Online retraining of physical layer transceivers based on machine learning

Abstract

Disclosed is a radio receiver system (1), comprising one or more digital sample buffers (11), a physical, PHY, layer circuit (12), retraining performer (13), and a retraining controller (14). The one or more digital sample buffers (11) are configured to buffer received blocks of receive signals relating to data packets. The PHY layer circuit (12) is configured to derive binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions (121), and send the correctness information to a retraining controller (14) of the system (1). The retraining controller (14) is configured to selectively trigger a retraining operation of the model by a retraining performer (13) of the system (1) in accordance with the buffered blocks of receive signals and the associated correctness information. The retraining performer (13) is configured to perform the retraining operation of the model in accordance with the retraining trigger by the retraining controller (14). The PHY layer circuit (12) is further configured to update the model in accordance with the retraining operation. This confers an online retraining capability to PHY transceivers.

Description

[0001] ONLINE RETRAINING OF PHYSICAL LAYER TRANSCEIVERS BASED ON MACHINE LEARNING

[0002] Technical Field

[0003] The present disclosure relates generally to the field of wireless communication systems, and more particularly to a radio receiver system and a method of operating the same.

[0004] Background Art

[0005] The use of Artifical Intelligence (Al) and Machine Learning (ML) models to jointly design transmitter (Tx) and receiver (Rx) functions in wireless communication systems attracts great attention.

[0006] For example, autoencoders (AEs) may be used to jointly design modulation and demodulation functions. AEs are a type of artificial neural networks being used to learn efficient codings for arbitrary input data. AEs have two components, namely an encoder and a decoder. The encoder compresses the input data into a more compact representation (a.k.a. “latent space”) of the data, while the decoder tries to restore the compact representation into the original input information. In other words, AEs learn a representation (encoding) for arbitrary input data by training the neural network to find a compact representation of the data. If the encoder and decoder parts of an AE are placed remotely and the encoder sends the encoded information over a communication channel, the concept may be used to perform communication. More sophisticated AEs have been proposed to perform modulation and demodulation in combination with a Low Density Parity Check (LDPC) channel decoder, thereby improving a performance in terms of Packet Error Rate (PER) considerably at similar complexity.

[0007] Al and ML models have also been applied to other Physical Layer (PHY) functions, such as joint design of transmit (Tx) and receive (Rx) filters of a communication system, which allows to overcome physical limiting factors of communication systems such as Peak to Average Power Ratio (PAPR) and Adjacent Channel Leakage Ratio (ACER). Al and ML models have also been applied to Multiple Input Multiple Output (MIMO) communication systems, Channel State Information (CSI) compression and adquisition, linear precoding, or coding, amongst many others.

[0008] All examples mentioned can be categorized as AI / ML-based communication systems having several issues that hinder their potential to be implemented in real systems:

[0009] First, AI / ML-based systems generally involve two independent operational phases for their design and utilization: a training phase and a deployment / inference phase. During training, the system learns from data how to perform a function. During deployment / inference, the system performs the learned function. If new data is available after training, the system would have to be retrained from the beginning, which induces unacceptable levels of computation, storage, and transmission in real systems.

[0010] Second, AI / ML-based systems are not able to adapt to environments unseen during training. This implies that a performance of AI / ML-based communication systems may degrade severely if the environment is new and unseen in training, limiting their potential to be implemented in real systems.

[0011] Summary

[0012] The foregoing and other drawbacks are overcome by the features of the independent claims. Further implementation forms are apparent from the dependent claims, the description and the figures.

[0013] According to a first aspect, a radio receiver system is provided, comprising one or more digital sample buffers, a physical, PHY, layer circuit, retraining performer, and a retraining controller. The one or more digital sample buffers are configured to buffer received blocks of receive signals relating to data packets. The PHY layer circuit is configured to derive binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions; send the correctness information to a retraining controller of the system; and update the model in accordance with a retraining operation. The retraining controller is configured to selectively trigger the retraining operation of the model by a retraining performer of the system in accordance with the buffered blocks of receive signals and the associated correctness information. The retraining performer is configured to perform the retraining operation of the model in accordance with the retraining trigger by the retraining controller.

[0014] A “radio receiver” as used herein may refer to an electronic device being configured to receive a radio signal and derive (usually demodulate, equalize and decode) information carried by the same.

[0015] A “digital sample buffer” as used herein, may refer to a digital memory being configured to store digital samples of a received radio signal.

[0016] A “PHY layer circuit” as used herein may refer to an electronic device implementing PHY layer functionality such as (de)modulation, equalization, (de)coding, and the like.

[0017] A “retraining controller” as used herein may refer to a logic, implemented in hardware and / or software, being configured to decide on a retraining of machine learning models.

[0018] A “retraining performer” as used herein may refer to a logic, implemented in hardware and / or software, being configured to carry out a retraining of machine learning models.

[0019] A “model” as used herein may refer to a representation of an object, such as a PHY layer circuit, covering those attributes of the object being relevant for its proper functioning. The model may particularly be based on artificial intelligence (Al) and / or machine learning (ML) concepts. Such models have the capacity to learn the particular function to perform from data. An AI / ML model consists of an architecture or design, e.g., in case of neural networks the number of layers, neurons per layer, and so on, constitute the design, and its parameters, i.e., the values of the variables of the model and that change during the learning process, such as the values of the weights corresponding to the layers of a neural network, the values of the splitting branches in a tree-based method (e.g. random forest), or a mathematical description (e.g. equation) of the boundaries between different classes in Support Vector Machines (SVMs). An AI / ML model may be exchanged between system entities by providing its parameters only, if the involved entities know the design, or by providing the model as such (i.e., the combination of design plus parameters) if there is no such a-priori knowledge of the model available.

[0020] Receive signals as used herein may refer to the discrete complex values forming the information being carried by (i.e., modulated on and demodulated from) a radio signal. For example, quadrature phase-shift keying (QPSK) involves four discrete complex values having unit magnitude and respective phases of 7t / 4, 3TC / 4, 5K / 4 and 7K / 4, which may require equalization (i.e., reversal of distortion incurred by the transmitted signal transmitted through the radio channel) on the receive side.

[0021] Binary data as used herein may refer to binary information being derivable from the receive signals being demodulated from a radio signal. For example, each modulation symbol derived from the receive signals may yield one or more bits of the binary data. Each of the four discrete complex values in the QPSK example respectively represents two bits, namely „11“, „01“, „00“ and „10“.

[0022] The radio receiver system facilitates local, continuous and proactive adaptation of the model of the PHY layer functions, and thereby allows for fast adaptation to changes in the environment e.g., changing channel distributions, channel conditions, etc. allows for system specialization to particular and complex environments (factories, indoors, etc). reduces / eliminates a need of generating big datasets for the training phase, reduces / eliminates feedback information related to the retraining operation, offloads computational burden of retraining to the receiver, eliminates additional signaling overhead, which implies an increase of spectral efficiency and reduced overhead, avoids a performance degradation of the communication system, and avoids unacceptable downtime of the communication system. In a possible implementation form, the retraining trigger may comprise one of: a negative retraining request, if a maximum number of HARQ retransmissions of any one of the buffered blocks of receive signals is reached; and a positive retraining request, if a retraining of the model is needed in accordance with the buffered blocks of receive signals and the associated correctness information.

[0023] Hybrid automatic repeat request (HARQ) as used herein may refer to a combination of the errorcontrol functions of forward error correction (FEC) and automatic repeat request (ARQ) signaling. FEC, also known as channel coding, encodes binary data to be sent in a redundant way, thereby enabling error detection and error correction at the receiver. For example, a parity bit represents a simple form of FEC enabling detection of single-bit errors and no correction. ARQ signaling requests retransmission of erroneous binary data by the transmitter if transmission errors remain.

[0024] This facilitates the retraining of the communication system in accordance with the existing HARQ functionality on the MAC layer, thereby integrating well with the existing base of communication systems.

[0025] In a possible implementation form, the retraining trigger of the positive retraining request may further comprise one or more of: an identifier of a user associated with the data packets to be used for the retraining operation; an identifier of a flow associated with the data packets to be used for the retraining operation; one or more identifiers of the data packets to be used for the retraining operation; one or more identifiers of the digital sample buffers buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and an indication of the buffered blocks of receive signals to be used for the retraining operation.

[0026] This facilitates the retraining of the communication system with respect to exactly the particular user, flow, data packets and buffers stimulating the retraining.

[0027] In a possible implementation form, the correctness information may comprise correctness probabilities and / or correctness checksums. This facilitates the decision-making for the retraining by providing a definitive judgment of incorrectness (correctness checksums) as well as a probabilistic degree of incorrectness (correctness probabilities, i.e. “beliefs”).

[0028] In a possible implementation form, the system may further comprise a medium access control, MAC, controller. The MAC controller may be configured to request a radio transmitter in communication with the system to perform hybrid automatic repeat request, HARQ, retransmissions of blocks of receive signals relating to incorrect binary data in accordance with the associated correctness checksums. The PHY layer circuit may further be configured to send the derived binary data and the correctness checksums to the MAC controller.

[0029] A MAC controller as used herein may refer to an electronic device implementing MAC layer functionality for control of the (PHY layer) hardware responsible for interaction with the actual transmission medium. The MAC layer provides a control abstraction of the PHY layer to upper layers of the network stack. For example, a MAC controller typically implements an (H)ARQ signaling of a HARQ funtionality.

[0030] This facilitates a guidance of the retraining of the communication system in accordance with the existing HARQ functionality on the MAC layer, thereby integrating well with the existing base of communication systems.

[0031] In a possible implementation form, the MAC controller may further be configured to send an indication of the blocks of receive signals relating to incorrect binary data to the retraining controller.

[0032] An indication as used herein may refer to a reference to a memory address (i.e., pointer), processor register, identifier or the like specifying where the blocks of receive signals are stored.

[0033] This facilitates the retraining of the communication system with respect to exactly the particular data stimulating the retraining. In a possible implementation form, the indication may comprise one or more of: an identifier of a user associated with the data packets to be used for the retraining operation; an identifier of a flow associated with the data packets to be used for the retraining operation; one or more identifiers of the data packets to be used for the retraining operation; one or more identifiers of the digital sample buffers buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and the maximum number of HARQ requests for each of the buffered blocks of receive signals.

[0034] This facilitates the retraining of the communication system with respect to exactly the particular user, flow, data packets and buffers stimulating the retraining.

[0035] In a possible implementation form, the radio receiver PHY layer functions may comprise a decoding function.

[0036] Decoding as used herein may refer to a PHY layer function being responsible for decoding, in particular channel decoding, of binary data derived (demodulated) from a received radio signal. One popular example of (de)coding involves low-density parity codes (LDPC).

[0037] This facilitates applying the radio receiver system for purely receiver-based decoding.

[0038] In a possible implementation form, the model may further comprise one or more radio transmitter PHY layer functions; and the PHY layer circuit may further be configured to send parameters of the model relating to the one or more radio transmitter PHY layer functions to the radio transmitter in communication with the system.

[0039] This facilitates applying the radio receiver system for joint retraining of radio transmitter PHY layer functions and radio receiver PHY layer functions. In a possible implementation form, the radio receiver PHY layer functions may comprise a demodulation function; and the radio transmitter PHY layer functions may comprise a modulation function.

[0040] Demodulation as used herein may refer to a PHY layer function being responsible for deriving binary data from receive signals carried by radio waves.

[0041] Modulation as used herein may refer to a PHY layer function being responsible for imparting binary data on radio waves in the form of modulation symbols.

[0042] This facilitates applying the radio receiver system for joint modulation and demodulation, and optionally decoding as well.

[0043] In a possible implementation form, the radio receiver PHY layer functions may comprise a beamcombiner function; and the radio transmitter PHY layer functions may comprise a codebook-based beam-former function.

[0044] Beam-combining as used herein may refer to a PHY layer function being responsible for directional reception of radio waves, by combining elements in antenna arrays in such a way that signals at particular angles experience constructive interference while others experience destructive interference. Beam-forming as used herein may refer to a PHY layer function being responsible for directional transmission of radio waves, using similar techniques.

[0045] This facilitates applying the radio receiver system for joint beam-forming and beam-combining.

[0046] In a possible implementation form, the system may be configured to implement a UE-to-UE interface of user equipments, UEs, in communication with one another.

[0047] User equipments as used herein may refer to terminals or terminal devices in cellular communication networks. In a possible implementation form, the UE-to-UE interface may comprise a PC5 sidelink, SL, interface.

[0048] In a possible implementation form, the PHY layer circuit may further be configured to send the parameters of the model relating to the one or more radio transmitter PHY layer functions via a Physical Sidelink Control Channel, PSCCH of the PC5 sidelink interface.

[0049] This facilitates applying the radio receiver system for device-to-device (D2D) communication in cellular communication networks, and in particular for PC5 SL communication via PSCCH in 5G cellular communication networks.

[0050] D2D or SL communication as used herein may refer to a direct communication between UEs in cellular communication networks.

[0051] In a possible implementation form, the system may be configured to implement a Uu downlink, DL, interface of a UE in communication with a base station, of the cellular communication network.

[0052] DL communication as used herein may refer to a communication from a base station to a UE in cellular communication networks.

[0053] In a possible implementation form, the PHY layer circuit may further be configured to send the parameters of the model relating to the one or more radio transmitter PHY layer functions via a Physical Uplink Control Channel, PUCCH of a corresponding Uu uplink, UL, interface of the implemented Uu DL interface.

[0054] UL communication as used herein may refer to a communication from a UE to a base station in cellular communication networks. This facilitates applying the radio receiver system for UL communication in cellular communication networks, and in particular for Uu UL communication via PUCCH in 5G cellular communication networks.

[0055] In a possible implementation form, the system may be configured to implement a Uu UL interface of a base station of the cellular communication network.

[0056] In a possible implementation form, the PHY layer circuit may further be configured to send the parameters of the model relating to the one or more radio transmitter PHY layer functions via one of: a Physical Downlink Control Channel, PDCCH of a corresponding Uu DL interface of the implemented Uu UL interface; and a Broadcast Control Channel, BCCH.

[0057] This facilitates applying the radio receiver system for DL communication in cellular communication networks, and in particular for Uu DL communication in 5G via PDCCH or BCCH cellular communication networks.

[0058] In a possible implementation form, the PHY layer circuit may form part of a distributed unit, DU, of the base station; the MAC controller may form part of a centralized unit, CU, of the base station; and the DU and the CU may be in communication via an Fl interface of a DU-CU split base station architecture.

[0059] A DU-CU split base station architecture as used herein may refer to a functional split of base station functionality into distributed units (DU) arranged closer to the UE and centralized units (CU) arranged closer to the core of the cellular communication network. Different functional splits are possible. For example, according to a particular functional split, DUs may comprise PHY layer functionality only and CUs may comprise MAC layer functionality only.

[0060] In a possible implementation form, the retraining controller may form part of the DU.

[0061] In a possible implementation form, the retraining controller may form part of the CU. This facilitates applying the radio receiver system to split base station architectures in cellular communication networks.

[0062] In a possible implementation form, the retraining performer may form part of the PHY layer circuit.

[0063] This facilitates applying the radio receiver system to straightforward, receiver-integrated implementations.

[0064] In a possible implementation form, the retraining controller and the retraining performer may form part of a RAN data analytics function, RANDAF, of the cellular communication network; and the RANDAF and the base station may be in communication via an Xn interface or an interface for data exchange associated with the model.

[0065] This facilitates applying the radio receiver system to implementations with a centralized radio access network (RAN) function in charge of dealing with the training / retraining / monitoring / maintenance of AI / ML models, which may be called RAN data analytics function.

[0066] In a possible implementation form, the retraining operation of the model may be based on one of: supervised learning in accordance with input data and output data; unsupervised learning in accordance with the input data; and reinforcement learning in accordance with the input data, an internal state of the model, and a plurality of given actions.

[0067] Supervised learning, unsupervised learning and reinforcement learning as used herein denote different machine learning techniques.

[0068] This facilitates retraining based on machine learning techniques.

[0069] In a possible implementation form, the input data may comprise one or more of: binary data; a number of Physical Resource Blocks, PRBs; channel state information, CSI; predictive quality of service, QoS, information; location information; velocity information; geographic information of the environment; a receive signal; blocks of receive signals; beamforming selection related information; and control information.

[0070] A PRB as used herein may refer to a smallest element of resource allocation assigned by a scheduler of a base station. Being a time-frequency radio resource, a PRB may, for example, span 12 consecutive subcarriers (corresponding to a bandwidth of 180 kHz with a 15-kHz subcarrier spacing) over 7 OFDM symbols (corresponding to a time slot of 0,5 ms). Moreover, 5G cellular communication networks support other numerologies that allow many combinations of sub-carrier spacing and number of OFDM symbols per sub-frame to support loT / URLLC communications.

[0071] CSI as used herein may refer to information on current channel conditions which can be viewed as knowing the time / frequency response of a digital filter. CSI acquisition facilitates adapting radio transmission to current channel conditions so as to achieve low bit error rates.

[0072] QoS as used herein may refer to a measure or figure of merit of an overall performance of a service, particularly the performance seen by a user of a communication network.

[0073] Location information as used herein may refer to geographic coordinates (latitude and longitude measured in degrees, minutes and seconds) obtainable by means of localization reference signals or satellite navigation, for example.

[0074] Velocity information as used herein may refer to an indication of speed (measured in m / s) and direction derivable from the localization information mentioned previously or obtainable by means of satellite navigation, for example.

[0075] Geographic information as used herein may refer to a geographical map of the environment, for example.

[0076] Beamforming selection related information as used herein may refer to an index into a finite codebook of a codebook-based beam-former function, for example. Control information as used herein may refer to PRB assignment, for example.

[0077] In a possible implementation form, the output data may comprise one or more of: a transmit signal; blocks of receive signals; binary data; and precoding / combining vectors / codewords.

[0078] According to a second aspect, a method of operating a radio receiver system is provided. The method comprises: buffering received blocks of receive signals relating to data packets; deriving binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions; sending the correctness information to a retraining controller of the system; selectively triggering by the retraining controller a retraining operation of the model in accordance with the buffered blocks of receive signals and the associated correctness information; performing the retraining operation of the model in accordance with the retraining trigger; and updating the model in accordance with the retraining operation.

[0079] The method may be performed by the system of the first aspect or any of its implementations.

[0080] According to a third aspect, a computer program is provided, comprising executable instructions which, when executed by a processor, cause the processor to perform the method of the second aspect or any of its implementations.

[0081] The technical effects and advantages described above in relation with the radio receiver system equally apply to the method and the computer program having corresponding features.

[0082] Brief Description of Drawings

[0083] The above-described aspects and implementations will now be explained with reference to the accompanying drawings, in which the same or similar reference numerals designate the same or similar elements. The drawings are to be regarded as being schematic representations, and elements illustrated in the drawings are not necessarily shown to scale. Rather, the various elements are represented such that their function and general purpose become apparent to those skilled in the art.

[0084] FIG. 1 illustrates a radio receiver system in accordance with the present disclosure;

[0085] FIGs. 2 - 5 illustrate various implementations of the PHY layer circuit of the system of FIG. 1;

[0086] FIGs. 6 - 9 illustrate various implementations of a PHY layer circuit of a radio transmitter in communication with the system of FIG. 1;

[0087] FIG. 10 illustrates a decoding use case of the radio receiver system in accordance with FIGs. 1, 3 and 7;

[0088] FIG. 11 illustrates a joint modulation and demodulation use case of the radio receiver system in accordance with FIGs. 1, 4 and 8;

[0089] FIG. 12 illustrates a joint modulation, demodulation and decoding use case of the radio receiver system in accordance with FIGs. 1, 3, 4, 7 and 8;

[0090] FIG. 13 illustrates a joint beamforming and combining use case of the radio receiver system in accordance with FIGs. 1, 5 and 9;

[0091] FIG. 14 illustrates deployment opportunities for the radio receiver system of FIG. 1 in a cellular communication network;

[0092] FIGs. 15 - 16 illustrate implementation options of the radio receiver system in accordance with the present disclosure in a cellular communication network having a split base station architecture;

[0093] FIG. 17 illustrates a further implementation option of the radio receiver system in accordance with the present disclosure in a cellular communication network having a separate RAN function dealing with training, retraining, monitoring, and maintenance of AI / ML models;

[0094] FIG. 18 illustrates a further implementation option of the radio receiver system in accordance with the present disclosure in a cellular communication network having a split base station architecture as well as a separate RAN function dealing with training, retraining, monitoring, and maintenance of AI / ML models ; and

[0095] FIG. 19 illustrates a method of operating a radio receiver system in accordance with the present disclosure.

[0096] Detailed Descriptions of Drawings

[0097] In the following description, reference is made to the accompanying drawings, which form part of the disclosure, and which show, by way of illustration, specific aspects of embodiments of the present disclosure or specific aspects in which embodiments of the present disclosure may be used. It is understood that embodiments of the present disclosure may be used in other aspects and comprise structural or logical changes not depicted in the figures. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present disclosure is defined by the appended claims.

[0098] For instance, it is understood that a disclosure in connection with a described method may also hold true for a corresponding apparatus or system configured to perform the method and vice versa. For example, if one or a plurality of specific method steps are described, a corresponding device may include one or a plurality of units, e.g. functional units, to perform the described one or plurality of method steps (e.g. one unit performing the one or plurality of steps, or a plurality of units each performing one or more of the plurality of steps), even if such one or more units are not explicitly described or illustrated in the figures. On the other hand, for example, if a specific apparatus is described based on one or a plurality of units, e.g. functional units, a corresponding method may include one step to perform the functionality of the one or plurality of units (e.g. one step performing the functionality of the one or plurality of units, or a plurality of steps each performing the functionality of one or more of the plurality of units), even if such one or plurality of steps are not explicitly described or illustrated in the figures. Further, it is understood that the features of the various exemplary embodiments and / or aspects described herein may be combined with each other, unless specifically noted otherwise.

[0099] FIG. 1 illustrates a radio receiver system 1 in accordance with the present disclosure.

[0100] The radio receiver system 1 shown on the right of FIG. 1 may be in communication with a radio transmitter (system) 3 suggested on the left of FIG. 1 via an intervening communication channel 2.

[0101] In accordance with FIG. 1, the radio transmitter 3 may comprise a medium access control, MAC, controller 35 and a physical, PHY, layer circuit 32. The radio receiver system 1 comprises a PHY layer circuit 12 on its part, and may further comprise a MAC controller 15.

[0102] The MAC controller 15 of the radio receiver system 1 may be configured to request the MAC controller 35 of the radio transmitter 3 to perform hybrid automatic repeat request, HARQ, retransmissions of blocks of receive signals relating to incorrect binary data in accordance with associated correctness checksums. In response to such HARQ requests, the MAC controller 35 of the radio transmitter system 3 may be configured to perform the requested packet retransmissions. In more detail, the MAC controller 35 of the radio transmitter system 3 may be arranged to determine if there are pending HARQ requests to retransmit. If so, the MAC controller 35 passes to its PHY layer circuit 32 previously transmitted information bits, otherwise it passes new information bits. The PHY layer circuit 32 is arranged to convert the information bits into transmit signals. To generate the transmit signals, the PHY layer circuit 32 can optionally exploit supplementary input data such as Channel State Information (CSI), number of Physical Resource Blocks (PRBs), or geographic location information of the involved radio transmitter 3 and radio receiver 1. The supplementary input depends on the exact function(s) implemented by the PHY layer circuit 32. The generated signals are then transmitted over the communication channel 2 to the radio receiver system 1. In accordance with FIG. 1, the radio receiver system 1 further comprises one or more digital sample buffers 11 being configured to buffer received blocks of receive signals relating to data packets. The received signals are preemptively stored, in case the received packets are wrongly decoded, in which case they might be used later on for retraining.

[0103] The PHY layer circuit 12 is configured to derive binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions 121.

[0104] The PHY layer circuit 12 is further configured to generate “beliefs” associated with the decoded bits, i.e., the probability that the decoder associates with each bit being 1 or 0, to check if those bits contain errors through a standard cyclic redundancy check (CRC) mechanism, and to delete derived binary data associated with successful decoding from the digital sample buffers 11 to avoid an overflow of the same.

[0105] The correctness information may therefore comprise correctness probabilities (“beliefs”) and / or correctness checksums.

[0106] The PHY layer circuit 12 may further be configured to send the derived binary data and the correctness checksums to the MAC controller 15. That is to say, the PHY layer circuit 12 is configured to pass the decoded bits / packets together with their CRC checksums to its MAC controller 15. The MAC controller 15 of the radio receiver system 1 is configured to send the afore-mentioned HARQ requests to the MAC controller 35 of the radio transmitter 3 for retransmission in response to incorrect binary data in accordance with associated correctness checksums, and to pass correct binary data to higher layers for further processing.

[0107] In accordance with FIG. 1, the radio receiver system 1 further comprises a retraining controller 14, being in charge of triggering a retraining when possible and necessary.

[0108] The PHY layer circuit 12 is configured to send © the correctness information, i.e., the generated CRCs and beliefs, to the retraining controller 14. The MAC controller 15 may further be configured to send ® an indication of the blocks of receive signals relating to incorrect binary data to the retraining controller 14.

[0109] The indication may comprise one or more of: an identifier of a user (“user ID”) associated with the data packets to be used for the retraining operation; an identifier of a flow (“flow ID”) associated with the data packets to be used for the retraining operation; one or more identifiers of the data packets (“packet IDs”) to be used for the retraining operation; one or more identifiers of the digital sample buffers 11 (“Rx buffer ID”) buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and the maximum number of HARQ requests for each of the buffered blocks of receive signals.

[0110] With this information, together with the CRCs and beliefs provided by the PHY layer circuit 12, the retraining controller 14 is in a position to decide if a retraining operation is both possible and necessary.

[0111] In more detail, the retraining of the model is possible if a set of the buffered blocks of receive signals relates to correct binary data in accordance with the associated correctness checksums, and the retraining of the model is necessary if if the correctness information indicates a transmission performance being worse than expected. As an example, successful decoding may be impossible if a signal-to-noise ratio (SNR) of the receive signal is -20 dB. So, even if there are errors, the retraining will be deemed unnecessary because the system is behaving as expected.

[0112] In accordance with FIG. 1, the radio receiver system 1 further comprises a retraining performer 13.

[0113] The retraining controller 14 is configured to selectively trigger ® the retraining operation of the model by the retraining performer 13 in accordance with the buffered blocks of receive signals and the associated correctness information.

[0114] The retraining trigger may comprise one of: a negative retraining request, if a maximum number of HARQ retransmissions of any one of the buffered blocks of receive signals is reached, in which case the PHY layer circuit 12 would flush the digital sample buffers 11 to empty space for new incoming Rx signals. Negative retraining request may be issued for other reasons as well.

[0115] Alternatively, the retraining trigger may comprise a positive retraining request, if a retraining of the model is needed in accordance with the buffered blocks of receive signals and the associated correctness information, in which case the radio receiver system 1 is subjected to the retraining operation.

[0116] The retraining trigger of the positive retraining request may further comprise one or more of: an identifier of a user (“user ID”) associated with the data packets to be used for the retraining operation; an identifier of a flow (“flow ID“) associated with the data packets to be used for the retraining operation; one or more identifiers of the data packets (“packet IDs”) to be used for the retraining operation; one or more identifiers of the digital sample buffers 11 (“Rx Buffer ID“) buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and an indication of the buffered blocks of receive signals to be used for the retraining operation.

[0117] The retraining performer 13, which according to FIG. 1 may form part of the PHY layer circuit 12, is configured to perform @ the retraining operation of the model in accordance with the retraining trigger by the retraining controller 14.

[0118] The retraining operation of the model may be based on one of: supervised learning in accordance with input data and output data; unsupervised learning in accordance the input data; and reinforcement learning in accordance with the input data, an internal state of the model, and a plurality of given actions.

[0119] The input data may comprise one or more of: binary data; a number of Physical Resource Blocks, PRBs; channel state information, CSI; predictive quality of service, QoS, information; location information; velocity information; geographic information of the environment; a receive signal; blocks of receive signals; beamforming selection related information; and control information. The output data may comprise one or more of: a transmit signal; blocks of receive signals; binary data; and precoding / combining vectors / codewords.

[0120] Subsequently to the retraining operation, the PHY layer circuit 12 of the radio receiver system 1 may be configured to flush the digital sample buffers 11.

[0121] If the retraining operation covers one or more radio receiver PHY layer functions 121 and one or more radio transmitter PHY layer functions 321 (“joint retraining”) in accordance with FIGs. 11 to 13 below, the PHY layer circuit 12 may further be configured to send ® parameters of the model relating to the one or more radio transmitter PHY layer functions 121 to the radio transmitter 3 in communication with the system 1.

[0122] The PHY layer circuit 12 of the radio receiver system 1 is further configured to update ® the model in accordance with the retraining operation. This relates to the one or more radio receiver PHY layer functions 121 as well as to the one or more radio receiver PHY layer functions 121, if any.

[0123] FIGs. 2 - 5 illustrate various implementations of the PHY layer circuit 12 of the system 1 of FIG. 1.

[0124] In accordance with FIG. 2, the PHY layer circuit 12 comprises the one or more radio receiver PHY layer functions 121.

[0125] In accordance with FIG. 3, the radio receiver PHY layer functions 121 may comprise a decoding function 121C.

[0126] In accordance with FIG. 4, the radio receiver PHY layer functions 121 may comprise a demodulation function 12 IB.

[0127] In accordance with FIG. 5, the radio receiver PHY layer functions 121 may comprise a beamcombiner function 121 A. The radio receiver PHY layer functions 121A, 121B, 121C may be combined to form more comprehensive models covering multiple PHY layer functions (see FIG. 12, for example).

[0128] In particular, the model of the one or more radio receiver PHY layer functions 121 may further comprise one or more radio transmitter PHY layer functions 321 for the purpose of joint retraining.

[0129] FIGs. 6 - 9 illustrate various implementations of a PHY layer circuit 32 of a radio transmitter 3 in communication with the system 1 of FIG. 1.

[0130] In accordance with FIG. 6, the PHY layer circuit 32 of the radio transmitter 3 being in communication with the radio receiver system 1 may comprise one or more radio transmitter PHY layer functions 321.

[0131] In accordance with FIG. 7, the radio transmitter PHY layer functions 321 may comprise a coding function 321C.

[0132] In accordance with FIG. 8, the radio transmitter PHY layer functions 321 may comprise a modulation function 32 IB.

[0133] In accordance with FIG. 9, the radio transmitter PHY layer functions 321 may comprise a codebook-based beam-former function 321 A.

[0134] The radio receiver PHY layer functions 121A, 121B, 121C may be combined to form more comprehensive models (see FIG. 12, for example).

[0135] FIG. 10 illustrates a decoding use case of the radio receiver system 1 in accordance with FIGs. 1, 3 and 7.

[0136] In particular, the radio receiver PHY layer functions 121 may comprise a decoding function 121C in accordance with FIG. 3, and the radio transmitter PHY layer functions 321 may comprise a corresponding coding function 321C in accordance with FIG. 7. In this use case, there is no need of exchange of information between PHY Rx and PHY Tx as the model covers (parts of) the PHY Rx only.

[0137] FIG. 11 illustrates a joint modulation and demodulation use case of the radio receiver system 1 in accordance with FIGs. 1, 4 and 8.

[0138] In particular, the radio receiver PHY layer functions 121 may comprise a demodulation function 121B in accordance with FIG. 4, and the radio transmitter PHY layer functions 321 may comprise a modulation function 321B in accordance with FIG. 8.

[0139] Additionally, the model of the one or more radio receiver PHY layer functions 121 may further comprise one or more radio transmitter PHY layer functions 321 for the purpose of joint retraining.

[0140] Thus, the PHY layer circuit 12 may further be configured to send ® parameters of the model relating to the one or more radio transmitter PHY layer functions 121 to the radio transmitter 3 in communication with the system 1.

[0141] That is to say, if the online retraining process of the model for both the PHY Tx and PHY Rx entities happens at the Rx side, then the PHY Rx entity provides to the PHY Tx entitiy an updated model (or parameters of the model) of the one or more physical layer functions.

[0142] FIG. 12 illustrates a joint modulation, demodulation and decoding use case of the radio receiver system 1 in accordance with FIGs. 1, 3, 4, 7 and 8.

[0143] In particular, the radio receiver PHY layer functions 121 may comprise a demodulation function 121B in accordance with FIG. 4 and may further comprise a decoding function 121C in accordance with FIG. 3.

[0144] Additionally, the model of the one or more radio receiver PHY layer functions 121 may further comprise one or more radio transmitter PHY layer functions 321 for the purpose of joint retraining. The radio transmitter PHY layer functions 321 may comprise a modulation function 32 IB in accordance with FIG. 8, and may further comprise a coding function 321C in accordance with FIG. 7.

[0145] The PHY layer circuit 12 may further be configured to send ® parameters of the model relating to the one or more radio transmitter PHY layer functions 121 to the radio transmitter 3 in communication with the system 1.

[0146] FIG. 13 illustrates a joint beamforming and combining use case of the radio receiver system 1 in accordance with FIGs. 1, 5 and 9.

[0147] In particular, the radio receiver PHY layer functions 121 may comprise a beam-combiner function 121 A in accordance with FIG. 5.

[0148] Additionally, the model of the one or more radio receiver PHY layer functions 121 may further comprise one or more radio transmitter PHY layer functions 321 for the purpose of joint retraining.

[0149] The radio transmitter PHY layer functions 321 may comprise a codebook-based beam-former function 321 A in accordance with FIG. 9.

[0150] The PHY layer circuit 12 may further be configured to send ® parameters of the model relating to the one or more radio transmitter PHY layer functions 121 to the radio transmitter 3 in communication with the system 1.

[0151] FIG. 14 illustrates deployment opportunities for the radio receiver system 1 of FIG. 1 in a cellular communication network.

[0152] First, the system 1 may be configured to implement a UE-to-UE interface of user equipments, UEs, in communication with one another.

[0153] The UE-to-UE interface may comprise a PC5 sidelink, SL, interface. In this use case, the PHY layer circuit 12 may be configured to send ® the parameters of the model relating to the one or more radio transmitter PHY layer functions 321 via a Physical Sidelink Control Channel, PSCCH of the PC5 sidelink interface.

[0154] Second, the system 1 may be configured to implement a Uu downlink, DL, interface of a UE in communication with a base station (BS), of the cellular communication network.

[0155] In this use case, the PHY layer circuit 12 may be configured to send ® the parameters of the model relating to the one or more radio transmitter PHY layer functions 321 via a Physical Uplink Control Channel, PUCCH of a corresponding Uu uplink, UL, interface of the implemented Uu DL interface.

[0156] Third, the system 1 may be configured to implement a Uu UL interface of a base station of the cellular communication network.

[0157] In this use case, the PHY layer circuit 12 may be configured to send ® the parameters of the model relating to the one or more radio transmitter PHY layer functions 321 via one of a Physical Downlink Control Channel, PDCCH of a corresponding Uu DL interface of the implemented Uu UL interface; and a Broadcast Control Channel, BCCH.

[0158] FIGs. 15 - 16 illustrate implementation options of the radio receiver system 1 in accordance with the present disclosure in a cellular communication network having a split base station architecture.

[0159] The PHY layer circuit 12 may form part of a distributed unit, DU 1’, of the base station; the MAC controller 15 may form part of a centralized unit, CU 1“, of the base station; and the DU 1’ and the CU 1“ may be in communication via an Fl control plane interface of the DU-CU split base station architecture (e.g. option 6 of functional splitting according to 3GPP TS 38.401, 3GPP TR 38.801).

[0160] In accordance with FIG. 15, the retraining controller 14 may form part of the DU 1’.

[0161] In accordance with FIG. 16, the retraining controller 14 may form part of the CU 1“. FIG. 17 illustrates a further implementation option of the radio receiver system 1 in accordance with the present disclosure in a cellular communication network having a separate RAN data analytics function (RANDAF) 4 dealing with training, retraining, monitoring, and maintenance of AI / ML models.

[0162] The retraining controller 14 and the retraining performer 13 may form part of the RANDAF 4 of the cellular communication network, and the PHY layer circuit 12 and the MAC controller 15 may respectively form part of the base station 1.

[0163] Collectively, the retraining controller 14 and the retraining performer 13 may also be termed a retraining function.

[0164] The RANDAF 4 and the base station 1 may be in communication via one of: an Xn interface; an N2 interface between BS and a core network (CN) of the cellular communication network in case the RANDAF is implemented in the CN or the Xn interface is not available at the base station 1; and an interface specifically for data exchange associated with the model. For example, a control channel dedicated to the exchange of information related to the training / retraining / monitoring / maintenance of AI / ML models between a RANDAF 4 and all or some base stations 1 contained in a logical cluster may be used.

[0165] The retraining controller 14 is configured to selectively trigger ® the retraining operation of the model by the retraining performer 13 in accordance with the buffered blocks of receive signals and the associated correctness information.

[0166] In this implementation option, the retraining performer 13 is configured to perform @ the retraining operation of the model in accordance with the retraining trigger by the retraining controller 14, and to send ® the updated model of the PHY layer circuits 12, 32 (or their parameters) to the radio receiver system 1 and to the radio transmitter 3 respectively being in communication with the system 1. FIG. 18 illustrates a further implementation option of the radio receiver system 1 in accordance with the present disclosure in a cellular communication network having a split base station architecture as well as a separate RAN function.

[0167] The retraining controller 14 and the retraining performer 13 may again form part of the RANDAF 4 of the cellular communication network, while the PHY layer circuit 12 may form part of the distributed unit, DU 1’, of the split base station and the MAC controller 15 may form part of the centralized unit, CU 1“, of the split base station.

[0168] The DU 1’ and the CU 1“ may be in communication as mentioned in connection with FIGs. 15 - 16.

[0169] The RANDAF 4 and the base station 1 may be in communication as mentioned in connection with FIG. 17.

[0170] The retraining controller 14 is configured to selectively trigger ® the retraining operation of the model by the retraining performer 13 in accordance with the buffered blocks of receive signals and the associated correctness information.

[0171] In this implementation option, the retraining performer 13 is configured to perform @ the retraining operation of the model in accordance with the retraining trigger by the retraining controller 14, and to send ® the updated model of the PHY layer circuits 12, 32 (or their parameters) to the radio receiver system 1 and to the radio transmitter 3 respectively being in communication with the system 1.

[0172] FIG. 19 illustrates a method of operating a radio receiver system 1 in accordance with the present disclosure.

[0173] The method may be performed by the system 1 of the first aspect or any of its implementations.

[0174] The method 5 comprises buffering 51 received blocks of receive signals relating to data packets. The method 5 further comprises deriving 52 binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions 121.

[0175] The method 5 further comprises sending 53, © the correctness information to a retraining controller 14 of the system 1.

[0176] The method 5 further comprises selectively triggering 54, ® by the retraining controller 14 a retraining operation of the model in accordance with the buffered blocks of receive signals and the associated correctness information.

[0177] The method 5 further comprises performing 55, @ the retraining operation of the model in accordance with the retraining trigger.

[0178] The method 5 further comprises updating 56, ® the model in accordance with the retraining operation.

[0179] The present disclosure has been described in conjunction with various embodiments as examples as well as implementations. However, other variations can be understood and effected by those persons skilled in the art and practicing the claimed matter, from the studies of the drawings, this disclosure and the independent claims. In the claims as well as in the description the word “comprising” does not exclude other elements or steps and the indefinite article “a” or “an” does not exclude a plurality. A single element or other unit may fulfill the functions of several entities or items recited in the claims. The mere fact that certain measures are recited in the mutual different dependent claims does not indicate that a combination of these measures cannot be used in an advantageous implementation. A computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

Claims

CLAIMS1. A radio receiver system (1), comprising one or more digital sample buffers (11), being configured to:- buffer received blocks of receive signals relating to data packets; a physical, PHY, layer circuit (12), being configured to:- derive binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions (121); and- send the correctness information to a retraining controller (14) of the system (1); and- update the model in accordance with a retraining operation; the retraining controller (14) being configured to:- selectively trigger the retraining operation of the model by a retraining performer (13) in accordance with the buffered blocks of receive signals and the associated correctness information; and the retraining performer (13) being configured to- perform the retraining operation of the model in accordance with the retraining trigger by the retraining controller (14).

2. The system (1) of claim 1, the retraining trigger comprising one of:- a negative retraining request, if a maximum number of HARQ retransmissions of any one of the buffered blocks of receive signals is reached; and- a positive retraining request, if a retraining of the model is needed in accordance with the buffered blocks of receive signals and the associated correctness information.

3. The system (1) of claim 2, the retraining trigger of the positive retraining request further comprising one or more of:- an identifier of a user associated with the data packets to be used for the retraining operation;- an identifier of a flow associated with the data packets to be used for the retraining operation;- one or more identifiers of the data packets to be used for the retraining operation;- one or more identifiers of the digital sample buffers (11) buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and- an indication of the buffered blocks of receive signals to be used for the retraining operation.

4. The system (1) of any one of the preceding claims, the correctness information comprising correctness probabilities and / or correctness checksums.

5. The system (1) of claim 4, further comprising a medium access control, MAC, controller (15), being configured to:- request a radio transmitter (3) in communication with the system (1) to perform hybrid automatic repeat request, HARQ, retransmissions of blocks of receive signals relating to incorrect binary data in accordance with the associated correctness checksums; the PHY layer circuit (12) further being configured to:- send the derived binary data and the correctness checksums to the MAC controller (15).

6. The system (1) of claim 5, the MAC controller (15) further being configured to:- send an indication of the blocks of receive signals relating to incorrect binary data to the retraining controller (14).

7. The system (1) of claim 6, the indication comprising one or more of:- an identifier of a user associated with the data packets to be used for the retraining operation;- an identifier of a flow associated with the data packets to be used for the retraining operation;- one or more identifiers of the data packets to be used for the retraining operation;- one or more identifiers of the digital sample buffers (11) buffering the buffered blocks of receive signals relating to the data packets to be used for the retraining operation; and- the maximum number of HARQ requests for each of the buffered blocks of receive signals.

8. The system (1) of any one of the preceding claims, the radio receiver PHY layer functions (121) comprising a decoding function (121C).

9. The system (1) of any one of the preceding claims, the model further comprising one or more radio transmitter PHY layer functions (321); and the PHY layer circuit (12) further being configured to:- send parameters of the model relating to the one or more radio transmitter PHY layer functions (121) to the radio transmitter (3) in communication with the system (1).

10. The system (1) of claim 9, the radio receiver PHY layer functions (121) comprising a demodulation function (121B); and the radio transmitter PHY layer functions (321) comprising a modulation function.

11. The system (1) of claim 9, the radio receiver PHY layer functions (121) comprising a beam-combiner function; and the radio transmitter PHY layer functions (321) comprising a codebook-based beam-former function.

12. The system (1) of any one of the claims 1 to 11, being configured to implement a UE-to-UE interface of user equipments, UEs, in communication with one another.

13. The system (1) of claim 12, the UE-to-UE interface comprising a PC5 sidelink, SL, interface.

14. The system (1) of claim 13,the PHY layer circuit (12) further being configured to:- send (5) the parameters of the model relating to the one or more radio transmitter PHY layer functions (321) via a Physical Sidelink Control Channel, PSCCH of the PC5 sidelink interface.

15. The system (1) of any one of the claims 1 to 11, being configured to implement a Uu downlink, DL, interface of a UE in communication with a base station, of the cellular communication network.

16. The system (1) of claim 15, the PHY layer circuit (12) further being configured to:- send (5) the parameters of the model relating to the one or more radio transmitter PHY layer functions (321) via a Physical Uplink Control Channel, PUCCH of a corresponding Uu uplink, UL, interface of the implemented Uu DL interface.

17. The system (1) of any one of the claims 1 to 11, the system (1) being configured to implement a Uu UL interface of a base station of the cellular communication network.

18. The system (1) of claim 17, the PHY layer circuit (12) further being configured to:- send (5) the parameters of the model relating to the one or more radio transmitter PHY layer functions (321) via one of:- a Physical Downlink Control Channel, PDCCH of a corresponding Uu DL interface of the implemented Uu UL interface; and- a Broadcast Control Channel, BCCH.

19. The system (1) of claim 17 or claim 18, the PHY layer circuit (12) forming part of a distributed unit, DU (U), of the base station; the MAC controller (15) forming part of a centralized unit, CU (1“), of the base station; and the DU (U) and the CU (1“) being in communication via an Fl interface of a DU-CU split base station architecture.

20. The system (1) of claim 19, the retraining controller (14) forming part of the DU (T).

21. The system (1) of claim 19 or claim 20, the retraining controller (14) forming part of the CU (1“).

22. The system (1) of any one of the claims 1 to 21, the retraining performer (13) forming part of the PHY layer circuit (12).

23. The system (1) of any one of the claims 17 to 19, the retraining controller (14) and the retraining performer (13) forming part of a RAN data analytics function, RANDAF (4), of the cellular communication network; and the RANDAF (4) and the base station being in communication via an Xn interface or an interface for data exchange associated with the model.

24. The system (1) of any one of the preceding claims, the retraining operation of the model being based on one of:- supervised learning in accordance with input data and output data;- unsupervised learning in accordance the input data; and- reinforcement learning in accordance with the input data, an internal state of the model, and a plurality of given actions;25. The system (1) of claim 24, the input data comprising one or more of:- binary data;- a number of Physical Resource Blocks, PRBs;- channel state information, CSI;- predictive quality of service, QoS, information;- location information;- velocity information;- geographic information of the environment;- a receive signal;- blocks of receive signals;- beamforming selection related information; and- control information.

26. The system (1) of claim 24 or claim 25, the output data comprising one or more of:- a transmit signal;- blocks of receive signals;- binary data; and- precoding / combining vectors / codewords.

27. A method (5) of operating a radio receiver system (1), the method (5) comprising- buffering (51) received blocks of receive signals relating to data packets;- deriving (52) binary data and associated correctness information from the blocks of receive signals in accordance with a model of one or more radio receiver PHY layer functions (121);- sending (53) the correctness information to a retraining controller (14) of the system (1);- selectively triggering (54) by the retraining controller (14) a retraining operation of the model in accordance with the buffered blocks of receive signals and the associated correctness information;- performing (55) the retraining operation of the model by a retraining performer (13) in accordance with the retraining trigger; and- updating (56) the model in accordance with the retraining operation.

28. The method of claim 27, being performed by the system (1) of any one of the claims 1 to 26.

29. A computer program comprising executable instructions which, when executed by a processor, cause the processor to perform the method of claim 27 or claim 28.

Images