Wireless access network pilot frequency transmission optimization and channel parameter dynamic estimation method and system based on multi-mode signal processing

Abstract

The invention provides a wireless access network pilot frequency transmission optimization and channel parameter dynamic estimation method and system based on multi-mode signal processing. According to the method, the density, the position and the power of the pilot frequency are dynamically adjusted by monitoring the channel state in real time and combining machine learning and reinforcement learning technologies, and efficient utilization of pilot frequency resources and accurate improvement of channel estimation are achieved. Under various communication scenes such as high-speed movement, multi-user dense and complex multipath, the system can intelligently switch pilot frequency strategies, support non-pilot frequency or low-pilot frequency channel estimation, and significantly reduce bandwidth occupied by pilot frequency and terminal energy consumption. Furthermore, the pilot layout is optimized by adopting an analog pilot technology and a reinforcement learning algorithm, and the robustness and the anti-interference capability of channel estimation are improved. According to the method, pilot frequency resources can be saved, channel estimation errors can be reduced, the error rate is greatly reduced, the spectrum efficiency and the communication quality of a wireless communication system are greatly improved, and the method is suitable for 5G and future 6G network environments.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and specifically to a multi-scenario adaptive channel estimation and pilot adjustment method, and more particularly to a pilot signal optimization and control method based on channel parameter classification, communication scenario matching, channel state prediction, and dynamic feedback of control parameters. The method belongs to the core physical layer processing technology suitable for improving spectrum utilization and channel estimation accuracy in 5G and subsequent generation mobile communication systems. Background Art

[0002] In wireless communication systems, channel estimation is a key technology for ensuring communication quality. Wireless channels are inherently time-varying and uncertain. Signal propagation is affected by multiple factors, such as multipath fading, noise interference, and frequency offset, leading to signal distortion and increased bit error rates. The core purpose of channel estimation is to accurately obtain channel state information by analyzing received signals, providing a reliable basis for subsequent signal demodulation and decoding. High-precision channel estimation not only effectively compensates for performance losses caused by channel fading, but also significantly improves signal transmission stability and overall communication quality. Pilot transmission, a fundamental tool for channel estimation, typically involves sending a known pilot signal, enabling the receiver to compare it with the received signal and estimate the channel characteristics. The design and transmission strategy of the pilot signal directly determine the accuracy of channel estimation and resource utilization efficiency. The convergence of the Internet of Things (IoT) and vertical industries will result in a comprehensive network and scenario encompassing multiple devices, networks, and applications, interconnected and integrated. Standardizing device interface standards, communication protocols, and management protocols is a systematic technological innovation. Addressing these challenges is crucial for the widespread application of IoT technology.

[0003] In practical wireless communication systems, resources (such as bandwidth and power) are limited. Pilot transmission, while important, inevitably consumes a certain amount of system resources. As wireless communication technology evolves toward higher data rates and greater capacity, minimizing pilot resource overhead while ensuring channel estimation performance has become a core issue that demands urgent attention. Excessive pilots not only reduce spectrum utilization and impair system data transmission capabilities, but also increase device power consumption and shorten battery life. This problem is particularly severe for mobile devices. Therefore, optimizing pilot transmission strategies to achieve resource conservation and efficient utilization is of great practical significance for improving the overall performance of wireless communication systems. Appropriate pilot optimization not only helps improve spectrum and energy efficiency but also reduces system operating costs, laying a solid foundation for the sustainable development of wireless communication technology. Summary of the Invention

[0004] This paper discloses a method for optimizing pilot transmission and dynamically estimating channel parameters in wireless access networks based on multimodal signal processing. By combining advanced pilot design concepts with intelligent channel estimation algorithms, it achieves significant savings in pilot transmission resources while improving channel estimation accuracy and system performance.

[0005] In terms of pilot design and transmission, the traditional uniformly distributed pilot approach struggles to adapt to the rapid changes in time-varying wireless channels, resulting in insufficient channel estimation accuracy. To address this, the present invention proposes an adaptive pilot design method that dynamically adjusts key parameters such as the spatial position, number, and transmit power of the pilots based on the time-varying characteristics of the channel and communication requirements. In high-speed mobility scenarios, the system automatically increases the pilot density to improve its ability to track rapid channel changes; when the channel is relatively stable, the number of pilots is reduced to reduce resource consumption. This method relies on real-time channel state information feedback, which, while increasing system complexity, significantly improves the flexibility and accuracy of channel estimation.

[0006] In complex multipath fading environments, where signals propagate along multiple paths and their arrival times and amplitudes vary significantly, channel estimation is an extremely challenging task. Even with sophisticated algorithms and carefully designed pilot sequences, perfectly capturing the dynamics of the channel is difficult, leading to increased estimation errors and impacting communication quality. As system bandwidth continues to expand, the demand for the proper allocation of pilot resources becomes even higher. Within limited bandwidth, ensuring that the pilot accurately reflects the channel state while minimizing the pilot's occupation of data transmission bandwidth remains a key challenge in this field.

[0007] To solve the above problems, the present invention is based on advanced channel estimation algorithms and innovative pilot design strategies to effectively reduce pilot transmission resource consumption while ensuring that channel estimation accuracy is not affected, thereby improving the overall performance of the wireless communication system.

[0008] This solution classifies communication scenarios in detail (such as static scenarios, high-speed movement, multi-user environments, etc.) and adjusts pilot resource allocation in different scenarios, successfully reducing channel resource consumption by 30% to 70%.

[0009] This solution also supports AI-assisted pilot-free channel estimation. In scenarios with high signal-to-noise ratios, the pilot signal is completely eliminated, and deep learning models such as autoencoders are used to directly extract channel features from the data signal to achieve channel estimation, further saving resources.

[0010] Compared with the traditional fixed pilot scheme that cannot adapt to high-speed movement or sudden interference, this scheme combines scenario classification with pre-configured EPC (pilot configuration scheme) to achieve millisecond-level switching of pilot modes, such as switching from sparse pilots to dense pilots when demand changes, improving system adaptability.

[0011] The present invention also introduces a reinforcement learning dynamic parameter adjustment mechanism, which automatically optimizes the pilot transmission interval and power based on real-time feedback, realizing the transformation of channel estimation from passive repair to active prediction. The system can predict environmental changes, intelligently configure pilot resources, and improve communication stability and efficiency.

[0012] In several embodiments, the system uses an adaptive mechanism to adjust the number of pilots, combined with machine learning to predict channel trends, to dynamically reduce the number of pilots and partially replace real pilots with simulated signals. In a multi-cell environment, the system automatically reduces the number of pilots in cells with severe interference and increases simulated signals, effectively reducing pilot contamination. In actual multi-cell testing, interference signal strength was reduced by 40%, and the bit error rate dropped from 0.03 to 0.01, significantly improving communication quality and anti-interference capabilities.

[0013] Furthermore, by analyzing extensive historical channel data, a genetic algorithm can optimize the pilot placement distribution to adapt to diverse channel environments. In a massive MIMO base station scenario, at a signal-to-noise ratio of 15dB, the optimized pilot placement reduced the mean square error of channel estimation from 0.07 to 0.03, while reducing the number of pilots by 30%, significantly improving system resource utilization.

[0014] When the channel changes rapidly, such as the Doppler effect in high-speed mobile scenarios or strong interference causing a sharp drop in signal-to-noise ratio and a surge in bit error rate, machine learning algorithms promptly capture these trends and predict channel deterioration. The system responds quickly, avoiding the need to blindly increase the number of pilot signals. By adding simulated pilot signals to fill resource requirements, the system improves the sensitivity and accuracy of channel estimation and enhances communication reliability.

[0015] For example, in high-speed rail communications, in response to the rapid changes in the channel caused by the high-speed movement of the train, the number of pilots is dynamically adjusted and supplemented by simulated pilots, which not only ensures the continuity of channel estimation, but also effectively saves resources and reduces communication interruptions.

[0016] In view of the fact that devices such as smart home sensors in the IoT environment frequently transmit small amounts of data, a low-frequency and high-efficiency pilot solution is adopted to balance communication needs and resource conservation, improve data transmission efficiency, and reduce pilot overhead.

[0017] Indoor and complex industrial environments, signal fading and interference are severe due to multipath reflection, electromagnetic interference, and physical obstructions. Pilot design must possess strong anti-interference capabilities. This, combined with AI-driven communication parameter acquisition and channel estimation algorithms, supplemented by simulated pilot technology, overcomes the transmission attenuation and interference issues that limit traditional pilots, ensuring stable and reliable communications.

[0018] The present invention also uses deep learning to automatically generate optimal pilot sequences adapted to multiple scenarios, constructs pilot templates using self-supervised learning, and improves robustness against multipath interference. Reinforcement learning optimizes the distribution of pilot positions, significantly reducing pilot density while ensuring channel estimation accuracy.

[0019] The TinyML model is deployed on the base station side to process channel status information in real time, with inference delay less than 50 milliseconds and computing resource consumption reduced by 70%, significantly improving system real-time performance and resource utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic diagram showing channel configuration loading.

[0021] Figure 2 A schematic diagram showing real-time channel data monitoring.

[0022] Figure 3 A schematic diagram showing signal noise reduction is shown.

[0023] Figure 4 A schematic diagram of signal filtering processing is shown.

[0024] Figure 5 A schematic diagram showing the conversion of feature extraction into feature vectors.

[0025] Figure 6 Schematic diagram showing several sample scenarios of model classification.

[0026] Figure 7 A schematic diagram showing the model classification operation process.

[0027] Figure 8 Shown in accordance with Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 ,and Figure 7 A diagram illustrating how to dynamically adjust and coordinate operations, describing how to systematically and continuously operate and optimize channels.

[0028] Figure 9 A schematic diagram showing how feedback is dynamically optimized for the system.

[0029] Figure 10 A schematic diagram showing the overall process of the system. DETAILED DESCRIPTION

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0031] Figure 1The figure shows a novel channel configuration loading process in the core network. The pilot predefined template node 110 includes the following parameters: pilot length 111, pilot interval 112, guard band width 113, and transmission period 114. The device management software node 120 receives configuration parameters such as the input frequency band 121, modulation and coding mode 122, and pilot predefined template 110, and encapsulates them into a binary configuration file 130 for loading and use when the device is started. After the parameters are loaded, the system enters the environment adaptation link 131, which monitors the network operation status through an AI algorithm and automatically completes parameter calibration. Subsequently, the hardware initialization process 140 is performed to complete the initialization operations of the analog-to-digital converter (ADC) 141, the digital-to-analog converter (DAC) 142, the filter 143, and the RF front end 144 in sequence.

[0032] Figure 2 The core components of real-time channel data monitoring are demonstrated. The signal acquisition node 210 uses software-defined radio (SDR) equipment to collect multi-band signals and perform comprehensive analysis of Doppler frequency shift, delay spread, signal-to-noise ratio (SNR) and interference level to evaluate signal quality. If the signal-to-noise ratio is lower than the preset threshold (such as 15dB), the corresponding data will be discarded. The data preprocessing node 220 includes a sliding window noise reduction 221 (the default window length is 100ms and the step size is 50ms), an FIR filter 222 (the cutoff frequency dynamically matches the current channel bandwidth), and a dual-path feature extraction module (combining time domain statistics and frequency domain energy distribution). The vectorization node 230 uses Fourier transform to generate a 512-dimensional frequency domain feature vector, and generates a time-frequency matrix through wavelet transform to support non-stationary signal analysis. All vectorization results are accompanied by timestamps and geographic location tags, which are uniformly managed by the data storage node and support queries.

[0033] Figure 3 The following details the sliding window noise reduction process. Initially, the window size W is set to control the noise reduction intensity. The data loading node then reads the signal points within the window, and the data processing node performs comprehensive processing on the time domain, frequency domain, and statistical dimensions. If the signal is stationary, a moving average algorithm is used to clean up abnormal fluctuations in the time series: Among them, y t is the moving average at time point t (output result). x i is the original data (input) of the i-th observation point in the time series. w is the window size, which determines the number of observations included in the mean calculation. i is the summation index, and w points are taken continuously from the current time t (covering the interval [t, t+w-1]). In a fast-response scenario, use the exponentially weighted moving average algorithm to quickly track changes: y t =αx t +(1-α)y t-1 Where yt is the smoothed estimated value (output result) at the current time t. x t is the actual observation value (original data input) at the current time t. y t-1 It is the smoothed estimated value (historical state) at the previous moment t-1. α is the attenuation factor (weight coefficient), which satisfies 0<α<1 and controls the weight distribution of new and old data. Choose a larger α (such as 0.9) to adapt to fast response. The Window Update node replaces the center point data with the noise reduction result, and then the window slides to the next position until the signal sequence is traversed.

[0034] Figure 4 This chapter describes the complete process of FIR digital filtering. The signal input node receives the discrete-time signal x[n]x[n], and the preprocessing node performs operations such as DC component removal, normalization, and outlier processing. Filter types include low-pass, high-pass, band-pass, and band-stop. Based on the target bandwidth and stop-band attenuation requirements, the system determines the filter order N and generates the filter coefficients h[n]h[n] according to the design algorithm, storing them in registers or memory. The filtering operation can use time-domain convolution: Where y[n] is the output signal of the system at discrete time point n. x[nk] is the value of the input signal at time point nk (the time-shifted input). h[k] is the kth coefficient of the system unit impulse response (Unit Impulse Response) (characterizing the system characteristics). N is the filter length (or duration of the impulse response). k is the sum index variable (covering integers from 0 to N). Or use FFT to speed up processing of long sequences: y=IFFT{FFT(h)·FFT(x)} The final filtered signal y[n] is output, and its frequency response, fidelity and processing complexity are evaluated; if it does not meet the standards, the filtering parameters are automatically adjusted through feedback.

[0035] Figure 5Three parallel processing paths for feature vector extraction are demonstrated. Path one involves time-domain feature extraction, calculating statistics such as mean, variance, skewness, and kurtosis to form an N-dimensional vector. Path two involves frequency-domain processing, which involves Fourier transforming and then uniformly sampling the spectrum to construct a 512-dimensional frequency-domain vector. Path three involves time-frequency analysis, using wavelet transform to generate a 128x128 time-frequency feature matrix and normalize it. The results from these three paths are dimensionally aligned and fused to generate a multidimensional composite feature vector, which is particularly suitable for AI modeling and anomaly detection of non-stationary signals.

[0036] Figure 6 This paper illustrates template classification models for various scenarios, including channel parameters (such as SNR and multipath delay), terminal status (speed and location), and environmental characteristics (base station density and obstacle distribution). The scenario classification node uses pre-trained models to determine scenarios such as indoor scenarios (low frequency and strong multipath), mobile scenarios (frequency shift and time-varying channels), and dense urban areas (high frequency attenuation and co-channel interference). Innovations include: after identification, collection parameters can be called back via the control bus, discard reason codes (such as low SNR and frame errors) can be marked, and a compressed storage strategy is enabled for high-frequency scenarios, while retaining the original waveform for low-frequency scenarios.

[0037] Figure 7 Describes the classification model operation process. Preprocessing nodes include denoising, wavelet filtering, normalization (Z-score / Min-Max), missing value interpolation and other operations. The model structure supports two types of paths: 1) Lightweight 1D-CNN branch, suitable for extracting time series structure features, the structure includes input layer, convolution layer, pooling layer and fully connected layer; 2) Decision tree structure based on feature importance, including information gain calculation and pruning optimization, suitable for scenarios with low feature dimension and high response speed requirements. Static scenes output labeled templates to facilitate historical matching; dynamic scenes trigger the reinforcement learning module, adjust parameters in real time, and feed the results back to the data storage node to continuously optimize model performance.

[0038] Figure 8 The system demonstrates an adaptive communication configuration system based on scenario recognition. The process is initiated by the monitoring node, which sequentially collects and preprocesses channel data (interference elimination and signal restoration). The data then enters the feature extraction module (such as SNR and spectrum indicators). The classification model then outputs a scenario label (such as "indoor"), which then loads the corresponding configuration strategy. This includes modulation optimization (such as QAM order switching), power adaptation, and resource mapping updates. Finally, the performance evaluation node determines whether the system meets the requirements. If the evaluation passes, a configuration report is generated and archived in a closed loop. If not, an alarm and manual intervention mechanism are triggered, forming a closed-loop control chain of "monitoring-analysis-adjustment-verification."

[0039] Figure 9The workflow of the dynamic optimization feedback mechanism is shown. The system starts the closed-loop optimization process through the monitoring node, initializes the sensor array (BER / SNR detection module, etc.), and collects time domain and frequency domain channel parameters, such as BER, SNR, multipath delay, frequency offset, phase noise, etc. The preprocessing node is responsible for burst noise elimination and feature normalization. AI predictive analysis uses a lightweight LSTM model to predict channel degradation trends and dynamically adjust the window (10 to 100ms). The optimization logic includes: Channel A determines whether to activate Turbo redundant coding and shorten the OFDM interval through BER monitoring; Channel B triggers power amplifier bias compensation based on SNR decline. The system generates a new EPC scheme based on feedback, including: enhanced physical layer configuration, dynamic scheduling of hybrid automatic repeat request HARQ process number, pilot map update, etc. The feedback mechanism refreshes data every 10ms. If it fails three times in a row, it switches to the backup channel. The Q-learning algorithm is used for weight adjustment: Q(s t ,α t )=(1-α)Q(s t ,α t )+α[r+γmaxQ(s t+1 ,α t+1 )] Where s is the channel level combination and a is the adjustable power / time parameter. Among them, Q(s,a) represents the Q value of taking action a in state s, α is the learning rate (controls the weight of each update), r is the immediate reward that the agent can get after performing action a, γ is the discount factor (controls the weight of future rewards, indicating the importance of future rewards), s' and a' represent the new state and new action entered after the current action is performed, and max(Q(s',a')) represents the maximum Q value of all possible actions taken in the next state s'. By continuously iteratively updating the Q value, the agent can eventually learn the optimal strategy for taking different actions in different states, thereby achieving autonomous decision-making.

[0040] Figure 10 The overall architecture of the system is summarized, including four technical modules: dynamic perception layer ( Figure 2-5 ) realizes channel monitoring and signal processing; AI intelligent layer ( Figure 6-7 ) completes scene recognition and EPC generation; resource configuration layer ( Figure 8 ) adjusts communication parameters according to model output; closed-loop optimization layer ( Figure 9 ) Real-time monitoring and feedback correction to achieve system self-learning, self-adaptation and self-optimization.

[0041] The adaptive wireless communication optimization system based on scene recognition proposed in this invention has achieved technological breakthroughs in multiple dimensions, including communication perception, channel modeling, intelligent classification, parameter optimization, and system feedback. It has the following significant advantages:

[0042] First, in terms of channel perception and data processing, this system integrates multimodal signal analysis techniques across the time, frequency, and time-frequency domains, enabling comprehensive modeling and analysis of complex channel environments characterized by non-stationary, multipath, and multi-interference. The system achieves high-precision monitoring of signal quality by acquiring core metrics such as bit error rate (BER), signal-to-noise ratio (SNR), and carrier frequency offset (CFO) in real time, combined with sliding window noise reduction, wavelet denoising, FIR filtering, and multidimensional feature extraction. In particular, during signal processing, dynamic adaptive window control and feature matrix normalization techniques effectively improve channel decodability and modeling stability in low signal-to-noise ratio environments.

[0043] Secondly, in terms of AI-assisted decision-making and scene recognition, the system introduces lightweight one-dimensional convolutional neural networks (1D-CNN) and long short-term memory network (LSTM) models to achieve efficient recognition and classification of communication scenarios, including typical application scenarios such as indoor, mobile, urban, remote suburbs, and dense base station areas. The scene model can not only perform static labeling management, but also introduce reinforcement learning strategies for dynamic environmental changes. According to the current channel status and historical data, it can autonomously adjust the feature selection weights and model structure, thereby continuously optimizing the classification accuracy. Through the joint analysis of signal statistics, spectrum distribution, and spatial correlation characteristics, the system can accurately judge complex channel change patterns, providing a reliable basis for subsequent communication parameter adjustments.

[0044] In terms of adaptive parameter adjustment, the system establishes a scenario-strategy mapping mechanism to automatically match and configure EPC solutions, including key aspects such as modulation switching (such as QAM order adjustment), dynamic adjustment of transmit power, and pilot pattern reconfiguration. The system determines whether current communication performance meets standards based on preset thresholds and real-time feedback mechanisms. This triggers configuration report generation or alarm mechanisms, thereby establishing a complete "monitoring-analysis-optimization-verification" closed-loop control chain.

[0045] In terms of closed-loop optimization control, the system introduces the Q-learning algorithm to intelligently search and update the weights of parameter adjustment strategies, forming a dual-index coupling optimization model based on the joint evaluation of BER and SNR. When the system detects that both are degraded at the same time, it prioritizes the execution of aggressive parameter combination strategies (such as increasing the transmit power while compressing the OFDM symbol interval) to improve communication recovery capabilities. In addition, the system has hardware-in-the-loop verification capabilities, and reconstructs the pilot pattern through FPGA real-time loading to ensure that the parameter adjustment action is physically feasible and avoid software and hardware mismatch problems. To ensure long-term evolution and parameter security management, the EPC configuration solution introduces a version control mechanism similar to Git to achieve rapid rollback of historical versions and synchronous updates of multi-environment deployments.

[0046] Finally, in terms of system deployment and engineering adaptation, the present invention adopts a standardized interface encapsulation method, and all modules support calling in the form of APIs, with good module reusability and cross-platform integration capabilities. The system architecture supports the coordinated operation of edge computing and cloud training, with the edge side processing real-time perception and preliminary judgment, and the cloud executing model iteration and strategy evolution, forming an intelligent communication platform with controllable resources, real-time response, and precise strategy. At the same time, the model update mechanism supports online hot loading and transfer learning, ensuring that the system has the ability to quickly adapt and continuously optimize when facing new scenarios, new frequency bands, and new terminal types.

[0047] In summary, the present invention not only significantly improves the stability and adaptability of the communication system by integrating key modules such as multimodal perception, AI intelligent classification, scenario-driven configuration, and reinforcement learning closed-loop optimization, but also provides a systematic and deployable technical solution for wireless communication optimization in 5G and future AIoT environments.

[0048] The claims and embodiments of the present invention are intended to be exemplary only. Those skilled in the art may make various modifications, substitutions, equivalent variations, and combinations based on the spirit and principles of the present invention. Any changes that do not depart from the essence and scope of the technical solutions of the present invention shall be included within the scope of protection of the present invention.

[0049] For example, the communication parameters, channel monitoring methods, machine learning model types, pilot configuration strategies, hardware initialization steps, etc. described in the claims can all be appropriately adjusted and improved according to actual application requirements, including but not limited to the use of different algorithm models, hardware structures, spectrum resource management solutions and adaptive optimization methods, which should all be regarded as equivalent solutions of the present invention.

Claims

1. A channel monitoring method, comprising: Monitor channel status based on received signals or network signaling; Extract channel features based on multimodal signal fusion perception; Match communication scenarios; Predict channel changes; as well as Process channel state information.

2. The method of claim 1, wherein the method further comprises: Signaling analysis based on the wireless access network and / or data collection through dedicated detection equipment.

3. The method according to claim 1, further comprising: Monitor channel status in real time and use machine learning algorithms to predict channel change trends.

4. The method according to claim 1, wherein the method further comprises extracting feature vectors through parameters to determine the channel environment; wherein the parameters include electromagnetic wave intensity, phase, visible light signal, sound wave, and / or scene perception.

5. A channel monitoring method, comprising: Monitor channel status based on received signals or network signaling; Classify channel parameters; Match communication scenarios; Predict channel changes; Replace the pilot signal with an analog signal; as well as Dynamic feedback of control parameters realizes closed-loop control.

6. The method according to claim 5, wherein the method further comprises using a lightweight neural network or a decision tree, combined with signal quality, propagation characteristics and / or spectrum parameters to perform scene classification.

7. The method according to claim 5 further comprises adjusting the number, position, and / or power allocation of pilot signals according to the communication scenario to optimize channel estimation performance and / or system resource utilization.

8. The method of claim 5, further comprising adjusting the pilot signal configuration, employing simulated pilot signals and / or channel estimation compensation to maintain communication quality when a degradation of channel quality is detected.

9. The method of claim 5, further comprising adjusting pilot configuration and / or channel equalization parameters based on feedback.

10. The method according to claim 5, further comprising adjusting channel parameters to adapt to different scene changes.

Images