A semi-dynamic scheduling method and system based on time-varying frequency hopping

Abstract

The application discloses a semi-dynamic scheduling method based on time-varying frequency hopping. For DL SPS and UL CG services, a base station introduces a semi-dynamic scheduling mechanism, taking a Hopping Pattern four-tuple model as the core, to jointly control time domain, frequency domain, space domain and MCS. The base station predicts CSI, beams and QoS based on an AI / ML module, autonomously decides whether to update the Hopping Pattern, and can classify UEs with weak channel correlation into the same multicast group for unified distribution, to realize space division multiplexing and free hopping of intra-group UEs in a frequency domain resource pool. The Hopping Pattern is carried by a semi-dynamic scheduling SDS dedicated MAC CE, a bitmap is used to indicate the hopping dimension on demand, and the semi-dynamic scheduling overhead is reduced. After receiving the SDS dedicated MAC CE, a UE performs multi-dimensional resource hopping from a specified period, and uses the latest valid Hopping Pattern before receiving a new indication.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and specifically to a semi-dynamic scheduling method and system based on time-varying frequency hopping. Background Technology

[0002] In 5G NR systems, downlink semi-static scheduling (SPS) and uplink pre-configured authorization (CG) are the main scheduling mechanisms for periodic services. SPS completes initial configuration via RRC signaling and, after activation by PDCCH, the UE automatically transmits and receives data at the same resource location at fixed intervals, without requiring dynamic scheduling each time. CG is divided into two types: Type 1 and Type 2. The former completes all parameter settings through pure RRC configuration and takes effect immediately, while the latter requires further activation or release via PDCCH. These mechanisms are widely used in periodic small-packet services such as VoIP and industrial IoT.

[0003] However, the aforementioned mechanisms exhibit significant limitations in 6G Ultra-Reliable Low-Latency Communication (HRLLC) scenarios. Firstly, both SPS and CG employ static, fixed resource reservation, failing to dynamically adjust time, frequency, and spatial resource allocation based on channel conditions. This results in low spectral efficiency and an inability to effectively combat frequency-selective fading. Secondly, the strictly periodic transmission opportunities of SPS lead to a mismatch between packet arrival times and transmission opportunities, causing latency jitter. In contrast, existing HARQ retransmission mechanisms have a fixed retransmission delay equal to the SPS period, reaching up to 160ms, far exceeding the 0.1ms to 1ms air interface latency target requirement of 6G HRLLC. Furthermore, existing mechanisms rely on semi-static configuration for MCS adjustment, lacking link adaptive capabilities based on real-time channel conditions, which is insufficient when reliability targets are increased from 10⁻⁻⁶. 5 Upgraded to 10⁻ 7 In 6G scenarios, relying solely on redundant retransmissions is extremely inefficient. These issues collectively limit the applicability of 5G SPS and CG mechanisms in 6G HRLLC service scenarios, necessitating a novel scheduling method that can retain the low signaling overhead advantages of semi-static scheduling while achieving dynamic adaptation of multi-dimensional resources. Summary of the Invention

[0004] To overcome the existing problems and shortcomings, this invention proposes a semi-dynamic scheduling method based on time-varying frequency hopping, applied to wireless communication networks, specifically:

[0005] When the base station configures DL SPS or UL CG for the UE, the initial configuration information includes a default Hopping Pattern configuration, with the default mode being non-hopping mode;

[0006] The base station groups the UEs that need to update their Hopping Pattern based on AI / ML prediction results, builds a multicast group, and uniformly distributes the Hopping Pattern to the UEs in the same group.

[0007] The Hopping Pattern employs a multi-dimensional quadruple model, which is defined as follows:

[0008] Hopping(i)={t_hop[i],f_hop[i],s_hop[i],m_hop[i]};

[0009] Where i is the business cycle number, t_hop[i] is the time domain jump parameter, f_hop[i] is the frequency domain jump parameter, s_hop[i] is the spatial domain jump parameter, and m_hop[i] is the MCS jump parameter;

[0010] After receiving the Hopping Pattern, the UE performs multi-dimensional resource transitions according to the transition parameters of the multi-dimensional transition model starting from the specified service period i; before receiving a new Hopping Pattern, the UE continues to execute using the most recently received Hopping Pattern.

[0011] When the base station decides to update the Hopping Pattern, it notifies the UE via RRC message or MACCE. The UE applies the new Hopping Pattern to subsequent service cycles from the time it receives the new Hopping Pattern.

[0012] Furthermore, the dimensions of the quadruple model are defined as follows:

[0013] t_hop[i]={i_start,duration,COUNT}, where i_start is the time domain starting business cycle number when the jump takes effect, duration is the number of business cycles that keep resources unchanged between the current jump and the next jump, and COUNT is the maximum number of effective jumps in this Hopping Pattern;

[0014] f_hop[i]={nPRB_offset,size,COUNT}, where nPRB_offset is the offset of the starting position of the frequency domain PRB in this jump relative to the previous position, and size is the number of PRB resources allocated in this jump;

[0015] s_hop[i]={port_bitmap,TCI,COUNT}, where port_bitmap is the antenna port bitmap, used to specify the antenna port used this time, and TCI is the transmission control indicator, used to specify the DL or UL beam used this time;

[0016] m_hop[i]={MCS_index[CW],COUNT}, where MCS_index[CW] is the index of the MCS modulation and coding scheme corresponding to each CW.

[0017] Furthermore, the AI / ML prediction results include one or more of the following:

[0018] Based on historical CSI sequences, the channel matrix H of future time slots is predicted using AR models, LSTM, Transformer, or EKF.

[0019] Based on historical beam measurements, location information, and motion trajectories, reinforcement learning or GNN is used to predict the optimal beam pair index.

[0020] Based on real-time CQI, BLER and HARQ feedback, real-time QoS status prediction is performed within a time scale of 1ms to 10ms.

[0021] Furthermore, the update decision of the Hopping Pattern includes one or more of the following: QoS quality requirement monitoring results, channel BLER change trends, CSI prediction results, and beam prediction results;

[0022] The update decision-making method of the Hopping Pattern includes one of the following: unidirectional autonomous decision-making by the base station; or the base station making a decision after the UE provides prediction assistance information to the base station.

[0023] Furthermore, the Hopping Pattern supports both fast frequency hopping mode and slow frequency hopping mode;

[0024] In fast frequency hopping mode, the UE transmits at different time-frequency resource locations in each service cycle;

[0025] In slow frequency hopping mode, the UE maintains the same resource location in multiple consecutive service cycles, and the hopping interval is configured by the duration parameter in t_hop[i].

[0026] The switching interval is dynamically configured by the base station based on the frequency selective fading monitoring results.

[0027] Furthermore, when the correct reception time of the Hopping Pattern is missed in a certain service cycle due to HARQ or ARQ retransmission, the UE checks whether a transition needs to be performed in the service cycle before the arrival of each service cycle i, and performs subsequent transmission based on the most recently successfully received Hopping Pattern until a new Hopping Pattern is received.

[0028] A semi-dynamic scheduling system based on time-varying frequency hopping includes:

[0029] On the base station side, including:

[0030] The AI / ML prediction unit is used to perform CSI prediction, beam prediction, and QoS prediction.

[0031] The grouping decision unit is used to perform multicast grouping for UEs that need to update their Hopping Pattern based on the prediction results;

[0032] Hopping Pattern Generation Unit, used to generate Hopping Patterns containing quadruples {t_hop, f_hop, s_hop, m_hop};

[0033] The SDS MAC CE encapsulation unit is used to encapsulate the Hopping Pattern into an SDS-specific SDS MAC CE and transmit it via a dedicated MAC PDU;

[0034] On the UE side, including:

[0035] The Hopping Pattern receiving and parsing unit is used to receive and parse the Hopping Pattern in the SDS MAC CE.

[0036] The jump execution unit is used to perform multi-dimensional resource jumps according to the four-tuple parameters starting from a specified business cycle i.

[0037] The business cycle monitoring unit is used to check whether a transition operation needs to be performed before each business cycle i arrives.

[0038] Furthermore, the SDS MAC CE uses an independent dedicated MAC PDU for transmission and is not multiplexed with DL-SCH data. The dedicated MAC PDU can carry multiple DL SPS or UL CG Hopping Pattern information.

[0039] The subheader of the SDS MAC CE includes an R field, an F field, an LCID field, and an eLCID field. The eLCID field distinguishes the value range of DL SDS MAC CE from the value range of UL SDS MAC CE. When DL and UL both require semi-dynamic scheduling and share the same subheader, the LCID value is located in the common interval. The L field in the subheader is divided into the DL part and the UL part, which respectively represent their length information.

[0040] Furthermore, the SDS MAC CE payload includes the following fields: Hopping dimensions bitmap field, used to indicate the dimensional information carried by the current MAC CE; Time domain Hopping duration field; PRBoffset field; PRB resource size field; Antenna port bitmap field; TCI state index field; and MCS field;

[0041] The Hopping dimensions bitmap uses a 4-bit plus 2-bit reserved format, and dimensions that do not jump do not carry corresponding information; when the Bits for CW1 field is 0, it implicitly indicates that only CW0 is used in this business data; when the T field is 00, it indicates that the TCI state index is not carried; when CW0 and CW1 use different MCS, the M field is set to 1, and when they use the same MCS, the M field is set to 0.

[0042] Furthermore, the base station configures a dedicated SDS LCID for the SDS MAC CE. The SDS LCID is distinct from the ordinary data LCID and is used for high-priority transmission. The SDS LCID supports dynamic allocation and is determined by the base station based on the LCID and priority mapping information pre-configured to the UE and then notified to the UE. The scheduling of the SDS MAC PDU adopts the same DCI Format as the ordinary MAC PDU, and the transmission reliability is ensured through the high aggregation level of CCE to reduce the probability of false detection and missed detection by the UE.

[0043] The semi-dynamic scheduling method and system based on time-varying frequency hopping provided by this invention have the following beneficial effects:

[0044] This invention uses a multi-dimensional quadrupole Hopping Pattern to jointly control the time domain, frequency domain, spatial domain, and MCS, enabling the UE to adaptively avoid channel fading in different service cycles, effectively reducing the block error rate, improving transmission reliability, reducing the number of retransmissions and the need for repeated transmissions, thereby reducing wireless transmission latency and ensuring bounded low jitter in latency.

[0045] This invention retains the advantages of low signaling overhead of SPS and CG semi-static scheduling, and uses a dedicated MAC CE for SDS to carry Hopping Pattern information. It also uses a Hopping dimensions bitmap to carry the dimension information that changes as needed, avoiding the transmission of redundant fields and achieving low-overhead semi-dynamic resource control signaling transmission.

[0046] This invention continuously predicts CSI, beam, and QoS through the base station-side AI / ML prediction module. Based on the prediction results, it autonomously decides whether to update the Hopping Pattern and assigns UEs with weak channel correlation to the same multicast group to uniformly distribute the Hopping Pattern, thereby realizing spatial multiplexing among UEs within the group and improving the overall spectrum efficiency of the system.

[0047] This invention supports both fast frequency hopping and slow frequency hopping modes, and can flexibly configure the hopping interval according to the monitoring results of frequency selective fading, taking into account both anti-interference performance and resource utilization efficiency, and is suitable for various 6G HRLLC service scenarios. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the overall process of a semi-static / pre-configured service based on semi-dynamic scheduling.

[0050] Figure 2 This is a schematic diagram of a semi-dynamic scheduling process;

[0051] Figure 3 This is a schematic diagram of the semi-dynamic scheduling SDSMACCE format;

[0052] Figure 4 This is a schematic diagram of the CSI prediction mechanism architecture;

[0053] Figure 5 This is a schematic diagram of the collaborative architecture between beam prediction and CSI prediction.

[0054] Figure 6 This is a schematic diagram of a time-scale QoS prediction framework.

[0055] Figure 7 A schematic diagram of a dedicated MACPDU format for semi-dynamic scheduling (independent sub-header format);

[0056] Figure 8 A schematic diagram of the MACPDU format for semi-dynamic scheduling (Common subheader format);

[0057] Figure 9 A schematic diagram of the SDSMACCE subheader format;

[0058] Figure 10 A schematic diagram showing the range of SDSLCID values;

[0059] Figure 11 This is a schematic diagram of the semi-dynamic scheduling MACCELCID mechanism. Detailed Implementation

[0060] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0061] In the description of this invention, the correspondence between Chinese names and English abbreviations is as follows:

[0062] UE (User Equipment)

[0063] XNBNode B base station;

[0064] DLDownlink downlink transmission;

[0065] ULUplink uplink transmission;

[0066] SPSSemi-Persistent Scheduling;

[0067] CGConfigured Grant configuration authorization;

[0068] SDSSemi-Dynamic Scheduling;

[0069] MACMedium Access Control;

[0070] CEControl Element control unit;

[0071] PDU (Protocol Data Unit)

[0072] RRCRadio Resource Control;

[0073] DCIDownlink Control Information;

[0074] LCID (Logical Channel ID) is the logical channel identifier.

[0075] eLCIDExtended Logical Channel ID;

[0076] PRBPhysical Resource Block;

[0077] MCSM modulation and coding scheme;

[0078] CWCodeword;

[0079] TCITransmission Configuration Indication;

[0080] CSIChannel State Information;

[0081] HARQHybrid Automatic Repeat Request;

[0082] ARQAutomatic Repeat Request automatically retransmits requests.

[0083] BLERBlock Error Rate;

[0084] CQIChannel Quality Indicator;

[0085] AI / ML (Artificial Intelligence / Machine Learning)

[0086] ARAuto-Regressive Model;

[0087] LSTMLong Short-Term Memory Network;

[0088] EKFExtended Kalman Filter;

[0089] GNNGraph Neural Network;

[0090] QoS Quality of Service;

[0091] PDBPacket Delay Budget;

[0092] Inertial Measurement Unit (IMU)

[0093] CCEControl Channel Element;

[0094] DL-SCHDownlink Shared Channel;

[0095] DUDistributed Unit;

[0096] CU-UPCentral Unit - User Plane;

[0097] OAMO Operation Administration and Maintenance;

[0098] HRLLC Hyper-Reliable and Low-Latency Communication;

[0099] URLLCUltra-Reliable Low-Latency Communication;

[0100] Example 1:

[0101] This embodiment describes the overall implementation of a semi-dynamic scheduling method based on time-varying frequency hopping, which is applied to the transmission scenarios of DL SPS and UL CG services between base stations and UEs in wireless communication networks.

[0102] like Figure 1 As shown, for DL ​​SPS semi-static scheduling and UL CG type services, the base station introduces the SDS mechanism to the UE, using Hopping Pattern as the core control method to achieve multi-dimensional joint transitions of time domain, frequency domain, spatial domain and MCS, to ensure reliable transmission of service flow, reduce the number of retransmissions and repeated transmission requirements, and ensure bounded low jitter of transmission delay.

[0103] When a base station configures DL SPS or UL CG for a UE, the initial configuration information includes a default Hopping Pattern configuration. The initial Hopping configuration information in the SPS / CG configuration is represented as Hopping(i) = (0, f, s, m), which defaults to non-hopping mode, meaning the domain hopping parameter is set to zero, and all dimension parameters retain their initial values. The service cycle number i is the most fundamental variable; changes in all other dimension resources are triggered by i. To avoid significant parameter processing overhead due to real-time calculation of Hopping at each i position, parameter updates and Hopping calculations are also managed in a semi-static manner.

[0104] like Figure 2 As shown, the base station continuously runs an AI / ML prediction module. Based on multiple prediction results, including CSI prediction, beam prediction, and QoS prediction, it comprehensively judges the channel state and service quality state of each UE and assesses whether the current Hopping Pattern used by each UE still meets the transmission reliability requirements. Based on this, it decides whether a new Hopping Pattern needs to be issued to a specific UE. When the base station decides to update the Hopping Pattern of a group of UEs, it assigns UEs with weak channel correlation characteristics to the same multicast group and generates a new Hopping Pattern for the same multicast group to achieve spatial multiplexing among UEs in the same group. The base station encapsulates the Hopping Pattern into a dedicated SDS MAC CE and issues it to the corresponding UE via a dedicated MAC PDU.

[0105] Hopping Pattern employs a multi-dimensional hop model, whose hop parameters include one or more of the following dimensions: time-domain hop parameter t_hop[i], frequency-domain hop parameter f_hop[i], spatial-domain hop parameter s_hop[i], and modulation and coding scheme (MCS) hop parameter m_hop[i], where i is the service cycle number. When the above four dimensions participate in the hop simultaneously, they form a four-tuple model Hopping(i) = {t_hop[i], f_hop[i], s_hop[i], m_hop[i]}. After receiving the SDSMACCE carrying Hopping Pattern, the UE parses it, extracts the hop parameters of each dimension, and performs multi-dimensional resource hops according to the parsed parameters starting from the service cycle i specified in the MACCE. Before each service cycle i arrives, the UE checks whether a hop needs to be performed for that service cycle. If a hop is required, the UE simultaneously determines and executes the hop content for each dimension.

[0106] Until a new Hopping Pattern is received, the UE continues to use the most recently received Hopping Pattern for resource hopping in subsequent service cycles. If the base station notifies the UE to stop frequency hopping via DCI, the UE will fall back to the default resource location or switch to the resource location specified by the base station according to the instruction. When the base station decides to update the Hopping Pattern, it notifies the UE via RRC message or MAC CE, and the UE applies the new Hopping Pattern to subsequent service cycles from the time it receives it.

[0107] Regarding the Hopping Pattern update decision, the input conditions for the decision include service-aware information such as CSI prediction results, beam prediction results, QoS prediction results, and network slice resource status prediction. The update decision supports two methods: one-way autonomous decision-making by the base station, and decision-making by the base station after the UE actively provides prediction assistance information to the base station; the latter is an optional enhancement method.

[0108] Example 2:

[0109] This embodiment, based on Embodiment 1, uses the four-tuple model in the Hopping Pattern multi-dimensional hopping model as an example to describe the parameters of each dimension of the Hopping Pattern in detail. In practical applications, one or more of the above dimensions can be selected for hopping control according to business requirements. The design principle of the Hopping Pattern is efficient and low-overhead indication, transmitting necessary information, with other information implicitly used for calculation, while also considering resource conflicts between UEs.

[0110] like Figure 3 As shown, the specific definitions of each dimension of the quadruple model are as follows.

[0111] t_hop[i] = {i_start, duration, COUNT}, where i_start represents the time-domain starting service period number where the hop takes effect; duration, or Time domain Hopping duration, represents the number of service periods between the current hop and the next hop where resources remain unchanged; and COUNT represents the maximum number of valid hops in this Hopping Pattern. The time-domain hop function controls the number of hop intervals in the time domain, adjusting the hop speed. For example, it can specify a fixed hop every X service periods, or use a variable interval associated with the service period number i to calculate whether a hop is needed for each service period. If a hop is needed, it also indicates which dimensions need to participate in the hop.

[0112] f_hop[i] = {nPRB_offset, size, COUNT}, where nPRB_offset is the PRB offset, representing the offset of the starting position of the frequency domain PRB in this jump relative to the previous position; size is the PRB resource size, representing the number of PRB resources allocated in this jump.

[0113] s_hop[i] = {port_bitmap, TCI, COUNT}, where port_bitmap is the antenna portbitmap, which specifies the antenna port to be used; TCI is the transmission control indicator TCI state, which specifies the DL or UL beam to be used.

[0114] m_hop[i] = {MCS_index[CW], COUNT}, where MCS_index[CW] is the index of the MCS modulation and coding scheme corresponding to each CW.

[0115] The comprehensive expression of the above quadruple is Hopping(i) = {(i_start, duration[], COUNT);(nPRB_offset, size)[]; (port_bitmap, TCI)[]; (MCS_index)[]}, which means that starting from the i-th service period, the UE determines the semi-dynamic resource allocation scheme for the current service period i and subsequent agreed periods according to the time domain window specified by duration, the number of periods specified by COUNT, the PRB starting position specified by nPRB_offset, the number of PRBs specified by size, the antenna port number specified by port_bitmap, the DL / UL beam specified by TCI, and the modulation and coding scheme specified by MCS_index.

[0116] Hopping Pattern supports two operating modes: fast frequency hopping and slow frequency hopping. In fast frequency hopping mode, the UE transmits at different time-frequency resource locations in each service cycle; in slow frequency hopping mode, the UE maintains the same resource location across multiple consecutive service cycles, with the hopping interval configured by the duration parameter in t_hop[i]. The base station dynamically configures the hopping interval based on the monitoring results of frequency-selective fading.

[0117] Regarding the error-tolerant handling of Hopping Pattern reception, when the correct reception time of Hopping Pattern is missed in a certain service cycle due to factors such as HARQ or ARQ retransmission, the UE checks whether a transition needs to be performed in the service cycle before the arrival of each service cycle i, and performs subsequent transmission based on the most recently successfully received Hopping Pattern until a new Hopping Pattern is received.

[0118] Example 3:

[0119] This embodiment, based on Embodiment 1, provides a detailed description of the AI / ML prediction mechanism upon which the base station drives SDS update decisions. The prediction types involved include three categories: CSI prediction, beam prediction, and QoS prediction.

[0120] Table 1 Comparison of Physical Layer CSI Prediction and Beam Prediction

[0121] Dimension CSI Forecast Beam prediction Predicted object Channel matrix / coefficients H(t)∈CNt×Nr Beam pair index b∗∈{1,...,B} or beamforming vector w Output granularity Fine-grained (subcarrier-level amplitude / phase) Coarse-grained (spatial direction / beam ID) Physical meaning Channel frequency response Spatial propagation direction (DoA / DoD) Time domain range Short (1-10 slots, subject to coherence time limitations) It can be relatively long (10ms-1s, based on trajectory prediction). Model Input Historical CSI sequence Historical beam measurement, location, IMU, environmental mapping Mathematical tools Linear prediction, LSTM, Kalman filtering Classification / Regression, Reinforcement Learning, Graph Neural Networks Core challenges High-dimensional complex sequence prediction Discrete space decision-making, robustness in multipath environments

[0122] As shown in Table 1, CSI prediction and beam prediction differ fundamentally in terms of prediction object, output granularity, physical meaning, time domain range, model input, and mathematical tools. CSI prediction targets the channel matrix H, representing fine-grained subcarrier-level amplitude and phase prediction. Its time domain range is relatively short, typically 1 to 10 time slots, and is limited by channel coherence time. The model input is a historical CSI sequence, and common mathematical tools include linear prediction, LSTM, and Kalman filtering. Beam prediction, on the other hand, targets the optimal beam pair index, representing coarse-grained spatial direction prediction. Its prediction time domain range can be longer, typically 10 ms to 1 s, and is based on trajectory prediction. Model inputs include historical beam measurement results, location information, IMU data, and environmental maps. Common mathematical tools include classification and regression models, reinforcement learning, and GNNs.

[0123] Table 2 CSI Prediction Model Classification

[0124] Case Target CSI slot UE usage history CSI Network Usage History CSI Priority illustrate 0 Present slot no no high Traditional instant feedback, no prediction 1 Present slot no yes Low The network side corrects the current CSI based on historical feedback. 2 Present slot yes no high The UE side uses historical measurements to predict the current CSI. 3 Future slots yes no high UE-side prediction of future CSI (core solution) 4 Future slots no yes Low Network side predicts the future based on historical feedback 5 Future slots yes yes Low Bilateral collaborative prediction of future CSI

[0125] As shown in Table 2, CSI prediction is categorized into six cases, from Case 0 to Case 5, based on whether the UE and network sides use historical CSI separately. Case 3 involves the UE side using historical measurement data to predict future time slot CSI, representing a true prediction scheme with the highest priority. Case 0 is the traditional real-time feedback mode, which does not rely on prediction. Case 2 involves the UE side using historical measurements to predict the current time slot CSI. Cases 1 and 4 involve the network side making corrections or predictions based on historical feedback, with lower priority. Case 5 is a bilateral collaborative prediction, also with lower priority. Cases 0, 2, and 3 do not rely on historical feedback from the network side, resulting in relatively low implementation complexity.

[0126] like Figure 4 As shown, AI / ML models are deployed on both the UE and base station sides. The UE-side model predicts the future channel matrix CSI h(t+Δ) based on the historical CSI sequence h(tk), while the base station-side model performs network-side prediction based on the historical feedback sequence h(tk). The two models can work together to improve prediction accuracy. Commonly used AI / ML CSI prediction model algorithms include: autoregressive prediction methods based on AR models, used as a baseline; prediction methods based on LSTM long short-term memory networks, currently the mainstream AI approach; prediction methods based on Transformers, considered advanced; and extended Kalman filtering methods based on EKF, a model-driven approach. Commonly used AI / ML beam prediction model algorithms include: beam prediction methods based on motion trajectories; LSTM prediction methods based on historical beam sequences; beam tracking methods based on reinforcement learning; and beam prediction methods based on GNNs.

[0127] like Figure 5 As shown, beam prediction provides spatial prior information, which can narrow the search space for CSI prediction and CSI feedback; CSI prediction provides channel quality information, assisting in the determination of beam prediction confidence; the goal of their joint optimization is to select the beam pair that maximizes the achievable rate of the predicted CSI. The SDS mechanism relies on the joint output results of beam prediction and CSI prediction to determine the Hopping decision in the spatial and temporal domains.

[0128] like Figure 6As shown, QoS prediction is divided into four layers according to the time scale. The first layer is the real-time prediction layer, with a time scale of 1ms to 10ms, which is predicted by DU / CU-UP based on real-time data such as CQI, BLER, and HARQ feedback. The second layer is the short-term prediction layer, with a time scale of 100ms to 1s, which is predicted by CU based on parameters such as QoS and PDB, using models such as ARIMA, LSTM, and Prophet. The third layer is the medium-term prediction layer, with a time scale of 1min to 60min, which uses models such as N-BEATS, DeepAR, and XGBoost for prediction. The fourth layer is the long-term prediction layer, with a time scale of 1 hour to more than 24 hours, which is predicted by the OAM layer. The SDS update decision is driven by the output results of the first-layer real-time prediction layer.

[0129] Example 4:

[0130] This embodiment provides a semi-dynamic scheduling system based on time-varying frequency hopping, and, based on the methods described in embodiments one to three, provides a detailed description of the complete design of the SDS-specific MAC CE, including the MAC PDU structure, subheader format, MAC CE payload structure, LCID design, and DCI Format design.

[0131] The system includes base station-side functional modules and UE-side functional modules. The base station-side module includes an AI / ML prediction unit, a packet decision unit, a Hopping Pattern generation unit, and an SDS MAC CE encapsulation unit. The AI / ML prediction unit performs CSI prediction, beam prediction, and QoS prediction. The packet decision unit performs multicast packet generation for UEs requiring Hopping Pattern updates based on the prediction results. The Hopping Pattern generation unit generates a Hopping Pattern containing one or more dimensional transition parameters, including a time-domain transition parameter t_hop, a frequency-domain transition parameter f_hop, a spatial-domain transition parameter s_hop, and an MCS transition parameter m_hop. The SDS MAC CE encapsulation unit encapsulates the Hopping Pattern into an SDS-specific MAC CE and transmits it via a dedicated MAC PDU. The UE side includes a Hopping Pattern Receiving and Parsing Unit, a Hopping Execution Unit, and a Service Cycle Monitoring Unit. The Hopping Pattern Receiving and Parsing Unit is used to receive and parse the Hopping Pattern in the SDS MACCE. The Hopping Execution Unit is used to perform multi-dimensional resource hopping according to the quadruple parameters starting from a specified service cycle i. The Service Cycle Monitoring Unit is used to check whether a hopping operation needs to be performed before each service cycle i arrives.

[0132] Regarding the MAC PDU structure, such as Figure 7 and Figure 8 As shown, the SDS MAC CE uses an independent dedicated MAC PDU for transmission and is not multiplexed with DL-SCH data, in order to adopt a reliable MCS transmission scheme and corresponding bearer resource allocation. The SDS MAC CE distinguishes between DL and UL directions. The DL SDS MAC CE is used for semi-dynamic adjustment of the resource hopping mode of DL SPS, and the UL SDS MAC CE is used for semi-dynamic adjustment of the resource hopping mode of UL CG. A single semi-dynamic dedicated MAC PDU can carry Hopping Pattern information for multiple DL SPS or UL CGs. The specific number depends on whether the UE has multiple corresponding service flows simultaneously and the monitoring and evaluation decision results of resource hopping mode updates.

[0133] Regarding the subheader format, such as Figure 9 As shown, the subheader of the SDS MAC CE includes the R field, F field, LCID field, and eLCID field. eLCID is an extended logical channel number LCID, which separately divides the LCID value range belonging to the DL SDS MAC CE and the LCID value range belonging to the UL SDS MAC CE; these two ranges do not overlap. For example... Figure 10 As shown, if multiple link directions such as DL and UL require semi-dynamic scheduling simultaneously and share the same subheader, the LCID value is located in the common interval, i.e., the eLCID interval C. For the common LCID interval, the L field in the subheader contains the length information of both DL and UL SDS MAC CE, that is, the L field is divided into two parts, for example, 4 bits for DL ​​and 4 bits for UL, or 1 byte for DL ​​and 1 byte for UL to represent their respective length information.

[0134] Regarding the MAC CE payload structure, as shown in Figure 3, the SDS MAC CE payload is used to carry Hopping Pattern information, and the design of each field is as follows. The Hopping dimensions bitmap field is used to indicate which resource dimensions' transition information is included in the current SDS MAC CE. It is recommended to use a 4-bit plus 2-bit reserved format. Dimension information that does not undergo transitions does not need to be transmitted to reduce signaling overhead. The Common field contains the i_start field and the Hopping COUNT field, which correspond to the starting service cycle number and the maximum number of valid transitions in the quadruple, respectively. The Time domain Hopping duration field indicates the number of continuous service cycles of this Hopping Pattern starting from the start position of this Hopping. When this field is set to 0, it means that the previous Hopping Pattern will continue to be used. For example, when the number of valid cycles of the Hopping Pattern exceeds the range that this field can represent, this field is set to 0. The PRB offset field and the PRB resource size field correspond to the nPRB_offset and size parameters in f_hop[i], respectively. The "Bits for CW1" field indicates the number of bits occupied by the antenna port of CW1. When this field is 0, it implicitly indicates that the service data only uses CW0, i.e., single-codeword transmission. The "Antenna port bitmap" field corresponds to the "port_bitmap" parameter in "s_hop[i]" and is used to specify the antenna port used this time. The "T" field indicates the number of TCI state indices. When T is 00, it indicates that no TCI state index is carried. In this case, the T bit is immediately followed by the MCS field. The TCI stateindex field is optional and corresponds to the TCI parameter in "s_hop[i]". The MCS field corresponds to the "MCS_index[CW]" parameter in "m_hop[i]". If there are two CW scheduling situations and CW0 and CW1 use different MCS, the M field is set to 1. If CW0 and CW1 use the same MCS, the M field is set to 0. Each Hopping Pattern entry contains the above fields. Multiple Hopping Pattern entries can be arranged sequentially in the same SDS MAC CE payload, up to the OctN byte position.

[0135] Regarding LCID design, such as Figure 11As shown, the base station configures a dedicated SDS LCID for the SDS MAC CE, dividing it into a dedicated LCID group to distinguish it from ordinary data LCIDs. This dedicated group is used for high-priority transmission and is granted higher-priority processing permissions within the UE. Sub-levels are further divided within the dedicated LCID group based on real-time and reliability requirements. The SDS LCID supports dynamic allocation; that is, the LCID corresponding to the SDS MAC CE is not a fixed value. It is determined by the base station based on the pre-configured LCID and priority mapping information given to the UE and notified to the UE. The UE identifies and parses the SDS MAC CE based on the allocated SDS LCID. The MAC CE format definition uses a Hopping dimensions bitmap to indicate which dimensions' transition information is carried by the current MAC CE. Dimensions that do not undergo transitions are not carried, enabling dynamic adaptive information transmission and reducing signaling overhead.

[0136] Regarding the DCI format design, the scheduling of SDS MAC PDUs uses the same DCI format as ordinary MAC PDUs for scheduling instructions. The transmission reliability of this DCI is ensured through CCE high aggregation level and other means to reduce the probability of false detection and missed detection by the UE, and to ensure that SDS control signaling can still reliably reach the UE when the radio channel conditions are poor.

[0137] It will be understood by those skilled in the art that the specific parameter values ​​described in the above embodiments are exemplary, and the scope of protection of the present invention is not limited to the specific values ​​described above. Without departing from the spirit and scope of the present invention, those skilled in the art can make various modifications and improvements to the above embodiments, and all such modifications and improvements fall within the scope of protection of the present invention.

[0138] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0139] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

Claims

1. A semi-dynamic scheduling method based on time-varying frequency hopping, applied to wireless networks, characterized in that: When the base station configures DL SPS or UL CG for the UE, the initial configuration information includes a default Hopping Pattern configuration, with the default mode being non-hopping mode; The base station groups the UEs that need to update their Hopping Pattern based on AI / ML prediction results, builds a multicast group, and uniformly distributes the Hopping Pattern to the UEs in the same group. The Hopping Pattern adopts a multi-dimensional jump model, which uses the business cycle number i as the basic variable. Its jump parameters include one or more of the following dimensions: time domain jump parameter t_hop[i], frequency domain jump parameter f_hop[i], spatial domain jump parameter s_hop[i], and modulation and coding scheme (MCS) jump parameter m_hop[i]. After receiving the Hopping Pattern, the UE performs multi-dimensional resource transitions according to the transition parameters of the multi-dimensional transition model starting from the specified service period i; before receiving a new Hopping Pattern, the UE continues to execute using the most recently received Hopping Pattern. When the base station decides to update the Hopping Pattern, it notifies the UE via RRC message or MACCE. The UE applies the new Hopping Pattern to subsequent service cycles from the time it receives the new Hopping Pattern.

2. The method according to claim 1, characterized in that, The jump parameters for each dimension of the multi-dimensional jump model are defined as follows: t_hop[i]={i_start,duration,COUNT}, where i_start is the time domain starting business cycle number when the jump takes effect, duration is the number of business cycles that keep resources unchanged between the current jump and the next jump, and COUNT is the maximum number of effective jumps in this Hopping Pattern; f_hop[i]={nPRB_offset,size,COUNT}, where nPRB_offset is the offset of the starting position of the frequency domain PRB in this jump relative to the previous position, and size is the number of PRB resources allocated in this jump; s_hop[i]={port_bitmap,TCI,COUNT}, where port_bitmap is the antenna port bitmap, used to specify the antenna port used this time, and TCI is the transmission control indicator, used to specify the DL or UL beam used this time; m_hop[i]={MCS_index[CW],COUNT}, where MCS_index[CW] is the index of the MCS modulation and coding scheme corresponding to each CW.

3. The method according to claim 1, characterized in that, The AI / ML prediction results include one or more of the following: Based on historical CSI sequences, the channel matrix H of future time slots is predicted using AR models, LSTM, Transformer, or EKF. Based on historical beam measurements, location information, and motion trajectories, reinforcement learning or GNN is used to predict the optimal beam pair index. Based on real-time CQI, BLER and HARQ feedback, real-time QoS status prediction is performed within a time scale of 1ms to 10ms.

4. The method according to claim 1, characterized in that, The update decision of the Hopping Pattern includes QoS quality monitoring or prediction results, channel BLER change trends, and one or more of CSI prediction results and beam prediction results; The update decision-making method of the Hopping Pattern includes one of the following: base station unidirectional autonomous decision-making; Alternatively, the UE may provide predictive information to the base station to assist the base station in making decisions.

5. The method according to claim 1, characterized in that, The Hopping Pattern supports both fast frequency hopping mode and slow frequency hopping mode; In fast frequency hopping mode, the UE transmits at different time-frequency resource locations in each service cycle; In slow frequency hopping mode, the UE maintains the same resource location in multiple consecutive service cycles, and the hopping interval is configured by the duration parameter in t_hop[i]. The switching interval is dynamically configured by the base station based on the results of frequency selective fading monitoring or prediction.

6. The method according to claim 1, characterized in that, When the correct reception time of the Hopping Pattern is missed in a service cycle due to HARQ or ARQ retransmission, the UE checks whether a transition needs to be performed in the service cycle before the arrival of each service cycle i, and performs subsequent transmission based on the most recently successfully received Hopping Pattern until a new Hopping Pattern is received.

7. A semi-dynamic scheduling system based on time-varying frequency hopping, comprising a base station side and a UE side, characterized in that: The base station side includes: The AI / ML prediction unit is used to perform CSI prediction, beam prediction, and QoS prediction. The grouping decision unit is used to perform multicast grouping for UEs that need to update their Hopping Pattern based on the prediction results; Hopping Pattern Generation Unit, used to generate Hopping Patterns containing quadruples {t_hop, f_hop, s_hop, m_hop}; The SDS MAC CE encapsulation unit is used to encapsulate the Hopping Pattern into an SDS-specific SDS MAC CE and transmit it via a dedicated MAC PDU; The UE side includes: The Hopping Pattern receiving and parsing unit is used to receive and parse the Hopping Pattern in the SDS MAC CE. The jump execution unit is used to perform multi-dimensional resource jumps according to the four-tuple parameters starting from a specified business cycle i. The business cycle monitoring unit is used to check whether a transition operation needs to be performed before each business cycle i arrives.

8. The system according to claim 7, characterized in that, The SDS MAC CE uses an independent dedicated MAC PDU for transmission and is not multiplexed with DL-SCH data. The dedicated MAC PDU can carry multiple DL SPS or UL CG Hopping Pattern information. The subheader of the SDS MAC CE includes an R field, an F field, an LCID field, and an eLCID field. The eLCID field distinguishes the value range of DL SDS MAC CE from the value range of UL SDS MAC CE. When DL and UL both require semi-dynamic scheduling and share the same subheader, the LCID value is located in the common interval. The L field in the subheader is divided into the DL part and the UL part, which respectively represent their length information.

9. The system according to claim 8, characterized in that, The SDS MAC CE payload includes the following fields: Hopping dimensions bitmap field, used to indicate the dimension information carried by the current MAC CE; Time domain Hopping duration field; PRB offset field; PRB resource size field; Antenna port bitmap field; TCI state index field; And the MCS field; The Hopping dimensions bitmap uses a 4-bit plus 2-bit reserved format, and dimensions that do not jump do not carry corresponding information; when the Bits for CW1 field is 0, it implicitly indicates that only CW0 is used in this business data; when the T field is 00, it indicates that the TCI state index is not carried; when CW0 and CW1 use different MCS, the M field is set to 1, and when they use the same MCS, the M field is set to 0.

10. The system according to claim 7, characterized in that, The base station configures a dedicated SDSLCID for the SDS MAC CE. The SDS LCID is distinct from the ordinary data LCID and is used for high-priority transmission. The SDS LCID supports dynamic allocation, which is determined by the base station based on the LCID and priority mapping information pre-configured to the UE and then notified to the UE. The dedicated MAC PDU is scheduled using the same DCI format as the ordinary MAC PDU, and transmission reliability is guaranteed through CCE high aggregation level to reduce the probability of false detection and missed detection by UE.

Images