Millimeter wave beam forming chip phase noise suppression method and system

Abstract

The invention relates to the technical field of communication, discloses a millimeter wave beam forming chip phase noise suppression method and system, and aims to solve the problems of high cost, high power consumption, remarkable processing delay and poor environmental adaptability of a phase noise suppression scheme in the prior art. The method comprises the following steps: coupling an output signal of a local oscillator in real time and digitizing to obtain a monitoring signal containing phase noise; an instantaneous phase error sequence is extracted through digital orthogonal demodulation and a CORDIC algorithm; and performing dynamic prediction on the phase noise by using the LMS or RLS adaptive prediction model. Through a feed-forward digital compensation mechanism, real-time prediction and offset of phase noise are realized in a baseband domain, the beam forming precision and the system robustness are improved, low-cost oscillator application is supported, the overall power consumption and cost are reduced, and meanwhile, the stability and the anti-interference capability of a communication link are improved.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a method and system for suppressing phase noise in millimeter-wave beamforming chips. Background Technology

[0002] With the rapid evolution of modern wireless communication technology, millimeter-wave communication has become a core supporting technology for achieving ultra-high bandwidth and massive connectivity in fifth-generation and even future sixth-generation mobile communication systems. To overcome the severe path loss in the millimeter-wave band, beamforming technology has been widely used. Among these technologies, the performance of millimeter-wave beamforming chips, as key core components, directly determines the signal quality and connection stability of the entire communication link.

[0003] Existing phase noise suppression schemes for millimeter-wave beamforming chips generally have limitations. On the one hand, hardware-level solutions often employ high-quality oscillators to reduce phase noise at its source, but this significantly increases chip manufacturing costs, area, and power consumption, making it unsuitable for large-scale integration and low-cost applications. On the other hand, software or algorithm-level compensation techniques, such as digital phase-locked loops or post-processing algorithms, while flexible, typically have high computational complexity, introducing significant signal processing delays that fail to meet the stringent low-latency requirements of millimeter-wave communication. Furthermore, existing solutions often struggle to effectively address the dynamic degradation of phase noise caused by external factors such as temperature changes and power fluctuations, resulting in insufficient robustness of the system in complex operating environments. These combined issues directly lead to beam pattern distortion and increased sidelobe levels, severely impacting communication coverage, anti-interference capabilities, and data transmission accuracy.

[0004] Therefore, given the inherent contradictions between cost, power consumption, processing latency, and environmental robustness in existing phase noise suppression technologies, there is an urgent need to propose a novel phase noise suppression method and system for millimeter-wave beamforming chips. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method and system for suppressing phase noise in millimeter-wave beamforming chips. This invention aims to solve the problem that in millimeter-wave communication systems, the phase noise of the local oscillator is directly modulated onto the radio frequency signal, causing random deterioration of the phase relationship between various antenna channels, which in turn seriously affects the accuracy of the beamforming pattern and the signal quality of the communication link. Existing technical solutions either rely on expensive and power-consuming high-performance phase-locked loops to reduce phase noise at the source, or the digital compensation methods used have problems such as large processing delays, limited compensation accuracy, and inability to adapt to dynamic changes in the working environment.

[0006] To achieve the above objectives, the present invention provides the following technical solution: In one aspect, a millimeter-wave beamforming chip phase noise suppression system, comprising the following components: The local oscillator signal monitoring module is configured to couple a small portion of the signal from the local oscillator output of the millimeter-wave beamforming chip and process the coupled signal to acquire a digital monitoring signal representing the phase noise characteristics of the local oscillator in real time. The phase noise feature extraction module is electrically connected to the local oscillator signal monitoring module. It is used to receive digital monitoring signals and demodulate and analyze the digital monitoring signals using digital signal processing algorithms, thereby accurately extracting instantaneous phase error sequence data. The phase noise prediction and compensation module, which is connected to the phase noise feature extraction module, is used to receive instantaneous phase error sequence data and, based on an adaptive prediction model, make a forward prediction of the phase noise for one or more future sampling periods, generating a digital compensation signal that is equal in magnitude but opposite in phase to the predicted phase noise. The multi-channel signal compensation application module is located between the digital baseband processing unit and the digital-to-analog converters of each antenna channel. It is connected to the phase noise prediction and compensation module to receive digital compensation signals and apply these signals synchronously and independently to the digital baseband signals leading to each antenna channel, thereby achieving pre-distortion bit correction of the baseband signal. The RF upconversion and transmission module contains multiple parallel antenna channels. Each channel's mixer receives a signal containing phase noise from the local oscillator as an upconversion signal source. This signal is used to upconvert the phase-corrected baseband signal to the millimeter-wave band and transmit it via the antenna. Pre-distortion phase correction and phase noise during the upconversion process cancel each other out, thereby outputting a phase-clean RF signal.

[0007] Preferably, the local oscillator signal monitoring module includes a directional coupler, a low-noise mixer, and an analog-to-digital converter; the directional coupler is used to losslessly couple the monitoring signal from the main output path of the local oscillator; the low-noise mixer mixes and down-converts the monitoring signal with a high-stability reference frequency signal to obtain an intermediate frequency signal containing phase noise information; the analog-to-digital converter is responsible for sampling and quantizing the intermediate frequency signal to convert it into a digital monitoring signal for subsequent digital domain processing.

[0008] Furthermore, the phase noise feature extraction module integrates a digital quadrature demodulation unit and a phase calculation unit. The digital quadrature demodulation unit decomposes the input digital monitoring signal into in-phase and quadrature components. The phase calculation unit uses a coordinate rotation digital calculation algorithm to calculate the instantaneous phase of the signal in real time based on the I / Q components. By comparing it with an ideal phase reference, it obtains instantaneous phase error sequence data, which accurately characterizes the change of the phase noise of the local oscillator over time.

[0009] Preferably, the adaptive prediction model in the phase noise prediction and compensation module is a least mean square adaptive filter or a recursive least squares adaptive filter. This adaptive filter takes historical phase error sequence data as input and iteratively updates its internal filter coefficients to learn and track the statistical characteristics and dynamic changes of phase noise, thereby making a high-precision prediction of future phase noise values. The prediction time margin is set to be greater than or equal to the signal path delay from the compensation application point to the RF mixer to ensure the real-time effectiveness of the compensation.

[0010] In addition, the multi-channel signal compensation application module consists of a set of digital complex multipliers, with each antenna channel corresponding to an independent digital complex multiplier. The digital compensation signal generated by the phase noise prediction and compensation module is first converted into a complex rotation factor with an amplitude of 1 and a phase value equal to the phase value of the compensation signal. The digital complex multiplier multiplies this rotation factor with the original digital baseband complex signal of the corresponding channel. This operation is equivalent to applying a precise phase shift to the baseband signal in the complex domain, thereby completing the pre-distortion bit correction.

[0011] Furthermore, the system also includes a system calibration and control unit, which is used to calibrate and compensate for the amplitude and phase imbalance of the local oscillator signal monitoring module, the sampling clock jitter of the analog-to-digital converter, and the delay differences of each channel signal path during system initialization or operating mode switching. The system calibration and control unit is also responsible for dynamically adjusting the key parameters of the adaptive prediction model, such as the filter order, step size factor, or forgetting factor, according to the current operating frequency, temperature, and communication mode, in order to maintain optimal phase noise suppression performance under different operating conditions.

[0012] Preferably, the phase noise prediction and compensation module also introduces a modeling and compensation mechanism for nonlinear noise components when generating the digital compensation signal. This module identifies and quantifies the phase noise components introduced by the nonlinear effects of the circuit by analyzing the higher-order statistical characteristics of the phase error sequence data, and incorporates them into the prediction model to generate a composite digital compensation signal containing linear and nonlinear compensation components, thereby improving the ability to suppress complex phase noise sources.

[0013] On the other hand, a method for suppressing phase noise in a millimeter-wave beamforming chip includes the following specific steps: Step S110: The output signal of the local oscillator is coupled in real time through the local oscillator signal monitoring module, and down-converted and digitized to obtain a digital monitoring signal containing phase noise information. Step S120: In the phase noise feature extraction module, the digital monitoring signal is digitally quadrature demodulated, decomposed into I / Q signals, and the instantaneous phase is calculated using a phase calculation algorithm. By comparing with an ideal phase reference, a high-resolution instantaneous phase error sequence is extracted. Step S130: The instantaneous phase error sequence is sent to the phase noise prediction and compensation module. An adaptive prediction model is used to process the sequence, learn its dynamic characteristics, and predict the phase noise value at future times. Step S140: Based on the predicted phase noise value, generate a digital compensation signal with opposite phase and normalized amplitude, which is represented in the form of a complex rotation factor; Step S150: In the multi-channel signal compensation application module, the complex rotation factor is multiplied one by one with the digital baseband signal leading to each antenna channel, thereby applying a pre-distortion bit to the baseband signal that is opposite to the prediction noise. In step S160, the phase-compensated digital baseband signal is processed by the digital-to-analog converters and RF upconversion and transmission modules of each channel. The upconversion is completed using the local oscillator signal containing phase noise. Finally, the transmitted RF signal exhibits a significantly reduced phase noise level due to the mutual cancellation between the pre-compensated phase and the actual noise phase.

[0014] Compared with the prior art, the present invention has the following beneficial effects: Through forward-looking prediction and feedforward compensation mechanisms, this invention can accurately cancel phase noise in the digital baseband domain before it affects the radio frequency signal, overcoming the inherent delay and stability problems of traditional feedback loops, and achieving real-time and efficient suppression of broadband phase noise.

[0015] This invention realizes phase noise monitoring, processing and compensation entirely digitally within the chip. It has a compact structure, is easy to integrate, and can dynamically adjust through an adaptive algorithm to cope with the drift of phase noise characteristics under different working environments (such as temperature changes and power fluctuations), thus having strong robustness and environmental adaptability.

[0016] Because this invention can effectively suppress phase noise, it allows system designs to use lower-cost, lower-power local oscillators (such as integrated voltage-controlled oscillators) while still achieving the phase noise performance required by high-performance communication systems, thereby significantly reducing the overall cost and power consumption of the millimeter-wave beamforming chip and even the entire system.

[0017] This invention performs independent phase compensation for each antenna channel, which can accurately correct the slight phase inconsistencies caused by physical differences between channels. It not only suppresses common-mode phase noise but also improves differential-mode phase noise, thereby improving the pointing accuracy and sidelobe suppression ratio of the multi-antenna beamforming pattern, and directly enhancing the beam gain and spatial selectivity of the communication link. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical architecture of the millimeter-wave beamforming chip phase noise suppression system proposed in this invention; Figure 2 This is a logical flowchart of the phase noise suppression method for millimeter-wave beamforming chips proposed in this invention. Detailed Implementation

[0019] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0020] Example 1 Please see Figure 1 , Figure 1 This is a schematic diagram of the overall technical architecture of a millimeter-wave beamforming chip phase noise suppression system according to the present invention. In this embodiment, the system is integrated into a 64-channel phased array transceiver chip operating in the 28 GHz band, aiming to provide high-purity beamforming signals for high-data-rate 5G mobile communication base stations. The system specifically includes a local oscillator signal monitoring module, a phase noise feature extraction module, a phase noise prediction and compensation module, a multi-channel signal compensation application module, and an RF up-conversion and transmission module. Furthermore, the system also includes a system calibration and control unit for calibrating and dynamically optimizing the performance of the entire system.

[0021] The local oscillator (LO) signal monitoring module's function is to capture the phase noise information of the local oscillator (LO) in real time and without loss, and convert it into a digital signal. The core of this module is a directional coupler located on the LO's main signal output path, with its coupling degree precisely designed to -20 dB to ensure that while coupling out a monitoring signal with sufficient power, the impact on the power and phase of the main path signal is minimized. The coupled monitoring signal has a power of approximately 0 dBm and is first conditioned by a low-noise amplifier before being fed into a low-noise mixer. The other input of this mixer is connected to a reference frequency signal source driven by an off-chip high-stability temperature-compensated crystal oscillator (TCXO) with a frequency of 27.9 GHz and a frequency stability of 0.5 parts per million. The mixer down-converts the 28 GHz monitoring signal with the 27.9 GHz reference signal, generating an intermediate frequency (IF) signal with a center frequency of 100 MHz. This IF signal completely preserves the phase noise spectral envelope of the original LO signal. The intermediate frequency (IF) signal is then filtered through a bandpass filter to remove out-of-band noise and spurious components before being fed into a high-speed analog-to-digital converter (ADC). This ADC is set to a sampling rate of 400 MHz and a quantization bit depth of 14 bits, ensuring oversampling and high dynamic range digitization of the IF signal. This converts the analog IF signal, which contains phase noise information, into a high-precision digital monitoring signal stream, which is output to subsequent modules at a rate of 400 million 14-bit samples per second.

[0022] The phase noise feature extraction module is directly connected to the output of the local oscillator signal monitoring module. Its core task is to accurately separate the instantaneous phase error from the digital monitoring signal. Internally, this module implements a digital quadrature demodulation unit. This unit contains a numerically controlled oscillator (NCO) to generate two phase-orthogonal digital carrier signals, namely sine and cosine sequences, with their frequencies precisely set to 100 MHz, consistent with the center frequency of the intermediate frequency signal. The input digital monitoring signal is simultaneously multiplied digitally by these two digital carriers. The resulting two signals are then passed through a 32nd-order finite impulse response (FIR) low-pass filter to remove high-frequency mixing products. The filter cutoff frequency is set to 50 MHz. Through this process, the original digital monitoring signal is successfully decomposed into in-phase (I) and quadrature (Q) components, forming a complex signal sequence. Subsequently, this I / Q data stream is fed into the phase resolution unit. This unit employs a coordinate rotation digital computation (CORDIC) algorithm implemented based on a hardware pipeline architecture. In each clock cycle, the phase calculation unit receives a pair of I / Q data and calculates the instantaneous phase angle of the complex sample through 14 iterations of micro-rotation operations. The calculated instantaneous phase sequence is then subtracted point-by-point from a linearly increasing ideal phase reference sequence generated at an ideal 100 MHz frequency. The result of this subtraction operation is the instantaneous phase error sequence data Δθ(n). This sequence data has a resolution of 0.01 degrees, accurately and in real-time reflecting the random fluctuations of the local oscillator phase noise over time.

[0023] The phase noise prediction and compensation module is the core intelligent unit of this system. It receives the instantaneous phase error sequence data Δθ(n) from the phase noise feature extraction module and performs forward prediction on it. In this embodiment, the module internally implements a 128th-order least mean square (LMS) adaptive filter as its adaptive prediction model. The LMS filter consists of a 128-stage tapped delay line, 128 adjustable filter coefficients (weights), and an error calculation and coefficient update unit. The historical phase error sequence data Δθ(n-1), Δθ(n-2), ..., Δθ(n-128) serve as the input vector of the filter. At the nth time step, the filter uses the current coefficient vector W(n) to linearly combine the input vector to generate an estimate of the current phase error. This estimate is compared with the actual input phase error Δθ(n) to obtain the prediction error e(n). Subsequently, the coefficient update unit updates the filter coefficients according to the iterative rules of the LMS algorithm. The update process can be described by the following formula: W(n+1) = W(n) + μ*e(n)*X(n) Where W(n+1) is the updated coefficient vector, W(n) is the current coefficient vector, μ is the step size factor, e(n) is the prediction error, and X(n) is the input vector containing historical phase error data. The step size factor μ is set by the system calibration and control unit to a value between 0.001 and 0.05 based on the current operating state to achieve a balance between convergence speed and steady-state error. Through thousands of iterations, the LMS filter learns the autocorrelation characteristics and statistical regularities of the phase noise. Once the filter converges, it can use the current input to predict the phase noise value for one or more future sampling periods. The prediction time margin, i.e., the predicted future time point, is precisely set to the total signal path delay from the multi-channel signal compensation application module to the mixer in the RF upconversion and transmit module, which is approximately 5 nanoseconds in this chip design. Therefore, the module predicts the phase noise value at time Δθ(n+2) (5 nanoseconds for 2 sampling periods at a 400 MHz sampling rate). Finally, the module generates a digital compensation signal that is equal in magnitude but opposite in phase to the predicted phase noise, namely -Δθ_pred(n+2).

[0024] The multi-channel signal compensation application module is deployed between the digital baseband processing unit and the digital-to-analog converters (DACs) of 64 antenna channels. This module consists of an array of 64 parallel, identically structured digital complex multipliers, one for each antenna channel. The scalar digital compensation signal -Δθ_pred(n+2) generated by the phase noise prediction and compensation module is first fed into an arctangent lookup table (LUT) to convert it into a complex rotation factor C(n) of the form exp(-jΔθ_pred(n+2)). The amplitude of this rotation factor is always 1, and its phase is the phase value of the compensation signal. This complex rotation factor is broadcast to all 64 digital complex multipliers. Simultaneously, 64 independent raw digital baseband complex signals S_k(n) = I_k(n) + jQ_k(n) (where k = 1, 2, ..., 64) from the digital baseband processing unit are input to their respective digital complex multipliers. Inside the multiplier, the input signal is multiplied by a rotation factor: S'_k(n) = S_k(n) * C(n). This multiplication is mathematically equivalent to rotating the phase of the original baseband signal by an angle -Δθ_pred(n+2). This applies a precise, synchronous pre-distortion phase correction to the baseband signals of all channels. The corrected 64-channel digital baseband signal S'_k(n) is then sent to its respective digital-to-analog converter.

[0025] The RF upconversion and transmit module comprises 64 parallel antenna channels. Each channel includes a digital-to-analog converter (DAC), a low-pass reconstruction filter, a mixer for upconversion, a power amplifier, and an antenna element. The digital baseband signal S'_k(n), after pre-distortion phase correction, is converted into an analog signal in the DAC, smoothed by the filter, and then input to the mixer. The other input of the mixer receives a 28 GHz upconverted signal source from the local oscillator, containing actual phase noise Δθ_lo(t). During upconversion, the phase of the baseband signal is superimposed on the phase of the local oscillator signal. Since the pre-compensated baseband signal phase is -Δθ_pred(t), and the phase noise of the local oscillator signal is Δθ_lo(t), and the prediction model ensures that Δθ_pred(t) is highly time-matched with Δθ_lo(t), the two precisely cancel each other out during the mixing process. Ultimately, the phase noise of the millimeter-wave radio frequency signal emitted from the antenna is significantly suppressed, exhibiting extremely high phase purity, thereby ensuring the stability and accuracy of the beamforming pattern.

[0026] The system calibration and control unit, acting as a background processor, plays a crucial role during system startup or operating mode switching. During initialization, it executes a calibration procedure by injecting a known single-tone test signal to measure and calibrate the amplitude and phase imbalance of the I / Q channels in the local oscillator signal monitoring module, generating correction coefficients for the digital quadrature demodulation unit. Simultaneously, it measures and records the precise signal path delay from the digital compensation point to the mixer for each antenna channel, providing the average delay value to the phase noise prediction and compensation module as a reference for prediction time margin. During system operation, this unit continuously monitors the chip's temperature sensor and operating frequency settings. Based on a multidimensional lookup table pre-stored in non-volatile memory, this unit can dynamically adjust the step factor μ of the LMS adaptive filter according to the current temperature and frequency. For example, when rising temperature exacerbates changes in phase noise characteristics, μ is appropriately increased to accelerate the filter's tracking speed; under stable operating conditions, μ is decreased to reduce steady-state error, thus maintaining optimal phase noise suppression performance under various operating conditions.

[0027] Please see Figure 2 , Figure 2 This is a logical flowchart of the phase noise suppression method for millimeter-wave beamforming chips according to the present invention. Based on the above system structure, the specific implementation steps of the method of the present invention are as follows: Step S110: Real-time monitoring and digitization of the local oscillator signal. After system startup, the local oscillator signal monitoring module operates continuously. Its internal directional coupler couples a -20 dB signal from the 28 GHz local oscillator main output path. This signal is mixed with a 27.9 GHz reference signal in a low-noise mixer to generate a 100 MHz intermediate frequency (IF) signal. Subsequently, an analog-to-digital converter samples and quantizes this IF signal at a rate of 400 MHz and a precision of 14 bits, generating a continuous digital monitoring signal data stream.

[0028] Step S120: Extract the instantaneous phase error sequence. The phase noise feature extraction module receives the digital data stream. The digital quadrature demodulation unit uses a 100 MHz quadrature digital carrier generated by an internal numerically controlled oscillator to demodulate the input signal into two baseband signals, I and Q. Then, the phase calculation unit applies the CORDIC algorithm to calculate the instantaneous phase angle of the I / Q signals point by point. By comparing this instantaneous phase sequence with an ideal linear phase ramp, the instantaneous phase error sequence Δθ(n), representing the phase noise of the local oscillator, is accurately extracted.

[0029] Step S130: Adaptive prediction of future phase noise. The phase error sequence Δθ(n) is fed into the phase noise prediction and compensation module in real time. The 128th-order LMS adaptive filter within the module uses this sequence as training data. Through continuous iteration, the filter adjusts its 128 internal weight coefficients to minimize the mean square value of the prediction error, thereby learning and establishing a prediction model that can reflect the dynamic characteristics of phase noise.

[0030] Step S140: Generate a digital compensation signal. After the LMS filter converges, for each new phase error sample input, the model prospectively predicts the phase noise value Δθ_pred(n+2) 5 nanoseconds in advance. Based on this prediction, the module generates a digital compensation signal -Δθ_pred(n+2) with opposite phase and converts it into a complex rotation factor C(n) = exp(-jΔθ_pred(n+2)) with an amplitude of 1 and a phase equal to the compensation value.

[0031] Step S150: Perform pre-distortion bit correction on the multi-channel baseband signal. In the multi-channel signal compensation application module, the generated complex twitch factor C(n) is synchronously broadcast to all 64 digital complex multipliers. Each multiplier performs a complex multiplication of the twitch factor with the original digital baseband signal S_k(n) of the corresponding channel. This operation uniformly applies a precise pre-distortion bit to the baseband signal of all channels, opposite to the phase noise to be encountered.

[0032] Step S160: Up-conversion transmission and phase noise cancellation. The 64 phase-compensated digital baseband signals S'_k(n) pass through their respective digital-to-analog converters and filters before entering the RF up-conversion and transmission module. In the mixer, these signals are up-converted with the local oscillator signal containing the actual phase noise Δθ_lo(t). Due to the precise matching of signal path delays and the accuracy of prediction, the pre-compensated phase -Δθ_pred(t) carried by the baseband signal and the phase noise Δθ_lo(t) of the local oscillator signal are precisely aligned in the time domain and cancel each other out. Finally, the synthesized beam transmitted from the 64 antennas has extremely low phase noise, ensuring the high quality and stability of the communication link.

[0033] Example 2 This embodiment provides a millimeter-wave beamforming chip phase noise suppression system that is similar to Embodiment 1 in overall system architecture but employs a different technical approach in its core prediction algorithm. The main difference between Embodiment 1 and Embodiment 2 is that the phase noise prediction and compensation module in this embodiment uses a recursive least squares (RLS) adaptive filter as its adaptive prediction model to cope with specific application scenarios where the statistical characteristics of phase noise change rapidly and non-stationarily, such as in industrial environments with strong electromagnetic interference or drastic power supply voltage fluctuations.

[0034] In this embodiment, the RLS adaptive filter implemented within the phase noise prediction and compensation module is also set to order 128. Unlike the LMS algorithm, which only uses the current error to adjust the coefficients, the RLS algorithm uses information from all historical data in each iteration to update the filter coefficients recursively, minimizing a weighted least squares cost function. The core of this algorithm is to recursively calculate the inverse matrix of a 128x128 dimensional input data autocorrelation matrix. To reduce computational complexity and enable it to track time-varying characteristics, the algorithm introduces a forgetting factor λ. The forgetting factor λ is a positive number close to 1, such as 0.999, which exponentially decays the weight of historical data, giving more weight to recent input data in parameter estimation.

[0035] The coefficient update process of an RLS filter is more complex, mainly involving the following micro-steps: Calculate the gain vector k(n): This vector determines the extent to which the current prediction error affects the coefficient update. Its calculation depends on the inverse autocorrelation matrix P(n-1) from the previous time step and the current input vector X(n).

[0036] Calculate the prior prediction error α(n): Filter the current input X(n) using the coefficient vector W(n-1) from the previous time step to obtain the predicted value, and compare it with the actual phase error Δθ(n) to obtain the prior error.

[0037] Update the filter coefficient vector W(n): Update the coefficient vector using the gain vector k(n) and the prior prediction error α(n): W(n) = W(n-1) + k(n) * α(n).

[0038] Update the inverse autocorrelation matrix P(n): Based on the new gain vector k(n) and input vector X(n), the inverse autocorrelation matrix is ​​recursively updated using the matrix inversion lemma to prepare for the next iteration.

[0039] Compared to the LMS algorithm in Example 1, the RLS algorithm used in this example has a significantly faster convergence speed. During system startup or sudden changes in the operating environment, the RLS filter only requires a few hundred iterations to converge to the optimal coefficient solution, while the LMS filter requires thousands or even tens of thousands of iterations. This fast convergence characteristic allows the system to quickly adapt to changes in phase noise statistics, achieving near-instantaneous optimal suppression performance. However, the computational complexity of the RLS algorithm is much higher than that of the LMS algorithm. The computational cost of each iteration of the LMS algorithm is proportional to the filter order N, i.e., O(N), while the computational cost of the standard RLS algorithm is proportional to the square of N, i.e., O(N^2). For a 128th-order filter, the hardware logic gate resources and power consumption required by the RLS algorithm increase significantly. Therefore, this example is suitable for high-performance applications with extremely high dynamic response speed requirements and relatively relaxed budgets for chip area and power consumption.

[0040] In this embodiment, the function of the system calibration and control unit has also been adjusted accordingly. Instead of adjusting the step size factor μ, it is responsible for dynamically adjusting the forgetting factor λ of the RLS filter. When the system detects drastic environmental changes (e.g., by monitoring power supply ripple or temperature change rate), it appropriately reduces the value of λ (e.g., from 0.999 to 0.995) to reduce reliance on historical data and enhance sensitivity to new data, thereby accelerating the filter's tracking speed. When the system enters a stable operating state, it restores λ to a value close to 1 to obtain smoother and more accurate coefficient estimates, reducing steady-state error. This adaptive adjustment of the forgetting factor allows the RLS scheme to maintain its fast convergence advantage while also considering steady-state performance under different operating conditions.

[0041] The execution flow of the method is largely the same as that of the embodiment, with the main difference being in step S130. In step S130, the phase error sequence fed into the phase noise prediction and compensation module is processed by the RLS adaptive filter. The filter efficiently utilizes the input data and quickly adjusts its 128 weight coefficients by recursively calculating the inverse autocorrelation matrix and updating the gain vector to minimize the weighted sum of squared errors, thereby rapidly establishing a prediction model that can accurately capture the non-stationary dynamic characteristics of phase noise. The remaining steps, such as signal monitoring, phase extraction, compensation signal generation and application, and the final RF cancellation, are consistent with the principles and implementation methods of embodiment one.

[0042] Example 3 This embodiment, based on the linear prediction framework of Embodiment 1 or Embodiment 2, further introduces a modeling and compensation mechanism for nonlinear noise components, aiming to solve the complex phase noise components introduced by circuit nonlinear effects (such as the nonlinearity of the tuning curve of the voltage-controlled oscillator (VCO), and the modulation of power supply noise to the output phase through a nonlinear path). These nonlinear components cannot be completely captured and eliminated by linear prediction models such as LMS or RLS; therefore, this embodiment provides a more refined and comprehensive compensation scheme.

[0043] In this embodiment, the internal structure of the phase noise prediction and compensation module has been enhanced. In addition to the original linear adaptive prediction model (which can be an LMS or RLS filter), a nonlinear noise analysis and modeling unit has been added in parallel. This unit is specifically responsible for processing the instantaneous phase error sequence Δθ(n) output by the phase noise feature extraction module.

[0044] The workflow of the nonlinear noise analysis and modeling unit is as follows: First, the unit performs high-order statistical analysis on the input phase error sequence Δθ(n). Specifically, it calculates the third-order cumulant, or bispectral quantity, of the sequence in real time. Nonlinear components in phase noise, especially second-order nonlinearity, produce non-zero peaks in the bispectral domain, while the bispectral quantity of linear Gaussian noise is always zero. By detecting and locating peaks in the bispectral domain, the unit can effectively identify the frequency characteristics and intensity of the phase noise components generated by nonlinear coupling.

[0045] After identifying the nonlinear components, this unit employs a simplified second-order Volterra series filter or a lookup table-driven third-order polynomial model to model and predict this nonlinear phase error. Taking the third-order polynomial model as an example, its input is the prediction error e(n) of the linear prediction model, i.e., the residual part that the linear model cannot predict. This residual part mainly contains nonlinear noise and a small amount of random noise. The model outputs a nonlinear compensation component Δθ_nl(n), which has the following form: Δθ_nl(n)=a*e(n)+b*e(n)^2+c*e(n)^3 The coefficients a, b, and c are optimized through offline system characterization or online slow adaptive algorithms (such as gradient descent) with the goal of minimizing the final compensation residual. These coefficients are stored in the registers of the system calibration and control unit.

[0046] Ultimately, the total digital compensation signal generated by the phase noise prediction and compensation module is a composite signal synthesized from linear and nonlinear compensation components. That is, the total compensation signal -Δθ_total_pred(n+2) = -(Δθ_linear_pred(n+2) + Δθ_nl(n)). Here, Δθ_linear_pred(n+2) is predicted by the LMS or RLS filter of the main path, while Δθ_nl(n) is generated by the nonlinear noise analysis and modeling unit. This composite digital compensation signal is then converted into a complex rotation factor and applied to the multi-channel signal compensation application module, where the subsequent compensation and cancellation processes are the same as in the aforementioned embodiments.

[0047] This embodiment utilizes a parallel architecture of linear and nonlinear processing to model phase noise more comprehensively. The linear filter handles the main, linear, autocorrelated noise components, while the nonlinear model specifically compensates for noise components generated by circuit nonlinearity that the linear model cannot handle. This divide-and-conquer strategy significantly improves the system's ability to suppress complex phase noise sources, especially in high-power output scenarios or those with significant power supply coupling noise. Compared to a simple linear compensation scheme, it can achieve an additional 3 to 5 dB improvement in phase noise suppression performance.

[0048] In terms of the method flow, this embodiment adds a parallel processing step after step S130. While the instantaneous phase error sequence is fed into the linear adaptive prediction model, it is also fed into the nonlinear noise analysis and modeling unit. This unit analyzes its higher-order statistical properties and performs nonlinear modeling on the prediction residuals of the linear model to generate a nonlinear compensation component. In step S140, the generated digital compensation signal is a composite result based on the linear prediction value and the nonlinear compensation component. This improvement makes the entire compensation method more accurate in describing phase noise, thus achieving a more thorough phase noise cancellation effect in step S160.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A millimeter-wave beamforming chip phase noise suppression system, characterized in that, include: The local oscillator signal monitoring module is configured to couple a signal from the output of the chip's local oscillator and convert the coupled signal into a digital monitoring signal carrying the phase noise characteristics of the local oscillator. The phase noise feature extraction module is connected to the local oscillator signal monitoring module. It is used to receive digital monitoring signals and analyze the digital monitoring signals based on digital signal processing algorithms to extract instantaneous phase error sequence data that characterizes the change of local oscillator phase noise over time. The phase noise prediction and compensation module, connected to the phase noise feature extraction module, is used to receive instantaneous phase error sequence data and, based on an adaptive prediction model, use historical instantaneous phase error sequence data to predict the phase noise at future moments, so as to generate a digital compensation signal that is phase-opposite to the predicted phase noise at future moments. The multi-channel signal compensation application module has its input end connected to a digital baseband processing unit and a phase noise prediction and compensation module. It is used to synchronously apply the digital compensation signal to the digital baseband signal leading to multiple antenna channels in order to perform pre-distortion phase correction on the digital baseband signal and generate a phase-corrected baseband signal. It also includes an RF upconversion and transmission module, which contains multiple parallel antenna channels. Each antenna channel is connected to the output of the multi-channel signal compensation application module and receives the output signal of the local oscillator as an upconversion signal source to upconvert the phase-corrected baseband signal to the millimeter-wave band and transmit it via the antenna.

2. The system according to claim 1, characterized in that, The local oscillator signal monitoring module includes: A directional coupler is used to couple the monitoring signal out from the main output path of the local oscillator; A low-noise mixer is used to mix and downconvert the monitoring signal with a high-stability reference frequency signal to obtain an intermediate frequency signal containing the phase noise information. And an analog-to-digital converter for sampling and quantizing the intermediate frequency signal to convert it into the digital monitoring signal.

3. The system according to claim 2, characterized in that, The phase noise feature extraction module includes: A digital quadrature demodulation unit is used to decompose the input digital monitoring signal into in-phase and quadrature components; And a phase calculation unit, which uses a coordinate rotation digital calculation algorithm to calculate the instantaneous phase of the signal in real time based on the in-phase component and the quadrature component, and obtains the instantaneous phase error sequence data by comparing it with an ideal phase reference.

4. The system according to claim 3, characterized in that, The adaptive prediction model in the phase noise prediction and compensation module is a least mean square adaptive filter or a recursive least squares self-adaptive filter; the adaptive filter takes the historical phase error sequence data as input, and iteratively updates its internal filter coefficients to learn and track the dynamic change law of the phase noise, thereby predicting the future phase noise value. The predicted time margin is set to be greater than or equal to the signal path delay from the multi-channel signal compensation application module to the RF upconversion and transmission module.

5. The system according to claim 4, characterized in that, The multi-channel signal compensation application module consists of a set of digital complex multipliers, with each antenna channel corresponding to an independent digital complex multiplier. The digital compensation signal generated by the phase noise prediction and compensation module is converted into a complex rotation factor, the amplitude of which is normalized and the phase is the phase value of the compensation signal. The digital complex multiplier multiplies the rotation factor with the original digital baseband complex signal of the corresponding channel to complete the pre-distortion bit correction.

6. The system according to claim 5, characterized in that, The system also includes a system calibration and control unit, which is used to calibrate and compensate for the amplitude and phase imbalance of the local oscillator signal monitoring module and the delay differences of each channel signal path during system initialization or working mode switching; the system calibration and control unit is also responsible for dynamically adjusting the key parameters of the adaptive prediction model according to the current working frequency, temperature and communication mode.

7. The system according to claim 6, characterized in that, The phase noise prediction and compensation module is also configured to introduce a modeling and compensation mechanism for nonlinear noise components. The module identifies and quantifies the phase noise components introduced by the nonlinear effects of the circuit by analyzing the higher-order statistical characteristics of the phase error sequence data, and incorporates them into the prediction model to generate a composite digital compensation signal containing linear and nonlinear compensation components.

8. A method for suppressing phase noise in a millimeter-wave beamforming chip phase noise suppression system according to any one of claims 1-7, characterized in that, Includes the following steps: A local oscillator signal monitoring module is used to couple the output signal of the local oscillator of the chip, and the coupled signal is processed to generate a digital monitoring signal, which carries the phase noise characteristics of the local oscillator. In the phase noise feature extraction module, the digital monitoring signal is received and analyzed to extract an instantaneous phase error sequence data, which characterizes the change of the local oscillator phase noise over time. In the phase noise prediction and compensation module, the instantaneous phase error sequence data is received, and an adaptive prediction model is used to predict the phase noise at future times. The prediction is based on the historical instantaneous phase error sequence data to generate a digital compensation signal that is phase-opposite to the predicted phase noise at future times. In the multi-channel signal compensation application module, the digital compensation signal is synchronously applied to the digital baseband signal leading to multiple antenna channels to perform pre-distortion phase correction on the digital baseband signal, thereby generating a phase-corrected baseband signal; In the radio frequency upconversion and transmission module, the output signal of the local oscillator is used as the upconversion signal source to upconvert the phase-corrected baseband signal to the millimeter-wave band and transmit it via an antenna. The phase applied by the pre-distortion phase correction cancels out the phase noise introduced by the local oscillator during the upconversion process.

9. The method according to claim 8, characterized in that, The steps for extracting the instantaneous phase error sequence data include: performing digital quadrature demodulation on the digital monitoring signal, decomposing it into in-phase and quadrature components, calculating the instantaneous phase using a phase calculation algorithm, and extracting the instantaneous phase error sequence by comparing it with an ideal phase reference; the steps for predicting using an adaptive prediction model include: feeding the instantaneous phase error sequence into a least mean square adaptive filter or a recursive least squares adaptive filter, and learning the dynamic characteristics of the phase noise and predicting the phase noise value at future times by iteratively updating the internal coefficients of the filter.

10. The method according to claim 9, characterized in that, Also includes: During system initialization or working mode switching, a calibration procedure is executed to calibrate and compensate for the amplitude and phase imbalance of the signal monitoring path and the delay differences of each channel signal path. During system operation, the key parameters of the adaptive prediction model are dynamically adjusted according to the current working conditions. The key parameters include the order of the filter, the step size factor, or the forgetting factor. The system analyzes the higher-order statistical characteristics of the instantaneous phase error sequence, identifies and quantifies the nonlinear phase noise component, and combines the nonlinear phase noise component with the linear component predicted by the adaptive prediction model to generate a composite digital compensation signal.

Images