Apparatus, method and computer program product for communication

Through neural network decoding and detection, the problem of noise damage control information and discontinuous transmission in cellular communication systems is solved, efficient control information decoding and transmission detection is realized, and the performance and stability of the system are improved.

CN113273108BActive Publication Date: 2024-11-08NOKIA TECHNOLOGIES OY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201980088250.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2019-01-07
Publication Date
2024-11-08
Estimated Expiration
2039-01-07

AI Technical Summary

Technical Problem

In modern cellular communication systems, control information is susceptible to noise during transmission, resulting in increased difficulty in decoding errors and discontinuous transmission detection, and it is difficult for the prior art to effectively deal with noise-destructive control information and discontinuous transmission.

Method used

A neural network is used to decode control information elements and detect discontinuous transmission. By training the neural network to simulate noise-destructed control information and discontinuous transmission data, the accurate decoding of control information in the received frame and the detection of discontinuous transmission.

Benefits of technology

It improves the decoding accuracy of control information and the efficiency of discontinuous transmission detection, reduces the computational complexity and delay, and maintains good performance under different channels and noise conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113273108B_ABST
    Figure CN113273108B_ABST
Patent Text Reader

Abstract

The present disclosure relates to apparatus, methods and computer program products for communication. A solution for employing a neural network in a radio receiver is provided. According to one aspect, the method comprises: obtaining a sample set associated with a position of a control information element in a received frame from a received frame; inputting the sample set to a node of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; processing the sample set in a trained neural network before receiving the sample set to decode one or more determined values ​​of the control information element and detect discontinuous transmission; outputting an indicator indicating a decoded value of a control information element included in the sample set, or an indicator indicating discontinuous transmission, in an output layer of the neural network.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to communications in wireless communication systems, and in particular to detecting control information transmitted in uplink or downlink frames. Background Art

[0002] The control information may carry various information related to radio link management. In modern cellular communication systems, the control information may include, for example, an acknowledgement message (ACK / NAK) confirming the reception of a frame, channel quality information (CQI) reporting the quality of a radio channel of a radio link, a rank indicator (RI) indicating the rank of a radio channel, etc. The control information is typically encoded in a transmitter and decoded in a receiver by mapping received control information symbols corrupted by noise in a radio channel to the closest codeword using, for example, a maximization likelihood criterion. Summary of the invention

[0003] According to an aspect, the subject matter of the independent claims is provided. Some embodiments are defined in the dependent claims.

[0004] According to one aspect, an apparatus is provided, comprising a component for performing: obtaining a sample set associated with a position of a control information element in a received frame; inputting the sample set to nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; processing the sample set in a trained neural network before receiving the sample set to decode one or more determined quantities of the control information element and detect discontinuous transmission; and outputting an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating discontinuous transmission in an output layer of the neural network.

[0005] In an embodiment, the output layer includes a number of nodes equal to the possible outputs of the neural network.

[0006] In an embodiment, the component is configured to output a non-zero value in only one of the nodes of the output layer and to output a zero value in the other one or more nodes of the output layer.

[0007] In an embodiment, the set of samples includes modulation symbol samples, and wherein the neural network is configured to decode one or more determined values ​​of a control information element from the modulation symbol samples.

[0008] In an embodiment, the neural network includes only two layers between an input layer and an output layer: a first layer is coupled to the output of the input layer and includes a number of nodes equal to the number of nodes in the input layer; and a second layer is coupled between the first layer and the output layer and includes a number of nodes equal to the number of nodes in the output layer.

[0009] In an embodiment, the neural network is fully connected.

[0010] In an embodiment, the control information element has one bit or two bits.

[0011] In an embodiment the control information element is an acknowledgement message or a rank indicator confirming the reception of the frame.

[0012] In an embodiment, the neural network is trained via training inputs to the neural network, the training inputs comprising all possible values ​​of the control information elements as corrupted by simulated noise, wherein the simulated noise simulates one or more radio channel models, and the training inputs further comprising discontinuous transmission training inputs.

[0013] In an embodiment, a component comprises: at least one processor; and at least one memory including computer program code, the at least one memory and the computer program code being configured to, with the at least one processor, cause execution of the apparatus.

[0014] According to another aspect, a method is provided, comprising: obtaining, by a wireless device, a sample set associated with a position of a control information element in a received frame; inputting, by the wireless device, the sample set to nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; processing, by the wireless device, the sample set in a trained neural network before receiving the sample set to decode one or more determined quantities of the control information element and detect discontinuous transmission; and outputting, by the wireless device, in an output layer of the neural network, an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating discontinuous transmission.

[0015] In an embodiment, the output layer includes a number of nodes equal to the possible outputs of the neural network.

[0016] In an embodiment, the neural network outputs a non-zero value in only one of the nodes of the output layer and outputs a zero value in the other one or more nodes of the output layer.

[0017] In an embodiment, the set of samples includes modulation symbol samples, and wherein the neural network decodes one or more determined values ​​of a control information element from the modulation symbol samples.

[0018] In an embodiment, the neural network includes only two layers between an input layer and an output layer: a first layer is coupled to the output of the input layer and includes a number of nodes equal to the number of nodes in the input layer; and a second layer is coupled between the first layer and the output layer and includes a number of nodes equal to the number of nodes in the output layer.

[0019] In an embodiment, the neural network is fully connected.

[0020] In an embodiment, the control information element has one bit or two bits.

[0021] In an embodiment, the control information element is an acknowledgement message or a rank indicator confirming the reception of the frame.

[0022] In an embodiment, the neural network is trained by training inputs to the neural network, the training inputs comprising all possible values ​​of the control information elements as corrupted by simulated noise, and wherein the simulated noise simulates one or more radio channel models, the training inputs further comprising discontinuous transmission training inputs.

[0023] According to another aspect, a computer program product is provided, which is implemented on a computer-readable medium and includes a computer program code readable by a computer, wherein the computer program code configures a computer to perform a computer process, including: obtaining a sample set associated with a position of a control information element in a received frame; inputting the sample set to nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; before receiving the sample set, processing the sample set in a trained neural network to decode one or more determined values ​​of the control information element and detect discontinuous transmission; and outputting an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating discontinuous transmission in an output layer of the neural network.

[0024] One or more examples of implementations are set forth in more detail in the drawings and the description that follows. Other features will be apparent from the description and drawings, and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Some embodiments will be described below with reference to the accompanying drawings, in which

[0026] Figure 1 illustrates an example of a wireless network in which embodiments of the present invention may be applied;

[0027] Figure 2 illustrates the transmission of a control information element from a sender to a receiver;

[0028] Figure 3 An embodiment for jointly performing decoding of control information elements and detection of discontinuous transmissions is illustrated;

[0029] Figure 4 illustrates a procedure for processing a control information element in a transmitter;

[0030] Figure 5 illustrates a neural network configured to jointly perform decoding of control information elements and detection of discontinuous transmissions according to some embodiments of the present invention;

[0031] Fig. 6A and Figure 6B illustrates a process for training a neural network according to an embodiment of the present invention;

[0032] Figure 7 illustrates a process for testing a neural network according to an embodiment of the present invention; and

[0033] Figure 8 An apparatus according to an embodiment of the invention is illustrated. DETAILED DESCRIPTION

[0034] The following embodiments are exemplary. Although this specification may refer to "one", "an", or "some" (multiple) embodiments at several locations in the text, this does not necessarily mean that the same (multiple) embodiments are referred to each time, or that a particular feature only applies to a single embodiment. Individual features of different embodiments may also be combined to provide other embodiments.

[0035] In the following, different exemplary embodiments will be described using a radio access architecture based on Advanced Long Term Evolution (LTE Advanced, LTE-A) or New Radio (NR, 5G) as an example of an access architecture to which the embodiments may be applied, but the embodiments are not limited to such architectures. It is obvious to a person skilled in the art that the embodiments may also be applied to other kinds of communication networks with suitable components by appropriately adjusting parameters and procedures. Some examples of other options for suitable systems are Universal Mobile Telecommunications System (UMTS) Radio Access Network (UTRAN or E-UTRAN), Long Term Evolution (LTE, same as E-UTRA), Wireless Local Area Network (WLAN or WiFi), Worldwide Interoperability for Microwave Access (WiMAX), Personal Communications Service (PCS), Wideband Code Division Multiple Access (WCDMA), systems using Ultra-Wideband (UWB) technology, sensor networks, mobile ad-hoc networks (MANET), and Internet Protocol Multimedia Subsystem (IMS), or any combination thereof.

[0036] Figure 1 An example of a simplified system architecture is depicted showing only some elements and functional entities, which are all logical units, the implementation of which may differ from known implementations. Figure 1 The connections shown in are logical connections; the actual physical connections may be different. It is obvious to those skilled in the art that the system typically includes in addition to Figure 1 Other functions and structures than those shown in .

[0037] However, the embodiments are not limited to the systems given as examples, but a person skilled in the art may apply the solutions to other communication systems provided with the necessary properties.

[0038] Figure 1 The example of FIG. 1 shows parts of an exemplary radio access network.

[0039] Figure 1 User equipment 100 and 102 are shown, and user equipment 100 and 102 are configured to be wirelessly connected to an access node (such as (e / g) Node B) 104 providing the cell on one or more communication channels in the cell. The physical link from the user equipment to the (e / g) Node B is called an uplink or reverse link, and the physical link from the (e / g) Node B to the user equipment is called a downlink or forward link. It should be understood that the Node Bs or their functions can be implemented (e / g) using any node, host, server, or access point, etc., which is suitable for such use.

[0040] The communication system usually includes more than one (e / g) node B, in which case, the (e / g) node B can also be configured to communicate with each other through a wired or wireless link designed for this purpose. These links can not only be used for signaling purposes, but also can be used to route data from one (e / g) node B to another (e / g) node B. (e / g) Node B is a computing device configured to control the radio resources of the communication system to which it is coupled. Node B can also be referred to as a base station, access point, access node, or any other type of interface device, including a relay station capable of operating in a wireless environment. (e / g) Node B includes or is coupled to a transceiver. The connection is provided to an antenna unit from the transceiver of the (e / g) node B, which establishes a two-way radio link to a user equipment. The antenna unit may include multiple antennas or antenna elements. (e / g) Node B is also connected to a core network 110 (CN or next generation core NGC). Depending on the system, the counterpart on the CN side may be a serving gateway (S-GW, routing and forwarding user data packets), a packet data network gateway (P-GW) for providing a connection for a user equipment (UE) to an external packet data network, or a mobility management entity (MME), etc.

[0041] A user equipment (also referred to as UE, user equipment, user terminal, terminal device, etc.) illustrates a type of device to which resources on the air interface are allocated and assigned, and therefore any features described herein using the user equipment may be implemented using a corresponding device, such as a relay node. An example of such a relay node is a layer 3 relay (self-backhaul relay) to a base station.

[0042] User equipment generally refers to portable computing devices, including wireless mobile communication devices that operate with or without a subscriber identity module (SIM), including but not limited to the following types of devices: mobile stations (mobile phones), smart phones, personal digital assistants (PDAs), handheld devices, devices using wireless modems (alarms or measuring devices, etc.), laptops and / or touch-screen computers, tablet computers, game controllers, notebook computers, and multimedia devices. It should be understood that user equipment can also be nearly exclusive uplink-only devices, examples of which are cameras or video cameras that load images or video clips to the network. User equipment can also be a device with the ability to operate in an Internet of Things (IoT) network, which is a scenario in which objects are provided with the ability to transmit data over a network without human-to-human or human-to-computer interaction. User equipment can also utilize the cloud. In some applications, a user device may include a small portable device (such as a watch, headset, or glasses) with a radio portion, and the calculation is performed in the cloud. The user device (or in some embodiments, a layer 3 relay node) is configured to perform one or more functions in the user equipment function. A user device may also be called a subscriber unit, a mobile station, a remote terminal, an access terminal, a user terminal, or user equipment (UE), to mention just a few names or devices.

[0043] Wireless device is a general term covering both access nodes and terminal devices.

[0044] The various techniques described herein can also be applied to cyber-physical systems (CPS) (systems of computing elements that collaborate to control physical entities). CPS can support the implementation and utilization of a large number of interconnected ICT devices (sensors, actuators, processor microcontrollers, etc.) embedded in physical objects at different locations. Mobile cyber-physical systems are a subcategory of cyber-physical systems, where the physical systems in question have inherent mobility. Examples of mobile physical systems include mobile robots and electronic devices transported by humans or animals.

[0045] Additionally, although the apparatus has been depicted as a single entity, different units, processors and / or memory units (not all of which are shown in FIG. Figure 1 ) can be implemented.

[0046] 5G supports the use of Multiple-Input Multiple-Output (MIMO) antennas, many more base stations and nodes than LTE (the so-called small cell concept), including macro sites operating in cooperation with smaller stations, and employing a wide variety of radio technologies depending on service needs, use cases and / or available spectrum. 5G mobile communications support a wide range of use cases and related applications, including video streaming, augmented reality, different ways of sharing data and various forms of machine type applications (such as (massive) machine type communications (mMTC), including vehicle safety, different sensors and real-time control. 5G is expected to have multiple radio interfaces, namely sub-6 GHz, cm wave and mm wave, and can also be integrated with existing traditional radio access technologies (injection, LTE). At least in the early stages, the integration with LTE can be implemented as a system, where macro coverage is provided by LTE and 5G radio interface access comes from small cells through aggregation to LTE. In other words, 5G is planned to support both inter-RAT operability (such as LTE-5G) and inter-RI operability (inter-radio interface operability, such as sub-6 GHz to cm wave, sub-6 GHz to cm wave to mm wave). One of the concepts considered for use in 5G networks is network slicing, in which multiple independent and dedicated virtual sub-networks (network instances) can be created within the same infrastructure to run services with different requirements for latency, reliability, throughput and mobility.

[0047] The current architecture in LTE networks is completely distributed in the radio and completely centralized in the core network. Low-latency applications and services in 5G require bringing content close to the radio leading to local grooming and multi-access edge computing (MEC). 5G enables analysis and knowledge generation to occur at the data source. This approach requires the use of resources that may not be continuously connected to the network, such as laptops, smartphones, tablets, and sensors. MEC provides a distributed computing environment for application and service hosting. It also has the ability to store and process content very close to cellular subscribers in order to speed up response time. Edge computing covers a wide range of technologies such as wireless sensor networks, mobile data acquisition, mobile signature analysis, cooperative distributed peer-to-peer self-organizing networking and processing (also classified as local cloud / fog computing and grid / mesh computing), dew computing, mobile edge computing, cloudlets, distributed data storage and retrieval, autonomous self-healing networks, remote cloud services, augmented and virtual reality, data caching, Internet of Things (massive connectivity and / or latency critical), critical communications (autonomous vehicles, traffic safety, real-time analysis, time-critical control, healthcare applications).

[0048] The communication system may also be capable of communicating with other networks, such as the public switched telephone network or the Internet 112, or utilizing services provided by them. The communication network may also be capable of supporting the use of cloud services, for example, at least part of the core network operations may be performed as a cloud service (this is in the Figure 1 114). The communication system may also include a central control entity or the like, providing facilities for networks of different operators to cooperate, for example in spectrum sharing.

[0049] Edge cloud can be introduced into the radio access network (RAN) by leveraging network function virtualization (NVF) and software defined networking (SDN). Using edge cloud may mean that access node operations will be performed at least partially in a server, host or node that is operatively coupled to a remote radio head or a base station including the radio part. Node operations may also be distributed among multiple servers, nodes or hosts. The application of cloud RAN architecture enables RAN real-time functions to be performed on the RAN side (in distributed units, DU 104), and non-real-time functions to be performed in a centralized manner (in centralized units, CU108).

[0050] It should also be understood that the functional distribution between core network operations and base station operations may be different from LTE, or even non-existent. Some other technological advances that may be used may be big data or all-IP, which may change the way the network is constructed and managed. 5G (or new radio, NR) networks are designed to support multiple layers, where MEC servers can be placed between the core and the base station or Node B (gNB). It should be understood that MEC can also be applied to 4G networks.

[0051] 5G can also use satellite communications to enhance or supplement the coverage of 5G services, for example by providing backhaul. Possible use cases are to provide service continuity for machine-to-machine (M2M) or Internet of Things (IoT) devices or passengers carried by vehicles, or to ensure service availability for critical communications and future railway, maritime, and / or aviation communications. Satellite communications can utilize geostationary earth orbit (GEO) satellite systems, as well as low earth orbit (LEO) satellite systems, in particular mega constellations (systems in which hundreds of (nano) satellites are deployed). Each satellite 106 in a mega constellation can cover several satellite-supported network entities that create ground cells. Ground cells can be created by ground relay nodes 104 or by gNBs located on the ground or in satellites.

[0052] It is obvious to a person skilled in the art that the depicted system is only an example of a part of a radio access system, and in practice, the system may include multiple (e / g) Node Bs, user equipment may access multiple radio cells and the system may also include other devices, such as physical layer relay nodes or other network elements. At least one of the (e / g) Node Bs may be a home (e / g) Node B. Additionally, in a geographical area of ​​the radio communication system, a plurality of different kinds of radio cells and a plurality of radio cells may be provided. The radio cells may be macro cells (or umbrella cells), which are large cells, typically having a diameter of up to tens of kilometers, or may be smaller cells, such as micro cells, femto cells or pico cells. Figure 1 The (e / g) Node B can provide any type of these cells. The cellular radio system can be implemented as a multi-layer network including several types of cells. Usually, in a multi-layer network, one access node provides one or more cells of one type, and therefore multiple (e / g) Node Bs are required to provide such a network structure.

[0053] In order to meet the needs of improving the deployment and performance of communication systems, the concept of "plug and play" (e / g) Node B has been introduced. Generally, a network capable of using "plug and play" (e / g) Node B includes a Home Node B Gateway or HNB-GW (not in Figure 1 ). The HNB Gateway (HNB-GW), which is usually installed in the operator's network, can aggregate traffic from a large number of HNBs back to the core network.

[0054] As is known in relation to wireless communication systems, control or management information is communicated over a radio interface, for example between the terminal device 100 and the access node 104. The control information may include various information, such as those control information fragments described in the background. Figure 2 A basic block diagram of a process for transmitting a codeword X representing the value of a control information element is shown. The codeword X is encoded in a transmitter 300 and allocated to a dedicated location in a transmission resource block. The transmitted control information element is then delivered to a receiver 304 via a radio channel 302, which corrupts the transmitted control information element with noise. The noise characteristics depend on the distance between the transmitter and the receiver, the speed of the transmitter and the receiver, and obstacles between the transmitter and the receiver, among other factors. Several radio channel models have been designed to simulate how the channel modifies the transmitted signal.

[0055] Conventionally, each control information segment is individually decoded in the receiver 304, where decoding is the reverse operation of encoding in the transmitter. The output of the decoder is an estimate of the codeword X. For example, when the control information element is encoded by using Reed-Muller block coding, the receiver may employ a Reed-Muller block decoder to decode the control information element. For convolutionally encoded control information elements, a Viterbi decoder may be employed. Furthermore, there may be situations where the transmitter cannot send a control information element in a dedicated position, thus resulting in a discontinuous transmission (DTX) in the dedicated position. In order to avoid erroneous detection of noise or payload data as control information elements, unexpected DTX should be correctly detected in the receiver. Conventionally, separate DTX detection is employed.

[0056] The present invention employs a neural network to jointly perform decoding of control information elements and DTX detection. Figure 3 An embodiment of such a process performed in a wireless device receiving a control information element via a radio channel is illustrated. The wireless device may be an access node, in which case the control information is uplink control information. However, the principles of the present invention also apply to downlink control information, where the wireless device is a terminal device.

[0057] refer to Figure 2 , the process includes: acquiring (block 300) a frame and extracting a sample set from the frame according to the position of the control information element; inputting (block 302) the sample set into nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; (block 304) processing the sample set in a trained neural network before receiving the sample set to decode one or more determined values ​​of the control information element and detect discontinuous transmission; outputting an indicator indicating the decoded value of the control information element included in the sample set (block 308) or an indicator indicating discontinuous transmission (block 310) in an output layer of the neural network.

[0058] In an embodiment, the neural network in block 306 determines whether the sample set contains a control information element or DTX. If a control information element is found in the sample set, the process proceeds to block 308. If the sample set is detected to contain DTX, the process continues to block 310.

[0059] Figure 4The uplink control information (UCI) elements according to the 3GPP specification are illustrated. The UCI on the physical uplink shared channel (PUSCH) contains channel quality information (CQI) 404, precoding matrix indicator (PMI) 404, rank indicator (RI) 406, and acknowledgement / negative acknowledgement (ACK / NACK) 406. ACK / NACK belongs to a hybrid automatic repeat request, which is a block retransmission method for preventing the loss of a transport block. When UCI is sent on the PUSCH, channel coding, scrambling, and modulation for ACK / NACK, RI, and CQI / PMI are performed separately for each piece of control information before multiplexing the control information into the same symbol block. ACK / NACK is punctured as payload data 400 at a predetermined position next to a demodulation reference signal (DRS) in an uplink subframe. CQI / PMI is placed at the beginning of data 400. The number of punctured resource elements (REs) depends on the number of ACK / NACK bits and the modulation scheme adopted. A resource block is used to describe a group of REs that spans 12 consecutive subcarriers in the frequency domain and 1 slot in the time domain.

[0060] like Figure 4 As depicted, four different channel coding methods are applied to the UCI on the PUSCH. If the length of the CQI / PMI 404 is less than 11 bits, only Reed-Muller channel coding 412 (32 input, 2 11 Output) is performed. If the length of the CQI / PMI is greater than 11 bits, a convolutional encoder 410 (rate = 1 / 3) is used and then a rate matching operation is performed. When CQI / PMI channel coding is performed, the output of the decoder is multiplexed 416 with a vector of bits from the payload data 400 (such as application data).

[0061] Since CQI / PMI is calculated based on the simultaneously reported (same subframe) or previously reported rank (different subframe), RI and CQI / PMI may be provided in the same subframe or different subframes on PUSCH.

[0062] Each positive ACK is encoded as a binary 1 and each NACK is encoded as a 0. The HARQ feedback coding depends on the number of information bits in the ACK / NACK beam, that is, the number of confirmations sent simultaneously. The coding for one-bit ACK / NACK and RI is achieved by repetition coding in block 414. For two-bit ACK / NACK and RI, simplex coding 414 is used together with optional cyclic repetition of coded data. Those coding techniques are used to maximize the Euclidean distance of data symbols in modulation schemes, such as orthogonal phase shift keying (QPSK), 16 orthogonal amplitude modulation (QAM) and 64QAM, representing two, four and six coded bits, respectively. Longer ACK / NACK (more than 2 bits) may be subject to Reed-Muller block coding 412.

[0063] For example, in the channel interleaving block 418, the encoded RI and ACK / NACK channel coding bits are inserted into predefined column positions, and block interleaving is performed by filling the array by rows and reading the array by columns. The interleaved bits are then scrambled using a UE-specific pseudo-random sequence, modulated using QPSK, 16QAM, 64QAM, or 256QAM configured by a higher layer to form complex-valued symbols. The next step is a transform precoding function using a digital Fourier transform, followed by mapping to resource elements (subcarriers), and generating a complex-valued time domain transmission signal for each antenna port. The 3GPP specification defines single carrier frequency division multiple access (SC-FDMA) as an access scheme. In the downlink, orthogonal frequency division multiple access (OFDMA) is adopted.

[0064] In the receiver 304 for the PUSCH channel, an inverse SC-FDMA operation is performed by using DFT, thus bringing the received signal to the frequency domain. Subsequently, resource element demapping is performed. The reference signal is extracted and transmitted to the channel and noise estimator. Therefore, by using the estimated channel, the signal samples are equalized, and in the case of using more than a single antenna, antenna combination can be performed. The signal is then converted into a time domain signal by an inverse DFT, followed by resource block demapping. As a result of resource block demapping, a set of complex-valued symbols representing UCI elements can be obtained.

[0065] In addition to PUSCH, UCI can be sent on the physical uplink control channel (PUCCH), and similar operations described above can also be performed for PUCCH. Regarding the downlink, downlink control information (DCI) can be sent on the physical downlink control channel (PDCCH). Regarding DCI or other forms of control information, other coding schemes and transmission functions can be used.

[0066] As above combined Figure 3As described, the present invention relates to achieving the required performance with low computational complexity without knowledge of channel and noise conditions by adopting machine learning methods for decoding control information elements and DTX detection. Compared to conventional decoders and DTX detection schemes based on mathematical models and information theory, a machine learning architecture (such as a neural network) does not require expertise because it learns the design directly from training data. After training the neural network decoder, the decoding process becomes simple and has low latency. The configurable multi-layer neural network architecture can learn from different types of input training data to output the value of the control information element or the DTX output in a reliable manner. The neural network topology and weights can be configured according to the type of control information element that needs to be decoded, and more configurations are described below in conjunction with training the neural network. The neural network is also able to discover the control information element and the estimate of DTX by passing through each layer of the neural network only once, thereby enabling a low latency implementation, also known as one-shot decoding.

[0067] Then, let's refer to Figure 5 Let us denote the received sample set 500 representing the control information elements extracted from the received frame as follows:

[0068] y k =Hq+n (1)

[0069] where y k represents a sample set 500 from the position of the control information element in the received frame, H represents the channel matrix describing the radio channel, q comprises the coded transmitted symbol of the control information element (if present in the frame), and n represents noise. The sample set may contain complex-valued samples. Standard machine learning and neural network tools assume real-valued samples, and thus the complex-valued sample set may be converted to a real-valued sample set (block 502). The sample set y in equation (1) is k can be viewed as a stack of the real and imaginary parts of a set of complex-valued samples, and so if the number of complex-valued samples in the set is P, then y k The number of real-valued samples in is 2P.

[0070] The set of real-valued samples can then be input to the input layer of a neural network as described above. The input layer can have 2P nodes, one for each data sample.

[0071] In an embodiment, the output layer of the neural network includes a number of nodes equal to the possible outputs of the neural network. Figure 5Some possible outputs of the neural network are illustrated on the right. For example, since the neural network is configured to detect DTX, one possible output is a DTX output 508. When the neural network is configured to detect a one-bit control information element (such as a one-bit ACK / NACK 504 or a one-bit RI 506), two possible outputs are achieved through the possible values ​​of the one-bit control information element: one output for a bit value of '0' and one output for a bit value of '1'. When the neural network is configured to detect a two-bit control information element (such as a two-bit ACK / NACK 504 or a two-bit RI 506), four possible outputs are achieved through the possible values ​​of the two-bit control information element: one output for each bit value of '00', '01', '10', '11'. In short, when the neural network is configured to jointly detect a one-bit control information element and DTX, the number of possible outputs and the number of nodes in the output layer is three. When the neural network is configured to jointly detect a two-bit control information element and DTX, the number of possible outputs and the number of nodes in the output layer is five. In total, the number of nodes in the output layer is 2 T +1, where T is the number of bits in the control information element that the neural network is configured to decode.

[0072] The output layer can be configured to output a non-zero value in only one of the nodes of the output layer and to output a zero value in the other one or more nodes of the output layer. Thus, the output layer can be configured to output a "one-hot vector". Tables 1 and 2 below illustrate a possible one-hot encoding method for the output layer for a one-bit control information element (Table 1) and for a two-bit control information element (Table 2).

[0073] send One-hot encoding tokens 0 001 1 010 DTX 100

[0074] Table 1

[0075] send One-hot encoding tokens 00 00001 01 00010 10 00100 11 01000 DTX 10000

[0076] Table 2

[0077] One way to understand the output of the neural network is that, in response to the input sample set in the input layer, only one "light" is lit in the output layer, and the light that is lit indicates the detection of DTX or the value of the control information element in the input sample set by using the one-hot encoded label in Table 1 or Table 2. Due to the mapping between the one-hot label and the transmitted value in Table 1 or Table 2, the value of the control information element or DTX can be determined.

[0078] In an embodiment, the output layer is configured to output a one-hot vector by using an activation function (e.g., a softmax function). The neural network may inherently output a value representing the probability for each possible output, i.e., the output layer may output a non-zero value in each output node without using an activation function, which represents the probability of the corresponding output. The activation function may select a value with the largest probability among the output values, and set the selected output to a non-zero value and set other outputs to zero to implement a one-hot output vector. An example of a Softmax function is the following function: Where Z j represents the value of the output vector before the excitation function, and j is the number of nodes in the output layer, for example three or five, depending on the length of the control information element being decoded. Another example of an excitation function is max(0, z j ), where the highest probability value is selected.

[0079] Then, let us describe the internal structure and operation of the neural network according to some embodiments. In an embodiment, the neural network includes only two layers between the input layer and the output layer: the first layer is coupled to the output of the input layer and includes a number of nodes equal to the number of nodes in the input layer; and the second layer is coupled between the first layer and the output layer and includes a number of nodes equal to the number of nodes in the output layer. The inventors have found that two layers provide a good combination of high performance and low complexity, especially for short control information elements (one or two bits). A larger number of layers may not provide additional performance improvement, but may increase complexity.

[0080] In an embodiment, Figure 5 The neural network shown is fully connected. Being fully connected means that every node in a layer is connected to every node in the previous layer and the subsequent layer. For example, every node in the input layer is coupled to every node in layer 1, every node in layer 1 is coupled to every node in layer 2, and every node in layer 2 is coupled to every node in the output layer.

[0081] In a neural network, the output of one layer is the input to the next layer. For example, the output of the input layer serves as the input to layer 1, the output of layer 1 serves as the input to layer 2, and the output of layer 2 serves as the input to the output layer. Each node in a layer can be viewed as a neuron using a nonlinear activation function σ(.), which gives the neuron the ability to learn. Therefore, a layer can be described by the following equation:

[0082] a1=y k (2)

[0083] Equation (2) represents the input sample set y k is fed into the input layer. Layer l can be defined as:

[0084] a l+1 =σ l (W l a l +b l ) (3)

[0085] W l and b l are weight parameters and bias parameters respectively, and a l is the input to layer l. Figure 1 For layer 1, l=1, and for layer 2, l=2. The output layer can be modeled as:

[0086]

[0087] where L represents the maximum value of l, and the function σ L It can be the activation function described above for achieving the one-hot output vector.

[0088] Layers 1 and 2 can be considered as hidden layers because their activation functions σ l This is achieved through training. The weight parameter W for each node in the hidden layer l and the bias parameter b l is discovered during training. Let us then describe some embodiments for performing training, which is a key factor in configuring neural networks for high performance.

[0089] In an embodiment, the input to the neural network is the demodulated soft symbol values ​​or log likelihood ratios. In another embodiment, the neural network is also trained to perform demodulation, in which case the input is a set of modulated symbol values.

[0090] In an embodiment, the neural network is trained by training inputs to the neural network, the training inputs comprising: all possible values ​​of the control information elements decoded by the neural network, the control information elements corrupted by simulated noise simulating one or more radio channel models; and DTX training inputs that do not carry control information elements. The DTX training inputs may comprise pure noise samples or payload data samples. Training may comprise a training phase and a testing phase. Training may be performed offline, i.e., training may be performed before any actual transmission of the control information elements. Let us now refer to Fig. 6A To describe the training phase.

[0091] refer to Fig. 6A, the control information bits to be decoded are generated in block 600. For example, the output of block 600 may be a one-bit or two-bit control information element. The control information element is then encoded and modulated in block 602, depending on the system specification to be used in the actual transmission of the control information element. The output of block 602 may be a sample of one or more modulated symbols of the control information element. In an embodiment where the neural network does not perform demodulation, modulation may be omitted from block 602, or a separate demodulation step may be provided between block 606 and block 608. In block 604, the sample(s) of modulated symbols are modified by samples of a selected radio channel model. Table 3 below lists a number of possible channel models that simulate real radio channels and that can be used to train a neural network. In block 604, noise samples are applied to the samples to achieve a selected signal-to-noise ratio (SNR) for the samples.

[0092] Now, with the knowledge of the true value of the control information element, the neural network can be trained in block 608 as input from block 600 to select weight parameters and bias parameters for the nodes of the hidden layer so that the control information element is extracted from the output samples of block 606. The process can employ a performance metric to determine the performance of the neural network during the execution of block 608. An example of a performance metric is a loss function, which is a measure of the difference between the output of the neural network and the desired output of the neural network (the output of block 600). The loss function can be a calculation of a performance metric of the neural network, such as a mean square error or binary cross entropy between the output of the neural network and the desired output. The set of weight parameters and bias parameters that provide the best performance metric can be selected, and the parameters can be stored in block 608.

[0093] An example of a loss function is:

[0094]

[0095] Where X k is the desired output (the value sent) of the neural network, and is the output of the neural network. K is the total number of outputs considered.

[0096] In an embodiment, the performance threshold may be set to a minimum performance requirement for the neural network. In other words, it may not be sufficient to select only the best performance metric, but the best performance metric should also provide performance that meets a threshold level defined by the performance threshold. The 3GPP specification defines the performance requirement represented by a minimum SNR to achieve a 1% miss detection probability. Therefore, a design rule for the neural network is to provide equal miss detection probabilities for erroneously detecting DTX as a control information element value (e.g., as ACK) and for erroneously detecting one value of a control information element as another value of a control information element (e.g., ACK→NACK):

[0097] P(DTX→ACK)=P(ACK→NACK) (6)

[0098] By assuming that the probability P(ACK→NACK) for falsely detecting an ACK as a NACK is negligibly small, the following approximation can be made:

[0099] P(ACK→DTX)~P(ACK→NACK)

[0100] P(DTX→ACK)~P(DTX→NACK) (7)

[0101] Then, the false alarm rate (FAR) can be calculated as a function of the missed detection probability:

[0102] FAR=P(DTX→ACK)+P(DTX→NACK)=2*P(ACK→DTX)=2*P(ACK→NACK)=2*P(DTX→ACK)=2*MDP (8)

[0103] where MDP represents the missed detection probability in equation (6).

[0104] Now, the false alarm rate sets a stricter requirement on the performance of the neural network, and therefore the loss function can be modified to include weight parameters in the form of a weight matrix. Then, equation (5) can be modified to the following form:

[0105]

[0106] The weight matrix can have 2 α +1, where α is the length of the control information element in bits and has the following structure:

[0107]

[0108] Thereafter, the next noise level may be selected in block 610, and another training iteration may be performed by returning to execute blocks 606 and 608 at the next noise level. When all noise levels for the radio channel models have been trained to the neural network and the corresponding optimal parameters have been found and stored, the process may proceed from block 610 to block 612 for selecting the next radio channel model to be trained to the neural network. In this way, the neural network may be trained to detect the control information elements output from block 600 in various radio channels having various noise levels. When all radio channels and noise levels have been trained to the neural network, the process may end.

[0109] The 3GPP specification defines some specifications that can be used in training. Table 3 below illustrates some channel models and noise levels that can be used for training a neural network.

[0110]

[0111] Table 3

[0112] Can be performed for multiple possible values ​​of the control information element Fig. 6A Additionally, the process may be performed for multiple DTX samples. Figure 6B An embodiment of block 600 and block 602 is illustrated. When training a neural network for detecting DTX, DTX samples may be generated in block 600 in addition to the value of the control information element. Figure 6B 6, (uplink) control information (UCI) element bits are generated in block 650 and DTX samples are generated in block 652. The DTX samples may include noise samples or randomly generated bits. Thus, blocks 650 and 652 represent embodiments of block 600. Accordingly, blocks 650 and 652 may include constructing a training data set including a sequence of DTX sample sets and a set of control information element samples. The control information element may be coded and modulated in block 602. In the case where the DTX samples include randomly generated bits, they may be coded and modulated. In the case of noise samples, block 602 may be omitted.

[0113] Training may be performed for multiple DTX sample sets and multiple sets of control information element values, and the DTX sample sets and control information value sample sets may be provided in an interleaved sequence, e.g., one DTX sample set followed by 5 ACK / NACK values ​​followed by 10 DTX sample sets, etc. The sequence may be selected in block 654. In an embodiment, at least 50% of the training data sets are DTX sample sets. For example, if the control information element is a one-bit indicator and the training of FIG. 6 is repeated N times for each value of the control information element, the number of DTX training events is at least 2N. In an embodiment, the total number of training data sample sets (DTX, different control information values) is higher than 10,000 to provide sufficient training statistics.

[0114] When training using the process of FIG. 6 is complete, multiple parameter sets are stored (block 608), each parameter set being optimized to detect a particular control information element in the presence of a given channel model and noise level.

[0115] Then, let's refer to Figure 7 The test phase is described in detail with reference to an embodiment of FIG. 6. Blocks with the same reference numbers as in FIG. 6 represent the same or substantially similar functions. Figure 7In an embodiment of the invention, after the channel model samples and noise have been added to the control information element symbols or DTX samples output from block 602, the samples represent the received control information (or DTX) samples corrupted by the noise and the radio channel. These samples are input to the neural network and to a reference receiver, which may be a conventional receiver including a conventional decoder (block 702). With respect to the neural network, the first set of neural network parameters stored in block 608 are loaded into the neural network in block 704. These parameters may be used until Figure 7 The process ends. The received samples are then processed by the reference receiver in block 702, and the bit error rate (BER) of the reference receiver is calculated in block 708 by using the decoded output of the reference receiver and the control information element generated in block 600. The reference receiver can also be configured for DTX detection, and successful / failed DTX detection can be incorporated into the performance evaluation of the reference receiver. Successful / failed DTX detection can be incorporated into the BER calculation in the same manner as successful / failed bit decoding.

[0116] The received samples are also processed by the neural network using the loaded parameters, and the BER of the output of the neural network is calculated in block 709 by using the output of the neural network and the control information element generated in block 600. As described above, the neural network can also be configured for DTX detection, and successful / failed DTX detection can be incorporated into the evaluation of BER performance, as described in the previous paragraphs for the reference receiver.

[0117] The BER values ​​calculated in blocks 708 and 709 may then be stored and the next noise level applied to the control information element samples in block 606. In this way, the process may be repeated in the same manner for different noise levels and channel models as described above in conjunction with FIG. 6. Accordingly, the neural network parameters may be tested with various noise levels and channel models and the BER calculated for each channel model and noise level. The test scenarios of Table 3 may be used. When all noise levels and channel models have been tested, a comparison is made in block 710 between the BER obtained with the reference receiver and the BER obtained with the neural network configured with the parameters loaded in block 704. The comparison may include the calculation of a normalized validation error (NVE) defined as follows:

[0118]

[0119] Where R represents the number of channel models trained, S represents the number of noise values ​​trained, and BER NN represents the BER of the neural network obtained as the output of block 709 in each iteration, and BER RRrepresents the BER of the reference receiver obtained in each iteration as the output of block 708. When a BER comparison is performed, the result of the comparison (e.g., the NVE value) can be stored for later selection of a neural network configuration.

[0120] By executing Figure 7 A neural network parameter set may be tested for a reference receiver under various channel models and various noise levels. A neural network parameter set may be performed for each neural network parameter set (i.e., a different neural network configuration obtained as a result of executing the process of FIG. 6 ). Figure 7 Since the process is unchanged for the reference receiver, blocks 702 and 708 do not need to be repeated. When all neural network parameter sets have been tested for the reference receiver, the neural network parameter set that provides the lowest NVE value can be selected for actual decoding.

[0121] In an embodiment, a deterministic selection logic may be used when selecting a set of neural network parameters. For example, if the channel model includes an additive white Gaussian noise (AWGN) channel, a NVE obtained using the AWGN channel model may be provided with a lower weight than NVEs associated with other channel models.

[0122] The combined decoding and DTX detection described above provides high performance for small control information elements (e.g., one-bit or two-bit codewords) using a neural network. When the size of the control information element is larger, more hidden layers may be required and the complexity of the neural network may increase. However, the neural network can provide acceptable performance when decoding longer codewords.

[0123] In an embodiment, for example, when the control information element is long, a neural network may be used in DTX detection, and the actual decoding of the control information element may be performed by a conventional decoder. In this case, the neural network may be trained only for DTX detection. The input to the neural network will be samples from the positions of the control information elements, and the neural network will only detect the presence / absence of DTX. The output layer may then have only two possible outputs: one for DTX and another for non-DTX. In an embodiment, the output layer of the neural network may contain one node indicating either DTX or non-DTX. In another embodiment, the output layer of the neural network may contain two nodes: one indicating DTX and the other indicating non-DTX. The one-hot notation may be as follows (TX indicates transmission of the control information element):

[0124] send One-hot encoding tokens TX 01 DTX 10

[0125] Table 4

[0126] Figure 8The apparatus is illustrated as including a communication controller 50, such as at least one processor or processor circuitry, and at least one memory 60 including computer program code (software) 64, wherein the at least one memory and the computer program code (software) are configured to, together with the at least one processor, cause the apparatus to perform Figure 3 The process or any one of the embodiments described above. The apparatus can be used for a terminal device or an access node (e.g., a gNB) or a controller that controls the operation of an access node according to an embodiment of the present invention. Figure 8 The device may be an electronic device.

[0127] refer to Figure 8 , the memory 60 may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memory, magnetic memory devices and systems, optical memory devices and systems, fixed memory, and removable memory. The memory may include a configuration database 66 for storing configuration parameters, such as neural network parameters selected as a result of the training phase and the testing phase described above. The parameters may be pre-stored in the configuration database 66, such as before the device is used for the first time. The memory 60 may also store a data buffer 68 for data awaiting transmission and / or data to be decoded.

[0128] refer to Figure 8 , the device may also include a communication interface 62, which includes hardware and / or software for implementing communication connectivity according to one or more radio communication protocols. The communication interface 62 can provide radio communication capabilities to devices in a wireless network (e.g., a radio access network of a cellular communication system). In an embodiment, the communication interface 62 includes one or more antenna arrays that provide the device with the ability to form directional transmit radio beams and receive radio beams. The communication interface may include well-known analog radio components, such as amplifiers, filters, frequency converters and circuits, conversion circuits for converting signals between analog and digital domains, and one or more antennas. Digital signal processing regarding decoding of received signals can be performed in the communication controller 50.

[0129] The communication controller 50 may include a control information element (IE) extractor 56 configured to extract control information elements from a received frame (e.g., from one or more subcarriers of a frame). A frame may be a transmitted signal in a certain time-frequency resource. The control information elements extracted from the received frame may include a set of samples from an ACK / NACK or RI position, for example, although other information elements are equally possible. When extracting a set of samples from a received frame, the set of samples may be input to a neural network decoder and a DTX detector 54, for example. Figure 5. The entity 54 may be configured to perform joint DTX detection and decoding on the sample set and, as described above, output an indication of the result of the decoding and DTX detection. When the result is a DTX detection, further processing involving the sample set may end. When the result is a determined value of the control information element, the value may be input to the transmission controller 52, which is configured to perform a determined function in response to the determined value. For example, when the value indicates NACK, the transmission controller 52 may be configured to trigger a retransmission of the message, such as a transmission resource block. When the value indicates ACK, the transmission controller may end a HARQ process associated with the message transmission confirmed by the ACK. When the value indicates a determined channel rank, the transmission controller may configure the transmission parameters according to the channel rank. In a similar manner, the transmission controller may respond to the result of the decoding according to the value of the decoded control information element.

[0130] Figure 8 At least some functions of the apparatus can be shared between two physically separate devices to form an operational entity. Therefore, the apparatus can be regarded as depicting an operational entity including one or more physically separate devices for performing at least some of the processes described with respect to the access point. For example, an access node utilizing such a shared architecture may include a remote control unit (RCU) or a central unit (CU), such as a host computer or a server computer, which is operably coupled (e.g., via a wireless or wired network) to a remote radio head (RRH) or a distributed unit (DU), such as a transmission reception point (TRP) located in an access node (e.g., gNB 104, 104A, 104B). The RCU can generate a virtual network through which the RCU communicates with the RRH. In general, a virtual network can involve the process of combining hardware and software network resources and network functions into a single software-based management entity (virtual network). Network virtualization can involve platform virtualization, which is often combined with resource virtualization. Network virtualization can be classified as combining many networks or parts of a network into an external virtual networking of a server computer or a host computer (i.e., an RCU). The goal of external network virtualization is to optimize network sharing. Another classification is internal virtual networking that provides network-like functions to software containers on a single system. Internal virtual networks can provide flexible distributed operations between RRHs and RCUs. In practice, any digital signal processing task can be performed both in the RRH and in the RCU, and the boundary of responsibility transfer between the RRH and RCU can be selected according to the implementation. As an example of a shared architecture case, once the SgNB-CU receives the combined indication packet and the scheduling window (such as Figure 5, 502), it will need to communicate the indication to the MAC layer in the gNB-DU. This can be performed in several ways. For example, by assigning increased packet priority or adjusted latency targets to packets, QoS information overlaid on the F1 interface between the DU and the CU is established (as part of the F1 process of UE context setup / DRB modification). Alternatively, the scheduling window can be provided directly to the MAC via F1.

[0131] At least some of the processes described above may be performed by the RCU or shared between the RRH and the RCU.

[0132] As used in this application, the term 'circuitry' refers to all of the following: (a) pure hardware circuit implementations, such as implementations in analog and / or digital circuits only, and (b) combinations of circuits and software (and / or firmware), such as (as applicable): (i) combinations of (multiple) processors or (ii) portions of (multiple) processors / software, including (multiple) digital signal processors, software and (multiple) memories, which work together to enable the device to perform various functions, and (c) circuits, such as (multiple) microprocessors or portions of (multiple) microprocessors, which require software or firmware for operation, even if the software or firmware is not present. This definition of "circuitry" applies to all uses of this application. As a further example, as used in this application, the term "circuitry" would also cover only a processor (or multiple processors) or portions of a processor and its (or their) accompanying software and / or firmware. The term "circuitry" would also cover, for example and if applicable to the particular element, a baseband integrated circuit or application processor integrated circuit for a mobile phone or a similar integrated circuit in a server, cellular network device, or another network device.

[0133] In an embodiment, in combination Figures 2 to 7 At least some of the processes described may be performed by an apparatus including corresponding components for performing at least some of the processes described. Some example components for performing the processes may include at least one of the following: a detector, a processor (including a dual-core or multi-core processor), a digital signal processor, a controller, a receiver, a transmitter, an encoder, a decoder, a memory, a RAM, a ROM, software, firmware, a display, a user interface, a display circuit, a user interface circuit, a user interface software, a display software, a circuit, an antenna, an antenna circuit system, and a circuit system. In one embodiment, at least one processor, a memory, and a computer program code form a processing component or include one or more computer program code parts for performing according to any one of the above. Figures 2 to 7 Or an operational embodiment thereof performs one or more operations.

[0134] According to another embodiment, an apparatus for performing an embodiment includes a circuit system including at least one processor and at least one memory including computer program code. When activated, the circuit system causes the apparatus to perform according to Figures 2 to 7 Any embodiment of the embodiments or at least some functions of its operation.

[0135] The techniques and methods described herein can be implemented by various components. For example, these techniques can be implemented in hardware (one or more devices), firmware (one or more devices), software (one or more modules), or a combination thereof. For hardware implementation, the (multiple) devices of the embodiment can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to perform the functions described herein, or a combination thereof. For firmware or software, the implementation can be performed by at least one module (e.g., process, function, etc.) of a chipset that performs the functions described herein. The software code can be stored in a memory unit and executed by a processor. The memory unit can be executed within or outside the processor. In the latter case, as known in the art, it can be communicatively connected to the processor through various components. Additionally, the components of the system described herein can be rearranged and / or supplemented by accessory components to facilitate the implementation of the various aspects described therewith, etc., and as will be understood by those skilled in the art, they are not limited to the precise configuration described in the given image.

[0136] The described embodiments may also be implemented in the form of a computer process, which is defined by a computer program or part thereof. By executing at least a part of a computer program comprising corresponding instructions, the Figures 2 to 7 Embodiments of the described methods are described herein. The computer may be in the form of source code, in the form of information of object code, or in the form of some intermediate type, and it may be stored in some type of carrier, which may be any entity or device capable of carrying the program. For example, a computer program may be stored on a computer program distribution medium readable by a computer or processor. For example, a computer program medium may be, but is not limited to, a recording medium, a computer memory, a read-only memory, an electrical carrier signal, a telecommunication signal, and a software distribution package. For example, a computer program medium may be a non-transient medium. Software coding for executing the embodiments shown and described is also within the scope of ordinary technicians in the art. In an embodiment, a computer readable medium includes the computer program.

[0137] Even though the present invention has been described above with reference to the examples according to the accompanying drawings, it is clear that the present invention is not limited thereto but can be modified in several ways within the scope of the appended claims. Therefore, all words and expressions should be interpreted broadly and they are intended to illustrate rather than limit the embodiments. It will be obvious to those skilled in the art that, as technology advances, the concepts of the present invention can be implemented in various ways. Furthermore, it is clear to those skilled in the art that the described embodiments can be, but are not required to be, combined with other embodiments in various ways.

Claims

1. A device for communication, comprising: at least one processor; as well as at least one memory including computer program code, wherein the at least one memory and the computer program code are configured to, together with the at least one processor, cause the apparatus to perform the following: obtaining, from a received frame, a set of samples associated with a position of a control information element in the frame; Inputting the sample set into nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; processing the set of samples in the neural network that has been trained to decode one or more determined values ​​of the control information element and detect discontinuous transmissions prior to receiving the set of samples; outputting, in an output layer of the neural network, an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating the discontinuous transmission, wherein the neural network is trained by training inputs to the neural network, the training inputs comprising all possible values ​​of the control information element corrupted by simulated noise, wherein the simulated noise simulates one or more radio channel models, the training inputs also comprising payload data samples or samples of randomly generated bits as discontinuous transmission training inputs.

2. The apparatus of claim 1 , wherein the output layer comprises a number of nodes equal to the possible outputs of the neural network.

3. The apparatus of claim 2, wherein the neural network is configured to output a non-zero value in only one of the nodes of the output layer and to output a zero value in the other one or more nodes of the output layer.

4. The apparatus of claim 3, wherein the set of samples comprises modulation symbol samples, and wherein the neural network is configured to decode one or more determined values ​​of the control information element from the modulation symbol samples.

5. An apparatus according to any preceding claim, wherein the neural network comprises only two layers between the input layer and the output layer: a first layer is coupled to the output of the input layer and comprises a number of nodes equal to the number of nodes in the input layer; and a second layer is coupled between the first layer and the output layer and comprises a number of nodes equal to the number of nodes in the output layer.

6. The apparatus of any one of the preceding claims 1 to 4, wherein the neural network is fully connected.

7. The apparatus according to any one of the preceding claims 1 to 4, wherein the control information element has one bit or two bits.

8. The apparatus according to any one of the preceding claims 1 to 4, wherein the control information element is an acknowledgement message confirming reception of a frame or a rank indicator.

9. A method for communication, comprising: Obtaining, by the wireless device, from a received frame, a set of samples associated with a location of a control information element in the frame; Inputting the sample set to nodes of an input layer of a neural network by the wireless device, the input layer having the same number of nodes as the number of samples in the sample set; processing, by the wireless device, the set of samples in the trained neural network prior to receiving the set of samples to decode one or more determined values ​​of the control information element and detect discontinuous transmissions; outputting, by the wireless device, in an output layer of the neural network, an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating the discontinuous transmission, wherein the neural network is trained by training inputs to the neural network, the training inputs comprising all possible values ​​of the control information element corrupted by simulated noise, wherein the simulated noise simulates one or more radio channel models, the training inputs also comprising payload data samples or samples of randomly generated bits as discontinuous transmission training inputs.

10. The method of claim 9, wherein the output layer comprises a number of nodes equal to the possible outputs of the neural network.

11. The method of claim 10, wherein the neural network outputs a non-zero value in only one of the nodes in the output layer and outputs a zero value in the other one or more nodes in the output layer.

12. The method of claim 11, wherein the set of samples comprises modulation symbol samples, and wherein the neural network decodes the one or more determined values ​​of the control information element from the modulation symbol samples.

13. A method according to any of the preceding claims 9 to 12, wherein the neural network comprises only two layers between the input layer and the output layer: a first layer is coupled to the output of the input layer and comprises a number of nodes equal to the number of nodes in the input layer; and a second layer is coupled between the first layer and the output layer and comprises a number of nodes equal to the number of nodes in the output layer.

14. The method of any one of the preceding claims 9 to 12, wherein the neural network is fully connected.

15. The method according to any of the preceding claims 9 to 12, wherein the control information element has one bit or two bits.

16. The method according to any of the preceding claims 9 to 12, wherein the control information element is an acknowledgement message confirming reception of a frame or a rank indicator.

17. A computer program product embodied on a computer readable medium and comprising computer program code readable by a computer, wherein the computer program code, when read and executed by the computer, configures the computer to perform a computer process comprising: obtaining, from a received frame, a set of samples associated with a position of a control information element in the frame; Inputting the sample set into nodes of an input layer of a neural network, the input layer having the same number of nodes as the number of samples in the sample set; processing the set of samples in the neural network that has been trained to decode one or more determined values ​​of the control information element and detect discontinuous transmissions prior to receiving the set of samples; outputting, in an output layer of the neural network, an indicator indicating a decoded value of the control information element included in the sample set or an indicator indicating the discontinuous transmission, wherein the neural network is trained by training inputs to the neural network, the training inputs comprising all possible values ​​of the control information element corrupted by simulated noise, wherein the simulated noise simulates one or more radio channel models, the training inputs also comprising payload data samples or samples of randomly generated bits as discontinuous transmission training inputs.