A time-domain signal generation method and device for AI compression encoding
By modulating the constellation points, rearranging the groups, and limiting the amplitude of AI compressed encoded data, the signal distortion problem when generating time-domain signals from AI compressed encoded data is solved, thereby improving the signal-to-noise ratio of the received signal and the data decoding performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA ACADEMY OF INFORMATION & COMM
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-05
AI Technical Summary
When AI-compressed data is directly used to generate time-domain signals, there is a problem of signal distortion caused by the peak-to-average power ratio, resulting in a decrease in the quality of the received signal.
By performing constellation point modulation, group rearrangement, and amplitude limiting on the bitstream output by AI compression encoding, the position of the signal peak is controlled, and amplitude is limited at a preset position to generate an OFDM time-domain waveform signal.
Without changing the transmit power, it effectively reduces the peak-to-average power ratio of the signal, improves the signal-to-noise ratio of the received signal, and ensures data decoding performance.
Smart Images
Figure CN122160217A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of wireless communication, and in particular to a time-domain signal generation method, transmitting device, receiving device, and system for AI compression coding. Background Technology
[0002] Using AI technology to compress and encode business data (such as streaming media data like images, videos, and audio) before transmission can significantly reduce the total amount of data transmitted, thereby effectively reducing the wireless communication resources required for transmission. However, because AI-compressed data often exhibits specific distribution characteristics, directly using this data to generate time-domain signals and transmitting them through radio frequency equipment can easily lead to signal distortion and other problems, resulting in a severe deterioration in the quality of the received signal. For the bitstream obtained after AI compression and encoding, the OFDM time-domain waveform generated using traditional methods is as follows: Figure 2 As shown, some peaks in the signal occupy a large portion of the transmitted power, resulting in insufficient energy allocation for the remaining data and thus a decrease in signal reception quality.
[0003] To address this issue, existing technologies employ traditional methods such as windowing or direct peak reduction. However, these methods suffer from limited improvement in peak-to-average power ratio (PAPR) or poor demodulation at the receiving end. Therefore, there is an urgent need for a time-domain signal generation method that can effectively solve the signal distortion problem, improve the signal-to-noise ratio (SNR) of the received signal without changing the transmit power, and thus ensure data decoding performance. Summary of the Invention
[0004] This application proposes a time-domain signal generation method and apparatus for AI compression coding, which solves the problem that when AI compressed coding data is directly generated into time-domain signals for transmission, the signal peak-to-average power ratio and distortion are caused by the data distribution characteristics, which leads to a decrease in the quality of the received signal.
[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a time-domain signal generation method for AI compression coding, applied at a transmitting end. The method includes: performing constellation point modulation on the bitstream output by AI model compression coding to obtain an initial constellation point sequence; grouping the initial constellation point sequence and rearranging the constellation point order within each group based on a specific random sequence, so that the high-amplitude peak value in the generated time-domain waveform signal appears at a preset position; performing an IFFT transform on the rearranged constellation point sequence to generate an OFDM time-domain waveform signal; performing amplitude limiting on the peak value appearing at the preset position in the OFDM time-domain waveform signal; and combining the multiple amplitude-limited OFDM time-domain waveform signals into a complete transmission waveform.
[0006] In some embodiments, grouping the initial constellation point sequence includes: dividing the initial constellation point sequence into multiple subsequences according to the transmission resource parameters of the communication system, each subsequence corresponding to a set of constellation points to be modulated onto its subcarriers by an OFDM symbol.
[0007] In some embodiments, the random sequence is reused in different groups.
[0008] In some embodiments, the limiting process includes: for each OFDM time-domain waveform symbol, dynamically determining a local limiting threshold based on the peak amplitude of the symbol at the preset position, and limiting the symbol.
[0009] Secondly, this application provides a transmitting device for AI compression coding, used to implement the method described in the first aspect. The device includes: a modulation module for performing constellation point modulation on the bitstream output by AI compression coding to generate an initial constellation point sequence; a constellation point rearrangement module for grouping the initial constellation point sequence and rearranging the constellation points within each group based on a random sequence, so that the high-amplitude peak value in the subsequently generated time-domain waveform signal appears at a preset position; an IFFT conversion module for converting the rearranged constellation point sequence into an OFDM time-domain waveform signal; a limiting module for limiting the peak value of the OFDM time-domain waveform signal at the preset position; and a combining module for combining multiple limited waveform signals into a complete transmission waveform.
[0010] Thirdly, this application provides a time-domain signal receiving method for AI compression coding, applied at a receiving end. The method includes: performing an FFT transform on the received time-domain waveform signal to obtain a received constellation point sequence; determining a constellation point offset compensation amount based on the amplitude information of the preset position peaks in the time-domain waveform signal; performing offset compensation on the received constellation point sequence according to the offset compensation amount; rearranging the offset-compensated constellation point sequence using a specific random sequence to restore the original constellation point sequence; and demapping the original constellation point sequence into a bitstream, inputting it into an AI decompression model to recover the original service data.
[0011] In some embodiments, the offset compensation amount is calculated by averaging the difference between the peak amplitude of multiple OFDM symbols at the preset position and a preset threshold.
[0012] In some embodiments, at least one OFDM symbol as a reference signal is extracted from the received time-domain waveform signal; the offset compensation amount is determined based on the peak amplitude of the reference signal at the fixed position and a preset reference threshold.
[0013] In some embodiments, the offset compensation and rearrangement restoration are accomplished through joint iterative processing, including the following steps: tentatively rearranging the received constellation point sequence using the candidate sorting of the random sequence; evaluating and updating the estimated value of the offset compensation amount based on the constellation point clustering characteristics of the rearranged sequence; iterating the above process until the offset compensation amount that makes the constellation point clustering of the rearranged sequence optimal is found.
[0014] Fourthly, this application provides a receiving end device for AI compression coding, used to implement the method described in the third aspect. The device includes: a demodulation module for converting a received time-domain waveform signal into a constellation point sequence; an offset estimation module for determining a constellation point offset compensation amount based on the amplitude information of the peak values at preset positions in the time-domain waveform signal; an offset compensation module for performing offset compensation on the constellation point sequence according to the offset compensation amount; a constellation point rearrangement module for rearranging the compensated constellation point sequence according to a random sequence used by the transmitting end to restore the original constellation point sequence; and a demapping module for demapping the original constellation point sequence into a bitstream and inputting it into an AI decompression model.
[0015] Fifthly, this application provides a communication system for AI compression coding. The system includes: a transmitting device as described in the second aspect; and a receiving device as described in the fourth aspect; the transmitting device is used to generate and transmit an OFDM time-domain waveform signal that has undergone amplitude limiting and rearrangement processing; the receiving device is used to receive the waveform signal and perform offset compensation and rearrangement to recover the original data.
[0016] The above-mentioned at least one technical solution adopted in the embodiments of this application can achieve the following beneficial effects: by rearranging the peak position of the control signal and limiting the amplitude, the signal distortion problem can be effectively solved, and the signal-to-noise ratio of the received signal can be improved without changing the transmission power, thereby ensuring the decoding performance of AI compressed encoded data. Attached Figure Description
[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a system structure diagram of this application; Figure 2 To compress and encode waveform signals generated by traditional methods for AI-oriented applications; Figure 3 This scheme is used for AI-oriented compression encoding of waveform signals generated based on the proposed solution. Figure 4 This is a flowchart illustrating the method used in this application at the sending end; Figure 5This is an example of a transmitter structure; Figure 6 This is a schematic diagram of the waveform at the transmitting end; Figure 7 This is a flowchart illustrating the method used in this application at the receiving end; Figure 8 This is an example of a receiver structure; Figure 9 To receive the constellation diagram, where (a) represents the directly demodulated constellation diagram and (b) represents the constellation diagram after translation; Figure 10 This is a comparison of signal reception performance under different schemes. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] This solution addresses the scenario of AI-compressed encoded service data bitstreams by providing improved methods for time-domain waveform signal generation and reception at the radio frequency end, ensuring signal transmission and reception quality. For the bitstream obtained after AI compression encoding, the OFDM time-domain waveform generated using traditional methods is as follows: Figure 2 As shown, some peaks in the signal occupy a large portion of the transmitted power, resulting in insufficient energy distribution to the remaining data and thus degrading the signal reception quality. Traditional methods such as windowing or directly attenuating peaks have limited effect on improving the peak-to-average power ratio (PAPR) or poor demodulation at the receiving end. To address this, this solution proposes a waveform improvement method. The core idea is to generate an OFDM signal by rearranging the order of the modulation constellation symbols, thus causing peaks to appear at fixed, finite positions. The peaks are then attenuated. The waveform generated using this method is shown below. Figure 3 As shown, compared to Figure 2 This allows for the generation of waveform signals with more uniform intensity. The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.
[0020] Figure 1This application describes a communication system that includes a transmitting device and a receiving device. The system's workflow is geared towards AI compressed encoded service data bitstream scenarios, aiming to provide improved methods for time-domain waveform signal generation and reception at the radio frequency end, ensuring signal transmission and reception quality. In this system, the transmitting device executes the transmitting method as described in Example 1, generating and transmitting an OFDM time-domain waveform signal that has undergone amplitude limiting and rearrangement processing. The receiving device executes the receiving method as described in Example 2, receiving the waveform signal and performing offset compensation and rearrangement to recover the original data. Through collaborative improvements at the transmitting and receiving ends, this system effectively solves the signal distortion problem that easily occurs when AI compressed encoded data is directly transmitted, improving the signal-to-noise ratio of the received signal without changing the transmit power.
[0021] Figure 2 The figure illustrates the waveform signal generated by a traditional scheme for AI compression coding. This figure is used to compare and explain the technical problem that this application aims to solve. As shown in the figure, if the bitstream obtained after AI compression coding is directly modulated and OFDM waveform generated, the resulting time-domain signal contains some randomly distributed peaks with high amplitude. These peaks occupy a large proportion of the transmitted power, resulting in insufficient energy distribution for the remaining part of the signal carrying effective data, thus causing a degradation in signal quality at the receiving end. This power amplifier nonlinear distortion problem caused by the signal peak-to-average power ratio (PAPR) is one of the core problems that this application aims to solve.
[0022] Figure 3 The waveform signal generated by this scheme for AI compression coding is shown. This figure is used to visually illustrate the beneficial effects of the technical solution presented in this application. Figure 2 Compared to traditional waveforms, the waveform generated using the method described in this application exhibits a more uniform signal intensity distribution. Specifically, by grouping the initial constellation point sequence and rearranging the order of the constellation points within each group based on a specific random sequence, the originally randomly distributed high-amplitude peaks of the signal are concentrated and shifted to preset fixed positions. Subsequently, by performing amplitude limiting processing on these peaks at preset positions, the waveform obtained is as follows: Figure 3 The waveform shown is an example of this. It is evident that this solution can effectively reduce the peak-to-average power ratio of the transmitted signal without significant information loss, thus creating conditions for subsequent RF amplification and high-quality reception.
[0023] Example 1: Transmitter Method and Device This embodiment describes in detail the time-domain signal generation method for AI compression coding and the implementation of the transmitting device.
[0024] refer to Figure 4 The sending process shown is compared to Figure 1 The flowchart shown adds two steps, constellation rearrangement and amplitude limiting, to the OFDM waveform generation and transmission steps. The specific steps are as follows: Step 110: Perform constellation point modulation on the bitstream output by the compressed encoding of the AI model to obtain the initial constellation point sequence.
[0025] Specifically, the bitstream obtained by compression encoding of the AI model is subjected to constellation point modulation and converted into an OFDM time-domain waveform signal for transmission. This is the process of mapping digital bit information to symbol points on the complex plane, such as using modulation methods like QPSK or 16QAM.
[0026] Step 120: Group the initial constellation point sequence, rearrange the constellation point order in each group based on a specific random sequence, so that the high amplitude peak in the generated time-domain waveform signal appears at a preset position.
[0027] In this step, "grouping" can be understood as: dividing the initial constellation point sequence into multiple subsequences according to the transmission resource parameters of the communication system, with each subsequence corresponding to a set of constellation points to be modulated onto its subcarriers by an OFDM symbol.
[0028] For example, the number of OFDM symbols needed to carry data can be determined based on the total number of constellation points to be transmitted and the number of constellation points that each OFDM symbol can carry, naturally forming packets. The number of packets depends on the system configuration, such as bandwidth, number of subcarriers, and modulation order.
[0029] In one specific implementation, a random sequence s for rearranging constellation points is first generated. Then, the initial constellation point sequence Q0 is rearranged according to s, and the new sequence is denoted as Q1. Taking 16QAM as an example, assuming the constellation point sequence obtained by modulation based on the bitstream sequence is [1,5,10,16,12], and the rearranged sequence s is [4,3,5,2,1], then the new rearranged sequence is [16,10,12,5,1]. The random sequence s can be reused in different groups to reduce the complexity of generation and synchronization.
[0030] Step 130: Perform IFFT transform on the rearranged constellation point sequence to generate OFDM time-domain waveform signal.
[0031] The rearranged constellation point sequence Q1 is allocated to each resource block of the OFDM. An IFFT is then performed on each OFDM symbol to convert the constellation point sequence into a time-domain waveform signal. The waveform effect is as follows: Figure 6As shown: For each waveform on OFDM, compared to the waveform of Q0, which has multiple peaks and whose peak positions are not fixed, the peak of Q1 only appears at the beginning position of the time domain signal (i.e., the preset position). This is because the rearrangement operation changes the phase relationship of the frequency domain symbols, causing the time domain energy to concentrate near the starting sampling point after IFFT.
[0032] Step 140: Limit the peak value of the OFDM time-domain waveform signal at the preset position.
[0033] For the initial peak value of the waveform signal of each OFDM symbol in Q1, it is determined whether it exceeds the threshold P. If it exceeds the threshold, it is limited to the value of P. This is to further reduce the peak-to-average power ratio (PAPR) of the signal.
[0034] In an optimized embodiment, the clipping process includes: for each OFDM time-domain waveform symbol, dynamically determining a local clipping threshold based on the peak amplitude of the symbol at a preset position, and clipping the symbol. This means that not all symbols use the same global threshold P, but rather a more reasonable clipping threshold can be adaptively determined based on the peak magnitude of each symbol, thereby achieving a better balance between suppressing peaks and maintaining signal quality. For example, a relatively strict clipping threshold can be set for symbols with higher peak values, while symbols with lower peak values can be treated more leniently or even not clipped, to minimize demodulation performance loss.
[0035] Step 150: Combine the multiple OFDM time-domain waveform signals after clipping into a complete transmit waveform.
[0036] By concatenating and combining the waveforms of each OFDM symbol, a complete transmitted waveform can be obtained, as shown in the image. Figure 3 As shown.
[0037] Figure 5 An embodiment of a transmitting device implementing the above method is shown. This device specifically includes the following functional modules, which work together to complete the process described in Embodiment 1: Modulation module 501: Its function corresponds to step 110, and it is used to perform constellation point modulation on the bit stream output by AI compression encoding to generate an initial constellation point sequence.
[0038] Constellation point rearrangement module 502: Its function corresponds to step 120, and it is used to group the initial constellation point sequence and rearrange the constellation points in each group based on a random sequence so that the high amplitude peak value in the subsequently generated time-domain waveform signal appears at a preset position. This module is the core of realizing peak position control.
[0039] IFFT conversion module 503: Its function corresponds to step 130, and it is used to convert the rearranged constellation point sequence into an OFDM time-domain waveform signal.
[0040] Limiting module 504: Its function corresponds to step 140, and it is used to limit the peak value of the OFDM time-domain waveform signal at the preset position. This module can be configured to implement the dynamic local limiting strategy as described in the embodiment.
[0041] Combination module 505: Its function corresponds to step 150, and it is used to combine multiple amplitude-limited waveform signals into a complete transmission waveform.
[0042] Figure 6 This diagram illustrates the waveform at the transmitting end, specifically showing the time-domain waveform signal of a single OFDM symbol generated by the IFFT transformation of the rearranged constellation point sequence Q1. This figure is a graphical representation of the effect of the technical feature in claim 1, "performing an IFFT transformation on the rearranged constellation point sequence to generate an OFDM time-domain waveform signal." As shown, for the rearranged constellation point sub-sequence Q1, the waveform characteristics generated by its IFFT transformation are: high-amplitude peaks only appear at the beginning position of the time-domain signal (i.e., the preset position), while the signal amplitude at other time points is relatively low and stable. This contrasts sharply with the waveform generated by the unrearranged sequence Q0, where the peak positions are randomly distributed. This feature is the physical basis for the controllable peak management of this scheme, enabling subsequent amplitude limiting operations to be performed efficiently and accurately at fixed positions.
[0043] Example 2: Receiving End Method and Device This embodiment describes in detail the implementation of a time-domain signal receiving method and receiving device for AI compression coding.
[0044] refer to Figure 7 The receiving end process shown in this diagram involves the receiving end recovering the original service data from the received time-domain waveform signal. Compared to traditional solutions, this solution adds features such as... Figure 8 The block diagram shows two steps (offset compensation and rearrangement restoration).
[0045] Step 210: Perform FFT transformation on the received time-domain waveform signal to obtain the received constellation point sequence.
[0046] This is the process of converting the received time-domain sampling points back into a frequency-domain constellation point sequence using a fast Fourier transform.
[0047] Step 220: Determine the constellation point offset compensation amount based on the amplitude information of the preset position peak in the time-domain waveform signal.
[0048] Because the transmitter performs a limiting operation, it causes a linear shift in the constellation points. For example... Figure 9As shown in (a), directly based on Figure 3 The waveform demodulated constellation diagram shown has a linear offset, so offset compensation is required.
[0049] In one embodiment, the offset compensation amount Δ can be calculated by averaging the differences between the peak amplitudes of multiple OFDM symbols at the preset position and a preset threshold. Specifically, the constellation point offset Δ is predetermined. Considering the differences in the peak values of the waveform signals corresponding to each OFDM symbol, the offset Δ can be calculated by averaging, as shown in the following formula: Δ i =(1 / N) * Σ (x i (j) - P), where x i (1) represents the initial position amplitude of the time-domain waveform signal of the i-th OFDM symbol, and the summation interval is j = 1 to N.
[0050] In another preferred embodiment, at least one OFDM symbol as a reference signal can be extracted from the received time-domain waveform signal; the offset compensation amount is determined based on the peak amplitude of the reference signal at the fixed position and a preset reference threshold. Here, the reference signal can be a known training sequence or pilot symbol inserted by the transmitter, whose theoretical peak value is known. Therefore, by measuring its actual received peak value, the offset introduced by the channel can be estimated more accurately and robustly, avoiding random errors caused by relying solely on data symbol estimation.
[0051] Step 230: Perform offset compensation on the received constellation point sequence according to the offset compensation amount.
[0052] Directly based Figure 3 The waveform demodulated shows a constellation diagram with a linear offset, therefore offset compensation is needed. For the received offset constellation sequence Q1' (e.g., ... Figure 9 (a) A translation operation with a translation amount of Δ is performed to obtain the unoffset constellation sequence Q2', as shown in (a). Figure 9 As shown in (b).
[0053] Step 240: Use a specific random sequence to rearrange the offset-compensated constellation point sequence to restore the original constellation point sequence.
[0054] It should be noted that both the sending and receiving ends use a "specific random sequence," meaning that the random sequence used by the receiving end is the same as that used by the sending end.
[0055] Based on the initial sequence s, the constellation sequence Q2' is reordered to restore Q0'. It should be noted that the order of steps 230 (offset compensation) and 240 (reordering restoration) can be interchanged in some implementations.
[0056] In a more complex embodiment, the offset compensation and rearrangement restoration are performed through a joint iterative process. This includes the following steps: tentatively rearranging the received constellation point sequence using candidate sorts of the random sequence; evaluating and updating the estimated offset compensation amount based on the constellation point clustering characteristics of the rearranged sequence (e.g., the degree to which constellation points cluster around a standard constellation diagram); and iterating the above process until the offset compensation amount that optimizes the clustering of constellation points in the rearranged sequence is found. This method, through joint optimization, may find better compensation and restoration results than individual sequential processing.
[0057] Step 250: Demap the original constellation point sequence into a bit stream, and input it into the AI decompression model to recover the original business data.
[0058] about Figure 8 Detailed description of the receiving device: Figure 8 A specific embodiment of the receiving device implementing claim 10 is shown. The device specifically includes the following functional modules, which work together to complete the method flow described in embodiment 2: Demodulation module 801: Its function corresponds to step 210, and it is used to convert the received time-domain waveform signal into a constellation point sequence.
[0059] Offset estimation module 802: Its function corresponds to step 220, and it is used to determine the constellation point offset compensation amount based on the amplitude information of the preset position peak in the time-domain waveform signal. This module can be configured to perform the average calculation method or the precise estimation method based on the reference signal.
[0060] Offset compensation module 803: Its function corresponds to step 230, and it is used to perform offset compensation on the constellation point sequence according to the offset compensation amount.
[0061] Constellation point rearrangement module 804: Its function corresponds to step 240. It is used to rearrange the compensated constellation point sequence according to the random sequence used by the transmitter to restore the original constellation point sequence. This module corresponds to the constellation point rearrangement module of the transmitter and is the key to restoring the data order.
[0062] Demapping module 805: Its function corresponds to step 250, and it is used to demap the original constellation point sequence into a bit stream and input it into the AI decompression model.
[0063] Figure 9 The diagram shows the receiving constellation diagram, where (a) represents the direct demodulation constellation diagram and (b) represents the shifted constellation diagram. This diagram visually illustrates the signal distortion introduced by the transmitting end's clipping operation during the receiving end's processing and the effectiveness of the compensation method proposed in this application. Figure 9The direct demodulation constellation diagram shown in (a) is the result of directly demodulating and FFT-transforming the received time-domain waveform signal that has been clipped at the transmitting end. As shown, since the clipping operation essentially performs nonlinear clipping on the signal, all the demodulated constellation points undergo a common, linear shift (e.g., a general translation towards the origin or in a certain direction), resulting in overall deformation of the constellation diagram and severely affecting the accuracy of the demapping. This is precisely the manifestation of the "signal distortion" problem pointed out in the background section on the constellation diagram. Figure 9 (b) illustrates the result after "determining the constellation point offset compensation amount based on the amplitude information of the preset position peak in the time-domain waveform signal" and "performing offset compensation on the received constellation point sequence according to the offset compensation amount" according to the method described in claim 6 of this application. After compensation, the constellation points are correctly translated back to their original theoretical positions, restoring a clear constellation diagram structure and laying the foundation for subsequent accurate demapping and AI decoding.
[0064] Example 3: System and Effects This embodiment describes a communication system comprised of the aforementioned transmitting and receiving devices and its gain effect. The transmitting device is used to generate and transmit an OFDM time-domain waveform signal that has undergone amplitude limiting and rearrangement processing; the receiving device is used to receive the waveform signal and perform offset compensation and rearrangement to recover the original data.
[0065] like Figure 10 As shown, the signal reception performance under different schemes is illustrated, with the transmitting power remaining consistent across all three schemes. (a) represents the waveform and receiving constellation diagram of the traditional scheme, (b) represents the waveform and receiving constellation diagram with the peak value directly truncated, and (c) represents the waveform and receiving constellation diagram of this scheme. The diagram demonstrates that this scheme effectively improves signal reception quality. This scheme is designed for AI compressed coding service data bitstream scenarios, providing an improved method for time-domain waveform signal generation and reception at the radio frequency end, ensuring signal transmission and reception quality.
[0066] It should also be noted that 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 process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0067] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when an element is “connected” or “coupled” to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein may include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.
[0068] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical, technical, and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0069] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.
Claims
1. A time-domain signal generation method for AI compression coding, applied at the transmitting end, characterized in that, include: Constellation point modulation is performed on the bitstream output by the compressed encoding of the AI model to obtain the initial constellation point sequence; The initial constellation point sequence is grouped, and the constellation points in each group are rearranged according to a specific random sequence so that the high amplitude peaks in the generated time-domain waveform signal appear at preset positions. Perform IFFT transform on the rearranged constellation point sequence to generate OFDM time-domain waveform signal; The peak value of the OFDM time-domain waveform signal at the preset position is subjected to amplitude limiting processing; The multiple OFDM time-domain waveform signals after being clipped are combined into a complete transmit waveform.
2. The method according to claim 1, characterized in that, Grouping the initial constellation point sequence includes: dividing the initial constellation point sequence into multiple subsequences according to the transmission resource parameters of the communication system, with each subsequence corresponding to a set of constellation points to be modulated onto its subcarriers by an OFDM symbol.
3. The method according to claim 1, characterized in that, The random sequence is reused in different groups.
4. The method according to claim 1, characterized in that, The limiting process includes: For each OFDM time-domain waveform symbol, a local limiting threshold is dynamically determined based on the peak amplitude of the symbol at the preset position, and the symbol is limited.
5. A transmitting device for AI compression encoding, used to implement the method according to any one of claims 1 to 4, characterized in that, include: The modulation module is used to perform constellation point modulation on the bitstream output by AI compression encoding to generate an initial constellation point sequence; The constellation point rearrangement module is used to group the initial constellation point sequence and rearrange the constellation points in each group based on a random sequence so that the high amplitude peaks in the subsequently generated time-domain waveform signal appear at preset positions. The IFFT conversion module is used to convert the rearranged constellation point sequence into an OFDM time-domain waveform signal; A limiting module is used to limit the peak value of the OFDM time-domain waveform signal at the preset position; The combination module is used to combine multiple limited waveform signals into a complete transmit waveform.
6. A time-domain signal receiving method for AI compression coding, characterized in that, Applied to the receiving end, including: The received time-domain waveform signal is subjected to FFT transformation to obtain the received constellation point sequence; Based on the amplitude information of the preset position peak in the time-domain waveform signal, the constellation point offset compensation amount is determined; The received constellation point sequence is offset compensated according to the offset compensation amount; Using a specific random sequence, the offset-compensated constellation point sequence is rearranged to restore the original constellation point sequence; The original constellation point sequence is demapped into a bitstream and input into an AI decompression model to recover the original business data.
7. The method according to claim 6, characterized in that, The offset compensation amount is calculated by averaging the difference between the peak amplitude of multiple OFDM symbols at the preset position and a preset threshold.
8. The method according to claim 6, characterized in that, From the received time-domain waveform signal, at least one OFDM symbol is extracted as a reference signal; based on the peak amplitude of the reference signal at the fixed position and a preset reference threshold, the offset compensation amount is determined.
9. The method according to claim 6, characterized in that, The offset compensation and rearrangement restoration are completed through joint iterative processing, including the following steps: The candidate sorting of the random sequence is used to perform a tentative rearrangement of the received constellation point sequence; Based on the constellation point clustering characteristics of the rearranged sequence, the estimated value of the offset compensation is evaluated and updated. Iterate through the above process until the offset compensation amount that makes the clustering of constellation points in the rearranged sequence optimal is found.
10. A receiving device for AI compression encoding, used to implement the method according to any one of claims 6 to 9, characterized in that, include: The demodulation module is used to convert the received time-domain waveform signal into a constellation point sequence; The offset estimation module is used to determine the constellation point offset compensation amount based on the amplitude information of the preset position peak in the time-domain waveform signal. An offset compensation module is used to perform offset compensation on the constellation point sequence according to the offset compensation amount. The constellation point rearrangement module is used to rearrange the compensated constellation point sequence according to the random sequence used by the transmitter, and restore the original constellation point sequence. The demapping module is used to demap the original constellation point sequence into a bitstream and input it into the AI decompression model.
11. A communication system for AI compression coding, characterized in that, include: At least one transmitting device as described in claim 5 and at least one receiving device as described in claim 10.