising, at a receiver: receiving an input signal based on the frequency domain signal; performing de-precoding on the received input signal; and The de-precoded signal is provided as an input to a machine learning model, wherein an output of the machine learning model is a likelihood ratio.

15. A device comprising: at least one processor and at least one memory including computer program code configured to, with the at least one processor, cause the apparatus at a receiver to at least: receiving an input signal based on the frequency domain signal; performing de-precoding on the received input signal; as well as The de-precoded signal is provided as an input to a machine learning model, wherein an output of the machine learning model is a likelihood ratio.

16. A computer-readable medium comprising program instructions for causing an apparatus to perform at least the following operations at a receiver: receiving an input signal based on the frequency domain signal; performing de-precoding on the received input signal; and The de-precoded signal is provided as an input to a machine learning model, wherein an output of the machine learning model is a likelihood ratio.

Images