Digital predistortion method for millimeter-wave communication transmission systems with subarray architecture

Abstract

This invention discloses a digital predistortion method for millimeter-wave communication transmission systems with subarray architectures. Based on the non-ideal characteristics of various devices in a hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture, the nonlinear effects of the system are analyzed. A receiving antenna is set in the main beam direction, and the transmitted RF signal is down-converted and sampled into a baseband digital signal by an analog-to-digital converter (ADC), which is then input to the digital predistortion module. The signal fed back from the ADC is differentially processed with the signal input to the RF chain after digital precoding. An adaptive algorithm is used to continuously correct the sampled baseband digital signal, thus achieving digital predistortion in the main beam direction. This method can dynamically adjust the predistortion parameters according to the actual nonlinear characteristics. By continuously updating the predistortion model, it can effectively cope with nonlinear changes in the system and has a wider range of applications.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, specifically relating to an adaptive digital predistortion method for a hybrid beamforming millimeter-wave communication transmission system with a subarray architecture. Background Technology

[0002] Millimeter-wave communication technology has become one of the key technologies for next-generation communication systems. Millimeter waves have a large bandwidth, supporting higher data transmission rates and lower latency, making their application prospects very broad. However, millimeter-wave signals experience significant attenuation during propagation and are significantly affected by obstructions. To overcome the propagation attenuation problem of millimeter waves, beamforming technology has emerged. Beamforming concentrates energy to form a directional transmission beam by adjusting the phase and amplitude of multiple antennas, thereby improving signal propagation efficiency and anti-interference capabilities. As the scale of antenna arrays increases, the complexity of beamforming technology and the demand for computing resources also increase. To address this issue, hybrid beamforming technology has received widespread attention. Hybrid beamforming combines the advantages of digital and analog beamforming, reducing hardware complexity and power consumption while ensuring system performance, making it particularly suitable for large-scale antenna arrays in millimeter-wave communication.

[0003] In large-scale antenna arrays, the sheer number of antennas leads to increased hardware complexity and power consumption. To optimize this situation, subarray architecture has emerged. This architecture divides the large-scale antenna array into several smaller subarrays, each operating independently, reducing system complexity while improving flexibility and resource utilization.

[0004] In millimeter-wave communication systems, devices such as power amplifiers often cause signal distortion due to their nonlinear characteristics, thus affecting communication quality. Digital predistortion technology can effectively compensate for the nonlinear distortion caused by power amplifiers and other devices. By preprocessing the transmitted signal, it makes the signal after amplification as linear as possible, reducing the impact of distortion on the communication system.

[0005] Existing digital predistortion methods only consider the nonlinear effects of the power amplifier, neglecting the nonlinear characteristics of other non-ideal components in the system. Moreover, since millimeter-wave communication environments and channel characteristics typically exhibit strong dynamic changes, traditional fixed digital predistortion methods may not be able to cope with various variations. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a digital predistortion method for millimeter-wave communication transmission systems with subarray architectures. A digital predistortion module is designed to compensate for nonlinear distortion in a hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture. A nonlinear model of the system is established based on the nonideal characteristics of each component. When the standard test signal transmitted by the source enters the MIMO channel through the nonlinear system, it is received by the receiving antenna in the main beam direction. The signal is then down-converted and sampled into a baseband digital signal. The signal fed back from the analog-to-digital converter (ADC) is differentially processed with the signal input to the RF chain after digital precoding. Under the action of an adaptive algorithm, the difference between the signal fed back from the ADC and the signal input to the RF chain after digital precoding eventually approaches zero. At this point, the adaptive algorithm converges, and digital predistortion is achieved.

[0007] The technical solution of the present invention includes the following steps:

[0008] Step 1: Analyze the nonlinear effects of the system based on the nonideal characteristics of each component in the millimeter-wave communication transmission system;

[0009] Step 2: Set up the receiving antenna in the direction of the main beam, and sample the transmitted radio frequency signal into a baseband digital signal through down-conversion and analog-to-digital converter, then input it into the digital predistortion module:

[0010] Step 3: Differential processing is performed on the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. The acquired baseband digital signal is continuously corrected using an adaptive algorithm, thus realizing digital predistortion in the main beam direction.

[0011] Step 4: Within the maximum scanning range of the phased array antenna array, continuously adjust the angle to achieve digital predistortion under different beam directions.

[0012] Furthermore, the specific method for step 1 is as follows:

[0013] A hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture consists of a signal source, a digital precoder, a digital-to-analog converter (DAC), an up-conversion converter, an analog beamformer, an antenna array, a MIMO channel, a down-conversion converter, and an analog-to-digital converter (ADC). A digital predistortion module is designed to compensate for nonlinear distortion in the system. The DAC, up-conversion converter, and analog beamformer form the RF chain. The analog beamformer consists of a phase shifter and a power amplifier. The digital predistortion module includes a predistorter (DPD) and DPD parameter estimation.

[0014] In a millimeter-wave communication transmission system based on a subarray architecture, nonlinear interference caused by the non-ideal characteristics of the DAC, upconversion module, power amplifier, downconversion module, and ADC is taken into account.

[0015] The resolution of a DAC determines the accuracy of the converted signal, and some distortion is unavoidable during the signal conversion process. When a digital-to-analog converter (DAC) discretizes a continuous analog signal into a digital signal, quantization error exists. The conversion function of a DAC is not perfectly linear, introducing nonlinear distortion.

[0016] The upconversion module mainly consists of two modules: an IQ modulator and an upconverter. An imperfect balance in the I and Q transmission links can lead to amplitude and phase imbalances.

[0017] For broadband and high-frequency applications, power amplifiers introduce nonlinear distortion. When a power amplifier is operating, the relationship between its input and output is not linear, resulting in nonlinear distortion of the output signal, including harmonic distortion and intermodulation distortion.

[0018] The downconversion module consists of a downconverter and an IQ demodulator. The nonlinear distortion of this module is similar to that of the upconversion module, with amplitude and phase imbalance caused by IQ imbalance.

[0019] ADCs also suffer from quantization errors when discretizing analog signals into digital signals. Similar to DACs, the limited number of digital quantization bits cannot perfectly represent a continuous analog signal, resulting in a difference between the output digital signal and the original analog signal. Furthermore, the conversion function of an analog-to-digital converter is not perfectly linear, introducing nonlinear distortion.

[0020] In a digital beamforming architecture, the digital predistortion module is implemented after the beamforming operation; that is, each power amplifier is equipped with a dedicated predistorter to compensate for the distortion it generates. In this case, the signal received by the user end takes the following form:

[0021] R = H T G[F D (W D X)]

[0022] in Represents the transmitted signal matrix. This represents the received signal matrix from different user terminal directions, where K is the sample length and Q is the number of transmitted signal streams; W D H and H are the P×Q digital beamformer and channel matrix, respectively, representing the number of power amplifiers; G(·)=[g1(·),g2(·),...,g P (·)]and This represents the nonlinear transfer function of the power amplifier and predistorter in the array.

[0023] In digital predistortion operation, the inverse behavior model of the amplifier is identified and used as the predistorter, i.e. p = 1, 2, ..., P. Therefore, the cascaded module of the predistorter and power amplifier is considered a linear system, in which case the user will receive a linearized signal containing only information from the corresponding RF link.

[0024] For a hybrid beamforming millimeter-wave communication transmitter system with P antennas and Q independent RF chains (Q << P), each digital stream simultaneously drives multiple antennas and power amplifiers. After analog beamforming, the received signal is described as:

[0025] R = H T G[WF A (X)]

[0026] in This represents the nonlinear function of the predistorter. In the SA (subarray) architecture, each Q RF chain is independently connected to a set of P antennas and a power amplifier via N phase shifters, resulting in an overall analog beamformer W SA To form a block diagonal matrix:

[0027]

[0028] in q = 1, ..., Q. Therefore, the input signal of the power amplifier can be expressed as:

[0029]

[0030] in It is the Kronecker product. Let represent the input signal of each power amplifier. From the above formula, it can be seen that each power amplifier in the subarray hybrid beamforming system is driven only by the phase signal of the corresponding RF chain. Therefore, the received signal of the i-th user can be expressed as:

[0031]

[0032] in i = 1, 2, ..., Q represents a set of univariate nonlinear transfer functions. Ignoring crosstalk, the i-th beam of the SA hybrid beamforming array contains only the transmitted signal from the corresponding RF chain. A digital predistortion (DPD) model configured independently for each RF chain is equivalent to the inverse model of multiple univariate nonlinear models.

[0033] Any continuously differentiable nonlinear function can be expanded using Taylor series; therefore, Taylor series are often used to approximate nonlinear functions. In particular, in modeling the nonlinear behavior of power amplifiers, the Volterra series model, which is very similar to Taylor series, is widely used. Its discretized bandpass expression is as follows:

[0034]

[0035] in Let M be the bandpass, real input signal and output signal, respectively, where M is the memory depth, P is the nonlinearity order, and h is the input signal and output signal. p (m1, ..., m) p ) is a p-order Volterra kernel.

[0036] In DPD processing, the signal is represented using a low-pass complex signal. Therefore, in practical applications, the bandpass Volterra series model above can be transformed into the following equivalent low-pass Volterra series model:

[0037]

[0038] In the above equation, x(n) and y(n) represent the low-pass and complex input and output signals, respectively.

[0039] Since the Volterra series model is only suitable for modeling weakly nonlinear systems and has high model complexity, the Memory Polynomial (MP) model is used instead of the equivalent low-pass Volterra series model. The Memory Polynomial (MP) model is a classic simplified model derived from the Volterra series model. The commonly used MP model expression now includes both odd-order and even-order distortion terms, as shown in the following equation:

[0040]

[0041] Furthermore, the specific method for step 2 is as follows:

[0042] Based on the construction and nonlinear analysis of a hybrid beamforming millimeter-wave communication transmission system with a subarray architecture, a receiving antenna is positioned along the main beam direction to receive the radio frequency (RF) signals generated by the transmitting system. Standard test signals, such as standard dual-tone signals, QPSK signals, and 16QAM signals, are transmitted from the signal source. These standard test signals undergo digital pre-coding via a digital pre-encoder, are split into several data streams, and input to the RF chain for up-conversion and analog beamforming. The beam signals are then transmitted through a power amplifier and a phased array antenna array. In actual experiments, the position of the receiving antenna is often fixed. The main beam direction is determined by the azimuth and elevation angles between the phased array antenna array and the receiving antenna.

[0043] The digital predistortion module's predistorter receives a pre-encoded digital signal as input at system startup. Since the DPD parameter estimation module hasn't yet received the feedback output signal, the predistorter doesn't process the signal at this point. The signal is input to the RF link, affected by the nonlinearity of the DAC, up-conversion module, and power amplifier. After passing through the phase shifter network, it's converted into an RF signal with nonlinear distortion. The RF signal is transmitted in the MIMO channel, received by the receiving antenna, and then sequentially down-converted and sampled by the ADC, converting it into a baseband digital signal fed back to the digital predistortion module. During this process, nonlinear interference from the down-converter and ADC is introduced again.

[0044] Furthermore, the specific method for step 3 is as follows:

[0045] In step 2, the system's transmitted signal has been down-converted and sampled into a baseband digital signal by an analog-to-digital converter. The baseband digital signal is then input into the DPD parameter estimation module in the digital predistortion module to extract the nonlinear model parameters. After inversion, the parameter matrix of the predistorter is obtained. The specific operation is as follows: First, the nonlinear model parameters are extracted by differential processing of the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. Under the action of the adaptive LMS algorithm, the parameter matrix of the predistorter is continuously learned and adjusted until the difference between the standard test signal and the signal fed back from the power amplifier's output signal approaches 0. At this time, the system response corresponding to the parameter matrix of the predistorter is the inverse effect of the nonlinear response of the entire system, thus obtaining a digital predistortion model suitable for non-ideal hardware conditions.

[0046] The LMS algorithm calculates the output error at each time step and then adjusts the predistorter coefficients based on this error. This update process is asymptotic, meaning that the LMS algorithm can gradually approach the ideal linearization effect by adjusting the coefficients in small steps.

[0047] Specifically, given an input signal x(n) and a desired output signal d(n), the LMS algorithm outputs the actual signal y(n) based on the current model parameters w(n), and then updates the next model parameters w(n+1) by comparing the difference between the actual output signal and the desired output signal. The update rule of the LMS algorithm is as follows:

[0048] w(n+1)=w(n)+μ·e(n)x(n)

[0049] e(n) = d(n) - y(n)

[0050] Where w(n) represents the model parameters at time n, μ is the learning rate (also called step size), which determines the amount of model parameter update in each iteration and affects the convergence speed of the algorithm, e(n) represents the degree to which the actual output signal y(n) deviates from the desired output signal d(n), and x(n) represents the current input signal. According to this update rule, the LMS algorithm continuously adjusts the model parameters so that the predicted output signal gradually approaches the desired output signal, thereby achieving the purpose of signal analysis and prediction.

[0051] When the system transmits a standard test signal under the action of beamforming, the digital predistortion model in the direction of the current main beam can be obtained by using the LMS algorithm.

[0052] Furthermore, the specific method for step 4 is as follows:

[0053] Steps 2 and 3 are used to achieve digital predistortion in all main beam directions. Since the phased array antenna array can change the pointing angle of the main beam by modifying the feed phase of the phase shifter, it is also necessary to continuously adjust the pointing angle of the main beam within the maximum scanning range of the phased array antenna array to obtain digital predistortion models for all different angles.

[0054] The beam scanning range of a phased array antenna is determined by the azimuth angle φ and the scanning angle θ, which are complementary to the elevation angle. The azimuth angle φ ranges from 0 to 360°, and the scanning angle θ ranges from 0 to 60°. Each set of azimuth and scanning angles corresponds to a beam pointing, and each beam pointing corresponds to a digital predistortion model. While satisfying the beam pointing accuracy, all possible combinations of azimuth and scanning angles are iterated to obtain digital predistortion models for different angles.

[0055] Digital predistortion models trained at different angles were used to process the standard test signal at the transmitting end. For each digital predistortion model, a set of baseband digital signals was obtained at the receiving end through down-conversion and ADC sampling by continuously changing the beam direction. The error between this set of signals and the standard test signal was calculated, and the optimal digital predistortion model was selected as the final adaptive digital predistortion model under the criterion of minimizing the output signal error.

[0056] The beneficial effects of this invention are as follows:

[0057] This invention proposes an adaptive digital predistortion method for hybrid beamforming millimeter-wave communication transmission systems based on subarray architectures. This method can compensate for nonlinear distortion of the transmitted signal caused by non-ideal components such as power amplifiers, IQ modulators, and upconverters. Due to its adaptive nature, the digital predistortion method can dynamically adjust the predistortion parameters according to the actual nonlinear characteristics. By continuously updating the predistortion model, it can effectively cope with nonlinear changes in the system, thus having a wider range of applications. Attached Figure Description

[0058] Figure 1 Hybrid millimeter-wave communication transmission system based on subarray architecture;

[0059] Figure 2 The RF chain structure of the subarray architecture;

[0060] Figure 3 The structure of a digital predistortion module;

[0061] Figure 4 LMS algorithm principle;

[0062] Figure 5 LMS algorithm flowchart;

[0063] Figure 6 Beam pointing of a phased array antenna;

[0064] Figure 7 The RF chain structure in the embodiment;

[0065] Figure 8 The DAC module in the embodiment;

[0066] Figure 9 The up-conversion module in the embodiment;

[0067] Figure 10 The simulated beamformer in the embodiment;

[0068] Figure 11 The digital predistortion module in the embodiment. Detailed Implementation

[0069] The specific implementation method of the present invention will be further described below with reference to the accompanying drawings.

[0070] Step 1: Construct a hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture:

[0071] like Figure 1 As shown, the hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture consists of a signal source, a digital precoder, a digital-to-analog converter (DAC), an up-conversion converter, an analog beamformer, an antenna array, a MIMO channel, a down-conversion converter, an analog-to-digital converter (ADC), and a digital predistortion module. The DAC, up-conversion converter, and analog beamformer form the radio frequency (RF) chain. The analog beamformer consists of a phase shifter and a power amplifier. The digital predistortion module includes a predistorter (DPD) and DPD parameter estimation.

[0072] The main function of the source module is to generate standard test signals, such as standard two-tone signals, QPSK signals, and 16QAM signals. The raw signal is modulated by the corresponding modulator and then transmitted through a root-raised-cosine transmit filter to eliminate inter-symbol interference during signal transmission.

[0073] The function of a digital precoder is to precode the baseband signal input from the signal source, using methods such as ZF precoding, MMSE precoding, and SVD precoding. In multi-user MIMO systems, the precoder utilizes channel state information to perform linear transformations on the signals of different users, effectively separating user signals and reducing mutual interference. The digital precoder uses algorithms to minimize interference between different users during transmission, ensuring that each user receives a signal with minimal interference. Simultaneously, the digital precoder enables spatial multiplexing, allowing multiple independent data streams to be transmitted to multiple users or via a multi-antenna system using the same time and frequency resources. This spatial multiplexing adjusts the directionality of the transmitted signal, enabling multiple data streams to be transmitted through different spatial paths, thereby improving the system's spectral efficiency and transmission capacity.

[0074] The DAC converts the pre-encoded digital signal into an analog signal for subsequent signal processing. The upconversion module mainly consists of two modules: an IQ modulator and a modulator. The IQ modulator first modulates the baseband signal to the intermediate frequency (IF), and then the modulator modulates the IF signal into a high-frequency signal.

[0075] Analog beamformers mainly consist of phase shifters and power amplifiers. The primary role of the phase shifter in hybrid beamforming is to adjust the phase of the signals, causing the signals emitted by multiple antennas to interfere and superimpose in a specific direction, thus achieving beamforming. Specifically, the phase shifter changes the signal phase of each antenna element, ensuring that the signals from different antenna elements have the same phase in the target direction, forming coherent superposition and maximizing signal strength in that direction. Simultaneously, in non-target directions, the signal phases are inconsistent and cancel each other out, thus suppressing interference. This method of adjusting the signal phase using phase shifters effectively achieves analog beamforming and controls the directionality of the transmitted signal. The power amplifier amplifies the signal power to a sufficient level to ensure that the signal can cover the target area and penetrate long-distance wireless communication environments.

[0076] The downconversion module converts the radio frequency signal output by the antenna array into a baseband signal, and the ADC module converts the feedback baseband analog signal into a baseband digital signal, which is then input into the DPD parameter estimation module to complete the extraction of system nonlinear model factors.

[0077] The MIMO channel module is considered as a MIMO channel under microwave anechoic chamber conditions, i.e., there is only one direct wave and the transmitting and receiving antennas are in the far field condition.

[0078] Step 2: Establish the nonlinear model of the system based on the nonideal characteristics of each component in the millimeter-wave communication transmission system.

[0079] In a millimeter-wave communication transmission system based on a subarray architecture, nonlinear interference caused by the non-ideal characteristics of the DAC, upconversion module, power amplifier, downconversion module, and ADC is taken into account.

[0080] The resolution of a DAC determines the accuracy of the converted signal, and some distortion is unavoidable during the signal conversion process. When a digital-to-analog converter (DAC) discretizes a continuous analog signal into a digital signal, quantization error exists. This is because the limited number of digital quantization bits cannot perfectly represent a continuous analog signal, resulting in a difference between the output signal and the original analog signal. The conversion function of a DAC is not perfectly linear, introducing nonlinear distortion. The dynamic range of the input and output signals of a DAC is limited. When the input signal is too large or too small, exceeding the effective range of the DAC, it will cause the output signal to be distorted or unable to correctly represent the input signal. A clock signal is used for timing control in a DAC. Clock jitter or clock noise is introduced into the DAC output, causing timing inaccuracies and jitter in the output signal.

[0081] The upconversion module mainly consists of two modules: an IQ modulator and an upconverter. Imperfect balance in the I and Q transmission links can lead to amplitude and phase imbalances. For example, imperfect amplifier gain matching and uneven filter response in the I and Q paths can cause balance errors. Phase shifts between the I and Q paths can cause phase errors. Phase errors are caused by signal source instability, clock jitter, or phase drift in the transmission link. Noise is random interference introduced during signal transmission. Noise originates from environmental interference, noise from the circuit components themselves, and interference during signal transmission. Noise affects the signal quality of both the I and Q paths. In wireless communication, multipath propagation is caused by the reflection and propagation of signals along multiple different paths in the transmission path. This leads to differences in signal arrival time and amplitude on different paths in the I and Q paths, causing demodulation errors. Due to the instability of the intermediate frequency clock in the RF signal source or receiver, frequency shifts can cause frequency differences between the I and Q paths, resulting in phase and timing misalignments.

[0082] For broadband and high-frequency applications, power amplifiers introduce nonlinear distortion. When a power amplifier operates, the relationship between its input and output is not linear. This causes distortion in the output signal, including harmonic distortion, intermodulation distortion, and other issues. Nonlinear distortion alters the signal shape, reducing the system's transmission accuracy and dynamic range.

[0083] The downconversion module consists of a downconverter and an IQ demodulator, which converts the radio frequency signal output by the antenna array into a baseband signal. The nonlinear distortion of this module is similar to that of the upconversion module, and there is amplitude and phase imbalance caused by IQ imbalance.

[0084] Quantization errors also exist when an ADC discretizes an analog signal into a digital signal. Similar to a DAC, the limited number of digital quantization bits cannot perfectly represent a continuous analog signal, resulting in a difference between the output digital signal and the original analog signal. The conversion function of an analog-to-digital converter (ADC) is not perfectly linear, introducing nonlinear distortion. This is due to factors such as the nonlinear characteristics of the ADC, the nonlinearity of the sample-and-hold circuit, and system bias, leading to distortion of the output digital signal. The sampling process of an ADC introduces sampling distortion. When the input signal frequency exceeds half the sampling frequency, a scrambling effect occurs, causing spectral distortion of the output signal.

[0085] In a digital beamforming architecture, the digital predistortion module is implemented after the beamforming operation; that is, each power amplifier is equipped with a dedicated predistorter to compensate for the distortion it generates. In this case, the signal received by the user end takes the following form:

[0086] R = H T G[F D (WD X)]

[0087] in Represents the transmitted signal matrix. This represents the received signal matrix from different user terminal directions, where K is the sample length and Q is the number of transmitted signal streams; W D H and H are the P×Q digital beamformer and channel matrix, respectively, representing the number of power amplifiers; G(·)=[g1(·),g2(·),...,g P (·)]and This represents the nonlinear transfer function of the power amplifier and predistorter in the array.

[0088] In digital predistortion operation, the inverse behavior model of the amplifier is identified and used as the predistorter, i.e. p = 1, 2, ..., P. Therefore, the cascaded module of the predistorter and power amplifier is considered a linear system, in which case the user will receive a linearized signal containing only information from the corresponding RF link.

[0089] For a hybrid beamforming millimeter-wave communication transmitter system with P antennas and Q independent RF chains (Q << P), each digital stream simultaneously drives multiple antennas and power amplifiers. After analog beamforming, the received signal is described as:

[0090] R = H T G[WF A (X)]

[0091] in This represents the nonlinear function of the predistorter. In the SA (subarray) architecture, each Q RF chain is independently connected to a set of P antennas and a power amplifier via N phase shifters, resulting in an overall analog beamformer W SA To form a block diagonal matrix:

[0092]

[0093] in q = 1, ..., Q. Therefore, the input signal of the power amplifier can be expressed as:

[0094]

[0095] in It is the Kronecker product. Let represent the input signal of each power amplifier. From the above formula, it can be seen that each power amplifier in the subarray hybrid beamforming system is driven only by the phase signal of the corresponding RF chain. Therefore, the received signal of the i-th user can be expressed as:

[0096]

[0097] in i = 1, 2, ..., Q represents a set of univariate nonlinear transfer functions. Ignoring crosstalk, the i-th beam of the SA hybrid beamforming array contains only the transmitted signal from the corresponding RF chain. A digital predistortion (DPD) model configured independently for each RF chain is equivalent to the inverse model of multiple univariate nonlinear models.

[0098] Any continuously differentiable nonlinear function can be expanded using Taylor series; therefore, Taylor series are often used to approximate nonlinear functions. In particular, the Volterra series model, which is very similar to Taylor series, is widely used in modeling the nonlinear behavior of power amplifiers. The difference between the Volterra series and Taylor series is that the Volterra series can capture memory effects, effectively modeling weakly nonlinear systems with memory. Its discretized bandpass expression is as follows:

[0099]

[0100] in Let M be the bandpass, real input signal and output signal, respectively, where M is the memory depth, P is the nonlinearity order, and h is the input signal and output signal. p (m1, ..., m) p ) is a p-order Volterra kernel.

[0101] In DPD processing, the signal is represented using a low-pass complex signal. Therefore, in practical applications, the bandpass Volterra series model above can be transformed into the following equivalent low-pass Volterra series model:

[0102]

[0103] In the above equation, x(n) and y(n) represent the low-pass and complex input and output signals, respectively.

[0104] The Memory Polynomial (MP) model is a simplified version of the classic Volterra series model. It adds the expression m1 = m2 = ... = m to the equivalent low-pass Volterra series model. p The constraints. The commonly used MP model expression now includes both odd-order and even-order distortion terms, as shown in the following equation:

[0105]

[0106] Step 3: Set up the receiving antenna in the direction of the main beam, and sample the transmitted radio frequency signal into a baseband digital signal through down-conversion and analog-to-digital converter, then input it into the digital predistortion module:

[0107] This step, based on the completion of the construction and nonlinear analysis of the hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture, involves setting up a receiving antenna in the main beam direction to receive the RF signals generated by the transmission system. Standard test signals, such as standard dual-tone signals, QPSK signals, and 16QAM signals, are transmitted from the signal source. These standard test signals undergo digital pre-coding via a digital pre-encoder, are split into several data streams, and input to the RF chain for up-conversion and analog beamforming. The beam signals are then transmitted through a power amplifier and a phased array antenna array. The specific structure of the RF chain is as follows... Figure 2 As shown. In actual experiments, the position of the receiving antenna is often fixed. The direction of the main beam is determined by the azimuth and elevation angles between the phased array antenna array and the receiving antenna.

[0108] The structure of the digital predistortion module is as follows: Figure 3 As shown, the predistorter in the digital predistortion module receives a pre-encoded digital signal as input when the system starts running. Since the DPD parameter estimation module has not yet received the feedback output signal, the predistorter does not process the signal at this time. The signal is input to the RF link and, affected by the nonlinearity of the DAC, up-conversion module, and power amplifier, is converted into an RF signal with nonlinear distortion after passing through the phase shifter network. The RF signal is transmitted in the MIMO channel, received by the receiving antenna, and then sequentially undergoes down-conversion and ADC sampling to be converted into a baseband digital signal fed back to the digital predistortion module. During this process, nonlinear interference from the down-converter and ADC is introduced again.

[0109] Step 4: Differentially process the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. Use an adaptive algorithm to continuously correct the acquired baseband digital signal to realize the function of the digital predistortion module.

[0110] In step 3, the system's transmitted signal has been down-converted and sampled into a baseband digital signal by an analog-to-digital converter. The baseband digital signal is then input into the DPD parameter estimation module of the digital predistortion module to extract nonlinear model parameter factors. After inversion, the parameter matrix of the predistorter is obtained. The specific operation is as follows: First, the nonlinear model parameters are extracted by differential processing of the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. Under the action of the adaptive LMS algorithm, the parameter matrix of the predistorter is continuously learned and adjusted until the difference between the standard test signal and the signal fed back from the power amplifier's output signal approaches 0. At this time, the system response corresponding to the parameter matrix of the predistorter is exactly the inverse effect of the nonlinear response of the entire system, thus obtaining a digital predistortion model suitable for non-ideal hardware conditions.

[0111] The LMS algorithm calculates the output error at each time step and then adjusts the predistorter coefficients based on this error. This update process is asymptotic, meaning that the LMS algorithm can gradually adjust the coefficients in small steps, allowing the predistorter to progressively approach the ideal linearization effect. This adaptive process enables the predistorter to dynamically compensate for nonlinear distortion as the operating environment changes.

[0112] The LMS algorithm simplifies gradient estimation in the steepest descent algorithm and is also known as the stochastic gradient algorithm. With a limited step size, the LMS algorithm converges, and its convergence speed is related to the step size factor. A larger step size factor speeds up convergence but may identify coefficients that are not optimal but rather suboptimal solutions surrounding the optimal solution; conversely, a smaller step size factor slows down convergence. Therefore, a carefully chosen step size factor is crucial when using the LMS algorithm to identify parameters.

[0113] The principle of the LMS algorithm is as follows: Figure 4 As shown, the LMS algorithm has a simple structure and low computational load. The operation only involves multiplication and addition / subtraction, without complex division operations. It has few intermediate parameters, and in the process of identifying predistorter parameters, there is only the estimation error, which is small. Therefore, the computational complexity is low, and it is a batch processing algorithm with strong hardware implementation.

[0114] The flowchart of the LMS algorithm is as follows: Figure 5 As shown. Specifically, for a given input signal x(n) and a desired output signal d(n), the LMS algorithm outputs the actual signal y(n) based on the current model parameters w(n), and then updates the next model parameters w(n+1) by comparing the difference between the actual output signal and the desired output signal. The update rule of the LMS algorithm is as follows:

[0115] w(n+1)=w(n)+μ·e(n)x(n)

[0116] e(n) = d(n) - y(n)

[0117] Where w(n) represents the model parameters at time n, μ is the learning rate (also called step size), which determines the amount of model parameter update in each iteration and affects the convergence speed of the algorithm, e(n) represents the degree to which the actual output signal y(n) deviates from the expected output signal d(n), and x(n) represents the current input signal. According to this update rule, the LMS algorithm continuously adjusts the model parameters so that the predicted output signal gradually approaches the expected output signal, thereby achieving the purpose of signal analysis and prediction.

[0118] When the system transmits a standard test signal under the action of beamforming, the digital predistortion model in the direction of the current main beam can be obtained by using the LMS algorithm.

[0119] Step 5: Within the maximum scanning range of the phased array antenna array, continuously adjust the angle to achieve digital predistortion under different beam directions:

[0120] Steps 3 and 4 are used to achieve digital predistortion in all main beam directions. Since the phased array antenna array can change the pointing angle of the main beam by modifying the feed phase of the phase shifter, it is also necessary to continuously adjust the pointing angle of the main beam within the maximum scanning range of the phased array antenna array to obtain digital predistortion models for all different angles.

[0121] The beam scanning range of a phased array antenna is determined by the azimuth angle φ and the scanning angle θ, which are complementary to the elevation angle. Its beam pointing is as follows: Figure 6 As shown. The azimuth angle φ ranges from 0 to 360°, and the scanning angle θ ranges from 0 to 60°. Each set of azimuth and scanning angles corresponds to a beam pointing, and each beam pointing corresponds to a digital predistortion model. While satisfying the beam pointing accuracy, all possible combinations of azimuth and scanning angles are traversed to obtain digital predistortion models for different angles.

[0122] Digital predistortion models trained at different angles were used to process the standard test signal at the transmitting end. For each digital predistortion model, a set of baseband digital signals was obtained at the receiving end through down-conversion and ADC sampling by continuously changing the beam direction. The error between this set of signals and the standard test signal was calculated, and the optimal digital predistortion model was selected as the final adaptive digital predistortion model under the criterion of minimizing the output signal error.

[0123] Example:

[0124] Simulink-based simulation was built as follows Figure 1The diagram illustrates a hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture. In the simulation, the azimuth angle φ is set to a range of 0-360°, and the scanning angle θ is set to a range of 0-60°. During simulation, a reasonable combination of azimuth and elevation angles is automatically generated to calculate beamforming-related parameters. The system first generates a sampled signal of modulated baseband digital or analog signal from the source module. This signal is then digitally pre-coded using a ZF pre-encoder. Next, the signal is split into four data streams and input to four beamforming lines. Figure 7 The RF link shown consists of a digital-to-analog converter, an IQ modulator, an up-converter, a power divider, a phase shifter, a power amplifier, and an antenna array. The digital-to-analog converter is as follows: Figure 8 As shown, the baseband digital signal is converted into a baseband analog signal by the upconverter, as shown. Figure 9 As shown, under the influence of local oscillation, the analog signal is first up-converted to an intermediate frequency by an IQ modulator, and then up-converted to the millimeter-wave band by an upconverter, thus becoming a radio frequency signal. The analog beamformer is as follows... Figure 10 As shown, the radio frequency signal is split into four paths by the power divider, simulated beamforming is performed by the phase shifter, and then transmitted through the power amplifier and the antenna array configured as a 2×2 uniform array to enter the MIMO channel for transmission.

[0125] In the RF link, the DAC, due to its precision limitations, introduces some distortion to the output signal, and the ADC also generates corresponding distortion. After passing through the DAC, the signal is split into I and Q paths for transmission, introducing IQ modulation distortion. Second-order and third-order nonlinear distortions are artificially introduced in the up-conversion and down-conversion modules. The nonlinear characteristics of the power amplifier are simulated based on a memory polynomial model; the simulated signal, when passing through the power amplifier, is affected by the amplifier's memory effect and intermodulation distortion caused by the amplifier's nonlinearity. The system's transmitted signal, affected by the nonlinearity of these non-ideal components, is received by the receiving antenna after transmission through the MIMO channel. It then passes sequentially through the down-conversion module and the ADC, combining into a single baseband digital signal that is fed back to the digital predistortion module.

[0126] The digital predistortion module is located between the digital precoder and the DAC, and its structure is as follows: Figure 11 As shown, the algorithm consists of two parts: a predistorter and a DPD parameter estimation section. The DPD parameter estimation section receives the combined feedback output signal, extracts the nonlinear parameters in the system using the least mean square algorithm, and calculates the minimum mean square error e(n) between the feedback output signal and the pre-encoded digital signal. The predistorter section predistorts the pre-encoded digital signal according to the nonlinear parameters and then inputs it into the RF chain. The flowchart of the LMS algorithm is shown below. Figure 5As shown. By dynamically adjusting the nonlinear parameters, when e(n) approaches 0, the baseband signal output by the predistorter and the feedback output signal after combination are linearly related, thus completing the linearization of the system's transmitted signal, that is, realizing the nonlinear distortion compensation of the system devices.

Claims

1. A digital predistortion method for millimeter-wave communication transmission systems with subarray architecture, characterized in that, The steps include the following: Step 1: Analyze the nonlinear effects of the system based on the nonideal characteristics of each component in the millimeter-wave communication transmission system; Step 2: Set up the receiving antenna in the direction of the main beam, and sample the transmitted radio frequency signal into a baseband digital signal through down-conversion and analog-to-digital converter, then input it into the digital predistortion module: Step 3: Differential processing is performed on the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. An adaptive algorithm is used to continuously correct the acquired baseband digital signal, thereby realizing digital predistortion in the main beam direction. Step 4: Within the maximum scanning range of the phased array antenna array, continuously adjust the angle to achieve digital predistortion under different beam directions; The specific method for step 1 is as follows: A hybrid beamforming millimeter-wave communication transmission system based on a subarray architecture consists of a signal source, a digital precoder, a digital-to-analog converter (DAC), an up-conversion converter, an analog beamformer, an antenna array, a MIMO channel, a down-conversion converter, and an analog-to-digital converter (ADC). A digital predistortion module is designed to compensate for nonlinear distortion in the system. The DAC, up-conversion converter, and analog beamformer form the RF chain. The analog beamformer consists of a phase shifter and a power amplifier. The digital predistortion module includes a predistorter (DPD) and DPD parameter estimation. In a millimeter-wave communication transmission system based on a subarray architecture, nonlinear interference caused by the non-ideal characteristics of the DAC, upconversion module, power amplifier, downconversion module, and ADC is considered. The resolution of the DAC determines the accuracy of the converted signal, and distortion will inevitably occur during the signal conversion process; The upconversion module mainly consists of two modules: an IQ modulator and an upconverter. Incomplete balance between the I and Q transmission links can lead to amplitude and phase imbalance. For broadband and high-frequency applications, power amplifiers introduce nonlinear distortion; when a power amplifier is operating, the relationship between its input and output is not linear, which causes nonlinear distortion in the output signal. The downconversion module consists of a downconverter and an IQ demodulator. The nonlinear distortion of this module is similar to that of the upconversion module, with amplitude and phase imbalance caused by IQ imbalance. Quantization error also exists when an ADC discretizes an analog signal into a digital signal; In a digital beamforming architecture, the digital predistortion module is implemented after the beamforming operation; that is, each power amplifier is equipped with a dedicated predistorter to compensate for the distortion it generates. In this case, the signal received by the user end takes the following form: in Represents the transmitted signal matrix. This represents the received signal matrix from different user terminal directions. It is the sample length. It is the number of transmitted signal streams; and They are The digital beamformer and channel matrix represent the number of power amplifiers; and This represents the nonlinear transfer function of the power amplifier and predistorter in the array; In digital predistortion operation, the inverse behavior model of the amplifier is identified and used as the predistorter, i.e. Therefore, the cascaded module of the predistorter and power amplifier is considered as a linear system. In this case, the user will receive a linearized signal containing only information from the corresponding RF link. For a having One antenna and An independent radio frequency chain ( A hybrid beamforming millimeter-wave communication transmission system, where each digital stream simultaneously drives multiple antennas and power amplifiers; after analog beamforming, the received signal is described as: in This represents the nonlinear function of the predistorter; in the SA (subarray) architecture, each Each radio frequency chain passes independently A phase shifter is connected to a set of Each antenna and power amplifier contributes to the overall analog beamformer. To form a block diagonal matrix: in Therefore, the input signal of the power amplifier can be expressed as: in It is the Kronecker product. This represents the input signal of each power amplifier; it can be seen from the above formula that each power amplifier in the subarray hybrid beamforming system is driven only by the phase signal of the corresponding RF chain. Therefore, the first... The received signal of a user can be represented as: in Represents a set of univariate nonlinear transfer functions; neglecting crosstalk, the first... Each beam contains only the transmitted signal from the corresponding radio frequency chain; the digital predistortion (DPD) model configured independently for each radio frequency chain is equivalent to the inverse model of multiple univariate nonlinear models; In modeling the nonlinear behavior of the power amplifier, the Volterra series model is used, and its discretized bandpass expression is as follows: in , Let M be the bandpass, real input signal and output signal, respectively, where M is the memory depth and P is the nonlinearity order. It is a p-order Volterra kernel; In DPD processing, the signal is represented using a low-pass complex signal. Therefore, in practical applications, the bandpass Volterra series model above can be transformed into the following equivalent low-pass Volterra series model: The above formula , These represent low-pass and complex input and output signals, respectively. Since the Volterra series model is only suitable for modeling weakly nonlinear systems and has high model complexity, the memory polynomial model is used instead of the equivalent low-pass Volterra series model; the memory polynomial model is as follows: 。 2. The digital predistortion method for a millimeter-wave communication transmission system with a subarray architecture according to claim 1, characterized in that, Step 2 is explained in the following steps: A receiving antenna is set up in the direction of the main beam to receive the radio frequency signal generated by the transmitting system; a standard test signal is transmitted by the signal source; the standard test signal is digitally pre-encoded by a digital pre-encoder, divided into several data streams and input into the RFchain for up-conversion and analog beamforming, and then the beam signal is transmitted through a power amplifier and a phased array antenna array. The direction of the main beam is determined by the azimuth and elevation angles between the phased array antenna array and the receiving antenna; When the system starts running, the predistorter of the digital predistortion module receives a pre-encoded digital signal as input. Since the DPD parameter estimation module has not yet received the feedback output signal at this time, the predistorter does not process the signal. The signal is input to the RF link and is affected by the nonlinearity of the DAC, upconversion module and power amplifier. After passing through the phase shifter network, it is converted into an RF signal with nonlinear distortion. Radio frequency signals are transmitted in the MIMO channel. After being received by the receiving antenna, they are successively processed by downconversion and sampled by ADC, and then converted into baseband digital signals that are fed back to the digital predistortion module. In this process, nonlinear interference from the downconverter and ADC is introduced.

3. The digital predistortion method for a millimeter-wave communication transmission system with a subarray architecture according to claim 2, characterized in that, The standard test signal used includes, but is not limited to, standard dual-tone signals, QPSK signals, or 16QAM signals.

4. The digital predistortion method for a millimeter-wave communication transmission system with a subarray architecture according to claim 2, characterized in that, The specific method for step 3 is as follows: In step 2, the system's transmitted signal has been down-converted and sampled into a baseband digital signal by an analog-to-digital converter. The baseband digital signal is then input into the DPD parameter estimation module in the digital predistortion module to extract the nonlinear model parameters. After inversion, the parameter matrix of the predistorter is obtained. The specific operation is as follows: First, the nonlinear model parameters are extracted by differential processing of the signal fed back from the analog-to-digital converter and the signal input to the RF chain after digital precoding. Under the action of the adaptive LMS algorithm, the parameter matrix of the predistorter is continuously learned and adjusted until the difference between the standard test signal and the signal fed back from the power amplifier's output signal approaches 0. At this time, the system response corresponding to the parameter matrix of the predistorter is the inverse effect of the nonlinear response of the entire system, thus obtaining a digital predistortion model suitable for non-ideal hardware conditions. The LMS algorithm calculates the output error at each time step and then adjusts the coefficients of the predistorter based on this error. This update process is gradual, meaning that the LMS algorithm can gradually approach the ideal linearization effect by adjusting the coefficients in small steps. Specifically, given an input signal x(n) and a desired output signal d(n), the LMS algorithm outputs the actual signal y(n) based on the current model parameters w(n), and then updates the next model parameters w(n+1) by comparing the difference between the actual output signal and the desired output signal. The update rule of the LMS algorithm is as follows: Where w(n) represents the model parameters at time n. Here, e(n) represents the degree to which the actual output signal y(n) deviates from the desired output signal d(n), and x(n) represents the current input signal. According to this update rule, the LMS algorithm will continuously adjust the model parameters so that the predicted output signal gradually approaches the desired output signal, thereby achieving the purpose of signal analysis and prediction. When the system transmits a standard test signal under the action of beamforming, the digital predistortion model in the direction of the current main beam can be obtained by using the LMS algorithm.

5. The digital predistortion method for a millimeter-wave communication transmission system with a subarray architecture according to claim 4, characterized in that, Step 4 is explained in detail below: The operations in steps 2 and 3 are used to achieve digital predistortion in all main beam directions. Since the phased array antenna array can change the pointing angle of the main beam by modifying the feed phase of the phase shifter, it is also necessary to continuously adjust the pointing angle of the main beam within the maximum scanning range of the phased array antenna array to obtain digital predistortion models for all different angles. The beam scanning range of a phased array antenna array is determined by the azimuth angle. and scanning angle Confirmed, the scan angle and elevation angle are complementary; azimuth angle... The value range is 0-360. Scan angle The value range is 0-60 Each set of azimuth and scanning angles corresponds to a beam pointing, and each beam pointing corresponds to a digital predistortion model. Under the condition of satisfying the beam pointing accuracy, all possible combinations of azimuth and scanning angles are traversed to obtain digital predistortion models for different angles. The digital predistortion models trained at different angles are used to process the standard test signal at the transmitting end. For each digital predistortion model, by continuously changing the beam direction, a set of baseband digital signals are obtained at the receiving end through downconversion and ADC sampling. The error between this set of signals and the standard test signal is calculated. Under the criterion of minimizing the output signal error, the optimal digital predistortion model is selected as the final adaptive digital predistortion model.

Images