An optical-reservoir-based channel equalization and frequency offset estimation device and method

Abstract

This application relates to the field of wireless communication technology and proposes a channel equalization and frequency offset estimation device and method based on an optical reservoir. The device includes a signal input unit, an optical reservoir computing network, a state acquisition unit, a channel equalization unit, a frequency offset estimation and compensation unit, and a signal output unit. The output of the signal input unit is connected to the input of the optical reservoir computing network. The state acquisition unit is located on the state output side of the optical reservoir computing network and is connected to both the channel equalization unit and the frequency offset estimation and compensation unit. The signal output unit is connected to the outputs of both the channel equalization unit and the frequency offset estimation and compensation unit. The method includes acquiring a received signal, inputting the received signal into the optical reservoir computing network and extracting reservoir state information, performing channel equalization and frequency offset estimation based on the reservoir state information, and performing compensation based on the frequency offset estimation value. This application improves the stability of received signal processing under conditions of Doppler frequency offset and low signal-to-noise ratio.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and more specifically, to a channel equalization and frequency offset estimation device and method based on an optical reservoir. Background Technology

[0002] In 5G non-terrestrial networks (NTN) and future satellite communication systems, satellite communication can extend the coverage of terrestrial cellular networks and provide communication access capabilities for remote areas, oceans, aviation, and other scenarios. Compared with terrestrial communication links, satellite communication links typically have characteristics such as longer transmission distances, greater path loss, and significant variations in the propagation environment. Simultaneously, the high relative speed between the satellite platform and the ground terminal can easily introduce significant Doppler frequency shifts into the received signal. This type of frequency shift disrupts the orthogonality of Orthogonal Frequency Division Multiplexing (OFDM) symbols and, together with channel fading, multipath effects, and noise interference, affects the synchronization, equalization, and demodulation reliability of the received signal.

[0003] Existing receivers typically employ ephemeris-assisted pre-compensation, cyclic prefix autocorrelation, primary synchronization signal / secondary synchronization signal (PSS / SSS) cross-correlation, cross-synchronization block time slot correlation processing, or digital signal processing algorithms to estimate frequency offset and equalize the channel in the received signal. Ephemeris-assisted pre-compensation relies heavily on the accuracy and timeliness of ephemeris information; when ephemeris updates are delayed, link states change rapidly, or the satellite-to-ground link is interrupted, the pre-compensation result can easily deviate from the actual frequency offset. Cyclic prefix autocorrelation and PSS / SSS cross-correlation methods primarily rely on repetitive structures or synchronization sequence correlation peaks in the received signal. Under large frequency offset conditions, these correlation peaks can broaden, shift, or decrease in amplitude. Furthermore, under low signal-to-noise ratio conditions, they are easily interfered with by noise and multipath components, thus limiting the frequency offset estimation range and reducing estimation accuracy. Some methods expand the observation time window through cross-synchronization block time slot correlation processing, but this requires buffering data from multiple time slots, increasing synchronization processing delay and potentially introducing additional estimation errors due to changes in channel state and time deviation between different time slots. Furthermore, traditional channel equalization and frequency offset estimation are often performed serially as two relatively independent digital processing steps, which makes it difficult to fully utilize the dynamic state characteristics of the received signal in nonlinear channels. When faced with complex link changes such as multipath fading, rain attenuation, and atmospheric scattering, the adaptive capability is insufficient, and the high computational load and power consumption are not conducive to the deployment of spaceborne platforms or low-power terminals.

[0004] Therefore, there is an urgent need for a device and method that can extract steady-state features from received signals under conditions of Doppler frequency offset and low signal-to-noise ratio, and collaboratively complete channel equalization and frequency offset estimation compensation based on these features, so as to reduce receiver processing delay and computational complexity and improve the reliability of received signal processing in satellite communication scenarios. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the prior art by providing a channel equalization and frequency offset estimation device and method based on an optical reservoir, so as to solve the problem that frequency offset estimation and channel equalization in the prior art are difficult to balance accuracy, delay and power consumption.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: This application provides a channel equalization and frequency offset estimation device based on an optical reservoir. The device includes a signal input unit, an optical reservoir computing network, a state acquisition unit, a channel equalization unit, a frequency offset estimation and compensation unit, and a signal output unit. The output end of the signal input unit is connected to the input end of the optical reservoir computing network. The state acquisition unit is located on the state output side of the optical reservoir computing network and is connected to the channel equalization unit and the frequency offset estimation and compensation unit, respectively. The signal output unit is connected to the output side of the channel equalization unit and the frequency offset estimation and compensation unit.

[0007] This application establishes an optical reservoir computing network, a state acquisition unit, a channel equalization unit, and a frequency offset estimation and compensation unit between the signal input unit and the signal output unit. This allows the received signal to form a nonlinear dynamic response within the optical reservoir computing network, and the state acquisition unit extracts the reservoir state information. This reservoir state information is not a single correlation peak result, but rather a state characteristic formed after the received signal is mapped through the optical reservoir computing network. Therefore, under low signal-to-noise ratio (SNR) conditions, it provides a more stable feature basis for subsequent processing, reducing the impact of noise on traditional correlation peak detection results. Since the state acquisition unit is connected to both the channel equalization unit and the frequency offset estimation and compensation unit, the same reservoir state information can enter the channel equalization branch and the frequency offset estimation and compensation branch respectively. The channel equalization unit compensates for multipath fading, link loss, and channel distortion based on this state information, while the frequency offset estimation and compensation unit extracts and compensates for Doppler frequency offset features based on this state information. Thus, this device reduces redundant calculations and processing delays caused by the decentralized processing of channel equalization and frequency offset estimation, and improves the stability of received signal processing under Doppler frequency offset and low SNR conditions.

[0008] Furthermore, the optical reservoir computing network includes at least two reservoir sub-networks, which include at least two of the following: a single-feedback-loop optical reservoir sub-network, a multi-layer cascaded optical deep reservoir sub-network, and a multiple-input multiple-output optical reservoir sub-network. The optical reservoir computing network can form a collaborative configuration of various reservoir structures according to the received signal processing requirements, enabling different reservoir sub-networks to perform low-complexity mapping, deep feature extraction, or multi-feature fusion functions respectively, thereby improving the device's adaptability to complex channel environments.

[0009] Furthermore, the state acquisition unit is connected to the state output side of at least two reserve pool sub-networks and acquires the reserve pool state information output by at least two reserve pool sub-networks. The state acquisition unit can aggregate the state information output by different reserve pool sub-networks, providing a richer state feature basis for the channel equalization unit and frequency offset estimation and compensation unit, which is beneficial to improving the stability of equalization processing and frequency offset estimation and compensation.

[0010] Furthermore, the single-feedback-loop optical reservoir subnetwork includes a silicon-based microring resonator and a single feedback loop coupled to the silicon-based microring resonator. The silicon-based microring resonator, in conjunction with the single feedback loop, can form a delayed feedback response with fewer optical components, reducing the structural complexity and power consumption of the optical reservoir computing network.

[0011] Furthermore, the multilayer cascaded optical depth reservoir subnetwork comprises multiple sequentially connected reservoir layers, with adjacent reservoir layers connected via optical waveguides. The received signal is transmitted and mapped layer by layer between the reservoir layers, enhancing the optical reservoir computing network's ability to extract complex nonlinear features; the connection between adjacent reservoir layers via optical waveguides also reduces the additional power consumption caused by interlayer photoelectric conversion.

[0012] Furthermore, the multiple-input multiple-output (MIMO) optical reservoir subnetwork includes multiple input channels, each connected to different channel characteristic parameters. After these different channel characteristic parameters enter the MIMO optical reservoir subnetwork through multiple input channels, they form fused state information within the reservoir. This enables the device to comprehensively utilize channel states such as received signal strength, phase changes, and multipath delay, improving its adaptive processing capabilities in complex environments.

[0013] Furthermore, the frequency offset estimation and compensation unit includes a coarse frequency offset estimation unit, a fine frequency offset estimation unit, a fusion unit, and a time-domain compensation unit. The coarse and fine frequency offset estimation units are connected to the state acquisition unit, respectively. The fusion unit connects the coarse, fine, and time-domain compensation units, and the time-domain compensation unit is connected to the signal output unit. The coarse and fine frequency offset estimation units generate frequency offset estimation results of different accuracy levels based on the state information of the reservoir. The fusion unit fuses the two-level estimation results and provides them to the time-domain compensation unit, which helps to improve the accuracy and stability of frequency offset compensation under Doppler frequency offset conditions.

[0014] This application also proposes a channel equalization and frequency offset estimation method based on an optical reservoir, which includes the following steps: S1. Acquire the received signal; S2. Input the received signal into the optical reservoir computing network to extract the reservoir status information; S3. Perform channel equalization on the received signal based on the reservoir status information to obtain the equalized signal; S4. Based on the status information of the same reserve pool, frequency offset estimation is performed to obtain the frequency offset estimate; S5. Perform frequency offset compensation on the received signal based on the frequency offset estimate.

[0015] This application inputs the received signal into an optical reservoir computing network, first extracts the reservoir state information corresponding to the received signal, and then performs channel equalization and frequency offset estimation based on this reservoir state information, enabling the channel distortion characteristics and frequency offset characteristics in the received signal to be expressed in the same state space. For the channel equalization process, the reservoir state information reflects the changes in the received signal after being affected by multipath fading, link loss, and noise interference, thus providing a state basis for obtaining the equalized signal. For the frequency offset estimation process, the reservoir state information reflects the signal phase and frequency change characteristics caused by Doppler frequency offset, thus providing a basis for obtaining the frequency offset estimate and further frequency offset compensation. Therefore, this method avoids the problem of insufficient stability at low signal-to-noise ratios when relying solely on the correlation peak of the synchronization sequence for frequency offset estimation, and also reduces the computational burden caused by the separate processing of channel equalization and frequency offset estimation, thereby improving the reliability and real-time performance of received signal processing.

[0016] Furthermore, in S3, the state information of the reservoir is used to construct a state vector. This state vector is then subjected to matrix operations with a pre-determined equalization output weight matrix to obtain the equalized signal. By organizing the reservoir state information into a state vector and performing matrix operations with the equalization output weight matrix, the nonlinear state features extracted by the optical reservoir computing network are transformed into equalization output results. This provides a clear computational basis for the channel equalization process and helps reduce the computational complexity of channel equalization processing.

[0017] Furthermore, in S4, the frequency offset estimate is generated based on the reserve pool state information corresponding to the primary and secondary synchronization signals within a single synchronization signal block time slot. Generating the frequency offset estimate includes a coarse estimation stage and a fine estimation stage. The coarse estimation stage yields a coarse frequency offset estimate, and the fine estimation stage yields a fine frequency offset estimate. The coarse and fine frequency offset estimates are then fused to obtain the total frequency offset estimate, which is used for time-domain frequency offset compensation. By utilizing the reserve pool state information corresponding to the primary and secondary synchronization signals within a single synchronization signal block time slot to generate the frequency offset estimate, processing delays and synchronization error accumulation caused by cross-synchronization signal block time slot buffering are avoided. Moreover, a broad range of coarse frequency offset estimates is obtained in the coarse estimation stage, and a more refined frequency offset estimate is obtained in the fine estimation stage. The fusion of the two-stage estimates for time-domain frequency offset compensation helps to balance the estimation range and compensation accuracy under Doppler frequency offset conditions.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This application introduces an optical reservoir computing network into the channel equalization and frequency offset estimation process, so that the received signal is mapped by the optical reservoir computing network to form reservoir state information, and channel equalization and frequency offset estimation compensation are completed based on the reservoir state information. Compared with the traditional method of setting and processing channel equalization and frequency offset estimation separately, this application can reduce the computational burden caused by repeated feature extraction and multi-level digital processing at the receiver, and improve the overall stability of received signal processing under Doppler frequency offset and low signal-to-noise ratio conditions.

[0019] (2) The optical reservoir computing network of this application can set up multiple reservoir sub-networks according to processing requirements, so that reservoir sub-networks with different structural forms can respectively undertake low-complexity state generation, deep feature extraction or multi-channel feature fusion tasks, thereby enhancing the device's adaptability to complex channel environments. This structure is beneficial to reduce the hardware resource and power consumption requirements of traditional high-complexity digital processing methods while ensuring the receiving signal processing effect.

[0020] (3) In the frequency offset estimation and compensation process, this application adopts a combination of coarse frequency offset estimation and fine frequency offset estimation, and performs time-domain compensation after fusing the two-level frequency offset estimation results. This can take into account both the estimation range under large frequency offset conditions and the estimation accuracy under residual frequency offset conditions, thereby improving the accuracy of frequency offset compensation in Doppler frequency offset scenarios.

[0021] (4) This application can generate frequency offset estimation based on the reserve pool status information corresponding to the main synchronization signal and the auxiliary synchronization signal in a single synchronization signal block time slot. It does not require buffering multiple segments of received data across synchronization signal block time slots, which reduces the buffering requirements and processing delay caused by cross-time slot processing, and also reduces the impact of channel changes and time deviations between different time slots on the frequency offset estimation results. Attached Figure Description

[0022] Figure 1 A schematic diagram of a channel equalization and frequency offset estimation device based on an optical reservoir provided by the present invention; Figure 2 This is a schematic diagram of the multilayer tandem optical depth reservoir subnetwork in this invention; Figure 3 A flowchart of a channel equalization and frequency offset estimation method based on an optical reservoir provided by the present invention; Figure 4 This is a flowchart of the frequency offset estimation and compensation process in this invention; Figure 5 The acquisition probability curves of the present invention under different signal-to-noise ratio conditions are shown below. Figure 6 The frequency offset estimation error curves of this invention under different signal-to-noise ratio conditions are shown. Figure 7 The bit error rate curves for this invention under different signal-to-noise ratio conditions are shown. Detailed Implementation

[0023] To make the implementation process of this invention clearer, a detailed description will be provided below in conjunction with the accompanying drawings.

[0024] Example 1:

[0025] This invention provides a channel equalization and frequency offset estimation device based on an optical reservoir, such as... Figure 1 As shown, the device includes a signal input unit, an optical reservoir computing network, a state acquisition unit, a channel equalization unit, a frequency offset estimation and compensation unit, and a signal output unit. The output of the signal input unit is connected to the input of the optical reservoir computing network. The state acquisition unit is located on the state output side of the optical reservoir computing network and is connected to the channel equalization unit and the frequency offset estimation and compensation unit, respectively. The signal output unit is connected to the output of the channel equalization unit and the frequency offset estimation and compensation unit.

[0026] In this embodiment, the signal input unit is used to access the received signal obtained by the satellite communication receiver. This received signal can be a baseband received signal transmitted over a long distance in a 5G NTN scenario, or a digital signal after down-conversion, sampling, and preprocessing by the radio frequency front-end. Due to the relatively high-speed motion between the satellite and the ground terminal, the received signal typically contains Doppler frequency offset, channel fading, multipath interference, and noise disturbances simultaneously. After the signal input unit inputs the received signal into the optical reservoir computing network, the optical reservoir computing network performs a nonlinear dynamic mapping on the received signal, transforming the amplitude changes, phase changes, frequency offset characteristics, and channel distortion characteristics in the received signal into a dynamic state within the reservoir.

[0027] The Optical Reservoir Computing (ORC) network comprises at least two reservoir sub-networks, which include at least two of the following: a single-feedback-loop optical reservoir sub-network, a multi-layer cascaded optical depth reservoir network, and a multi-input multi-output optical reservoir sub-network. In practice, at least two of these three types of reservoir sub-networks can be selected and combined based on processing accuracy, power consumption constraints, and channel environment complexity; alternatively, all three types of reservoir sub-networks can be configured simultaneously. Multiple reservoir sub-networks can receive the received signals output from the signal input unit in parallel, or they can separately receive the received signals and channel characteristic parameters obtained from the received signals to form reservoir state information at different levels or dimensions.

[0028] The state acquisition unit is connected to the state output side of at least two reserve pool sub-networks and acquires the reserve pool state information output by at least two reserve pool sub-networks. Specifically, the state acquisition unit may include a virtual node sampling unit, which samples the delayed state in the reserve pool sub-network according to a preset sampling interval to obtain multiple virtual node states. The reserve pool state information output by different reserve pool sub-networks can each form a corresponding state vector, or they can be concatenated, selected, or weighted in the state acquisition unit to form a unified state vector. This unified state vector is then transmitted to the channel equalization unit and the frequency offset estimation compensation unit, enabling the channel equalization and frequency offset estimation compensation to share the dynamic state characteristics of the received signal after mapping through the optical reserve pool computing network.

[0029] The single-feedback-loop optical reservoir subnetwork comprises a silicon-based microring resonator and a single feedback loop coupled to the silicon-based microring resonator. The received signal, after preprocessing, is injected into the silicon-based microring resonator. Under the influence of the input signal, the silicon-based microring resonator generates a nonlinear optical response. The single feedback loop delays the output state of the silicon-based microring resonator and feeds it back to its input or coupling side, allowing the current input signal and historical states to participate in the reservoir state evolution. Therefore, the single-feedback-loop optical reservoir subnetwork can generate multiple virtual node states with a relatively small number of physical optical nodes.

[0030] For ease of explanation, the dynamic response of the single-feedback-loop optical reservoir subnetwork can be expressed as: Where a is the amplitude of the electric field inside the cavity. This is the resonant frequency of the cold cavity. It is a nonlinear frequency shift. This represents the total loss within the cavity. For input signal, As a feedback signal, To provide feedback on the delay time, For feedback phase, and These represent the input coupling coefficient and the feedback coupling coefficient, respectively. This formula is used to explain the process by which the input signal, the intracavity nonlinear response, and the delayed feedback state jointly influence the state evolution of the reservoir.

[0031] During the delayed feedback process, the state acquisition unit samples the states of multiple virtual nodes at equal intervals within one mask period. Let the mask period be... The sampling interval between adjacent virtual nodes is Then the number of virtual nodes is: The states of multiple virtual nodes extracted within the t-th sampling period constitute a state vector: in, Let the state vector of the reservoir be... Let be the state value of the i-th virtual node during the t-th sampling period. Through the above virtual node sampling method, the single-feedback-loop optical reservoir subnetwork can form a high-dimensional state representation with fewer physical components.

[0032] like Figure 2 As shown, the multilayer cascaded optical depth reservoir subnetwork includes multiple reservoir layers connected in sequence, with adjacent reservoir layers connected by optical waveguides. Figure 2 In this diagram, ORC modules represent reservoir layers. Multiple ORC modules are connected sequentially along the signal transmission direction; the ellipsis indicates that the number of reservoir layers can be expanded according to processing requirements. The state information of each ORC module can be transmitted to the output side to participate in subsequent channel equalization or frequency offset estimation processing. Each reservoir layer may include a silicon-based microring resonator and a corresponding delay feedback structure. After the received signal enters the first reservoir layer, it forms the first layer of state information. This first layer of state information is transmitted to the next reservoir layer through an optical waveguide, and the next reservoir layer continues to perform nonlinear mapping based on the previous layer's state information. After the received signal is transmitted and mapped layer by layer among multiple reservoir layers, the multi-layer cascaded optical deep reservoir subnetwork can obtain a stronger ability to extract complex channel features. The use of optical waveguides to connect adjacent reservoir layers can reduce the inter-layer photoelectric conversion process, thereby reducing additional power consumption.

[0033] For ease of explanation, the hierarchical state of the multi-layer cascaded optical depth reservoir sub-network can be represented as: in, For the first The state vector of the layered reservoir, For the first Input to the layered storage pool, For the first The input weight matrix of the layered reserve pool, For the first Internal connection weight matrix of the layered reserve pool This is a non-linear activation function. Between two adjacent pool layers, the input of the next layer can be provided by the state vector of the previous layer, i.e.: In practice, the multi-layer cascaded optical depth reservoir subnetwork can be trained layer by layer. That is, the output weights corresponding to the first reservoir layer are first determined, then the parameters of the first reservoir layer are fixed and the output weights corresponding to the next reservoir layer are determined, and so on, until the training of multiple reservoir layers is completed. This method avoids global backpropagation training of the entire multi-layer network, reducing the computational overhead of the training process.

[0034] In this embodiment, the surface of the silicon-based microring resonator in at least one reservoir layer can be covered with a material having a negative thermo-optic coefficient. Silicon-based microring resonators are prone to resonance drift under temperature changes. The material with the negative thermo-optic coefficient can compensate for the thermo-optic effect of the silicon-based microring resonator, reducing the impact of temperature changes on the resonance state, thereby reducing the power consumption of the temperature control circuit and improving the stability of the reservoir state.

[0035] The multi-input multi-output (MIMO) optical reservoir subnetwork comprises multiple input channels, each connected to different channel characteristic parameters. These channel characteristic parameters may include at least one of the following: received signal strength, phase change rate, multipath delay distribution, signal-to-noise ratio (SNR) estimate, and environmental attenuation parameters. After different channel characteristic parameters are injected into the MIMO optical reservoir subnetwork through their respective input channels, they undergo nonlinear fusion within the reservoir to form a fused state containing multiple types of channel state information. This fused state reflects the changes in received signals under complex channel conditions, enabling the channel equalization unit and frequency offset estimation and compensation unit to process data based on more comprehensive state characteristics.

[0036] Multiple channel feature parameters can form an input vector: in, Let be the multi-channel input vector for the t-th sampling period. Let M be the channel characteristic parameters corresponding to the m-th input channel, where M is the number of input channels. (Multi-Input Multiple-Output Optical Reservoir Subnetwork) After performing nonlinear mapping, fused state information is output. This fused state information, as one of the state information in the reserve pool collected by the state acquisition unit, is used together with the state information output by other reserve pool sub-networks for channel equalization and frequency offset estimation compensation.

[0037] The channel equalization unit is connected to the state acquisition unit and obtains the equalized signal based on the reservoir state information output by the state acquisition unit. After the state acquisition unit transmits the acquired reservoir state information to the channel equalization unit, the channel equalization unit compensates for channel distortion in the received signal based on this reservoir state information to reduce the impact of multipath effects, link losses, and noise disturbances on the received signal. The channel equalization unit can process the reservoir state information using a pre-determined equalization output weight matrix to output the equalized signal.

[0038] The frequency offset estimation and compensation unit includes a coarse frequency offset estimation unit, a fine frequency offset estimation unit, a fusion unit, and a time-domain compensation unit. The coarse and fine frequency offset estimation units are connected to the state acquisition unit, respectively. The fusion unit connects the coarse, fine, and time-domain compensation units, and the time-domain compensation unit is connected to the signal output unit. The coarse frequency offset estimation unit obtains a coarse frequency offset estimate based on the reservoir state information corresponding to the received signal. The fine frequency offset estimation unit obtains a fine frequency offset estimate based on the reservoir state information corresponding to the signal after coarse frequency offset compensation. The fusion unit merges the coarse and fine frequency offset estimates to obtain a total frequency offset estimate. The time-domain compensation unit performs time-domain frequency offset compensation on the received signal based on the total frequency offset estimate. Specifically, the coarse frequency offset estimation unit can obtain a coarse frequency offset estimate based on the reservoir state information corresponding to the primary and secondary synchronization signals within a single synchronization signal block time slot. The coarse frequency offset estimate is used to perform preliminary frequency offset compensation on the received signal, ensuring that the residual frequency offset in the compensated signal falls within the range that the fine frequency offset estimation unit can further estimate. The fine frequency offset estimation unit further extracts the reservoir state information corresponding to the residual frequency offset based on the signal after coarse frequency offset compensation, and obtains the fine frequency offset estimate. The fusion unit fuses the coarse and fine frequency offset estimates to form the total frequency offset estimate, and the time-domain compensation unit performs frequency offset compensation on the received signal based on the total frequency offset estimate.

[0039] In practical applications, after the receiver receives a signal from the satellite communication link, the signal input unit sends the received signal to the optical reservoir computing network. Multiple reservoir sub-networks within the optical reservoir computing network perform nonlinear mapping on the received signal or channel characteristic parameters. The state acquisition unit collects reservoir state information from the state output side of each reservoir sub-network and forms a state vector for subsequent processing. The channel equalization unit outputs the equalized signal based on the state vector, and the frequency offset estimation and compensation unit generates coarse and fine frequency offset estimates based on the state vector, fusing them to obtain the total frequency offset estimate. The time-domain compensation unit uses the total frequency offset estimate to compensate the received signal, and the compensated signal is output by the signal output unit. In this embodiment, the signal input unit, optical reservoir computing network, state acquisition unit, channel equalization unit, and frequency offset estimation and compensation unit form a continuous received signal processing structure. The optical reservoir computing network performs nonlinear dynamic mapping on the received signal. The state acquisition unit provides the reservoir state information to the channel equalization unit and the frequency offset estimation and compensation unit respectively, enabling the device to complete channel equalization and frequency offset estimation and compensation based on the same state characteristics, thereby improving the stability of received signal processing under conditions of large frequency offset and low signal-to-noise ratio.

[0040] Example 2:

[0041] Based on the apparatus of Embodiment 1, this application also proposes a channel equalization and frequency offset estimation method based on an optical reservoir, such as... Figure 3As shown, this method first extracts the reservoir state information corresponding to the received signal, and then performs channel equalization and frequency offset estimation based on the reservoir state information. This allows the channel equalization and frequency offset estimation to share the same state feature basis, thereby reducing the repetitive feature extraction process and improving the stability of received signal processing under Doppler frequency offset and low signal-to-noise ratio conditions.

[0042] The method includes the following steps: S1. Acquire the received signal; In this embodiment, the received signal is the baseband received signal obtained by the receiver in a 5G NTN communication scenario. After transmission through the satellite communication link, the received signal may simultaneously contain Doppler frequency offset, channel distortion, multipath interference, and noise disturbance. After acquiring the received signal, the receiver performs down-conversion, sampling, normalization, or segmentation processing on the received signal to make it suitable for input into the optical reservoir computing network. This step is used to obtain the signal to be processed required for subsequent channel equalization and frequency offset estimation.

[0043] S2. The received signal is input into the optical reservoir computing network to extract reservoir state information. In this embodiment, after the received signal is input into the optical reservoir computing network, the network generates reservoir state information through nonlinear dynamic mapping. The reservoir state information originates from a single-feedback-loop optical reservoir subnetwork, a multi-layer cascaded optical depth reservoir subnetwork, or a multi-input multi-output optical reservoir subnetwork, or can be formed by combining the state information output from multiple reservoir subnetworks. In this step, the optical reservoir computing network does not directly output the final demodulated data, but first maps the received signal into reservoir state information containing time dynamic characteristics, nonlinear response characteristics, and historical state information. For received signals with multipath fading, low signal-to-noise ratio, and Doppler frequency offset, the reservoir state information can reflect the changes in the received signal after being affected by the channel, and provide a common data basis for subsequent channel equalization and frequency offset estimation.

[0044] S3. Channel equalization is performed on the received signal based on the reservoir state information to obtain the equalized signal. In this embodiment, the channel equalization process constructs a state vector from the reservoir state information and performs matrix operations on the state vector with a pre-determined equalization output weight matrix to obtain the equalized signal. The equalized signal can be represented as: in, The signal after equalization To balance the output weight matrix, This represents the state vector of the reservoir. The above process means that the optical reservoir computing network is responsible for mapping the received signal to a high-dimensional state space, and the equalization output weight matrix is ​​responsible for extracting the output result corresponding to the desired equalization signal from the high-dimensional state space. In this way, the channel equalization process does not need to directly rely on a single correlation peak or a single linear filtering result, but instead utilizes the state characteristics formed by the received signal in the reservoir network to compensate for distortions caused by multipath fading, link loss, and noise disturbances.

[0045] The equilibrium output weight matrix can be predetermined using training samples. During training, the desired equilibrium signal Y is used as the target output, and the reservoir state matrix X is used as the input state. Ridge regression is employed to determine the equilibrium output weight matrix. The regularized loss function of ridge regression can be expressed as: in, For regularization parameters, This represents the target output signal, i.e., the desired signal after equalization. This training method primarily determines the equalization output weight matrix. Furthermore, the state mapping structure within the optical reservoir computing network can remain relatively fixed, thus avoiding complex global backpropagation training of the entire optical reservoir computing network. During online processing, the channel equalization process only requires inputting the real-time acquired reservoir state vector into the predetermined equalization output weight matrix to obtain the equalized signal, thereby reducing the computational complexity of channel equalization processing.

[0046] Furthermore, in complex channel environments, the channel equalization process can also incorporate a multi-input multi-output (MIMO) optical reservoir subnetwork. This subnetwork includes multiple input channels, each connected to channel characteristic parameters such as received signal strength, phase change rate, multipath delay distribution, and signal-to-noise ratio (SNR) estimate. After these different channel characteristic parameters enter the subnetwork, they undergo nonlinear fusion within the reservoir to form fused state information. This fused state information, along with the reservoir state information corresponding to the received signal, serves as the state basis for the channel equalization process, enabling it to compensate for changes in the received signal based on current channel conditions.

[0047] To improve the adaptability of the channel equalization process to channel changes, the hyperparameters of the optical reservoir computing network are adjusted based on real-time channel state information. Let the network hyperparameter vector be... The learning rate is The instantaneous bit error rate estimate is The update process of network hyperparameters can then be represented as: Network hyperparameters include at least one of injection strength, feedback gain, feedback delay time, or angular frequency detuning. Through these adjustments, the optical reservoir computing network can adjust its state mapping process according to changes in channel state, thereby improving the stability of channel equalization results.

[0048] Furthermore, the channel equalization process employs a recursive least squares algorithm to update the output weight matrix online. Let... Let be the output weight matrix updated after the t-th sampling period. This is the output weight matrix of the previous sampling period. For the gain vector, To represent the output error, the update process of the output weight matrix can be expressed as: Among them, the output error The channel equalization process is determined based on the difference between the equalized output signal and the desired output signal. By updating the output weight matrix online, the channel equalization process can adapt to changes in channel conditions such as multipath effects, rain attenuation, or atmospheric scattering without retraining the entire optical reservoir computing network.

[0049] S4. Frequency offset estimation is performed based on the same reservoir state information to obtain the frequency offset estimate. In this embodiment, the frequency offset estimate is generated based on the reservoir state information corresponding to the primary synchronization signal (PSS) and secondary synchronization signal (SSS) within a single synchronization signal block (SSB) time slot. Specifically, the receiver acquires the received signal segments corresponding to the PSS and SSS within one SSB time slot and inputs these received signal segments into the optical reservoir calculation network. The optical reservoir calculation network performs nonlinear mapping on the phase changes, frequency offsets, and timing characteristics contained in the PSS and SSS, and the state acquisition unit extracts the corresponding reservoir state information. The frequency offset estimation compensation unit generates the frequency offset estimate based on the above reservoir state information. Since this frequency offset estimation process only uses the corresponding state information of the PSS and SSS within a single SSB time slot, it does not need to cache synchronization signals in multiple SSB time slots and perform cross-time slot correlation processing. Therefore, it can reduce the processing delay caused by cross-time slot caching and also reduce the impact of channel state changes or time deviations between different time slots on the frequency offset estimation result.

[0050] The generation of frequency offset estimates includes a coarse estimation stage and a fine estimation stage, such as... Figure 4 As shown, the coarse estimation stage is used to obtain a wide range of frequency offset estimation results from the received signal, so that subsequent processing can first eliminate the main frequency offset components; the fine estimation stage is used to further estimate the residual frequency offset based on the coarse frequency offset compensation, thereby improving the frequency offset estimation accuracy.

[0051] In the coarse estimation stage, the received signal is input into the optical reservoir computing network, and the coarse frequency offset estimate is obtained using the reservoir state information. Let the state vector extracted in the coarse estimation stage be... The coarse-frequency biased output weight matrix obtained through pre-training is: Then the coarse frequency bias estimate can be expressed as: The coarse frequency offset estimate corresponds to the main Doppler frequency offset component in the received signal. It is used to perform preliminary frequency offset compensation on the received signal so that the residual frequency offset in the compensated signal falls into the range that can be further estimated in the fine estimation stage.

[0052] In the fine estimation stage, the signal after coarse frequency offset compensation is input again into the optical reservoir computing network, which extracts the reservoir state information corresponding to the residual frequency offset. Let the state vector extracted in the fine estimation stage be... If the pre-trained fine-frequency bias output weight matrix is ​​given, then the fine-frequency bias estimate can be expressed as: The fine frequency offset estimate is used to characterize the residual frequency offset that still exists after coarse frequency offset compensation. By inputting the value back into the optical reservoir computational network after coarse compensation, the fine estimation stage can extract more detailed frequency offset features within a smaller frequency offset range, thereby improving the accuracy of the final frequency offset estimate.

[0053] The coarse frequency offset estimate and the fine frequency offset estimate are fused to obtain the total frequency offset estimate, which can be expressed as: in, This is the estimated value of the total frequency offset. This is a dynamic weighting factor. The dynamic weighting factor can be adjusted based on the signal-to-noise ratio estimate, or based on the confidence levels of the coarse and fine frequency offset estimation results. By fusing the coarse and fine frequency offset estimates, the ability to capture a wide range of frequency offsets in the coarse estimation stage can be retained, while the ability to refine the residual frequency offset in the fine estimation stage can be utilized, thus balancing the estimation range and accuracy of the frequency offset.

[0054] After obtaining the total frequency offset estimate, the receiver adapts the frequency offset estimation process to the current channel state. Specifically, when the receiver detects a decrease in signal-to-noise ratio, changes in multipath delay distribution, fluctuations in received signal strength, or increased environmental attenuation, these channel state characteristics are input as auxiliary information into the optical reservoir computing network. This allows the optical reservoir computing network to consider both the phase change of the received signal itself and the changes in the current channel environment when extracting residual frequency offset features. This approach does not change the basic process of coarse estimation, fine estimation, and fusion of the two-stage estimation results. Instead, it introduces auxiliary features related to the current channel environment during the reservoir state generation process, thereby improving the adaptability of the frequency offset estimation results to complex link changes. When the frequency offset estimation error increases or the residual frequency offset changes rapidly, the receiver can also adjust the injection strength, feedback gain, feedback delay time, or angular frequency detuning of the optical reservoir computing network to match the nonlinear response state of the optical reservoir computing network with the current degree of frequency offset change, thereby improving the stability of frequency offset estimation under low signal-to-noise ratio and dynamic channel conditions.

[0055] S5. Perform frequency offset compensation on the received signal based on the frequency offset estimate. In this embodiment, time-domain frequency offset compensation is performed based on the total frequency offset estimate. If the received signal is represented as... The sampling interval is The compensated signal can then be expressed as: in, This is the signal after frequency offset compensation. The compensation process uses a phase rotation factor corresponding to the frequency offset estimate to cancel the frequency shift in the received signal, thus attenuating the Doppler frequency offset. The frequency offset-compensated signal can then proceed to subsequent demodulation, decoding, or other baseband processing stages.

[0056] In this embodiment, the receiver first acquires the received signal and inputs it into the optical reservoir computing network. The optical reservoir computing network performs a nonlinear mapping on the received signal, and the state acquisition unit extracts the reservoir state information. Subsequently, the reservoir state information is used for channel equalization to obtain the equalized signal, and for frequency offset estimation to obtain coarse and fine frequency offset estimates, which are then fused to obtain the total frequency offset estimate. Finally, time-domain frequency offset compensation is performed on the received signal based on the total frequency offset estimate. Through the above method, channel equalization and frequency offset estimation can share the same reservoir state extraction process. The primary and secondary synchronization signals within a single synchronization signal block time slot can provide the state basis for frequency offset estimation, thereby reducing cross-time slot buffering and repeated feature extraction, and improving the reliability of received signal processing under Doppler frequency offset and low signal-to-noise ratio conditions.

[0057] To verify the processing effect of the channel equalization and frequency offset estimation method based on optical reservoir in this embodiment, simulation tests were conducted using a Rayleigh multipath channel as specified by 3GPP. The maximum Doppler frequency offset was set to 52kHz, the frequency offset change rate was set to 750Hz / s, and the signal-to-noise ratio range was set to -8dB to 4dB. Figure 5 The above are the acquisition probability curves of the present invention under different signal-to-noise ratio conditions. Figure 6 The present invention provides frequency offset estimation error curves under different signal-to-noise ratio conditions. Figure 7 This shows the bit error rate curves of the present invention under different signal-to-noise ratio conditions. Figure 5 It can be seen that the acquisition probability of this invention gradually increases with the increase of the signal-to-noise ratio; at a signal-to-noise ratio of -2dB, the acquisition probability of this invention can reach 0.99, which is higher than the acquisition probability of the traditional overall cross-correlation method under the same conditions. Figure 6 It can be seen that as the signal-to-noise ratio (SNR) increases, the overall frequency offset estimation error of this invention decreases; at an SNR of -4dB, the average frequency offset estimation error of this invention is approximately 0.5kHz, which is lower than the average frequency offset estimation error of the traditional overall cross-correlation method under the same conditions. Figure 7 It can be seen that the overall bit error rate of this invention decreases as the signal-to-noise ratio (SNR) increases. The above results demonstrate that this invention can achieve relatively stable channel equalization and frequency offset estimation compensation under conditions of Doppler frequency offset and low SNR.

[0058] Example 3:

[0059] Based on the channel equalization and frequency offset estimation device based on optical reservoir provided in Embodiment 1, this embodiment further illustrates the application process of this device in a satellite intelligent communication system. This embodiment is aimed at 5G NTN communication scenarios under conditions of large frequency offset and low signal-to-noise ratio. The satellite intelligent communication system includes a signal transmission intelligent unit and a signal reception intelligent unit. The signal reception intelligent unit is equipped with the channel equalization and frequency offset estimation device based on optical reservoir as described in Embodiment 1, and performs channel equalization and frequency offset estimation compensation on the received signal according to the method in Embodiment 2.

[0060] The intelligent signal transmission unit includes an optical depth reservoir calculation model, a Multiple Input Multiple Output (MIMO) antenna array, and an intelligent frequency offset pre-compensation module. The optical depth reservoir calculation model, serving as the core of intelligent processing at the transmitter, employs a multi-layer cascaded structure. Each layer uses a silicon-based microring resonator as a nonlinear node and generates a nonlinear dynamic response through a delay feedback mechanism to preprocess the signal to be transmitted. This optical depth reservoir calculation model corresponds logically to the multi-layer cascaded optical depth reservoir sub-network in Example 1, both utilizing a multi-layer reservoir structure to perform layer-by-layer nonlinear mapping of the signal.

[0061] MIMO antenna arrays are used to improve channel capacity and interference immunity through spatial multiplexing and beamforming techniques. For ease of explanation, the transmit signal model of a MIMO system can be represented as: in, This represents the transmit signal vector of the MIMO antenna array. The original data stream vector, This is the precoding matrix. The optical depth reservoir computing network optimizes and generates signals in real time based on channel state information, enabling the transmitted signal to adjust its spatial transmission characteristics according to the channel conditions. This part belongs to the auxiliary enhancement design at the transmitter end, which is used to improve the signal state before entering the satellite communication link and works in conjunction with the channel equalization and frequency offset estimation compensation at the receiver end.

[0062] The intelligent frequency offset pre-compensation module is embedded in the optical depth reservoir computing network. Based on ephemeris information and channel estimation results, the transmitter uses the reservoir network to predict the Doppler frequency offset trend and pre-compensates the baseband signal. For ease of explanation, the frequency offset pre-compensation of the baseband signal can be expressed as: in, This is the baseband signal after frequency offset pre-compensation. The baseband signal before pre-compensation. This is the pre-estimated frequency offset value output by the reservoir network. Through this pre-compensation process, the transmitter can reduce some of the Doppler frequency offset effects before signal transmission, thereby reducing the processing pressure of frequency offset estimation and compensation at the receiver. The receiver still performs frequency offset estimation and time-domain compensation based on the reservoir state information corresponding to the received signal, according to the method in Example 2.

[0063] The intelligent signal transmission unit can also jointly optimize the optical depth reservoir network parameters and the MIMO precoding matrix, enabling the transmitter to adaptively adjust the transmit power, beam direction, and frequency offset precompensation amount. This joint optimization process allows the transmitter's reservoir network parameters, spatial transmission parameters, and frequency offset precompensation parameters to be coordinated around the current channel state, thereby improving the transmit signal's adaptability to Doppler shift and time delay spread.

[0064] In this embodiment, the intelligent signal receiving unit includes an adaptive noise suppression mechanism and a dynamic channel tracking module. The adaptive noise suppression mechanism adjusts the injection strength and feedback gain of the optical reservoir computing network based on the real-time signal-to-noise ratio (SNR) estimate, enabling the optical reservoir computing network to maintain an appropriate nonlinear response state under low SNR conditions. When the real-time SNR is low, the receiver adjusts the injection strength and feedback gain to retain more stable dynamic characteristics of the received signal, reducing the impact of random noise on the state extraction results. When the real-time SNR is high, the receiver can enable the optical reservoir computing network to retain more detailed changes, thereby improving the precision of channel equalization and frequency offset estimation. This adaptive noise suppression mechanism adjusts the state generation process of the optical reservoir computing network without changing the basic connection relationships between the signal input unit, optical reservoir computing network, state acquisition unit, channel equalization unit, and frequency offset estimation compensation unit in Embodiment 1.

[0065] The dynamic channel tracking module utilizes the short-time memory characteristic of the optical reservoir computing network to continuously monitor changes in the channel state of the received signal and update the output weights based on the reservoir state information and output error. When the channel environment of the ground receiver changes, such as changes in multipath distribution, increased rain attenuation, intensified atmospheric scattering, or changes in the receiver's motion state, the dynamic channel tracking module can adjust the output weights corresponding to the channel equalization unit to ensure that the equalized signal matches the current channel state. For the frequency offset estimation and compensation unit, the dynamic channel tracking module can adjust the output weights related to frequency offset estimation based on changes in residual frequency offset, enabling the frequency offset estimation and compensation process to adapt to continuous changes in Doppler frequency offset.

[0066] In practical applications, the satellite transmits downlink signals to the ground receiver. After passing through the satellite communication link, the signal is received at the ground receiver with channel distortion, noise disturbance, and Doppler frequency offset. The ground receiver first preprocesses the received signal through the signal input unit, and then inputs the received signal into the optical reservoir computing network. The optical reservoir computing network extracts reservoir state information, and the state acquisition unit provides this information to the channel equalization unit and the frequency offset estimation and compensation unit. The channel equalization unit outputs the equalized signal based on the reservoir state information, and the frequency offset estimation and compensation unit obtains the frequency offset estimate based on the reservoir state information and performs time-domain frequency offset compensation based on the frequency offset estimate. Simultaneously, an adaptive noise suppression mechanism adjusts the injection strength and feedback gain of the optical reservoir computing network based on the real-time signal-to-noise ratio estimate, and the dynamic channel tracking module updates the output weights according to changes in the received signal state, enabling the ground receiver to continuously adjust the reception processing as the channel state changes.

[0067] Through the above application methods, the channel equalization and frequency offset estimation device based on optical reservoir in Example 1 can be used as the core processing structure of the ground receiver in a satellite intelligent communication system. This device completes channel equalization and frequency offset estimation compensation based on the same reservoir state information, which can reduce the repetitive feature extraction and processing delays caused by the decentralized processing of channel equalization and frequency offset estimation. Simultaneously, through an adaptive noise suppression mechanism and a dynamic channel tracking module, it can improve the stability of received signal processing in low signal-to-noise ratio, Doppler frequency offset, and complex channel environments.

[0068] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A channel equalization and frequency offset estimation device based on an optical reservoir, comprising a signal input unit, an optical reservoir computing network, a state acquisition unit, a channel equalization unit, a frequency offset estimation and compensation unit, and a signal output unit, characterized in that: The output of the signal input unit is connected to the input of the optical reservoir computing network; the state acquisition unit is located on the state output side of the optical reservoir computing network and is connected to the channel equalization unit and the frequency offset estimation compensation unit respectively; the signal output unit is connected to the output side of the channel equalization unit and the frequency offset estimation compensation unit.

2. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 1, characterized in that: The optical reservoir computing network includes at least two reservoir sub-networks, and the at least two reservoir sub-networks include at least two of the following: a single feedback loop optical reservoir sub-network, a multi-layer cascaded optical depth reservoir sub-network, and a multi-input multi-output optical reservoir sub-network.

3. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 2, characterized in that: The status acquisition unit is connected to the status output side of at least two of the reserve pool sub-networks, and acquires the reserve pool status information output by at least two of the reserve pool sub-networks.

4. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 2, characterized in that: The single-feedback-loop optical reservoir subnetwork includes a silicon-based microring resonator and a single-feedback loop coupled to the silicon-based microring resonator.

5. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 2, characterized in that: The multi-layer cascaded optical depth reservoir subnetwork includes multiple reservoir layers connected in sequence, with adjacent reservoir layers connected by optical waveguides.

6. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 2, characterized in that: The multiple input multiple output optical reservoir subnetwork includes multiple input channels, each of which is connected to different channel characteristic parameters.

7. The channel equalization and frequency offset estimation device based on optical reservoir according to claim 1, characterized in that: The frequency offset estimation and compensation unit includes a coarse frequency offset estimation unit, a fine frequency offset estimation unit, a fusion unit, and a time-domain compensation unit. The coarse frequency offset estimation unit and the fine frequency offset estimation unit are respectively connected to the state acquisition unit. The fusion unit is connected between the coarse frequency offset estimation unit, the fine frequency offset estimation unit, and the time-domain compensation unit. The time-domain compensation unit is connected to the signal output unit.

8. A channel equalization and frequency offset estimation method based on optical reservoir, characterized in that: The method includes the following steps: S1. Acquire the received signal; S2. Input the received signal into the optical reservoir computing network to extract the reservoir status information; S3. Perform channel equalization on the received signal based on the reservoir status information to obtain the equalized signal; S4. Based on the same state information of the reserve pool, frequency offset estimation is performed to obtain the frequency offset estimate; S5. Perform frequency offset compensation on the received signal based on the frequency offset estimate.

9. The channel equalization and frequency offset estimation method based on optical reservoir according to claim 8, characterized in that: In S3, the state information of the reserve pool is used to form a state vector, and the state vector is used to perform matrix operations with a pre-determined equalization output weight matrix to obtain the equalized signal.

10. The channel equalization and frequency offset estimation method based on optical reservoir according to claim 8, characterized in that: In S4, the frequency offset estimate is generated based on the reserve pool status information corresponding to the main synchronization signal and the auxiliary synchronization signal within a single synchronization signal block time slot. Generating the frequency offset estimate includes a coarse estimation stage and a fine estimation stage. The coarse estimation stage obtains a coarse frequency offset estimate, and the fine estimation stage obtains a fine frequency offset estimate. The coarse frequency offset estimate and the fine frequency offset estimate are fused to obtain a total frequency offset estimate, and time-domain frequency offset compensation is performed based on the total frequency offset estimate.

Images