Generating training examples for machine learning based receivers

Abstract

A method performed by a user equipment (UE) for generating training data. The method comprises receiving a signal transmitted by a network node, where the signal is modulated using a training sequence (TS). The method also includes generating a feature based on the received signal (e.g., I or Q components of baseband samples of the signal). The method further includes generating a TS for modulating the signal, where the generating includes generating the TS using a pseudorandom sequence generator (pRSG). The method also includes generating a training example including the feature and a tag, where the tag includes the generated TS.

Description

[0001] This application has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 101015956. Technical Field

[0002] Embodiments related to generating training examples for a machine learning (ML)-based receiver are disclosed. Background Art

[0003] A receiver is an entity that receives the transmitted data and detects the transmitted data. This data can be transmitted via a wireless channel, an optical fiber, or a wired channel. The transmitted data may undergo some distortion depending on the transmission channel and the hardware at the transmitter and the receiver. The operating conditions of a receiver typically vary over time and / or frequency and / or space (e.g., a wireless transmission channel may change due to weather conditions, and the performance of a power amplifier (PA) in a transmitter may change over time). Since these changes are generally unknown, the receiver needs to estimate them to achieve optimal performance. That is, to cope with these changing operating conditions, the receiver can be adjusted (e.g., change parameter values) depending on the conditions under which the receiver operates.

[0004] Machine Learning

[0005] Machine learning (ML) refers to the technique of using training data (also known as a training data set or training examples) to train a model that can be used for various applications including inference, classification, and / or prediction. ML algorithms can be classified as online algorithms and offline algorithms, where offline algorithms rely on a pre-trained model, while online algorithms can train the model in real time when new data samples are received. Another fundamental difference lies in supervised ML, unsupervised ML, and reinforcement learning. In the supervised learning paradigm, each training example in the training data set includes input data (also known as features) paired with corresponding output data (also known as labels). That is, in supervised learning, the model is trained using a labeled data set, while in unsupervised learning, the training data set is unlabeled.

[0006] Artificial Neural Network

[0007] An artificial neural network is a class of machine learning algorithms that are widely used due to their ability to approximate any general function based on a training data set and their inherent parallel processing ability (which makes these techniques attractive candidates for implementation on emerging artificial intelligence (AI) accelerator hardware). As Figure 1 shown, a neural network is based on interconnected processing units (referred to as neurons), where each neuron receives a weighted version of the outputs of other neurons and uses an activation function to compute the output based on a non-linear transformation of the aggregated input.

[0008] ML Receiver Method

[0009] As Figure 2 shown, in a transmitter / receiver architecture, an ML method can be used on the receiver side to optimize one or more functions of the receiver. For example, a neural network (NN)-based ML receiver method has been proposed in References [1] and [2] to optimize a demapper (a single function) to compensate for hardware impairments due to oscillator phase noise. Figure 2 The receiver chain 200 is shown, where an example of a single function (soft demapper) is replaced by an ML agent (i.e., an ML-based soft demapper).

[0010] Figure 3 A possible implementation of a neural network receiver is shown in. Figure 3 The structure in performs symbol-by-symbol soft demapping, taking as input the I and Q components of the complex baseband samples, context information, and the signal-to-noise ratio (SNR) estimate, and generating soft bits as output. Such a demapper will help improve the performance of a system affected by radio frequency (RF) impairments such as power amplifier (PA) non-linearity. Figure 4 A comparison of the performance of this ML-based method with that of a baseline method is shown in.

[0011] Neural Network (NN) Training

[0012] Training of an NN refers to adjusting the parameters (e.g., weight and bias values) of the NN based on training examples (i.e., a set of feature-label tuples). The purpose of training is to improve the accuracy of the results of the NN, e.g., reducing classification errors or prediction errors. This training is performed by adapting the parameters of the NN to minimize a loss function, where the loss function is defined as a measure of the distance between the neural network output for the training features and the training labels. There are several methods for training neural networks, such as gradient descent-based methods, which update the network parameters by backpropagating the error of the output of the neural network, aiming to minimize the loss function. Summary of the Invention

[0013] There are certain challenges currently. For example, although ML-based receiver methods have been proposed (e.g., see References [1, 2] discussed in the "Background Art" section above), these methods assume that training examples are available at the receiver for performing model training, and this assumption is rarely true. That is, in general, the receiver does not have access to a good set of training examples.

[0014] The transmitter can send training examples directly to the receiver (in which case it would mean that the examples are sent individually and additionally ensuring that they are received correctly), which increases the transmission overhead and complexity; or the transmitter can attempt to construct the label part of the training data set using signals that are successfully decoded by a conventional receiver or a trained ML-based receiver (i.e., data blocks received that pass the CRC check) (see, for example, references [5] and [6]). However, the problem with the latter approach is that it only collects samples for which the decoder can successfully decode the transmitted coded blocks. Therefore, samples with lower signal quality will be lost because there is no guarantee that they can be successfully decoded. This poses a problem when training the receiver because the receiver will not be able to continuously train and adapt to the training data (only the successfully received blocks) and it will not be able to handle specific conditions representing such low signal quality receptions. These conditions particularly include lower S(I)NR and / or larger RF impairments (e.g., high levels of phase noise or non-linear PA operation), etc. It is worth noting that these operating regions with lower signal quality receptions are the most promising regions for performance gains in receiver calibration (see Figure 4 the 24 dB SNR point in

[0015] Accordingly, on the one hand, a method for generating training data performed by a UE is provided. The method includes receiving a signal transmitted by a network node, where the signal is modulated using a training sequence (TS). The method further includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). The method further includes generating a TS for modulating the signal, where the generating includes using a pseudo-random sequence generator (pRSG) to generate the TS. The method further includes generating a training example including the features and a label, where the label includes the generated TS.

[0016] On the other hand, a UE is provided that is configured to perform a method for generating training data. The method includes receiving a signal transmitted by a network node, where the signal is modulated using a training sequence (TS). The method further includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). The method further includes generating a TS for modulating the signal, where the generating includes using a pseudo-random sequence generator (pRSG) to generate the TS. The method further includes generating a training example including the features and a label, where the label includes the generated TS.

[0017] On the other hand, a method for generating training data performed by a network node is provided. The method includes receiving a signal sent by a UE, where the signal is modulated using a training sequence (TS) generated by the user equipment. The method further includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). The method further includes obtaining the TS used to modulate the signal. The method further includes generating a training example including the features and a label, where the label includes the generated TS.

[0018] On the other hand, a network node is provided, which is configured to perform a method for generating training data. The method includes receiving a signal sent by a UE, where the signal is modulated using a training sequence (TS) generated by the UE. The method further includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). The method further includes obtaining the TS used to modulate the signal. The method further includes generating a training example including the features and a label, where the label includes the generated TS.

[0019] On the other hand, a computer program including instructions is provided, which, when executed by a processing circuit of a device (e.g., a UE or a network node), causes the device to perform any method disclosed herein. In one embodiment, a carrier containing a computer program is provided, where the carrier is one of an electrical signal, an optical signal, a radio signal, and a computer-readable storage medium.

[0020] One advantage of the embodiments disclosed herein is that, since there is no need to transmit labels, they can reduce the overhead in transmission, and since there is no need to encode / decoder the labels for reliable transmission to the receiver, they have lower complexity. Additionally, these embodiments are able to collect samples in cases where a block is not successfully received (e.g., at low S(I)NR or when the distortion level due to hardware impairments is high). These samples are crucial for training receiver methods to compensate for hardware impairments, as the gains of these methods are most promising under such operating conditions. Furthermore, these embodiments are able to collect training examples in scenarios or operating conditions where the performance of a pre-trained receiver model is poor and retraining is required. Description of the Drawings

[0021] The drawings included herein and forming a part of this specification illustrate various embodiments.

[0022] Figure 1 A neural network is shown.

[0023] Figure 2 Components of an ML-based receiver according to an embodiment are shown.

[0024] Figure 3Shows a neural network receiver.

[0025] Figure 4 Is a data graph comparing the performance of an ML-based method with the performance of a baseline method.

[0026] Figure 5 Shows a signaling diagram according to an embodiment.

[0027] Figure 6 Shows a signaling diagram according to an embodiment.

[0028] Figure 7A Shows a transmission format according to an embodiment.

[0029] Figure 7B Shows a transmission format according to an embodiment.

[0030] Figure 8 Shows a DCI for multiplexing the transmission of user plane (UP) data and training sequence (TS), and shows how to map UP data and TS to physical resources.

[0031] Figure 9 Shows a signaling diagram according to an embodiment.

[0032] Figure 10 Shows a system according to an embodiment.

[0033] Figure 11 Is a flowchart showing a process according to an embodiment.

[0034] Figure 13 Is a block diagram of a UE according to an embodiment.

[0035] Figure 14 Is a block diagram of a network node according to an embodiment. Detailed Description

[0036] Figure 10 Shows a system 100 according to an embodiment. System 100 includes a user equipment (UE) 102 and a network node 104 (in the example shown, network node 104 is a base station or a component of a base station). As used herein, a UE is any device capable of wireless communication with a network node, such as a phone, tablet computer, computer, sensor, home appliance, etc.; and a network node is any device that provides network access to a UE. UE 102 and / or network node 104 may include an ML-based receiver chain (e.g., see receiver chain 200). This disclosure provides embodiments for obtaining training examples for training an ML model used by an ML-based receiver.

[0037] As described above, each training example includes features paired with a label. One problem addressed by these embodiments is obtaining the label portion of the training example. Compared to existing systems, in the proposed embodiments, the label is generated at the ML-based receiver and / or pre-stored at the ML-based receiver. For example, in one embodiment, the label is a pseudo-randomly generated bit string, also referred to as a "bit stream" or "training sequence (TS)", which is created on the transmitter side, e.g., by using a pRSG, and generated at the receiver by using the same pRSG. By ensuring that both the transmitter (Tx) and the receiver (Rx) obtain the same training sequence (TS), these embodiments reduce overhead because these embodiments do not require the label to be sent to the Rx, and these embodiments also ensure that new training examples can be obtained regardless of the signal quality. The configuration of the pRSG used during the training process can be transmitted to the Rx as part of the training configuration process or hard-coded as specified in the 3rd Generation Partnership Project (3GPP) standards.

[0038] The pRSG is described in various 3GPP publications and is used in various parts of the 3GPP system to generate pseudo-random number sequences for various purposes, such as generating spreading codes in spread-spectrum communication, generating keys for encryption, and generating masks for channel coding. 3GPP specifies multiple pRSGs that can be used in different parts of the system and provides a detailed description of their design and operation. For example, Sections 7.4.1.1.1 and 5.2.1 of 3GPP Technical Specification (TS) 38.211 V17.4.0 describe the use of a binary pRSG to generate demodulation reference signals for the Physical Downlink Shared Channel (PDSCH) in 5G NR. The pRSG uses a seed (usually in the form of a number or a vector) to initialize the pseudo-random number generator.

[0039] In the context of the present disclosure, the ML-based receiver employs a supervised learning paradigm, where, during the training phase, training examples (i.e., feature-label tuples) are used to iteratively adjust the ML parameters. In one embodiment, the label portion of the training example is a vector of binary values of length N (N > 0), which represents a bit sequence that is associated with symbols (e.g., There is a one-to-one mapping relationship between the complex-valued symbols in a set (of QAM symbols) - this binary-valued vector of length N is referred to as the training sequence (TS) in this article. Typical values of N are 4 (which maps to 16-level modulation, such as 16-QAM) and 6 (which maps to 64-level modulation, such as 64-QAM). The feature part of the training example includes the I and Q values of the complex symbols to which the training sequence is mapped, where the signal is received at the receiver, i.e., corrupted by the transmitter and receiver non-linearity, radio channel effects, and thermal noise. In another embodiment, in addition to the I and Q values of the received complex symbols, the feature also includes an estimated value of the SNR and context information (e.g., the power level of the PA). In another embodiment, the feature is the value of the complex-valued symbol received after channel equalization.

[0040] In one embodiment, multiple TSs are created from a single L-length sequence (i.e., a binary-valued vector). For example, a pRSG and a seed value can be used to generate the L-length sequence, which is then divided into L / N N-length sub-parts to create L / N TSs, where each TS is mapped to a complex-valued symbol, and the received version of this complex-valued symbol is used as part of the feature, thus forming L / N training examples. Generally, training an ML-based receiver requires a large number of training examples, so L is usually equally large, e.g., on the order of thousands or tens of thousands. In one embodiment, the L-length sequence is encoded by an error-correcting code before being divided into sub-parts and mapped to complex-valued symbols. In one embodiment, the label is the sub-part of the L-length sequence before applying the error-correcting code, i.e., the ML-based receiver jointly implements the demapping of symbols to bits and channel decoding. In another embodiment, the label is the sub-part of the L-length sequence generated by the error-correcting code encoding the sequence, i.e., the ML-based receiver only implements the demapping of symbols to bits.

[0041] Figure 5 is a signaling diagram showing a process according to one embodiment. In this embodiment, the UE 102 includes an ML-based receiver and generates a group of one or more training examples for training the ML model adopted by the ML-based receiver.

[0042] In Figure 5 the example shown, the network node 104 detects a training trigger (e.g., an event), which triggers the network node to initiate a training process. This trigger can be, for example, the expiration of a timer, receiving a control message from an operation and management system, or receiving a training data request message from the UE 102.

[0043] After detecting a training trigger, the network node obtains a training sequence (TS), e.g., a pseudo-random bit stream. In one embodiment, the network node obtains the TS by generating a pseudo-random sequence (PRS) using a pRSG. In one embodiment, the generated PRS is the TS. In one embodiment, the process for generating the TS is standardized. In one embodiment, the TS is generated after a single pRSG process that generates the PRS. After this step, there may be a scrambling process for scrambling the PRS, where it is ensured that this scrambling varies over time, e.g., when generating the scrambling sequence, timing information such as on a time slot, frame, or radio frame ID is used. For example, a scrambling code (i.e., one or more bits) may be selected based on the time slot or frame in which the TS will be transmitted, and the scrambled TS can be generated by performing an exclusive OR operation on the PRS generated by the pRSG and the selected scrambling code, thereby generating the TS to be transmitted.

[0044] Additionally, after detecting a training trigger, the network node sends a control message m502 to the UE, e.g., downlink control information (DCI), radio resource control (RRC) message, media access control (MAC) control element (CE), which indicates to the UE that the TS for generating training examples will be transmitted.

[0045] In one embodiment, the control message includes a training data indicator (TDI) for providing this indication to the UE. The TDI may consist of one or more bits. In one embodiment, the TDI is explicitly included in the control message. In another embodiment, the TDI is bitwise exclusive ORed with the cyclic redundancy check (CRC) bits of the control message.

[0046] The control message may also contain transmission format information, which indicates the transmission format that the network node will use to transmit the TS. This transmission format information enables the UE to receive the TS. For example, the transmission format information may include: a channel coding identifier indicating the channel coding that will be used to transmit the TS, information identifying the resource units (e.g., physical resources in time / frequency) that will be used to transmit the RS, an MCS indicator indicating the modulation and coding scheme (MCS) used to modulate the TS, information indicating the scrambling sequence used to transmit the TS, and layer mapping information.

[0047] The TS obtained by the network node (e.g., pRSG TS or scrambled TS) may be associated with state information (e.g., a seed value or a scrambling code), where this state information ensures that the generated TS is randomized over time. This state information may be explicitly signaled in the control message, or it may be implicitly determined, e.g., by defining it to be generated using the above example timing information.

[0048] After sending control message m502, the network node sends a TS. For example, the TS is sent according to the modulation and coding scheme (MCS) indicated in the control message. More specifically, the network node sends a signal modulated with the TS. Preferably, the modulation is multi-level modulation, i.e., non-binary modulation. For example, 64 QAM modulation is used.

[0049] In one embodiment, as Figure 7A shown, the process of sending the TS includes channel coding and scrambling of the TS; while in another embodiment, as Figure 7B shown, channel coding and scrambling are omitted. In either case, the TS is mapped to a data channel, such as the physical downlink shared channel (PDSCH) or the physical uplink shared channel (PUSCH). That is, one or more specific resources (e.g., orthogonal frequency division multiplexing (OFDM) symbols) are used to send the TS.

[0050] To enable the sending node and the receiving node to coordinate the mapping of the TS to radio resources, the MAC / PHY procedures of conventional downlink (DL) data transmission can be followed (e.g., see Figure 7A ). This is referred to here as the transmission format and can include, for example, channel coding, resource allocation for physical resources in time / frequency, modulation and coding scheme (MCS), scrambling sequence, layer mapping. As described above, in one embodiment, the TS used to train the ML model can be channel coded and possibly scrambled. Thus, in this case, the TS can be regarded as the payload of data transmission.

[0051] As Figure 5 shown, the UE receives the modulated signal carrying the TS and uses conventional receiver techniques to process the signal (e.g., demodulation, equalization, signal-to-noise ratio (SNR) measurement, etc.) to create features, such as the I and Q components of the complex baseband samples of the signal, context information (e.g., power amplifier (PA) back-off, speed of movement, etc.), and signal-to-noise ratio (SNR) estimates.

[0052] The UE also obtains a tag, i.e., the same TS sent by the network node. For example, the UE obtains (e.g., generates) the TS using the same state information as that used by the network node. Alternatively, the UE may retrieve the TS from a pool of TSs stored at the UE. For example, each TS included in the pool may be associated with an index value, and the control message includes the index value of the TS sent to the UE. In this way, the UE will obtain the same TS obtained by the network node and sent to the UE. Then, the UE generates a training example, where the features of the training example include the above information, such as I / Q symbols, SNR values, context information, and the label of the training example includes the obtained TS. Then, the training example can be used to train the ML model used by the UE. The UE may perform the training by itself, or the UE may provide the training example to a server, which then uses the training example to perform the training.

[0053] The process of generating TSs at the transmitter and receiver is similar to the process of generating reference signals (e.g., DMRS, CSI-RS, PTRS). However, standardized reference signals are typically optimized to confer certain waveform characteristics (e.g., limited envelope variation), which are different from the waveform characteristics of the data-carrying waveform and are not sufficient to train a machine learning-based receiver. The training data for an ML-based receiver needs to be statistically and in terms of waveform shape similar to the user plane data. This data and its corresponding waveform will activate some operating points that may not be activated by the standard reference signals, e.g., resulting in excessive nonlinear distortion.

[0054] Figure 6 is a signaling diagram showing a process according to another embodiment. In this embodiment, the network node 104 includes an ML-based receiver and generates a group of one or more training examples for training the ML model adopted by the ML-based receiver.

[0055] In Figure 6 the example shown, the network node 104 detects a training trigger, which triggers the network node to initiate a training process. The trigger may be, for example, a timer expiration or a control message received from an operation and management system.

[0056] After detecting a training trigger, the network node sends a control message m602 (e.g., DCI, RRC message, MAC CE) to the UE 102, which indicates that the UE sends a specific TS so that the network node can generate training examples. In one embodiment, the control message includes a TDI, which provides the indication to the UE. The TDI can consist of a single bit or multiple bits. In one embodiment, the TDI is explicitly included in the control message. In another embodiment, the TDI is bitwise XORed with the cyclic redundancy check (CRC) bits of the control message (affecting the false detection rate of control message decoding but keeping the control message definition unchanged).

[0057] The control message can also contain (as described above) transmission format information, which indicates the transmission parameters that the UE must use to send the TS. The control message can also include information for enabling the UE to generate a specific TS. For example, the information can include a seed value and / or a scrambling code. In another embodiment, the control message can also contain information for enabling the UE to retrieve a specific TS from a TS pool stored at the UE. For example, the information can include a TS index value.

[0058] In response to receiving the control message and determining that the control message indicates that the UE sends a specific TS, the UE performs a process including obtaining the TS. In one embodiment, the UE obtains the TS by generating the TS using a pRSG. In one embodiment, the process of generating the TS is standardized. In one embodiment, the TS is generated after a single pRSG process. The scrambling process described above can be performed after this step.

[0059] The TS obtained by the UE can be associated with state information, where such state information ensures that the generated TS is randomized over time. Such state information can be explicitly signaled in the control message, or can be implicitly determined by, for example, defining it to be generated using the example timing information described above.

[0060] After obtaining the TS, the UE uses radio resources (e.g., the radio resources identified in the control message) to send the TS. In some embodiments, the UE encodes the TS before transmission. More specifically, the network node sends a signal modulated with the TS. Preferably, the modulation is multi-level modulation, i.e., non-binary modulation. For example, 64 QAM modulation is used.

[0061] As Figure 6 shown, the network node receives the modulated signal carrying the TS and processes the signal using conventional receiver techniques (e.g., demodulation, equalization, signal-to-noise ratio (SNR) measurement, etc.) to create features (e.g., in-phase (I) and quadrature (Q) symbol values, measured SNR values, context information about the receiver (e.g., power amplifier (PA) back-off)).

[0062] The network node also obtains the tag, i.e., the same TS sent by the UE. For example, the network node uses the same process as the process used by the UE to obtain (e.g., generate) the TS. In this way, the network node will obtain the same TS that the UE has obtained and sent to the network node. Then, the network node generates a training example, where the features of the training example include one or more of the above features, and the tag of the training example includes the obtained TS. Then, this training example can be used to train the ML model used by the network node.

[0063] Figure 7A An embodiment of Tx (UE or network node) processing the TS before transmission is shown. Specifically, in the shown embodiment, the TS is channel-coded and then scrambled. Then the encoded and scrambled TS is mapped to a data channel, e.g., selecting a resource element (RE) where the TS will be sent. Figure 7B An embodiment is shown where Tx does not encode and / or scramble the TS before mapping.

[0064] In the above embodiment, the transmission format of the TS follows the same process as the process used for existing user plane (UP) data transmission, and the TS essentially replaces the regular UP transmission, but this is not necessary. For example, it may be beneficial to multiplex the UP data with the TS (using frequency and / or time-interleaved radio resources for transmission). In this case, there may be different transmission formats for the UP data and the TS, and part or all of the transmission formats can be dynamically conveyed in a control message, semi-statically configured by RRC, or configured by a specification text. Combinations of one or more configurations can also be used to obtain the required information. For example, the resource mapping can be defined by RRC, the process for determining the seed used to generate the TS can be defined by the specification, and the data aspects related to training data collection (e.g., the modulation scheme used, the number of spatial layers) can be dynamically indicated by a control message.

[0065] More specifically, the generated TS can be mapped to a specific resource set, e.g., one or more OFDM symbols in an NR transmission. That is, for example, a specific set of OFDM symbols (e.g., the first OFDM symbol in a subframe) is used to send the TS. Such resources can be rate-matched or punctured, i.e., around the payload transmission taking place in the same resource set, the training data will overwrite / replace the mapped payload resource values. In some embodiments, one modulation and coding scheme (MCS) is used to send the UP data, while a different MCS is used to send the TS.

[0066] Figure 8 An exemplary mapping is shown, Figure 8The DCI (indicated by the arrow) that schedules DL or UL transmissions shown in the figure includes not only information about the UP data and associated reference signals (RS) to be transmitted, but also information about the TS. In this case, the information about the TS includes a modulation indicator (which may be different from the UP data), a seed for TS generation, and a resource mapping. The resource mapping in this example is simply to place the TS in the first time resource symbol. Other more complex mapping options are also envisioned, such as interleaved allocation on the time / frequency resource grid.

[0067] In an embodiment, the TS can be used as a demodulation reference signal (DMRS) for the UP data, in which case the UP DMRS can be omitted.

[0068] In addition to the aperiodic / dynamic transmission of training data, the training data can also be configured for periodic or semi-persistent transmission, which follows principles and signaling similar to those for reference signal scheduling.

[0069] To reduce overhead and / or UE complexity, for example, by restricting training to a subset of UEs that need it, the TDI, state information, and / or TS for training can be sent in a broadcast or multicast manner (e.g., via the multicast and broadcast service (MBS) control channel (MCCH) and / or MBS traffic channel (MTCH) and / or group-based DCI). For example, a network node can broadcast a group consisting of one or more TSs that can be used by several UEs for training.

[0070] In some embodiments, the Rx (network node 104 or UE 102) can indicate when and which training data are needed to train its model (as used herein, training a model refers not only to initial training but also to retraining). For example, the Rx can monitor its performance, and if the Rx detects that the performance has dropped below a threshold, this event triggers the Rx to request the Tx to send the TS so that the Rx can generate new training examples. Figure 9 An example of this embodiment is shown. Figure 9 It shows the Rx sending a training data request (TDR) to the Tx.

[0071] In one embodiment, the TDR is a single bit. When the Rx is a UE, the TDR can be included in the UL control message (e.g., transmitted via the physical uplink control channel (PUCCH)), similar to the existing functionality in the NR specification (using the PUCCH for uplink control information (UCI) transmission); when the Rx is a network node, the TDR can be included in the DL control message (e.g., DCI, MAC CE, RRC message), similar to the existing functionality in the NR specification (e.g., using the physical downlink control channel (PDCCH) transmission to trigger the UE to send a sounding reference signal (SRS)).

[0072] In another embodiment, the TDR consists of two or more bits. The TDR can be carried in the UCI or DCI, sent on a specific control channel (e.g., the uplink PUCCH in NR), or multiplexed with data (e.g., the PUSCH in NR). Alternatively, the TDR can be included in the MAC control element (MAC CE). In this embodiment where the TDR is more than one bit, the TDR can not only indicate the request for the Tx to send the TS (i.e., the request for training data), but also indicate: what type of TS is requested, the number (N) of TSs to be sent, and the transmission format to be used for sending the TS. The indicated transmission format can include, for example: the modulation scheme to be used (e.g., if the receiver estimates that it performs poorly under 64 QAM modulation, it requests additional training data using that modulation order); the number of spatial layers / streams for which the receiver wants to receive training data (similar to modulation, it estimates that the model needs to be improved in this operating mode); and the hardware configuration used by the receiver to receive the training data.

[0073] Similar to the request to activate training data, the Rx can use a similar channel and procedure to stop sending training data. For example, as Figure 9 shown, the Rx can send a stop training message to the Tx to cause the Tx to stop sending the TS.

[0074] Figure 11 is a flowchart showing a process 1100 for generating training data performed by the UE 102 according to an embodiment. The process 1100 can start at step s1102. Step s1102 includes receiving a signal sent by a network node (104), where the signal is modulated using a training sequence (TS). Step s1104 includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). Step s1106 includes generating the TS for modulating the signal, where the generation includes using a pseudo-random sequence generator (pRSG) to generate the TS. Step s1108 includes generating a training example including the features and a label, where the label includes the generated TS.

[0075] Figure 12is a flowchart showing process 1200 for generating training data performed by network node 104 according to an embodiment. Process 1200 may begin at step s1202. Step s1202 includes receiving a signal transmitted by a user equipment (UE) (102), where the signal is modulated using a training sequence (TS) generated by the UE. Step s1202 includes generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal). Step s1206 includes obtaining the TS used to modulate the signal. Step s1208 includes generating a training example including the features and a label, where the label includes the generated TS.

[0076] Figure 13 is a block diagram of UE 102 according to some embodiments. As Figure 13 shown, UE 102 may include: a processing circuit (PC) 1302, which may include one or more processors (P) 1355 (e.g., one or more general microprocessors and / or one or more other processors, such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), etc.); a communication circuit 1348, which is coupled to an antenna arrangement 1349 including one or more antennas and includes a transmitter (Tx) 1345 and a receiver (Rx) 1347 for enabling UE 102 to transmit and receive data (e.g., wirelessly transmit / receive data); and a storage unit (also known as a "data storage system") 1308, which may include one or more non-volatile storage devices and / or one or more volatile storage devices. In embodiments where PC 1302 includes a programmable processor, a computer-readable storage medium (CRSM) 1342 may be provided. CRSM 1342 may store a computer program (CP) 1343 including computer-readable instructions (CRI) 1344. CRSM 1342 may be a non-transitory computer-readable medium, such as a magnetic medium (e.g., a hard disk), an optical medium, a storage device (e.g., a random access memory, a flash memory), etc. In some embodiments, the CRI 1344 of computer program 1343 is configured such that when executed by PC 1302, the CRI causes UE 102 to perform the steps described herein (e.g., the steps described herein with reference to the flowchart). In other embodiments, UE 102 may be configured to perform the steps described herein without code. That is, for example, PC 1302 may consist of only one or more ASICs. Thus, the features of the embodiments described herein may be implemented in hardware and / or software.

[0077] Figure 14 is a block diagram of network node 104 for performing the network node methods disclosed herein according to some embodiments. As Figure 14As shown, network node 104 may include: a processing circuit (PC) 1402, which may include one or more processors (P) 1455 (e.g., general microprocessors and / or one or more other processors such as application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), etc.), which may be co-located in a single housing or a single data center, or may be geographically distributed (i.e., the network node may be a distributed computing device); a network interface 1468, including a transmitter (Tx) 1465 and a receiver (Rx) 1467, for enabling network node 104 to send data to and receive data from other nodes connected to network 110 (e.g., an Internet Protocol (IP) network), the network interface 1468 being connected to the network; a communication circuit 1448 (e.g., a radio transceiver circuit including Rx 1447 and Tx 1445), coupled to an antenna system 1449, for wireless communication with a UE or other nodes; and a storage unit (aka "data storage system") 1408, which may include one or more non-volatile storage devices and / or one or more volatile storage devices. In embodiments where PC 1402 includes a programmable processor, a computer-readable storage medium (CRSM) 1442 may be provided. CRSM 1442 may store a computer program (CP) 1443 including computer-readable instructions (CRI) 1444. CRSM 1442 may be a non-transitory computer-readable medium such as a magnetic medium (e.g., a hard disk), an optical medium, a storage device (e.g., random access memory, flash memory), etc. In some embodiments, the CRI 1444 of computer program 1443 is configured such that when executed by PC 1402, the CRI causes network node 104 to perform the steps described herein (e.g., the steps described herein with reference to one or more flowcharts). In other embodiments, network node 104 may be configured to perform the steps described herein without code. That is, for example, PC 1402 may consist of only one or more ASICs. Thus, the features of the embodiments described herein may be implemented in hardware and / or software fashion.

[0078] Overview of various embodiments

[0079] A1. A method for generating training data performed by a user equipment (UE), the method comprising: receiving a signal sent by a network node, wherein the signal is modulated using a training sequence (TS); generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal); generating a TS for modulating the signal, wherein the generating includes using a pseudo-random sequence generator (pRSG) to generate the TS; and generating a training example including the features and a label, wherein the label includes the generated TS.

[0080] A2. The method according to embodiment A1 further includes training a machine learning (ML) model using training examples.

[0081] A3. The method according to embodiment A1 or A2 further includes: before receiving the signal, receiving a control message (m502) from a network node, the control message (m502) indicating that the network node will send the signal.

[0082] A4. The method according to embodiment A3, wherein the control message includes a seed value, and generating the TS includes using the seed value as an input to the pRSG.

[0083] A5. The method according to embodiment A3, wherein the method further includes: determining a seed value based on information indicated by the control message (i.e., information indicating the time slot / frame in which the signal will be sent), and generating the TS includes using the seed value as an input to the pRSG.

[0084] A6. The method according to any one of embodiments A3 to A5, wherein the control message is a downlink control information (DCI), the control message is a MAC control element (MAC CE), the control message is a radio resource control (RRC) message, the control message is a broadcast message, or the control message is a multicast message.

[0085] A7. The method according to any one of embodiments A1 to A6, wherein generating the TS includes: using the seed value as an input to the pRSG to generate a pseudo-random sequence (PRS), and scrambling the PRS with a scrambling code to generate the TS.

[0086] A8. The method according to any one of embodiments A1 to A7 further includes: before receiving the signal, sending a training data request message to the network node, the training data request message requesting the network node to send one or more TSs.

[0087] A9. The method according to embodiment A8, wherein the data request message indicates the number of TSs being requested to be sent.

[0088] A10. The method according to embodiment A2, wherein the ML model acts as a symbol-to-bit demapper (i.e., is configured to generate soft bits).

[0089] A11. The method according to any one of embodiments A1 to A10, wherein using the pRSG to generate the TS includes: using the pRSG to generate a vector of binary values, wherein the vector has a length of L and L>1; and using the vector of binary values to generate the TS.

[0090] A12. The method according to embodiment A11, wherein TS is a sub - part of a binary - valued vector.

[0091] A13. The method according to embodiment A11, wherein generating TS using a binary - valued vector includes: generating an encoded version of the binary - valued vector using an error - correcting code, wherein the TS is a sub - part of the encoded version of the binary - valued vector.

[0092] B1. A method for generating training data performed by a network node, the method comprising: receiving a signal transmitted by a user equipment (UE), wherein the signal is modulated using a training sequence (TS) generated by the UE; generating features based on the received signal (e.g., the I or Q component of the baseband samples of the signal); obtaining the TS used to modulate the signal; and generating a training example including the features and a label, wherein the label includes the generated TS.

[0093] B2. The method according to embodiment B1, further comprising training a machine learning (ML) model using the training example.

[0094] B3. The method according to embodiment B1 or B2, further comprising sending a control message (m602) to the UE, the control message (m602) instructing the UE to send the signal.

[0095] B4. The method according to embodiment B3, wherein the control message includes a seed value for enabling the UE to generate the TS.

[0096] B5. The method according to embodiment B4, wherein obtaining the TS includes generating the TS using the seed value as an input to a pseudo - random sequence generator (pRSG).

[0097] B6. The method according to any one of embodiments B3 to B5, wherein the control message is a downlink control information (DCI), the control message is a MAC control element (MAC CE), or the control message is a radio resource control (RRC) message.

[0098] B7. The method according to any one of embodiments B1 to B6, wherein obtaining the TS includes: generating a pseudo - random sequence (PRS) using the seed value as an input to a pRSG, and scrambling the PRS using a scrambling code to produce the TS.

[0099] B8. The method according to any one of embodiments B1 to B5, further comprising: before receiving the signal, sending a training data request message to the UE, the training data request message requesting the UE to send one or more TSs.

[0100] B9. The method according to embodiment B8, wherein the training data request message indicates the number of TSs being requested to be sent.

[0101] B10. The method according to embodiment B2, wherein the ML model acts as a symbol-to-bit demapper (i.e., is configured to generate soft bits).

[0102] C1. A computer program comprising instructions which, when executed by a processing circuit of a UE, cause the UE to perform the method according to any one of embodiments A1 to A10.

[0103] C2. A computer program comprising instructions which, when executed by a processing circuit of a network node, cause the network node to perform the method according to any one of embodiments B1 to B10.

[0104] C3. A carrier containing the computer program according to embodiment C1 or C2, wherein the carrier is one of an electrical signal, an optical signal, a radio signal, and a computer-readable storage medium (1342, 1442).

[0105] D1. A user equipment (UE) configured to generate training data, the UE operable to: receive a signal transmitted by a network node, wherein the signal is modulated using a training sequence (TS); generate features based on the received signal (e.g., the I or Q component of the baseband samples of the signal); generate a TS for modulating the signal, wherein the generation includes using a pseudo-random sequence generator (pRSG) to generate the TS; and generate a training example including the features and a label, wherein the label includes the generated TS.

[0106] D2. A user equipment (UE) configured to generate training data, the UE including a processing circuit and a storage unit, the storage unit including computer-readable instructions executable by the processing circuit, whereby the UE is operable to: receive a signal transmitted by a network node, wherein the signal is modulated using a training sequence (TS); generate features based on the received signal (e.g., the I or Q component of the baseband samples of the signal); generate a TS for modulating the signal, wherein the generation includes using a pseudo-random sequence generator (pRSG) to generate the TS; and generate a training example including the features and a label, wherein the label includes the generated TS.

[0107] D3. The UE according to embodiment D1 or D2, wherein the UE is further operable to perform the method according to any one of embodiments A2 to A13.

[0108] E1. A network node operable to: receive a signal transmitted by a user equipment (UE), wherein the signal is modulated using a training sequence (TS) generated by the UE; generate features based on the received signal (e.g., the I or Q component of the baseband samples of the signal); obtain the TS used to modulate the signal; and generate a training example including the features and a label, wherein the label includes the generated TS.

[0109] E2. A network node configured to generate training data, the network node including processing circuitry and a storage unit including computer-readable instructions executable by the processing circuitry, whereby the network node is operable to:

[0110] E3. The network node according to embodiment E1 or E2, wherein the network node is further operable to perform the method according to any one of embodiments B2 to B10.

[0111] Conclusion

[0112] The advantages of the embodiments disclosed above are that they enable the Rx to generate training examples in a way that does not consume valuable network resources.

[0113] Although various embodiments are described herein, it should be understood that they are presented by way of example and not limitation. Thus, the breadth and scope of the present disclosure should not be limited by any of the above exemplary embodiments. Additionally, any combination of the above elements in all possible variations is included in the present disclosure unless otherwise indicated or clearly conflicts with the context in some other way.

[0114] As used herein, sending a message "to" or "towards" an intended recipient includes sending the message directly to the intended recipient or indirectly to the intended recipient (i.e., using one or more other nodes to relay the message from the source node to the intended recipient). Similarly, as used herein, receiving a message "from" a sender includes receiving the message directly from the sender or indirectly from the sender (i.e., using one or more nodes to relay the message from the sender to the receiving node). Additionally, "a" herein means "at least one" or "one or more".

[0115] Additionally, although the processes described above and shown in the figures are shown as a series of steps, they are for illustrative purposes only. Thus, it is contemplated that some steps may be added, some steps may be omitted, the order of steps may be rearranged, and some steps may be performed in parallel.

[0116] References

[0117] [1] International Patent Application Publication No. WO2021262052, titled, “A CONTEXT AWARE DATA RECEIVER FOR COMMUNICATION SIGNALS BASED ON MACHINE LEARNING.”

[0118] [2] H. Farhadi and M. Sundberg, "Machine learning empowered context-aware receiver for high-band transmission," IEEE Globecom Workshops, 2020.

[0119] [3] International Patent Application No. PCT / SE2022 / 050730, titled “A method of adaptive neural network receive.”

[0120] [4] International Patent Application No. PCT / SE2022 / 050667, titled, “A method of adaptive transmit signal quality for capable receivers.”

[0121] [5] Fischer, Moritz Benedikt, et al. "Adaptive Neural Network-based OFDM Receivers." 2022 IEEE 23rd International Workshop on Signal Processing Advances in Wireless Communication (SPAWC), IEEE, 2022.

[0122] [6] Schibisch, Stefan, et al. "Online label recovery for deeplearning-based communication through error correcting codes." 2018 15thInternational Symposium on Wireless Communication Systems (ISWCS), IEEE,2018。

Claims

1. A method (1100) for generating training data performed by a user equipment UE (102), the method comprising: Receiving (s1102) a signal sent by a network node (104), wherein the signal is modulated using a training sequence TS; Generating (s1104) features based on the received signal; Generating (s1106) the TS for modulating the signal, wherein the generating comprises: using a pseudo-random sequence generator pRSG to generate the TS; and Generating (s1108) a training example comprising the features and a label, wherein the label comprises the generated TS.

2. The method according to claim 1, wherein, The method further comprises training a machine learning ML model using the training example.

3. The method according to claim 1 or 2, wherein The method further comprises: Before receiving the signal, receiving a control message (m502) from the network node (104), the control message (m502) indicating that the network node (104) will send the signal.

4. The method according to claim 3, wherein, The control message comprises a seed value, and Generating the TS comprises: using the seed value as an input to the pRSG.

5. The method according to claim 3, wherein, The method further comprises determining a seed value based on information indicated by the control message, and Generating the TS comprises: using the seed value as an input to the pRSG.

6. The method according to any one of claims 3 to 5, wherein, The control message is a downlink control information DCI, The control message is a media access control MAC control element MAC CE, The control message is a radio resource control RRC message, The control message is a broadcast message, or, The control message is a multicast message.

7. The method according to any one of claims 1 to 6, wherein, Generating the TS comprises: using a seed value as an input to the pRSG to generate a pseudo-random sequence PRS; and Scrambling the PRS using a scrambling code to generate the TS.

8. The method according to any one of claims 1 to 7, wherein The method further comprises: Before receiving the signal, sending a training data request message to the network node, the training data request message requesting the network node to send one or more TSs.

9. The method according to claim 8, wherein The data request message indicates the number of TSs being requested to be sent.

10. The method according to claim 2, wherein, The ML model acts as a symbol-to-bit demapper.

11. The method according to any one of claims 1 to 10, wherein, Using the pRSG to generate the TS comprises: Using the pRSG to generate a binary value vector, wherein the vector has a length L and L>1; and Using the binary value vector to generate the TS.

12. The method according to claim 11, wherein The TS is a sub-part of the binary value vector.

13. The method according to claim 11, wherein, Using the binary value vector to generate the TS comprises: using an error correction code to generate an encoded version of the binary value vector, wherein the TS is a sub-part of the encoded version of the binary value vector.

14. A method (1200) for generating training data performed by a network node (104), the method comprising: Receive (s1202) a signal sent by a user equipment UE (102), where the signal is modulated using a training sequence TS generated by the UE; Generate (s1204) features based on the received signal; Obtain (s1206) the TS used to modulate the signal; and Generate (s1208) a training example including the features and a label, where the label includes the generated TS.

15. The method according to claim 14, wherein, The method further includes using the training example to train a machine learning ML model.

16. The method according to claim 14 or 15, wherein The method further includes sending a control message (m602) to the UE, where the control message (m602) instructs the UE to send the signal.

17. The method according to claim 16, wherein, The control message includes a seed value for enabling the UE to generate the TS.

18. The method according to claim 17, wherein, Obtaining the TS includes: using the seed value as an input to a pseudo-random sequence generator pRSG to generate the TS.

19. The method according to any one of claims 16 to 18, where the control message is a downlink control information DCI, the control message is a MAC control element MAC CE, or the control message is a radio resource control RRC message.

20. The method according to any one of claims 14 to 19, where obtaining the TS includes: using a seed value as an input to the pRSG to generate a pseudo-random sequence PRS; and scrambling the PRS using a scrambling code to generate the TS.

21. The method according to any one of claims 14 to 18, wherein, The method further includes: before receiving the signal, sending a training data request message to the UE, where the training data request message requests the UE to send one or more TSs.

22. The method according to claim 21, wherein The training data request message indicates the number of TSs being requested to be sent.

23. The method according to claim 15, wherein, The ML model acts as a symbol-to-bit demapper.

24. A computer program (1343) including instructions (1344), where when the instructions are executed by a processing circuit (1302) of a UE, the UE is caused to execute the method according to any one of claims 1 to 13.

25. A computer program (1443) including instructions (1444), where when the instructions are executed by a processing circuit (1402) of a network node, the network node is caused to execute the method according to any one of claims 14 to 23.

26. A carrier, comprising the computer program according to claim 24 or 25, wherein, The carrier is one of an electrical signal, an optical signal, a radio signal, and a computer-readable storage medium (1342, 1442).

27. A user equipment UE (102) configured to generate training data, where the UE (102) is operable to: Receiving a signal transmitted by a network node (104), wherein, the signal is modulated using a training sequence TS; generate features based on the received signal; generate the TS used to modulate the signal, where the generation includes: using a pseudo-random sequence generator pRSG to generate the TS; and generate a training example including the features and a label, where the label includes the generated TS.

28. The UE (102) according to claim 27, wherein, The UE is further operable to execute the method according to any one of claims 2 to 13.

29. A network node (104), where the network node is operable to: Receives a signal sent by a user equipment UE (102), wherein, The signal is modulated using a training sequence TS generated by the UE; Features are generated based on the received signal; The TS used to modulate the signal is obtained; and A training example including the features and a label is generated, where the label includes the generated TS.

30. The network node (104) according to claim 29, wherein, The network node is also operable to perform the method according to any one of claims 15 to 23.

Images