A lightweight receiving demodulation method and system based on tree-shaped sub-channel decomposition

Abstract

The present application relates to the technical field of wireless communication and optical communication digital baseband demodulation, and discloses a light-weight receiving demodulation method and system based on tree-shaped sub-channel decomposition, which discards the discrete redundant architecture of traditional DDC digital down-conversion and FFT frequency domain decomposition, adopts multi-stage tree-shaped sub-band decomposition to perform stream fine division on a received time domain signal, each stage of decomposition is executed recursively through digital frequency mixing rotation and half-band filter extraction, and continuous sub-band time sequence data stream is output. The present application integrates channel decomposition and digital down-conversion functions, does not need whole data buffering and large-scale FFT operation, has non-uniform sparse spectrum fine decomposition capability, can activate effective channel branches as needed, retains complete time-frequency holographic information, supports multi-dimensional stereoscopic joint decision, and reduces chip area and power consumption by about 50% and pure signal processing delay by about 56% compared with traditional schemes. The present application can be widely applied to 5G / 6G / 7G base stations, terminals and high-speed optical communication receiving equipment.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication, optical communication digital baseband demodulation, subband signal detection, and digital downconversion technology. Specifically, it relates to a lightweight receiver demodulation technology based on tree-structured subchannel decomposition, and in particular, a lightweight receiver demodulation method and system based on tree-structured subchannel decomposition. This method can completely replace the traditional receiver FFT (Fast Fourier Transform) and DDC (Digital Down Converter) combined architecture, and is widely applicable to mobile terminals, 5G commercial base stations, 6G next-generation communication, future 7G communication systems, and high-speed optical communication systems. Background Technology

[0002] In contemporary wireless and optical communication systems, receiver-side digital baseband signal processing is one of the most crucial and critical components of the entire physical layer link. Taking 5G NR (New Radio) systems as an example, the base station-side receiving channel needs to simultaneously process wide-bandwidth signals from multiple frequency bands and carriers. In typical scenarios, the ADC (Analog-to-Digital Converter) sampling rate can reach as high as 983.04 Msps or even higher. The receiving baseband needs to complete a series of complex operations, such as wideband signal acquisition, down-conversion, channel decomposition, demodulation, and decoding, within an extremely short time. On the smartphone terminal side, limited by battery capacity and heat dissipation conditions, the power consumption and area of ​​the receiving baseband chip directly determine the battery life and market competitiveness of the terminal product. In the field of high-speed optical communication, as single-wavelength rates evolve from 400G to 800G and 1.6T, receiver-side DSP (Digital Signal Processor) chips face unprecedented power consumption and area pressures, and traditional receiving architectures have become a bottleneck restricting further improvements in system performance. In scenarios with low signal-to-noise ratio and strong interference, such as satellite communication and deep space communication, the receiver's sensitivity and anti-interference capability directly determine the reliability and survivability of the communication link.

[0003] In all the above application scenarios, the receiver's digital baseband processing generally adopts a discrete architecture of "DDC digital downconversion + FFT frequency domain decomposition". Since the large-scale commercial application of OFDM (Orthogonal Frequency Division Multiplexing) technology in the 1990s, this architecture has evolved through three generations of communication systems: 3G, 4G, and 5G, and has formed a highly solidified industry standard solution: the receiver first completes carrier frequency offset correction and sampling rate conversion through a DDC multi-stage filtering link, downconverting the broadband intermediate frequency signal to baseband, and then converts the time-domain baseband signal to the frequency domain through large-scale point-based FFT operations, realizing subcarrier-level channel decomposition and equalization demodulation. This architecture is widely used in 3GPP (3rd Generation Partnership Project) standards, IEEE 802.11 wireless LAN standards, DVB (Digital Video Broadcasting) digital television standards, and OIF (Optical Internetworking Forum) high-speed optical communication standards, and is the de facto standard architecture for receiver baseband chips in the global communications industry.

[0004] The traditional DDC+FFT discrete receiver architecture has revealed multiple inherent technical shortcomings in long-term industrial practice. Moreover, in next-generation mainstream communication scenarios such as sparse spectrum and non-uniform heterogeneous spectrum, these defects have already affected the technological progress of the entire industry, as detailed below:

[0005] First, high hardware redundancy, large chip area, and high power consumption constitute the core bottleneck for the lightweight development of communication baseband chips. To match the transmitter's IFFT (Inverse Fast Fourier Transform) transmission spectrum design, which nearly fills the digital frequency circle and has a steep transition band, the receiver must be equipped with a large-scale DDC multi-stage high-order steep filtering link. This typically includes complex circuits such as NCO (Numerically Controlled Oscillator) carrier frequency offset correction, 7-tap / 15-tap / 31-tap multi-stage half-band decimation filters, and 47-tap programmable shaping FIR (Finite Impulse Response) filters. These DDC filtering links occupy a large area in the chip and all operate at the highest sampling rate, resulting in extremely high dynamic power consumption. After this, a separate large-scale FFT operation unit is also required, such as a 4096-point pipelined FFT, which contains a large number of complex multipliers, butterfly operation units, and block cache RAM (Random Access Memory). The DDC and FFT modules are designed, powered, and occupy chip area independently, and there are many repetitive filtering and frequency conversion functions between them, resulting in serious architectural redundancy. Taking a 4096-point FFT receiving channel under 7nm process as an example, the traditional DDC+FFT discrete architecture has an equivalent total number of gates of about 600,000, a chip area of ​​about 0.22 square millimeters, and a dynamic power consumption of about 53mW at 400MHz. With the continuous increase of 5G millimeter wave and 6G terahertz communication bandwidth, the hardware overhead almost doubles every time the number of FFT points increases, which makes the chip area and power consumption problems continue to worsen, which has seriously restricted the iterative development of communication baseband chips towards lower cost and lower power consumption.

[0006] Secondly, there is a serious waste of systemic spectrum resources. The entire industry is forced to reserve approximately 20% of blank subcarriers as spectrum guard bands. Due to the steepness of the transition band of non-ideal high-order filters in traditional transceiver architectures and the influence of spectrum leakage, the IFFT output signal at the transmitter has a non-negligible out-of-band leakage energy at the edge of the digital frequency circle. To suppress interference between adjacent channels, the system must reserve a large number of blank subcarriers as spectrum guard intervals. In a typical OFDM system, approximately 20% of the subcarriers do not carry any effective information and exist only as spectrum "isolation bands." This means that in actual commercial systems, only about 80MHz of every 100MHz of licensed spectrum is used for effective data transmission, while the remaining 20MHz is systematically wasted. Against the backdrop of increasingly scarce spectrum resources and record-high spectrum auction prices, the economic losses caused by this systematic waste are enormous. More seriously, once these 20% of blank subcarriers are put into effective data transmission, because brick-wall filters are not feasible in engineering, aliasing at both ends of the frequency band will directly pollute adjacent channels, causing the orthogonality between subcarriers after FFT demodulation at the receiver to be destroyed, and the system performance to deteriorate sharply.

[0007] Third, the global block integral transform of FFT completely loses temporal holographic information, has a single decision dimension, and has inherent upper limits on anti-interference capability and receiver sensitivity. The essence of FFT is to perform global block integration on the time-domain signal within a time window and output the average frequency domain energy result within that window. The inherent characteristics of this mathematical transformation determine that it completely erases all information about the changes in the signal in the time dimension—including the arrival time, duration, power fluctuation mode, instantaneous frequency stability, and other rich features. Traditional receivers rely solely on the single-dimensional frequency domain energy information output by FFT to determine the presence of the signal and identify the modulation type. This "single-dimensional energy decision" mechanism is extremely susceptible to random noise fluctuations in low signal-to-noise ratio environments, resulting in false alarms or missed detections. In complex electromagnetic environments, when interference signals and useful signals overlap in the frequency domain, the single-dimensional energy decision cannot distinguish them at all, leading to a significant decrease in the receiver's anti-interference performance. This limitation is particularly fatal in adversarial scenarios such as satellite communication and emergency communication, directly determining the survivability of communication links in harsh environments.

[0008] Fourth, the forced uniform grid decomposition of FFT cannot natively adapt to the heterogeneous channel characteristics of non-uniform, sparse, and asymmetric channels, resulting in a large amount of invalid computational overhead in dynamic spectrum scenarios. The mathematical nature of FFT dictates that its decomposition result is a uniformly spaced frequency grid. Regardless of the actual signal distribution, the system must perform a complete decomposition operation on all grids. In 5G / 6G / 7G dynamic sparse spectrum scenarios, many frequency bands are idle or allocated to other services, but the FFT receiver still has to perform complete complex multiplication and accumulation operations on these invalid empty frequency bands. This "forced global uniform grid" approach... The "FFT" mechanism leads to a huge waste of computing power in scenarios with fragmented frequency domain resources. For example, in a typical IoT sparse spectrum access scenario, the effective signal accounts for less than 10% of the entire analysis bandwidth, but the traditional FFT receiver still needs to perform full-load decomposition calculations on 100% of the frequency band, resulting in extremely low hardware resource utilization. In the optical communication flexible grid scenario, the spectrum division itself is non-uniform and asymmetric, and the FFT uniform grid is completely incompatible with it. It must be adapted through additional frequency domain interpolation or resampling methods, which further increases the system complexity and hardware overhead.

[0009] In summary, the traditional DDC+FFT discrete receiver architecture suffers from serious deficiencies in four dimensions: hardware efficiency, spectral efficiency, information utilization efficiency, and heterogeneous adaptability. These deficiencies are rooted in the mathematical nature and engineering implementation of the architecture and cannot be fundamentally improved through local optimization or process upgrades. Therefore, those skilled in the art urgently need a lightweight receiver demodulation technology solution based on tree-shaped N-molecule channel decomposition that does not rely on FFT, fundamentally breaking the triangular constraint of "area-power consumption-performance". Summary of the Invention

[0010] The core technical problem addressed by this invention is: how to construct a novel receiver demodulation architecture that can simultaneously eliminate hardware redundancy, improve spectrum utilization, retain time-frequency holographic information, and natively adapt to sparse and non-uniform spectrum from the underlying architecture, so as to fundamentally replace the traditional DDC+FFT discrete combination scheme.

[0011] For those skilled in the art, solving the aforementioned core technical problems has significant technical implications and industrial value:

[0012] From a technological evolution perspective, the communications industry is currently at a critical juncture, transitioning from 5G to 6G and even 7G. Typical characteristics of 6G communication include integrated air-space-ground-sea coverage, convergence of sensing, computing, and intelligence, extreme connection density, and ultra-low latency and high reliability. This places unprecedented demands on the lightweight, high-sensitivity, high-spectral-efficiency, and high-flexibility of the receiving baseband. The traditional FFT+DDC architecture has reached its physical limits—the computational load of FFT increases exponentially with the number of points. As bandwidth increases, the overhead of DDC filtering increases linearly, and the combination of the two causes the power consumption benefits brought by advanced processes below 7nm to be quickly swallowed up by the inefficiency of the architecture itself.

[0013] From the perspective of industry competition, major global communication chip manufacturers are actively exploring alternative solutions for receiver architectures in the post-FFT era. Whoever can break through this core technology bottleneck first will dominate the next-generation communication baseband chip market.

[0014] For those skilled in the art, solving the aforementioned core technical problems is essential, because continuing to follow the traditional FFT+DDC approach can no longer meet the demands of 6G / 7G for extreme energy efficiency and spectrum efficiency. Solving these core technical problems will enable breakthroughs in the following technical bottlenecks:

[0015] First, it completely breaks the bottleneck of linear growth in chip area and power consumption with bandwidth, enabling an order-of-magnitude improvement in the energy efficiency ratio of communication baseband chips;

[0016] Secondly, it breaks through the 20% systematic loss limit of spectrum efficiency, achieving 100% effective utilization of the entire frequency band;

[0017] Third, it breaks through the sensitivity limit of receiver single-dimensional energy decision-making, realizing three-dimensional intelligent perception and decision-making based on time-frequency holographic features.

[0018] To address the aforementioned core technical challenges, this invention presents a lightweight receiving and demodulation method and system based on tree-structured subchannel decomposition. The aim is to abandon the traditional two-level discrete architecture of "DDC down-conversion → FFT frequency domain decomposition," integrating channel decomposition and digital down-conversion functions into a single process. A multi-level tree-structured subband decomposition architecture is employed to progressively refine the frequency domain division of streaming time-domain signals. This technical solution should be applied to all scenarios requiring baseband receiving processing, such as wireless communication base stations, mobile terminals, and optical communication transceivers, to overcome industry-wide bottlenecks such as hardware redundancy, spectrum waste, information loss, and grid rigidity in traditional architectures. The core idea of ​​this solution is to recursively decompose wideband signals into multiple narrowband subchannels using multi-level digital mixing rotation combined with half-band filtering decimation. Each decomposition level outputs a complete and continuous time-series data stream, and the decomposition structure at each level can be adaptively configured according to the actual channel distribution, fundamentally achieving architecture fusion, holographic preservation, and on-demand computation.

[0019] To achieve the above objectives, the specific technical solution of the present invention is a lightweight receiving and demodulation method based on tree-structured subchannel decomposition, the method comprising the following steps:

[0020] Step A: Using the preset sampling rate The received broadband analog signal is converted from analog to digital to obtain a digital time-domain signal sequence. ;

[0021] Step B: Process the digital time-domain signal sequence Perform multi-level tree-structured subband decomposition;

[0022] Step C: In the decomposition process of the multi-level tree sub-bands, the branch sub-band signals output by each level of the decomposition process are continuous time-series data streams;

[0023] Step D: Extract the multidimensional feature parameters of each branch subband signal based on the continuous time-series data stream output from each decomposition process.

[0024] Step E: Perform signal existence determination and demodulation processing on each branch sub-band channel based on the multi-dimensional feature parameters.

[0025] Preferably, the number of levels in the multi-level tree subband decomposition Based on the total number of target sub-channels Confirmed, satisfied ,and It is a positive integer greater than or equal to 1.

[0026] Preferably, the rotational angular frequency parameter In the The value in the level decomposition is The rotation factor sequence is This is used to shift the two right quadrants of the current stage input signal spectrum to near the zero frequency.

[0027] Preferably, the half-band filtering process employs a half-band decimation filter with an equiripple design, and the number of taps in the filter is selected from any one of 7 taps, 15 taps, or 31 taps.

[0028] Preferably, each level of the multi-level tree subband decomposition includes three processing operations: digital mixing and rotation, half-band filtering, and 2x decimation, and each level of processing operation is recursively executed according to the same structure;

[0029] Each level of tree-like subband decomposition specifically includes the following sub-steps:

[0030] Sub-step B1: Perform digital mixing and rotation processing on the digital time-domain signal sequence input to the current stage, using the rotation factor sequence. The target frequency band was moved to the vicinity of the baseband, among which This refers to the rotational angular frequency parameter of the current stage;

[0031] Sub-step B2: Perform half-band filtering on the digital time-domain signal sequence after frequency mixing and rotation to filter out out-of-band image components;

[0032] Sub-step B3: Perform a 2x decimation process on the filtered digital time-domain signal sequence to halve the sample rate.

[0033] Preferably, in step C, the first... The number of branches output by the level decomposition is The sample rate of each branch is ,in For level index, , This represents the total decomposition series.

[0034] Preferably, the multidimensional feature parameters extracted in step D include at least one of the following:

[0035] D1, the average energy value of the subband output at each decomposition stage;

[0036] D2, statistical variance of power fluctuation of sub-band signal within a preset time window;

[0037] D3, Cordic phase detectors obtain sub-band instantaneous frequency values ​​and frequency jitter statistics;

[0038] D4. The duration characteristic of the subband signal in terms of timing, that is, the continuous length of time during which the signal energy exceeds the preset threshold.

[0039] Preferably, the signal existence determination in step E includes:

[0040] E1. Calculate the first decision metric based on the average energy value of the sub-band and the statistical variance of power fluctuation;

[0041] E2. Calculate the second decision metric based on the instantaneous frequency jitter statistical characteristics;

[0042] E3. Calculate the third decision metric based on the aforementioned temporal duration characteristics;

[0043] E4. The first, second and third decision metrics are weighted and fused. When the fusion result exceeds the preset decision threshold, it is determined that there is a valid signal in the current sub-band channel.

[0044] Preferably, in the multi-level tree subband decomposition process of step B, a process limiting operation is embedded before or after each level of half-band filtering. The process limiting operation is used to limit the amplitude of the input signal within a preset amplitude threshold range to suppress short-time pulse spike interference.

[0045] Preferably, the process limiting operation is performed according to the following formula:

[0046] ,when hour;

[0047] ,when ;

[0048] in, These are the sample values ​​of the input signal before the amplitude limiting operation. These are the sample values ​​of the output signal after the amplitude limiting operation. To preset the amplitude threshold, It is a symbolic function.

[0049] Preferably, the configuration parameters for each level of the tree-like subband decomposition in step B are adaptively adjusted based on the spectral distribution characteristics of the currently received signal, channel state information, or interference distribution information. These configuration parameters include the number of decomposition levels. Tap coefficients and twitch factor sequence parameters for each stage of the filter.

[0050] Preferably, the adaptive adjustment includes: performing deep multi-level decomposition in signal-dense frequency bands to obtain fine-grained subband division, and stopping further decomposition or performing shallow coarse-grained decomposition in signal-sparse or idle frequency bands, so as to achieve custom subband channel division with non-uniformity, asymmetry, unequal intervals, and unequal bandwidth.

[0051] Preferably, in step B, the tree decomposition branch corresponding to the invalid frequency band is clock-gated or power-gated to prevent the invalid frequency band from participating in the operation and consuming dynamic power.

[0052] Preferably, the present invention also discloses a lightweight receiver demodulation system based on tree-structured subchannel decomposition, the system being used to implement the method, comprising:

[0053] The analog-to-digital conversion unit operates at a preset sampling rate. The received broadband analog signal is converted from analog to digital, and a digital time-domain signal sequence is output.

[0054] A tree-shaped multi-level decomposition network, connected to the analog-to-digital conversion unit, is used to perform multi-level tree-shaped sub-band decomposition on the digital time-domain signal sequence;

[0055] The tree-structured multi-level decomposition network includes: cascaded multi-level decomposition nodes, each level of decomposition node including:

[0056] A digital mixing rotation module is used to perform spectral shifting on the digital time-domain signal sequence input to the analog-to-digital conversion unit using a rotation factor sequence;

[0057] A half-band filtering module, connected to the digital mixing and rotating module, is used to filter out the out-of-band mirror component after the digital time-domain signal sequence has been mixed.

[0058] A 2x decimation module, connected to the half-band filtering module, is used to halve the sample rate of the filtered digital time-domain signal sequence.

[0059] Each decomposition node outputs a continuous time-series data stream of multiple branches, and the decomposition nodes at each level are recursively connected according to the same structure.

[0060] Preferably, each decomposition node of the tree-shaped multi-level decomposition network further includes a process limiting module, which is connected between the half-band filter module and the 2x decimation module, or between the digital mixing and rotating module and the half-band filter module, for limiting the signal amplitude within a preset threshold range.

[0061] Preferably, the system further includes:

[0062] The multidimensional feature extraction module is connected to the output of each level of the tree-like multi-level decomposition network and is used to extract at least one of the following from the continuous time-series data stream output at each level: subband average energy value, power fluctuation statistical variance, instantaneous frequency jitter feature, and time-series duration feature;

[0063] The joint decision module, connected to the multi-dimensional feature extraction module, is used to perform weighted fusion decision based on the extracted multi-dimensional features to determine the signal existence of each sub-band channel.

[0064] Preferably, the multidimensional feature extraction module includes a Cordic phase detection submodule, which is used to extract instantaneous phase information from the continuous time-series data stream of each sub-band, and calculate the instantaneous frequency value and its jitter statistical characteristics based on the instantaneous phase information.

[0065] Preferably, the system further includes:

[0066] The recursive statistics maintenance module is connected to the output of each level of the tree-shaped multi-level decomposition network. It is used to maintain multiple recursive statistical scalar variables in real time during the decomposition process. The recursive statistical scalar variables include the first moment estimate and the second moment estimate of each sub-band signal, which are used to characterize the energy mean and power fluctuation of the sub-band signal.

[0067] The recursive statistics maintenance module does not contain a large-capacity RAM cache, but only maintains a limited number of scalar registers.

[0068] Preferably, the system further includes:

[0069] An adaptive configuration controller, connected to the tree-like multi-level decomposition network, is used to dynamically adjust the configuration parameters of the tree-like multi-level decomposition network based on spectrum sensing results, channel quality indications, or interference detection results. The configuration parameters include the currently active decomposition level, the tap coefficients of each half-band filtering module, and the rotation factor parameters of each digital mixing rotation module.

[0070] Preferably, the adaptive configuration controller is further configured to: generate a branch activation control signal based on the spectrum distribution information, wherein the branch activation control signal is used to enable or disable the clock signal or power supply of a specific branch in the tree-like multi-level decomposition network, so that the branch corresponding to the invalid frequency band does not participate in the operation and does not consume dynamic power consumption.

[0071] Preferably, the half-band filtering module adopts a half-band decimation filter with equal ripple design. The number of taps of the filter is selected from any one of 7 taps, 15 taps or 31 taps according to the signal-to-noise ratio or bandwidth requirements of the current stage input signal, and different stages of the half-band filtering module can be configured with different numbers of taps.

[0072] Preferably, in each decomposition node of the tree-like multi-level decomposition network, the rotation angular frequency parameter of the digital mixing rotation module is... Values The rotation factor sequence is These are used to shift the right semicircle of the current stage input signal spectrum, specifically the first and fourth quadrants, to near zero frequency.

[0073] Preferably, in the system, when the tree-structured multi-level decomposition network has 12 levels, the 12th level decomposition outputs 4096 sub-band channels, and the sample rate of each sub-band channel is... .

[0074] Preferably, the present invention also discloses a wireless communication receiving device, the device comprising the system described above.

[0075] Preferably, the wireless communication receiving device is any one of a 5G base station, a 6G base station, a 7G base station, a mobile terminal, an Internet of Things terminal, a satellite communication terminal, or a high-speed optical communication receiver.

[0076] Preferably, the present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is capable of implementing the steps of the method.

[0077] Compared with the prior art, the technical solution disclosed in this invention has the following non-obvious technical features:

[0078] The first technical feature is the deep integration of the two major functional modules of DDC (Digital Down-Conversion) and FFT (Frequency Filtration) in the traditional discrete architecture, constructing a unified decomposition framework of "recursive digital mixing rotation + half-band filtering and decimation". In traditional solutions, DDC and FFT are two completely independent chip modules implemented by different design teams. They are connected through an on-chip bus or FIFO (First In First Out) buffer, and each has its own independent control logic, clock domain, and power management strategy. When faced with the overhead problems of DDC and FFT, those skilled in the art usually optimize them separately—designing a lower-power DDC filtering structure or a more efficient FFT butterfly unit. No one has ever proposed an architecture that integrates the two into one. This invention breaks this mindset. Through a tree-like multi-level decomposition structure, each level of decomposition simultaneously undertakes the down-conversion, filtering, and decimation functions in the traditional DDC, as well as the frequency domain decomposition function in the traditional FFT, completely eliminating redundancy between the two modules.

[0079] Technical Feature Two: Breaking through the mathematical constraints of the FFT's forced uniform grid, this invention natively achieves custom subband partitioning that is non-uniform, sparse, unequally spaced, and unequal in bandwidth based on a tree-based decomposition architecture. The uniform grid of the FFT is determined by the mathematical properties of the Fourier transform. Those skilled in the art have long believed that the concept of frequency domain decomposition inevitably leading to uniform frequency resolution is the foundation of Fourier analysis and cannot be changed. However, this invention achieves completely non-uniform frequency domain partitioning through the configurability of the tree-based decomposition architecture—each level of decomposition branch can choose whether to continue decomposing downwards and what decomposition ratio to use based on the actual channel distribution. For example, it can perform deep and fine decomposition in densely populated frequency bands and stop decomposition or only perform coarse-grained partitioning in idle frequency bands. This on-demand decomposition capability is completely impossible to achieve under the FFT framework. This technical feature essentially changes the industry axiom that "frequency domain decomposition = uniform grid," exhibiting significant unconventionality.

[0080] Technical Feature 3: By utilizing a tree-like multi-level decomposition structure to retain and extract complete continuous temporal information, a "multi-level time-frequency holographic" decision space is constructed, distinct from the single-dimensional energy decision of FFT. The loss of temporal information during FFT block integration is an inherent property of Fourier transform. When using FFT for signal detection, the generally accepted optimization direction is to improve frequency domain resolution, improve the energy detection threshold algorithm, or adopt second-order statistical methods such as cyclic stationary detection. No one has proposed "bypassing" the FFT information loss problem by changing the decomposition architecture itself. This invention, through the essential characteristic that each level of tree decomposition outputs a continuous sub-band temporal data stream, allows the extraction of multi-dimensional information such as hierarchical energy convergence trajectory, power fluctuation variance, instantaneous frequency jitter, and temporal duration during the decomposition process, forming a novel "three-dimensional evidence chain" decision mechanism that differs from existing technologies.

[0081] Technical Feature 4: Embedding recursive statistics and process limiting mechanisms during tree decomposition enables robust detection under low signal-to-noise ratio and source suppression of impulse interference. Existing anti-interference measures typically employ post-processing methods such as frequency domain zeroing or time domain threshold detection after FFT demodulation, which are "post-hoc remedial" strategies. Interference energy has already participated in the FFT operation and may cause spectral spread effects. This invention maintains recursive statistical variables in real time during the decomposition process and can embed limiting processing at each filtering node, so that impulse interference is suppressed before entering subsequent decomposition branches, preventing the propagation and amplification of interference pollution from the source. This "process-as-process" architectural feature is completely absent in existing FFT+DDC schemes.

[0082] Technical Feature 5: End-to-end streaming online computation, with no need for large-block data caching, maintaining only a small number of recursive statistical scalar variables, eliminating large-capacity RAM overhead from the root of the architecture; Traditional FFT processing must cache an entire FFT window (such as 4096 consecutive samples) before starting the computation, which means that the chip must be equipped with a RAM array matching the size of the window, and the area overhead of this RAM is extremely considerable at high sampling rates; This invention adopts a sample-by-sample streaming processing architecture, where samples are processed as soon as they enter and output as soon as they are processed, without relying on any large-block data caching, fundamentally eliminating the RAM area overhead and waiting latency caused by FFT block processing.

[0083] Compared with existing technologies (traditional FFT+DDC architecture), the present invention has the following advantages:

[0084] (1) This invention can achieve a significant reduction of approximately 50% in hardware power consumption and chip area, further expanding its advantages in sparse and non-uniform spectrum scenarios. This invention innovatively combines the two high-overhead modules of DDC digital down-conversion and FFT frequency domain decomposition into one, completely eliminating the redundant overhead of the multi-level DDC filtering link and the large-scale complex number operation unit and block cache RAM in the traditional architecture. Specifically, this invention adopts a tree-shaped multi-level recursive decomposition structure of "digital mixing rotation + half-band filtering extraction," where each level of decomposition simultaneously performs the dual functions of down-conversion and frequency domain partitioning, avoiding the two-level serial processing mode of first down-converting the signal to baseband and then independently performing FFT decomposition in the traditional architecture. Therefore, this invention can reduce the equivalent total gate count from [previous value] in 7nm process technology. The traditional solution reduces the number of gates from approximately 600,000 to approximately 305,000, the chip area from approximately 0.22 square millimeters to approximately 0.112 square millimeters, and the dynamic power consumption at 400MHz from approximately 53mW to approximately 27mW, achieving an overall hardware resource saving of approximately 49%. More importantly, in sparse dynamic spectrum scenarios of 5G / 6G / 7G, the tree-like architecture of this invention, which performs channel decomposition only for valid signal frequency bands, does not participate in the calculation of invalid frequency bands and does not occupy hardware resources. Its power consumption and area advantages will be orders of magnitude superior to the FFT solution that forces global uniform calculation. This technical effect is the result of the combined effect of architecture integration and on-demand calculation mechanisms, directly solving the core bottleneck of lightweight development of communication baseband chips across the industry.

[0085] (2) This invention constructs a multi-dimensional holographic joint decision space, which significantly improves receiver sensitivity and anti-interference capability; this invention breaks through the limitation of the traditional FFT, which can only output the single decision dimension of the average frequency domain energy of a single window, and utilizes the continuous sub-band time-series data stream output by each level in the tree-like multi-level decomposition process to simultaneously extract hierarchical energy convergence features, sub-band power fluctuation statistical features, and Cordic (Coordinate Rotation Digital Computer). DigitalComputer (DC) uses multi-dimensional information such as instantaneous frequency jitter characteristics and temporal duration characteristics to form a three-dimensional, multi-dimensional joint decision evidence chain. Specifically, real communication signals exhibit a top-down, orderly energy convergence pattern and stable instantaneous frequency characteristics during the tree-like multi-level decomposition process, while environmental noise exhibits random energy distribution, violent and irregular instantaneous frequency jitter, and unstable temporal duration at each level of decomposition. This invention utilizes these essential differences to construct a joint decision space, which can effectively distinguish between random noise fluctuations, short-term burst interference, and real signals, significantly reducing the probability of false alarms and missed detections in low signal-to-noise ratio environments. At the same time, the process limiting mechanism embedded in the streaming decomposition link can suppress short-term pulse spike interference from the source in advance, avoiding the contamination and diffusion of interference energy to subsequent branches. The synergistic effect of multi-dimensional holographic feature extraction and process limiting makes the anti-interference performance of this invention achieve a qualitative leap compared with the traditional FFT single-dimensional energy decision scheme.

[0086] (3) This invention can eliminate systemic spectrum redundancy across the entire industry, achieving 100% effective utilization of the entire frequency band, and overwhelmingly improving spectrum efficiency in sparse / non-uniform scenarios; the traditional engineering scheme of full-frequency circular transform of FFT / IFFT combined with non-ideal high-order filters inevitably produces spectrum leakage due to insufficient transition band steepness, forcing the system to reserve about 20% of blank subcarriers as guard bands, forming a systemic spectrum waste common across the entire industry; this invention relies on a tree-shaped multi-level narrowband phase rotation synthesis and decomposition architecture, each level of subband division has sufficient transition band margin, and out-of-band leakage is strictly constrained within the boundaries of each subband. After multi-level cascading, the overall out-of-band leakage is extremely low, and no blank guard spectrum needs to be set; this technical effect comes from the essential feature of "recursive decomposition" rather than "global transformation" of this invention - each The first stage only processes the half-band division under the current sample rate. The filter design only needs to meet the local transition band requirements, instead of the traditional scheme where the first-stage filter needs to complete the steep filtering task from high sampling rate to baseband. The engineering feasibility is greatly improved, and the filter order and design difficulty are significantly reduced. Therefore, this invention does not need to sacrifice any effective spectrum resources to absorb aliasing leakage. Under the same licensed spectrum bandwidth, it can achieve 100% effective data transmission across the entire frequency band. Compared with the 80% effective utilization rate of the traditional scheme, the spectrum efficiency is improved by 25%. In mainstream sparse and non-uniform scenarios such as 5G / 6G dynamic spectrum access and flexible grid for optical communication, this invention can also finely divide the effective channel and merge idle frequency bands as needed, completely getting rid of the rigid constraints of the traditional FFT uniform grid. The spectrum utilization rate and resource utilization efficiency form an overwhelming advantage.

[0087] (4) This invention utilizes a streaming pipeline architecture to achieve ultra-low processing latency, perfectly adapting to time-varying channels and low-latency services. This invention performs real-time streaming decomposition, real-time statistics, and real-time decision-making on a sample-by-sample basis throughout the entire process. Once a sample enters the decomposition network, pipeline processing begins immediately. There is only filter group delay (typically about 4.8 μs) between each level, eliminating the need to wait for the entire block of data to be filled before starting the operation. Compared to the traditional FFT scheme, which requires caching 4096 samples (about 10.9 μs delay) before performing block transformation, this invention reduces the pipeline latency of pure signal processing by about 56%. In scenarios that are extremely sensitive to latency, such as high-speed mobile communication, low-orbit satellite communication, and industrial IoT, this latency reduction directly translates into faster link establishment time, more accurate channel tracking capability, and better service experience quality. Attached Figure Description

[0088] Figure 1 This is a flowchart of the method described in this invention;

[0089] Figure 2 This is a block diagram of the pipeline architecture for synthesizing 4096 sub-channels as described in this invention;

[0090] Figure 3This is a block diagram of the preprocessing architecture before tree decomposition as described in this invention;

[0091] Figure 4 This is a block diagram of the architecture of the 2nd to 12th levels of the tree-shaped binary channel decomposition described in this invention;

[0092] Figure 5 This is a replacement diagram of the tree-shaped binary channel decomposition and DDC+FFT described in this invention;

[0093] Figure 6 This is a block diagram of the traditional 5G mobile phone DDC+FFT architecture described in this invention;

[0094] Figure 7 This is a schematic diagram of the system described in this invention;

[0095] Figure 8 This is a table comparing the number of branches and the sample rate output by each level of decomposition nodes as described in this invention. Detailed Implementation

[0096] The following is in conjunction with the appendix Figures 1 to 8 The main technical features of the present invention will be described in detail below;

[0097] In the technical solution of this invention, the tree-shaped N-band decomposition architecture is a method that decomposes a broadband digital signal step by step into its constituent parts through a multi-level recursive operation of "digital mixing and rotation → half-band filtering → decimation by 2". This invention presents a streaming processing architecture for narrowband sub-channel signals. Each decomposition stage divides the current frequency band into two equal sub-bands. The next stage continues the same binary division operation on each sub-band, forming a binary tree-shaped signal decomposition network. This architecture is the core foundation of this invention, determining the essential characteristics of the entire receiving and demodulation scheme. It replaces the traditional two-stage discrete architecture of "DDC down-conversion + FFT frequency domain decomposition," merging the functions that originally required two independent large modules into a single recursive structure, fundamentally solving the hardware redundancy problem. Simultaneously, the recursive nature of this architecture ensures that each stage outputs a continuous time-series data stream, providing a data foundation for subsequent holographic information extraction. This architecture is configurable and customizable; its topology directly affects the system's hardware overhead, processing latency, and adaptability.

[0098] In the technical solution of this invention, the recursive processing unit for step-by-step digital mixing rotation and half-band filtering decimation consists of three sequentially executed operations at each decomposition node:

[0099] (1) Digital mixing rotation: using rotation factor sequence Shift the target quadrant of the current stage input signal spectrum to near zero frequency;

[0100] (2) Half-band filtering: A half-band decimation filter (typically 7 / 15 / 31 taps) with equiripple design is used to filter out the out-of-band image components after shifting;

[0101] (3) 2x Decimation: Halve the sample rate of the filtered signal. These three operations are recursively applied to each level of decomposition in a fixed order;

[0102] The aforementioned unit is the core of the engineering implementation of this invention and the concrete execution carrier of the "architectural fusion" concept. Digital mixing rotation replaces the NCO mixing function in traditional DDC, half-band filtering replaces the multi-stage filtering links in DDC, and 2x decimation simultaneously accomplishes the dual tasks of sampling rate adaptation and frequency domain refinement. The regularity and repeatability of this three-pronged structure make the hardware design highly modular, facilitating pipeline implementation and power management. The number of filter taps and rotation parameters at each stage directly affect the frequency response characteristics and hardware overhead of that stage.

[0103] In the technical solution of this invention, the on-demand activated adaptive non-uniform decomposition capability allows the tree-based decomposition network to selectively perform deep decomposition only on the frequency bands containing valid signals, based on the spectral distribution characteristics of the currently received signal. It stops further decomposition of idle or invalid frequency bands or performs only shallow decomposition, thereby achieving custom subband partitioning that is non-uniform, asymmetric, unequal in spacing, and unequal in bandwidth. This capability is achieved through an adaptive configuration controller that dynamically manages the activation / disabling states of each branch level. This capability directly solves the problem of redundant computation in traditional FFT-forced uniform grids under sparse / non-uniform spectrum scenarios. It transforms the hardware advantage of this invention in mainstream scenarios such as 5G / 6G dynamic spectrum, fragmented spectrum in the Internet of Things, and flexible grids in optical communication from a "slight advantage" to a "crushing lead." This feature is the key differentiator of this invention compared to other filter bank-based frequency domain decomposition schemes and is also the core technical support for achieving extreme lightweight design.

[0104] In the technical solution of this invention, the holographic information output of the continuous sub-band time-series data stream differs from the traditional FFT, which only outputs the average frequency domain energy within a window. In this invention, each sub-band branch output by each level of the decomposition network is a complete and continuous time-domain data stream, containing all the signal's changes in the time dimension—including amplitude envelope trajectory, continuous phase changes, power fluctuation patterns, signal duration, etc. This holographic information output is the data foundation for achieving "multi-dimensional holographic joint decision-making," directly determining whether the receiving sensitivity and anti-interference capability of this invention can surpass the FFT scheme. The continuous time-series data stream allows subsequent processing modules to extract various statistical features (energy convergence trajectory, power fluctuation variance, instantaneous frequency jitter, duration, etc.) to construct a three-dimensional evidence chain. If the output of each level is not a continuous data stream but remains a blocky result, this invention will lose this core differentiating advantage.

[0105] In the technical solution of this invention, the recursive statistical scalar maintenance mechanism (RAM-free) eliminates the need for caching large amounts of historical sample data during the tree decomposition process. Instead, it maintains only a small number of recursive statistical scalar variables (such as the recursive update amounts of the first and second moment estimates), updating them in real-time on a sample-by-sample basis. The entire processing flow does not rely on any large-capacity RAM array; this mechanism is one of the technical guarantees for achieving approximately 50% reduction in chip area. In traditional FFT schemes, block cache RAM can account for 15%-25% of the total chip area; this invention completely eliminates this overhead through streaming processing and recursive statistics. Simultaneously, the recursive statistical mechanism allows statistical features to reflect the latest state changes of the signal in real time, which is of great significance for rapid tracking of time-varying channels.

[0106] In the technical solution of this invention, the process limiting mechanism embeds amplitude limiting processing units before and after the filtering nodes at each stage of the tree decomposition. These units monitor the amplitude of signal samples in real time, and when the amplitude exceeds a preset threshold, the samples are truncated to the threshold value. This suppresses short-time pulse spike interference in the early stages of signal propagation down the decomposition link. This mechanism is one of the key technologies that makes this invention superior to traditional solutions in terms of anti-interference capability. In traditional FFT solutions, after the interference signal enters the FFT processor, the energy of the pulse interference spreads to all output frequencies through butterfly operations, causing a "global interference pollution" effect. This invention, through process limiting, blocks interference propagation during the decomposition process, protecting the purity of subsequent decomposition results from the source.

[0107] In the technical solution of this invention, the oversampling rate maintenance in each level of decomposition is achieved by rationally designing the transition band and passband ripple of the half-band filter at each level of the multi-level tree decomposition. This ensures a sufficient oversampling rate relationship between the useful signal frequency components and the current level's sample rate, so that aliasing components only fall within a controllable protection margin range and do not affect the purity of the effective signal sub-bands. This oversampling rate maintenance is the technical basis for this invention's elimination of the need to reserve a 20% blank protection subcarrier. In traditional schemes, FFT requires strict signal band limiting and no aliasing; otherwise, subcarrier orthogonality is destroyed, thus requiring a large amount of protection band to be reserved. However, this invention maintains the oversampling rate relationship in each level of decomposition and adopts a controllable aliasing strategy, so that aliasing components are limited to the transition band of the sub-band boundary without sacrificing any effective spectrum resources. If this characteristic is not satisfied, aliasing will pollute the effective signals of adjacent sub-bands, and the full-band utilization advantage of this invention will be lost.

[0108] In the technical solution of this invention, the pipelined streaming processing architecture (block-free processing with no waiting delay) adopts a sample-by-sample streaming processing mode throughout the entire receiving and demodulation link. Samples are processed as soon as they enter and passed to the next level after processing. There is only filter group delay between nodes, and there is no need to wait for a preset number of samples to accumulate before starting the operation. This architecture is the technical guarantee for achieving ultra-low processing latency (approximately 4.8μs, a reduction of about 56% compared to traditional solutions) and is also a necessary condition for achieving the elimination of large-capacity RAM. Streaming processing makes this invention naturally adaptable to scenarios such as time-varying channel tracking and low-latency services in high-speed mobile communications, and is a key architectural attribute for meeting the ultra-low latency requirements of 5G / 6G.

[0109] The following is in conjunction with the appendix Figures 1 to 8 The embodiments of the present invention will be described in detail below;

[0110] refer to Figure 1 This document illustrates a flowchart of a lightweight receiver demodulation method based on tree-structured subchannel decomposition according to one or more embodiments of this application. The method is applied to the digital baseband processing chip of wireless communication receiving devices (including but not limited to 5G / 6G / 7G base stations, mobile terminals, and high-speed optical communication receivers) to replace the traditional discrete receiver architecture of DDC digital down-conversion and FFT frequency domain decomposition. The method includes the following steps:

[0111] Step A: Using the preset sampling rate The received broadband analog signal is converted from analog to digital to obtain a digital time-domain signal sequence. ;

[0112] Step B: Process the digital time-domain signal sequence Perform multi-level tree-structured subband decomposition;

[0113] Step C: In the decomposition process of the multi-level tree sub-bands, the branch sub-band signals output by each decomposition process are continuous time-series data streams;

[0114] Step D: Extract the multidimensional feature parameters of each branch subband signal based on the continuous time-series data stream output from each decomposition process;

[0115] Step E: Perform signal existence determination and demodulation processing on each branch sub-band channel based on the multi-dimensional feature parameters.

[0116] In one or more embodiments, step A utilizes an analog-to-digital converter (ADC) at a preset sampling rate. (For example, a typical value of 983.04 Msps) The received broadband intermediate frequency analog signal is sampled and quantized to output a digital time-domain signal sequence. .

[0117] In one or more embodiments, to avoid the significant power consumption and area constraints caused by the design of high-order analog filters, the ADC sampling rate is configured to be more than 8 times the signal bandwidth, so that the useful signal is distributed on the initial digital frequency circle. Within this interval, sufficient transition band margin is reserved for subsequent digital domain filtering processing. This digital time-domain signal sequence It is a complex I / Q data stream, containing in-phase and quadrature components.

[0118] In one or more embodiments, the tree-structured multilevel decomposition network in step B decomposes the digital time-domain signal sequence. Perform a step-by-step recursive decomposition. The number of levels N in the multi-level tree-shaped sub-band decomposition is based on the total number of target sub-channels. Confirmed, satisfied ,and It is a positive integer greater than or equal to 1. For example, when it needs to be decomposed into 4096 sub-channels, =12. Each level of tree-structured subband decomposition includes three processing operations: digital mixing and rotation, half-band filtering, and decimation by 2x. All processing operations at each level are executed recursively according to the same structure.

[0119] In one or more embodiments, each level of tree subband decomposition includes the following sub-steps:

[0120] Sub-step B1: Perform digital mixing and rotation processing on the digital time-domain signal sequence input to the current stage, using the rotation factor sequence. The target frequency band was moved to the vicinity of the baseband, among which This refers to the rotational angular frequency parameter of the current stage. In a preferred embodiment of the invention, the rotational angular frequency parameter... In the The value in the level decomposition is The rotation factor sequence is It is used to shift the two right quadrants (first quadrant and fourth quadrant) of the current stage input signal spectrum to near zero frequency.

[0121] Sub-step B2: Perform half-band filtering on the mixed-rotated digital time-domain signal sequence, using a half-band decimation filter (HBF) with equiripple design to filter out out-of-band image components. The number of taps in the filter is selected from 7, 15, or 31 taps based on the signal-to-noise ratio or bandwidth requirements of the current stage input signal, and different stages of the half-band filtering module can be configured with different numbers of taps. For example, a 7-tap filter can be used in the pre-stage to reduce delay, while a 31-tap filter can be used in the post-stage to obtain better stopband attenuation performance.

[0122] Sub-step B3: Perform a 2x decimation process on the filtered digital time-domain signal sequence to halve the sample rate.

[0123] In one or more embodiments, a digital signal preprocessing operation prior to tree decomposition is included before step B. For example... Figure 3 As shown, in the preprocessing stage, a 7-tap, equiripple half-band decimation filter is first used to reduce the sample rate by half. At this time, the useful signal is distributed in the digital frequency circle. The interval is then filtered using a 15-tap, equiripple-designed half-band decimation filter to halve the sample rate. At this time, the useful signal is distributed in the digital frequency circle. The interval completes signal conditioning before entering the tree-like multi-level decomposition network.

[0124] In one or more embodiments, the number of branches output by the k-th level decomposition in step C is The sample rate of each branch is ,in For level index, , This represents the total decomposition series. For example, when... When =12, the number of branches and sample rate output by each decomposition node are as follows: Figure 8 As shown.

[0125] In one or more embodiments, the multidimensional feature parameters extracted in step D include at least one of the following:

[0126] D1, the average energy value of the subband output at each decomposition stage;

[0127] D2, statistical variance of power fluctuation of sub-band signal within a preset time window;

[0128] D3, Cordic phase detectors obtain sub-band instantaneous frequency values ​​and frequency jitter statistics;

[0129] D4. The duration characteristic of the subband signal in terms of timing, that is, the continuous length of time during which the signal energy exceeds the preset threshold.

[0130] In one or more embodiments, the multidimensional feature extraction in step D is implemented in the following manner:

[0131] First, regarding the first Level 1 branch road ( ) continuous time-series data stream Maintain the recursive statistical scalar variable. First-order moment estimate (mean energy). Updated in real time according to the following recursive formula:

[0132]

[0133] in, This is the forgetting factor (typically ranging from 0.875 to 0.999), used to control the effective length of the statistical window; .

[0134] Second-order moment estimate (second-order statistic of power fluctuations) Updated in real time according to the following recursive formula:

[0135]

[0136] in, The second forgetting factor (typically ranging from 0.875 to 0.999), and... It can be configured independently. Through the above recursive statistics, the extraction of statistical features at each level does not require caching a large amount of historical sample data, but only requires maintaining a limited number of scalar registers. The entire processing flow does not depend on any large-capacity RAM array.

[0137] Then, for the extraction of instantaneous frequency values ​​in D3, the Cordic phase detection submodule performs an arctangent operation on the continuous time-series data stream of each branch subband to extract instantaneous phase information. Then calculate the instantaneous frequency value. ,in This is the current sampling period for this branch. The frequency jitter statistical characteristic is the variance of the instantaneous frequency value within a preset statistical window. .

[0138] For the timing duration characteristic in D4, a counter is used to record whether the current sub-band signal energy continuously exceeds a preset threshold. The number of sample points is counted, and the counter is reset to zero when the energy drops below the threshold.

[0139] In one or more embodiments, in order to ensure data consistency and decision stability in the multidimensional feature extraction process, the signal existence decision in step E is performed in the following manner.

[0140] In one or more embodiments, step E includes:

[0141] E1, based on the average energy value of the sub-band and power fluctuation statistical variance Calculate the first decision metric:

[0142]

[0143] in It is a very small positive number (such as 10). -9 This is used to prevent the denominator from being zero. When the signal energy is significantly higher than its own variance, A larger value indicates a higher probability of the signal's existence.

[0144] E2. Calculate the second decision metric based on the instantaneous frequency jitter statistical characteristics:

[0145]

[0146] in, This is a preset reference frequency jitter threshold (typically 1% to 5% of the subcarrier spacing). When the instantaneous frequency jitter is less than the reference threshold... A larger value indicates that the signal has stable frequency characteristics, which conforms to the characteristics of a real modulated signal.

[0147] E3. Calculate the third decision metric based on the aforementioned temporal duration characteristics:

[0148]

[0149] in, The current continuous duration (in units of sample points). This is a preset minimum duration threshold (e.g., the number of samples corresponding to several times the OFDM symbol length). When the signal duration exceeds the threshold, The value is relatively large.

[0150] E4. Weighted fusion of the first, second, and third decision metrics; if the fusion result exceeds a preset decision threshold... When this occurs, it is determined that a valid signal exists in the current sub-band channel:

[0151]

[0152] in, , , The preset weighting coefficients (typically set to...) =0.35~0.50, =0.25~0.40, =0.15~0.30, and satisfying ).when If the signal is valid, the sub-band channel is determined to have a valid signal; otherwise, it is determined to be a noise or interference channel.

[0153] In one or more embodiments, a channel shutdown operation is performed in response to the fusion decision result being lower than a preset threshold, i.e., when the current sub-band channel is determined to be an invalid or noisy channel. Specifically, the adaptive configuration controller generates a branch shutdown control signal to enable or disable the clock signal or power supply of the specific branch in the tree-like multi-level decomposition network, so that the branch corresponding to the invalid frequency band stops participating in subsequent operations and does not consume dynamic power. The branch shutdown operation adopts a two-stage mechanism of clock gating and power gating: first, the clock signal of the branch is turned off, reducing its dynamic power consumption to near zero; if the branch remains in an invalid state for a long statistical period, its power supply is further turned off, so that the static power consumption also returns to zero. In response to subsequent spectrum changes, the branch can be reactivated within microseconds.

[0154] In one or more embodiments, the method achieves the integrated fusion of channel decomposition and digital downconversion functions through the coordinated operation of the above steps. It eliminates the need for bulk data buffering and large-scale FFT operations, reduces chip area and power consumption by about 50% compared to the traditional DDC+FFT discrete architecture, and reduces pure signal processing pipeline latency by about 56%, significantly improving the energy efficiency and real-time performance of the receiving baseband chip.

[0155] refer to Figure 7 This paper illustrates a schematic diagram of a lightweight receiver demodulation system based on tree-structured subchannel decomposition according to one or more embodiments of this application. The system is applied in the digital baseband processing chips of devices such as wireless communication base stations, mobile terminals, and high-speed optical communication receivers, and includes the following core modules: an analog-to-digital converter (ADC), a tree-structured multi-level decomposition network, a multi-dimensional feature extraction module, a joint decision module, an adaptive configuration controller, and a recursive statistical maintenance module.

[0156] In one or more embodiments, the analog-to-digital conversion unit operates at a preset sampling rate. (Typical value is 983.04 Msps) The received broadband analog signal is converted from analog to digital, and the output is a digital time-domain signal sequence. .

[0157] In one or more embodiments, a tree-structured multilevel decomposition network is connected to the analog-to-digital conversion unit for processing the digital time-domain signal sequence. A multi-level tree-structured subband decomposition is performed. The tree-structured multi-level decomposition network includes cascaded multi-level decomposition nodes. to Each level of decomposition node include:

[0158] Digital mixing rotation module, used to utilize rotation factor sequence The input digital time-domain signal sequence undergoes spectral shifting. In this embodiment, the rotation angular frequency parameter of the digital mixing rotation module... Values The rotation factor sequence is These are used to shift the right semicircle of the current stage input signal spectrum, specifically the first and fourth quadrants, to near zero frequency.

[0159] A half-band filtering module, connected to the digital mixing and rotating module, is used to filter out the out-of-band image components after mixing the digital time-domain signal sequence. The half-band filtering module adopts a half-band decimation filter with equal ripple design. The number of taps of the filter is selected from any one of 7 taps, 15 taps, or 31 taps according to the signal-to-noise ratio or bandwidth requirements of the current stage input signal, and different stages of the half-band filtering module can be configured with different numbers of taps.

[0160] The 2x decimation module, connected to the half-band filtering module, is used to halve the sample rate of the filtered digital time-domain signal sequence.

[0161] Each level of decomposition node Output a continuous time-series data stream with multiple branches, and each level of decomposition node... Connect recursively level by level according to the same structure. Level decomposition nodes The number of branches output is The sample rate of each branch is .

[0162] The multidimensional feature extraction module is connected to the output of each level of the tree-like multi-level decomposition network and is used to extract multidimensional feature parameters from the continuous time-series data stream output from each level, including at least one of the following: subband average energy value, power fluctuation statistical variance, instantaneous frequency jitter feature, and time-series duration feature.

[0163] The joint decision module is connected to the multi-dimensional feature extraction module and is used to perform weighted fusion decision based on the extracted multi-dimensional features to determine the signal existence of each sub-band channel.

[0164] In one or more embodiments, each decomposition node of the tree-like multilevel decomposition network It also includes a process limiting module. The process limiting module is connected between the half-band filter module and the 2x decimation module, or between the digital mixer rotation module and the half-band filter module, and is used to limit the signal amplitude within a preset threshold range to suppress short-time pulse spike interference.

[0165] In one or more embodiments, the process limiting module performs the process limiting operation according to the following formula:

[0166] ,when hour;

[0167] ,when ;

[0168] in, These are the sample values ​​of the input signal before the amplitude limiting operation. These are the sample values ​​of the output signal after the amplitude limiting operation. To preset the amplitude threshold, It is a symbolic function.

[0169] In one or more embodiments, the multidimensional feature extraction module includes a Cordic phase detection submodule. The Cordic phase detection submodule is used to extract instantaneous phase information from the continuous time-series data stream of each sub-band. And calculate the instantaneous frequency value based on the instantaneous phase information. and its jitter statistics .

[0170] In one or more embodiments, the system further includes a recursive statistical maintenance module. This module is connected to the outputs of each level of the tree-like multi-level decomposition network and is used to maintain multiple recursive statistical scalar variables in real time during each level of decomposition. The recursive statistical scalar variables include the first-order moment estimates of each sub-band signal. and second moment estimates This is used to characterize the average energy and power fluctuation of the subband signal. The recursive statistics maintenance module does not contain a large-capacity RAM cache, but only maintains a limited number of scalar registers.

[0171] In one or more embodiments, the system further includes adaptive configuration control. This adaptive configuration controller is connected to the tree-like multilevel decomposition network and is used to dynamically adjust the configuration parameters of the tree-like multilevel decomposition network based on spectrum sensing results, channel quality indicators, or interference detection results. The configuration parameters include the currently active decomposition level. Tap coefficients of each stage of the half-band filter module, and rotation factor parameters of each stage of the digital mixer rotation module.

[0172] In one or more embodiments, the configuration adjustment logic of the adaptive configuration controller is implemented in the following manner:

[0173] First, the adaptive configuration controller receives frequency domain resource allocation information from the upper-layer protocol stack or spectrum sensing module to obtain the frequency band range currently occupied by the effective signal. ( , (Number of effective signal frequency bands);

[0174] Then, the adaptive configuration controller calculates the target decomposition depth of each tree branch based on the distribution of effective signal frequency bands. For frequency bands containing valid signals, set deep decomposition targets (such as...). The corresponding sub-channel bandwidth is For idle frequency bands or pure noise frequency bands, set shallow decomposition or no decomposition targets (e.g., That is, stop further decomposition operations on that branch).

[0175] Next, the adaptive configuration controller generates branch activation control signals based on the target decomposition depth. For branches requiring deep decomposition, it enables the decomposition nodes at each level of that branch. The clock signal and power supply; for branches that do not need to be decomposed, generate a branch shutdown control signal to enable or shut down the clock signal or power supply of a specific branch in the tree-shaped multi-level decomposition network, so that the branches corresponding to the invalid frequency band do not participate in the operation and do not consume dynamic power consumption.

[0176] Finally, the adaptive configuration controller also dynamically adjusts the decision threshold of each branch based on the current signal-to-noise ratio (SNR) estimate. and weighting coefficients , , In low signal-to-noise ratio scenarios, reduce and increase (Energy feature weighting) to improve detection sensitivity; in high signal-to-noise ratio scenarios, improve To reduce the probability of false alarms.

[0177] In one or more embodiments, when the adaptive configuration controller detects three non-contiguous valid signal frequency bands in the current spectrum (e.g., band A: 0–20MHz, band B: 50–60MHz, band C: 80–100MHz), it performs a complete 12-level decomposition only on the tree branches corresponding to bands A, B, and C, and performs clock gating shutdown on the branches corresponding to other idle frequency bands. Compared to traditional FFT schemes that require performing 4096-point uniform FFT operations across the entire 0–100MHz frequency band, the on-demand activated tree decomposition architecture of this invention reduces the computational load by approximately 70% in sparse spectrum scenarios, achieving extreme hardware resource utilization efficiency.

[0178] In one or more embodiments, to ensure the accuracy and stability of the decision results, the joint decision module employs atomic weighted fusion operations to perform signal existence decisions. Specifically, this includes:

[0179] First, the joint decision module obtains multi-dimensional feature data of the same sub-band channel from the multi-dimensional feature extraction module. );

[0180] Then, the joint decision module calculates the three decision metrics respectively. ;

[0181] Next, the joint judgment module performs weighted fusion to determine... Has the threshold been exceeded? Generate the judgment result;

[0182] Finally, the joint decision module outputs the decision result to the subsequent channel estimation equalization and demodulation module, and feeds the decision result back to the adaptive configuration controller as the basis for the next round of configuration adjustment.

[0183] In one or more embodiments, in response to any overflow or invalid data indication generated by any submodule in the multidimensional feature extraction module, the system executes an error handling procedure by the joint decision module. The error handling procedure includes at least one of the following operations: terminating the decision operation of the current subband channel, recording an error log, and reverting to the feature values ​​from the previous valid statistical period as the basis for the current decision.

[0184] In one or more embodiments, the system responds to the adaptive configuration controller detecting that configuration parameters exceed a preset valid range (such as the requested decomposition level). (If the value is greater than 12 or the extraction factor exceeds the hardware support range), a configuration error handling process is executed, wherein the configuration error handling process includes at least one of the following operations: maintaining the current valid configuration unchanged, rejecting new configuration requests and reporting configuration error flags, and reverting to the default configuration parameters.

[0185] In one or more embodiments, the system, through the collaborative operation of the aforementioned modules, achieves the architectural integration of channel decomposition and digital downconversion functions, completely eliminating redundant overhead at the hardware level between the traditional DDC multi-stage filtering link and the FFT large-scale complex number operation unit and block cache RAM. At the 7nm process, the system of this invention has an equivalent total gate count of approximately 305,000, a chip area of ​​approximately 0.112 square millimeters, and a dynamic power consumption of approximately 27mW at 400MHz. Compared to the traditional DDC+4096 pipelined FFT discrete architecture (equivalent total gate count of approximately 600,000, chip area of ​​approximately 0.22 square millimeters, and dynamic power consumption of approximately 53mW at 400MHz), it achieves approximately 49% hardware resource savings and approximately 56% pipeline latency reduction.

[0186] refer to Figure 7 The accompanying drawings illustrate the structure of a wireless communication receiving device according to one or more embodiments of this application. The device includes the aforementioned lightweight receiving and demodulation system based on tree-structured subchannel decomposition, and is specifically applied to 5G base stations, 6G base stations, 7G base stations, mobile terminals, IoT terminals, satellite communication terminals, or high-speed optical communication receivers.

[0187] Taking a 5G NR base station as an example, the device includes a radio frequency front-end module, an analog-to-digital converter (ADC), and a digital baseband processing chip. The digital baseband processing chip integrates the aforementioned lightweight receiver demodulation system based on tree-structured subchannel decomposition.

[0188] In one or more embodiments, the digital baseband processing chip of the wireless communication receiving device includes the following core components:

[0189] The radio frequency front-end module is used to receive spatial electromagnetic wave signals captured by the antenna, and outputs an analog intermediate frequency signal after low-noise amplification, mixing and downconversion to intermediate frequency;

[0190] An analog-to-digital converter (ADC) is connected to the radio frequency front-end module at a preset sampling rate. (Typical value is 983.04 Msps) The analog intermediate frequency signal is sampled and quantized to output a digital time-domain signal sequence. ;

[0191] A tree-shaped multi-level decomposition network, integrated inside the digital baseband processing chip and connected to the output of the analog-to-digital converter, is used to perform multi-level tree-shaped subband decomposition on the digital time-domain signal sequence.

[0192] A multi-dimensional feature extraction module is integrated inside the digital baseband processing chip and connected to the outputs of each level of the tree-like multi-level decomposition network;

[0193] The joint decision module is integrated inside the digital baseband processing chip and is connected to the multi-dimensional feature extraction module.

[0194] An adaptive configuration controller is integrated inside the digital baseband processing chip, connected to the tree-like multi-level decomposition network, and communicates with the upper-layer protocol stack or spectrum management module to obtain spectrum resource allocation information.

[0195] In one or more embodiments, the wireless communication receiving device is a mobile phone terminal. In this embodiment, the maximum number of decomposition levels of the tree-like multi-level decomposition network... The processing bandwidth requirements of the mobile terminal are set to 10 levels (corresponding to 1024 sub-channels) to balance the power consumption and performance requirements of the terminal device. The battery management unit of the mobile terminal is connected to the adaptive configuration controller. When the battery level is lower than a preset threshold (e.g., 15%), the adaptive configuration controller further limits the number of activated decomposition branches and the maximum decomposition level to extend the terminal's battery life.

[0196] In one or more embodiments, the wireless communication receiving device is a high-speed optical communication receiver. In this embodiment, the analog-to-digital converter is a high-speed ADC with a sampling rate of... Typical values ​​are 64 Gsps or higher. The number N of the tree-like multi-level decomposition network is set according to the spectral partitioning granularity requirements of the Flex-Grid optical communication system, supporting non-uniform and unequal bandwidth sub-band partitioning to adapt to the heterogeneous spectral characteristics of the optical communication system. The adaptive configuration controller dynamically adjusts the activation state of each decomposition node based on the spectral occupancy information provided by Optical Channel Monitoring (OCM), performing decomposition processing only on the frequency band carrying the optical carrier signal.

[0197] In one or more embodiments, the wireless communication receiving device is an Internet of Things (IoT) terminal. In this embodiment, since IoT terminals typically operate in narrowband, low-data-rate scenarios, the maximum number of decomposition levels in the tree-structured multi-level decomposition network is... It is limited to 6 levels (corresponding to 64 sub-channels), and most branches are turned off when idle, with only the decomposed branches corresponding to the frequency band of the target signal remaining active. The multi-dimensional feature extraction module and the joint decision module operate at a low sampling rate (e.g., This further reduces system power consumption.

[0198] In one or more embodiments, the adaptive configuration controller of the device implements multi-mode operation according to the following logic:

[0199] First, when the device starts up, the adaptive configuration controller executes the full-band scanning mode—activating all branches of the tree-like multi-level decomposition network, performing a complete N-level decomposition, identifying all valid signal frequency bands in the current environment through multi-dimensional feature extraction and joint decision-making, and constructing an initial spectrum occupancy map;

[0200] Then, in steady-state operation mode, the adaptive configuration controller selectively activates the decomposed branches corresponding to valid signal frequency bands and shuts down invalid frequency band branches based on the spectrum occupancy map. In steady-state mode, the device operates in an on-demand decomposition mode, reducing power consumption to 30%–50% of the fully activated mode.

[0201] Next, in spectrum change monitoring mode, the adaptive configuration controller periodically (e.g., every 10ms) performs a fast coarse-grained scan of the original idle frequency band (only performing the first 3-4 levels of shallow decomposition) to detect whether any new signals appear. If a new signal is detected, the decomposition depth corresponding to that frequency band is dynamically expanded, switching from shallow decomposition to deep decomposition;

[0202] Finally, in low-power standby mode, the tree-like multi-level decomposition network only keeps the first level of decomposition active, monitoring energy changes across the entire frequency band. Only when the total energy exceeds the wake-up threshold is the deeper decomposition branches activated step by step. This hierarchical wake-up mechanism reduces standby power consumption to less than 10% of that in the fully active mode.

[0203] In one or more embodiments, to ensure the continuity of signal processing when the device switches between different operating modes, the adaptive configuration controller adopts a "build first, tear down later" branch switching strategy. Specifically:

[0204] When it is necessary to switch a certain idle frequency band branch from the "off" state to the "active" state, first enable the clock signal and power supply of the branch, wait for the transient response of the filter to stabilize (the typical waiting time is 3 to 5 times the filter group delay), and then switch the data stream to the branch for normal decomposition processing.

[0205] When it is necessary to switch a branch from the "active" state to the "off" state, the current sample sequence being processed is completed first, then clock-gated shutdown is performed, and finally power-gated shutdown is performed after confirming that there is no data residue.

[0206] In one or more embodiments, in response to a device operating mode switching failure (such as the filter failing to lock within a preset time after branch activation or abnormal data from the feature extraction module), the adaptive configuration controller executes a mode switching error handling process, including at least one of the following operations: maintaining the previous valid operating mode, resetting the abnormal decomposition branch to the default state, recording an error log and reporting it to the device management unit.

[0207] In one or more embodiments, in response to a scenario where the device detects an abnormally strong interference signal causing frequent triggering of the process limiting module, the device executes an anti-interference enhancement mode: the threshold of each stage of the process limiting module... Automatically reduce by 6dB, while the joint decision module will (Frequency jitter feature weights) and The (persistence feature weight) is increased to 1.5 times the normal value to enhance the ability to identify real signals under strong interference environments.

[0208] In one or more embodiments, through the above configuration, the wireless communication receiving device achieves end-to-end streaming processing from antenna to demodulation output without requiring large-capacity data buffering. In typical 5G NR base station applications, the receiving sensitivity of this device is improved by approximately 3–6 dB compared to the traditional DDC+FFT architecture, and the false alarm rate is reduced by approximately one order of magnitude in low signal-to-noise ratio (SNR = -5 dB to 0 dB) environments. This device can comprehensively replace the traditional DDC+FFT-based receiving baseband architecture in various RF and optical communication devices, solving the common industry pain points of high power consumption, large area, low sensitivity, and rigid spectrum grid in FFT-type signal processing.

[0209] In one or more embodiments, this application also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it can implement the steps of the aforementioned lightweight reception and demodulation method based on tree subchannel decomposition. The computer-readable storage medium includes, but is not limited to, non-volatile storage media such as ROM (Read-Only Memory), RAM (Random Access Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, solid-state drive (SSD), embedded multimedia card (eMMC), USB flash drive, CD-ROM, and DVD.

[0210] In one or more embodiments, when the computer program is executed by a processor, step A is first performed: at a preset sampling rate. The received broadband analog signal is converted from analog to digital to obtain a digital time-domain signal sequence. Specifically, the processor reads the ADC sampling data through the ADC driver interface and configures the parameters according to the preset sampling rate. (e.g., 983.04 Msps) controls the sampling clock of the ADC, converting the analog signal into a digital time-domain signal sequence and storing it in the buffer.

[0211] Then proceed to step B: for the digital time-domain signal sequence Perform multi-level tree-structured subband decomposition. The processor calls the tree decomposition algorithm module to perform multi-level recursive processing operations of "digital mixing and rotation → half-band filtering → decimation by 2" on the input digital time-domain signal sequence.

[0212] Next, proceed to step C: During each level of decomposition, the sub-band signals of each branch output from each level of decomposition are output to the next level or stored in an intermediate buffer in the form of a continuous time-series data stream.

[0213] Then, proceed to step D: extract the multidimensional feature parameters of each branch subband signal based on the continuous time-series data stream output from each level of decomposition.

[0214] Finally, step E is executed: based on the multidimensional feature parameters, signal existence determination and demodulation processing are performed on each branch sub-band channel.

[0215] In one or more embodiments, the computer program is configured to execute within an embedded processor core of a digital signal processor (DSP) or application-specific integrated circuit (ASIC). In this embodiment, the tree decomposition algorithm module is a compiler-optimized fixed-point implementation, the tap coefficients of each stage filter are stored in read-only memory in fixed-point (e.g., Q15 format), and the digital mixing rotation operation is implemented via a lookup table—a rotation factor sequence. The cosine and sine values ​​are pre-calculated and stored as a lookup table, and then read through the index at runtime to reduce the complexity of real-time calculation.

[0216] In one or more embodiments, the computer program is configured to execute on a general-purpose processor (such as an ARM Cortex-A series or x86 architecture) for simulation verification of communication baseband algorithms or a software-defined radio (SDR) platform. In this embodiment, the tree decomposition algorithm module is implemented in floating-point numbers, the tap coefficients of each stage of the filter are stored in double-precision floating-point numbers, and the digital mixing rotation operation is calculated in real time by calling trigonometric functions from a standard math library. Although this software implementation is less power-efficient and has lower real-time performance than the hardware implementation, it has the advantages of convenient development and debugging, and rapid iterative optimization.

[0217] In one or more embodiments, the computer program also implements the logical function of an adaptive configuration controller: dynamically adjusting the configuration parameters of the tree decomposition based on the spectrum sensing results. Specifically, during the initialization phase or operation, the computer program reads spectrum sensing data or receives frequency domain resource allocation instructions from the upper-layer protocol stack, calculates the optimal tree decomposition configuration (including activated branches, decomposition depth at each level, filter parameters, etc.), and then writes the configuration parameters into hardware registers or updates the algorithm parameters in memory through the register configuration interface.

[0218] In one or more embodiments, in order to ensure data consistency in a multi-threaded or parallel processing environment, when the computer program is executed on a multi-core processor, the tree decomposition processing at each level is assigned to different processing cores for parallel execution, and data transfer between each level is achieved through a lock-free queue to avoid performance loss caused by lock contention.

[0219] In one or more embodiments, inter-level data transfer is implemented using a lock-free circular buffer with a single producer-single consumer (SPSC) architecture: After each branch of the first level completes a 2x extraction, it writes the output data to the second level. In the ring buffer zone corresponding to the first-level branch; Each processing thread reads data from this circular buffer for further processing at the next level. The read / write pointers of the circular buffer are updated atomically to ensure the atomicity and sequential consistency of data transfer.

[0220] In one or more embodiments, in response to an abnormal condition of a circular buffer overflow (i.e., data generation outpacing consumption), the computer program executes an error handling procedure. This error handling procedure includes at least one of the following operations: discarding the most recently generated data sample to avoid blocking, sending a backpressure signal to the adaptive configuration controller requesting a reduction in the preceding processing rate, and logging an overflow error.

[0221] In one or more embodiments, in response to a computer program detecting a mismatch between configuration parameters and hardware resources during execution (such as the requested decomposition level exceeding the maximum number of levels supported by the processor or insufficient memory), a configuration verification error handling process is executed. The error handling process includes at least one of the following operations: terminating program execution and returning an error code, automatically downgrading to the maximum parameter configuration supported by the hardware and continuing execution, logging an error, and prompting the user to modify the configuration.

[0222] In one or more embodiments, the processor, through the execution of the aforementioned computer program, achieves the fusion processing of channel decomposition and digital downconversion functions, eliminating the need for bulk data caching and large-scale FFT operations. Experimental results on a software-defined radio platform show that, compared to traditional FFT+DDC software implementations, the computer program implementation of this invention reduces the CPU utilization rate for receiving and processing from approximately 65% ​​(4096-point FFT+DDC) to approximately 35% (tree decomposition), memory usage from approximately 2MB to approximately 0.5MB, and processing latency from approximately 15μs to approximately 6μs. This makes it particularly suitable for resource-constrained embedded software-defined radio platforms and real-time communication signal processing scenarios.

[0223] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0224] The above technical solutions only embody the preferred technical solutions of the present invention. Any modifications that may be made by those skilled in the art to certain parts therein embody the principles of the present invention and fall within the protection scope of the present invention.

Claims

1. A lightweight receiver demodulation method based on tree-structured subchannel decomposition, characterized in that, The method includes the following steps: Step A: Using the preset sampling rate The received broadband analog signal is converted from analog to digital to obtain a digital time-domain signal sequence. ; Step B: Process the digital time-domain signal sequence Perform multi-level tree-structured subband decomposition; Step C: In the decomposition process of the multi-level tree sub-bands, the branch sub-band signals output by each level of the decomposition process are continuous time-series data streams; Step D: Extract the multidimensional feature parameters of each branch subband signal based on the continuous time-series data stream output from each decomposition process. Step E: Perform signal existence determination and demodulation processing on each branch sub-band channel based on the multi-dimensional feature parameters; The signal existence determination in step E includes: E1. Calculate the first decision metric based on the sub-band average energy value and power fluctuation statistical variance; E2. Calculate the second decision metric based on the statistical characteristics of instantaneous frequency jitter; E3. Calculate the third decision metric based on the duration characteristics of the time series; E4. The first, second and third decision metrics are weighted and fused. When the fusion result exceeds the preset decision threshold, it is determined that there is a valid signal in the current sub-band channel.

2. The method according to claim 1, characterized in that, Each level of the multi-level tree subband decomposition in step B includes three processing operations: digital mixing rotation, half-band filtering, and 2x decimation. All processing operations at each level are executed recursively according to the same structure. The multi-level tree subband decomposition process also includes embedding a process limiting operation before or after each level of half-band filtering. The process limiting operation is used to limit the amplitude of the input signal within a preset amplitude threshold range to suppress short-time pulse spike interference. The configuration parameters for each level of the tree-structured subband decomposition are adaptively adjusted based on the spectral distribution characteristics of the currently received signal, channel state information, or interference distribution information. These configuration parameters include the number of decomposition levels. The tap coefficients and twitch factor sequence parameters of each stage of the filter; The adaptive adjustment includes: performing deep multi-level decomposition in signal-dense frequency bands to obtain fine-grained sub-band division, and stopping further decomposition or performing shallow coarse-grained decomposition in signal-sparse or idle frequency bands, so as to achieve custom sub-band channel division with non-uniformity, asymmetry, unequal intervals, and unequal bandwidth.

3. The method according to claim 1, characterized in that, In step C, the first The number of branches output by the level decomposition is The sample rate of each branch is ,in For level index, , This represents the total decomposition series.

4. The method according to claim 1, characterized in that, The multidimensional feature parameters extracted in step D include at least one of the following: D1, the average energy value of the subband output at each decomposition stage; D2, statistical variance of power fluctuation of sub-band signal within a preset time window; D3, Cordic phase detectors obtain sub-band instantaneous frequency values ​​and instantaneous frequency jitter statistics; D4. Timing duration characteristics of subband signals, i.e., the continuous duration for which the signal energy exceeds a preset threshold.

5. A lightweight receiver demodulation system based on tree-structured subchannel decomposition, the system being used to implement the method as described in any one of claims 1 to 4, characterized in that, include: The analog-to-digital conversion unit operates at a preset sampling rate. The received broadband analog signal is converted from analog to digital, and a digital time-domain signal sequence is output. A tree-shaped multi-level decomposition network, connected to the analog-to-digital conversion unit, is used to perform multi-level tree-shaped sub-band decomposition on the digital time-domain signal sequence; The tree-structured multi-level decomposition network includes: cascaded multi-level decomposition nodes, each level of decomposition node including: A digital mixing rotation module is used to perform spectral shifting on the digital time-domain signal sequence input to the analog-to-digital conversion unit using a rotation factor sequence; A half-band filtering module, connected to the digital mixing and rotating module, is used to filter out the out-of-band mirror component after the digital time-domain signal sequence has been mixed. A 2x decimation module, connected to the half-band filtering module, is used to halve the sample rate of the filtered digital time-domain signal sequence. Each decomposition node outputs a continuous time-series data stream of multiple branches, and each level of decomposition node is recursively connected according to the same structure. Also includes: The multidimensional feature extraction module is connected to the output of each level of the tree-like multi-level decomposition network and is used to extract at least one of the following from the continuous time-series data stream output at each level: subband average energy value, power fluctuation statistical variance, instantaneous frequency jitter feature, and time-series duration feature; The joint decision module, connected to the multi-dimensional feature extraction module, is used to perform weighted fusion decision based on the extracted multi-dimensional features to determine the signal existence of each sub-band channel. The multidimensional feature extraction module includes a Cordic phase detection submodule, which is used to extract instantaneous phase information from the continuous time-series data stream of each sub-band, and calculate the instantaneous frequency value and its jitter statistical characteristics based on the instantaneous phase information.

6. The system according to claim 5, characterized in that, Each decomposition node of the tree-shaped multi-level decomposition network also includes a process limiting module, which is connected between the half-band filter module and the 2x decimation module, or between the digital mixing and rotation module and the half-band filter module, for limiting the signal amplitude within a preset threshold range.

7. The system according to claim 5, characterized in that, The system also includes: The recursive statistics maintenance module is connected to the output of each level of the tree-shaped multi-level decomposition network. It is used to maintain multiple recursive statistical scalar variables in real time during the decomposition process. The recursive statistical scalar variables include the first moment estimate and the second moment estimate of each sub-band signal, which are used to characterize the energy mean and power fluctuation of the sub-band signal. The recursive statistics maintenance module does not contain a large-capacity RAM cache, but only maintains a limited number of scalar registers; The system also includes: An adaptive configuration controller, connected to the tree-like multi-level decomposition network, is used to dynamically adjust the configuration parameters of the tree-like multi-level decomposition network according to spectrum sensing results, channel quality indications, or interference detection results. The configuration parameters include the currently active decomposition level, the tap coefficients of each half-band filter module, and the rotation factor parameters of each digital mixing rotation module. The adaptive configuration controller is further configured to: generate a branch activation control signal based on the spectrum distribution information. The branch activation control signal is used to enable or disable the clock signal or power supply of a specific branch in the tree-like multi-level decomposition network, so that the branch corresponding to the invalid frequency band does not participate in the calculation and does not consume dynamic power consumption.

8. A wireless communication receiving device, characterized in that, Includes the system according to any one of claims 5 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it is able to implement the steps of the method as described in any one of claims 1 to 4.

Images