Randomization of deep neural networks for telecommunications

Abstract

End-to-end deep network (DNN) communication is provided between a first device (104a) and a second device (104b). In the first device, a transmitting DNN (106) processes input communication data to generate an output communication signal (118). In response that the predicted transmission of the generated output communication signal does not meet the white noise interference level, the first device performs a scrambling DNN operation, which includes selecting neural network scrambling information (NNSI) to reconfigure the transmitting DNN to process the input communication data and generate a scrambled output communication signal that meets the white noise interference level at the time of transmission. The first device transmits a control message having an indication of the NNSI and scrambling timing information to indicate when the second device should reconfigure the corresponding receiving DNN (109) to generate reconstructed communication data. Based on the scrambling timing information, the first device transmits the scrambled output communication signal.

Description

[Background technology]

[0001] Conventional fourth-generation (4G) and fifth-generation (5G) communication systems have complex transmitter and receiver processing chains in which multiple processing components encode and modulate input communication data for wireless transmission from a first device and reception and reconstruction by a second device. The complexity of the current transmitter and receiver processing chains can be mitigated by training machine learning (ML) algorithms, such as deep neural networks (DNNs), to form transmitter (TX) DNN models (TX DNNs) and receiver (RX) DNN models (RX DNNs) (also referred to as transmit DNNs and receiver DNNs) capable of providing end-to-end communication. Such transmitter and receiver DNN models can extend and / or replace conventional transmitter and receiver processing chains. For example, a trained transmit DNN can generate a transmit waveform that is suitable for efficiently overcoming the numerous channel environments, faults, and interferences (e.g., multipath interference, multi-access interference, narrowband interference) found in current communication systems, and further improving performance. Furthermore, such transmit DNN models and receive DNN models are well-suited to supporting end-to-end communication systems where constructing conventional transmitter and receiver processing chains is impractical.

[0002] Transmitting and receiving DNN models are being considered for deployment in advanced communication systems, but conventional methods often fail to adequately control the DNN model to rapidly generate an output that becomes a whitened physical transmit signal in order to reduce interference as needed.

[0003] For example, a transmit DNN model can be trained to generate an output communication signal to whiten the transmit signal, but it is impractical to maintain a specific level of white noise interference in the transmit signal (e.g., the amount of white noise interference tolerated by the system, or a specific power spectral density (PSD) level of the white noise spectrum tolerated) for various combinations of input communication data that can be processed by the transmit DNN model.

[0004] While still maintaining the advantages of end-to-end communication using transmit and receive DNNs, there is an opportunity to develop an efficient and dynamic mechanism that can control the transmit DNN model to adjust the transmit signal so that the transmit signal is whitened during transmission, i.e., appears random to adjacent cells (e.g., whitened interference), and / or satisfies a specific level of white noise interference. [Overview of the Initiative]

[0005] In a first aspect, the Disclosure provides a method performed by a first device communicating with a second device, the method comprising: processing input communication data with a transmitting deep neural network (DNN) to generate an output communication signal for transmission to the second device; and performing a scrambling DNN operation in response to the predicted transmission of the generated output communication signal not satisfying a white noise interference level, the scrambling DNN operation further comprising: selecting neural network scrambling information (NNSI) for processing the input communication data to reconfigure the transmitting DNN to generate a scrambled output communication signal that satisfies a white noise interference level at transmission; sending a control message to the second device indicating the NNSI and scrambling timing information to instruct the second device when it should reconfigure the receiving DNN; and sending a scrambled output communication signal that satisfies a white noise interference level to the second device based on the scrambling timing information.

[0006] In a second aspect, the Disclosure provides a method performed by a second device communicating with a first device, the method comprising: receiving a control message from the first device indicating an NNSI and scramble timing information; receiving a communication signal transmitted from the first device in accordance with the scramble timing information; reconstructing a receiving DNN of the second device in accordance with the NNSI and scramble timing information; processing the received communication signal in the receiving DNN to generate reconstructed communication data represented by the received communication signal; and transmitting the reconstructed communication data to a data sink of the second device, or transmitting the reconstructed communication data to one or more higher protocol layers of the protocol stack of the second device.

[0007] Further embodiments provide apparatus and systems for carrying out the methods of the first and second embodiments.

[0008] Embodiments of the method, apparatus, and system offer numerous advantages, including, for example, efficient design and control of transmit DNN and receive DNN structures capable of maintaining a transmit signal that satisfies a white noise interference level, and reducing interference to receivers of other DNNs or non-DNNs in cells or regions around the first and second devices. The scrambled DNN operation used in the transmit DNN of the first device maintains a white noise interference level for the transmit signal from the first device without causing accidental transmit spikes due to the diversity of different combinations of input communication data processed by the transmit DNN. The scrambled DNN operation mitigates, reduces, and / or prevents transmit spikes in the transmit signal of the first device when using the transmit DNN while satisfying the white noise interference level. A further advantage is that the corresponding scrambled DNN operation used in the transmit DNN of the first device allows for efficient control of the reconfiguration of the receive DNN of the second device. The transmit DNN and receive DNN of the first and second devices are efficiently, rapidly, and dynamically reconfigured in real time according to scrambled DNN operation to change the white noise interference level of the transmit signal generated using the output communication signal of the transmit DNN, while simultaneously maintaining the transmit power or bit / symbol error rate of the signal under consideration. A further advantage is the efficient synchronization between the first device using the transmit DNN and the second device using the corresponding receive DNN, enabling dynamic whitening of the transmit signal from the first device and reception and decoding of the dynamically whitened transmit signal by the second device.

[0009] By referring to the accompanying drawings, this disclosure will be better understood and many of its features and advantages may become apparent to those skilled in the art. The use of the same reference numeral in different drawings indicates similar or identical items. Embodiments of the present invention will be described, by example, with reference to the following drawings. [Brief explanation of the drawing]

[0010] [Figure 1]This schematic diagram shows a comparison between exemplary conventional transmitter and receiver structures in several embodiments and exemplary end-to-end communication transmitter and receiver structures using a deep neural network. [Figure 2a] This is a schematic diagram illustrating an exemplary scrambled DNN communication system according to several embodiments. [Figure 2b] This schematic diagram shows an exemplary power spectral density of a transmitted signal that satisfies a white noise interference level according to several embodiments. [Figure 2c] This schematic diagram shows other exemplary power spectral densities of a transmitted signal having transmitted spikes that do not meet the white noise interference level, according to several embodiments. [Figure 2d] This schematic diagram shows further exemplary power spectral densities of other transmitted signals exceeding the white noise interference level, according to several embodiments. [Figure 2e] This schematic diagram shows further exemplary power spectral densities of other transmitted signals that satisfy the white noise interference level according to several embodiments. [Figure 3a] This is a schematic diagram illustrating exemplary input-based scrambling configurations for transmit and receive DNNs according to several embodiments. [Figure 3b] This is a schematic diagram showing exemplary output-based scrambling configurations for transmit and receive DNNs according to several embodiments. [Figure 3c] This is a schematic diagram illustrating exemplary hidden layer-based scrambling configurations for transmit and receive DNNs according to several embodiments. [Figure 4a] This flowchart illustrates an exemplary DNN scrambling process for generating an output communication signal from a transmitting DNN that satisfies a white noise interference level during transmission, according to several embodiments. [Figure 4b] This flowchart illustrates an exemplary process for analyzing whether the spectral density of a transmitted signal, representing the output communication signal from a transmitting DNN, satisfies the white noise interference level, according to several embodiments. [Figure 4c] A flowchart showing an exemplary process for receiving, at a second device, one or more control messages including neural network scramble information according to some embodiments. [Figure 4d] A flowchart showing an exemplary process for receiving, at a second device, an output communication signal from a transmission DNN of a first device and reconstructing communication data according to some embodiments. [Figure 5] A schematic diagram showing an exemplary first device transmitter having a transmission buffer and an exemplary second device receiver having a reception buffer according to some embodiments. [Figure 6a] A schematic diagram showing an exemplary random permutation and random inverse permutation (or permutation解除) used to scramble one or more neural network layers of a transmission DNN according to some embodiments. [Figure 6b] A schematic diagram showing an exemplary random permutation sequence starting from an initial seed according to some embodiments. [Figure 6c] A schematic diagram showing an exemplary permutation matrix obtained from a selected i-th random permutation sequence used to scramble / descramble one or more neural network layers of a transmission / reception DNN according to some embodiments. [Figure 6d] A flowchart showing an exemplary iterative process for selecting an i-th random permutation sequence used to randomize the order of neural network nodes in one or more neural network layers of a transmission DNN such that the transmission of an output communication signal meets a white noise interference level according to some embodiments. [Figure 7] A signal flowchart showing the activation and deactivation of a scramble DNN communication session between a first device and a second device according to some embodiments. [Figure 8]Figure 7 is a signal flow diagram illustrating exemplary DNN scrambled communication between a first device and a second device during a DNN communication session, according to several embodiments. [Figure 9] Figure 7 is a signal flow diagram illustrating another exemplary DNN scrambled communication between a first device and a second device during a DNN communication session, according to several embodiments. [Figure 10] This is a signal flow diagram illustrating a scrambled DNN communication session between a first device, a second device, and a third device according to several embodiments. [Figure 11] This is a signal flow diagram illustrating exemplary DNN scrambled communication between a first device, a second device, and a third device during a DNN communication session, as shown in Figure 7 or Figure 10, according to several embodiments. [Figure 12] This is a signal flow diagram showing the activation and deactivation of uplink (UL) / downlink (DL) scrambled DNN communication sessions between a base station and user equipment, according to several embodiments. [Figure 13] Figure 12 is a signal flow diagram illustrating exemplary DL DNN scrambled communication between a base station and user equipment during a UL / DL DNN communication session, according to several embodiments. [Figure 14] Figure 12 is a signal flow diagram illustrating exemplary UL DNN scrambled communication between user equipment and a base station during a UL / DL DNN communication session, according to several embodiments. [Figure 15] The following is a signal flow diagram showing other exemplary UL DNN scrambled communication between user equipment and a base station during a UL / DL DNN communication session in Figure 12, according to several embodiments. [Figure 16] This is a signal flow diagram illustrating another exemplary DNN scrambled communication session between a base station and user equipment, according to several embodiments. [Figure 17] This is a schematic diagram of an exemplary computer-readable medium according to several embodiments. [Modes for carrying out the invention]

[0011] Figure 1 shows a comparison between exemplary conventional transmitter and receiver structures, namely a first conventional communication device 102a and a second conventional communication device 102b, and exemplary end-to-end communication transmitter and receiver structures, namely a first DNN device 104a and a second DNN device 104b (also referred to herein as the first device 104a and the second device 104b), which use a transmitting deep neural network (DNN) structure 106 and a receiving DNN structure 108 (TX DNN and RX DNN), respectively. Conventional 4G and 5G communication systems have complex transmitter and receiver processing chains in which multiple processing components encode and modulate input communication data 101a for wireless transmission from the first conventional device 102a and reception of reconstructed communication data 101b by the second conventional communication device 102b.

[0012] In this embodiment, the transmitter processing chain of the first conventional communication device 102a processes input communication data 101a (e.g., a block of bits, a bitstream, or other digital data) from a data source (not shown) in order to transmit from the first conventional communication device 102a to the second conventional communication device 102b. The transmitter processing chain includes an array of processing blocks, such as coding blocks, interleaving blocks, scrambling blocks, pre-coding blocks, and modulation blocks, which process the input communication data 101a into an output communication signal 118 for transmission. The first communication device 102a includes a radio frequency (RF) front end, which includes an RF analog transmit / transmitter / transmit (TX) component (RF analog TX) 103a that processes the output data signal to perform radio frequency upconversion (e.g., digital-to-analog conversion and / or radio frequency upconversion, etc.) and transmits it as a transmit signal 105a via the antenna over the communication channel 105. A second conventional communication device 102b receives the transmit signal 105a. The second communication device 102b includes an RF front end, which includes an RF analog receiver / receiver (RX) component (RF analog RX) 103b of the second conventional communication device 102b for receiving the transmit signal 105a (for example, for analog-to-digital conversion and / or frequency down-conversion to baseband), and the receiver processing chain generates reconstructed communication data 101b from the consequently received transmit signal 105a. The receiver processing chain includes an array of processing blocks, such as demodulation components, descrambling components, deinterleaving components, and decoding components, to process the received transmit signal 105a and reconstruct the input communication data 101a as reconstructed communication data 101b. As communication systems evolve (e.g., from 5G to 6G communication standards), and with the fusion of various heterogeneous technologies, they provide greater capacity and lower latency, various updates and modifications are made to the already complex transmitter and receiver processing chains.This creates a strictly controlled and complex transmitter and receiver processing chain.

[0013] The complexity of the current transmitter and receiver processing chains can be mitigated by training machine learning (ML) algorithms, such as deep neural networks (DNNs), to integrate the transmitter and receiver processing blocks into a form of a transmit DNN structure 106 and a receive DNN structure 108 (hereinafter also referred to herein as the transmitter DNN structure and the receiver DNN structure, respectively) capable of providing end-to-end communication. Such transmit and receive DNN structures 106 and 108 may extend and / or replace the conventional transmitter and receiver processing chains used in a first conventional communication device 102a and a second conventional communication device 102b. For example, the transmit DNN structure 106 includes a transmit DNN model 107 (TX DNN) trained to replace the transmitter processing chain of the first conventional device 102a, which includes coding blocks, interleaving blocks, scrambling blocks, and modulation blocks. The first device 104a configures a transmit DNN structure 106 to process input communication data 101a'' (e.g., blocks of input communication data / bits) to generate an output communication signal 118 for transmission as a transmit signal 105a'' via the RF analog TX 103a'' over the communication channel 105''. In such a case, the transmit DNN model 107 of the transmit DNN structure 106 generates the output communication signal 118 during training. The RF analog TX 103a'' processes the output communication signal 118 for transmission as a transmit signal 105a'' over the communication channel 105''. The resulting transmit signal 105a'' generated from the output communication signal 118 of the transmit DNN model 107 is a transmit waveform suitable for efficiently dealing with the numerous channel environments, faults, and interferences (e.g., multipath interference, multiaccess interference, narrowband interference) found in current communication systems, depending on the training / situation.

[0014] The second device 104b receives the transmit signal 105a'' via the RF analog RX 103b'', and the RF analog RX 103b'' processes the received transmit signal 105a'' (e.g., by performing at least down-conversion) to produce a baseband received communication signal 119 for input to the receive DNN structure 108. The receive DNN structure 108 includes a receive DNN model 109 (RX DNN) trained to generate reconstructed communication data 101b'' given a properly formatted baseband (or down-converted) input data signal as input. The second device 104b configures the receive DNN model 109 to perform a reverse operation of the transmit DNN model 107 to generate reconstructed communication data 101b'' representing the input communication data 101a'' input to the transmit DNN model 107.

[0015] In this embodiment, the second device 104b has a protocol stack comprising multiple protocol layers. After generating reconstructed communication data 101b' for a specific time slot or one or more time slots, the second device 104b transmits the reconstructed communication data 101b' for one or more time slots to one or more higher-layer protocols of the protocol stack of the second device 104b. In the protocol stack of the second device 104b, lower layers serve to provide services to higher layers, and higher layers use these services to provide their own functionality. For example, the reconstructed communication data 101b' is generated in the physical layer of the protocol stack and passed up to the application layer of the protocol stack by each of the higher layers for processing, and the corresponding reconstructed communication data 101b' is used, non-limitingly, for example, to display to a user, to process further, and / or to transmit to one or more applications of the second device 104b for further processing and / or consumption of the reconstructed communication data 101b'.

[0016] In this embodiment, once trained, the receiving DNN model 109 replaces a receiver processing chain, including, for example, demodulation blocks, descramble blocks, deinterleave blocks, and decoding blocks, to generate reconstructed communication data 101b' from the received transmitted signal 105a'. The reconstructed communication data 101b' represents the input communication data 101a'. The transmitting DNN model 107 and the receiving DNN model 109 are well-suited to supporting end-to-end communication systems where constructing conventional transmitter and receiver processing chains is impractical. As the complexity and requirements of the transmitter and receiver chains increase, the transmitting DNN structure 106 and the receiving DNN structure 108 become important components in 5G advanced communication systems or 6G and beyond.

[0017] Transmitting DNN Model 107 is described as performing the functions of the coding block, interleaving block, scrambling block, and modulation block of the transmitter chain, but this is merely an example and not limited thereto. It will be understood by those skilled in the art that Transmitting DNN Model 107 can be trained to perform any one or more functions of the transmitter processing chain, including at least one or more of coding, interleaving, scrambling, pre-coding, and modulation, combinations thereof, variations thereof, and / or applications-dependent forms. Receiving DNN Model 109 is described as being trained to perform the functions of demodulation, descrambling, deinterleaving, and decoding of the receiver processing chain, but this is merely an example and not limited thereto. It will be understood by those skilled in the art that Receiving DNN Model 109 can be trained to perform one or more functions of the receiver processing chain, including at least one or more of demodulation, descrambling, deinterleaving, decoding, and / or any other receiver processing chain functions, combinations thereof, variations thereof, and / or applications-dependent forms. For example, the transmit DNN model 107 is trained to perform many, if not all, of the functions of the transmit processing chain, apart from the modulation block and the RF analog TX103a', and the corresponding receive DNN model is trained to perform most, if not all, of the functions of the transmit processing chain, apart from the RF analog RX103b' and the demodulation block.

[0018] As described herein, the respective transmit DNN models 107 and receive DNN models 109 of the corresponding transmit DNN structure 106 and receive DNN structure 108 are trained to replace the functions of conventional transmit chains / receive chains, respectively, and / or to overcome various channel situations, and / or to meet performance requirements such as 5G communication standards, 6G communication standards, and future communication standards. The transmit DNN model 107 and receive DNN model 109 are deployable to constitute the transmit DNN structure 106 and receive DNN structure 108, respectively, which are used to perform DNN communication between a first device 104a and a second device 104b. For example, the first DNN model 107 includes at least an input neural network layer, one or more hidden neural network layers, and an output neural network layer. The first DNN model 107 processes the input communication data to generate an output communication signal which is processed and transmitted (e.g., digital-to-analog conversion and radio frequency upconversion) by the RF analog TX 103a' as a transmission signal 105a'. For example, the second DNN model 109 includes at least an input neural network layer, one or more hidden neural network layers, and an output neural network layer. The RF analog RX 103b' processes the transmission signal 105a' (e.g., performing analog-to-digital conversion and / or frequency downconversion to baseband) to generate a received communication signal 119 for input to the second DNN model 109, and the second DNN model 109 processes the received communication signal to generate reconstructed communication data which represents the input communication data incorporated into the transmission signal 105a'. In other embodiments, a pair of first DNN model 107 and / or second DNN model 109 may be selected by the first device 104a depending on the communication performance requirements for a DNN communication session with the second device 104b, and the first device 104a may communicate this selection to the second device 104b during the establishment of the DNN communication session.

[0019] As described herein, the ML model algorithms / architectures used to train the transmitting DNN model 107 and the receiving DNN model 109 include, but are not limited to, neural networks, fully connected neural networks, convolutional neural networks, long short-term memory (LSTM) neural networks, transformer neural networks, and / or any other suitable DNN architectures, combinations thereof, modifications thereof, the forms described herein, and / or forms depending on the application, based on or based on one or more of the above. Depending on the application, supervised training and / or unsupervised training may be performed. For example, supervised training of the first DNN model 107 and the second DNN model 109 used by the first device 104a and the second device 104b may use, for example, gradient backpropagation-based techniques to update the node weights / parameters of the neural network layers and any other DNN architecture components of the corresponding first DNN model 107 and / or second DNN model 109, for example, using a preferred or appropriate loss function. With respect to the first and second DNN structures of the first and second devices, it is assumed that various DNN models and / or architectures, as well as DNN model / structure arrangements, as described with reference to Figures 3a and 3c, have already been trained and determined. Multiple different first DNN models 107 and / or second DNN models 109 used in different communication scenarios may be stored and / or mapped, along with appropriate identifiers / indexes, in storage accessible by the first device 104a and / or second device 104b, which are retrieved to constitute the first DNN structure 106 and second DNN structure 108 of the first device 104a and second device 104b when the first device 104a and second device 104b establish DNN communication with each other.

[0020] While there are significant advantages to using the transmit DNN structure 106 and receive DNN structure 108 to replace one or more functions of conventional transmit and receive chains, there are also several challenges to consider when implementing the transmit DNN structure 106 and receive DNN structure 108 to meet the performance requirements of 5G, 6G, and future communication standards. For example, reducing interference or rapidly generating a whitened physical transmit signal that meets a specific white noise interference level is beneficial for increasing capacity and reducing latency in communication systems. Currently, the scramble / descramble blocks of conventional 4G / 5G transmitter and receiver chains are controlled individually to rapidly whiten the physical transmit signal. However, incorporating such functions into current transmit and receive DNN structures is difficult. In the transmit DNN model 107 and receive DNN model 109 of the transmit DNN structure 106 and the receive DNN structure 108, training may be performed together (e.g., jointly) to produce an output communication signal 118 that satisfies a specific white noise interference noise level (e.g., the amount of white noise interference tolerated by the system, or a specific PSD level of the tolerable white noise spectrum) when upconverted and transmitted as the transmit signal 105a'. However, with respect to subsequent input communication data blocks, it may be impossible to maintain this specific white noise interference level because the input bitstream changes.

[0021] Furthermore, different receivers within a cell or region may tolerate different levels of interference, and the transmitter chain of the first device 104a dynamically adjusts the white noise interference level of the transmitted signal. In the case of the transmit DNN structure 106, which replaces the conventional transmitter processing chain, the remaining components, such as the RF analog TX 103a', dynamically adjust the final transmitted signal to satisfy the white noise interference level. However, using the RF analog TX 103a' results in coarser adjustment of the transmit power, increasing the symbol error rate or bit error rate over the communication link, and consequently, there is a risk of reduced throughput due to increased retransmission between the first device 104a and the second device 104b, for example. Another possible approach is to train multiple transmit DNN models (and corresponding receive DNN models), each configured to produce output communication signals 118 that satisfy different white noise interference levels for the same input communication data 101a' when transmitted by the RF analog TX 103a'. The first device 104a selects a transmit DNN model (and corresponding receive DNN model) that produces the output communication signal 118 with the lowest predicted white noise interference level during transmission. This may be impractical because satisfying all reasonable white noise levels would require a large number of transmit DNN models and corresponding receive DNN models. This also represents an inefficient and impractical use of computational resources in both the first device 104a and the second device 104b.

[0022] Training multiple different transmit DNN models (and corresponding receive DNN models) used to whiten the transmit signal 105a' to satisfy different white noise interference levels involves selecting transmit and receive DNN pairs to be used by the transmit DNN structure 106 and receive DNN structure 108 of the first device 104a and the second device 104b, respectively, depending on the white noise interference level. This training is a resource-intensive process requiring significant computing, storage, and transmission resources to ensure it can reliably handle all types of interference and white noise interference levels. This means that the first and second devices store multiple transmit and receive DNN structures within themselves for recall.

[0023] The above challenges are addressed by including a scrambling operation in the transmit DNN structure 106 of the first device 104a, which is controllable to generate an output communication signal 118 from input communication data 101a', and the output communication signal 118 generates a transmit signal 105a' that satisfies a white noise interference level when processed and transmitted by the RF analog TX 103a'. The receive DNN structure 108 of the second device 104b performs descrambling using the reverse operation of the transmit DNN structure 106 to generate reconstructed communication data 101b'. The scrambling DNN operation is controllable using neural network scrambling information (NNSI). Each NNSI is associated with a different white noise interference level, and the first device 104a selects an NNSI from a set of NNSIs. For example, each NNSI describes the type and / or location of scrambling performed within one or more neural network layers of the transmit DNN model 107 of the transmit DNN structure 106. When the first device 104a predicts that the resulting transmitted signal 105a' satisfies a specific white noise interference level, the first device 104a selects an NNSI for scrambling the output communication signal 118 from a set of NNSIs. The first device 104a transmits the selected NNSI to the second device 104b before transmitting the corresponding output communication signal 118 as the transmitted signal 105a'. This allows the second device 104b to reconstruct the received DNN structure 108 to generate reconstructed communication data 104b' corresponding to the input communication data 101a' represented by the transmitted signal 105a'.

[0024] For example, when the first device 104a communicates with the second device 104b, the transmit DNN structure 106 of the first device 104a processes the input communication data 101a' using the transmit DNN model 107 of the transmit DNN structure 106 to generate an output communication signal 118 for transmission to the second device 104b. When the first device 104a analyzes the generated output communication signal 118 and estimates or predicts that the resulting transmission will not meet the white noise interference level, in response, the first device 104a and the second device 104b perform scrambling DNN operation. The scrambling DNN operation in the first device 104a includes the first device 104a selecting an NNSI to reconfigure the transmit DNN model 107 of the transmit DNN structure 106 so that it processes the input communication data 101a' and generates a scrambled output communication signal 118 that meets the white noise interference level when transmitted as the transmit signal 105a'. The first device 104a transmits the selected NNSI to the second device 104b in a control message. The control message includes an indication of the NNSI and scramble timing information. The scramble timing information indicates when the second device 104b should reconstruct the received DNN model 109 of the received DNN structure 108 to generate reconstructed communication data 101b' when the second device 104b receives the transmit signal 105a' corresponding to the scrambled output communication signal 118. Based on the scramble timing information, at a suitable time, the first device 104a processes the scrambled output communication signal 118 as the transmit signal 105a' to the second device 104b via the RF analog TX component, and the transmission of the scrambled output communication signal 118 satisfies the white noise interference level.

[0025] The scrambled DNN operation in the second device 104b includes the second device 104b receiving a control message indicating the NNSI and corresponding scramble timing information. The RF analog RX 103b' of the second device 104b receives a transmit signal 105a' from the first device via the communication channel 105 according to the scramble timing information and outputs a received communication signal 119 for processing by the received DNN model 109 of the second device 104b. Before processing the received communication signal 119, the second device 104b reconstructs the received DNN model 109 using the received NNSI and associated scramble timing information. After reconstruction, the received DNN model 109 processes the received communication signal 119 to generate reconstructed communication data 101b' represented by the received communication signal 119. The second device 104b transmits the reconstructed communication data 101b' to the data sink of the second device 104b, or transmits the reconstructed communication data 101b' to one or more higher protocol layers of the protocol stack of the second device 104b (for example, to the application protocol layer of the protocol stack used by one or more applications running on the second device 104b). The scrambled DNN operation continues as long as the transmit signal 105a' transmitted by the first device 104a satisfies the white noise interference level and / or as further input communication data for transmission is available.

[0026] During a communication session, if the first device 104a performs multiple scramble DNN operations, the first device 104a selects a different NNSI because, due to different input communication data 101a', the transmit DNN model 107 generates an output communication signal 118 that does not meet the current white noise interference level when transmitted as a transmit signal 105a' by the RF analog TX 103a'. The first device 104a transmits the selected different NNSI, along with the relevant scramble timing information used in the corresponding scramble DNN operation, to the second device 104b in a control message. The first device 104a selects a different NNSI as the white noise interference level changes, and as a result, the sacrificial device may receive unacceptable interference from the first device 104a, i.e., the first device 104a receives a request to adjust the white noise interference level to an acceptable level.

[0027] The scrambled DNN operation of the first device 104a and the second device 104b provides the advantage that the first device 104a will not transmit the output communication signal 118 of the transmit DNN until the resulting transmit signal satisfies the white noise interference level set by the first device 104a. This means that the transmission of the first device 104a satisfies the white noise interference level without causing interference to adjacent devices by transmit spikes. The scrambled DNN operation also synchronizes the transmit DNN structure 106 and the receive DNN structure 108 to work together in order to reconstruct the input communication data 101a' at the second device 104b.

[0028] The first device 104a and the second device 104b may be any type of communication device used in the communication system 100, including, but not limited to, a base station (BS), a network device, a radio access network (RAN) element including a user equipment (UE), or any combination of other RAN elements in the communication system 100. For example, the first device 104a and the second device 104b may be two BSs, or two UEs, or a BS and an UE, or an UE and a BS, or any other combination of communication devices depending on the application. Figure 2a shows a scrambled DNN communication system 200 in which the first device 210 and the second device 220 are a BS and an UE, respectively.

[0029] Figure 2a shows an exemplary scrambled DNN communication system 200 in which a first device 210 communicates with a second device 220. In this example, the first device 210 is a BS and the second device 220 is an UE. The BS 210 is connected to the core network (not shown) of the scrambled DNN communication system 200 via one or more interfaces. For example, the communication system could be a 5G / 6G communication system or a New Radio (NR) communication system. The UE 220 and the BS 210 communicate via a radio communication channel 205, via downlink transmit signals 205a (e.g., downlink transmit) and uplink transmit signals 205b (e.g., uplink transmit). The wireless communication channel 205 may include a downlink communication channel (e.g., a physical downlink shared channel (PDSCH)) used by BS210 to transmit a downlink transmit signal 205a to UE220, and an uplink communication channel (e.g., a physical uplink shared channel (PUSCH)) used by UE220 to transmit an uplink transmit signal 205b to BS210. The downlink communication channel may also include a downlink control channel (e.g., a physical downlink control channel (PDCCH)), and the uplink communication channel may also include an uplink control channel (e.g., a physical uplink control channel (PUCCH)).

[0030] BS210 may be implemented as a computing system / device to perform any of the corresponding methods, scrambled DNN operations, scramble / randomize / descramble operations or processes described herein, and / or to implement any of the corresponding systems, units, and / or devices as described herein. BS210 includes an RF front end 203a / b, including RF analog TX components / antennas and RF analog RX components / antennas, one or more transceivers 211, one or more processors 212, and a memory unit 213, which are interconnected. Those skilled in the art will understand that, depending on the application, other types of computing devices / systems / platforms, such as distributed computing systems, may be used instead to implement BS210 and the methods described herein. BS210 includes one or more processors 212. One or more processors 212 control the operation of other components of BS210, such as the RF front end 203a / b, one or more transceivers 211, and the memory unit 213. One or more processors 212 may be single-core devices or multi-core devices. One or more processors 212 may include a central processing unit (CPU), one or more CPUs, a graphics processing unit (GPU), and / or one or more GPUs. Alternatively, one or more processors 212 may include dedicated processing hardware, such as a reduced instruction set computer (RISC) processor, or programmable hardware with built-in firmware. Multiple processors may be included in the BS210. In some embodiments, one or more processors 212 may be part of a distributed computing system, such as a cloud computing system and / or a cloud computing platform.

[0031] Depending on the application, one or more processors 212 of the BS210 may be connected to a network interface, such as a transceiver 211 including a transmitter (TX) and a receiver (RX), to communicate with other devices and systems via the RF front end 203a / b over the wireless communication channel 205 of the network, such as the UE220, other communication devices, network equipment, RAN entities or RAN devices, operators, and / or any other devices, services, systems, and / or devices. Optionally, one or more processors 212 may be connected to a user interface (UI) for user or operator inputs to instruct or use the BS210 and / or the underlying computing system, and / or for user or operator inputs to output data from them. Optionally, one or more processors 212 may be connected to a display to show output to a user or operator.

[0032] BS210 includes a memory system or memory unit 213 that includes working memory or volatile memory. One or more processors 212 may access the volatile memory to process data and may control the storage of data in the memory. The volatile memory may include any type of random access memory (RAM), such as static RAM (SRAM) or dynamic RAM (DRAM), or the volatile memory may include flash memory such as a secure digital (SD) card. In some embodiments, the memory unit 213 and / or one or more volatile memories may have a number of memories that form part of a distributed computing system such as a cloud computing system and / or a cloud computing platform. BS210 also includes non-volatile memory. Non-volatile memory may store a set of operation instructions or operating system instructions for controlling the operation of processor 212 in computer-readable instruction form, and / or software instructions in computer-readable instruction form, which, when executed by one or more processors, cause the processors to perform methods, processes, operations, and / or functions of scrambled DNN operation, scramble / randomization operation, processes, and / or methods, as described herein. Depending on the application, non-volatile memory may be any type of memory, such as read-only memory (ROM), flash memory, SD drive, magnetic drive memory, or magnetic disk drive memory. In some embodiments, non-volatile memory may consist of a number of non-volatile memories that form part of a distributed computing system such as a cloud computing system and / or cloud computing platform.

[0033] The non-volatile memory of the BS210's memory unit 213 contains computer program code and / or instructions for implementing the BS DNN controller (DNNC) 214 and / or the BS downlink transmit DNN structure (BS DL TX DNN) 206 or the BS uplink receive DNN structure (BS UL RX DNN) 208. When the BS DNNC 214 is executed on one or more processors 212, it uses the BS DL TX DNN 206 and / or the BS UL RX DNN 208, the BS neural network scramble information (NNSI) store / table / buffer 215a stored in the memory unit 213, and the BS transmit / receive (TX / RX) DNN store or table 215b (also referred to as the BS TX DNN / RX DNN store or table 215b) stored in the memory unit 213 to control scramble DNN operation between the BS210 and the UE220. Although BS DNNC214 is shown as part of memory unit 213, this is merely an example, and BS DNNC214 is not limited to this. Those skilled in the art will understand that BS DNNC214 may be implemented in hardware and / or software of BS210, depending on the application.

[0034] At least one processor 212, together with at least one memory unit 213 and computer program code or instructions stored in the memory unit 213, is configured to cause the computing system of BS210 to execute at least one corresponding operation, method, and / or process disclosed with respect to a schematic diagram, flow diagram, or operation and related features, for example, as described in any of Figures 1 to 17.

[0035] Similarly, UE220 may be implemented as a computing system / device to perform any of the corresponding methods, scrambled DNN operations, scramble / randomize / descramble operations or processes described herein, and / or to implement any of the corresponding systems, units, and / or devices described herein. UE220 includes an RF front-end 203a / b, one or more transceivers 221, one or more processors 222, and a memory unit 223, which are interconnected. Those skilled in the art will understand that other types of computing devices / systems / platforms may be used alternatively to implement the UE220 and methods described herein. UE220 includes one or more processors 222 (e.g., CPUs). One or more processors 222 control the operation of other components of UE220, such as the RF front-end 203a / b, one or more transceivers 221, and the memory unit 223. One or more processors 222 may be single-core or multi-core devices. One or more processors 222 may include a CPU and / or a GPU. Alternatively, one or more processors 222 may include dedicated processing hardware, such as a RISC processor, or programmable hardware with built-in firmware. Multiple processors may be included in the UE220.

[0036] Depending on the application, one or more processors 222 of the UE220 may be connected to a network interface, such as a transceiver 221 including a transmitter (TX) and a receiver (RX), to communicate with other devices and systems via the RF front end 203a / b over the wireless communication channel 205 of the network, such as the BS210, other communication devices, network equipment, RAN entities or RAN devices, users or operators, and / or any other devices, services, systems, and / or devices. Optionally, one or more processors 222 may be connected to a UI for user inputs to instruct or use the UE220 and / or the underlying computing system, and / or to output data from them. Optionally, one or more processors 222 may be connected to a display for displaying output to the user.

[0037] The UE220 includes a memory system or memory unit 223, which includes working memory or volatile memory. One or more processors 222 may access the volatile memory to process data and may control the storage of data into the memory. The volatile memory may include any type of RAM, e.g., SRAM, DRAM, or the volatile memory may include flash memory such as an SD card. The UE220 also includes non-volatile memory. The non-volatile memory may store a set of operation instructions or operating system instructions for controlling the operation of the processors 222 in the form of computer-readable instructions, and / or software instructions, which, when executed by one or more processors 222, cause the processors 222 to perform the corresponding methods, processes, operations, and / or functions of scrambled DNN operation, scramble / randomize / descramble operation, and / or methods, as described herein. The non-volatile memory may be any type of memory, depending on the application, such as ROM, flash memory, SD drive, magnetic drive memory, or magnetic disk drive memory.

[0038] The non-volatile memory of the UE220's memory unit 223 contains computer program code and / or instructions for implementing the UE DNN controller (UE DNNC) 224 and / or the UE uplink transmit DNN structure (UE UL TX DNN) 226 and / or the UE downlink receive DNN structure (UE DL RX DNN) 228. When the UE DNNC 224 is running on one or more processors 222, it uses the UE UL TX DNN 226 and / or the UE DL RX DNN 228, the UE NNSI store / table / buffer 225a stored in the memory unit 223, and the UE TX / RX DNN store / table 225b stored in the memory unit 223 to control scrambled DNN operation between the BS210 and the UE220. The UE DNNC 224 is shown as part of the memory unit 223, but this is merely an example, and the UE DNNC 224 is not limited to this. Those skilled in the art will understand that the UE DNNC224 can be implemented with any combination of the UE220 hardware and / or software, and / or depending on the application.

[0039] During operation, BS210 and UE220 establish a DL / UL DNN communication session between them. The DL / UL DNN communication session includes DL DNN communication from BS210 to UE220 and UL DNN communication from UE220 to BS210. While establishing the DL / UL DNN communication session, BS210 and UE220 communicate with each other to define, agree, and / or configure the type of pair of BS DL TX DNN206 and UE DL RX DNN228 to be used for DL ​​DNN communication, and the type of pair of UE UL TX DNN226 and BS uplink receive DNN structure (BS UL RX DNN)208 to be used for UL DNN communication. For example, BS210 selects a pair of BS DL TX DNN206 and UE DL RX DNN228 suitable for DL ​​DNN communication from the BS TX DNN / RX DNN store or table 215b (e.g., DNN configuration table). Similarly, BS210 selects a pair of UE UL TX DNN226 and BS UL RX DNN208 suitable for UL DNN communication from the BS TX DNN / RX DNN store or table 215b. Alternatively, UE220 may optionally select a pair of UE UL TX DNN226 and BS UL RX DNN208 suitable for UL DNN communication from the UE TX DNN / RX DNN store 225b.

[0040] The BS TX / RX DNN store or table 215b includes a set and / or pair of DL transmit DNN structures / DL receive DNN structures and a set and / or pair of UL transmit DNN structures / UL receive DNN structures, each of which is mapped to a DL DNN identifier / UL DNN identifier and stored in the BS TX / RX DNN store or table 215b (e.g., a lookup table) on BS210. Each transmit DNN structure / receive DNN structure is further trained to transmit / receive a transmit waveform suitable for efficiently overcoming, for example, a specific channel environment found in the current communication system, one or more specific channel faults, and / or one or more different types of interference (e.g., multipath interference, multiaccess interference, narrowband interference), depending on the training / situation, in order to further improve performance. Training the DL / UL TX DNN and the corresponding DL / UL RX DNN to overcome one or more types of channel faults and / or channel environments uses supervised DNN training across multiple scenarios. For example, supervised DNN training jointly trains DL TX DNN / DL RX DNN pairs and UL TX DNN / UL RX DNN pairs. For instance, BS210 and UE220 use DL transmit DNN and receive DNN pairs trained specifically for downlink communication channels (e.g., PDSCH), while UE220 and BS210 use uplink transmit DNN and receive DNN pairs trained specifically for uplink communication channels (e.g., PUSCH).

[0041] UE220 also has a corresponding set of UL / DL transmit DNN structures / receive DNN structures, which are also mapped to the same UL / DL DNN identifier and stored in UE220's UE TX DNN / RX DNN store 225b (e.g., lookup table). BS210 selects the DL transmit DNN structure and DL receive DNN structure, as well as the UL transmit DNN structure and UL receive DNN structure, based on the DL and / or UL communication channel / environment, the communication performance requirements of the DL and / or UL DNN connection, and the type of DL and / or UL data communication of the communication session (e.g., voice communication, data communication, and multimedia streaming).

[0042] BS210 transmits one or more control messages indicating the selected DL / UL receive DNN structure (e.g., DL / UL DNN identifier) ​​for the DL communication channel and UL communication channel (e.g., PDSCH and PUSCH), DL control channel and UL control channel (e.g., PDCCH / PUCCH) used for DL ​​DNN communication and UL DNN communication between UE220 and BS210. For DL ​​DNN communication, BS210 and UE220 configure the respective BS DL TX DNN206 and UE DL RX DNN228 based on the selected DL DNN identifier. For UL DNN communication, BS210 and UE220 configure the respective BS UL RX DNN208 and UE UL TX DNN226 based on the selected UL DNN identifier. After BS210 and UE220 establish a DL / UL DNN communication session, BS210 and UE220 perform DL DNN communication for one or more time slots using BS DL TX DNN206 and UE DL RX DNN228, respectively. UE220 and BS210 also perform UL DNN communication for one or more time slots using UE UL TX DNN226 and BS UL RX DNN208, respectively.

[0043] During DL DNN communication, if BS210 detects that the transmission from BS210 does not meet a specific white noise interference level, BS210 performs DL DNN scrambling. During UL DNN communication, if UE220 detects that the transmission from UE220 does not meet a specific white noise interference level, UE220 performs UL DNN scrambling. In the case of DL DNN scrambling communication from BS210 to UE220, the TX DNN controller of BS DNNC214 controls the DL DNN scrambling operation on BS210, and the RX DNN controller of UE DNNC224 controls the corresponding DL DNN scrambling operation on UE220. In the case of UL DNN scrambling communication from UE220 to BS210, the TX DNN controller of UE DNNC224 controls the UL DNN scrambling operation on UE220, and the RX DNN controller of BS DNNC214 controls the corresponding UL DNN scrambling operation on BS210.

[0044] For example, in DL DNN scrambling, if the output communication signal from BS DL TX DNN206 does not meet the white noise interference level when predicting transmission via PDSCH, BS210 enables the execution of scrambled DL DNN operation for one or more time slots. The DL TX DNN controller of BS210 detects that BS DL TX DNN206 has generated an output communication signal from the input communication data of a particular time slot, but that this output communication signal will become a downlink transmission signal 205a via PDSCH that does not meet the white noise interference level. When this occurs, DL scramble DNN operation is enabled, and the DL TX DNN controller of BS210 selects neural network scramble information (NNSI) to use to reconstruct the BS DL TX DNN206 so that it generates an output communication signal from the same input communication data for a particular time slot that becomes a downlink transmit signal 205a over the PDSCH that satisfies the white noise interference level set on the PDSCH. The NNSI may be obtained from the BS NNSI store 215a or may be determined iteratively by randomizing one or more neural network layers of the BS DL TX DNN206. For example, an NNSI (e.g., randomization / scrambling parameters and / or specific / selective neural network layers of the transmit DNN to be randomized) is selected to randomize or scramble the order of the set of network nodes in one or more specific neural network layers of the BS DL TX DNN206 of BS210, such that the output communication signal resulting from the reconstructed BS DL TX DNN206 produces a downlink transmit signal 205a that satisfies the white noise interference level of the PDSCH. The specific neural network layers of the BS DL TX DNN206 to be randomized are specified in the NNSI.

[0045] Similarly, if UE220 (or BS210) detects that the uplink transmit signal 205b via PUSCH does not meet the white noise interference level set for PUSCH, the UL scrambled DNN operation is activated. For example, BS210 selects an NNSI to use to reconfigure UE220's UE UL TX DNN226 so that it generates an output communication signal from UE220's UE UL TX DNN226 that will be the uplink transmit signal 205b via PUSCH that meets the white noise interference level. In the embodiment, when BS210 selects an NNSI for UE220, BS210 simulates UL with random UE UL input communication data and UE UL TX DNN226 to generate an output communication signal and selects an NNSI that produces an output communication signal that meets the white noise interference level for UL. BS210 transmits the selected NNSI to UE220 in a control message. NNSI does not necessarily depend on specific UE UL input communication data; rather, NNSI primarily depends on the UE UL TX DNN226 used in the UE UL. For example, an NNSI (e.g., randomization / scrambling parameters, and / or specific layers of the transmit DNN to be randomized) is selected to randomize or scramble the order of a set of network nodes in one or more specific neural network layers of the UE UL TX DNN226 of the UE220, such that the output communication signal resulting from the reconstructed UE UL TX DNN226 produces an uplink transmit signal 205b that satisfies the PUSCH white noise interference level.

[0046] In DL scrambled communication, after selecting the NNSI of the BS DL TX DNN206 that produces a downlink transmit signal 205a satisfying the white noise interference conditions, the BS210 sends a control message to the UE220 that includes indications of the NNSI and scramble timing information. This control message is used by the UE DNNC224 to reconfigure the UE DL RX DNN228 according to the scramble timing information to process the downlink transmit signal 205a obtained from the output communication signal generated by the reconfigured BS DL TX DNN206. The received scramble timing information indicates when the UE DNNC224 should reconfigure the UE DL RX DNN228 using the selected NNSI. For example, the UE DL RX DNN228 processes the received downlink transmit signal 205a to generate reconfigured communication data for a specific time slot, corresponding to the input communication data transmitted in that particular time slot. As an example, the UE220 may send the reconstructed communication data to the UE220's data sink, or to one or more higher protocol layers of the UE220's protocol stack (for example, to the application protocol layer of the protocol stack used by one or more applications running on the UE220).

[0047] Similarly, in the case of UL scrambled communication, after selecting the NNSI of the UE UL TX DNN226 that produces an uplink transmit signal 205b satisfying the white noise interference conditions, BS210 sends a control message to UE220 that includes indications of the NNSI and scramble timing information, which is used by UE DNNC224 to reconfigure the UE UL TX DNN226 according to the scramble timing information to process the uplink transmit signal 205b for transmission to BS210 via PUSCH, and BS210 receives and processes the uplink transmit signal using the BS UL RX DNN208 reconfigured based on the selected NNSI and corresponding scramble timing information. The scramble timing information indicates when BS DNNC214 should reconfigure the BS DL RX DNN208 using the selected NNSI. For example, the BS UL RX DNN208 processes the received uplink transmit signal 205b to generate reconstructed communication data for a specific time slot, corresponding to the input communication data transmitted in that particular time slot.

[0048] Before transmitting the output communication signal generated by the BS DL TX DNN 206, the BS TX DNNC of the BS DNN controller 214 analyzes the output communication signal to determine whether the output communication signal whitens the downlink transmit signal 205a and whether it satisfies the white noise interference level of the PDSCH. If the analysis indicates that RF processing of the output communication signal will generate a transmit signal that does not satisfy the white interference noise level (e.g., a forecast transmit signal or forecast transmission signal), the BS 210 and UE 220 select the NNSI and scrambled DNN operation described above.

[0049] The DL / UL scrambling process offers the advantage that the BS210 / UE220 will not transmit the output communication signals of the DL / UL transmit DNNs until the resulting predicted downlink transmit signals 205a / uplink transmit signals 205b meet the white noise interference level set by the BS210 / UE220. The DL / UL transmits from the BS210 and UE220 meet the corresponding white noise interference levels, respectively, without causing interference to adjacent devices and / or cells through transmit spikes. The DL / UL scrambling process also synchronizes the DL / UL transmit DNNs 206 / 226 and their corresponding DL / UL receive DNNs 228 / 208, respectively, so that they work together to reconstruct the corresponding input communication data at the UE220 or BS210, respectively.

[0050] DL / UL scrambled communication performed by BS210 and UE220 offers many advantages, including the efficient design and control of the BS DL TX DNN206 and UE DL RX DNN228, and / or UE UL TX DNN226 and BS UL RX DNN208, to maintain a white noise interference level while reducing interference to other DNN or non-DNN receivers in cells or areas around BS210 and UE220, or uplink transmit signal 205b. Scrambled DNN operation allows BS210 or UE220 to maintain a white noise interference level when transmitting downlink transmit signal 205a and / or uplink transmit signal 205b, respectively, without causing accidental transmit spikes due to the diversity of different combinations of input communication data processed by each of the BS DL TX DNN206 and UE UL TX DNN226. Therefore, controlling the reconfiguration of the BS DL TX DNN206 or UE UL TX DNN226 and the corresponding UE DL RX DNN228 or BS UL RX DNN208, respectively, reduces, minimizes, and / or prevents transmit spikes in the transmitted signals from the BS210 or UE220 while maintaining white noise interference levels. The BS DL TX DNN206 (or UE UL TX DNN226) and the corresponding UE DL RX DNN228 (or BS UL RX DNN208) of the BS210 and UE220 (or UE220 and BS210) are reconfigured efficiently, quickly, and dynamically in real time to change the white noise interference level of the downlink transmit signal 205a / uplink transmit signal 205b while maintaining the transmit power or bit / symbol error rate of the target signal. A further advantage is the efficient synchronization between the BS210 and UE220 during DL / UL scrambled communication, which allows for the dynamic whitening of the 205a / 205b transmission signals from the BS210 or UE220, enabling the corresponding UE220 or BS210 to receive and decode the dynamically whitened transmission signals.

[0051] A wireless communication network / system will be described with reference to Figures 1 to 2a and / or as described herein, but this is merely an example and not limited to such examples. Any type of communication network / system is applicable, and it will be understood by those skilled in the art that this includes, for example, any telecommunications network, any wired communication network, any wireless communication network, satellite network, peer-to-peer communication network, communication network using third-generation (3G), fourth-generation (4G), fifth-generation (5G), and / or sixth-generation (6G) or higher standards, Wi-Fi communication network, optical communication network, fiber optic communication network, and / or any other network for communication between a first device and a second device, combinations thereof, variations thereof, and / or forms depending on the application. As described with reference to Figure 2a and / or herein, the first device is described as BS210, but this is merely an example and not limited thereto. The first device can be any type of communication device capable of communicating with the second device, including, but not limited to, UE, BS, satellite, mobile phone or smartphone, laptop, computing device, device using 3G, 4G, 5G and / or 6G or higher standard technologies, and / or any other device used for communication with the second device, combinations thereof, variations thereof, and / or forms depending on the application, as will be understood by those skilled in the art. As described with reference to Figure 2a and / or herein, the second device is described as UE220, but this is merely an example and not limited thereto. The second device may be any type of communication device capable of communicating with the first device, including, but not limited to, UE, BS, satellite, mobile phone or smartphone, laptop, computing device, device using 3G, 4G, 5G and / or 6G or higher standards and technologies, and / or any other device used for communication with the first device, combinations thereof, variations thereof, and / or forms depending on the application, as will be understood by those skilled in the art.

[0052] Figure 2b shows an exemplary power spectral density (PSD) graph 230 representing the PSD of the transmitted signal 235 satisfying the white noise interference level 232. In this embodiment, the RF analog transmit component of the RF front end processes the output communication signal generated by the transmit DNN model and the resulting transmit signal 235 has a bandwidth 2f1 f c The signal is upconverted to the carrier frequency. The PSD of the transmitted signal 235 across the entire bandwidth is below the white noise interference level 232. In this embodiment, Figure 2b shows that the white noise interference level 232 is set to the frequency of the bandwidth in question, i.e., frequency f c -f1 and frequency f c It is shown as having a constant amplitude of flat white noise power spectral density over a bandwidth 231 frequency between +f1 and +f1. Although the white interference noise level is described as flat white noise power spectral density, this is merely an example, and it will be understood by those skilled in the art that the white interference noise level can be any appropriate measure of white noise interference or interference, for example, the total power of the flat white noise power spectral density over the frequency of the bandwidth in question, and / or any other appropriate measure of interference. In this example, spectral analysis of the transmitted signal 235 shows that the output communication signal generated by the transmitted DNN model is the transmitted signal 235 that satisfies the white noise interference level.

[0053] Figure 2c shows another PSD graph 240 representing the PSD of the transmitted signal 245 that does not satisfy the white noise interference level 232. In this embodiment, the RF analog transmit component processes the output communication signal generated by the transmit DNN model and the resulting transmitted signal 245 has a bandwidth 231 of 2f1. cThe signal is upconverted to the carrier frequency. The PSD of the transmitted signal 245 has large transmitted spikes in the form of PSD peaks 246a, 246b, 246c, and 246d that exceed the white noise interference level 232. When the first device 210 processes the output communication signal for transmission, these PSD peaks 246a, 246b, 246c, and 246d cause significant interference to other devices within the domain of the first device 210. In this case, spectral analysis of the transmitted signal 245 shows that the output communication signal generated by the transmitted DNN model becomes the transmitted signal 245 that does not meet the white noise interference level. In another embodiment, even if the average PSD of the transmitted signal 245 is below the white noise interference level 232, analysis of the transmitted spikes or PSD peaks 246a, 246b, 246c, and 246d of the transmitted signal 245 may reveal that the average PSD of PSD peaks 246a, 246b, 246c, and 246d exceeds the tolerable transmitted spike PSD threshold, and as a result, it may be shown that the transmitted signal 245 does not meet the white noise interference level 232.

[0054] Figure 2d shows a further exemplary PSD graph 250 representing the PSD of a transmitted signal 255 that similarly does not satisfy the white noise interference level 232. In this embodiment, the RF analog transmit component processes the output communication signal generated by the transmit DNN model and the resulting transmitted signal 255 has a bandwidth 231 of 2f1. c The signal is upconverted to the carrier frequency. The PSD of the transmitted signal 255 across the entire bandwidth exceeds the white noise interference level 252. In this case, spectral analysis of the transmitted signal 255 shows that the output communication signal generated by the transmitted DNN model becomes the transmitted signal 255, which exceeds the white noise interference level, and therefore does not meet the white noise interference level requirement.

[0055] Figure 2e shows a further exemplary PSD graph 260 representing the PSD of the transmitted signal 265 satisfying the white noise interference level 232. In this embodiment, the RF analog transmit component processes the output communication signal generated by the transmit DNN model and the resulting transmitted signal 265 has a bandwidth 2f1 of fc The signal is upconverted to the carrier frequency. The PSD of the transmitted signal 265 has small transmitted spikes in the form of PSD peaks 266a, 266b, 266c, and 266d that exceed the white noise interference level 232. When the first device 210 processes the output communication signal for transmission, these PSD peaks 266a, 266b, 266c, and 266d do not cause significant interference to other devices within the domain of the first device 210. In this case, spectral analysis of the PSD of the transmitted signal 265 shows that the output communication signal generated by the transmitted DNN model is the transmitted signal 265 that satisfies the white noise interference level. For example, the average PSD of the transmitted signal 265 is below the white noise interference level 232, and analysis of the small transmitted spikes or PSD peaks 266a, 266b, 266c, and 266d of the transmitted signal 265 may reveal that the average PSD of PSD peaks 266a, 266b, 266c, and 266d is below the tolerable transmitted spike PSD threshold. Given these two conditions, an analysis of the PSD of the transmitted signal 265 may show that the transmitted signal 265 satisfies the white noise interference level 232.

[0056] Figure 3a shows an exemplary communication system 300a including a first device 310 and a second device 320 that perform scrambling of the input layer of a transmit DNN structure 306 and regressive descrambling of the output layer of a receive DNN structure 308. The transmit DNN structure 306 of the first device 310 includes a transmit DNN model comprising an input neural network layer 316 and a further DNN layer 307 (e.g., one or more hidden and output layers) represented by a block labeled DNN1. The input neural network layer 316 receives input communication data 301a, and the further DNN layer 307 processes the output of the input neural network layer 316 to generate an output communication signal 318. A TX RF front-end component 303a processes the output communication signal 318 and transmits it via an antenna as a transmit signal 305.

[0057] In this embodiment, the transmit DNN model of the transmit DNN structure 306 is reconfigured based on a randomization operation being performed on the order of the neural network nodes of the input neural network layer 316. In this example, the scrambling DNN operation scrambles the nodes of the input neural network layer. Equivalent to scrambling the input neural network layer, the randomization operation can be used to scramble the input communication data 301a before it is input to the input neural network layer of the transmit DNN model of the transmit DNN structure 306. For example, the first device 310 selects an NNSI that specifies that the input layer of the transmit DNN model is randomized / scrambled, and selects one or more specific time slots to use this configuration of the transmit DNN structure 306. For example, the selected NNSI includes a neural network layer indicator that specifies that the input layer of the transmit DNN model is randomized / scrambled using this configuration of the transmit DNN structure 306 during one or more specific time slots. The first device 310 sends a control message containing the selected NNSI to the second device 320, along with one or more designated time slots in which scrambling occurs when the selected NNSI is used.

[0058] The receiving DNN structure 308 of the second device 320 includes a receiving DNN model comprising a DNN layer 309 (e.g., an input layer and one or more hidden layers) represented by a block labeled DNN2, and an output neural network layer 317. After the transmission signal 305 for a specific time slot is received, the DNN layer 309 of the receiving DNN model receives the communication signal 319 output from the RX RF front-end component 303b. The DNN layer 309 processes the received communication signal 319 and outputs the scrambled, reconstructed communication data to the output neural network layer 317 of the receiving DNN structure. In this example, the receiving DNN model of the receiving DNN structure 308 is reconstructed using NNSI, and the output neural network layer 317 performs a de-randomization operation (de-scrambling operation) on the neural network nodes of the output neural network layer 317. The de-randomization operation corresponds to the randomization operation performed on the neural network nodes of the input neural network layer 316 of the transmitting DNN of the transmitting DNN structure 306. The receiving DNN structure 308 sends the descrambled reconstructed communication data 301b corresponding to the input communication data 301a to the data sink of the second device 320, or sends the reconstructed communication data 301b to one or more higher protocol layers of the protocol stack of the second device 320 (for example, to the application protocol layer of the protocol stack used by one or more applications running on the second device 320).

[0059] The input neural network layer 316 of the transmit DNN structure 306 contains a set of N neural network nodes, where N > 1. Reconstructing the transmit DNN structure 306 involves performing a randomization operation on the input neural network layer 316 of the transmit DNN structure 306 by randomizing the set of N neural network nodes (or a subset of these N neural network nodes) of the input neural network layer 316. When the NNSI includes one or more random substitution parameters (e.g., seed, type of random function), randomizing the set of N neural network nodes of the input neural network layer 316 may involve performing random substitutions on the set of N neural network nodes of the input neural network layer 316. For example, the randomization operation generates an N-dimensional substitution matrix using a random substitution sequence of length N with the random substitution parameters. Randomizing the set of N neural network nodes of the input neural network layer 316 is based on randomizing the order of the N neural network nodes by multiplying the order of the N neural network nodes by the N-dimensional substitution matrix. In other embodiments, this corresponds to multiplying the input communication data 301a (i.e., its order) by an N-dimensional permutation matrix before inputting it to the input neural network layer 316.

[0060] After the second device 320 receives the NNSI for a specific time slot, the second device 320 reconstructs the received DNN structure 308 before processing the received communication signal 319 corresponding to that time slot. Assuming that the NNSI for this specific time slot includes a neural network layer indicator specifying the input neural network layer 316 of the transmit DNN structure 306 to be randomized / scrambled, the received DNN structure 308 of the second device 320 is reconstructed by performing an inverse randomization operation on the set of neural network nodes of the output neural network layer 317 of the received DNN structure 308. The output neural network layer 317 generates descrambled reconstructed communication data 301b corresponding to the input communication data 301a. In the embodiment, the second device 320 transmits the reconstructed communication data 301b to the data sink of the second device 320, or transmits the reconstructed communication data 301b to one or more higher protocol layers of the protocol stack of the second device 320.

[0061] The output neural network layer 317 of the receiving DNN structure 308 contains a set of N neural network nodes, where N > 1. When the NNSI includes random substitution parameters (e.g., seed, type of random function), the descramble operation or inverse randomization operation scrambles the set of N neural network nodes of the output neural network layer 317. For example, this involves performing random inverse substitution (or de-substitution) on the set of N neural network nodes of the output neural network layer 317. For example, the descramble operation or inverse randomization operation generates an N-dimensional substitution matrix using a random substitution sequence of length N that uses the random substitution parameters received in the NNSI for a particular time slot. To generate an inverted N-dimensional substitution matrix, the N-dimensional substitution matrix is ​​inverted. The descramble operation is performed by multiplying the N-dimensional output vector of the output neural network layer by the inverted N-dimensional substitution matrix, and the receiving DNN structure 308 generates descrambled reconstructed communication data 301b corresponding to the input communication data 301a.

[0062] Randomizing the input neural network layer and / or one or more hidden neural network layers offers the advantage of increasing the NNSI search space and making it more likely to repeatedly identify an NNSI suitable for the transmitting DNN structure 306'' to produce an output communication signal having a spectrum that satisfies the white noise interference level at transmission.

[0063] Figure 3b shows an exemplary communication system 300b including a first device 310 and a second device 320, further modifying the first device 310 and the second device 320 from Figure 3a to perform scrambling / descrambling of the output layer of the transmit DNN structure 306' and regressive descrambling of the input layer of the receive DNN structure 308'. In this embodiment, the transmit DNN structure 306' of the first device 310 includes a transmit DNN model comprising an output neural network layer 316' and a DNN layer 307' (e.g., an input layer and one or more hidden layers) represented by DNN1. The DNN layer 307' receives and processes input communication data 301a, and the output neural network layer 316' receives the input communication data 301a processed by DNN1 to generate an output communication signal 318. A TX RF front-end component 303a processes the output communication signal 318 and transmits it over the antenna as a transmit signal 305.

[0064] In this embodiment, the transmit DNN structure 306' is reconstructed based on a scrambling DNN operation or a randomizing DNN operation being performed on the order of the neural network nodes of the output neural network layer 316', in a manner similar to that described with reference to the neural network nodes of the input neural network layer 316 in Figure 3a. The receive DNN structure 308' of the second device 320 includes a receive DNN model comprising a DNN layer 309' (e.g., one or more hidden layers and an output layer) represented by a block labeled DNN2, and an input neural network layer 317'. The descrambling operation descrambles the input neural network layer 317', as described with reference to the descrambling operation of the output neural network layer 317 in Figure 3a. This effectively descrambles the received communication signal 319. The DNN layer 309' processes the descrambled received communication signal 319 to output descrambled reconstructed communication data 301b corresponding to the input communication data 301a.

[0065] The output neural network layer 316' of the transmit DNN structure 306' contains a set of N neural network nodes, where N > 1. Reconstructing the transmit DNN structure 306' involves performing a scrambling or randomization operation on the output neural network layer 316' of the transmit DNN structure 306', as described with reference to the scrambling or randomization of the input neural network layer 316 in Figure 3a. For example, randomizing the set of N neural network nodes in the output neural network layer 316' is based on randomizing the order of the N neural network nodes by multiplying the order of the N neural network nodes in the output neural network layer 316' by an N-dimensional permutation matrix. In other embodiments, this corresponds to multiplying the output communication signal 318 (i.e., its order) by an N-dimensional permutation matrix before input to the TX RF front-end component 303a.

[0066] After the second device 320 receives the NNSI for a specific time slot, the second device 320 reconstructs the received DNN structure 308' before processing the received communication signal 319 corresponding to that time slot. Assuming that the NNSI for this specific time slot includes a neural network layer indicator specifying the output neural network layer 316' of the transmitted DNN structure 306' to be randomized / scrambled, the received DNN structure 308' of the second device 320 is reconstructed by performing a descramble or derandomization operation on the set of neural network nodes of the input neural network layer 317' of the received DNN structure 308', similar to the method described with reference to the descramble or derandomization operation on the set of neural network nodes of the output neural network layer 317' in Figure 3a. The remaining DNN layer 309', represented by DNN2 of the received DNN structure 308', generates the descrambled reconstructed communication data 301b corresponding to the input communication data 301a. The second device 320 transmits the reconstructed communication data 301b to the data sink of the second device 320, or transmits the reconstructed communication data 301b to one or more higher protocol layers of the protocol stack of the second device 320.

[0067] Randomizing the output neural network layer of the transmit DNN structure 306' offers the advantage of reducing the computational resources required to iteratively identify updated NNSIs, since only the output neural network layer representing the output communication signal needs to be processed to determine whether a possible NNSI results in an output communication signal that can adjust the transmit signal to satisfy the white noise spectrum or white noise interference level.

[0068] Figure 3c shows an exemplary communication system 300c including a first device 310 and a second device 320, and Figure 3c further modifies the first device 310 and the second device 320 from Figure 3a or Figure 3b to more comprehensively perform scrambling / de-scrambling of one or more hidden neural network layers 316'' of the transmit DNN structure 306'' and one or more hidden neural network layers 317'' of the receive DNN structure 308''. The transmit DNN structure 306'' of the first device 310 includes a transmit DNN model that includes one or more hidden neural network layers 316'' used for hidden layer scrambling operations and a further DNN layer 307'' represented by a block labeled DNN1 (e.g., an input layer, one or more hidden layers if any, and an output layer). The DNN layer 307'' receives and processes the input communication data 301a, and passes the processed data to one or more hidden neural network layers 316'' for scrambling as appropriate. The one or more hidden neural network layers 316'' then pass the scrambled processed data to the output neural network layer of the DNN1 to generate the output communication signal 318. The TX RF front-end component 303a processes the output communication signal 318 to be transmitted via the antenna as the transmission signal 305.

[0069] In this embodiment, the transmitting DNN structure 306'' is reconstructed based on a scrambling DNN operation or randomization operation being performed on the order of the neural network nodes of one or more hidden neural network layers 316'', similar to the methods described with reference to the input neural network layer 316 or output neural network layer 316' in Figure 3a or 3b, respectively.

[0070] The receiving DNN structure 308'' of the second device 320 includes a receiving DNN model comprising a DNN layer 309'' represented by DNN2 (e.g., an input neural network layer, one or more hidden layers if any, and an output neural network layer) and one or more hidden neural network layers 317'' for descrambling, i.e., performing a descramble operation. After the transmission signal 305 for a particular time slot is received, the input neural network layer of DNN2309'' receives the communication signal 319 output from the RX RF front-end component 303b. In this example, the receiving DNN structure 308'' is reconstructed using NNSI in a manner similar to that described with reference to the neural network nodes of the output neural network layer 317 or input neural network layer 317'' in Figure 3a or Figure 3b, respectively, and for descrambling of one or more hidden neural network layers 317'', inverse random operation (descramble operation) is performed on the corresponding neural network nodes of the hidden neural network layer 317''. As a result, the received communication signal 319 is effectively descrambled, and the hidden neural network layer 317'' and the DNN layer 309'' process the communication signal 319 to output the descrambled reconstructed communication data 301b corresponding to the input communication data 301a. The symmetric DNN architecture of the transmitting DNN structure 306'' and the receiving DNN structure 308'' simplifies the scrambling and descrambling operations.

[0071] Randomizing one or more hidden neural network layers offers the advantage of increasing the NNSI search space, which increases the likelihood of repeatedly identifying an NNSI suitable for the transmitting DNN structure 306'' to produce an output communication signal with a spectrum that satisfies the white noise interference level at transmission.

[0072] Referring to Figures 3a to 3c, the scrambling of the input layer, output layer, and the Ith or (L-I+1)th hidden layer of the transmit DNN structure 306 has been described separately, but this is merely an example, and it will be understood by those skilled in the art that one or more of the input layer, output layer, and hidden layers of the transmit DNN structure 306 of the first device 310 may be scrambled and / or randomized. More generally, NNSI specifies the neural network layers to be scrambled along with scrambling parameters, and standards such as the Third Generation Partnership Project (3GPP®) standard may predefine the neural network layers to be scrambled. Alternatively or additionally, NNSI specifies a neural network layer indicator (e.g., a flag, bit, or field) that specifies one or more neural network layers of the transmit DNN structure 306 to be scrambled. More generally, NNSI includes a neural network layer indicator that specifies the scrambling of one or more neural network layers of the transmit DNN structure 306. The transmitting DNN structure 306 includes at least one input layer, one or more hidden layers, and one output layer. The first device 310 uses an NNSI to scramble / randomize the indicated one or more neural network layers. For each of the one or more neural network layers, the Ith neural network layer of the transmitting DNN structure 306 is randomized / scrambled, where 1 ≤ I ≤ L, where L is the number of neural network layers, I=1 is the input layer, and I=L is the output layer. Meanwhile, the second device 320 reconstructs the receiving DNN structure 308 by performing an inverse random operation on the set of neural network nodes of the (L-I+1)th neural network layer of the receiving DNN using the corresponding NNSI (received in a control message for a specific time slot).

[0073] When the NNSI includes random substitution parameters, the reconstruction of the first neural network layer of the transmitting DNN structure 306 is performed by generating an N-dimensional substitution matrix based on the random substitutions of the first neural network layer, where N > 1 and N is the number of neural network nodes in the first neural network layer, and by multiplying the N-dimensional substitution matrix by a specific order of the set of N neural network nodes in the first neural network layer so as to form a randomized order of the set of N neural network nodes. The first device 310 uses the reconstructed transmitting DNN structure 306 to process the input communication data 301a and generate a scrambled output communication signal 318. The TX RF front-end component 303a processes the output communication signal 318 and transmits it to the second device 320 as a transmit signal 305 via the antenna at a specific time slot. The transmit signal 305 satisfies the white noise interference level because the NNSI was selected to satisfy this criterion.

[0074] The second device 320 receives a corresponding NNSI, which includes random substitution parameters, a specific one or more neural network layers, and specific time slot(s) or scramble timing information associated with when the receiving DNN structure 308 should be reconfigured to descramble the corresponding received communication signal 319. At a suitable time, for example, before processing the received communication signal 319 corresponding to a specific time slot, each of the specified one or more neural network layers of the receiving DNN structure 308 is reconfigured to be descrambled based on the corresponding NNSI. The (L-I+1)th neural network layer of the receiving DNN structure 308 has a set of N (N>1) neural network nodes in a specific order, and its reconstruction is performed by generating an N-dimensional permutation matrix based on random permutations of the (L-I+1)th neural network layer, inverting the N-dimensional permutation matrix to generate an inverted N-dimensional permutation matrix, and multiplying the specific order of the set of N neural network nodes in the (L-I+1)th neural network layer by the inverted N-dimensional permutation matrix to form an inversely randomized order of the set of N neural network nodes. After reconstruction, the receiving DNN structure 308 processes the received communication signals 319 received in a specific time slot(s) to generate reconstructed communication data 301b corresponding to the input communication data 301a transmitted in a specific time slot(s).

[0075] Figure 4a shows an exemplary transmit DNN scrambling process 400 for generating a scrambled output communication signal that satisfies a white noise interference level during transmission from a transmit DNN model. Reference numerals in Figure 1 are reused for similar or identical components, features, and / or devices in Figures 4a to 4d. In this embodiment, a first device 104a communicates with a second device 104b, the first device 104a includes a transmit DNN structure 106 having a transmit DNN model 107, and the second device includes a receive DNN structure 108 having a receive DNN model 109. The transmit DNN model 107 is trained to substitute for several functions of the transmit processing chain (e.g., coding, interleaving, scrambling, pre-coding, etc.). The receive DNN model 109 is also trained to perform the reverse operation of the transmit DNN model 107 to generate reconstructed communication data 101b'. The transmit DNN scrambling process 400 includes the following steps:

[0076] In step 402, the first device 104a acquires input communication data 101a' for transmission in a specific time slot. The input communication data 101a' may be a bitstream from a data source or any type of digital data.

[0077] In step 404, the first device 104a processes the acquired input communication data 101a' in the transmission DNN structure 106 in order to generate an output communication signal 118 to be transmitted to the second device 104b.

[0078] In step 406, the first device 104a checks whether the transmission of the generated output communication signal 118 satisfies the white noise interference level. If the transmission of the generated output communication signal 118 satisfies the white noise interference level (e.g., "Y"), the process proceeds to step 412; otherwise (e.g., "N"), the process proceeds to step 408. The first device 104a performs spectral analysis to determine whether the output communication signal 118 becomes a transmission signal 105a' that satisfies the white noise interference level. That is, device 104a uses spectral analysis to predict whether the output communication signal 118 becomes a transmission that satisfies the white noise interference level. For example, spectral analysis includes calculating the power spectral density of the predicted transmission signal resulting from transmitting the output communication signal 118 after RF analog TX processing. The predicted transmission signal is not transmitted, but rather an estimate of the transmission signal that would be obtained if the output communication signal 118 were transmitted by the first device 104a.

[0079] In step 408, in response to the fact that the transmission of the generated output communication signal does not satisfy the white noise interference level, the first device 104a performs a scrambling DNN operation, which is done by selecting an NNSI to reconfigure the transmit DNN model 107 of the transmit DNN structure 106 so as to process the input communication data 101a' and generate a scrambled output communication signal that satisfies the white noise interference level at the time of transmission.

[0080] For example, NNSI may include data representing the randomization of the order of the inputs, outputs, or sets of neural network nodes in one or more specified neural network layers of a transmitting DNN model. For instance, NNSI may include one or more random substitution parameters for randomizing the order of the sets of neural network nodes in one or more specified neural network layers of the transmitting DNN model 107 of the first device 104a. Randomizing the order involves using the random substitution parameters to randomly replace the order of the sets of neural network nodes in one or more specified neural network layers of the transmitting DNN model 107.

[0081] In step 410, the first device 104a reconfigures the transmit DNN model 107 based on the selected NNSI. Proceeding to step 404, the reconfigured transmit DNN model 107 processes the input communication data 101a' acquired in step 402. For example, reconfiguring the transmit DNN model involves reconfiguring the transmit DNN model 107 using the NNSI to randomize the input, output, or set of neural network nodes in one or more specific neural network layers specified by the NNSI.

[0082] In step 412, the first device 104a checks whether the NNSI has been updated. If the NNSI has been updated (e.g., "Y"), i.e., the updated NNSI is different from the previously selected NNSI, proceed to step 414; otherwise (e.g., "N"), proceed to step 416.

[0083] In step 414, the first device 104a sends a control message to the second device 104b, the control message including data representing an NNSI indication and scramble timing information including a specific time slot (e.g., one or more time slots) indicating when the second device 104b should reconfigure the received DNN model 109 to generate reconstructed communication data 101b' corresponding to the input communication data 101a'. Proceed to step 416.

[0084] In step 416, the first device 104a transmits a scrambled output communication signal 118 with respect to scramble timing information to the second device 104b, and the transmission of the scrambled output communication signal satisfies the white noise interference level. Proceed to step 402 to further acquire input communication data 101a' for transmission.

[0085] Figure 4b shows an exemplary spectral analysis process of step 406 in Figure 4a for analyzing whether the spectral density of the transmitted signal, representing the output communication signal generated by the transmitted DNN model 107, satisfies the white noise interference level. The spectral analysis process includes the following steps:

[0086] In step 406a, the spectral density of the predicted transmit signal, which represents the output communication signal 118 after transmission processing, is estimated. The transmit signal is a predicted transmit signal because it has not yet been transmitted. For example, the spectral density of the predicted transmit signal is estimated by simulating or modeling the characteristics of the RF front-end TX component, antenna, and communication channel, given the output communication signal 118. Additionally or alternatively, the estimated spectral density of the predicted transmit is estimated for each antenna output of the RF front-end TX component. The estimated spectral density of the predicted transmit signal may be estimated based on combining the estimated spectral density of the predicted transmit at each output of the antenna of the RF front-end TX component.

[0087] In step 406b, check whether the estimated spectral density of the predicted transmitted signal satisfies the white noise interference level. If the estimated spectral density of the predicted transmitted signal satisfies the white noise interference level (e.g., "Y"), proceed to step 406d; otherwise, proceed to step 406c.

[0088] The checking further includes detecting or identifying whether the estimated spectral density meets the white noise interference level. For example, an analysis of the power spectral density of the transmitted signal detects or identifies that the spectral density of the predicted transmitted signal does not meet the white noise interference level when the spectral density of the predicted transmitted signal forms one or more interference spikes that exceed the white noise interference level. In other embodiments, an analysis of the power spectral density of the predicted transmitted signal detects or identifies that the spectral density of the predicted transmitted signal meets the white noise interference level when the power spectral density of the transmitted signal over the target bandwidth is below the white noise interference level or within a predetermined threshold region of the white noise interference level.

[0089] In further embodiments, the check and analysis in 406b includes comparing the estimated spectral density of the predicted transmit signal with the white noise spectral density associated with the white noise interference level. If the estimated spectral density of the predicted transmit signal is less than or substantially equal to the white noise spectral density associated with the white noise interference level, proceed to step 406d. Alternatively, if the estimated spectral density of the predicted transmit signal is greater than or substantially different from the white noise spectral density associated with the white noise interference level, proceed to step 406c.

[0090] In step 406c, data is shown that indicates that the white noise interference level is not met. The indication can be based on setting a predetermined negative flag / field value (e.g., "0", "N", a negative binary value) that indicates that the white noise interference level is not met.

[0091] In step 406d, data is shown that indicates that the white noise interference level is met. The indication can be obtained by setting a predetermined positive flag / field value (e.g., "1", "Y", a positive binary value) that indicates that the white noise interference level is met.

[0092] For example, a processor or other computing device (e.g., BS DNNC214 or UE DNNC224 in Figure 2a) can automatically perform the steps of the spectral analysis process in step 406 in Figure 4a. While several embodiments have been described in which a check is performed in step 406b in relation to whether the predicted transmit signal meets the white interference noise level, these are merely examples, and it will be understood by those skilled in the art that other suitable automated methods or techniques can be applied to check whether the predicted transmit signal meets the white interference noise level, etc.

[0093] Figure 4c shows an exemplary control message process 430 in the second device 104b for receiving one or more control messages, including the NNSI, transmitted by the first device 104a in step 414 of Figure 4a. The control message process 430 includes the following steps:

[0094] In step 432, the second device 104b receives a control message from the first device 104a that includes the NNSI and scramble timing information, including a specific time slot associated with the NNSI. The specific time slot indicates when the first device 104a transmits a transmit signal 105a', which represents the output communication signal generated from the transmit DNN model 107 during scramble DNN operation.

[0095] In step 434, the second device 104b stores the received NNSI mapped to a specific time slot (or scramble timing information). The second device 104b retrieves the stored NNSI mapped to a specific time slot in order to reconstruct the received DNN model 109 before the second device 104b processes the received communication data signal corresponding to the reception of the transmitted signal 105a' for the specific time slot.

[0096] Figure 4d shows an exemplary descrambled DNN process 440 for generating reconstructed communication data 101b' by receiving the output communication signal from the transmit DNN model 107 of the first device 104a in the second device 104b. The descrambled DNN process 440 includes the following steps:

[0097] In step 442, the second device 104b receives a communication data signal 119 based on the reception of a transmission signal 105a' sent from the first device, according to scramble timing information that includes a specific time slot. For example, the received communication data signal 119 is obtained based on the reception of a transmission signal 105a' obtained from the transmission of an output communication signal generated in the transmission DNN scrambling process 400 in Figure 4a, and the transmission of the output communication signal is performed in a specific time slot or according to specific scramble timing information.

[0098] In step 444, the second device 104b obtains the NNSI for a specific time slot of scramble timing information, if any. As illustrated in Figure 4c, the NNSI is received from one or more control messages transmitted by the first device 104a and stored in the second device 104b. The second device 104b maps the received NNSI to the corresponding specific time slot(s) in which the NNSI should be used.

[0099] In step 446, the second device 104b checks whether any NNSI has been acquired for a particular time slot. For example, the second device 104b checks whether the NNSI has changed from the previous NNSI for a particular time slot. If no NNSI has been acquired or has not changed (e.g., "N"), the process proceeds to step 450, where the current received DNN model 109 is used to descramble the received communication data signal 119. If an NNSI has been acquired or has changed (e.g., "Y"), the process proceeds to step 448, where the received DNN model 109 is reconfigured according to the NNSI.

[0100] In step 448, the second device 104b configures (for the first time) or reconfigures the second device 104b's receiving DNN model 109 according to the NNSI in order to process the received communication data signal 119 according to scramble timing information or a specific time slot. Proceed to step 450.

[0101] For example, NNSI includes data representing the randomization of the input, output, or order of a set of neural network nodes in one or more specified neural network layers of the transmitting DNN model 107 of the first device 104a. Reconstruction of the receiving DNN model 109 of the second device 104b further includes reconstructing the receiving DNN model 109 of the second device 104b using NNSI to reverse the randomization applied to the specified neural network layers of the transmitting DNN model 107.

[0102] In step 450, the second device 104b processes the received communication data signal 119 with the received DNN model 109 to reconstruct the communication data represented by the received communication data signal 119.

[0103] In step 452, the second device 104b transmits the reconstructed communication data 101b' to one or more upper layers of the data sink and / or protocol stack of the second device 104b.

[0104] Figure 5 shows an exemplary DNN communication system 500 in which a first device 510 communicates with a second device 520. In this embodiment, the first device 510 includes a transmit DNN structure 506 (or TX DNN 506), a TX DNN controller 514, an output communication (OC) signal buffer 536a (also referred to as OC buffer 536a), an NNSI buffer 537a, and an RF front-end TX component 503a. Input communication data 501a transmitted in each specific time slot (e.g., time slot q(TS(q)), TS(q+1), ..., TS(q+Q), etc.) is applied to the input of the TX DNN 506. The TX DNN 506 generates corresponding OC signals 518a to 518q for transmission in each of the specific time slots (e.g., TS(q), TS(q+1), ..., TS(q+Q), etc.). Each of the generated OC signals 518a to 518q passes through the TX DNN controller 514, which acts as a gate determining whether the OC signal 518a should be passed to the RF front-end TX component 503a so that the OC signal 518a is transmitted as the transmit signal 505a for TS(q) that satisfies the white noise interference level. As described with reference to Figures 1a to 4b and / or as described herein, the TX DNN controller 514 performs the scrambled DNN operation performed by the first device 510. If the TX DNN controller 514 detects that the generated OC signal 518a satisfies the white noise interference level when transmitted by the first device 510 as the transmit signal 505a for TS(q), the TX DNN controller 514 sends the OC signal 518a to the OC buffer 536a for transmission in a specific time slot TS(q). If the TX DNN controller 514 detects that the generated OC signal 518a does not meet the white noise interference level when transmitted by the first device 510 as the transmit signal 505a at a specific TS(q), the TX DNN controller 514 does not allow the generated OC signal 518a to be supplied to the OC buffer 536a.Instead, the TX DNN controller 514 selects the NNSI for a specific time slot TS(q) to reconfigure the TX DNN 506 to generate the OC signal 518a for that specific time slot TS(q) which is expected to satisfy the white noise interference level when transmitted in that time slot TS(q).

[0105] For example, the TX DNN controller 514 selects an NNSI signal 527a from an NNSI table (not shown) in storage, and each NNSI entry is associated with a specific white noise interference level characteristic that may satisfy the white noise interference level at the time of transmission of the resulting OC signal 518a. As described with reference to Figures 1a to 4b and / or as described herein, the TX DNN controller 514 instructs the TX DNN 506 to be reconfigured for scrambled DNN operation based on the selected NNSI signal 527a, and the reconfigured TX DNN 506 reprocesses the input communication data 501a of a particular time slot TS(q) to generate an updated or rescrambled OC signal 518a. If the TX DNN controller 514 detects that the generated OC signal 518a still does not meet the white noise interference level when transmitted by the first device 510 as the transmit signal 505a, the TX DNN controller 514 repeatedly selects a new NNSI signal until the OC signal 518a meets the white noise interference level when transmitted. If the TX DNN controller 514 detects that the generated OC signal 518a now meets the white noise interference level when transmitted by the first device 510 as the transmit signal 505a, the TX DNN controller 514 sends the OC signal 518a to the OC buffer 536a for transmission in a specific time slot TS(q). The TX DNN controller 514 also sends the corresponding selected NNSI signal 527a to the NNSI buffer 537a. Each NNSI signal 527a includes data representing a specific applicable time slot (e.g., TS(q)) and an indication of the randomization / scrambling parameters used to reconstruct the TX DNN 506, which the RX DNN controller 524 may use to reconstruct the RX DNN 508 to perform descrambling and generate reconstructed communication data 501b.To instruct the second device 520 to reconfigure the RX DNN 508 so that it processes the corresponding received OC signal for a specific time slot TS(q) and descrambles it to generate reconstructed communication data 501b for the specific time slot TS(q), the first device 510 sends each NNSI signal 527a in the NNSI buffer 537a to the second device 520 in a control message 505b at a suitable time (for example, before sending the corresponding OC signal 518a). Typically, the first device 510 sends the selected NNSI signal 527a for a specific time slot TS(q) before sending the OC signal 518a for that time slot TS(q). In other embodiments, the first device 510 sends the selected NNSI signal 527a in the control message after sending the OC signal 518a, and the second device 520 uses the received NNSI signal 527a to reconfigure the RX DNN 508 before processing the corresponding received OC signal.

[0106] Each of the OC signals 518a-518q and / or NNSI signals 527a-527p in the OC buffer 536a and NNSI buffer 537a is output for RF processing at a suitable time. For example, each of the OC signals 518a-518q is RF processed by the RF front-end TX component 503a for transmission in their specific time slots TS(q), TS(q+1), ..., TS(q+Q) and transmitted over the data communication channel as a data transmission signal 505a. For example, each of the NNSI signals 527a-527p is RF processed by the RF front-end TX component 503a for transmission before the specific time slots TS(q), TS(q+1), ..., TS(q+Q) used to transmit the OC signals 518a-518q and transmitted over the control communication channel as a control message 505b. In other embodiments, each of the NNSI signals 527a to 527p is RF processed by the RF front-end TX component 503a for transmission before the second device 520 processes the corresponding received OC signals 518a to 518q, and is transmitted as one or more control messages 505b over the control communication channel.

[0107] For example, when the first device 510 is a base station and the second device 520 is user equipment, the data communication channel includes a downlink data channel, for example, a physical downlink shared channel (PDSCH). Alternatively, when the first device 510 is user equipment and the second device 520 is a base station, the data communication channel includes an uplink data channel, for example, a physical uplink shared channel (PUSCH). Similarly, when the first device 510 is a base station and the second device 520 is user equipment, the control communication channel includes a downlink control channel, for example, a physical downlink control channel (PDCCH). Alternatively, when the first device 510 is user equipment and the second device 520 is a base station, the control communication channel includes an uplink control channel, for example, a physical uplink control channel (PUCCH).

[0108] The second device 520 includes a receive DNN structure 508 (or RX DNN 508), an RX DNN controller 524, a receive output communication (Rx OC) signal buffer 536b (also referred to as Rx OC buffer 536b), an NNSI buffer 537b, and an RF front-end RX component 503b. The RF front-end RX component 503b receives a transmit signal 505a over a data communication channel at a specific time slot (e.g., TS(Q+1)) and processes the received transmit signal 505a to generate a baseband receive OC data signal (Rx OC signal) 519a. The RF front-end RX component 503b supplies the Rx OC signal 519a for the specific time slot (e.g., TS(Q+1)) to the Rx OC buffer 536b. In this embodiment, the Rx OC buffer 536b is a first-in, first-out (FIFO) queue, but it may be an embodiment of any type of buffer / queue. The RF front-end RX component 503b also occasionally receives a control message 505b, which includes an NNSI signal 529a for a specific time slot (e.g., TS(Q+1)), transmitted via the control communication channel, and processes the received control message 505b to generate the NNSI signal 529a. The RF front-end RX component 503b supplies the NNSI signal 529a for the specific time slot (e.g., TS(Q+1)) to the NNSI buffer 537b. The Rx OC buffer 536b is connected to the RX DNN controller 524, which receives an Rx OC signal 519q from the Rx OC buffer 536b corresponding to a specific time slot (e.g., TS(Q+P)). As described with reference to Figures 1a to 4d and / or as described herein, the RX DNN controller 524 performs a reverse descrambled DNN operation in contrast to the scrambled DNN operation performed by the first device 510.

[0109] For example, the RX DNN controller 524 controls when the Rx OC signal 519q of a specific time slot TS(Q+P) is input to the RX DNN 508 in order to generate reconstructed communication data 501b corresponding to the input communication data 501a of that specific time slot TS(Q+P). In this case, the Rx DNN controller 524 identifies the existence of an NNSI signal 529q of a specific time slot TS(Q+P) to reconfigure the RX DNN 508 so as to descramble the Rx OC signal 519q received at the time slot TS(Q+P). Before processing the Rx OC signal 519q of the time slot TS(Q+P), the RX DNN controller 524 uses the NNSI signal 529q of the time slot TS(Q+P) to reconfigure the RX DNN 508. After reconfiguration, the RX DNN controller 524 inputs the Rx OC signal 519q to the reconfigured RX DNN 508 to generate reconfigured communication data 501b for a specific time slot TS(Q+P). The second device 520 either sends the reconfigured communication data 501b for the specific time slot TS(Q+P) to its data sink or to one or more upper protocol layers of the second device 520's protocol stack.

[0110] As can be seen in the figure, the second device 520 receives several NNSI signals 529a, 529b, and 529p into the NNSI buffer 537b, which correspond to the Rx OC signals 519a, 519b, and 519p for specific time slots TS(Q+1), TS(Q+2), and TS(Q+P-1). The RX DNN controller 524 uses the corresponding NNSI signals 529a, 529b, 529o, and 529p to reconfigure the RX DNN 508 before processing each of the corresponding Rx OC signals 519a, 519b, and 519p. Since the TX DNN controller 514 may determine that the same NNSI is applicable to multiple consecutive time slots, the TX DNN controller 514 does not necessarily have to update the NNSI signal for each time slot. In such cases, when a control message containing updated NNSIs for several subsequent time slots is received, the RX DNN controller 524 only needs to reconfigure the RX DNN 508. For example, there may be no further NNSI changes between TS(Q+P-2) and TS(Q+2). Therefore, with respect to time slots TS(Q+3) to TS(Q+P-2), the RX DNN controller 524 only needs to reconfigure the RX DNN 508 using the NNSI signal 529o until after the Rx OC signal 519b is processed at TS(Q+2), so that the Rx OC signals 519c up to Rx OC signal 519p, which do not include Rx OC signal 519p, are processed by the same configuration of the RX DNN 508.

[0111] Figure 6a shows an exemplary substitution operation 600 for performing random substitution on a specific order of one or more neural network nodes 641 in the neural network layer of the transmitting DNN model for scrambled DNN operation, and for performing random reverse substitution (also called de-substitution) on a specific order of a set of neural network nodes 642 in the neural network layer of the receiving DNN model for de-scrambled DNN operation.

[0112] In this embodiment, the set of neural network nodes 641 in a particular neural network layer of the transmit DNN model has a specific order, for example, indicated by node labels 1, 2, 3, 4, and 5. The transmit DNN model is reconfigured by randomizing the specific order of the set of neural network nodes 641 in a particular neural network layer using a random replacement operation 616. The random replacement operation 616 replaces or randomizes the specific order of the neural network nodes 641, resulting in a replaced or randomized order of the neural network nodes 642 for that particular neural network layer of the transmit DNN model. The specific order of the set of neural network nodes 642 in a particular neural network layer after the random replacement operation 616 is performed is indicated, for example, by the reordered node labels 2, 1, 3, 5, and 4. The input connection to neural network node 1 before the random substitution operation 616 is now input to neural network node 2, the input connection to neural network node 2 before the random substitution operation 616 is now input to neural network node 1, the input connection to neural network node 3 before the random substitution operation 616 is still input to neural network node 3, the input connection to neural network node 4 before the random substitution operation 616 is now input to neural network node 5, and the input connection to neural network node 5 before the random substitution operation 616 is now input to neural network node 4. The outputs of the set of neural network nodes 642 (e.g., nodes 1, 2, 3, 4, and 5) are still connected to the same neural network nodes in other neural network layers of the reconstructed transmit DNN model, but the random substitution of inputs to the set of neural network nodes 642 results in scrambling of the resulting transmit DNN model outputs. This may be referred to as scrambled DNN operation.

[0113] For a particular random permutation operation 616, the NNSI includes the data necessary to reverse the random permutation operation 616, which a second device uses to reconstruct the corresponding incoming DNN model to reverse the random permutation operation 616. The second device identifies the inverse neural network layer of the incoming DNN model and uses the NNSI to reconstruct the incoming DNN model using a random inverse permutation operation or random depermutation operation 617 for a specific order of the set of neural network nodes in the identified neural network layer of the incoming DNN model. In this embodiment, the set of neural network nodes 643 in a particular neural network layer of the incoming DNN model has a specific order, for example, indicated by node labels 2, 1, 3, 5, 4. The incoming DNN model is reconstructed by randomizing the specific order of the set of neural network nodes 643 in the particular neural network layer using the random depermutation operation 617. Random de-substitution operation 617 de-substitutes a specific order of neural network nodes 643, resulting in a de-substituted or de-randomized order of neural network nodes 644 for that particular neural network layer of the receiving DNN model. The specific order of the set of neural network nodes 644 in a particular neural network layer after random de-substitution operation 617 is indicated by the reordered node labels 1, 2, 3, 4, and 5.This means that the input connection to neural network node 1 before the random substitution de-operation 617 is now input to neural network node 2, the input connection to neural network node 2 before the random substitution de-operation 617 is now input to neural network node 1, the input connection to neural network node 3 before the random substitution de-operation 617 is still input to neural network node 3, the input connection to neural network node 4 before the random substitution de-operation 617 is now input to neural network node 5, and the input connection to neural network node 5 before the random substitution de-operation 617 is now input to neural network node 4. The outputs of neural network nodes 1, 2, 3, 4, and 5 are still connected to the same neural network nodes in other neural network layers of the reconstructed receiving DNN model, but the random substitution of inputs to the set of neural network nodes 644 results in descrambling of the resulting receiving DNN model output. This may be referred to as descrambled DNN operation.

[0114] NNSI may include, for example, randomization parameters / randomization functions for randomizing the order of a set of neural network nodes in the input, output, or one or more specified neural network layers of the transmit DNN that are to be randomized. The one or more specific neural network layers of the transmit DNN may include, but are not limited to, one or more of the input neural network layers (or input layer), output neural network layers (or output layer), and / or hidden neural network layers of the transmit DNN model, and / or combinations thereof. The specified neural network layers of the transmit DNN model are randomized / scrambled. NNSI is used to reconfigure the transmit DNN model and the receive DNN model to perform scrambled DNN operation.

[0115] The second device uses NNSI to reconfigure the receiving DNN model to synchronize with the reception of a transmit signal representing the output communication signal of the reconfigured transmit DNN model. To perform random substitution operations, NNSI further includes data representing one or more random substitution parameters and / or random substitution functions, which include one or more of the following: seed data for performing random substitutions on a specific order of a set of neural network nodes in a given neural network layer of the transmit DNN model, the number of iterations or sequence number of the random substitution, or identification of the seed generation and pseudo-randomization functions. The specific seed data may include, but is not limited to, identification information of the second device (e.g., UE), identification information of the first device (e.g., BS), identification information of the cell where the second device is located, identification information of the cell where the first device is located, timing slot information, frame identification number, and / or any other information associated with the first or second device, the number of iterations or sequence number of a particular random substitution. For example, an initial seed or initial seed value may be generated from seed data using a seed generation and pseudo-randomization function, which is used to randomize and / or derandomize neural network nodes in one or more specified neural network layers of a sending DNN model (TX DNN) and / or receiving DNN model (RX DNN), respectively.

[0116] For example, a second device receives an NNSI containing one or more random replacement parameters, such as seed data, the number of random replacement iterations or sequence number, or an identifier for seed generation, an indication of one or more neural network layers scrambled by the first device, and a pseudo-randomization function, and uses this to perform random replacements on a set of neural network nodes in the indicated one or more neural network layers. The second device also receives corresponding timing information associated with when the RX DNN should be reconfigured (e.g., when the corresponding neural network layers of the RX DNN should be descrambled). The timing information includes one or more time slots. Reconfiguring the RX DNN of the second device at a particular time slot may involve generating an inverse random replacement sequence corresponding to the number of random replacement iterations or sequence number for each neural network layer of the RX DNN corresponding to each indicated neural network layer of the TX DNN. The inverse random replacement sequence has a length equal to the number of neural network nodes in the set of neural network nodes in each of the neural network layers of the RX DNN. For each neural network layer of the RX DNN corresponding to each neural network layer of the TX DNN, derandomization or inverse permutation of the set of neural network nodes in that RX DNN layer is performed by applying the generated inverse random permutation sequence to the set of neural network nodes. Alternatively, an inverse random permutation matrix may be generated and applied to the order of the set of neural network nodes, etc.

[0117] Figure 6b shows an exemplary random permutation sequence 650 starting with an initial seed 652 (also referred to as the initial seed value 652). For a particular neural network layer having a set of N neural network nodes, the initial seed 652 is input to the random permutation function to generate the first random permutation sequence 651a of size N. The first random permutation sequence 651a can be an integer permutation in the range [1,N], where the integer represents the node label of the set of N neural network nodes. The first random permutation sequence 651a is generated from permutation iteration 1 of the random permutation function using the initial seed 652. Further iterations or cycles of the random permutation function using the initial seed 652 generate different random permutation sequences of size N, for example, the second random permutation sequence 651b in the second permutation iteration 2, the third random permutation sequence 651c in the third permutation iteration 3, ..., the i-th random permutation sequence 651i in the i-th permutation iteration i, ..., the n-th random permutation sequence 651n in the n-th permutation iteration n, and so on. Given an initial seed value of 652 and a pair of random permutation functions, the i-th random permutation sequence of length N (or the i-th cycle of random permutation sequences of length N) can be generated by simply specifying the number of random permutation iterations as i, and repeating the process of generating a random permutation sequence using the initial seed 652 and the random permutation function until the i-th cycle.

[0118] For example, the initial control message may include an NNSI, which has an initial seed value, a random substitution function indication, the number of random substitution iterations, and data representing one or more specified neural network layers of a randomized transmitting DNN model as described with reference to Figures 1a to 6a. The first device sends an initial control message to the second device to reconstruct the corresponding neural network layers of the receiving DNN model as described with reference to Figures 3a to 3c, using the received initial seed value, random substitution function indication, the number of random substitution iterations, and the specified neural network layers, in order to generate a corresponding random substitution sequence associated with the number of random substitution iterations. The generated random substitution sequence generates a random substitution de-(or reverse substitution) operation 617, which is used to reconstruct the corresponding neural network layers of the receiving DNN model in the second device to perform a de-scramble DNN operation and generate reconstructed communication data. The second device sends the reconstructed communication data to the data sink of the second device, or sends the reconstructed communication data to one or more higher protocol layers of the protocol stack of the second device. Furthermore, the first device transmits subsequent updates to the NNSI used to reconstruct the transmitting DNN model in subsequent control messages, which include an indication of the updated random permutation iteration count. When the initial seed and / or random permutation function are updated or modified, these are transmitted in further control messages or one or more further control messages.

[0119] Figure 6c shows the permutation and depermutation (or reverse permutation) operation 660, and the random permutation operation 616 uses the random permutation matrix 616' obtained from the selected i-th random permutation sequence 652i in Figure 6b, which is used to scramble / descramble one or more neural network layers of the transmit / receive DNN model. The i-th N×N random permutation matrix P iTo generate 616’, the i-th random permutation sequence 652i is used. For example, the i-th N×N random permutation matrix P i Calculating 616’ is done by generating an N×N identity matrix I where each column is consecutively labeled from 1 to N, and then replacing the columns of the identity matrix I using the generated i-th random permutation sequence 652i. Thereby, the random permutation matrix P i 616’ is generated, which represents the i-th random permutation sequence 652i. The random permutation operation 616 is i performed by multiplying 616’ by an N-dimensional vector, thereby generating a permuted N-dimensional vector replaced according to the i-th random permutation sequence 652i. The random permutation matrix P i After 616’ is calculated, the permutation matrix P i is inverted to generate an inverse permutation matrix D i 617’ (based on D i =(P i )), the random inverse permutation operation 617 is derived. The random inverse permutation operation 617 can be performed by multiplying the inverse permutation matrix D i 617’ by an N-dimensional vector that requires inverse permutation, thereby generating an inverse permuted N-dimensional vector.

[0120] For example, regarding a selected NNSI indicating the i-th random permutation iteration count of the i-th random permutation sequence 652i, where 1≦I≦L, I = 1 is the input layer, I = L is the output layer, and L is the number of neural network layers including one or more hidden neural network layers, reconstructing the I-th neural network layer of the transmitting DNN model involves generating the i-th random permutation sequence 652i and, assuming N>1 and N is the number of neural network nodes in the I-th neural network layer, using the i-th random permutation sequence 652i for the I-th neural network layer to form an N-dimensional random permutation matrix P i ​​The process involves generating 616' and assigning a specific order vector representing the current specific order of the set of N neural network nodes in the Ith neural network layer to an N-dimensional random permutation matrix P. i This is done by multiplying by 616' to generate a randomized ordered vector representing the set of N substituted neural network nodes.

[0121] For example, the first device selects an NNSI that indicates the number of random permutation iterations of the i-th random permutation sequence 652i and that the i-th neural network layer of the transmitting DNN model has been reconstructed. The first device transmits the selected NNSI to the second device in a control message as described herein. Given the selected NNSI, the second device reconstructs the (L-I+1)-th neural network layer of the receiving DNN model by generating the i-th random permutation sequence 652i and using the i-th random permutation sequence 652i in the (L-I+1)-th neural network layer, where N>1 and N is the number of neural network nodes in the (L-I+1)-th neural network layer, to create an N-dimensional random permutation matrix P i The process involves generating 616' and creating an N-dimensional random permutation matrix P. i By inverting 616', the N-dimensional depermutation matrix D i The process involves generating 617' and assigning a specific order vector representing the current specific order of the set of N neural network nodes in the (L-I+1)th neural network layer to an N-dimensional depermutation matrix D. i This is done by multiplying by 617' to generate a de-substituted or de-scrambled ordered vector (or inverse random permutation sequence) representing a set of N de-substituted neural network nodes, and the de-substituted or de-scrambled ordered vector (or inverse random permutation sequence) is used to reconstruct the (L-I+1)th neural network layer of the received DNN model.

[0122] Figure 6d shows an exemplary iterative NNSI selection process 670 performed by the first device to select an i-th random substitution sequence to use in a randomization operation that randomizes the order of neural network nodes in one or more neural network layers of a transmit DNN model, the i-th random substitution sequence reconstructs the transmit DNN model, and the reconstructed transmit DNN model, given input communication data as input, generates an output communication signal that satisfies the white noise interference level at transmission. The iterative NNSI selection process 670 performed by the first device includes one or more of the following steps:

[0123] In step 672, the first device selects or generates an i-th random permutation sequence (or i-th randomization operation) to randomize one or more neural network layers of the transmit DNN model. For example, the first device selects the i-th random permutation iteration from the NNSI table that is mapped to an indication of a white noise interference level characteristic that satisfies the current white noise interference level. Alternatively, the first device generates an i-th random permutation iteration (e.g., selects the value i) to be used to generate the i-th random permutation sequence. The i-th random permutation iteration (selected or generated) is used to generate the i-th cycle random permutation sequence (referred to as the i-th random permutation sequence) using an initial seed and a specific random permutation function. The i-th random permutation sequence has a length N > 1, where N is equal to the number of neural network nodes in the set of neural network nodes in a particular neural network layer of the transmit DNN model. The i-th random permutation sequence is used to randomly permut a set of neural network nodes in a particular neural network layer.

[0124] In step 674, the first device reconstructs the transmission DNN model using the i-th random substitution sequence selected / generated in step 672.

[0125] In step 676, the first device processes the input communication data with the reconstructed transmission DNN model in order to generate an output communication signal for transmission.

[0126] In step 678, the first device analyzes and determines whether the transmission of the generated output communication signal satisfies the white noise interference level. This may include analyzing and / or estimating the spectral density of the predicted transmitted signal waveform if the first device transmits the generated output communication signal. If it is determined that the generated output communication signal produces a predicted transmitted signal waveform that satisfies the white noise interference level when transmitted (e.g., "Y"), proceed to step 682; otherwise (e.g., "N"), proceed to step 680.

[0127] For example, the determination in step 678 is made in relation to analyzing the spectral density of a predicted transmitted signal obtained by processing the output communication signal for transmission, as described with reference to Figures 1a to 4b, in particular Figures 3a to 3b and 4b. The first device (or its components) automatically analyzes the output communication signal to determine whether the transmission of the output communication signal satisfies the white noise interference level. For example, a trained spectral density estimation model processes the output communication signal to predict or estimate the power spectral density of the transmitted signal obtained from the output communication signal. In other embodiments, the output communication signal is input to a simulation model to generate a predicted transmitted signal, and the power spectral density of the predicted transmitted signal is analyzed with respect to the white interference noise level, as described with reference to Figures 3a to 4b and / or as described herein. In other embodiments, a simulation of the transmission of the output communication signal over a simulated communication channel is performed, and the simulation analyzes whether the simulated transmission satisfies the white noise interference level and shows the results.

[0128] In step 680, the first device updates the iteration count i to i=i+1, and the iterative NNSI selection process 670 proceeds to step 672 for the first device to generate / select another i-th random substitution sequence. That is, the generation step 672, reconstruction step 674, processing step 676, and analysis step 678 are repeated for the next i-th iteration.

[0129] In step 682, the first device shows the selected / generated i-th substitution sequence or i-th substitution iteration, which is used to reconfigure the transmit DNN model to produce an output communication signal that satisfies the white noise interference level during transmission.

[0130] For example, in step 682, indicating the selected / generated i-th permutation sequence includes indicating the i-th random permutation iteration to be included in the NNSI, and as described herein, the first device sends the NNSI, including the i-th random permutation iteration, to the second device in a control message. Initially, the NNSI includes at least an initial seed, a random permutation function, the i-th random permutation iteration, and one or more neural network layers of the transmit DNN model, which are used to generate the i-th random permutation sequence(s) for reconstructing the transmit DNN. Subsequent updates to the NNSI may include indication of the selected / generated i-th random permutation iteration.

[0131] In addition to selecting the i-th random permutation sequence / i-th random permutation iteration, step 672 includes the first device selecting a set of one or more neural network layers, the selected set of one or more neural network layers being different from the previous selection of one or more neural network layers. The NNSI includes the selected set of one or more neural network layers that produce an output communication signal satisfying the white interference level at transmission.

[0132] The iterative NNSI selection process 670 is performed by the first device or its controller component (e.g., BS DNNC214 in Figure 2a) and, based on the analysis performed in step 678, a mapping may be used between the whitening characteristics of the output communication signal at transmission and the i-th random substitution sequence / i-th random substitution iteration and / or other NNSI data (e.g., one or more selected neural network layers) to populate an NNSI lookup table accessible by the first device. A further modification includes, for example, in step 672, the first device selecting the i-th random substitution sequence by obtaining from the NNSI lookup table a random substitution sequence or random substitution iteration mapped to a white noise interference characteristic that may satisfy the white noise interference level. This is based on the possibility that the corresponding whitening characteristics indicate that the selected NNSI will result in the transmission of the output communication signal of the transmission DNN that satisfies the white noise interference level. In other embodiments, the first device uses an NNSI table to bootstrap an iterative NNSI selection process 670 when searching for, for example, the i-th random substitution sequence / i-th random substitution iteration and / or other NNSI data (e.g., one or more selected neural network layers) that produce an output communication signal satisfying a white noise interference level during transmission.

[0133] Optionally, one or more neural network layers in a transmit DNN model include one or more neural network layers of a transmit DNN model consisting of an input neural network layer, an output neural network layer, and one or more hidden neural network layers. One or more neural network layers may be predefined or preselected before the scrambled DNN operation. The selection of one or more neural network layers may be made considering the capabilities of the first and / or second devices. For example, if the second device does not have the capability to descramble the hidden neural network layer, the input and / or output neural network layers may be selected to reduce complexity and / or computational resource consumption. Alternatively or additionally, one or more neural network layers may be randomly selected before the scrambled DNN operation and / or when a new NNSI is generated for whitening the transmit signal.

[0134] Figure 7 shows an exemplary signal flow of scrambled DNN communication 720 during a communication session between the first device 104a and the second device 104b. The first device 104a and the second device 104b in Figure 1 perform scrambled DNN communication 720 using any of the embodiments described with reference to Figures 1 to 6c. For example, the first device 104a and the second device 104b perform scrambled DNN communication 720 as the base station 210 and user equipment 220 in Figure 2a, or as the first device 310 and the second device 320 in Figures 3a to 3c. The signal flow of scrambled DNN communication 720 for a communication session between the first device 104a and the second device 104b includes the following signal flow operation.

[0135] In operation 721a, the first device 104a and the second device 104b establish a DNN connection during a communication session between them. While establishing the DNN connection, the first and second devices communicate with each other to define, agree on, and / or configure the types of transmit DNN structures and receive DNN structures that will enable end-to-end communication between them.

[0136] For example, after the first device 104a initiates a standard communication session with the second device 104b, the first device 104a selects the type of transmit DNN structure to use for the DNN connection with the second device 104b, depending on the communication channel conditions / environment, the communication performance requirements for the DNN connection, and the type of data communication being performed (e.g., voice communication, data communication, and multimedia streaming). The first device 104a requests the machine learning processing capabilities of the second device 104b to assist in selecting the transmit DNN structure and / or receive DNN structure to use during the DNN connection. The first device 104a accesses a DNN lookup table (or neural network table), which contains a set and / or pair of transmit / receive DNN structures / models mapped to DNN identifiers stored therein. In this embodiment, the first device 104a stores the DNN lookup table within itself. The second device 104b similarly accesses a corresponding DNN lookup table in which a set of corresponding transmit / receive DNN structures / models is mapped to the same DNN identifier. In the embodiment, the corresponding DNN lookup table stores each receive DNN structure / model and a mapping of the DNN identifiers corresponding to the transmit / receive DNN structure / model pair stored in the DNN lookup table accessible by the first device 104a. In the embodiment, the second device stores the corresponding DNN lookup table within itself.

[0137] In this embodiment, the first device 104a selects a suitable transmitting DNN structure / model and / or a corresponding receiving DNN structure / model for the DNN connection. After selecting the transmitting DNN structure and / or the corresponding receiving DNN structure, the first device 104a initiates the DNN connection by sending a DNN connection request message to the second device 104b, the DNN connection request message including a field requesting the establishment of a DNN connection, a DNN identifier corresponding to the selected transmitting DNN structure / receiving DNN structure and / or pair thereof, or an indication of the type of receiving DNN structure that the second device 104b should use and / or the type of transmitting DNN structure that the first device 104a will use, so that the second device 104b selects a suitable receiving DNN structure. In some embodiments, when full-duplex communication is performed, i.e., when both downlink and uplink communication is performed, the first device 104a has a transmit DNN structure and a receive DNN structure for communicating with the second device 104b, and the second device 104b also has a corresponding receive DNN structure and a corresponding transmit DNN structure for communicating with the first device 104a. The first device 104a and the second device 104b establish a DNN connection, and the first device 104a configures a selected transmit DNN structure (and / or its receive DNN structure) to perform DNN communication with the second device 104b via the communication channel in one or more time slots. The second device 104b configures a corresponding receive DNN structure (and / or its corresponding transmit DNN structure) to perform DNN communication with the first device 104a via the communication channel in one or more time slots.

[0138] In operation 721a, the white noise interference level is also initially set during the establishment of the DNN connection, either by the first device 104a or in response to a request from the second device 104b seeking improved block error rate performance (for example, an increase in the white noise level is requested).

[0139] In operation 721b, after a DNN connection is established between a first device 104a that operates a transmit DNN structure (and / or receive DNN structure) and a second device 104b that operates a corresponding receive DNN structure (and / or corresponding transmit DNN structure), the first device 104a uses the transmit DNN structure and receive DNN structure configured by the first device 104a and the second device 104b, respectively, to perform DNN communication with the second device 104b in one or more time slots. In this embodiment, the first device 104a processes input communication data using the transmit DNN structure to transmit it to the second device 104b, and the second device 104b processes the received transmission using the receive DNN structure to generate reconstructed communication data. Additionally or alternatively, the second device 104b processes input communication data using a transmit DNN structure to transmit it to the first device 104a, and the first device 104a processes the received transmit using a receive DNN structure to generate reconstructed communication data. The second device 104b transmits the reconstructed communication data to its data sink or to one or more higher protocol layers of the second device 104b's protocol stack.

[0140] In operation 792, if the first device 104a detects that the transmission from the first device 104a does not meet a certain white noise interference level, the first device 104a and the second device 104b perform scrambled DNN operation. In the case of scrambled DNN communication from the first device 104a to the second device 104b (e.g., DL DNN scrambling), as described with reference to Figures 1 to 6d, in particular Figures 4a to 5, the TX DNN controller of the first device 104a controls the scrambled DNN operation of the first device 104a, and the RX DNN controller of the second device 104b controls the descrambled DNN operation of the second device 104b. In the case of scrambled DNN communication from the second device 104b to the first device 104a (e.g., UL DNN scrambling), the TX DNN controller of the second device 104b controls the scrambled DNN operation of the second device 104b, and the RX DNN controller of the first device 104a controls the descrambled DNN operation of the first device 104a. For brevity and as a mere example, the following steps describe scrambled DNN communication from the first device 104a to the second device 104b, but similar or identical operation is applicable to scrambled DNN communication from the second device 104b to the first device 104a, with the first device 104a exchanging roles with the second device 104b.

[0141] In this embodiment, the first device 104a sets the white noise interference level in response to a request from a second device 104b seeking improved block error rate performance (for example, an increase in the white noise interference level is requested), or in response to a request from a third device (not shown) that is experiencing unacceptable white noise interference caused by transmissions from the first device 104a during DNN communication. In operation 792, the first device 104a and the second device 104b perform scrambled DNN operations based on, but not limited to, the following scrambled DNN operations:

[0142] In operation 722a, if the output communication signal from the transmit DNN structure does not meet the white noise interference level during transmission, the first device 104a enables the execution of scrambled DNN operation for one or more time slots. For example, the TX DNN controller of the first device 104a detects that the output communication signal generated by the transmit DNN model of the transmit DNN structure from the input communication data is a transmit signal that does not meet the white noise interference level. In such cases, scrambled DNN operation is enabled, and the TX DNN controller of the first device 104a selects an NNSI, which is used to reconfigure the transmit DNN model of the transmit DNN structure (referred to as reconfiguring the transmit DNN structure) so that it generates an output communication signal that meets the white noise interference level from the same input communication data, as described with reference to Figures 1, 2, and 3a-6d, particularly Figures 3a-6d.

[0143] In operation 723a, the first device 104a communicates to the second device 104b when and how the scrambled DNN operation is enabled by sending a control message, the control message including a selected NNSI and scramble timing information regarding when the received DNN structure should be reconfigured in accordance with the scrambled DNN operation at the second device 104b. For example, the first device 104a sends a control message to the second device 104b including data representing the selected NNSI and an indication of one or more time slots representing scramble timing information regarding when the first device 104a uses the selected NNSI to reconfigure the transmitted DNN structure. Initially, the selected NNSI includes initial seed information, the type of random substitution function, one or more specific neural network layers of the transmitting DNN structure to be reconfigured for scrambling, random substitution iterations, and any other data that enables the second device 104b to reconfigure the receiving DNN model of the receiving DNN structure (also referred to as reconfiguring the receiving DNN structure) to generate the descrambled and reconstructed communication data. Subsequent control messages for further scrambling DNN operation include an updated NNSI, such as the number of selected random substitution iterations or one or more selected neural network layers reconfigured for scrambling.

[0144] In operation 723b, the second device 104b sends an acknowledgment (e.g., ACK) indicating that it has received the selected NNSI configuration and that it has successfully enabled the scrambled DNN operation on the second device 104b.

[0145] In operations 724a and 724b, the first device 104a and the second device 104b enable scramble DNN operations based on scramble timing information. For example, in a scramble DNN operation from the first device 104a to the second device 104b, the first device 104a prepares to perform a scramble TX DNN operation on the first device 104a in order to transmit to the second device 104b, and the second device 104b prepares to perform a descramble RX DNN operation on the second device 104b based on scramble timing information (e.g., one or more time slots) in order to receive a transmission from the first device 104a. For example, a scramble TX DNN operation includes the first device 104a reconfiguring the transmit DNN structure based on a selected NNSI for each specific time slot associated with the scramble timing information. The descramble RX DNN operation involves the second device 104b reconfiguring the receive DNN structure based on a selected NNSI for each specific time slot before the receive DNN configuration processes the transmissions arriving for each specific time slot associated with the scramble timing information. Similarly, if there is a DNN scramble operation from the second device 104b to the first device 104a (e.g., an uplink scramble DNN operation), the second device 104b prepares to perform a scramble TX DNN operation similar to that described above with reference to the first device, and the first device 104a prepares to perform a descramble RX DNN operation as described above with reference to the second device 104b.

[0146] In operation 725, the first and second devices communicate with each other by performing scramble DNN operations based on scramble timing information (e.g., one or more specific time slots in which scrambling takes place), using corresponding transmit DNN and receive DNN structures. For example, the first device 104a performs a scramble TX DNN operation in the first device 104a in order to transmit to the second device 104b. The second device 104b performs a reverse descramble RX DNN operation in the second device 104b based on scramble timing information (e.g., one or more time slots) in order to receive a transmission from the first device 104a. For example, the scramble TX DNN operation includes the first device 104a reconfiguring the transmit DNN structure based on a selected NNSI for each specific time slot. Scrambled RX DNN operation includes the second device 104b reconfiguring the received DNN structure based on a selected NNSI for a particular time slot before the received DNN structure processes the transmissions arriving for a particular time slot. If the first device 104a and the second device 104b are performing scrambled DNN communication from the second device 104b to the first device 104a (e.g., uplink scrambled DNN operation), the second device 104b performs scrambled TX DNN operation in the same manner as described above for the first device 104a, and the first device 104a performs scrambled RX DNN operation in the same manner as described above for the second device 104b.

[0147] Furthermore, when performing a scrambled DNN operation during communication between the first device 104a and the second device 104b, the first device 104a repeatedly determines a subsequent NNSI using a common / initial seed and random function already transmitted in the first control message of operation 723a to transmit in one or more subsequent time slots. This subsequent NNSI contains the minimum randomization information necessary to enable the second device 104b to reconfigure the receiving DNN structure to process the transmission received in the subsequent time slot. For example, the subsequent NNSI includes, for example, a specified random substitution sequence number or order. The first device 104a transmits the subsequent NNSI to the second device 104b in one or more fields of the subsequent control message. The subsequent NNSI also includes, for example, one or more specific neural network layers that perform randomization / scrambling. In subsequent scrambled DNN operation, when the output communication signal becomes a transmit signal that does not meet the white noise interference level, only the updated number of random substitution iterations or sequence number is required to synchronize the reconfiguration of the receive DNN structure of the second device 104b to process the transmit received in the subsequent time slot. The second device 104b reuses the common / initial seed data, common seed generation, and pseudo-randomization function indication transmitted in the initial control message.

[0148] In operation 726, the first device 104a disables the scrambled DNN operation.

[0149] In operation 727a, the first device 104a sends a communication to the second device 104b informing it that the scrambled DNN operation is being disabled / stopped.

[0150] In operation 727b, the second device 104b sends an acknowledgment (e.g., ACK) regarding the reception of the communication.

[0151] When the first device 104a receives an acknowledgment (ACK) from the second device 104b in operation 727b, in operation 728a it stops the scrambled DNN operation and restores the transmit DNN structure to its original configuration. If the first device 104a and the second device 104b are performing scrambled DNN communication from the second device 104b to the first device 104a (e.g., uplink scrambled DNN operation), the second device 104b has a transmit DNN structure, the first device 104a has a corresponding receive DNN structure, and the first device 104a also restores its own receive DNN structure to its original configuration.

[0152] When the second device 104b receives a communication in operation 727a to disable the scrambled DNN operation, in operation 728b it stops the scrambled DNN operation and restores the received DNN structure to its original configuration. If the first device 104a and the second device 104b are performing scrambled DNN communication from the second device 104b to the first device 104a (e.g., uplink scrambled DNN operation), the second device 104b has a transmit DNN structure, the first device 104a has a corresponding receive DNN structure, and the second device 104b also restores its transmit DNN structure to its original configuration.

[0153] In operation 729, after the first device 104a and the second device 104b disable the scrambled DNN connection, the first device 104a and the second device 104b continue DNN communication by the first device 104a activating the original transmit DNN structure (and / or receive DNN structure) and the second device 104b activating the original corresponding receive DNN structure (and / or corresponding transmit DNN structure). For example, the first device 104a uses the transmit DNN and receive DNN configured by the first device 104a and the second device 104b, respectively, to communicate with the second device 104b in one or more time slots. In this embodiment, the first device 104a processes the input communication data using the transmit DNN structure to send it to the second device 104b, and the second device 104b processes the received transmission using the receive DNN structure to generate reconstructed communication data. Additionally or alternatively, the second device 104b processes the input communication data using a transmit DNN structure to send it to the first device 104a, and the first device 104a processes the received transmit using a receive DNN structure to generate reconstructed communication data. After the DNN communication is completed, the first device 104a and the second device 104b return to running a standard or conventional communication session.

[0154] Figure 8 shows the signal flow of an embodiment of the scrambled DNN operation 825 between the first device 104a and the second device 104b during the operation 725 of the scrambled DNN communication 720 shown in Figure 7. The first device 104a and the second device 104b perform the scrambled DNN operation 825 of communication from the first device 104a to the second device 104b in Figure 1 using any of the embodiments described with reference to Figures 1 to 6c. In other embodiments, the base station 210 and user equipment 220 in Figure 2a perform the scrambled DNN operation 825. In other embodiments, the first device 310 and the second device 320 in any of the figures 3a to 3c perform the scrambled DNN operation 825, or the first device 510 and the second device 520 in Figure 5, etc., perform the scrambled DNN operation 825. The first device 104a and the second device 104b perform the scrambled DNN operation 825, but this is merely an example and not limiting. It will be understood by those skilled in the art that the scrambled DNN operation 825 is also applicable to communication from the second device 104b to the first device 104a, and that the roles of the first device 104a and the second device 104b can be reversed with respect to some or all of the scrambled DNN operation 825. In this embodiment, the first device 104a and the second device 104b establish the scrambled DNN operation as described in operation 792 of Figure 7, in particular operations 722a, 723a, 723b, 724a, and 724b. The signal flow of the scrambled DNN operation 825 from the first device 104a to the second device 104b includes the following signal flow operation.

[0155] In operation 802, the first device 104a obtains input communication data from a data source for input to the transmission DNN structure of the first device 104a.

[0156] In operation 804, the transmit DNN structure of the first device 104a processes the input communication data to generate an output communication signal to be transmitted to the second device 104b in a specific time slot.

[0157] In operation 806, before transmitting the generated output communication signal in a specific time slot, the first device 104a performs a spectral analysis to determine whether the predicted transmission of the output communication signal satisfies the white noise interference level. For example, Figures 1, 2a, 2b-2d and / or 4a-4d and / or 5 illustrate the execution of the spectral analysis. If the predicted transmission of the generated output communication signal does not satisfy the white noise interference level, operations 808 to 834 are performed; however, if the white noise interference level is satisfied, the scrambled DNN operation 825 proceeds to operation 816 for transmitting the output communication signal.

[0158] In operation 808, the first device 104a selects an updated NNSI to reconfigure the transmit DNN structure to produce an output communication signal that satisfies the white noise interference level during transmission. As illustrated with reference to Figures 1 to 6d, the updated NNSI may be determined from an iterative and / or NNSI lookup table. While the updated NNSI is selected, the transmit DNN structure reprocesses the input communication data to be reconfigured and produce an output communication signal. After determining / selecting the updated NNSI that reconfigures the transmit DNN structure to produce an output communication signal that satisfies the white noise interference level, the signal flow of scrambled DNN operation 825 proceeds to operation 810.

[0159] In operation 810, the first device 104a reconstructs the transmit DNN structure using the updated NNSI for a specific time slot.

[0160] In operation 814, the first device 104a transmits the updated NNSI and specific time slot information to the second device 104b. For example, the first device 104a transmits a control message to the second device 104b via a control channel, the control message including an indication of the updated NNSI and specific time slot information, which instructs the second device 104b to reconfigure its receiving DNN structure to process the transmission from the first device 104a corresponding to the specific time slot information. Upon receiving the control message from the first device 104a, the second device 104b performs operation 834, storing the updated NNSI and specific time slot information in an NNSI buffer or NNSI table and using it at a suitable time related to the specific time slot information.

[0161] In operation 816, the first device 104a transmits the output communication signal of the reconfigured transmit DNN structure according to specific time slot information when the generated output communication signal satisfies the white noise interference level.

[0162] When the second device 104b receives a transmission of an output communication signal in a particular time slot, it may buffer the received output communication signal until it is ready for processing by the receiving DNN structure. In operation 848, the second device 104b reconstructs the receiving DNN structure using the updated NNSI for the particular time slot before processing the received output communication signal corresponding to that time slot. The second device 104b retrieves the updated NNSI for the particular time slot from storage (e.g., from an NNSI buffer or NNSI table) and uses that updated NNSI to reconstruct the receiving DNN structure, for example, as described with reference to Figures 1 to 6d, particularly Figures 3a to 3c, 4c and 4d and 5, and / or as described herein. Reconstructing the receiving DNN structure modifies the receiving DNN structure to perform a reverse operation of the operation performed by the reconstructed transmitting DNN structure, i.e., a descramble operation, in order to generate reconstructed communication data corresponding to the input communication data transmitted in a particular time slot.

[0163] In operation 850, the receiving DNN structure processes the received output communication signal for a specific time slot to generate reconstructed communication data corresponding to the input communication data transmitted in that time slot. The second device 104b transmits the reconstructed communication data to its data sink or to one or more higher protocol layers of the second device 104b's protocol stack.

[0164] In operation 852, if the generation of the reconstructed communication data is successful, the second device 104b sends an acknowledgment to the first device 104a indicating successful reception of the input communication data for a specific time slot. In operation 853, the second device 104b continues to prepare to receive the next transmission from the first device 104a in one or more subsequent time slots.

[0165] In operation 825a, if the first device 104a receives an acknowledgment in operation 852, it proceeds to transmit further input communication data in one or more subsequent time slots if there is further input communication data from the data source, and the scrambled DNN operation 825 is repeated until it is decided to disable scrambled DNN communication as described in operation 726 of Figure 7.

[0166] Figure 9 shows a signal flow diagram of another embodiment of scrambled DNN operation 925 between the first device 104a and the second device 104b during operation 725 of the scrambled DNN communication 720 shown in Figure 7. Scrambled DNN operation 925 further modifies the operation of scrambled DNN operation 825 in Figure 8 by inserting operation 917 in which the first device 104a identifies further NNSIs that satisfy the white noise interference level during transmission. The signal flow of scrambled DNN operation 925 from the first device 104a to the second device 104b includes the following signal flow operations.

[0167] Operations 902–916 in the first device substantially correspond to operations 802–816 in Figure 8. Similarly, operations 934, 948, 950, 952, and 953 in the second device substantially correspond to operations 834, 848, 850, 852, and 853 in Figure 8. In the first device 104a, after operation 916, the first device 104a performs operation 917, and the first device 104a continues to identify further NNSIs to help reconfigure the transmit DNN structure to produce an output communication signal that is expected to satisfy the white noise interference level at transmit. The first device 104a stores any identified NNSIs mapped to NNSI identifiers and associated or estimated white noise interference levels in an NNSI lookup table for use in operation 908 (or operation 808 in Figure 8). This allows the search for the updated NNSI in operation 908 to be bootstrapped, enabling faster selection of the updated NNSI, which results in the generation of an output communication signal that satisfies the white noise interference level during transmission. In the embodiment, the first device 104a performs operation 917 in the background and / or between operations 916 and 925a.

[0168] In operation 925a, if the first device 104a receives an acknowledgment in operation 952, it proceeds to transmit further input communication data in one or more subsequent time slots if there is further input communication data from the data source, and the scrambled DNN operation 925 is repeated until it is decided to disable scrambled DNN communication as described in operation 726 of Figure 7.

[0169] Optionally, the first device 104a transmits updates regarding any identified NNSI, NNSI identifier, and / or the white noise interference level associated with or estimated to the NNSI to the second device 104b via one or more control messages, which the second device 104b uses to update its NNSI lookup table, including the NNSI and NNSI identifier. This provides the advantage that, for an NNSI selected by the first device 104a, the control message for reconstructing the received DNN structure of the second device 104b can send the NNSI identifier and timing information of the selected NNSI, allowing the second device 104b to retrieve the selected NNSI as appropriate.

[0170] Figure 10 shows the signal flow of another embodiment of scrambled DNN communication 1000 between the first device 104a, the second device 104b, and the third device 1003. Scrambled DNN communication 1000 further modifies scrambled DNN communication 720 in Figure 7 by inserting operations 1005a and 1005b in which the third device 1003 communicates with the first device 104a with respect to an acceptable level of white noise interference. The signal flow of scrambled DNN communication 1000 from the first device 104a to the second device 104b includes the following signal flow operations.

[0171] Operations 1021a, 1021b, 1092, 1026, 1027a, 1027b, 1028a, 1028b, and 1029 of the first device 104a and the second device 104b substantially correspond to operations 721a, 721b, 792, 726, 727a, 727b, 728a, 728b, and 729 described with reference to Figure 7. In operation 1021a, the white noise interference level is initially set during the establishment of the DNN connection by the first device 104a or in response to a request from the second device 104b seeking improved block error rate performance (e.g., an increase in the white noise level is required). When the first device 104a and the second device 104b establish a DNN communication connection and begin DNN communication with each other in operation 1021b, the third device 1003 may experience unacceptable white noise interference from transmissions from the first device 104a (or the second device 104b) during the DNN communication in operation 1021b.

[0172] In operation 1005a, the third device 1003 sends a notification to the first device 104a (or the second device 104b) indicating that the transmission of an output communication signal from the first device 104a to the second device 104b exceeds the acceptable white noise interference level associated with the third device 1003. In this embodiment, the third device 1003 is a sacrificial device that is being interfered with by the transmission from the first device 104a. In the embodiment, when the first device 104a is a base station, the third device 1003 communicates with the first device 104a via an uplink channel. In other embodiments, the third device 1003 is also a base station in another cell and requests the first device 104a to reduce interference occurring to other user equipment in that cell via the core network. In the embodiment, the notification from the third device 1003 indicates an acceptable or tolerable white noise interference level. In other embodiments, a notification from the third device 1003 indicates that the current white noise interference level output by the first device 104a is unacceptable.

[0173] In operation 1005b, the first device 104a adjusts the white noise interference level by reducing the white noise interference level when it receives a notification from the third device 1003 in operation 1005a. In one embodiment, if the notification from the third device 1003 indicates an acceptable or tolerable white noise interference level, the first device 104a adjusts the white noise interference level to an acceptable or tolerable level. In another embodiment, if the notification from the third device 1003 indicates that the current white noise interference level output by the first device 104a is unacceptable, the first device 104a adjusts the white noise interference level by gradually reducing the white noise interference level, or until the third device 1003 stops notifying to reduce the white noise interference level. For example, the third device 1003 (e.g., a sacrificial UE) will no longer experience an unacceptable white noise interference level once it moves away from the first device 104a. In other embodiments, when the third device 1003 is far from the first device 104a, the third device 1003 sends further notifications to the first device 104a indicating further tolerable white noise interference levels. These further white noise interference levels may be higher than the previous white noise interference levels as the distance from the third device 1003 to the first device 104a increases.

[0174] The first device 104a and the second device 104b perform operation 1092 at the adjusted white noise interference level. The remaining operations 1026, 1027a, 1027b, 1028a, 1028b, and 1029 in Figure 10 substantially correspond to operations 726, 727a, 727b, 728a, 728b, and 729 in Figure 7, respectively.

[0175] Figure 11 shows a signal flow diagram of another embodiment of scrambled DNN operation 1125 between the first device 104a and the second device 104b in operation 725 of scrambled DNN communication 720 or operation 1092 of scrambled DNN communication 1000 shown in Figure 7 or Figure 10. Scrambled DNN operation 1125 further modifies the operation of scrambled DNN operation 825 in Figure 8 or scrambled DNN operation 925 in Figure 9 by inserting operations 1105a and 1105b in which a third device 1103 notifies the first device 104a to adjust the white noise interference level to an acceptable level. The signal flow of scrambled DNN operation 1125 from the first device 104a to the second device 104b includes the following signal flow operations.

[0176] Operations 1102, 1104, 1106, 1108, 1110, 1114, 1116, and 1125a of the first device 104a substantially correspond to operations 802, 804, 808, 810, 814, 816, and 825a in Figure 8, or operations 902, 904, 908, 910, 914, 916, and 925a in Figure 9. Operations 1134, 1148, 1150, 1152, and 1153 of the second device 104b substantially correspond to operations 834, 848, 850, 852, and 853 described with reference to Figure 8, or operations 934, 948, 950, 952, and 953 described with reference to Figure 9. When the first device 104a and the second device 104b begin performing scrambled DNN operation 1125, the first device 104a and the second device 104b perform operations 1102, 1104, 1106, 1108, 1110, 1114, and 1116, and operations 1134, 1148, 1150, 1152, and 1153, respectively, in one or more time slots. During these one or more time slots, despite it being scrambled DNN communication between the first device 104a and the second device 104b, where the transmitting DNN structure of the first device 104a generates an output communication signal that satisfies the current white noise interference level at the time of transmission, the third device 1103 experiences an unacceptable level of white noise interference. This means that the current white noise interference level set by the first device 104a is too high, and the transmission from the first device 104a still results in an unacceptable or unforgivable level of interference to the third device 1103.

[0177] As described in operation 1005a of Figure 10, in operation 1105a, the third device 1103 sends a notification to the first device 104a (or the second device 104b) indicating that the transmission of an output communication signal from the first device 104a to the second device 104b exceeds the acceptable white noise interference level associated with the third device 1103. As described in operation 1005b of Figure 10, in operation 1105b, upon receiving the notification from the third device 1103 in operation 1105a, the first device 104a adjusts the white noise interference level by reducing it to an acceptable quantity if the third device 1103 indicates an acceptable quantity, or by reducing it by a stepwise quantity. In the latter case, the third device 1103 sends a further notification indicating that the adjusted white noise interference level is still unacceptable or unacceptable, and the first device 104a adjusts in further stepped quantities until the third device 1103 stops sending further notifications or sends a notification indicating that the white noise interference level is acceptable. Further embodiments described with respect to operations 1005a and 1005b are also applicable to operations 1105a and 1105b in Figure 11.

[0178] In operation 1125a, if the first device 104a receives an acknowledgment from the second device 104b in operation 1152, it proceeds to transmit further input communication data in one or more subsequent time slots if there is further input communication data from the data source, and the signal flow of scrambled DNN operation 1125 is repeated until it is decided to disable scrambled DNN communication as described in operation 726 in Figure 7 or operation 1026 in Figure 10.

[0179] Figure 12 shows the signal flow of an embodiment of scrambled DL / UL DNN communication 1220 during a DL / UL DNN communication session between BS210 and UE220. In this embodiment, the reference numbers in Figure 2a are used for the same or similar components. Scrambled DNN communications 720 and 1020 are further modified to include scrambled DL / UL DNN communication 1220 between BS210 and UE220 during a DL / UL DNN communication session between BS210 and UE220. BS210 and UE220 in Figure 2a perform scrambled DL / UL DNN communication 1220 using any of the embodiments described with reference to Figures 1 to 11. Specifically, operations 1221a, 1221b, 1292, 1226, 1227a, 1227b, 1228a, 1228b, and 1229 of BS210 and UE220 are further modified versions of the corresponding operations 721a, 721b, 792, 726, 727a, 727b, 728a, 728b, and 729 described with reference to Figure 7, and / or the corresponding operations 1021a, 1021b, 1092, 1026, 1027a, 1027b, 1028a, 1028b, and 1029 of Figure 10 for use in DL / UL DNN communication sessions between BS210 and UE220. The signal flow of scrambled DL / UL DNN communication 1220 for communication sessions between BS210 and UE220 includes the following signal flow operations.

[0180] In operation 1221a, the BS and UE perform the establishment of a Radio Resource Control (RRC) DNN connection to establish a DL / UL DNN communication session between them. The DL / UL DNN communication session includes DL DNN communication from BS210 to UE220 and UL DNN communication from UE220 to BS210. During the establishment of the DL / UL DNN communication session, BS210 and UE220 communicate with each other to define, agree, and / or configure the types of DL / UL transmit DNN structures and DL / UL receive DNN structures that they will each use for the DL DNN and UL DNN communications that provide end-to-end communication between them.

[0181] For example, when UE220 is in the RRC_IDLE state, UE220 sends an RRC DNN connection request to BS210 to establish a DL / UL DNN communication session. The RRC DNN connection request may include UE identification information and UE capabilities, which allow BS210 to select DL / UL transmit DNN structures and DL / UL receive DNN structures suitable for the DL communication channel and UL communication channel (e.g., physical downlink shared channel (PDSCH) and physical uplink shared channel (PUSCH)), and at the same time, define the control channel (e.g., physical downlink control channel (PDCCH) or physical uplink control channel (PUCCH)) to be used for DL ​​DNN communication and / or UL DNN communication between UE220 and BS210. BS210 has a set and / or pair of DL transmit DNN structures / DL receive DNN structures and a set and / or pair of UL transmit DNN structures / UL receive DNN structures, each of which is mapped to a DL DNN identifier / UL DNN identifier and stored in the DNN lookup table in BS210. UE220 has a corresponding set of UL / DL transmit DNN structures / receive DNN structures, which are also mapped to the same UL / DL DNN identifiers and stored in the DNN lookup table in UE220. BS210 selects the DL transmit DNN structures and DL receive DNN structures, and / or UL transmit DNN structures and UL receive DNN structures, based on the DL and / or UL communication channel / environment, the communication performance requirements of the DL and / or UL DNN connection, and the type of DL and / or UL data communication (e.g., voice communication, data communication, and multimedia streaming). BS210 sends an RRC DNN connection setup message containing data representing the selected DL / UL receive DNN structure for the DL communication channel and UL communication channel (e.g., PDSCH and PUSCH), DL control channel and UL control channel (e.g., PDCCH / PUCCH) used for DL ​​DNN communication and UL DNN communication between UE220 and BS210 (e.g., DL / UL DNN identifier).Upon receiving an RRC DNN connection setup message, UE220 may use DL DNN identifiers and UL DNN identifiers in its DNN lookup table to select and configure the corresponding DL receiving DNN structure and UL transmitting DNN structure. UE220 may send an RRC connection setup complete message to BS210 indicating that it has appropriately configured the corresponding DL receiving DNN structure and UL transmitting DNN structure and is ready for DL ​​DNN and UL DNN communication with BS210. BS210 also configures the corresponding DL transmitting DNN structure and UL receiving DNN structure for DL ​​DNN and UL DNN communication with UE220.

[0182] Alternatively, when UE220 is in the RRC_CONNECTED state, BS220 sends an RRC DNN connection reconfiguration message to UE220, which includes data representing DL / UL transmit DNN structures and DL / UL receive DNN structures suitable for DL ​​communication channels and UL communication channels (e.g., PDSCH and PUSCH) in order to change the existing RRC connection to an RRC DNN connection, and simultaneously defines the control channels (e.g., PDCCH / PUCCH) to be used for DL ​​and UL communication between UE220 and BS210. UE220 may send an RRC reconfiguration complete message to BS210 indicating that UE220 has appropriately configured the corresponding DL receive DNN structures and UL transmit DNN structures and is ready for DL ​​and UL DNN communication with BS210. BS210 also configures the corresponding DL transmit DNN structures and UL receive DNN structures for DL ​​and UL DNN communication with UE220.

[0183] In operation 1221a, during or after establishing the DL / UL DNN connection, BS210 sets the white noise interference level of the DL communication channel and / or UL communication channel to the default setting or to a specific white noise interference level depending on the performance requirements of the DL / UL DNN communication session. Alternatively or additionally, in the embodiment, UE220 requests a specific white noise interference level to improve the block error rate performance of the DL DNN communication and / or UL DNN communication (e.g., an increase in the white noise interference level is requested).

[0184] In operation 1221b, after BS210 and UE220 establish a DL / UL DNN communication session, BS210 and UE220 perform DL DNN communication in one or more time slots using the configured DL transmit DNN structure and DL receive DNN structure, respectively. UE220 and BS210 also perform UL DNN communication in one or more time slots using the configured UL transmit DNN structure and UL receive DNN structure, respectively.

[0185] In operation 1292, when BS210 detects that a transmission from BS210 in DL DNN communication does not meet a specific white noise interference level, or that a transmission from UE220 in UL DNN communication does not meet a specific white noise interference level, BS210 and UE220 perform DL DNN scrambling and / or UL DNN scrambling. In the case of DL DNN scrambling communication from BS210 to UE220, the TX DNN controller of BS210 and the RX DNN controller of UE220 control the scrambled DL DNN operation, as described with reference to Figures 1 to 11, particularly Figures 4a to 5. Similarly, in the case of UL DNN scrambling communication from UE220 to BS210, the TX DNN controller of UE220 and the RX DNN controller of BS210 control the scrambled UL DNN operation.

[0186] As described, the white noise interference level is set by BS210 in response to a request from UE220 regarding improved block error rate performance, or in response to a request from another UE or BS experiencing unacceptable white noise interference resulting from transmissions from BS210 during DL DNN communication or from UE220 during UL DNN communication (see, for example, Figure 10 or 11 for a third device 1003 or 1103). BS210 and UE220 perform DL / UL DNN operations 1292 based on, for example, the following scrambled DL / UL DNN operations.

[0187] In operation 1222a, if it is predicted that the output communication signal from the DL transmit DNN structure of the BS210 will not satisfy the white noise interference level of the DL communication channel (e.g., PDSCH) when transmitted over the DL communication channel, the BS210 enables the execution of scrambled DL DNN operation for one or more time slots. For example, the DL TX DNN controller of the BS210 detects that the output communication signal generated by the DL transmit DNN structure from the input communication data is predicted to be a transmit signal over a PDSCH that does not satisfy the white noise interference level. In such a case, DL scrambled DNN operation is enabled, and the DL TX DNN controller of the BS210 selects an NNSI, which is used to reconfigure the DL transmit DNN structure to generate an output communication signal from the same input communication data that will be a transmit signal over a PDSCH that satisfies the white noise interference level, as described with reference to Figures 1, 2, and 3a-6d, particularly Figures 3a-6d. Similarly, UL scrambled DNN operation is enabled when UE220 or BS210 detects that the transmitted signal over the UL communication channel (e.g., PUSCH) does not meet the white noise interference level set for the UL communication channel. BS210 selects an NNSI to be used by UE220 to reconfigure the UL transmit DNN structure so that it can be used to generate an output communication signal from the UL transmit DNN structure that will be the transmitted signal over PUSCH that meets the white noise interference level.

[0188] In operation 1223a, BS210 communicates when and how the scrambled DL DNN operation and scrambled UL DNN operation are enabled by sending, for example, an RRC control message (e.g., RRC connection reconfiguration), a medium access control (MAC) message, or a downlink control information (DCI) message to UE220 via PDCCH, which includes fields that also include DL and / or UL timing information regarding the selected NNSI configuration and when the DL receive DNN structure should be reconfigured at UE220 and / or when the UL transmit DNN structure should be reconfigured at UE220. Initially, the selected NNSI includes initial seed information, the type of random substitution function, specific layers of the DL and / or UL transmit DNN structure to be reconfigured, random substitution iterations, and any other data that enables UE220 to reconfigure the DL receive DNN structure to generate reconstructed communication data and / or reconfigure the UL transmit DNN structure to generate output communication signals for transmission via PUSCH. Subsequent RRC / MAC / DCI messages following further DL or UL scrambling DNN operations may include an updated NNSI configuration, such as the number of random substitution iterations or one or more selected neural network layers reconfigured for DL ​​and / or UL scrambling.

[0189] In operation 1223b, UE220 receives the selected NNSI configuration for scrambled / descrambled DL DNN operation (e.g., descrambled DL RX DNN operation performed by UE220) and / or scrambled UL DNN operation (e.g., scrambled UL TX DNN operation performed by UE220), and sends an acknowledgment (e.g., ACK) for enabling the scrambled DL / UL DNN operation.

[0190] In operations 1224a and 1224b, BS210 and UE220 enable scrambled DL and / or UL DNN operations based on DL and / or UL timing information. For example, BS prepares to perform a scrambled DL TX DNN operation for DL ​​transmission to UE220, while UE220 prepares to perform a descrambled DL RX DNN operation to receive DL transmission from BS210 using DL timing information (e.g., one or more DL time slots). In another embodiment, UE220 prepares to perform a scrambled UL TX DNN operation for UL transmission from UE220 to BS210, while BS210 prepares to perform a descrambled UL RX DNN operation for UL transmission from UE220 using UL timing information (e.g., one or more UL time slots).

[0191] In operation 1225, BS210 and UE220 communicate with each other via PDSCH and / or PUSCH in a manner similar to that described with reference to Figures 1 to 11, particularly Figures 7 and 10, to perform scrambled DL / UL DNN operation.

[0192] In operation 1226, after determining that scrambled DL / UL DNN operation / communication is unnecessary, BS210 disables scrambled DL / UL DNN operation.

[0193] In operation 1227a, when BS210 disables or stops scrambled DL and / or UL DNN operation, BS210 notifies UE220 that scrambled DL and / or UL DNN operation is disabled / stopped by sending, for example, an RRC message, a media access control (MAC) message, or a DCI message to UE220.

[0194] In operation 1227b, UE220 sends an acknowledgment (e.g., ACK) regarding the receipt of the RRC / MAC / DCI message sent by BS210 in operation 1227a.

[0195] In operation 1228a, upon receiving an acknowledgment (ACK) from UE220 in operation 1227b, BS210 stops the scrambled DL / UL DNN operation and reverts the DL transmit DNN structure and / or UL receive DNN structure to their original configuration. Alternatively, BS210 retains the current DL transmit DNN structure and / or UL receive DNN structure (e.g., the latest reconfiguration), which still makes it more likely that any further DNN communication will meet the white noise interference level.

[0196] In operation 1228b, if UE220 receives an RRC / DCI message in operation 1227a regarding the disabling of scrambled DL / UL DNN operation, it stops scrambled DL and / or UL DNN operation and restores the received DL DNN structure and / or UL transmitted DNN structure to their original configuration. Alternatively, if BS210 also indicates that the current structure should be retained, UE220 retains the current DL received DNN structure and / or the current UL transmitted DNN structure (e.g., the latest reconfiguration), which still makes it more likely that any further DNN communication will meet the white noise interference level.

[0197] In operation 1229, after BS210 and UE220 disable the scrambled DNN connection, BS210 and UE220 continue DNN communication. When DNN communication is not needed, BS210 and UE220 return to a state where they communicate with each other using, for example, standard or conventional 3G-4G, 5G, or 6G communication sessions.

[0198] Figure 13 shows the signal flow of an embodiment of scrambled DL DNN operation 1325 between BS210 and UE220 during operation 1225 of scrambled DL / UL DNN communication 1220 in Figure 12. In this embodiment, the reference numbers in Figure 2a are used for the same or similar components. The exemplary scrambled DNN operation 825 is further modified to include scrambled DL DNN operation between BS210 and UE220 during a DL DNN communication session between BS210 and UE220. BS210 and UE220 in Figure 2a perform scrambled DL DNN operation 1325 using any of the embodiments described with reference to Figures 1 to 12. Specifically, operations 1302 to 1316 of BS210 substantially correspond to operations 802 to 816 described with reference to Figure 8, but are modified for scrambled DL DNN communication. Similarly, UE operations 1334, 1348, 1350, 1352, and 1353 substantially correspond to operations 834, 848, 850, 852, and 853 described with reference to Figure 8, but are modified for scrambled DL DNN communication. In this embodiment, DNN operation 1292 in Figure 12 configures BS210 and UE220 to perform scrambled DL DNN operation 1325. The signal flow of scrambled DL DNN operation 1325 from BS210 to UE220 includes the following signal flow operations.

[0199] Operations 1302-1306 are substantially equivalent to operations 802-804 in Figure 8, except that the DL transmit DNN structure of BS210 processes the input communication data to generate an output communication signal for transmission to UE220 via PDSCH in a specific time slot.

[0200] Operation 1306 further modifies Operation 806 in Figure 8 by having BS210 communicate any updated NNSI and specific time slot information to UE220, for example, using RRC / DCI messaging over PDCCH. When UE220 receives the RRC / DCI messaging from BS210, it performs Operation 1334, similar to Operation 834 in Figure 8, where an NNSI buffer or NNSI table accessible to UE220 stores the updated NNSI and specific time slot information, which is used to descramble transmissions received over PDSCH with respect to the specific time slot information.

[0201] In operation 1316, when the generated output communication signal satisfies the white noise interference level, BS210 transmits the output communication signal of the reconstructed DL transmit DNN structure via the PDSCH according to specific time slot information. When UE220 receives the transmission of the output communication signal via the PDSCH in a specific time slot, it may buffer the received output communication signal until it is ready for processing by the DL receive DNN structure.

[0202] Operations 1348 and 1350 are substantially equivalent to operations 848 and 850, except that the DL receive DNN structure is reconstructed using the updated NNSI for a particular time slot, and the DL receive DNN structure processes the received output communication signal for a particular time slot to generate reconstructed communication data corresponding to the input communication data transmitted for a particular time slot. The UE220 transmits the reconstructed communication data to the UE220's data sink or to one or more higher protocol layers of the UE220's protocol stack.

[0203] Operations 1352 and 1325a substantially correspond to operations 852 and 825a in Figure 8. Scrambled DL DNN operation 1325 is repeated until it is decided to disable scrambled DL / UL DNN communication, as described in operation 1226 in Figure 12.

[0204] Figure 14 shows the signal flow of an embodiment of scrambled UL DNN communication 1420 between UE220 and BS210. In this embodiment, the reference numbers in Figure 2a are used for the same or similar components. In this embodiment, operations 1421a / b, 1422a, 1423a, 1423b, 1424a, and 1424b correspond to operations 1221a / b, 1222a, 1223a, 1223b, 1224a, and 1224b, except that UE220 performs scrambled UL DNN communication via PUSCH using a UL transmit DNN structure (e.g., UL TX DNN) and BS210 performs scrambled UL DNN communication via PUSCH using a UL receive DNN structure (e.g., UL RX DNN).

[0205] When UE220 has input communication data for transmission, UE220 uses a UL transmit DNN structure to generate a UL output communication signal for transmission via PUSCH. Operation 1406 is substantially the same as operation 806 in Figure 8 or operation 1306 in Figure 13, except that UE220 instead detects whether the predicted transmission of the UL output communication signal does not meet the white noise interference level of PUSCH. When UE220 detects that the white noise interference level is not met, it performs operations 1408, 1410, and 1414 based on the following:

[0206] Operation 1408 is substantially equivalent to Operation 808 or 1308, except that UE220 selects an updated NNSI to reconfigure the UL transmit DNN structure so as to produce a UL output communication signal that satisfies the white noise interference level when transmitted via PUSCH in a specific time slot. This can be selected iteratively, as described with reference to Operation 808 in Figures 1-6d and / or Figure 8 or Operation 1308 in Figure 13. After UE220 has selected the updated NNSI, UE220 proceeds to Operation 1410.

[0207] Operations 1410 and 1414 substantially correspond to operations 810 and 814, or operations 1310 and 1314, except that UE220 reconstructs the UL transmit DNN structure using the updated NNSI for a specific time slot, and UE220 communicates the updated NNSI to BS210 using uplink control signaling via PUCCH. BS210 already possesses specific time slot information, which is identified by BS210 in RRC / DCI messaging to UE220. Upon receiving the uplink control signaling, BS210 performs operation 1434, which substantially corresponds to operations 834 or 1334, except that BS210 stores the updated NNSI, along with the corresponding specific time slot, in BS210's NNSI buffer or NNSI table in order to descramble the transmission received via PUSCH with respect to the specific time slot information using the UL receive DNN structure.

[0208] In operation 1416, when the generated UL output communication signal satisfies the white noise interference level, the UE220 transmits the reconstructed UL transmit DNN structure via PUSCH according to a specific time slot.

[0209] Operations 1448 and 1450 are substantially equivalent to operations 848 and 850 in Figure 8, or operations 1348 and 1350 in Figure 13, except that BS210 performs these operations when it receives a transmission of a UL output communication signal via PUSCH in a particular time slot, and buffers the received UL output communication signal for a particular time slot before processing it using a UL receive DNN structure reconstructed with the updated NNSI for that time slot. The UL receive DNN structure processes the received UL output communication signal for a particular time slot to generate reconstructed communication data corresponding to the input communication data transmitted by UE220 via PUSCH in that time slot. BS210 transmits the reconstructed communication data to its data sink or to one or more higher protocol layers of BS210's protocol stack.

[0210] Operations 1452 and 1453 are substantially equivalent to operations 852 and 853 in Figure 8, or operations 1352 and 1353 in Figure 13, except that they are performed by BS210.

[0211] In operation 1420a, if UL220 receives an acknowledgment in operation 1452, and there is further input communication data from the UE220 data source, it proceeds to the next transmission via PUSCH of further input communication data in one or more subsequent time slots, and the scrambled UL DNN communication 1420 is repeated until it is decided to disable the scrambled UL DNN communication as described in operation 1226 in Figure 12.

[0212] Figure 15 shows the signal flow of another embodiment of scrambled UL DNN communication 1520 between UE220 and BS210. In this embodiment, the reference numbers in Figure 2a are used for the same or similar components. In this embodiment, operations 1521a / b, 1522a, 1523a, 1523b, 1524a, and 1524b correspond to operations 1421a / b, 1422a, 1423a, 1423b, 1424a, and 1424b. When UE220 has input communication data for transmission, UE220 uses the UL transmit DNN structure to generate a UL output communication signal for transmission via PUSCH. Operation 1506 further modifies operation 1406 in Figure 14 based on the following changes.

[0213] In operation 1507, after detecting that the UL output communication signal does not meet the white interference noise level when transmitted, UE220 uses uplink control signaling via PUCCH to communicate an updated NNSI request to BS210.

[0214] Operation 1508 is substantially equivalent to Operation 808 in Figure 8 or Operation 1308 in Figure 13, respectively, except that BS210 selects an updated NNSI to reconfigure the UE220's UL transmit DNN structure so as to produce a UL output communication signal that satisfies the white noise interference level when transmitted via PUSCH in a specific time slot. As described with reference to Figures 1-6d and / or Operation 808 in Figure 8, Operation 1308 in Figure 13, and Operation 1408 in Figure 14, this may be selected iteratively by BS210. In embodiments, when BS210 selects an NNSI for UE220, as described with reference to Figure 2a, BS210 simulates UL with random UE UL input communication data and UE UL transmit DNN structure to generate an output communication signal and selects an NNSI that produces an output communication signal that satisfies the UL's white noise interference level.

[0215] Operation 1514 is substantially equivalent to operations 814 and 1314, except that BS210 communicates updated NNSI and specific time slot information for UL to UE220, for example, using RRC / DCI messaging over PDCCH. In operation 1534a, BS210 stores the updated NNSI and specific time slot information in its NNSI buffer or NNSI table, which are used to descramble transmissions received via PUSCH with respect to the specific time slot information. Also, when UE220 receives RRC / DCI messaging from BS210, it performs operation 1534b, storing the updated NNSI and specific time slot information in its NNSI buffer or NNSI table, which are used to scramble input communication data for transmission via PUSCH according to the specific time slot information. In operation 1510, UE220 uses the updated NNSI for the specific time slot to reconstruct the UL transmit DNN structure of UE220.

[0216] Operation 1516 is substantially equivalent to Operation 816 in Figure 8 and Operation 1316 in Figure 13, respectively, except that the UE220 transmits UL output communication signals via PUSCH as appropriate according to a specific time slot.

[0217] Operations 1516, 1548, 1550, 1552, 1553, and 1520a correspond to operations 1416, 1448, 1459, 1452, 1453, and 1420a in Figure 14. If there is further input communication data from the UE220's data source, the UE220 proceeds to the next transmission of further input communication data via PUSCH in one or more subsequent time slots, and the scrambled UL DNN communication 1520 is repeated until it is decided to disable the scrambled UL DNN communication as described in operation 1226 in Figure 12.

[0218] Figure 16 shows the signal flow of a scrambled DL / UL DNN communication session 1600 between BS210 and UE220, as described with reference to Figures 2a and 12-15. In this embodiment, the reference numbers in Figure 2a are used for the same or similar components. BS210 includes a BS DNN controller 214 (BS DNNC), a BS DL transmit DNN structure (BS DL TX DNN) for processing and scrambling input communication data to be transmitted downlink to UE220 via PDSCH, and a BS UL receive DNN structure (BS UL RX DNN) for processing uplink transmissions received from UE220 via PUSCH. The UE220 includes a UE DNN controller 224 (UE DNNC), a UE DL receive DNN structure (UE DL RX DNN) for processing downlink transmissions received from BS210 via the PDSCH, and a UE UL transmit DNN structure (UE UL TX DNN) for uplink transmissions to BS210 via the PUSCH. It is also assumed that BS210 has already allocated suitable frequency / time slots to the UE220 for control plane signaling via the corresponding PDCCH and PUCCH.

[0219] In operation 1621a, BS DNNC214 establishes a UL / DL DNN communication session between BS210 and UE220. BS DNNC214 selects pairs of DL transmit DNN structures and DL receive DNN structures (BS DL TX DNN and UE DL RX DNN) to be used in the scrambled DL / UL DNN communication session 1600. In the scrambled DL / UL DNN communication session 1600, BS210 sends an RRC establishment request message (e.g., RRC DNN Establishment Request (DL DNN Type / Id)) containing the DL DNN type or identifier associated with the selected DL TX / RX DNN pair. BS DNNC214 retrieves the TX DNN configuration data for the BS DL TX DNN206 corresponding to the selected DL DNN type or identifier from the BS TX / RX DNN store or table 215b. In operation 1621b, BS DNNC214 sends a configuration instruction (e.g., Cfg(DLDNN Type)) to configure the BS DL TX DNN structure 206. In operation 1621c, when UE DNNC224 receives an RRC establishment request message containing the DL DNN type or identifier associated with the selected DL TX / RX DNN pair, UE DNNC224 retrieves the RX DNN configuration data for the UE DL RX DNN228 corresponding to the received DL DNN type or identifier from the UE TX / RX DNN storage / table 225b of UE220, and sends a configuration instruction (e.g., Cfg(DL DNN Type)) to the UE DL RX DNN structure 228 to configure the UE DL RX DNN228 based on the retrieved RX DNN configuration data. In operation 1621d, after the UE DL RX DNN228 is configured, the UE DNNC224 on the UE220 sends an RRC response (e.g., RRC DNN Establishment Resp(ACK)) to the BS210, indicating an acknowledgment of configuring the UE DL RX DNN228.

[0220] Upon receiving an acknowledgment, in operation 1602i, input communication data (e.g., I_Data Xi) for transmission at a specific time slot TS i is applied to the BS DL TX DNN206, which generates a DL output communication (OC) signal (e.g., OC_Data Xi) corresponding to I_Data Xi. In operation 1604i, DL OC_Data Xi is provided to the BS DNNC214 to be transmitted to TS i via the PDSCH as a transmit waveform signal to the UE220. As illustrated with reference to Figure 5, upon receiving OC_Data Xi, the BS DNNC214 detects whether the transmission of OC_Data Xi satisfies the white interference noise level. If it does, OC_Data Xi is, for example, buffered and transmitted to the specific TS i via the PDSCH as a transmit waveform signal to the UE220. If not, an NNSI is generated / selected to reconfigure the DL TX DNN206 to produce OC_Data Xi that satisfies the white interference noise level, as described with reference to Figure 5 and / or Figures 1-4d and / or Figures 6a-15. In operation 1616i, the BS DNNC214 transmits OC_Data Xi to the UE220 as a transmit waveform signal via the PDSCH to TS i (e.g., PDSCH RF TX WAVEFORM(OC_DataXi,TS i)). The UE220 receives the transmit signal waveform of TS i via the PDSCH, processes it and converts it to a received OC_Data Xi (e.g., Rx OC Xi) (e.g., downconverts it), and in operation 1650i-1, the Rx OC Xi of TS i is input to the UE DL RX DNN228 for processing. In operation 1650i-2, the UE DL RX DNN228 processes the Rx OC Xi to generate the reconstructed communication data for TS i corresponding to I_Data Xi (e.g., R_IData Xi). In operation 1652i, after the UE DL RX DNN228 has successfully generated the R_IData Xi for TS i, the UE DNNC224 uses uplink control plane signaling to send an acknowledgment (e.g., ACK) to BS210.Upon receiving an ACK from UE220, operations 1602i to 1652i are repeated for further input communication data to be transmitted from BS210 to UE220 in the following time slot.

[0221] In operation 1622, if BS DNNC214 detects that the DL output communication signal does not meet the white interference noise level during transmission, as described with reference to Figures 1 to 15, scrambled DL DNN communication is enabled, and BS210 selects an NNSI to reconfigure BS DL TX DNN206 to produce a DL output communication signal that meets the white interference noise level, as described with reference to Figures 1 to 15. The selected NNSI includes the specific neural network layer(s) to be scrambled (e.g., NN layer #), initial seed (e.g., Seed), identifier of the substitution random function (PRF), and number of random substitution iterations (e.g., P, RPermIt #) (e.g., NN layer #, Seed, PRF ID, RPermIt #). In operation 1622a / 1623a, BS DNNC214 enables scramble DL DNN communication and sends the selected NNSI to UE220 using RRC control plane signaling via PDCCH (e.g., RRC Scramble DNN Enabled Req(NN layer #,Seed,PRF ID,RPermIt #)). In operation 1624b, when UE DNNC224 receives the RRC Scramble DNN Enabled Req message containing the selected NNSI associated with BS DL TX DNN206, UE DNNC224 generates configuration data for reconfiguring UE DL RX DNN228 based on the NNSI, as described with reference to Figures 3a-3c and / or 4a-4b, and sends a configuration command (e.g., Cfg(Scramble)) for scramble configuration information to reconfigure UE DL RX DNN228 on UE220.

[0222] In operation 1623b, after the UE DL RX DNN228 is reconfigured, the UE DNNC224 on the UE220 uses uplink control plane signaling to send an RRC response message (e.g., RRC Scramble DNN Enabled Resp(ACK)) to the BS210 via PUSCH, indicating acknowledgment of the reconfiguration. In operation 1624c, upon receiving the acknowledgment, the BS DNNC214 sends a configuration command (e.g., Cfg(Scramble)) along with scramble configuration information based on the selected NNSI for reconfiguring the BS DL TX DNN structure 206.

[0223] In operation 1602j, the input communication data for time slot j (e.g., I_Data Yj) is applied to the reconstructed BS DL TX DNN206, which generates a DL output communication signal (e.g., OC_Data Yj) corresponding to I_Data Yj of TS j. In operation 1604j, the BS DL TX DNN206 outputs OC_Data Yj to the BS DNNC214 for transmission to TS j via PDSCH as a transmit waveform signal to the UE220. As described with reference to Figure 5, when the BS DNNC214 receives OC_Data Yj, it detects whether OC_Data Yj satisfies the white interference noise level at the time of transmission. If it does, OC_Data Yj is, for example, buffered and transmitted to TS j via PDSCH as a transmit waveform signal. In operation 1616i, the BS DNNC214 transmits OC_Data Yj to TS i via the PDSCH as a transmit waveform signal (e.g., PDSCH RF TX WAVEFORM(OC_DataYj,TS j)). The UE220 receives the transmit signal waveform of TS j via the PDSCH, processes it, and converts it to the received OC_Data Yj (e.g., Rx OC Yj) (e.g., downconverts to baseband). In operation 1650j-1, the UE DNNC224 applies or inputs the Rx OC Yj of TS j to the reconfigured UE RX DL DNN228 for DNN processing. In operation 1650j-2, the UE RX DL DNN228 processes the Rx OC Yj to generate the reconstructed communication data of TS j corresponding to I_Data Yj (e.g., R_IData Yj). In operation 1652j, after UE DL RX DNN228 successfully generates R_IData Yj, UE DNNC224 sends an acknowledgment (e.g., ACK) to BS210 using uplink control plane signaling via PUCCH.When an ACK is received from UE220, operations 1602j to 1652j are repeated until it is detected that the transmission of the DL output communication signal generated by BS DL TX DNN206 in the subsequent time slot does not meet the white noise interference level for further input communication data to be transmitted from BS210 to UE220 in the subsequent time slot.

[0224] In operation 1606a, BS DNNC214 detects that the current DL output communication signal of the current TS does not meet the white interference noise level when transmitted via PDSCH, as described with reference to Figures 1 to 15. BS210 selects an updated NNSI to reconfigure BS TX DL DNN206 to generate DL output communication signals for TS a and any subsequent time slots that are expected to meet the white interference noise level, as described with reference to Figures 1 to 15. The selected updated NNSI includes data (e.g., RPermIt #) representing at least the number of random substitution iterations, which is associated with generating a random substitution sequence to substitute the order of the current particular neural network layer(s) as defined in operation 1623a. In operation 1614a, BS DNNC214 sends an RRC / MAC control message (e.g., RRC / MAC Control message(RPermIt #,TS{a,b,c,d})) to UE220 using RRC control plane signaling via PDCCH, containing the selected update NNSI and corresponding specific timing information (e.g., TS a, TS b, TS c, and TS d). The specific timing information (e.g., TS{a,b,c,d}) describes when UE DNNC224 should apply the update NNSI to UE DL RX DNN228. In operation 1614b, when UE DNNC224 receives the RRC / MAC control message containing the selected update NNSI and specific timing information, UE DNNC224 stores the update NNSI and corresponding specific timing information (e.g., TS{a,b,c,d}) in UE NNSI storage / lookup table / buffer 225a. In operation 1614b, the UE DNNC224 uses uplink control plane signaling via PUCCH to send an acknowledgment (e.g., ACK) to the BS210 regarding the reception of the updated NNSI and specific timing information.When BS DNNC214 receives an acknowledgment in operation 1614b, in operation 1610a it sends a configuration instruction (e.g., Cfg(RPermIt#)) containing scrambled configuration information associated with a selected number of random substitution iterations of the selected updated NNSI to reconfigure BS DL TX DNN206.

[0225] In operation 1602a, BS210 inputs or applies the input communication data (e.g., I_Data Za) for time slot a to the reconfigured BS DL TX DNN206, which generates the DL output communication signal (e.g., OC_Data Za) for TS a corresponding to I_Data Za. In operation 1604a, OC_Data Za is provided to BS DNNC214 to be transmitted to UE220 as a transmit waveform signal via PDSCH for a specific time slot TS a. As illustrated with reference to Figure 5, upon receiving OC_Data Za, BS DNNC214 detects whether the transmission of OC_Data Za satisfies the white interference noise level. If it does, OC_Data Za is, for example, buffered and transmitted as a transmit waveform signal for a specific time slot TS a. If the transmission of OC_Data Za does not satisfy the white interference noise level, operation 1606a is performed again.

[0226] In operation 1616a, BS DNNC214 transmits OC_Data Za to TS a via PDSCH as a transmit waveform signal (e.g., PDSCH RF TX WAVEFORM(OC_DataZa,TS a)). UE220 receives the transmit signal waveform from TS a via PDSCH and downconverts it to the received OC_Data Za (e.g., Rx OC Za). As illustrated with reference to Figure 5, UE DNNC224 may buffer the Rx OC Za of TS a until UE DL RX DNN228 is ready to process TS a. In operation 1648a, the UE DNNC224 obtains the NNSI associated with TS a before processing TS a, as described with reference to Figures 3a-3c and / or 4c-4d, generates configuration data for reconfiguring the UE DL RX DNN228 based on the obtained NNSI, and sends a configuration instruction (e.g., Cfg(RPermit#)) with a scrambled configuration for reconfiguring the UE DL RX DNN228. In operation 1650a-1, the UE DNNC224 applies the Rx OC Za of TS a to the reconfigured UE DL RX DNN228 for processing. In operation 1650a-2, the UE DL RX DNN228 processes the Rx OC Za to generate the reconfigured communication data for TS a (e.g., R_IData Za) corresponding to the I_Data Za. In operation 1652a, after UE DL RX DNN228 successfully generates R_IData Za, UE DNNC224 sends an acknowledgment (e.g., ACK) to BS210 using uplink control plane signaling via PUCCH. Upon receiving the ACK from UE220, operations 1602a–1652a are repeated, with BS210 sending subsequent input communication data (e.g., I_Data Zb, I_Data Zc, I_Data Zd) to UE220 for at least TS b, c, and d, as well as other subsequent time slots. As illustrated, operations 1602a–1652a are also repeated for TS d, and BS210 sends the input communication data (e.g., I_Data Zd) for TS d, as described in operations 1602d–1652d.

[0227] During DL DNN communication between BS210 and UE220, UE220 and BS210 also perform UL DNN communication. UE220 uses the UL transmit DNN structure 226 (e.g., UE UL TX DNN) to transmit to BS via PUSCH, and BS210 uses the UL receive DNN structure 208 (e.g., BS UL RX DNN) to receive and process the transmission. It is assumed that the UE UL TX DNN 226 and BS UL RX DNN 208 are already configured, as described with reference to Figures 12-15. In operation 1606b, UE DNNC 224 detects that the current UL output communication signal does not meet the white interference noise level during transmission, as described with reference to Figures 1-15. UE220 reconfigures its UE UL TX DNN 226. In this embodiment, UE220 does not have the ability to select an updated NNSI; instead, in operation 1607, UE DNNC224 requests a communication resource (e.g., UL time slot / frequency) for an uplink scrambling transmission containing the updated NNSI using uplink control plane signaling via PUCCH (e.g., PUCCH Request(UL NNSI for Scrambling transmission)). Upon receiving the request for the communication resource and the updated NNSI, BS210 selects the updated NNSI to reconfigure UE TX UL DNN226 to produce a UL output communication signal that satisfies the white interference noise level of PUSCH, as described with reference to Figures 1–15 (e.g., see operation 1508 in Figures 2a and 15). The selected updated NNSI includes data (e.g., RPermIt #) representing at least the number of random substitution iterations, which is associated with generating a random substitution sequence to substitute the order of the current particular neural network layer(s) of UE UL TX DNN226.In operation 1614e, BS DNNC214 uses DCI control plane signaling via PDCCH to send the selected update NNSI and corresponding specific transmission timing information (e.g., TS e) to UE220 (e.g., DCI Control message(RPermlt #,TS e)), where TS e indicates the time slot to which the update NNSI should be applied. When UE DNNC224 receives the DCI message containing the selected update NNSI and specific timing information TS e, UE DNNC224 stores the update NNSI and corresponding specific timing information TS e in UE NNSI storage 225a. In operation 1614f, UE DNNC224 on UE220 uses uplink control plane signaling via PUCCH to send an acknowledgment (e.g., ACK) to BS210 regarding the receipt of the update NNSI and specific timing information.

[0228] In operation 1648b, to perform a UL transmission to TS e, the UE DNNC224 sends a configuration command (e.g., Cfg(RPermIt#)) containing scrambled configuration information based on the number of random substitution iterations of the selected updated NNSI of TS e to reconfigure the UE UL TX DNN226. In operation 1602e, the reconfigured UE UL TX DNN226 processes the input communication data of TS e (e.g., I_Data Ze), thereby generating the UL output communication signal of TS e (e.g., OC_Data Ze) corresponding to I_Data Ze. In operation 1604e, the UE DNNC224 of the UE220 processes OC_Data Ze for transmission to BS210 as the transmit waveform signal of TS e via PUSCH. As illustrated with reference to Figure 5, when UE DNNC224 receives OC_Data Ze, it detects whether the transmission of OC_Data Ze meets the white interference noise level. If it does, OC_Data Ze is, for example, buffered and transmitted to BS210 via PUSCH as the transmitted waveform signal of TS e. If the transmission of OC_Data Ze does not meet the white interference noise level, operation 1606b is repeated.

[0229] In operation 1616e, the UE DNNC224 transmits OC_DataZe to the BS210 via PUSCH as the transmit waveform signal of TS e (e.g., PUSCH RF TX WAVEFORM(OC_DataZe,TS e)). The BE210 receives the transmit waveform signal of TS e via PUSCH, processes it, and converts it to the received OC_DataZe (e.g., Rx OC Ze) (e.g., downconverting to baseband). As illustrated with reference to Figure 5, the BS DNNC214 may buffer the Rx OC Ze of TS e until the BS UL RX DNN208 is ready to process TS e. In operation 1648e, BS DNNC214 obtains the NNSI associated with TS e before processing TS e, as described with reference to Figures 3a-3c and / or 4c-4d, generates configuration data for reconfiguring BS UL RX DNN208 based on the obtained NNSI, and sends a configuration instruction (e.g., Cfg(RPermit#)) with a scrambled configuration for reconfiguring BS UL RX DNN208. In operation 1650e-1, BS DNNC214 applies the Rx OC Ze of TS e to the BS UL RX DNN208 reconfigured for DNN processing. In operation 1650e-2, BS DL RX DNN208 processes the Rx OC Ze to generate the reconfigured communication data (e.g., R_IData Ze) of TS e corresponding to I_Data Ze. In operation 1652e, after BS DL RX DNN208 successfully generates R_IData Ze, BS DNNC214 sends an acknowledgment (e.g., ACK) to UE220 using downlink control plane signaling via PDCCH. Upon receiving the ACK from BS210, the signal flow repeats operations 1602e-1652e for subsequent time slots, when UE220 has further input communication data to send to BS210.

[0230] In this embodiment, in operation 1627a, BS210 determines that scrambled DNN communication should be terminated (e.g., the UE requests termination of the communication session, the UE switches to the RRC_IDLE state, the UE switches to the RRC_INACTIVE state, there is a UE / BS connection failure, or the communication session returns to conventional communication use, etc.), and in that case, BS210 uses RRC control plane signaling to indicate that scrambled DNN communication is to be disabled (e.g., RRC Scramble DNN Disable Req()). In operation 1628a, BS DNNC214 sends a configuration message to the corresponding BS TX / RX DNN206 / 208 in order to return the corresponding BS TX / RX DNN206 / 208 to its original configuration for standard DNN communication, and / or to release the associated DNN computing resources used for DNN communication used for conventional communication, and / or to terminate the communication session. In operation 1628b, UE DNNC224 sends a configuration message to the corresponding UE TX / RX DNN226 / 228 in order to return it to its original configuration for standard DNN communication, and / or to release the associated DNN computing resources for DNN communication used in legacy communication, and / or to terminate the communication session.

[0231] Figure 17 shows non-temporary computer-readable media 1700 in several embodiments. The non-temporary computer-readable media 1700 may include a computer-readable storage medium 1702 and / or an input / output mechanism 1704 for enabling a computing system to access the computer-readable media 1702. In this embodiment, the non-temporary computer-readable media 1700 is a USB stick, but this is merely an example and is not limiting, and it will be understood by those skilled in the art that the non-temporary computer-readable media 1700 may be any other type of computer-readable media or computer program product, and / or any other computer-readable media depending on the application, such as a CD, DVD, USB stick, Blu-ray disc, flash drive, etc. The non-temporary computer-readable medium 1700 stores computer programs, computer program code, and / or instructions which, when executed by one or more processors of the device or system, cause one or more processors of the device or system to execute one or more methods, operations, and processes from any signal flows, flowcharts, methods, and / or processes described herein, as disclosed, for example, with respect to the signal flow diagrams, flowcharts, and schematic diagrams of Figures 1 to 16 and their associated features.

[0232] Embodiments of the methods or processes described herein may be implemented in digital electronic circuits, integrated circuits, specially designed ASICs (application-specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These also include computer program products having computer-readable instructions (e.g., software stored on magnetic disks, optical disks, memory, programmable logic devices, etc.), which, when executed by a processor, cause the processor to execute one or more of the methods and / or processes described herein.

[0233] Any system feature described herein may also be provided as a method feature or a process feature, and vice versa. Means-plus-function features may be expressed alternatively in accordance with their corresponding structures, as used herein. In particular, method embodiments may be applied to system embodiments, and vice versa.

[0234] Furthermore, any, some, and / or all features in one embodiment can be applied to any, some, and / or all features in any other embodiment, in any preferred combination. It should also be understood that certain combinations of the various features described and defined in any embodiment of the present invention can be carried out and / or supplied and / or used independently.

[0235] While several embodiments have been shown and described, modifications to these embodiments may be made without departing from the principles of the present disclosure, and it will be understood by those skilled in the art that the scope of the present disclosure is defined by the claims and their equivalents.

Claims

1. A method performed by a first device (104a) communicating with a second device (104b), wherein the method is To generate an output communication signal (118) to be transmitted to the second device, the input communication data is processed by a transmitting deep neural network (DNN) (106) (404), The system includes (406) performing a scramble DNN operation in response to the predicted transmission of the generated output communication signal not meeting the white noise interference level, wherein the scramble DNN operation is (408) Selecting neural network scrambling information (NNSI) for reconstructing the transmit DNN so as to process the input communication data and generate a scrambled output communication signal that satisfies the white noise interference level at the time of transmission, Sending a control message to the second device (414) having an indication of the NNSI and scramble timing information to indicate when the second device should reconfigure the received DNN (109), The scrambled output communication signal that satisfies the white noise interference level is transmitted to the second device based on the scrambling timing information (416), Methods that include...

2. The NNSI has data representing the randomization of the input of the transmit DNN to be randomized, the output of the transmit DNN, and / or the order of the set of neural network nodes in one or more neural network layers of the transmit DNN, and performing the scrambled DNN operation further, Reconstructing the transmit DNN based on the NNSI by randomizing the order of the input, the output, or the set of neural network nodes in the one or more neural network layers (410) The method according to claim 1, including the method described in claim 1.

3. Performing the aforementioned scrambled DNN operation further means that The process includes reconstructing the transmit DNN of the first device by performing a random replacement of a set of neural network nodes in one or more neural network layers of the transmit DNN using one or more random replacement parameters, wherein the NNSI specifies data representing the one or more random replacement parameters. The method according to either claim 1 or 2.

4. The one or more random replacement parameters include one or more of the following, used to perform the random replacement on the set of neural network nodes: seed data, the number of iterations or sequence number of the random replacement, or identification of the seed generation and pseudo-randomization function, and performing the random replacement on the set of neural network nodes further, For each neural network layer, this includes generating a random permutation sequence corresponding to the number of iterations or sequence number of the random permutation, wherein the random permutation sequence has a length equal to the number of neural network nodes in the set of neural network nodes for each neural network layer, and performing the random permutation of the set of neural network nodes further includes, The set of neural network nodes for each neural network layer is randomized by applying the generated random substitution sequence to the set of neural network nodes. The method according to claim 3, including the method described in claim 3.

5. The seed data further includes one or more of the following: identification information for the first device, identification information for the second device, identification information for the cell in which the second device is located, identification information for the cell in which the first device is located, timing slot information, frame identification number, and / or any other information associated with the first device or the second device. The method further includes generating a seed from the seed data in order to perform the random replacement of the set of neural network nodes in the one or more neural network layers. The method according to claim 4.

6. Performing the aforementioned scrambled DNN operation further means that The NNSI includes reconstructing the transmit DNN based on performing a randomization operation on the neural network nodes of the selected Ith neural network layer of the transmit DNN, where 1 ≤ I ≤ L, and L is the number of neural network layers, wherein the NNSI specifies that the selected Ith neural network layer of the transmit DNN is randomized. The method according to any of the prior claims.

7. The selected I-th neural network layer includes a set of N (N>1) neural network nodes arranged in a specific order, and performing the randomization operation on the selected I-th neural network layer of the transmit DNN further includes randomizing the set of N neural network nodes in the selected I-th neural network layer, and the randomization is To randomize the set of N neural network nodes in the selected I-th neural network layer, an N-dimensional permutation matrix is ​​generated using a random permutation sequence of length N. Randomizing the selected I-th neural network layer by multiplying the specific order of the set of N neural network nodes in the selected I-th neural network layer by the N-dimensional permutation matrix, so as to form a randomized order of the set of N neural network nodes. The method according to claim 6, based on the present invention.

8. The selection of the aforementioned NNSI further means that This includes iteratively selecting an NNSI, and said iterative selection is In the i-th iteration (i > 0), generate the i-th randomization operation with respect to the input, output, or set of neural network nodes in the one or more neural network layers of the transmitting DNN (672), Based on the i-th randomization operation, the transmit DNN is reconstructed (674), To generate a scrambled output communication signal, the input communication data is processed by inputting it into the reconstructed transmit DNN (676), To determine whether the transmission of the scrambled output communication signal satisfies the white noise interference level, the scrambled output communication signal is analyzed (678), In response to the analysis showing that the transmission of the scrambled output communication signal of the reconstructed transmit DNN satisfies the white noise interference level, the NNSI including the i-th randomization operation used to reconstruct the transmit DNN is shown (682), In response to the analysis indicating that the transmission of the output communication signal of the reconstructed transmit DNN does not satisfy the white noise interference level, update to the next i-th iteration (680), and repeat the generating step (672), the reconstructing step (674), the processing step (676), and the analysis step (678), A method according to any of the prior claims, based on the above.

9. The method of claim 8, wherein the i-th randomization operation is a random permutation of the order of the set of neural network nodes, and for each i-th random permutation iteration, further comprising iteratively selecting an NNSI using the transmit DNN of the first device to process the input communication data.

10. The method according to claim 8 or 9, further comprising the first device iteratively selecting an NNSI based on a simulation of the transmission of the scrambled output communication signal over a simulated communication channel.

11. The method according to any prior claim, further comprising storing an arbitrary selected NNSI and a corresponding whitening characteristic in an NNSI lookup table of the first device, wherein the selection of the NNSI further comprises obtaining an NNSI from the NNSI lookup table on the basis that the selected NNSI indicates that the corresponding whitening characteristic results in the transmission of the output communication signal of the transmit DNN that satisfies the white interference noise level.

12. In order to define and configure the transmit DNN and receive DNN that perform end-to-end communication between them, a DNN connection is established with the second device, Using the configured transmit DNN and receive DNN of the first device, DNN communication is performed with the second device in one or more time slots. When the output communication signal from the transmitting DNN does not satisfy the white noise interference level during transmission, the execution of the scrambled DNN operation for one or more time slots is enabled according to the scrambled timing information. The method according to any prior claim, further comprising:

13. The method according to any of the prior claims, further comprising determining whether the estimated spectral density of the predicted transmission of the output communication signal satisfies the white noise interference level before performing the scrambled DNN operation.

14. The determination of whether the estimated spectral density satisfies the white noise interference level further includes: When the power spectral density of the predicted transmission across the target bandwidth is below the white noise interference level, or within a predetermined threshold region of the white noise interference level, it is identified that the estimated spectral density of the predicted transmission satisfies the white noise interference level. The method according to claim 13, including the method described in claim 13.

15. The determination of whether the estimated spectral density satisfies the white noise interference level further includes: When the estimated spectral density of the predicted transmission forms one or more interference spikes that exceed the white noise interference level, it is determined that the estimated spectral density of the predicted transmission does not satisfy the white noise interference level. The method according to claim 13 or 14, including the method described in claim 13 or 14.

16. The method according to claim 13 or 14, wherein the estimated spectral density of the predicted transmission is estimated for each antenna output of the first device.

17. Identifying whether the estimated spectral density of the predicted transmission satisfies the white noise interference level means that Performing spectral density estimation in the predictive transmission representing the output communication signal (406a), With respect to the white noise power spectral density corresponding to the white noise interference level, the estimated spectral density is analyzed (406b), The method according to any one of claims 13 to 16, including the method described above.

18. (406b) The estimated spectral density of the predicted transmission is compared with the white noise spectral density associated with the white noise interference level. In response to the estimated spectral density of the predicted transmission being less than or substantially equal to the white noise spectral density associated with the white noise interference level, the output communication signal indicates that it satisfies the white noise interference level during transmission (406d), In response to the estimated spectral density of the predicted transmission being greater than or substantially different from the white noise spectral density associated with the white noise interference level, the output communication signal indicates that it does not meet the white noise interference level during transmission (406c), The method according to claim 17, further comprising:

19. The aforementioned white noise interference level is, A constant amplitude of flat white noise power spectral density across the target bandwidth, The total power of the flat white noise power spectral density across the target bandwidth, A method according to any of the prior claims, defining any of the following.

20. Receiving a notification from the third device (220b) indicating that the transmission of an output communication signal to the second device exceeds the acceptable white noise interference level associated with the third device (1105a), Adjusting the white noise interference level to satisfy the acceptable white noise interference level (1105b), Performing the scrambled DNN operation, which includes analyzing whether the output communication signal satisfies the adjusted white noise interference level at the time of transmission, The method according to any prior claim, including

21. The method according to any prior claim, wherein the NNSI includes data specifying one or more selected random substitution sequences to be used to randomize one or more neural network layers of the transmitting DNN, and transmitting the control message includes transmitting a further control message specifying the selected random substitution sequences to be used.

22. The method according to claim 21, wherein the transmission of the control message comprises using control plane signaling to transmit a selected random substitution iteration or sequence, seed information, and corresponding time slot information.

23. The method according to claim 21 or 22, wherein transmitting the control messages includes transmitting each control message as a radio resource control (RRC) message, a medium access control (MAC) message, or a downlink control information (DCI) message.

24. The method according to any prior claim, further comprising using downlink control information (DCI) to transmit further control messages from the first device to the second device, wherein the DCI is used to indicate a selected random substitution sequence iteration or sequence used to reconstruct the transmit DNN according to the scramble timing information.

25. A method performed by a second device communicating with a first device, Receiving a control message from the first device indicating neural network scrambling information (NNSI) and scrambling timing information (432), Receiving a communication signal transmitted from the first device in accordance with the scramble timing information (442), Reconstructing the received deep neural network (DNN) of the second device according to the received NNSI and the scramble timing information (448), To generate reconstructed communication data represented by the received communication signal, the received communication signal is processed by the receiving DNN (450), Transmitting the reconstructed communication data to the data sink of the second device, or transmitting the reconstructed communication data to one or more upper protocol layers of the protocol stack of the second device (452), Methods that include...

26. Receiving the aforementioned control message (432) further means The first device receives one or more control messages, each of which indicates an NNSI and corresponding scramble timing information (432), The received NNSI is stored in storage for use in relation to the scramble timing information (434), The method according to claim 25, including the method described in claim 25.

27. The scramble timing information includes one or more time slots for receiving transmissions from the first device, and the method further includes The receiving of the communication signal from the first device (442) further includes receiving the communication signal from the first device in a specific time slot among the one or more time slots, Using the NNSI obtained from the storage, the received DNN of the second device is reconstructed (448), The receiving DNN processes the receiving communication signal to generate reconstructed communication data represented by the receiving communication signal of the specific time slot (450), The method according to claim 26, including the method described in claim 26.

28. The NNSI includes data representing the randomization of the input, output, or set of neural network nodes in one or more neural network layers of the first device's transmit DNN, Reconfiguring the receiving DNN of the second device further includes reconfiguring the receiving DNN of the second device using the NNSI according to the scramble timing information so as to reverse the randomization applied to the input, output, and / or neural network layer of the transmitting DNN. The method according to any one of claims 25 to 27.

29. The method according to claim 28, wherein the NNSI specifies a random substitution of the order of the set of neural network nodes in the one or more neural network layers of the transmitting DNN.

30. The NNSI includes one or more random replacement parameters, the one or more random replacement parameters include seed data, the number of iterations or sequence number of the random replacement, or identification of seed generation and pseudo-randomization functions, and is used to perform the random replacement on a set of neural network nodes in one or more neural network layers, and the reconfiguration of the receiving DNN of the second device is The process includes generating an inverse random permutation sequence corresponding to the number of iterations or sequence number of the random permutation for each neural network layer of the receiving DNN that corresponds to each neural network layer of the transmitting DNN, wherein the inverse random permutation sequence has a length equal to the number of neural network nodes in the set of neural network nodes for each neural network layer of the receiving DNN, and the reconfiguration of the receiving DNN of the second device further includes: With respect to each neural network layer of the receiving DNN corresponding to each neural network layer of the transmitting DNN, the generated inverse random replacement sequence is applied to the set of neural network nodes to derandomize the set of neural network nodes in each neural network layer of the receiving DNN. The method according to claim 29, including the method described in claim 29.

31. The seed data further includes one or more of the following: identification information for the first device, identification information for the second device, identification information for the cell in which the second device is located, identification information for the cell in which the first device is located, timing slot information, frame identification number, and / or any other information associated with the first device or the second device. The method further includes generating a seed from the seed data in order to perform inverse random replacement of the set of neural network nodes in the neural network layer of the receiving DNN corresponding to each neural network layer of the transmitting DNN. The method according to claim 30.

32. Assuming 1 ≤ I ≤ L, where L is the number of neural network layers in the transmit DNN, including the input and output layers, the NNSI specifies that the Ith neural network layer of the transmit DNN is randomized. Reconstructing the receiving DNN further includes performing an inverse random operation on the set of neural network nodes in the (L-I+1)th neural network layer of the receiving DNN. The method according to any one of claims 25 to 31.

33. The (L-I+1)th neural network layer of the receiving DNN has a set of N (N>1) neural network nodes in a specific order, and performing the inverse random operation on the set of neural network nodes of the (L-I+1)th neural network layer of the receiving DNN is: Based on the random permutations of the (L-I+1)th neural network layer, an N-dimensional permutation matrix (N>1) is generated, Inverting the N-dimensional permutation matrix to generate an inverted N-dimensional permutation matrix, The specific order of the set of N neural network nodes in the (L-I+1)th neural network layer is multiplied by the inverted N-dimensional permutation matrix to form an inverse randomized order of the set of N neural network nodes. The method according to claim 32, including the method described in claim 32.

34. The method according to any one of claims 25 to 33, wherein the NNSI specifies a random substitution iteration or sequence, seed information, and corresponding time slot information, the control message is received using control plane signaling, and the reconfiguration of the received DNN of the second device further includes reconfiguring the received DNN of the second device using the random substitution iteration or sequence and the seed to process the received communication signal transmitted from the first device according to the time slot information.

35. The method according to claim 34, wherein the control message is a radio resource control (RRC) message, a medium access control (MAC) message, or a downlink control information (DCI) message.

36. The method according to any one of claims 25 to 35, further comprising receiving one or more further control messages from the first device using downlink control information (DCI), wherein the DCI is used to indicate a selected random substitution sequence iteration or sequence and the scramble timing information in order to reconstruct the received DNN according to the scramble timing information.

37. The method according to any prior claim, wherein the first device (104a) is a base station (210) and the second device (104b) is user equipment (220).

38. The method according to any one of claims 1 to 36, wherein the first device (104a) is a user device (220) and the second device (104b) is a base station (210).

39. A computer-readable medium (1700) storing computer-readable instructions, wherein, when executed by a computer, the computer-readable instructions cause the computer to execute the method according to any of the prior claims.

40. A first device (210) comprising one or more processors (212) and memory (213), wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the first device (210) causes the first device (210) to execute the method according to any one of claims 1 to 24 and 37 to 38.

41. A second device (220) comprising one or more processors (222) and a memory (223), wherein the memory (223) stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors (222), the second device (220) causes the second device (220) to execute the method according to any one of claims 25 to 38.

42. Apparatus (210, 220), One or more antennas (203a / b), One or more processors (212, 222), Memory (213, 223) and Equipped with, The one or more processors (212, 222) are connected to the memory (213, 223) and the one or more antennas (203a / b), and the memory (213, 223) further stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors (212, 222), they cause the device (210, 220) to perform the method according to any one of claims 1 to 38. Device (210, 220).

43. A communication system (200), A first device (210) configured according to claim 40, A second device (220) configured according to claim 41, Equipped with, The first device (210) and the second device (220) establish a deep neural network communication session. Communication system (200).

Images