Digital radio signal receiver

Abstract

One example discloses a digital radio signal receiver comprising: wherein the receiver is configured to be coupled to a device; wherein the device is coupled to receive an RF signal; wherein the RF signal comprises a desired signal and interference; wherein the receiver is configured to receive a first signal and an interference model from a turn-off device training stage; an on-device training stage configured to construct a second signal and interference model based on the first signal and interference model and the RF signal received by the device; and a decoder configured to generate a data set from the RF signal based on the second signal and an interference model.

Description

Technical Field

[0001] This specification relates to systems, methods, devices, apparatuses, articles of manufacture, and instructions for receiving digital radio signals. Background Art

[0002] Electric vehicles (EVs) generate a wide variety of interfering signals that degrade or even make impossible the reception of Digital Audio Broadcasting (DAB) data and audio streams. Example EV noise sources include chargers, DC-DC converters, electric motor powertrains, and any other high-power switches. Summary of the Invention

[0003] According to an example embodiment, a digital radio signal receiver includes: wherein the receiver is configured to be coupled to a device; wherein the device is coupled to receive an RF signal; wherein the RF signal includes a desired signal and interference; wherein the receiver is configured to receive a first signal and an interference model from a shutdown device training stage; an on device training stage configured to construct a second signal and an interference model based on the first signal and the interference model and the RF signal received by the device; and a decoder configured to generate a data set from the RF signal based on the second signal and the interference model.

[0004] In another example embodiment, the digital radio signal receiver is a Digital Audio Broadcasting (DAB) radio receiver.

[0005] In another example embodiment, the shutdown device training stage is a general shutdown device training stage; and the on device training stage is a device-specific on device training stage.

[0006] In another example embodiment, the interference is non-radiated and independent of the RF signal.

[0007] In another example embodiment, the device is at least one of the following: an electronic device, a machine, a vehicle, an electric vehicle (EV), a battery EV, a hybrid EV, and a set of vehicles having a common manufacturer.

[0008] In another example embodiment, the first signal and the interference model and the second signal and the interference model use a trellis representation to represent the desired signal and the interference.

[0009] In another example embodiment, the trellis representation is characterized by a transition probability matrix and an observation likelihood.

[0010] In another example embodiment, the shutdown device training stage and the on device training stage use a first trellis diagram to represent the desired signal and a second trellis diagram to represent the interference.

[0011] In another exemplary embodiment, the off-device training stage and the on-device training stage create a first signal and interference model and a second signal and interference model through Kronecker product matrix multiplication of a first trellis diagram and a second trellis diagram.

[0012] In another exemplary embodiment, the off-device training stage is configured to parameterize the first signal and interference model by translating a first trellis representation and a second trellis representation into an implicit Markov model (HMM).

[0013] In another exemplary embodiment, the off-device training stage is configured to parameterize the first signal and interference model by translating a first trellis representation and a second trellis representation into a shallow neural network (NN).

[0014] In another exemplary embodiment, a first signal and interference model is constructed based on a set of domain knowledge.

[0015] In another exemplary embodiment, the set of domain knowledge is based on a set of different RF signals received by a set of different digital radio receivers coupled to a set of different devices.

[0016] In another exemplary embodiment, the first signal and interference model and the second signal and interference model are implicit Markov models (HMMs).

[0017] In another exemplary embodiment, Bayesian probability theory and message passing are used to generate the first signal and interference model and the second signal and interference model.

[0018] In another exemplary embodiment, the second signal and interference model is constructed by the on-device training stage when the device acquires an RF signal.

[0019] In another exemplary embodiment, an on-device update stage is further included, which includes iteratively updating the second signal and interference model using a decoder.

[0020] In another exemplary embodiment, the update includes tracking real-time changes in the received RF signal.

[0021] In another exemplary embodiment, the off-device training stage is configured to initialize the first signal and interference model.

[0022] In another exemplary embodiment, the off-device training stage is configured to parameterize the first signal and interference model by translating a first trellis representation and a second trellis representation into a combination of an implicit Markov model (HMM) and a shallow neural network (NN).

[0023] According to an example embodiment, a method for receiving a digital radio signal by a device includes: receiving an RF signal including a desired signal and interference; constructing a first signal and interference model during a device-off training stage; constructing a second signal and interference model during a device-on training stage based on the first signal and interference model and the RF signal received by the device; and decoding a data set from the RF signal based on the second signal and interference model.

[0024] The foregoing discussion is not intended to represent every example embodiment or every implementation within the scope of the current or future set of technical solutions. The drawings and the following detailed description also illustrate various example embodiments.

[0025] Various example embodiments can be more fully understood by considering the following detailed description in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Shows an example of a digital radio system.

[0027] Figure 2 Illustrates an example embodiment of the construction of a first signal and interference model for a digital radio system.

[0028] Figure 3A Represents a first example embodiment of the parameterization of a first signal and interference model for a digital radio system.

[0029] Figure 3B Represents a second example embodiment of the parameterization of a first signal and interference model for a digital radio system.

[0030] Figure 4 Represents an example embodiment of the initialization of a first signal and interference model for a digital radio system.

[0031] Figure 5 Represents an example embodiment of a possible marker function for device-specific turn-on of a DAB device / receiver section.

[0032] Figure 6 Represents an example embodiment of adapting a first signal and interference model for a digital radio system into a second signal and interference model.

[0033] Figure 7 Shows an example of the operation of a digital radio signal receiver.

[0034] Figure 8 Represents an example system for escrowing instructions for enabling a digital radio signal receiver.

[0035] While the present disclosure admits of various modifications and alternative forms, the details thereof have been shown by way of example in the drawings and will be described in detail. It should be understood, however, that other embodiments are possible in addition to the specific embodiments described. All modifications, equivalents, and alternative embodiments falling within the spirit and scope of the appended claims are also covered. Detailed Description

[0036] Some DAB spatial noise cancellation receivers use multiple antennas to improve DAB reception in the presence of EV interference. Other DAB receivers may use an iterative decoding method, where dedicated silicon is implemented to support the iterative decoding architecture; however, these receivers treat EV interference as noise.

[0037] Now described are hybrid artificial intelligence / machine learning (AI / ML) EV interference suppression systems and methods that reduce interference by considering the received signal and the EV interference infrastructure. The systems and methods define a solution space based on Bayesian probability theory and message passing, such as used in an implicit Markov model (HMM). AI / ML methods for learning and adapting these models are used to adopt novel probability models for mitigating the harmful effects of EV interference.

[0038] In some example embodiments, a three - level AI / ML - based interference suppression architecture is used.

[0039] In the first level, with the DAB device / receiver powered off, a first joint signal and EV interference model is generated / constructed based on graph theory and Bayesian probability theory. This model is initialized via offline optimization using synthetic and field test data to ensure fast convergence of the powered - on DAB device / receiver optimization.

[0040] In some example embodiments of the powered - off device training stage, a large training database is collected via extensive field recordings. A hybrid AI / ML method based on Bayesian probability theory and message passing is used. This method is applied to learn the HMM models of the signal and interference to obtain likelihoods and transition matrices. The models can be specific to different types (e.g., battery - electric vehicles and hybrid vehicles), different brands, and different models of vehicles. The result of this stage is a good initialization / prior for the subsequent stage.

[0041] In the second level, with the DAB device / receiver powered on, a second model is generated that adapts / optimizes the first model to further improve the initialization from the powered - off device training. The model is first adapted for operation in the DAB context during the received acquisition phase and takes advantage of short data acquisition cycles due to the less - stringent latency requirements of DAB. Adapting the powered - off device training enables faster convergence.

[0042] In some example embodiments of the turn-on device training stage, a hybrid AI / ML approach is also applied to further improve the signal and interference models and the likelihood for a particular vehicle. Due to the less stringent latency requirements of the digital broadcast system and the low-complexity hybrid method used, turn-on device training can be performed during the acquisition phase of reception. Due to good initialization, turn-on device training can converge faster.

[0043] In the third stage, turn-on device learning / updating based on real-time operating conditions is used to continuously fine-tune the second model via a unique iterative decoding architecture. In some example embodiments, a soft information feedback method is used to follow the real-time behavior of the system.

[0044] In some example embodiments of the turn-on device learning / updating stage, the second model and the likelihood are further updated via an iterative method (e.g., using an iterative decoder) to track the real and time-varying nature of the channel (e.g., channel state information (CSI)) and interference conditions. This is performed for the channel to which the receiver is tuned. In some embodiments, a virtual tuner or a background tuner can be used to learn / update the HMM model for different channels in the background.

[0045] Overview

[0046] Figure 1 An example of a digital radio system 100 is shown. The system 100 includes a general off DAB device / receiver section 102 that performs model building 108, parameterization, and initialization. The system 100 also includes a device-specific turn-on DAB device / receiver section 104 that performs model adaptation and continuous fine-tuning. In some example embodiments, the system 100 uses a hybrid AI / ML EV interference suppression method in the DAB system.

[0047] In the first stage, the general off DAB device / receiver section 102 includes a domain knowledge set 106, model building 108, model parameterization 110, and model initialization 112.

[0048] Model building 108 constructs First signal and interference models (e.g., an advanced joint signal and EV interference model) using the available domain knowledge 106 via graph theory and Bayesian probability theory.

[0049] Model parameterization 110 parameterizes First the signal and interference model 117 such that it can be optimized in a data-driven manner.

[0050] Model initialization 112 initializes the parameterized first model 117 using both synthetic and collected field test data 114 stored in a large augmented database 116.

[0051] In a second stage, the first model 117 is then transmitted (e.g., deployed) to the digital radio signal receiver 118 in the switched-on DAB device / receiver section 104. The digital radio signal receiver 118 is configured to be coupled to a device (not shown) to receive the RF signal 120.

[0052] In various example embodiments, the RF signal 120 includes an interference component. In many various example embodiments, the device (not shown) can be at least one of the following: an electronic device, a machine, a vehicle, an electric vehicle (EV), a battery EV, a hybrid EV, and a collection of vehicles having a common manufacturer.

[0053] In many example embodiments, good initialization is crucial for the fast switched-on DAB device / receiver 104 convergence and overall system 100 performance, and there is an inherent deviation between the data used in the general switched-off DAB device / receiver section 102 and the specific communication context of the device in which the first model 117 is applied.

[0054] Accordingly, the device-specific switched-on DAB device / receiver section 104 includes fast data acquisition 122 and model adaptation 124 for adapting the first model 117 to be device-specific Second signal and interference model 125.

[0055] The fast data acquisition 122 is stored in the real-time data storage device 126. Note that due to the less stringent latency requirements in the DAB system, the fast data acquisition 122 stage and storing the data collected at runtime in the real-time data storage device 126 can conveniently achieve the direct adaptation of the initialization model.

[0056] In a third stage, the successive fine-tuning 128 and the decoder 130 receive the RF signal 120 with interference via real-time reception 132. The successive fine-tuning 128 and the decoder 130 perform iterative decoding such that the second model 125 can be continuously fine-tuned during real-time reception 132 with the changing communication / channel conditions.

[0057] Model construction 108

[0058] Figure 2 Shows an example embodiment 200 of the construction 108 of the first signal and interference model 117 for the digital radio system 100.

[0059] The construction 108 of the first model 117 utilizes graph theory and Bayesian probability theory. The first model 117 uses a lattice representation of multipath signal propagation 204 and time-varying interference 206, where the time-varying interference 206 is 'unradiated' (i.e., independent of the broadcast signal).

[0060] Two trellis representations 204 and 206 are characterized by a transition probability matrix (Ps and Pi) and corresponding observation likelihoods (s(ot) and bi(ot)).

[0061] Subsequently, a first model 117 (e.g., a joint signal and EV interference model) is obtained via an S+I combined model 202, which is a combination of two component models 204, 206 (e.g., a Kronecker product matrix multiplication operation).

[0062] The obtained model inherently incorporates all possible (allowed) transitions between joint multipath and interference states, and thus captures the temporal structure of the underlying process affecting the received RF signal 120. Trellis-based multipath (intersymbol interference) propagation models are also used in wireless communication, and trellis-based interference (impulse noise) models are also used in power line communication.

[0063] Model parameterization 110

[0064] Figure 3A A first exemplary embodiment 300 of a parameterization 110 representing a first signal and interference model 117 for a digital radio system 100. The first example 300 employs a fully learnable transition matrix and observation likelihoods.

[0065] After constructing the first model 117 using a trellis diagram incorporating domain knowledge 106, the parameterization 110 generates a learnable model 302, which can then be systematically further optimized into a second model 125. The parameterization 110 ensures that both the observation likelihood function (b(ot)) and the transition matrix (P * ) are parameterized and thus available for learning.

[0066] In the literature, it has been proposed that the likelihood of a trellis-based model can be learned from data using a (shallow) neural network (NN) trained on a relatively small labeled dataset. However, this NN method cannot be directly applied to learning the transition matrix of the first model 117.

[0067] Therefore, here the parameterization 110 treats the trellis as representing an implicit Markov model (HMM) and uses available techniques to optimize the HMM. That is, both the likelihood (b(ot)) and the transition matrix (P * ) can be trained, for example, via a sequence of training data input to the Baum-Welch algorithm for training an HMM, which in turn includes forward-backward Bayesian calculations of probabilities and expectation maximization (EM).

[0068] Figure 3B Represents a second exemplary embodiment 304 of the parameterization 110 of the first signal and interference model 117 for a digital radio system 100. The second example 304 utilizes a shallow neural network (NN) for observing likelihood estimation to generate a second learnable model 306. Alternatively, an implicit Markov model technique (e.g., the Baum-Welch algorithm) can be used to learn both the transition matrix and the observation likelihood.

[0069] Model initialization 112

[0070] Figure 4 Represents an exemplary embodiment 400 of the initialization 112 of the first signal and interference model 117 for a digital radio system 100. An exemplary set of initialization elements 402 is shown.

[0071] During model initialization 112, synthetic data from the interference source generator and / or other field tests is used to optimize the developed learnable models 302, 306 of the first model 117 in an "offline" manner before deployment. Formal initialization 112 prevents any random initialization of the first model 117, which may cause convergence problems and thus limit the performance during the operation of the system 100.

[0072] Note that if sufficiently accurate and representative model parameters are available as prior knowledge, these model parameters can also be utilized before optimization using synthetic and field test data. Depending on the algorithm selection, Bayesian probability theory can be used to perform the initialization via the forward-backward and expectation maximization algorithms. Alternatively, if a neural network is employed to compute the likelihood, the initialization will also involve a gradient descent algorithm.

[0073] Note that while in principle the HMM optimization framework can be performed in an unsupervised manner via the Baum-Welch algorithm, using an NN will typically require a labeled dataset for supervised learning. In the case of shutting down the DAB device / receiver, optimization and synthetically generated data labeling may typically be readily available. On the other hand, during field test data collection, ground truth data (e.g., recorded near the base station) also needs to be incorporated.

[0074] Turning on the DAB device / receiver data acquisition

[0075] Figure 5 Represents an exemplary embodiment 500 of a possible labeling function 502 for the device-specific turning on of the DAB device / receiver section 104. The first model 117 is initialized in the general turning off DAB device / receiver section 102 and then sent to the device-specific turning on DAB device / receiver section 104 for use in a specific device (not shown but introduced above).

[0076] Inherently, there will be a deviation between the conditions under which the first model 117 is initialized and the device-specific communication scenario and interference characteristics. Thus, as introduced above, the device-specific turn-on DAB device / receiver section 104 employs fast data acquisition 122 of the received RF signal 120 and then uses it to generate the second model 125.

[0077] The DAB application allows for relaxed latency requirements, which can be exploited to collect a small database (i.e., the real-time data storage device 126) "on the fly" at runtime.

[0078] The selected optimization framework for the parameterized joint signal and interference model may require supervised learning and thus labeled data (e.g., when using an NN). In this case, frame preambles or short pilot sequences can be exploited. Alternatively, the initialized model can be employed in a semi-supervised framework such as pseudo-labeling, where labels are assigned to unlabeled data based on predictions of the initialized NN model.

[0079] Model adaptation

[0080] Figure 6 An example embodiment 600 showing the model adaptation 124 of the first signal and interference model 117 for use in the digital radio system 100 into the second signal and interference model 125.

[0081] The first model 117 can be adapted into a second model 125 that is specific to both the device type and the real-time communication environment. Note that when using an NN, this can be done once the necessary labels 502 are available. The algorithms used for NN / HMM optimization (i.e., gradient descent and / or the forward-backward algorithm and EM) are reused. Note that during the general turn-off DAB device / receiver section 102 learning phase, these algorithms can run for several iterations to provide a good set of starting parameters. At the device-specific turn-on DAB device / receiver section 104 model adaptation 124 stage, only a small amount of data is needed to quickly adapt the already optimized first model 117.

[0082] Continuous fine-tuning

[0083] As Figure 1 introduced above, the third stage involves continuous fine-tuning 128 of the second model 125 in real time during the reception of the RF signal 120 using iterative decoding of the decoder 130. This allows for continuous "fine" adaptation as the RF signal 120 rapidly changes and / or environmental / channel conditions that may not have been captured by the acquisition stage.

[0084] Since the first model 117 and the second model 125 are constructed as lattice diagrams, they can be easily used with signal reception algorithms such as BCJR and Viterbi. This lattice diagram approach is a much less complex approach than a comprehensive deep learning solution. This lattice diagram also specifically captures the infrastructure, propagation effects, and any interference of the RF signal 120. The full parameterization of the first model 117 and the second model 125 using these lattice diagrams allows for dynamic adaptation in a data-driven manner.

[0085] Figure 7 Represents an operational example 700 of the digital radio signal receiver 100. The operational example 700 shows exemplary functionality performed in the general-off DAB device / receiver section 102 and the device-specific on DAB device / receiver section 104.

[0086] Figure 8 Represents an example system 800 for escrowing instructions for enabling the digital radio signal receiver 100. The system 800 shows an interface with the input / output data 802 of the electronic device 804. The electronic device 804 includes a processor 806, a storage device 808, and a non-transitory machine-readable storage medium 810. The machine-readable storage medium 810 includes instructions 812 that control how the processor 806 receives the input data 802 and transforms the input data into the output data 802 using data within the storage device 808. Example instructions 812 stored in the machine-readable storage medium 810 are discussed elsewhere in this specification. In an alternative example embodiment, the machine-readable storage medium is a non-transitory computer-readable storage medium.

[0087] A processor (such as a central processing unit, CPU, microprocessor, application-specific integrated circuit (ASIC), etc.) controls the overall operation of the storage device (such as random access memory (RAM) for temporary data storage, read-only memory (ROM) for permanent data storage, firmware, flash memory, external and internal hard disk drives, etc.). The processor device communicates with the storage device and the non-transitory machine-readable storage medium using a bus and performs operations and tasks implementing one or more instructions stored in the machine-readable storage medium. In an alternative example embodiment, the machine-readable storage medium is a computer-readable storage medium.

[0088] Example embodiments of the materials discussed in this specification can be implemented, in whole or in part, via a network, computer, or data-based device and / or service. These can include the cloud, the Internet, an intranet, a mobile device, a desktop computer, a processor, a lookup table, a microcontroller, a consumer device, an infrastructure, or other enabling devices and services. As used herein and in the claims, the following non-exclusive definitions are provided.

[0089] It will be readily understood that the components of the embodiments, as generally described herein and illustrated in the accompanying drawings, can be arranged and designed in a wide variety of different configurations. Thus, the detailed description of the various embodiments represented in the figures is not intended to limit the scope of the present disclosure, but is merely representative of the various embodiments. While aspects of the embodiments are presented in the drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0090] Without departing from the spirit or essential characteristics of the invention, the invention may be embodied in other specific forms. The described embodiments are to be considered in all respects only as illustrative and not restrictive. Thus, the scope of the invention is indicated by the appended claims rather than by this detailed description. All changes that come within the meaning and range of equivalency of the claims are to be embraced within their scope.

[0091] References throughout this specification to features, advantages, or similar language do not imply that all of the features and advantages that can be realized by the invention should be in or in any single embodiment of the invention. In fact, the language referring to the features and advantages should be understood to mean that a particular feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of the invention. Thus, the discussions of the features and advantages throughout this specification and similar language may, but do not necessarily, refer to the same embodiment.

[0092] Furthermore, the described features, advantages, and characteristics of the invention can be combined in any suitable manner in one or more embodiments. Based on the description herein, those of ordinary skill in the relevant art will recognize that the invention can be practiced without one or more of the specific features or advantages of a particular embodiment. In other instances, additional features and advantages may be identified in certain embodiments that may not be present in all embodiments of the invention.

[0093] References throughout this specification to "one embodiment," "an embodiment," or similar language mean that a particular feature, structure, or characteristic described in connection with the indicated embodiment is included in at least one embodiment of the invention. Thus, the phrases "in one embodiment," "in an embodiment," and similar language throughout this specification may, but do not necessarily, all refer to the same embodiment.

Claims

1. A digital radio signal receiver, characterized in that, Comprising: wherein the receiver is configured to be coupled to a device; wherein the device is coupled to receive an RF signal; wherein the RF signal includes a desired signal and interference; wherein the receiver is configured to receive a first signal and an interference model from an off-device training stage; an on-device training stage configured to construct a second signal and an interference model based on the first signal and the interference model and the RF signal received by the device; and a decoder configured to generate a data set from the RF signal based on the second signal and the interference model.

2. The receiver according to claim 1: It is characterized in that The digital radio signal receiver is a digital audio broadcast DAB radio receiver.

3. The receiver according to claim 1: It is characterized in that The off-device training stage is a general off-device training stage; and wherein the on-device training stage is a device-specific on-device training stage.

4. The receiver according to claim 1: It is characterized in that The interference is unradiated and independent of the RF signal.

5. The receiver according to claim 1: It is characterized in that, The first signal and the interference model and the second signal and the interference model use a lattice representation to represent the desired signal and the interference.

6. The receiver according to claim 5: It is characterized in that The lattice representation is characterized by a transition probability matrix and an observation likelihood.

7. The receiver according to claim 1: It is characterized in that The off-device training stage and the on-device training stage use a first lattice diagram to represent the desired signal and use a second lattice diagram to represent the interference.

8. The receiver according to claim 7: It is characterized in that The off-device training stage and the on-device training stage create the first signal and the interference model and the second signal and the interference model by Kronecker product matrix multiplication of the first lattice diagram and the second lattice diagram.

9. The receiver according to claim 7: Characterized in that, The off-device training stage is configured to parameterize the first signal and the interference model by interpreting the first lattice representation and the second lattice representation as an implicit Markov model HMM.

10. A method for receiving digital radio signals by a device, characterized in that, Comprising: Receiving an RF signal including a desired signal and interference; Constructing a first signal and an interference model during an off-device training stage; Constructing a second signal and an interference model during an on-device training stage based on the first signal and the interference model and the RF signal received by the device; and Decoding a data set from the RF signal based on the second signal and the interference model.

Images