WO2024032775A1 - 量化方法和装置 - Google Patents
量化方法和装置 Download PDFInfo
- Publication number
- WO2024032775A1 WO2024032775A1 PCT/CN2023/112591 CN2023112591W WO2024032775A1 WO 2024032775 A1 WO2024032775 A1 WO 2024032775A1 CN 2023112591 W CN2023112591 W CN 2023112591W WO 2024032775 A1 WO2024032775 A1 WO 2024032775A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- quantization
- dictionary
- information
- sub
- dictionaries
- Prior art date
Links
- 238000013139 quantization Methods 0.000 title claims abstract description 1025
- 238000000034 method Methods 0.000 title claims abstract description 211
- 238000004891 communication Methods 0.000 claims description 84
- 238000011002 quantification Methods 0.000 claims description 53
- 230000015654 memory Effects 0.000 claims description 40
- 230000008569 process Effects 0.000 claims description 31
- 238000004590 computer program Methods 0.000 claims description 22
- 238000012549 training Methods 0.000 claims description 21
- 230000002950 deficient Effects 0.000 claims description 16
- 230000007547 defect Effects 0.000 claims description 7
- 230000006870 function Effects 0.000 description 39
- 238000012545 processing Methods 0.000 description 38
- 239000011159 matrix material Substances 0.000 description 17
- 238000013473 artificial intelligence Methods 0.000 description 16
- 230000005540 biological transmission Effects 0.000 description 14
- 238000013461 design Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 230000004913 activation Effects 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 7
- 230000001360 synchronised effect Effects 0.000 description 7
- 230000009471 action Effects 0.000 description 6
- 230000006835 compression Effects 0.000 description 6
- 238000007906 compression Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000007774 longterm Effects 0.000 description 4
- 238000010295 mobile communication Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000011084 recovery Methods 0.000 description 3
- 230000011664 signaling Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000008713 feedback mechanism Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000003491 array Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000001356 surgical procedure Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
- H04B7/0613—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission
- H04B7/0615—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal
- H04B7/0619—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal using feedback from receiving side
- H04B7/0621—Feedback content
- H04B7/0626—Channel coefficients, e.g. channel state information [CSI]
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/0413—MIMO systems
- H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B7/00—Radio transmission systems, i.e. using radiation field
- H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
- H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
Definitions
- the embodiments of the present application relate to the field of wireless communication technology, and more specifically, to a quantification method and device.
- UE user equipment
- CSI channel state information
- the UE measures the downlink reference signal sent by the base station, and based on the received The downlink reference signal estimates the downlink channel matrix, and finally generates CSI and feeds it back to the base station.
- the scale of antenna arrays in communication systems increases, the number of supported antenna ports increases.
- the UE feeds back the complete channel matrix to the base station through CSI, which means huge feedback overhead.
- the machine learning method has stronger nonlinear feature extraction capabilities and can extract channel matrix features more effectively.
- it can contain more channel information in feedback of the same scale, thereby reducing CSI compresses the information loss of feedback and improves the accuracy of base station channel recovery.
- less feedback amount can be used to represent the same channel information, which can further reduce feedback overhead.
- a deep learning-based encoder is used on the UE side to compress the channel information.
- the compressed information is quantified through a quantization dictionary and fed back to the base station.
- the base station stores the same quantization dictionary as the UE. According to the quantization dictionary, the base station can restore the received index to a quantized representation, and restore the channel information by inputting the quantized representation into a decoding process based on deep learning.
- the quantization dictionary is usually large, resulting in huge storage overhead on the base station side and the UE side.
- This application provides a quantization method and device in order to reduce the storage overhead of the quantization dictionary.
- a quantification method is provided, which is applied to a communication device (for example, the first device in this application).
- the method includes:
- the first device obtains the first quantization dictionary
- the first device uses a first quantization dictionary to quantize the information to be quantified, where the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries are used to quantize R quantized parts of the information to be quantified, and both M and R are greater than 1. a positive integer; or,
- the first device uses a first quantization dictionary to dequantize the information obtained by quantization of the received R quantized parts, wherein the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries are used to dequantize the received R quantized parts.
- the quantized information that is, the information obtained by quantization, is dequantized, and M and R are both positive integers greater than 1;
- the first quantification dictionary includes the dictionary usage method and the dictionary content of each of the M sub-dictionaries.
- the dictionary usage method indicates the correspondence between the M sub-dictionaries and the R quantization parts.
- the first sub-dictionary in the M sub-dictionaries and the first sub-dictionary in the R quantization parts Corresponding to a quantization part, the first sub-dictionary is used to quantize the first quantization part, or the first sub-dictionary is used to de-quantize the information obtained by quantization in the first quantization part; each sub-dictionary in the M sub-dictionaries
- the dictionary content includes at least one dictionary element, and each dictionary element included in the first sub-dictionary among the M sub-dictionaries is a candidate for the quantization information of the first quantization part.
- a quantified dictionary representation format is proposed. Based on this representation format, a quantified dictionary can be represented by the dictionary usage method and dictionary content. This dictionary representation format actually establishes a corresponding relationship between the R quantized parts of the information to be quantified and the M quantized sub-dictionaries (referred to as sub-dictionaries for short).
- the first sub-dictionary in the M sub-dictionaries corresponds to the first quantization part among the R quantization parts
- the first sub-dictionary is used to quantize the first quantization part, or the first sub-dictionary used for
- the quantized information obtained by the first quantization part is dequantized, and each dictionary element in the dictionary content of the first sub-dictionary is a candidate for the first quantization part.
- the cost of storing the quantization dictionary on the device can be reduced.
- the first device uses the first quantization dictionary to quantize the information to be quantified, and after obtaining the quantized information, the quantized information can be sent to the second device.
- the quantized information of the R quantized parts received by the first device comes from the second device.
- the base station side and the UE side each store an identical quantization dictionary.
- This quantization dictionary is used for quantization (or dequantization) in various scenarios, or in other words, This mechanism does not differentiate between different scenarios. For example, high-speed scenes, low-speed scenes, indoor scenes or outdoor scenes, etc., all use the same quantification dictionary. Considering that in different scenarios, the value distribution of the information to be quantified may be completely different or have their own characteristics. If the same quantization dictionary is used uniformly without any distinction between scenarios, the quantification performance will be poor.
- this application proposes another quantization method, which aims to use different quantization dictionaries in different scenarios, so that the quantization dictionary can adapt to the value distribution of the information to be quantified in the current scenario, so as to improve the quantization performance.
- a quantification method is provided, applied to a communication device (for example, the first device in this application), and the method includes:
- the first device obtains the first quantization dictionary
- the first device uses the first quantization dictionary to quantize the first information to be quantified, or,
- the first device dequantizes the received first quantized information using the first quantization dictionary
- the first device obtains the second quantization dictionary
- the first device uses the second quantization dictionary to quantize the second information to be quantified, or,
- the first device dequantizes the received second quantized information using the second quantization dictionary
- the first quantization dictionary includes a plurality of dictionary elements, and each dictionary element included in the first quantization dictionary is a candidate for the information to be quantized generated in the first quantization scenario;
- the second quantization dictionary includes a plurality of dictionary elements, and each dictionary element included in the second quantization dictionary is a candidate for the information to be quantized generated in the second quantization scenario;
- the first quantization dictionary and the second quantization dictionary are different.
- the first quantization dictionary corresponds to a first quantization scenario
- the second quantization dictionary corresponds to a second quantization scenario.
- the value distribution of the first information to be quantized is different from the value distribution of the second information to be quantized.
- the value distribution of the information to be quantified generated in the first quantization scenario is the same as or close to the value distribution of the information to be quantified generated in the second quantization scenario, for example, does not exceed a certain determination threshold, then the If the first quantization scene and the second quantization scene actually belong to the same quantization scene, then the first quantization scene changes to the second quantization scene, or the second quantization scene changes to the first quantization scene, and there is no need to update the quantization dictionary.
- the first quantization is considered
- the scene and the second quantization scene are different quantization scenes. If the first quantization scene changes to the second quantization scene, or the second quantization scene changes to the first quantization scene, you need to select a quantization dictionary that is suitable for the current quantization scene.
- a change in location distribution means a change in the positioning scenario.
- a change in channel conditions cause changes in the distribution of CSI values, which can be regarded as changes in the CSI feedback scenario.
- the value distribution dimensions of the information to be quantified may be different.
- it may be two-dimensional, three-dimensional or more, which is not limited.
- the positioning information can be three-dimensional spatial coordinates.
- the first device obtains the quantization dictionary, including: the first device obtains the quantization dictionary from the quantization dictionary table.
- the quantization dictionary table contains at least one candidate quantization dictionary.
- the first quantization dictionary and/or the second quantization dictionary are obtained (or selected) from the quantization dictionary table.
- the first device may select an appropriate quantization dictionary from the quantization dictionary table according to different quantization scenarios.
- the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries are used to separately quantize R quantized parts of the information to be quantified or to quantize R Part of the information obtained through quantization is dequantized, and M and R are both positive integers;
- the second quantization dictionary includes C sub-dictionaries.
- the C sub-dictionaries are used to separately quantize the S quantized parts of the information to be quantified or to de-quantize the information obtained by respectively quantizing the S quantized parts.
- Both S and C are positive. integer;
- the first quantization dictionary includes a dictionary usage method and the dictionary contents of each of the M sub-dictionaries.
- the dictionary usage method indicates the correspondence between the M sub-dictionaries and the R quantization parts.
- the first sub-dictionary in the M sub-dictionaries and the R quantization parts Corresponding to the first quantization part of
- the dictionary content of the dictionary includes at least one dictionary element, and each dictionary element included in the first sub-dictionary in the M sub-dictionaries is a candidate for the quantization information of the first quantization part;
- the second quantification dictionary includes a dictionary usage method and the dictionary contents of each of the C sub-dictionaries.
- the dictionary usage method indicates the correspondence between the C sub-dictionaries and the S quantization parts.
- the first sub-dictionary in the C sub-dictionary and the S quantization parts Corresponding to the first quantization part of
- the dictionary content of the dictionary includes at least one dictionary element, and each dictionary element included in the first sub-dictionary in the C sub-dictionaries is a candidate for the quantization information of the first quantization part.
- the following description of the first quantization dictionary may also apply to the second quantization dictionary.
- the following description will take the first quantization dictionary as an example.
- the dictionary contents of the quantification dictionary in the second aspect, as well as the specific implementation and process of obtaining the quantification dictionary by the first device can be referred to the specific implementation and process of obtaining the quantization dictionary by the first device in the first aspect, and will not be described again in this article.
- each dictionary element included in the first sub-dictionary corresponds to a quantitative information and an index
- Each dictionary element included in the first sub-dictionary is a candidate for the quantization information of the first quantization part, including:
- the quantization information corresponding to each dictionary element included in the first sub-dictionary is a candidate for the quantization information of the first quantization part.
- the dictionary content of each sub-dictionary in the quantization dictionary is composed of dictionary elements, and each dictionary element corresponds to a quantification information and an index.
- the index of a dictionary element can uniquely indicate the dictionary element in the sub-dictionary, and thus can also uniquely correspond to the quantified information corresponding to the dictionary element. Therefore, when transmitting the updated part of the quantization dictionary, if a sub-dictionary only involves the update of part of the dictionary elements, only the updated dictionary elements in the sub-dictionary need to be transmitted, without the need for the complete quantization dictionary, which can reduce the transmission quantization Dictionary overhead.
- the first device obtains the first quantization dictionary, including:
- the first device updates the second quantization dictionary to obtain the first quantization dictionary
- the method also includes:
- the first device sends all or part of the first quantization dictionary to the second device.
- the first device updates the quantization dictionary and provides the updated quantization dictionary to the second device for use.
- the first device is a terminal device
- the second device is an access network device. Since the function of the quantization dictionary is to quantize the output of the encoder on the terminal side, the terminal device can obtain a quantization dictionary with higher quantization accuracy based on the output distribution of the encoder.
- the M sub-dictionaries include P sub-dictionaries and Q sub-dictionaries, where,
- P sub-dictionaries are selected by the first device from the quantization dictionary table stored in the first device, and the quantization dictionary table includes at least P candidate sub-dictionaries;
- the Q sub-dictionary is obtained by the first device through dictionary training
- the first quantization dictionary updated by the first device includes M sub-dictionaries.
- the M sub-dictionaries can be selected from the quantization dictionary table, or obtained by dictionary training, or a combination of the two methods.
- the first device updates the quantization dictionary according to changes in application scenarios (for example, from low-speed scenarios to high-speed scenarios), and can always obtain better quantization performance in various application scenarios.
- the first device sends a part of the first quantization dictionary to the second device, including:
- the first device sends a first message to the second device, the first message indicating an updated portion of the first quantization dictionary relative to the second quantization dictionary.
- the first message "instructs" the update part, which can also be expressed as the first message "carries" the update part.
- This does not limit the specific indication method of the update part, that is, it can be implemented by including the update part, or , which can be achieved through other information corresponding to the updated part.
- the corresponding relationship between the update part and the other information may be predefined by the protocol, for example, pre-stored in the first device and the second device, or pre-configured to the first device and/or the second device.
- the configuration may be implemented by the first device and/or the second device, or by another device different from the first device and the second device.
- the first quantization dictionary corresponds to a first quantization scenario
- the second quantization dictionary corresponds to a second quantization scenario.
- the value distribution of the first information to be quantized is different from the value distribution of the second information to be quantized.
- the value distribution of the information to be quantified generated in the first quantization scenario is the same as or close to the value distribution of the information to be quantified generated in the second quantization scenario, for example, does not exceed a certain determination threshold, then the If the first quantization scene and the second quantization scene actually belong to the same quantization scene, then the first quantization scene changes to the second quantization scene, or the second quantization scene changes to the first quantization scene, and there is no need to update the quantization dictionary.
- the first quantization is considered
- the scene and the second quantization scene are different quantization scenes. If the first quantization scene changes to the second quantization scene, or the second quantization scene changes to the first quantization scene, you need to select a quantization dictionary that is suitable for the current quantization scene.
- a change in location distribution means a change in the positioning scenario.
- a change in channel conditions cause changes in the distribution of CSI values, which can be regarded as changes in the CSI feedback scenario.
- the value distribution dimensions of the information to be quantified may be different.
- it may be two-dimensional, three-dimensional or more, which is not limited.
- the positioning information can be three-dimensional spatial coordinates.
- the first device after the quantization dictionary is updated, the first device only needs to transmit the updated part of the quantization dictionary to the second device, instead of the complete updated quantization dictionary. That is, the updated quantization dictionary can be transmitted from the first device to the second device. Since the complete quantization dictionary is huge, transmitting only the updated part can reduce the overhead of transmitting the quantization dictionary.
- the updated part includes:
- the dictionary content of the sub-dictionary corresponding to the Z quantization parts among the R quantization parts, 1 ⁇ Z ⁇ R, Z is an integer.
- the update of the quantification dictionary may involve an update of the dictionary usage method, an update of the dictionary content, or an update of both the dictionary usage method and the dictionary content.
- the first device transmits the update part to the second device, it only needs to transmit part or all of the corresponding relationships involved in the update, and/or part or all of the dictionary content involved in the update, which can reduce the overhead of transmitting the quantized dictionary.
- update part includes the corresponding relationship between the L quantization parts and the sub-dictionary, which represents the dictionary usage update of the first quantization dictionary, and the update part includes the dictionary contents of the sub-dictionaries corresponding to the Z quantization parts, which represents the first quantization dictionary. Dictionary content updated.
- the partial correspondence involving updating means that only some of the correspondences between the R quantization parts and the sub-dictionaries are updated, rather than that all the correspondences between the R quantization parts and the sub-dictionaries are updated.
- the partial dictionary content involved in the update is also similar, which refers to the update of part of the dictionary content in the quantified dictionary, for example, the update of dictionary elements in some sub-dictionaries, rather than the update of the dictionary content of all sub-dictionaries.
- only some of the dictionary elements may be updated, rather than every dictionary element.
- this implementation can reduce transmission overhead except when all correspondences are updated and all dictionary elements in all sub-dictionaries are updated.
- the fewer parts of the quantization dictionary involved in updating the more obvious the transmission overhead is reduced compared to transmitting the complete quantization dictionary.
- the first message indicates the L quantization parts, and L The identifier of the sub-dictionary corresponding to each quantization part; and/or,
- the first message indicates the identification of the Z sub-dictionaries, and the Z sub-dictionaries each involve updated dictionary elements and indexes.
- the first message is not limited to one or more.
- the information included in the first message can be carried in the same message, or in more than one message.
- the first message includes multiple messages, for example, the first message A includes the L quantized parts, and the first message B carries the subdictionaries corresponding to the L quantized parts in a predetermined order. logo. or,
- multiple messages carry a quantized part and its corresponding identifier in a preset order.
- the preset order includes the order of the quantized part in the information.
- the first message a includes the first quantization part among the L quantization parts and the identifier of its corresponding sub-dictionary, and includes the second quantization part among the L quantization parts and the identifier of its corresponding sub-dictionary in the first message b, so that analogy.
- updated dictionary elements and indexes include one or more of the following:
- the sub-dictionary when the identifier of the sub-dictionary is a newly added identifier in the quantification dictionary, the sub-dictionary can be considered to be a newly-added sub-dictionary.
- the dictionary content of the sub-dictionary included in the update part may also include all dictionary elements in the sub-dictionary and their corresponding index of.
- the first device can reduce transmission overhead by indicating the updated portion of the quantization dictionary to the second device.
- the method before the first device obtains the first quantization dictionary, the method further includes:
- the first device sends first information to the second device, and the first information is used to instruct the first device to update the second quantization dictionary.
- the update of the quantization dictionary cannot be performed independently by the first device and the second device, but needs to be completed on one side and then transmitted to the other side. side. If the first information indicates that the first device updates the quantization dictionary, the second device will not modify the contents of the quantization dictionary during the process of updating its own encoder or decoder, which can ensure that the first device and the second device each save Quantization dictionaries are always consistent.
- the accuracy of the CSI information recovered by the network side can be ensured.
- the method further includes:
- the first device receives second information from the second device, and the second information is used to request enabling of the first quantization dictionary.
- the first device sends the first quantization dictionary to the second device.
- the second device accurately receives the first quantization dictionary and determines to use the first quantization dictionary
- requesting the first device to enable the first quantization dictionary can improve the reliability of synchronously updating the quantization dictionary between the two devices. For example, if the second device does not accurately receive the first quantization dictionary, or if the second device receives the first quantization dictionary and evaluates it, and the quantization performance does not meet the requirements, the second device may not accept the activation of the new quantization dictionary.
- the method further includes:
- the first device sends third information to the second device, and the third information is used to indicate enabling the first quantization dictionary.
- the first device sends third information to the second device based on the information requested by the second device to enable the new quantization dictionary, indicating to enable the new quantization dictionary to maintain quantization between the first device and the second device. Synchronization of dictionary updates.
- the first device uses the first quantization dictionary to quantize the information to be quantified, including:
- the first device Starting from the moment when the third information is sent, the first device enables the first quantization dictionary to quantize the information to be quantified after a time interval T, where the time interval T is based on the protocol agreement or indicated by the third information; or,
- the first device uses the first quantization dictionary to dequantize the information obtained by quantization of the received R quantized parts, including:
- the first device Starting from the moment when the third information is sent, after a time interval T, the first device enables the first quantization dictionary to dequantize the information obtained by quantization of the R quantized parts received, where the time interval T is based on the protocol agreement or by indicated by the third information.
- the time interval T may be determined by the second device and sent to the first device.
- the time interval T may be carried in the second information used to request activation of the first quantization dictionary, without limitation.
- the time interval T may also be determined through negotiation between the first device and the second device through other signaling or messages, and is not limited to being carried or indicated through the second information or the third information.
- the first device obtains the first quantization dictionary, including:
- the first device receives all or part of the first quantization dictionary from the second device.
- the quantization dictionary is updated by the second device, and is transmitted to the first device after the new quantization dictionary is updated.
- the first device receives the updated first quantization dictionary, specifically, the dictionary usage method and the dictionary content of the first quantization dictionary are each related to the updated part. Compared with transmitting the complete quantization Dictionary, which can reduce the overhead of transmitting quantized dictionaries.
- the first device receives a portion of the first quantization dictionary from the second device, including:
- the first device receives a second message from the second device.
- the second message indicates an updated portion of the first quantization dictionary relative to the second quantization dictionary.
- the first quantization dictionary is obtained by updating the second quantization dictionary by the second device.
- the updated part includes:
- the dictionary content of the sub-dictionary corresponding to the Z quantization parts among the R quantization parts, 1 ⁇ Z ⁇ R, Z is an integer.
- the second message indicates the L quantization parts and the identifiers of the sub-dictionaries corresponding to the L quantization parts;
- the second message indicates the identification of the Z sub-dictionaries, and the Z sub-dictionaries each involve updated dictionary elements and indexes.
- the method further includes:
- the first device determines that the obtained first quantization dictionary is defective, the first device sends fourth information to the second device.
- the fourth information is used to instruct the second device to send the defective part of the first quantization dictionary, and the defective part includes the first Defects in the way the dictionary is used and/or defects in the content of the quantified dictionary;
- the first device receives the defective portion from the second device.
- the device receiving the quantization dictionary from the other party needs to determine whether it has received the quantization dictionary completely. If the quantization dictionary is defective, it can request the other party to send the defective part, which can ensure the reliability of transmitting the quantization dictionary.
- the method before the first device obtains all or part of the first quantization dictionary, the method further includes:
- the first device receives fifth information from the second device, and the fifth information is used to instruct the second device to update the quantization dictionary.
- the method further includes:
- the first device sends sixth information to the second device, and the sixth information is used to request activation of the first quantization dictionary.
- the method further includes:
- the first device receives seventh information from the second device, and the seventh information is used to indicate enabling the first quantization dictionary.
- the first device uses the first quantization dictionary to quantize the information to be quantified, including:
- the first device Starting from the moment when the seventh information is received, the first device enables the first quantization dictionary to quantize the information to be quantified after a time interval T, where the time interval T is based on the protocol agreement or indicated by the seventh information;
- the first device uses the first quantization dictionary to dequantize the information obtained by quantization of the received R quantization parts, including:
- the first device starts from the moment when the seventh information is received, after a time interval T, the first device enables the first quantization dictionary to dequantize the information obtained by quantization of the R quantized parts received, where the time interval T is based on the protocol agreement or Indicated by the seventh message.
- the time interval T may be determined by the first device and sent to the second device.
- the time interval T may be carried in the sixth information for requesting activation of the first quantization dictionary, without limitation.
- the time interval T may also be negotiated and determined by the first device and the second device through other signaling or messages, and is not limited to being carried or indicated by the sixth information or the seventh information.
- the second device updates the quantization dictionary and transmits it to the first device.
- the technical effects of these implementation methods are the same as the technical effects of the corresponding implementation method in which the first device updates the quantization dictionary and transmits it to the second device. Those who are equivalent will not be described again.
- the first device is an access network device and the second device is a terminal device;
- the method further includes:
- the first device evaluates the quantization performance of the first quantization dictionary
- the first device updates the decoder according to the first quantization dictionary.
- the first device is an access network device and the second device is a terminal device;
- the method further includes:
- the first device receives the first quantified information from the second device
- the first device dequantizes the first quantized information according to the first quantized information and the second quantization dictionary.
- the access network device before enabling the updated quantization dictionary, uses the pre-updated quantization dictionary to dequantize the received quantized information (or, in other words, recover the channel information), which can ensure Always use the same quantization dictionary synchronously with the terminal device.
- the first device is an access network device and the second device is a terminal device;
- the method further includes:
- the first device receives second quantified information from the second device
- the first device dequantizes the second quantized information according to the second quantized information and the first quantization dictionary.
- the access network device uses the new quantization dictionary to dequantize the received quantized information, which can ensure that the same quantization dictionary is always used synchronously with the terminal device.
- the first device is a terminal device and the second device is an access network device;
- the method further includes:
- the first device evaluates the quantization performance of the first quantization dictionary
- the first device updates the encoder according to the first quantization dictionary.
- the quantization dictionary is updated by the base station and transmitted to the terminal device after the update.
- the terminal device evaluates the quantization performance of the quantization dictionary, and if the encoder does not match the new quantization dictionary, updates the encoder based on the new quantization dictionary, which can make the encoder more consistent with the new quantization dictionary, thereby improving quantization performance.
- the first device is a terminal device and the second device is an access network device;
- the method further includes:
- the first device determines the quantized information of the first channel information to be fed back according to the second quantization dictionary
- the first device sends the quantized information of the first channel information to the second device.
- the terminal device uses the pre-updated quantization dictionary to quantize the channel information before enabling the updated quantization dictionary, which can ensure that the same quantization dictionary is always used synchronously with the network device.
- the first device is a terminal device and the second device is an access network device;
- the method further includes:
- the first device determines the quantized information of the second channel information to be fed back according to the first quantization dictionary
- the first device sends the quantized information of the second channel information to the second device.
- the terminal device uses the updated quantization dictionary to quantize the channel information, which can ensure that the terminal device and the network device always use the same quantization dictionary synchronously.
- both the first channel information and the second channel information in this implementation are examples of information to be quantized.
- an example of the information to be quantified may be channel information, such as CSI.
- the channel information may be output information of an encoder on the UE side, such as CSI compression information.
- the first arrangement Prepare to use the first quantification dictionary to quantify the information to be quantified, including:
- the information to be quantized is the output information of the encoder, and the first device uses the first quantization dictionary to quantize the output information of the encoder;
- the first device uses the first quantization dictionary to dequantize the information obtained by quantization of the received R quantized parts.
- a third aspect provides a quantization or dequantization device, which has the function of implementing the method in the first aspect or the second aspect, or any possible implementation thereof.
- Functions can be implemented by hardware, or by hardware executing corresponding software.
- Hardware or software includes one or more units corresponding to the above functions.
- the device may be an access network device, a device, a module or a chip (or a chip system) in the access network device, or a device that can be used in conjunction with the access network device.
- the device may be a terminal device, a device, a module or a chip (or a chip system) in the terminal device, or a device that can be used in conjunction with the access network device.
- the device may include modules that perform one-to-one correspondence with the methods/operations/steps/actions described in the first aspect.
- the modules may be hardware circuits, software, or hardware circuits combined with software.
- the apparatus may include a processing module and a communication module.
- the processing module is used to implement the above-mentioned first aspect or the second aspect, or the processing/operations/steps/actions, etc. executed internally by the device in any possible implementation thereof;
- the communication module is used to execute the processes/operations/steps/actions etc. executed by the device.
- actions such as sending/receiving/transmitting, please refer to the description of the device embodiment for details.
- the communication module may also be called a transceiver module, including the functions of receiving and/or transmitting.
- a fourth aspect provides a quantization or dequantization device, including a processor and a memory.
- a transceiver may also be included.
- the memory is used to store computer programs
- the processor is used to call and run the computer programs stored in the memory, and control the transceiver to send and receive signals, so that the communication device performs the method of the first aspect or the method of the second aspect, or these aspects. method in any possible implementation.
- a quantization or dequantization device including a processor and a communication interface, the communication interface being used to receive data and/or information, and transmit the received data and/or information to the processor,
- the processor processes the data and/or information
- the communication interface is also used to output the data and/or information processed by the processor, so as to achieve the method of the first aspect or the method of the second aspect, or these Methods in any possible implementation of the aspect are executed.
- the communication interface may be a transceiver, a circuit, a bus, a module, a pin, or other types of communication interfaces.
- a sixth aspect provides an apparatus for quantization or dequantization, including at least one processor coupled to at least one memory, and the at least one processor is configured to execute a computer program stored in the at least one memory. or instructions to cause the communication device to perform the method of the first aspect or the method of the second aspect, or the method in any possible implementation of these aspects.
- the above device may be a chip or a chip system.
- a computer-readable storage medium is provided.
- Computer instructions are stored in the computer-readable storage medium.
- the computer instructions are run on a computer, the first aspect or the second aspect, or any of these aspects, is achieved.
- One possible implementation of the method is executed.
- An eighth aspect provides a computer program product, the computer program product comprising computer program code, when the computer program code is run on a computer, the first aspect or the second aspect, or any of these aspects, is achieved.
- the method in any possible implementation is executed.
- a ninth aspect provides a chip, which is used to read a computer program stored in a memory and execute the method provided by the above-mentioned first or second aspect, or any possible implementation of these aspects.
- the chip includes a circuit for performing the method provided by the above-mentioned first aspect or second aspect, or any possible implementation of these aspects.
- a chip system in a tenth aspect, includes a processor and is used to support a device to implement the method provided by the first aspect or the second aspect, or any possible implementation of these aspects.
- the chip system further includes a memory, and the memory is used to save necessary programs and data for the device.
- the chip system can be composed of chips or include chips and other discrete devices.
- a wireless communication system including the quantization or dequantization device as described in the third aspect, the fourth aspect, the fifth aspect or the sixth aspect.
- it may include a first device and a second device, or a first device and a third device, or a second device and a third device as in the embodiment of the present application.
- Figure 1 is a schematic diagram of a communication scenario suitable for embodiments of the present application.
- Figure 2 is an example of deep learning-based channel compression feedback suitable for embodiments of the present application.
- Figure 3 is a schematic diagram suitable for the quantification method provided in this application.
- Figure 4 is a schematic diagram of updating the quantification dictionary using clustering method.
- Figure 5 is an example of the quantification method provided by this application.
- Figure 6 is an example of the quantification method provided by this application.
- Figure 7 is a schematic block diagram of a device for quantization or dequantization provided in this application.
- Figure 8 is a schematic structural diagram of a device for quantization or dequantization provided in this application.
- the technical solutions of the embodiments of this application can be applied to various communication systems, including but not limited to: the fifth generation (the 5th generation, 5G) system, which may also be called the new radio (NR) system, the long-term evolution ( long term evolution (LTE) system, long term evolution-advanced (LTE-A) system, LTE frequency division duplex (FDD) system, LTE time division duplex (TDD) system etc. It can also be applied to future communication systems, such as the sixth generation mobile communication system. In addition, it can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, and machine-type communication (machine type).
- NR new radio
- LTE long-term evolution
- LTE-A long term evolution-advanced
- FDD frequency division duplex
- TDD time division duplex
- future communication systems such as the sixth generation mobile communication system.
- D2D device-to-
- MTC mobile-fidelity
- IoT Internet of things
- WiFi wireless-fidelity
- WIMAX worldwide interoperability for microwave access
- 3gpp third generation partner program
- One network element in the communication system can send signals to or receive signals from another network element.
- the signal may include information, signaling or data, etc.
- the network element can also be replaced by an entity, a network entity, a device, a communication device, a communication module, a node, a communication node, etc.
- This application uses equipment as an example for description.
- the communication system applicable to this application may include one or more first devices, and one or more second devices.
- one of the first device and the second device may be a terminal device, and the other may be a network device.
- the first device and the second device may both be network devices or both terminal devices.
- Figure 1 is a schematic diagram of a communication scenario suitable for embodiments of the present application.
- the embodiment of this application is mainly applicable to the scenario where the terminal device sends information to the network device.
- the terminal device sends channel state information (CSI) to the network device.
- CSI channel state information
- the technical solution of this application is introduced by taking the terminal device to send CSI to the network device as an example. Based on this example, those skilled in the art can apply this technical solution to other scenarios, for example, one device sends CSI to another device.
- the communication system may also include more terminal devices and more access networks.
- the equipment may also include other network elements, such as core network equipment, and/or network elements used to implement artificial intelligence functions.
- the terminal equipment may also be called user equipment (UE), access terminal, subscriber unit, user station, mobile station, mobile station, mobile terminal (mobile terminal, MT), remote station, remote terminal, Mobile device, user terminal, terminal, drone, wireless communication device, user agent or user device.
- UE user equipment
- access terminal subscriber unit
- user station mobile station
- mobile station mobile station
- mobile terminal mobile terminal
- remote station remote terminal
- Mobile device user terminal
- terminal, drone wireless communication device
- user agent or user device may be a device that provides voice and/or data connectivity to users, and may be used to connect people, things, and machines, such as handheld devices and vehicle-mounted devices with wireless connection functions.
- the terminal device in the embodiment of the present application can be a mobile phone (mobile phone), a tablet computer (pad), a notebook computer, a handheld computer, a mobile internet device (mobile internet device, MID), a wearable device, a virtual reality (virtual reality, VR) equipment, augmented reality (AR) equipment, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, smart Wireless terminals in smart grid, wireless terminals in transportation safety, wireless terminals in smart city, wireless terminals in smart home, etc.
- a mobile phone mobile phone
- tablet computer tablet computer
- MID mobile internet device
- AR augmented reality
- the device used to implement the function of the terminal may be a terminal, or may be a device that can support the terminal to implement the function.
- a device such as a chip system or chip, a hardware circuit, a software module, or a combination of a hardware circuit and a software module, can be installed in a terminal or used in conjunction with the terminal.
- the chip system may be composed of chips, or may include chips and other discrete devices.
- the network device may be a device with wireless transceiver functions.
- the network device may be a device that provides wireless communication function services. It is usually located on the network side, including but not limited to the fifth generation (5th generation, 5G) communication system.
- 5G fifth generation
- the next generation base station gNodeB,
- the network device may include a centralized unit (CU) node, a distributed unit (DU) node, or a RAN device including a CU node and a DU node, or a control plane.
- CU nodes and user plane CU nodes, as well as RAN equipment of DU nodes, or network equipment can also be wireless controllers, relay stations, vehicle-mounted equipment and wearable devices in the cloud radio access network (cloud radio access network, CRAN) scenario wait.
- the base station may be a macro base station, a micro base station, a relay node, a donor node, or a combination thereof.
- a base station may also refer to a communication module, modem or chip used in the aforementioned equipment or devices.
- the base station can also be a mobile switching center and equipment that performs base station functions in D2D, V2X, and M2M communications, network-side equipment in 6G networks, equipment that performs base station functions in future communication systems, etc.
- the base station can support networks with the same or different access technologies without limitation.
- the device used to implement the function of the network device may be a network device, or may be a device capable of supporting the network device to implement the function, such as a chip system or chip, a hardware circuit, a software module, or a hardware circuit and software. With the combination of modules, the device can be installed in network equipment or used in conjunction with network equipment.
- the chip system may be composed of chips, or may include chips and other discrete devices.
- the number of supported services continues to increase, which places higher requirements on communication systems in terms of system capacity, communication delay and other indicators.
- the communication rate can be effectively increased.
- multiple contiguous or non-contiguous frequency bands may be allocated to the UE for communication.
- the base station needs to obtain channel state information (CSI) on each frequency band.
- CSI channel state information
- FDD Frequency Division Duplex
- the uplink and downlink channels do not have mutuality, and the base station needs to obtain the downlink CSI through uplink feedback from the UE.
- the base station sends a downlink reference signal to the UE, and the UE receives the downlink reference signal. Since the UE knows the transmission information of the downlink reference signal, the UE can estimate (measure) the downlink channel experienced by the downlink reference signal based on the received downlink reference signal. The UE generates CSI based on the measured downlink channel matrix and feeds the CSI back to the base station. .
- PMI precoding matrix indicator
- codebook design a basic issue in mobile communication systems.
- the traditional codebook design method is to predefine (agree) a series of precoding matrices and corresponding numbers in the protocol. These precoding matrices are called codewords.
- the channel matrix or precoding matrix can be approximated using a predefined codeword or a linear combination of multiple predefined codewords.
- the UE can feed back one or more of the codeword corresponding numbers and weighting coefficients to the base station through the PMI, which is used for the base station side to restore the channel matrix or precoding matrix.
- the higher the accuracy of the CSI fed back by the UE the richer the information, the more accurate the channel information restored by the base station based on the CSI, and the more accurate the precoding matrix determined by the base station, which makes the downlink spatial division multiplexing performance better and the UE's receiving signal-to-interference-noise ratio
- the size of the complete channel matrix is proportional to the number of antenna ports, as the scale of the antenna array in the system increases, the number of supported antenna ports increases. In the massive MIMO system, the UE feeds back the complete channel matrix to the base station through CSI The feedback overhead is also huge.
- the machine learning method has stronger nonlinear feature extraction capabilities and can extract channel matrix features more effectively. Compared with traditional solutions, it can contain more channel information in feedback of the same scale, thus reducing the cost of CSI compression feedback. information loss and improve the accuracy of channel recovery at the base station.
- AI artificial intelligence
- the terminal equipment uses the AI model to compress and feedback the CSI
- the access network equipment uses the AI model to restore the compressed CSI.
- this scheme can represent the same channel information with less feedback amount, thereby further reducing feedback overhead.
- the AI model can include neural networks.
- Figure 2 is an example of deep learning-based channel compression feedback suitable for embodiments of the present application.
- an encoder based on deep learning is used on the UE side to compress the channel information.
- the compressed information is quantified through a quantization dictionary (the compressed information is quantified using the element closest to it in the quantization dictionary), and the quantization is represented in the dictionary.
- the index is fed back to the base station with limited bits.
- the base station side has the same quantization dictionary as the UE side, so the feedback index can be restored to a quantized representation of the compressed information, and then the quantized representation of the compressed information is input into the decoder based on deep learning to recover the channel information.
- the UE and the base station need to share the same quantization dictionary.
- the quantization dictionary is usually very large, the overhead of storing the quantization dictionary for the UE and the base station is also huge.
- the quantization dictionary of the UE and the quantization dictionary of the base station need to be updated synchronously, and the updated quantization dictionary usage time also needs to be synchronized.
- the encoding process on the UE side, the decoding process and the quantization process on the base station side are jointly trained as an AI model as a whole. After the training is completed, configure the trained AI model on the UE side and the base station side. If the AI model or quantification process needs to be updated later, the updated AI model also needs to be transmitted to the base station side or UE side, and each update needs to transmit the complete AI model, resulting in huge transmission overhead.
- this application provides a quantization method aimed at reducing the storage overhead of the quantization dictionary.
- quantization dictionary when the quantization dictionary is updated, there is no need to transmit the complete quantization dictionary, but only the updated part, which can reduce the transmission cost. Quantify dictionary overhead.
- the quantification method provided by this application can be applied to any scenario where information needs to be quantified. For example, one of the two devices needs to quantify the information and transmit it to another device, and the other device needs to quantify the received information.
- the quantified information can be dequantized.
- the method of quantizing information using the solution provided by this application and the method of dequantizing the received information can be used decoupled.
- the quantization process is not limited to be used in conjunction with the dequantization process.
- quantization and dequantization are often used together.
- the UE needs to feed back channel information, such as CSI information, to the base station.
- the UE needs to quantize the compressed information of the CSI information to be fed back, and feed back the quantization to the base station. information after.
- the base station receives the quantized information from the UE and needs to dequantize it to finally obtain the CSI information.
- one device needs to feed back location information to another device, and the location information may be quantified.
- the location information may be quantified.
- another device receives the quantized information, it needs to dequantize it to obtain position information. Therefore, this application does not limit specific application scenarios, and the above-mentioned quantification of channel information, such as CSI information or location information, is only an example.
- Figure 3 is a schematic diagram suitable for the quantification method provided in this application.
- the first device obtains the first quantization dictionary.
- the first quantization dictionary may be initially generated by the first device, or may be obtained by updating the second quantization dictionary by the first device, without limitation.
- a quantization dictionary is represented by the dictionary usage method of the quantization dictionary and the dictionary contents of each of the M sub-dictionaries contained in the quantization dictionary.
- a quantitative dictionary includes the dictionary usage method and dictionary content.
- the dictionary content includes the dictionary content of each of the M sub-dictionaries included in the quantification dictionary.
- the dictionary usage method of the first quantization dictionary indicates the corresponding relationship between the M sub-dictionaries and the R quantization parts.
- the M sub-dictionaries include the first sub-dictionary, and the first sub-dictionary corresponds to the first quantization part among the R quantization parts.
- the first sub-dictionary is used to quantize the first quantization part, and the first sub-dictionary is the M sub-dictionary. any subdictionary in .
- the first quantization dictionary is used to quantize the information to be quantified.
- the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries are used to quantize R quantization parts of the information to be quantized, and M and R are both positive integers.
- the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries correspond to R quantized parts of the information to be quantized.
- M and R are both positive integers greater than 1.
- dequantization is the inverse process of quantization. Therefore, the M sub-dictionaries included in the first quantization dictionary are also used to dequantize the information obtained by quantization of the R quantization parts. Both M and R are greater than 1. Positive integer.
- the information obtained through quantization that is, the quantized information, may be the index of the quantized information corresponding to the quantized part in the sub-dictionary. For the sake of simplicity in description, the information obtained through quantification can also be called "quantified index" in this article.
- M sub-dictionaries correspond to M quantization parts one-to-one, and each sub-dictionary in the M sub-dictionaries is used to quantize all M quantization parts.
- the corresponding quantization part is quantified, as shown in the example in Table 1.
- R is greater than M. At this time, some of the R quantized parts can correspond to the same sub-dictionary in the M sub-dictionaries, as shown in the example in Table 2.
- the dictionary usage method of the first quantization dictionary indicates the corresponding relationship between the M sub-dictionaries and the R quantization parts of the information to be quantized.
- Each sub-dictionary in the M sub-dictionaries is used to quantize the corresponding quantization part.
- the first sub-dictionary corresponds to the first quantization part among the R quantization parts, and the first sub-dictionary is used to quantize the first quantization part.
- the first quantization dictionary includes 3 sub-dictionaries, namely Y1, Y2 and Y3.
- the information to be quantified includes three quantization parts, namely quantization part 1, quantization part 2 and quantization part 3. Therefore, Y1 corresponds to the quantization part 1 among the three quantization parts.
- Y2 corresponds to quantization part 2 among the three quantization parts.
- Y3 corresponds to quantization part 3 among the three quantization parts. It can be seen that the dictionary usage method is used to indicate the quantization part corresponding to a sub-dictionary.
- the first quantization dictionary includes two sub-dictionaries, namely Y1 and Y2, and the information to be quantized includes three quantization parts.
- the sub-dictionary Y1 corresponds to the quantization part 1 and the quantization part 3 among the three quantization parts. Therefore, the sub-dictionary Y1 is used to quantize the quantization part 1 and the quantization part 3 of the information to be quantized.
- the dictionary content of each sub-dictionary in the M sub-dictionaries includes at least one dictionary element.
- at least one dictionary element included in the dictionary content of the first sub-dictionary is a candidate for the quantization information of the first quantization part.
- the dictionary elements included in the dictionary content of the first sub-dictionary are candidates for the quantization information of the quantization part corresponding to the first sub-dictionary.
- each dictionary element contained in the dictionary content of Y1 is a candidate for the quantization information of the quantization part 1
- each dictionary element contained in the dictionary content of Y2 is a candidate for the quantization information of the quantization part 2.
- each dictionary element in the dictionary content of Y1 is not only a candidate for the quantization information of quantization part 1, but also a candidate for the quantization information of quantization part 3. candidate.
- the first quantization dictionary includes M sub-dictionaries, and the dictionary content of any one of the M dictionaries contains at least one dictionary element. Furthermore, each dictionary element corresponds to a quantitative information and an index. Optionally, the "quantified information" corresponding to each dictionary element can also be expressed as a "quantified representation" instead. Therefore, it can also be said that if the first quantized part of the information to be quantized corresponds to the first sub-dictionary in the M sub-dictionaries, then the quantization corresponding to each dictionary element in the first sub-dictionary is the quantization of the first quantized part. Candidates represented, as shown in Table 3.
- the dictionary content of sub-dictionary Y1 includes 4 dictionary elements, respectively a1, a2, a3, a4; the dictionary content of sub-dictionary Y2 includes 8 dictionary elements, respectively b1, b2,..., b8; sub-dictionary
- the dictionary content of Y3 includes 2 dictionary elements, c1 and c2 respectively.
- dictionary elements a1, a2, a3, and a4 are all candidates for the quantization information of quantization part 1.
- a dictionary element can be selected from the dictionary elements a1, a2, a3, a4 in Y1 as the quantization information of the quantization part 1.
- a dictionary element closest to the Euclidean distance of quantization part 1 may be selected from Y1 as the quantization information of quantization part 1. Assume that if the Euclidean distance between dictionary element a3 and quantized part 1 is closest, then dictionary element a3 is the quantified information of quantized part 1, or in other words, dictionary element a3 is the quantified representation of quantized part 1.
- the information to be quantized is a real vector of length N. If the real vector representation is divided into multiple quantized parts that do not overlap or omit each other, the multiple quantized parts can be quantized separately.
- a quantization part can be expressed as [x1,x2), where x1 represents the index of the starting point of the quantization part, and x2 represents the index of the starting point of the next quantization part.
- x1 represents the quantization part.
- the index of the starting point of 1, x2 represents the index of the starting point of quantization part 2. Therefore, the dictionary usage method of the quantization dictionary can be as shown in Table 4.
- the quantized part [x1, x2) of the encoder's output information corresponds to the sub-dictionary Y1, and Y1 is used to quantize the quantized part [x1, x2).
- the quantized part [x2, x3) corresponds to the sub-dictionary Y2, then Y2 is used for the quantized part [x2, x3) and so on.
- the real number vector is divided according to Table 5
- the quantization part is quantized separately for each quantization part, that is, using the quantization dictionary Y1, the first quantization part (the 0th value to the 1st value, that is, [0.6, 0.2]) and the third quantization part (the 4th value to the fifth value, that is, [0.3, 0.8]); use the quantization dictionary Y2 to quantize the second quantization part (that is, the second value to the third value, that is, [-0.7,-0.1]) and the The four quantization parts (i.e. the 6th value to the 7th value, i.e. [-0.1,-0.2]) are quantized.
- sub-dictionary Y2 includes 4 dictionary elements.
- the corresponding indexes of these four dictionary elements in sub-dictionary Y2 are (0,1,2,3) in sequence.
- the real number vector can be obtained, that is, [0.6,0.2,-0.7,-0.1,0.3,0.8,-0.1, -0.2]
- the quantization information of the four quantization parts is [0.6,0.1], [-0.6,-0.1], [0.1,0.6], [-0.1,-0.1].
- the four quantization information corresponds to Y1 respectively. index 2 in Y2, index 1 in Y1, and index 0 in Y2.
- the quantization error is the quantized information [0.6,0.1], [-0.6,-0.1], [0.1,0.6], [-0.1,-0.1] and the information before quantization [0.6,0.2,-0.7, -0.1,0.3,0.8,-0.1,-0.2].
- the quantization error can be calculated according to the above steps, and the average value of multiple quantization errors can be used as the evaluation criterion for the quantization performance of the quantization dictionary.
- the average quantization error exceeds a threshold, it can be considered that the quantization dictionary cannot quantize the information to be quantified well.
- the average quantization error does not exceed (less than or equal to) the threshold, it can be considered that the quantization dictionary can quantize the information to be quantified well, and the quantization performance can meet the quantization requirements.
- the first quantization dictionary includes M sub-dictionaries, and each sub-dictionary in the M sub-dictionaries can be determined by two parameters (K, B), where K means that the sub-dictionary is used for a length of K, that is, it includes K
- K means that the sub-dictionary is used for a length of K, that is, it includes K
- B means that each input of the quantization part corresponding to the sub-dictionary is quantized using B bits. Different values of the B bits may correspond to different candidate quantization information. In other words, different values of the B bits are indices of different candidate quantization information. It can be understood that for a sub-dictionary with parameters (K, B), it can include up to 2 B dictionary elements.
- sub-dictionary Y1 can be determined by (K 1 , B 1 )
- sub-dictionary Y2 can be determined by (K 2 , B 2 ) Sure.
- the sub-dictionary can include up to 2 B dictionary elements (or simply elements), where each element Corresponding to a real number vector of length K and an index, the real number vector corresponding to each element is a candidate for the quantization information of the quantization part corresponding to the sub-dictionary.
- different elements in the sub-dictionary have different indexes. The index is used to uniquely identify the element in this subdictionary. For example, take the information to be quantized as the output information of the UE's encoder.
- a quantization part of length K of the encoder's output information is quantized using a sub-dictionary with parameters (K, B).
- the encoder In an output of the processor the value of the quantized part of length K is recorded as o1, then find an element from the sub-dictionary that is closest to o1 (can refer to the Euclidean distance) as an approximation of o1, then the element corresponding to The real number vector of length K is the quantization information of o1, and the index corresponding to this element can be used to uniquely identify the element in the sub-dictionary, so that it can also uniquely correspond to the quantization information of o1.
- the base station can find the corresponding real vector of length K from the same sub-dictionary as an approximation of o1.
- a sub-dictionary with parameters K, B
- K, B a sub-dictionary with parameters
- the input of the quantization function is a real vector of length K to be quantized
- the output of the quantization function is a binary bit string of B bits.
- the quantization function corresponds to an inverse quantization function.
- the input of the inverse quantization function is a binary bit string of B bits
- the output of the inverse quantization function is a real number vector of length K.
- inverse quantization and solution quantization can be interchanged.
- the first device obtains the first quantization dictionary, which may include multiple possible implementations.
- the first device updates the second quantization dictionary to obtain the first quantization dictionary.
- the second quantization dictionary may refer to the information used by the first device and the second device before the first device obtains the updated first quantization dictionary, and is used for the communication between the first device and the second device.
- the interactive information is quantized or dequantized using the quantization dictionary.
- the first device and the second device use the same quantization dictionary, if the quantization dictionary is updated and obtained by the first device, the first device needs to send the updated first quantization dictionary to the second device.
- the first device and the second device may each store multiple sub-dictionaries for quantization in different application scenarios (or quantization scenarios), or for quantizing different quantized parts of the information to be quantized.
- the first device may store a quantization dictionary table, and the quantization dictionary table includes one or more candidate sub-dictionaries.
- the first device updates the second quantization dictionary to obtain the first quantization dictionary.
- comparing the first quantization dictionary with the second quantization dictionary it may be an update of the dictionary usage method, or it may be an update of the dictionary content of a sub-dictionary corresponding to one or more quantized parts of the information to be quantified, Or updates in both of the above aspects are involved.
- updating the second quantization dictionary to the first quantization dictionary may involve a change in the number of quantization parts of the information to be quantized, for example, the number of quantization parts increases or the number of quantization parts decreases.
- the number of quantified parts does not change and is not limited.
- the update of the number of quantified parts is also included in the above-mentioned update of the dictionary usage method or the update of the dictionary content.
- the first device may select one or more candidate sub-dictionaries from the candidate sub-dictionaries in the quantization dictionary table as part of the M sub-dictionaries included in the updated first quantization sub-dictionary.
- the first device can also perform quantization dictionary training to obtain a new sub-dictionary.
- the sub-dictionary obtained by the first device through quantization dictionary training is not included in the quantization dictionary table stored by the first device by default. In other words, what the first device obtains through training is a new sub-dictionary. The first device can subsequently add the new sub-dictionary to the quantization dictionary table.
- the first device can add the new sub-dictionary to the quantization dictionary table. Assign a new identifier to distinguish it from the existing sub-dictionaries in the quantization dictionary table.
- the quantization dictionary table stores sub-dictionaries Y1, Y2 and Y3.
- the first device assigns the identifier Y4 to the new sub-dictionary obtained through training. , and add Y4 to the quantization dictionary table.
- the quantization dictionary table contains sub-dictionaries Y1, Y2, Y3 and Y4.
- the first device may also assign an existing identifier to the new sub-dictionary, or in other words, may replace an existing sub-dictionary with a new sub-dictionary.
- the first device assigns the identifier Y3 to the new sub-dictionary obtained by training. Therefore, the sub-dictionaries saved in the quantization dictionary table are the existing sub-dictionaries Y1, Y2 and the new sub-dictionary obtained by training (identified as Y3). .
- the first quantization dictionary obtained by the first device includes M sub-dictionaries, and the M sub-dictionaries include P sub-dictionaries and Q sum sub-dictionaries.
- the transmission of a complete quantization dictionary includes the transmission of the dictionary usage method and the transmission of the dictionary content. Since the dictionary content is usually large and the required transmission overhead is also large, the number of transmissions of the complete dictionary should be minimized after the dictionary is updated.
- the update of the quantization dictionary can be divided into the following scenarios.
- different methods of transmitting the updated quantization dictionary can be used to reduce transmission overhead.
- the first device and the second device can deploy multiple sub-dictionaries. These multiple sub-dictionaries can be used to adapt to the quantification requirements of different scenarios, or to adapt to the quantification requirements of different quantification parts of the information to be quantified.
- the first device can select a sub-dictionary that is more suitable for the current scene from multiple existing sub-dictionaries for quantization or de-quantization processing.
- the first device and the second device deploy multiple sub-dictionaries to cope with the quantitative requirements of low-speed scenarios and high-speed scenarios respectively.
- the user speed changes it is only necessary to update the dictionary usage of the quantization dictionary. In this scenario, the first device does not need to transmit the dictionary content of the updated quantization dictionary, but only needs to transmit the updated part of the dictionary usage method of the updated quantization dictionary.
- the first device and the second device are deployed with sub-dictionaries Y1, Y2, Y3 and Y4.
- the second quantized word before updating The dictionary is shown in Table 6, and the updated first quantization dictionary is shown in Table 7.
- the update of the quantification dictionary may only involve the update of the way the dictionary is used, or in other words, the correspondence between the quantified part of the information to be quantified and the sub-dictionary is updated.
- the correspondence relationship indicated by the dictionary usage method of the second quantification dictionary may be partially changed or may be completely changed, without limitation.
- Table 7 takes the change in the correspondence part as an example. For example, the correspondence between quantization part 1 and sub-dictionary Y1 has not changed, only the correspondence between quantization part 2 and quantification part 3 and the sub-dictionary has changed.
- the first device may indicate to the second device the updated portion of the quantization dictionary.
- the first device sends a first message to the second device, the first message indicating an updated portion of the first quantization dictionary relative to the second quantization dictionary.
- the first message indicates the corresponding relationship between the quantization part 2 and the sub-dictionary Y3, and the corresponding relationship between the quantization part 3 and the sub-dictionary Y4.
- the first message contains the identification and quantification part 2 of the sub-dictionary Y3, and the identification and quantization part 3 of the sub-dictionary Y4.
- the identification of Y3 corresponds to the quantization part 2
- the identification of Y4 corresponds to the quantization part 3.
- sub-dictionaries Y1, Y2, Y3, Y4 and Y5 are stored in the quantization dictionary tables of the first device and the second device. Assume that the second quantization dictionary before updating is shown in Table 6, and the first quantization dictionary after updating is shown in Table 8.
- the dictionary usage mode of the quantization dictionary is updated, specifically, the 3 quantization parts (quantization part 1, quantization part 2 and quantization part 3) and the 3 sub-dictionaries (Y1, Y2) indicated by the second quantization dictionary , Y3) is updated to the correspondence between the four quantization parts (quantization part 1 to quantization part 4) and the four sub-dictionaries (Y1, Y2, Y3, Y4).
- the dictionary usage method is updated, the information to be quantified is re-divided into multiple quantified parts, and the number of quantified parts increases.
- the first message may include the respective identifiers of Y1, Y3, Y4 and Y5, and the corresponding quantized parts of Y1, Y3, Y4 and Y5.
- the first device needs to retrain the quantization dictionary. . It should be understood that the existing sub-dictionary cannot adapt to the current application scenario, which means that the sub-dictionary used before the update and the sub-dictionary in the quantization dictionary table cannot be adapted to the current application scenario, and the first device needs to obtain new quantization through re-training. dictionary.
- one way of training or updating the quantization dictionary is to update the quantization dictionary using a clustering method.
- Figure 4 is a schematic diagram of updating the quantization dictionary using the clustering method.
- the value distribution of the quantized part [0,2) is as shown in the left picture of Figure 4.
- only one quantized part of the information to be quantified is used as an example for explanation.
- the update of the sub-dictionaries corresponding to the remaining three quantized parts is also similar and will not be described again.
- the dictionary elements can be updated according to the distribution of the information to be quantified.
- updating dictionary elements according to the distribution of information to be quantified is only an implementation method of updating dictionary content. If the value distribution of only some quantized parts of the information to be quantified changes, then only the dictionary content of the sub-dictionary corresponding to the quantized part whose value distribution changes can be updated.
- the quantization dictionary can be retrained.
- updating the dictionary usage method may include re-dividing the quantization part, and updating the sub-dictionary identification corresponding to each quantization dictionary.
- the first device selects a sub-dictionary from the quantization dictionary table, and/or the first device re-trains the dictionary to obtain a new sub-dictionary, so as to obtain the M sub-dictionaries contained in the updated first quantization dictionary. Dictionary to update the quantification dictionary.
- the first device sends first information to the second device before updating, and the first information indicates that the first device updates the quantization dictionary. After updating and obtaining the first quantization dictionary, the first device sends the first quantization dictionary to the second device.
- the first device and the second device may agree that the update of the quantization dictionary is performed by the first device or the second device.
- the first device before updating the quantization dictionary, the first device There is no need to exchange information about who will update the quantization dictionary with the second device. Instead, after the agreed party completes the update of the quantization dictionary, the updated quantization dictionary is directly sent to the other party.
- the first device sends the first quantization dictionary to the second device, the first device sends to the second device the values of the first quantization dictionary relative to the second quantization dictionary.
- Update section Specifically, these updated parts are sent in the form of dictionary usage or dictionary content.
- the first device sends a first message to the second device, the first message indicating the updated portion.
- the first message may be one or more, without limitation.
- the update part includes the correspondence between L quantization parts out of the R quantization parts and the sub-dictionary, 1 ⁇ L ⁇ R, L is an integer.
- the first device notifies the second device of the updated part, for example, the first device sends a first message to the second device.
- the first message indicates the L quantization parts and the correspondence between the L quantization parts and the sub-dictionary.
- the L quantized parts refer to the quantized parts whose corresponding relationship with the sub-dictionary changes.
- the update part includes the dictionary content of the sub-dictionary corresponding to Z quantization parts among the R quantization parts, 1 ⁇ Z ⁇ R, Z is an integer.
- the first device notifies the second device of the updated part, the first device sends a first message to the second device.
- the first message indicates the identification of the Z sub-dictionaries, and the Z sub-dictionaries each involve updated dictionary elements and indexes.
- the Z quantified parts refer to the quantified parts where the dictionary content of the corresponding sub-dictionary changes.
- the second device when the second device receives the first quantization dictionary from the first device, it may specifically receive the dictionary usage method and dictionary content of the first quantization dictionary, and the updated dictionary usage method and dictionary content are involved, so that in the second quantization dictionary On the basis of the dictionary, the second quantization dictionary is simultaneously updated to the first quantization dictionary.
- the second device sends second information to the first device, where the second information is used to request enabling the first quantization dictionary.
- the first device sends third information to the second device, where the third information is used to indicate enabling the first quantization dictionary.
- the third information may carry a time interval T. If the time error between the time when the first device sends the third information and the time when the second device receives the third information is ignored, the first device starts from the time when it sends the third information and activates the first quantization dictionary after the time interval T, Starting from the moment when the third information is received and after the time interval T, the second device starts to activate a new quantization dictionary, that is, the first quantization dictionary.
- the time interval T can also be specified by the protocol. Starting from the moment when the second device receives the third information, the default is the elapsed time interval T, and both parties enable a new quantization dictionary.
- the second device parses the received first quantization dictionary according to the representation format of the quantization dictionary, and determines whether the received first quantization dictionary is complete. , that is, the dictionary content of the first quantification dictionary and the way the dictionary is used are intact. If it is determined that the received first quantization dictionary has a defective part, the second device may send fourth information to the first device, and the fourth information is used to instruct the first device to resend the defective part. After receiving the missing portion resent by the first device, the second device can obtain the complete first quantization dictionary.
- the first device receives a first quantization dictionary from the second device.
- the first quantization dictionary may be initially generated by the second device, or obtained by updating the second quantization dictionary. After the second device obtains the first quantization dictionary, the second device sends the first quantization dictionary to the first device.
- the second device before updating the quantization dictionary, sends fifth information to the first device, and the fifth information indicates that the second device updates the quantization dictionary.
- the first device sends sixth information to the second device, and the sixth information indicates a request to enable the new quantization dictionary.
- the second device sends seventh information to the first device, and the seventh information indicates enabling a new quantization dictionary.
- the time error between the time when the second device sends the seventh information and the time when the first device receives the seventh information is taken into account, then the first device starts from the time when the seventh information is received by default.
- a new quantization dictionary is enabled; from the moment when the second device sends the seventh information, the default is the elapsed time interval T2, and a new quantization dictionary is enabled.
- the first device After different time intervals have elapsed with the second device, a new quantization dictionary is opened.
- the design of T1 and T2 can take the above time difference into consideration, so that the first device and the second device start the new quantization dictionary at the same time.
- the second device indicates the time interval T1 in the seventh information, so that the first device determines the time to enable the new quantization dictionary based on the time when the seventh information is received and the time interval T1.
- the first device after receiving the first quantization dictionary, the first device sends a third message to the first device for requesting to enable the first quantization dictionary. Before six pieces of information, the first device can evaluate the quantization performance of the first quantization dictionary.
- the first device inputs the channel information of a pre-saved test set to the encoder, uses the first quantization dictionary to quantize the output information of the encoder, and calculates the quantization error, thereby evaluating the quantization of the first quantization dictionary. performance. If the quantization performance meets the quantization requirements, for example, the quantization error is less than or equal to a threshold, it is considered that the quantization requirements are met, and the first device sends the above-mentioned sixth information to the second device to request activation of the first quantization dictionary.
- the first device can also configure the encoder (the first device is the UE and the second device is the access network device) or the decoder (the first device) according to the new quantization dictionary. is the access network device, and the second device is the UE) for updating.
- the first device as an access network device as an example, if the access network device determines that the decoder does not match the first quantization dictionary obtained from the UE, the access network device updates the decoder according to the first quantization dictionary, for example, updates Decoder parameters.
- the access network equipment inputs the quantified information candidates in the quantization dictionary into the decoder to obtain the output information, calculates the difference between the probability distribution of the output information and the probability distribution of the channel information in the training set, such as mutual information, and uses this as the loss function to update the decoding device.
- the first device as a UE as an example
- the UE updates the encoder according to the first quantization dictionary, for example, updates the parameters of the encoder .
- the UE uses a training set, inputs the channel data in the training set into the encoder to obtain the output, and uses a new dictionary to quantize the encoder output.
- the average quantization error on the training set can be calculated as a loss function.
- the loss function is the encoding is a function of the weight of this AI model. If the loss function is less than a threshold, stop training, otherwise update the encoder weights to reduce the loss function, such as calculating the gradient of the quantization error loss function relative to the encoder weights, and update the encoder through the gradient descent method.
- the encoder weight may include the weight of each neuron in the neural network when the encoder is implemented by a neural network.
- the first device uses the first quantization dictionary to quantize the information to be quantized, or the first device uses the first quantization dictionary to dequantize the information obtained by quantization of the received R quantized parts.
- the first device obtains the first quantization dictionary. Afterwards, the first device uses the first quantization dictionary to perform quantization, or uses the first quantization dictionary to perform dequantization.
- the information obtained by quantization of the R quantized parts received by the first device may be R quantization indexes.
- the first quantization dictionary is used to quantize the compressed information of the CSI. After that, the UE reports the CSI to the network side.
- Quantified information is the index of the quantized information corresponding to the compressed information in the corresponding sub-dictionary (for example, referred to as the quantization index for short).
- the network side receives the index of the quantization information reported by the terminal device (i.e., the quantization index), determines the information corresponding to the index according to the first quantization dictionary, and then recovers the quantization corresponding to the compressed information of the CSI information, and after decoding and compressing the information, the channel information is finally obtained. Therefore, in one example, if the first device is a UE, the UE uses the first quantization dictionary to quantize the compressed information of the channel information. If the first device is an access network device, the access network device uses the quantized information received by the first quantization dictionary, such as the index corresponding to the quantized information, to perform dequantization to recover the compressed information of the channel information.
- the first device continues to use the second quantization dictionary to quantize the information to be quantified, and the second device continues to use the second quantization dictionary to quantize the information to be quantized.
- the received quantized information such as the index corresponding to the quantized information, is dequantized.
- the access network device receives the first quantized information from the UE, and the access network device uses the second quantization dictionary to The information obtained once quantified is dequantized. After enabling the first quantization dictionary, if the access network device receives the second quantized information from the UE, the access network device uses the first quantization dictionary to dequantize the second quantized information.
- the UE uses the second quantization dictionary to quantize the first channel information and sends the first channel information. Quantified information. After enabling the first quantization dictionary, if the UE needs to feed back the second channel information to the access network device, the UE uses the first quantization dictionary to quantize the second channel information and sends the quantized information of the second channel information.
- the quantization dictionary initially used or previously used by the base station and the UE is the second quantization dictionary.
- Figure 5 is an example of the quantification method provided by this application.
- the base station sends the first information to the UE, and the first information instructs the base station to update the quantization dictionary.
- the base station updates the second quantization dictionary and obtains the first quantization dictionary.
- the base station sends the first message to the UE, where the first message is used to indicate the updated part of the quantization dictionary.
- the UE receives the updated part of the quantization dictionary.
- the UE parses the updated part of the quantization dictionary and obtains the first quantization dictionary.
- the UE determines that the received first quantization dictionary has a missing part, the UE sends fourth information to the base station, and the fourth information is used to request the base station to resend the first quantization dictionary. defective part.
- the base station resends the missing part to the UE.
- the UE After receiving the defective part, the UE obtains the complete first quantization dictionary.
- steps 506-507 are interrelated steps, and the method 500 includes steps 506-507 at the same time, or does not include steps 506-507.
- the UE evaluates the quantization performance of the first quantization dictionary.
- the UE After the UE completes the evaluation of the quantization performance of the first quantization dictionary, if it meets the quantization requirements, it may request to enable the first quantization dictionary, as shown in step 510.
- the UE may also test the matching degree between the first quantization dictionary and the encoder, and update the encoder according to the first quantization dictionary, as shown in step 509.
- the UE updates the encoder according to the first quantization dictionary.
- the UE sends second information to the base station, and the second information is used to request activation of the first quantization dictionary.
- the base station receives the second information from the UE.
- the base station sends third information to the UE, where the third information is used to indicate enabling the first quantization dictionary.
- the UE receives third information from the base station.
- the base station After enabling the first quantization dictionary, the base station uses the first quantization dictionary to dequantize the received quantized information, and the UE uses the first quantization dictionary to quantize the information to be quantized.
- the information that needs to be quantified on the UE side may be CSI, or other information that needs to be reported to the base station, without limitation.
- the process shown in Figure 5 is an example of updating the quantization dictionary by the base station.
- An example of the process of updating the quantization dictionary by the UE is given below, as shown in Figure 6.
- the UE sends the first information to the base station, and the first information instructs the UE to update the quantization dictionary.
- the UE updates the second quantization dictionary and obtains the first quantization dictionary.
- the UE sends a first message to the base station, where the first message is used to indicate the updated part of the first quantization dictionary relative to the second quantization dictionary.
- the base station receives the update part.
- the base station parses the updated part and obtains the first quantization dictionary.
- the base station may request the UE to resend the defective part. After receiving the defective part, the base station obtains the complete first quantization dictionary.
- the base station can evaluate the quantization performance of the first quantization dictionary, and when the evaluation is completed and meets the quantization requirements, request the UE to enable the first quantization dictionary, as shown in step 606.
- the base station may also determine the matching degree between the decoder and the first quantization dictionary. If the decoder does not match the first quantization dictionary, the base station updates the decoder according to the first quantization dictionary, for example, updates the parameters of the decoder.
- the base station sends second information to the UE, and the second information is used to request activation of the first quantization dictionary.
- the UE receives the second information. In response to the second information, the UE sends third information to the base station.
- the base station receives third information from the UE, and the third information is used to indicate enabling the first quantization dictionary.
- the base station uses the first quantization dictionary to dequantize the received quantized information
- the UE uses the first quantization dictionary to quantize the information to be quantized.
- the quantization device (such as the first device) performs an initial generation and/or update of the quantization dictionary using one of the dequantization devices (such as the second device).
- the initially generated quantization dictionary or the updated quantization dictionary can also be provided by a third device other than the quantization device or the dequantization device.
- the quantization device and the dequantization device both receive the updated quantization dictionary from the third device.
- the quantization device receives all or part of the updated quantization dictionary or the initially generated quantization dictionary from a third device, and converts all or part of the updated quantization dictionary or part or the initially generated quantization dictionary is sent to the dequantization device; or the dequantization device receives all or part of the updated quantization dictionary or the initially generated quantization dictionary from the third device, and sends the updated quantization dictionary All or part of the quantization dictionary or the initially generated quantization dictionary is sent to the quantization device.
- the third device may be an OTT server (over-the-top server).
- Figure 7 is a schematic block diagram of a device for quantization or dequantization provided by the present application.
- the device 1000 includes a processing unit 1100, a storage unit 1200 and a transceiver unit 1300.
- the information to be quantified is quantized using a first quantization dictionary, wherein the first quantization dictionary includes M sub-dictionaries, the M sub-dictionaries are used to quantize R quantized parts of the information to be quantified, M and R are all positive integers greater than 1; or,
- the first quantization dictionary is used to dequantize the information obtained by quantization of the R quantized parts received, and the dequantized information is used as the input of the decoder, wherein the first quantization dictionary includes M sub-dictionaries, and the M sub-dictionaries are used to dequantize the quantized information of the received R quantized parts.
- M and R are both positive integers greater than 1;
- the first quantization dictionary includes the dictionary usage mode and the dictionary contents of each of the M sub-dictionaries.
- the dictionary usage mode indicates the corresponding relationship between the M sub-dictionaries and the R quantization parts.
- the M sub-dictionaries The first sub-dictionary in corresponds to the first quantization part among the R quantization parts, and the first sub-dictionary is used to quantize the first quantization part, or the first sub-dictionary is used to quantize all
- the quantized information of the first quantized part is dequantized;
- the dictionary content of each sub-dictionary in the M sub-dictionaries includes at least one dictionary element, and each of the first sub-dictionaries in the M sub-dictionaries includes Dictionary elements are candidates for the quantization information of the first quantization part, and M is a positive integer;
- the first quantization dictionary is used to perform quantization or recovery processing of channel information.
- each dictionary element included in the first sub-dictionary corresponds to a quantification information and an index
- Each dictionary element included in the first sub-dictionary is a candidate for the quantization information of the first quantization part, including:
- the quantization information corresponding to each dictionary element included in the first sub-dictionary is a candidate for the quantization information of the first quantization part.
- the processing unit 1100 is configured to update the second quantization dictionary to obtain the first quantization dictionary
- the transceiver unit 1300 is also configured to send all or part of the first quantization dictionary to the second device.
- the M sub-dictionaries include P sub-dictionaries and Q sub-dictionaries, where,
- the P sub-dictionaries are selected by the first device from a quantization dictionary table stored in the first device, and the quantization dictionary table includes at least P candidate sub-dictionaries;
- the Q sub-dictionary is obtained by the first device through dictionary training
- the transceiver unit 1300 is used for:
- a first message is sent to the second device, the first message indicating an updated portion of the first quantization dictionary relative to the second quantization dictionary.
- the correspondence between L quantization parts and the sub-dictionary among the R quantization parts is 1 ⁇ L ⁇ R, L is an integer; and/or,
- the dictionary content of the sub-dictionary corresponding to the Z quantization parts among the R quantization parts is 1 ⁇ Z ⁇ R, and Z is an integer.
- the update part includes the correspondence between L quantization parts and the sub-dictionary among the R quantization parts
- the first message indicates the L quantization parts
- the identification of the sub-dictionary corresponding to each of the L quantization parts and/or,
- the update part includes the dictionary contents of the sub-dictionaries corresponding to Z quantization parts among the R quantization parts, then the first message indicates the identification of the Z sub-dictionaries, and the Z sub-dictionaries each relate to Updated dictionary elements and indexes.
- the transceiver unit 1300 is also used to:
- the transceiver unit 1300 is also used to:
- Second information is received from the second device, the second information being used to request enabling of the first quantization dictionary.
- the transceiver unit 1300 is also used to:
- the processing unit 1100 is used to:
- the first quantization dictionary is enabled to quantize the information to be quantified, wherein the time interval T is based on the protocol agreement or indicated by the third information. ;
- the first quantization dictionary is enabled to dequantize the information obtained by quantization of the R quantized parts received, where the time interval T is Based on the agreement or indicated by the third information.
- the transceiver unit 1300 is configured to receive the dictionary usage method and dictionary content of the first quantization dictionary from the second device;
- the processing unit 1100 is configured to receive all or part of the first quantization dictionary from the second device.
- the transceiver unit 1300 is used for:
- a second message is received from the second device, the second message indicating an updated portion of the first quantization dictionary with respect to a second quantization dictionary, the first quantization dictionary being an update of the first quantization dictionary by the second device. Obtained by updating the binary quantization dictionary.
- the update part includes:
- the corresponding relationship between the L quantization parts and the sub-dictionary among the R quantization parts is 1 ⁇ L ⁇ R, and L is an integer;
- the dictionary content of the sub-dictionary corresponding to the Z quantization parts among the R quantization parts is 1 ⁇ Z ⁇ R, and Z is an integer.
- the second message indicates the L quantization parts, and the sub-dictionaries corresponding to each of the L quantized parts. logo; and/or,
- the second message indicates the identification of the Z sub-dictionaries, and the Z sub-dictionaries each relate to Updated dictionary elements and indexes.
- the processing unit 1100 is configured to determine that the obtained first quantization dictionary is defective
- transceiver unit 1300 is used for:
- the missing portion is received from the second device.
- the transceiver unit 1300 is configured to receive fifth information from the second device, where the fifth information is used to indicate that the second device updates the quantization dictionary. .
- the transceiver unit 1300 is also used to:
- Sixth information is sent to the second device, where the sixth information is used to request enabling of the first quantization dictionary.
- the transceiver unit 1300 is also used to:
- the seventh information is used to indicate enabling the first quantization dictionary.
- the processing unit 1100 is used to:
- the first quantization dictionary is enabled to quantize the information to be quantified, where the time interval T is based on the protocol agreement or indicated by said seventh information;
- the first quantization dictionary is enabled to dequantize the information obtained by quantization of the received R quantization parts, wherein, The time interval T is based on a protocol agreement or indicated by the seventh information.
- the device is an access network device or a chip (or chip system) configured in the access network device
- the second device is a terminal device
- the information to be quantified is The output information of the encoder
- the first quantization dictionary is used by the device to dequantize the information obtained by quantization of the received R quantization parts to obtain the input of the decoder
- processing unit 1100 is used for:
- the decoder is updated according to the first quantization dictionary.
- the device is an access network device or a chip (or chip system) configured in the access network device;
- the transceiver unit 1300 is configured to receive the first quantized information from the second device;
- the processing unit 1100 is configured to dequantize the first quantized information according to the first quantized information and the second quantization dictionary.
- the device is an access network device or a chip (or chip system) configured in the access network device;
- the transceiver unit 1300 is configured to receive the second quantized information from the second device
- the processing unit 1100 is configured to dequantize the second quantized information according to the second quantized information and the first quantization dictionary.
- the device is a terminal device or a chip (or chip system) configured in the terminal device
- the second device is an access network device
- the information to be quantified is an encoder.
- the output information of the first quantization dictionary is used by the device to quantize the output information of the encoder;
- the processing unit 1100 is used for:
- the encoder is updated according to the first quantization dictionary.
- the device is a terminal device or a chip (or chip system) configured in the terminal device;
- the processing unit 1100 is configured to determine the quantized information of the first channel information to be fed back according to the second quantization dictionary
- the transceiving unit 1300 is configured to send the quantized information of the first channel information to the second device.
- the device is a terminal device or a chip (or chip system) configured in the terminal device;
- the processing unit 1100 is configured to determine the quantized information of the second channel information to be fed back according to the first quantization dictionary
- the transceiving unit 1300 is configured to send the quantized information of the second channel information to the second device.
- the information to be quantified is the output information of the encoder; the processing unit 1100, Specifically used to use the first quantization dictionary to quantize the output information of the encoder; or,
- the processing unit 1100 is specifically configured to use the first quantization dictionary to process the received R quantized parts.
- the information obtained by quantification is dequantized.
- the processing unit 1100 is configured to perform processing and/or operations implemented internally by the first device in addition to the actions of sending and receiving.
- the transceiver unit 1300 is configured to perform an action of receiving or sending by the first device.
- the storage unit 1200 is used to store quantization dictionary, quantization dictionary table and other information.
- the processing unit 1100 is configured to perform a process of updating the quantization dictionary to obtain a first quantization dictionary; or the transceiver unit 1300 is configured to perform an operation of receiving a first quantization dictionary from a second device to obtain a first quantization dictionary. dictionary.
- the processing unit 1100 is also used to perform step 320.
- the transceiver unit 1300 is used to perform step 501, step 503, steps 506-507, and steps 510-511 performed by the base station; the processing unit 1100 is used to perform step 502 and step 512 performed by the base station, etc. .
- the transceiver unit 1300 is used to perform step 601, step 603, and steps 606-607 performed by the UE; the processing unit 1100 is used to perform step 602 and step 608 performed by the UE.
- each functional unit in each example of this disclosure may Integrated into a processor, it can also exist physically alone, or two or more units can be integrated into one unit.
- the above integrated units can be implemented in the form of hardware or software function modules.
- this application also provides a device for quantization or dequantization, as shown in Figure 8.
- FIG. 8 is a schematic structural diagram of a device for quantization or dequantization provided by the present application.
- the device 10 includes: one or more processors 11 , one or more memories 12 and one or more communication interfaces 13 .
- the processor 11 is used to control the communication interface 13 to send and receive signals
- the memory 12 is used to store a computer program
- the processor 11 is used to call and run the computer program from the memory 12, so that the device 10 executes the method described in the embodiments of the present application. The processing performed by a device.
- the processor 11 may have the function of the processing unit 1100 shown in FIG. 7
- the memory 12 may have the function of the storage unit 1200
- the communication interface 13 may have the function of the transceiver unit 1300 .
- the processor 11 may be used to perform processing or operations performed internally by the device
- the memory may be used to perform operations of caching and deleting information units from the cache
- the communication interface 13 is used to perform operations of sending and/or receiving by the device. .
- the device 10 may be applied to access network equipment.
- the specific device 10 may be an access network device, or may be an access network capable of supporting the access network device to implement any of the above-mentioned examples.
- the function of the device may be applied to access network equipment.
- the device 10 may be the first device in the method embodiment.
- the communication interface 13 may be a transceiver of the first device.
- a transceiver may include a receiver and/or a transmitter.
- the processor 11 may be a baseband device of the first device, and the communication interface 13 may be a radio frequency device.
- the device 10 may be a chip (or chip system) installed in the first device.
- communication interface 13 may be an interface circuit or an input/output interface.
- the dotted box behind the device indicates that there can be more than one device.
- the memory and the processor in the above device embodiments may be physically independent units, or the memory may be integrated with the processor, which is not limited herein.
- this application also provides a computer-readable storage medium.
- Computer instructions are stored in the computer-readable storage medium. When the computer instructions are run on the computer, the steps executed by the first device in each method embodiment of the application are caused. Operations and/or processing are performed.
- this application also provides a computer program product.
- the computer program product includes computer program code or instructions. When the computer program code or instructions are run on the computer, the operations performed by the first device in each method embodiment of the application are performed. /or processing is performed.
- the present application also provides a chip, the chip includes a processor, and a memory for storing a computer program is independent of the chip. And it is configured that the processor is used to execute the computer program stored in the memory, so that the device installed with the chip performs the operations and/or processing performed by the first device in any one method embodiment.
- the chip may also include a communication interface.
- the communication interface may be an input/output interface, or an interface circuit, etc.
- the chip may further include the memory.
- this application also provides a communication device (for example, it can be a chip or a chip system), including a memory, a processor and a communication interface.
- the communication interface is used to receive the first information and send the first information to the processing processor; the processor is configured to process the first information unit cached in the memory according to the first information.
- the present application also provides a communication device, including at least one processor, the at least one processor is coupled to at least one memory, and the at least one processor is used to execute a computer program or instructions stored in the at least one memory, The communication device is caused to perform the operations and/or processing performed by the first device in any method embodiment.
- this application also provides a wireless communication system, including the first device and the second device in various embodiments of this application.
- At least one (item) involved in the embodiments of this application indicates one (item) or multiple (items). Multiple (items) refers to two (items) or more than two (items).
- “And/or” describes the relationship between related objects, indicating that there can be three relationships. For example, A and/or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. The character “/” generally indicates that the related objects are in an "or” relationship.
- first, second, etc. may be used to describe various objects in this disclosure, these objects should not be limited to these terms. These terms are only used to distinguish objects from each other.
- the processor mentioned in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processor, digital signal processor (DSP), application specific integrated circuit (application specific integrated circuit) circuit, ASIC), off-the-shelf programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
- a general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.
- the memory in the embodiment of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories.
- non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (erasable PROM, EPROM), electrically removable memory. Erase electrically programmable read-only memory (EPROM, EEPROM) or flash memory.
- Volatile memory can be random access memory (RAM), which is used as an external cache.
- RAM static random access memory
- DRAM dynamic random access memory
- SDRAM synchronous dynamic random access memory
- double data rate SDRAM double data rate SDRAM
- DDR SDRAM double data rate SDRAM
- ESDRAM enhanced synchronous dynamic random access memory
- SLDRAM synchronous link dynamic random access memory
- direct rambus RAM direct rambus RAM
- the disclosed systems, devices and methods can be implemented in other ways.
- the device embodiments described above are only illustrative.
- the division of the units is only a logical function division. In actual implementation, there may be other division methods.
- multiple units or components may be combined or can be integrated into another system, or some features can be ignored, or not implemented.
- the coupling or direct coupling or communication connection between each other shown or discussed may be through some interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical or other forms.
- the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
- each functional unit in each embodiment of the present application can be integrated into one processing unit, each unit can exist physically alone, or two or more units can be integrated into one unit.
- the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.
- the technical solution of the present application is essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product.
- the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application.
- the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program code. .
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
本申请提供一种量化方法,可应用于两个设备共用同一个量化字典的应用场景中。在该方法中,提出将待量化的信息划分为多个量化部分,并针对每个量化部分分别进行量化的思路。提出一种量化字典的表示格式,一个量化字典可以由字典使用方式和字典内容来表示,这种表示格式实际上将待量化的信息的R个量化部分和M个量化子字典建立起一种对应关系。若M个子字典中的第一子字典和R个量化部分中的第一量化部分对应,则第一子字典用于对第一量化部分进行量化,或者用于对第一量化部分经量化获得的信息进行解量化,第一子字典中的每个字典元素均是第一量化部分的量化信息的候选。采用该量化字典的表示格式,可以降低设备存储量化字典的开销。
Description
本申请要求于2022年8月12日递交中国国家知识产权局、申请号为“202210971983.8”、申请名称为“量化方法和装置”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请实施例涉及无线通信技术领域,更具体地,涉及一种量化方法和装置。
在无线通信技术领域的很多场景下,用户设备(user equipment,UE)需要向基站反馈信道状态信息(channel state information,CSI),通常是由UE测量基站发送的下行参考信号,并基于接收到的下行参考信号估计出下行信道矩阵,最终生成CSI反馈给基站。UE反馈的CSI的精度越高,信息越丰富,基站根据CSI或恢复的信道信息就越准确,从而基站确定的预编码矩阵越准确。但是随着通信系统中天线阵列规模的增大,可支持的天线端口数增多,由于完整的信道矩阵大小与天线端口数成正比,在大规模多入多出(multiple input multiple output,MIMO)系统中由UE通过CSI反馈完整信道矩阵给基站,意味着巨大的反馈开销。机器学习的方法具有更强的非线性特征提取能力,可以对信道矩阵特征做更有效的提取,相比于传统的CSI反馈方案,可在同等规模的反馈中包含更多的信道信息,从而降低CSI压缩反馈的信息损失,提高基站信道恢复的精度。相比于传统方案可用更少的反馈量表示同样的信道信息,可进一步降低反馈开销。
现有一种基于深度学习的信道信息压缩反馈机制,在UE侧使用基于深度学习的编码器对信道信息进行压缩信道信息处理,经过量化字典将压缩信息处理后的信息量化表示,并向基站反馈该量化表示在量化字典中的索引。基站保存有和UE相同的量化字典,根据该量化字典,基站可以将接收到索引恢复为量化表示,并通过将量化表示输入基于深度学习的解码处理,恢复出信道信息。
在上述机制中,量化字典通常较大,导致基站侧和UE侧的存储开销是巨大的。
发明内容
本申请提供一种量化方法和装置,以期降低量化字典的存储开销。
第一方面,提供了一种量化方法,应用于通信设备(例如,本申请中的第一设备),该方法包括:
第一设备获取第一量化字典;
第一设备使用第一量化字典对待量化的信息进行量化,其中,第一量化字典包括M个子字典,M个子字典用于对待量化的信息的R个量化部分进行量化,M和R均为大于1的正整数;或者,
第一设备使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,其中,第一量化字典包括M个子字典,M个子字典用于对所接收到的R个量化部分经量化后的信息,即经量化获得的信息,进行解量化,M和R均为大于1的正整数;
第一量化字典包括字典使用方式和M个子字典各自的字典内容,字典使用方式指示M个子字典和R个量化部分的对应关系,M个子字典中的第一子字典和R个量化部分中的第一量化部分对应,第一子字典用于对第一量化部分进行量化,或者,第一子字典用于对第一量化部分经量化获得的信息进行解量化;M个子字典中的每个子字典的字典内容包括至少一个字典元素,M个子字典中的第一子字典所包括的每个字典元素为第一量化部分的量化信息的候选。
在本技术方案中,提出将待量化的信息划分为多个量化部分,并针对每个量化部分分别进行量化的思路。在此基础上,提出一种量化字典的表示格式,基于该表示格式,一个量化字典可以由字典使用方式和字典内容来表示。这种字典表示格式实际上将待量化的信息的R个量化部分和M个量化子字典(简称为子字典)建立起一种对应关系。在该对应关系中,若M个子字典中的第一子字典和R个量化部分中的第一量化部分对应,则第一子字典用于对第一量化部分进行量化,或者,第一子字典用于
对第一量化部分经量化获得的信息进行解量化,第一子字典的字典内容中的每个字典元素均是该第一量化部分的候选。基于本申请提供的量化字典的表示格式,可以降低设备存储量化字典的开销。
可选地,第一设备采用第一量化字典对待量化的信息进行量化,获得经量化后的信息之后,可以将该经量化后的信息发送给第二设备。
可选地,第一设备接收到的R个量化部分分别经过量化后的信息来自于第二设备。
此外,在已知的基于深度学习的信道信息压缩反馈机制中,基站侧和UE侧各自保存一个相同的量化字典,该量化字典用于各种场景下的量化(或解量化),或者说,该机制对于不同的场景不作区分。例如,高速场景、低速场景,室内场景或室外场景等,均是采用同一个量化字典。考虑到不同的场景中,待量化的信息的取值分布可能完全不同或者有着各自的特点,如果对场景不作任何区分而统一采用同一个量化字典,量化性能差。
为此,本申请提出另一种量化方法,该量化方法旨在不同的场景下采用不同的量化字典,以使得量化字典可以适应当前场景中待量化的信息的取值分布,以提高量化性能。
第二方面,提供一种量化方法,应用于通信设备(例如,本申请中的第一设备),该方法包括:
第一设备获取第一量化字典;
第一设备使用第一量化字典对第一待量化的信息进行量化,或者,
第一设备使用第一量化字典对接收到的第一经量化获得的信息进行解量化;
第一设备获取第二量化字典;
第一设备使用第二量化字典对第二待量化的信息进行量化,或者,
第一设备使用第二量化字典对接收到的第二经量化获得的信息进行解量化,
其中,第一量化字典包括多个字典元素,所述第一量化字典所包括的每个字典元素为第一量化场景中产生的待量化的信息的候选;
所述第二量化字典包括多个字典元素,所述第二量化字典所包括的每个字典元素为第二量化场景中产生的待量化信息的候选;
所述第一量化字典和所述第二量化字典不同。
可选地,所述第一量化字典对应第一量化场景;所述第二量化字典对应第二量化场景。
可选地,所述第一待量化的信息的取值分布与所述第二待量化的信息的取值分布不同。
也即,待量化的信息的取值分布是否相同,可以作为判断是否为不同的量化场景的依据。
示例性地,如果第一量化场景中产生的待量化信息的取值分布,与第二量化场景中产生的待量化信息的取值分布相同或者接近,例如,不超过某个判定阈值,则第一量化场景和第二量化场景实际上属于相同的量化场景,则第一量化场景变化到第二量化场景,或者第二量化场景变化到第一量化场景,不需要更新量化字典。相反,如果第一量化场景中产生的待量化信息的取值分布,与第二量化场景中产生的待量化信息的取值分布差异较大,例如,超出某个判定阈值,则认为第一量化场景和第二量化场景为不同的量化场景,第一量化场景变化到第二量化场景,或者第二量化场景变化到第一量化场景,则需要选择与当前量化场景相适应的量化字典。
作为一个示例,在一些通信系统的定位场景中,如果需要对定位信息,例如待定位设备的位置坐标,进行量化,位置分布的改变意味着定位场景的变化。又例如,在UE向网络侧反馈CSI的场景下,信道状况的变化导致CSI的取值分布也产生变化,从而可以看作是CSI反馈场景的变化等。这些不同的场景都是上述不同的量化场景的例子。
可选地,不同量化场景下,待量化的信息的取值分布维度可能是不同的,例如,可能是二维,三维或者更多维度的,不作限定。例如,对于定位场景下,定位信息可以是三维的空间坐标。
可选地,在第二方面的方法中,第一设备获取量化字典,包括:第一设备从量化字典表中获取量化字典。其中,量化字典表中包含至少一个候选量化字典。例如,在第二方面中,第一量化字典和/或第二量化字典是从量化字典表中获取(或者说选取)的。示例性地,第一设备可以根据量化场景的不同,从量化字典表中选择合适的量化字典。
结合第二方面,在第二方面的第一种实现方式中,第一量化字典包括M个子字典,M个子字典用于对待量化的信息的R个量化部分进行分别量化或者用于对R个量化部分分别经量化获得的信息进行解量化,M和R均为正整数;
第二量化字典包括C个子字典,C个子字典用于对待量化的信息的S个量化部分进行分别量化或者用于对S个量化部分分别经量化获得的信息进行解量化,S和C均为正整数;
所述第一量化字典包括字典使用方式和M个子字典各自的字典内容,字典使用方式指示M个子字典和R个量化部分的对应关系,M个子字典中的第一子字典和R个量化部分中的第一量化部分对应,第一子字典用于对第一量化部分进行量化,或者,第一子字典用于对第一量化部分经量化获得的信息进行解量化;M个子字典中的每个子字典的字典内容包括至少一个字典元素,M个子字典中的第一子字典所包括的每个字典元素为第一量化部分的量化信息的候选;
所述第二量化字典包括字典使用方式和C个子字典各自的字典内容,字典使用方式指示C个子字典和S个量化部分的对应关系,C个子字典中的第一子字典和S个量化部分中的第一量化部分对应,第一子字典用于对第一量化部分进行量化,或者,第一子字典用于对第一量化部分经量化获得的信息进行解量化;C个子字典中的每个子字典的字典内容包括至少一个字典元素,C个子字典中的第一子字典所包括的每个字典元素为第一量化部分的量化信息的候选。
由于第二量化字典的结构和第一量化字典的结构类似,只是子字典数目或一个或多个子字典的字典内容不同,以下针对第一量化字典的描述也可适用于第二量化字典。为了简略,后续以第一量化字典为例,进行描述。
此外,第二方面中的量化字典的字典内容,以及第一设备获取量化字典的具体实现以及流程均可以参考第一方面中,第一设备获取量化字典的具体实现以及流程,文中不再赘述。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一子字典所包括的每个字典元素对应一个量化信息和一个索引;
第一子字典所包括的每个字典元素为第一量化部分的量化信息的候选,包括:
第一子字典所包含的每个字典元素所对应的量化信息为第一量化部分的量化信息的候选。
在该实现方式中,基于量化字典的表示格式,量化字典中的每个子字典的字典内容由字典元素构成,而每个字典元素对应一个量化信息和一个索引。一个字典元素的索引可以在该子字典中唯一指示该字典元素,从而也可以唯一对应该字典元素对应的量化信息。由此,在传输量化字典的更新部分时,如果一个子字典仅涉及部分字典元素的更新,则只需要传输该子字典中涉及更新的字典元素,而不需要完整的量化字典,可以降低传输量化字典的开销。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备获取第一量化字典,包括:
第一设备对第二量化字典进行更新,获得第一量化字典;
以及,该方法还包括:
第一设备向第二设备发送第一量化字典的全部或部分。
在该实现方式中,第一设备对量化字典进行更新,并将更新获得的量化字典提供给第二设备使用。
可选地,第一设备为终端设备,第二设备为接入网设备。由于量化字典的作用是对终端侧的编码器的输出进行量化,因此,终端设备可基于编码器的输出分布获得量化精度更高的量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,M个子字典包括P个子字典和Q个子字典,其中,
P个子字典是第一设备从第一设备存储的量化字典表中选取的,量化字典表中包括至少P个候选子字典;
Q个子字典是第一设备通过进行字典训练获得的;
0≤P≤M,0≤Q≤M,P+Q=M,P和Q均为整数。
在该实现方式中,第一设备更新获得的第一量化字典中包括M个子字典,该M个子字典可以是从量化字典表中选取,或者进行字典训练获得的,或者是这两种方式的结合。第一设备根据应用场景的变化(例如,低速场景到高速场景),对量化字典进行更新,可以在各种不同的应用场景下始终获得较好的量化性能。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面某些实现方式中,第一设备向第二设备发送第一量化字典的部分,包括:
第一设备向第二设备发送第一消息,第一消息指示第一量化字典相对于第二量化字典的更新部分。
可选地,第一消息“指示”更新部分,也可以替换表述为第一消息“携带”更新部分,并不限定该更新部分的具体指示方式,即,可以通过包含该更新部分来实现,或者,可以通过与该更新部分对应的其它信息来实现。其中,该更新部分与该其他信息的对应关系可以为协议预定义,比如,预先存储在第一设备和第二设备中,或者,通过预先配置给第一设备和/或第二设备,该预先配置可以由第一设备和/或第二设备实现,或者,由不同于第一设备和第二设备的另一设备实现。
可选地,所述第一量化字典对应第一量化场景;所述第二量化字典对应第二量化场景。
可选地,所述第一待量化的信息的取值分布与所述第二待量化的信息的取值分布不同。
也即,待量化的信息的取值分布是否相同,可以作为判断是否为不同的量化场景的依据。
示例性地,如果第一量化场景中产生的待量化信息的取值分布,与第二量化场景中产生的待量化信息的取值分布相同或者接近,例如,不超过某个判定阈值,则第一量化场景和第二量化场景实际上属于相同的量化场景,则第一量化场景变化到第二量化场景,或者第二量化场景变化到第一量化场景,不需要更新量化字典。相反,如果第一量化场景中产生的待量化信息的取值分布,与第二量化场景中产生的待量化信息的取值分布差异较大,例如,超出某个判定阈值,则认为第一量化场景和第二量化场景为不同的量化场景,第一量化场景变化到第二量化场景,或者第二量化场景变化到第一量化场景,则需要选择与当前量化场景相适应的量化字典。
作为一个示例,在一些通信系统的定位场景中,如果需要对定位信息,例如待定位设备的位置坐标,进行量化,位置分布的改变意味着定位场景的变化。又例如,在UE向网络侧反馈CSI的场景下,信道状况的变化导致CSI的取值分布也产生变化,从而可以看作是CSI反馈场景的变化等。这些不同的场景都是上述不同的量化场景的例子。
可选地,不同量化场景下,待量化的信息的取值分布维度可能是不同的,例如,可能是二维,三维或者更多维度的,不作限定。例如,对于定位场景下,定位信息可以是三维的空间坐标。
在该实现方式中,由于本申请中提供的量化字典的表示格式,使得量化字典更新后,第一设备只需要向第二设备传输量化字典的更新部分,而不是完整的更新后的量化字典,即可实现更新后的量化字典从第一设备到第二设备的传输,由于完整的量化字典巨大,因此仅传输更新部分可以降低传输量化字典的开销。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,更新部分包括:
R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;
和/或,
R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
在该实现方式中,量化字典的更新可能涉及字典使用方式的更新,或字典内容的更新,或者字典使用方式和字典内容均有更新。第一设备向第二设备传输更新部分时,仅需要传输涉及更新的部分或全部对应关系,和/或,涉及更新的部分或全部字典内容即可,可以降低传输量化字典的开销。
应理解,上述更新部分包括L个量化部分和子字典之间的对应关系表示第一量化字典的字典使用方式更新,更新部分包括Z个量化部分所对应的子字典的字典内容表示第一量化字典的字典内容更新。
应理解,涉及更新的部分对应关系是指R个量化部分中只有一些量化部分和子字典之间的对应关系更新,而不是R个量化部分和子字典之间的对应关系全部更新。此外,涉及更新的部分字典内容也是类似的,是指量化字典中的部分字典内容更新,例如,部分子字典中的字典元素更新,而不是全部的子字典的字典内容更新。并且,对于字典内容发生更新的一个子字典而言,可能只有部分字典元素更新,而不是每个字典元素更新。相比于传输一个完整的量化字典,除了在全部对应关系更新且全部的子字典中的所有字典元素更新的情况,该实现方式就可以降低传输开销。并且,量化字典涉及更新的部分越少,和传输完整的量化字典相比,传输开销降低的越明显。
结合第一方面,在第一方面的某些实现方式中,若更新部分包括R个量化部分中的L个量化部分和子字典之间的对应关系,则第一消息指示L个量化部分,以及L个量化部分各自对应的子字典的标识;和/或,
若更新部分包括R个量化部分中的Z个量化部分所对应的子字典的字典内容,则第一消息指示Z个子字典的标识,以及Z个子字典各自涉及更新的字典元素以及索引。
可选地,在该实现方式中,第一消息不限于为一个或多个。或者说,第一消息中所包括的信息可以携带在同一条消息中,或,多于一条消息中。
第一消息包括多条消息的情况,比如,在第一消息A中包括所述L个量化部分,在第一消息B中以预定的顺序携带与所述L个量化部分各自对应的子字典的标识。或者,
再比如,在多条消息中以预设的顺序分别携带一个量化部分和其对应的标识,比如,预设的顺序包括量化部分在信息中的前后顺序,举例来说,在第一消息a中包括L个量化部分中的第一个量化部分和其对应的子字典的标识,在第一消息b中包括L个量化部分中的第二个量化部分和其对应的子字典的标识,以此类推。
此外,所述更新的字典元素以及索引包括以下中的一项或多项:
新增的字典元素以及新增的索引,
修改的字典元素以及该字典元素所对应的索引。
可选地,该子字典的标识为量化字典中新增的标识时,可以认为该子字典为新增的子字典。当所述更新部分所包括的字典使用方式的更新中包括该新增的子字典的标识时,所述更新部分所包括的子字典的字典内容还可以包括该子字典中的全部字典元素以及对应的索引。
在该实现方式中,第一设备通过向第二设备指示量化字典的更新部分,可以降低传输开销。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备获取第一量化字典之前,方法还包括:
第一设备向第二设备发送第一信息,第一信息用于指示由第一设备对第二量化字典进行更新。
在该实现方式中,由于第一设备和第二设备共用一个量化字典,因此,量化字典的更新不能由第一设备和第二设备各自独立进行,而是需要在单侧完成之后传输给另一侧。如果第一信息指示由第一设备对量化字典进行更新,第二设备在更新自己的编码器或解码器的过程中不会修改量化字典的内容,可以保证第一设备和第二设备各自保存的量化字典始终一致。可选地,在终端设备向网络设备反馈CSI的场景下,可以保证网络侧恢复出CSI信息的准确性。
结合第一方面,在第一方面的某些实现方式中,第一设备向第二设备发送第一量化字典的全部或部分之后,方法还包括:
第一设备从第二设备接收第二信息,第二信息用于请求启用第一量化字典。
在实现方式中,第一设备向第二设备发送第一量化字典。第二设备准确接收到第一量化字典并确定使用第一量化字典的情况下,向第一设备请求启用第一量化字典,可以提高两个设备之间同步更新量化字典的可靠性。例如,如果第二设备未准确接收到第一量化字典,或者第二设备接收到第一量化字典之后经过评估,量化性能不符合要求,第二设备可以不接受启用新的量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,方法还包括:
第一设备向第二设备发送第三信息,第三信息用于指示启用第一量化字典。
在该实现方式中,第一设备基于第二设备请求启用新的量化字典的信息,向第二设备发送第三信息,指示开启新的量化字典,以保持第一设备和第二设备之间量化字典更新的同步。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备使用第一量化字典对待量化的信息进行量化,包括:
第一设备从发送第三信息的时刻开始,经过时间间隔T,启用第一量化字典对待量化的信息进行量化,其中,时间间隔T是基于协议约定或由第三信息指示的;或者,
第一设备使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,包括:
第一设备从发送第三信息的时刻开始,经过时间间隔T,启用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,其中,时间间隔T是基于协议约定或由第三信息指示的。
可选地,时间间隔T可以由第二设备决定并发送给第一设备,例如,时间间隔T可以携带在用于请求启用第一量化字典的第二信息中,不作限定。或者,时间间隔T还可以是第一设备和第二设备通过其它的信令或消息来协商确定的,不限于通过第二信息或第三信息来携带或指示。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备获取第一量化字典,包括:
第一设备从第二设备接收第一量化字典的全部或部分。
在该实现方式中,量化字典由第二设备更新,并在更新获得新的量化字典之后传输给第一设备。根据本申请提供的量化字典的表示格式,第一设备接收更新后的第一量化字典,具体是接收第一量化字典的字典使用方式和字典内容各自涉及更新的部分,相比于传输完整的量化字典,可以降低传输量化字典的开销。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备从第二设备接收第一量化字典的部分,包括:
第一设备从第二设备接收第二消息,第二消息指示第一量化字典相对于第二量化字典的更新部分,第一量化字典是由第二设备对第二量化字典进行更新获得的。
结合第一方面,在第一方面的某些实现方式中,更新部分包括:
R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;
和/或,
R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,若更新部分包括R个量化部分中的L个量化部分和子字典之间的对应关系,则第二消息指示L个量化部分,以及L个量化部分各自对应的子字典的标识;和/或,
若更新部分包括R个量化部分中的Z个量化部分所对应的子字典的字典内容,则第二消息指示Z个子字典的标识,以及Z个子字典各自涉及更新的字典元素以及索引。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,该方法还包括:
若第一设备确定所获得的第一量化字典有缺损,第一设备向第二设备发送第四信息,第四信息用于指示第二设备发送第一量化字典的缺损部分,缺损部分包括第一量化字典的字典使用方式的缺损和/或字典内容的缺损;
第一设备从第二设备接收缺损部分。
在该实现方式中,从对方接收量化字典的设备需要确定是否完整接收该量化字典,如果量化字典有缺损,可以请求对方发送缺损部分,可以保证传输量化字典时的可靠性。
结合第一方面,在第一方面的某些实现方式中,第一设备获取第一量化字典的全部或部分之前,方法还包括:
第一设备接收来自于第二设备的第五信息,第五信息用于指示由第二设备对量化字典进行更新。
结合第一方面,在第一方面的某些实现方式中,第一设备获取到第一量化字典的全部或部分之后,方法还包括:
第一设备向第二设备发送第六信息,第六信息用于请求启用第一量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,该方法还包括:
第一设备接收来自于第二设备的第七信息,第七信息用于指示启用第一量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备使用第一量化字典对待量化的信息进行量化,包括:
第一设备从接收到第七信息的时刻开始,经过时间间隔T,启用第一量化字典对待量化的信息进行量化,其中,时间间隔T是基于协议约定或由第七信息指示的;
或者,第一设备使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,包括:
第一设备从接收到第七信息的时刻开始,经过时间间隔T,启用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,其中,时间间隔T是基于协议约定或由第七信息指示的。
可选地,时间间隔T可以由第一设备决定并发送给第二设备,例如,时间间隔T可以携带在用于请求启用第一量化字典的第六信息中,不作限定。或者,时间间隔T还可以是第一设备和第二设备通过其它的信令或消息来协商确定的,不限于通过第六信息或第七信息来携带或指示。
上述几个实现方式中,第二设备更新量化字典并传输给第一设备,这几个实现方式的技术效果,与第一设备更新量化字典并传输给第二设备的相应实现方式的技术效果是对等的,不再赘述。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为接入网设备,第二设备为终端设备;
第一设备在获取到第一量化字典之后,该方法还包括:
第一设备评估第一量化字典的量化性能;
若确定第一设备的解码器与第一量化字典不匹配,第一设备根据第一量化字典对解码器进行更新。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为接入网设备,第二设备为终端设备;
第一设备在启用第一量化字典之前,该方法还包括:
第一设备接收来自于第二设备的第一经量化获得的信息;
第一设备根据第一经量化获得的信息和第二量化字典,对第一经量化获得的信息进行解量化。
在该实现方式中,接入网设备在启用更新后的量化字典之前,采用更新前的量化字典对接收到的经量化后的信息进行解量化(或者说,进行信道信息的恢复),可以保证和终端设备总是同步使用同一个量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为接入网设备,第二设备为终端设备;
第一设备在启用第一量化字典之后,方法还包括:
第一设备接收来自于第二设备的第二经量化获得的信息;
第一设备根据第二经量化获得的信息和第一量化字典,对第二经量化获得的信息进行解量化。
在该实现方式中,接入网设备在启用新的量化字典之后,采用新的量化字典对接收到的经量化后的信息进行解量化,可以保证和终端设备总是同步使用同一个量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为终端设备,第二设备为接入网设备;
第一设备获得第一量化字典之后,方法还包括:
第一设备评估第一量化字典的量化性能;
若确定第一设备的编码器的输出与第一量化字典不匹配,第一设备根据第一量化字典对编码器进行更新。
在该实现方式中,量化字典由基站进行更新,并在更新后传输给终端设备。终端设备评估量化字典的量化性能,并在编码器与新的量化字典不匹配的情况下,根据新的量化字典对编码器进行更新,可以使得编码器和新的量化字典更匹配,进而提升量化性能。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为终端设备,第二设备为接入网设备;
第一设备在启用第一量化字典之前,方法还包括:
第一设备根据第二量化字典,确定待反馈的第一信道信息的经量化后的信息;
第一设备向第二设备发送第一信道信息的经量化后的信息。
在该实现方式中,终端设备在启用更新后的量化字典之前,采用更新前的量化字典进行信道信息的量化,可以保证和网络设备总是同步使用同一个量化字典。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设备为终端设备,第二设备为接入网设备;
第一设备在启用第一量化字典之后,方法还包括:
第一设备根据第一量化字典,确定待反馈的第二信道信息的经量化后的信息;
第一设备向第二设备发送第二信道信息的经量化后的信息。
在该实现方式中,终端设备在启用更新后的量化字典之后,采用更新后的量化字典进行信道信息的量化,可以保证和网络设备总是同步使用同一个量化字典。
应理解,该实现方式中的第一信道信息和第二信道信息均为待量化的信息的示例。或者说,待量化的信息的一个示例可以为信道信息,例如CSI。具体地,作为一个示例,信道信息可以为UE侧的编码器的输出信息,例如CSI的压缩信息。
结合第一方面或第二方面的第一种实现方式,在第一方面或第二方面的某些实现方式中,第一设
备使用第一量化字典对待量化的信息进行量化,包括:
若第一设备为终端设备,待量化的信息为编码器的输出信息,第一设备使用第一量化字典对编码器的输出信息进行量化;
或者,
若第一设备为接入网设备,第一设备使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化。
第三方面,提供了一种量化或解量化装置,装置具有实现第一方面或第二方面,或其任一可能的实现方式中的方法的功能。功能可以通过硬件实现,也可以通过硬件执行相应的软件实现。硬件或软件包括一个或多个与上述功能相对应的单元。
可选地,该装置可以为接入网设备,也可以是接入网设备中的装置、模块或芯片(或芯片系统)等,或者是能够和接入网设备匹配使用的装置。或者,该装置也可以为终端设备,也可以是终端设备中的装置、模块或芯片(或芯片系统)等,或者是能够和接入网设备匹配使用的装置。
一种设计中,该装置可以包括执行第一方面中所描述的方法/操作/步骤/动作所一一对应的模块,该模块可以是硬件电路,也可是软件,也可以是硬件电路结合软件实现。另一种设计中,该装置可以包括处理模块和通信模块。
示例性地,处理模块用于实现上述第一方面或第二方面,或其任意可能的实现方式中由该装置内部执行的处理/操作/步骤/动作等;通信模块用于执行由该装置执行的发送/接收/传输等动作,具体参见装置实施例的说明。可替换地,通信模块也可以称为收发模块,包括接收和/或发送的功能。
第四方面,提供一种量化或解量化装置,包括处理器和存储器。可选地,还可以包括收发器。其中,存储器用于存储计算机程序,处理器用于调用并运行存储器中存储的计算机程序,并控制收发器收发信号,以使通信装置执行如第一方面的方法或第二方面的方法,或这些方面的任一可能的实现方式中的方法。
第五方面,提供一种量化或解量化装置,包括处理器和通信接口,所述通信接口用于接收数据和/或信息,并将接收到的数据和/或信息传输至所述处理器,所述处理器处理所述数据和/或信息,以及,通信接口还用于输出经处理器处理之后的数据和/或信息,以使得如第一方面的方法或第二方面的方法,或这些方面的任一可能的实现方式中的方法被执行。
示例性的,通信接口可以是收发器、电路、总线、模块、管脚或其它类型的通信接口。
第六方面,提供一种量化或解量化的装置,包括至少一个处理器,所述至少一个处理器与至少一个存储器耦合,所述至少一个处理器用于执行所述至少一个存储器中存储的计算机程序或指令,以使所述通信装置执行如第一方面的方法或第二方面的方法,或这些方面的任一可能的实现方式中的方法。
可选地,上述装置可以为芯片或芯片系统。
第七方面,提供一种计算机可读存储介质,所述计算机可读存储介质中存储有计算机指令,当计算机指令在计算机上运行时,使得如第一方面或第二方面,或这些方面的任一可能的实现方式中的方法被执行。
第八方面,提供一种计算机程序产品,所述计算机程序产品包括计算机程序代码,当所述计算机程序代码在计算机上运行时,使得如第一方面或第二方面,或这些方面的任一方面的任一可能的实现方式中的方法被执行。
第九方面,提供一种芯片,该芯片用于读取存储器中存储的计算机程序,执行上述第一方面或第二方面,或这些方面的任一可能的实现方式所提供的方法。或者,所述芯片包括用于执行上述第一方面或第二方面,或者这些方面的任一可能的实现方式所提供的方法的电路。
第十方面,还提供了一种芯片系统,该芯片系统包括处理器,用于支持装置实现上述第一方面或第二方面,或这些方面的任一可能的实现方式所提供的方法。在一种可能的设计中,所述芯片系统还包括存储器,所述存储器用于保存该装置必要的程序和数据。该芯片系统可以由芯片构成,也可以包含芯片和其他分立器件。
第十一方面,提供一种无线通信系统,包括如第三方面,第四方面,第五方面或第六方面所述的量化或解量化装置。示例性地,可以包括如本申请实施例中的第一设备和第二设备,或者,第一设备和第三设备,或者,第二设备和第三设备。
图1为适用于本申请实施例的通信场景的示意图。
图2为适用于本申请实施例的基于深度学习的信道压缩反馈的一个示例。
图3为适用于本申请提供的量化方法的示意图。
图4为采用聚类法更新量化字典的示意图。
图5为本申请提供的量化方法的一个示例。
图6为本申请提供的量化方法的一个示例。
图7为本申请提供的用于量化或解量化的装置的示意性框图。
图8为本申请提供的用于量化或解量化的装置的示意性结构图。
下面将结合附图,对本申请中的技术方案进行描述。
本申请实施例的技术方案可以应用于各种通信系统,包括但不限于:第五代(the 5th generation,5G)系统,也可称为或新无线(new radio,NR)系统,长期演进(long term evolution,LTE)系统、长期演进高级技术(long term evolution-advanced,LTE-A)系统、LTE频分双工(frequency division duplex,FDD)系统、LTE时分双工(time division duplex,TDD)系统等。还可以应用于未来的通信系统,例如第六代移动通信系统。此外,还可以应用于设备到设备(device to device,D2D)通信,车辆外联(vehicle-to-everything,V2X)通信,机器到机器(machine to machine,M2M)通信,机器类型通信(machine type communication,MTC),以及物联网(internet of things,IoT)通信系统、卫星通信系统或者其它通信系统,等。此外,还可以扩展到类似的无线通信系统中,如无线保真(wireless-fidelity,WiFi),全球微波互联接入(worldwide interoperability for microwave access,WIMAX),以及第三代合作伙伴计划(3rd generation partnership project,3gpp)相关的通信系统等,不作限定。
通信系统中的一个网元可以向另一个网元发送信号,或从另一个网元接收信号。其中信号可以包括信息、信令或者数据等。其中,网元也可以被替换为实体、网络实体、设备、通信设备、通信模块、节点、通信节点等等。本申请中以设备为例进行描述。
适用于本申请的通信系统,可以包括一个或多个第一设备,以及,一个或多个第二设备。可选地,第一设备和第二设备中的一个可以为终端设备,另一个可以为网络设备。或者,第一设备和第二设备可以均为网络设备或均为终端设备。
参见图1,图1为适用于本申请实施例的通信场景的示意图。如图1,本申请实施例主要适用于终端设备向网络设备发送信息的场景。作为一个示例,适用于终端设备向网络设备发送信道状态信息(channel state information,CSI)的场景。本申请实施例中以终端设备向网络设备发送CSI为例介绍本申请的技术方案,本领域技术人员基于该示例,可以将该技术方案应用于其它场景中,例如,一个设备向另一个设备发送信息的场景下,不作限定。应理解,图1所示的通信系统中各个设备的数量、类型仅作为示意,本公开并不限于此,实际应用中在通信系统中还可以包括更多的终端设备、更多的接入网设备,还可以包括其它网元,例如可以包括核心网设备,和/或用于实现人工智能功能的网元。
示例性地,终端设备也可以称为用户设备(user equipment,UE)、接入终端、用户单元、用户站、移动站、移动台、移动终端(mobile terminal,MT)、远方站、远程终端、移动设备、用户终端、终端、无人机、无线通信设备、用户代理或用户装置。本申请实施例中的终端设备可以是指向用户提供语音和/或数据连通性的设备,可以用于连接人、物和机,例如具有无线连接功能的手持式设备、车载设备等。本申请的实施例中的终端设备可以是手机(mobile phone)、平板电脑(pad)、笔记本电脑、掌上电脑、移动互联网设备(mobile internet device,MID)、可穿戴设备,虚拟现实(virtual reality,VR)设备、增强现实(augmented reality,AR)设备、工业控制(industrial control)中的无线终端、无人驾驶(self driving)中的无线终端、远程手术(remote medical surgery)中的无线终端、智能电网(smart grid)中的无线终端、运输安全(transportation safety)中的无线终端、智慧城市(smart city)中的无线终端、智慧家庭(smart home)中的无线终端等。
本申请实施例中,用于实现终端的功能的装置可以是终端,也可以是能够支持终端实现该功能的
装置,例如芯片系统或芯片、硬件电路、软件模块、或者硬件电路和软件模块的结合,该装置可以被安装在终端中,或者与终端匹配使用。本申请实施例中,芯片系统可以由芯片构成,也可以包括芯片和其他分立器件。
示例性地,网络设备可以是具有无线收发功能的设备,该网络设备可以是提供无线通信功能服务的设备,通常位于网络侧,包括但不限于第五代(5th generation,5G)通信系统中的下一代基站(gNodeB,gNB)、第六代(6th generation,6G)移动通信系统中的基站、未来移动通信系统中的基站,或无线保真(wireless fidelity,WiFi)系统中的接入节点,长期演进(long term evolution,LTE)系统中的演进型节点B(evolved node B,eNB)、无线网络控制器(radio network controller,RNC)、节点B(node B,NB)、基站控制器(base station controller,BSC)、家庭基站(例如,home evolved NodeB,或home Node B,HNB)、基带单元(base band unit,BBU),传输接收点(transmission reception point,TRP)、发射点(transmitting point,TP)、基站收发台(base transceiver station,BTS)、卫星、无人机等。在一种网络结构中,网络设备可以包括集中单元(centralized unit,CU)节点,或包括分布单元(distributed unit,DU)节点,或者为包括CU节点和DU节点的RAN设备,或者为包括控制面CU节点和用户面CU节点,以及DU节点的RAN设备,或者,网络设备还可以为云无线接入网络(cloud radio access network,CRAN)场景下的无线控制器、中继站、车载设备以及可穿戴设备等。此外,基站可以是宏基站、微基站、中继节点、施主节点,或其组合。基站还可以指用于设置于前述设备或装置内的通信模块、调制解调器或芯片。基站还可以是移动交换中心以及D2D、V2X、M2M通信中承担基站功能的设备、6G网络中的网络侧设备、未来的通信系统中承担基站功能的设备等。基站可以支持相同或不同接入技术的网络,不作限定。
本申请实施例中,用于实现网络设备的功能的装置可以是网络设备,也可以是能够支持网络设备实现该功能的装置,例如芯片系统或芯片、硬件电路、软件模块、或者硬件电路和软件模块的结合,该装置可以被安装在网络设备中,或者与网络设备匹配使用。本申请实施例中,芯片系统可以由芯片构成,也可以包括芯片和其他分立器件。
随着无线通信技术的发展,所支持的业务不断增多,在系统容量、通信延时等指标上均对通信系统提出了更高的要求。通过扩展UE可使用的带宽,可有效提升通信速率。为了支持更宽的可用带宽,可为UE分配多个连续或非连续的频带用于通信。为支持每个频带上的有效通信,基站需要获取每个频带上的信道状态信息(Channel State Information,CSI)。但是,在目前广泛应用的基于频分双工(Frequency Division Duplex,FDD)的通信系统中,上下行信道不具备互异性,基站需要通过UE进行上行反馈的方式获得下行CSI。即,基站向UE发送下行参考信号,UE接收下行参考信号。由于UE已知下行参考信号的发送信息,UE基于接收到的下行参考信号可以估计(测量)出下行参考信号所经历的下行信道,UE基于测量得到下行信道矩阵生成CSI,并将CSI反馈给基站。
在FDD系统中,CSI反馈的一个重要部分是预编码矩阵指示(precoding matrix indicator,PMI),即在CSI中采用0、1比特来量化信道矩阵或预编码矩阵。PMI的设计(也可称为码本设计)是移动通信系统中的一个基本问题。传统的码本设计方法是在协议中预定义(约定)一系列预编码矩阵及相应编号,这些预编码矩阵称为码字。采用预定义码字或多个预定义码字的线性组合可近似信道矩阵或预编码矩阵。因此,UE可通过PMI向基站反馈码字相应的编号以及加权系数中的一个或多个,用于基站侧恢复信道矩阵或预编码矩阵。UE所反馈的CSI精度越高,信息越丰富,基站根据CSI所恢复的信道信息越准确,基站确定的预编码矩阵越准确,使得下行空分复用性能越好,UE的接收信干噪比越高,系统容量越高。但是,由于完整的信道矩阵大小与天线端口数成正比,随着系统中天线阵列规模的增大,可支持的天线端口数增多,在大规模MIMO系统中,UE通过CSI向基站反馈完整信道矩阵的反馈开销也是巨大的。
机器学习的方法具有更强的非线性特征提取能力,可以对信道矩阵特征做更有效的提取,相比于传统方案可在同等规模的反馈中包含更多的信道信息,从而降低CSI压缩反馈的信息损失,提高基站端信道恢复的精度。已知一种技术,将人工智能(artificial Intelligence,AI)引入无线通信网络中,产生了一种基于AI模型的反馈CSI的方案。终端设备利用AI模型对CSI进行压缩反馈,接入网设备利用AI模型对压缩的CSI进行恢复。该方案相比于传统方案,可用更少的反馈量表示同样的信道信息,从而进一步降低反馈开销。其中,AI模型可以包括神经网络。
参见图2,图2为适用于本申请实施例的基于深度学习的信道压缩反馈的一个示例。如图2,在UE侧使用基于深度学习的编码器压缩信道信息,经量化字典将压缩信息量化表示(用量化字典中与其最接近的元素量化表示压缩信息),将该量化表示在字典中的索引用有限比特反馈给基站。基站侧有与UE侧相同的量化字典,因此可将反馈的索引恢复为压缩信息的量化表示,进而将该压缩信息的量化表示输入基于深度学习的解码器,恢复出信道信息。在上述过程中,UE和基站需要共用同一个量化字典。但是由于量化字典通常非常大,因此,UE和基站存储量化字典的开销也是巨大的。
此外,当环境或配置变化导致量化字典需要更新时,需要UE的量化字典和基站的量化字典同步更新,以及并且更新后的量化字典使用时间也是需要同步的。由于目前的很多应用中,UE侧的编码过程、基站侧的解码过程和量化过程是作为一个AI模型整体来进行联合训练的。训练完成之后,在UE侧和基站侧配置训练好的AI模型。后续如果AI模型或者量化过程需要更新,更新后的AI模型也需要传输到基站侧或UE侧,并且每次更新都需要传输完整的AI模型,传输开销巨大。
为此,本申请提供一种量化方法,旨在降低量化字典的存储开销,另外,在量化字典发生更新的情况,不需要传输完整的量化字典,而是仅传输更新部分即可,可以降低传输量化字典的开销。
下面介绍本申请提供的技术方案。
本申请提供的量化方法,可以应用于任何需要对信息进行量化的场景下,例如,两个设备中有一个设备需要对信息进行量化并传输给另一个设备,而该另一个设备需要对接收到的经量化获得的信息进行解量化即可。
此外,采用本申请提供方案对信息进行量化的方法,和对接收到的信息进行解量化的方法,是可以解耦使用的。但是,也不限定量化的过程和解量化的过程配合使用。例如,在很多的通信场景中,量化和解量化通常是配合使用的。例如,在上述列举的通信系统中,例如FDD通信场景下,UE需要向基站反馈信道信息,如CSI信息,此场景下,UE需要对待反馈的CSI信息的压缩信息进行量化,并向基站反馈量化后的信息。基站从UE接收量化后的信息,并需要对其解量化,以最终获得CSI信息。再例如,在一些通信定位场景中,一个设备需要向另一个设备反馈位置信息,该位置信息可以是经过量化的。而另一个设备接收到量化后的信息,需要对其解量化,以获得位置信息。因此,本申请对具体的应用场景不予限制,上述列举的对信道信息,如CSI信息或位置信息进行量化也仅是作为示例。
参见图3,图3为适用于本申请提供的量化方法的示意图。
310、第一设备获取第一量化字典。
可选地,第一量化字典可以是由第一设备初始生成,或者是由第一设备对第二量化字典进行更新获得,不作限定。
在本申请中,一个量化字典是通过该量化字典的字典使用方式和该量化字典所包含的M个子字典各自的字典内容来表示的。也或者说,一个量化字典包括字典使用方式和字典内容。字典内容包括该量化字典所包括的M个子字典各自的字典内容。
第一量化字典的字典使用方式指示M个子字典和R个量化部分的对应关系。M个子字典中包括第一子字典,第一子字典和R个量化部分中的第一量化部分对应,则第一子字典用于对第一量化部分进行量化,第一子字典为M个子字典中的任意一个子字典。
第一量化字典用于对待量化的信息进行量化。具体地,第一量化字典包括M个子字典,M个子字典用于对待量化的信息的R个量化部分进行量化,M和R均为正整数。或者说,第一量化字典包括M个子字典,M个子字典和待量化的信息的R个量化部分对应。
可选的,M和R均为大于1的正整数。
可以理解的是,解量化为量化的逆过程,因此,第一量化字典所包括的M个子字典也用于对R个量化部分经量化获得的信息进行解量化,M和R均为大于1的正整数。其中,经量化获得的信息,即,经量化后的信息,可以为量化部分所对应的量化信息在子字典中的索引。为了描述上的简洁,文中也可以将经量化获得的信息称为“量化索引”。
在一个示例中,R=M,即M个子字典用于对M个量化部分进行量化,此时,M个子字典和M个量化部分一一对应,M个子字典中的每个子字典用于对所对应的量化部分进行量化,如表1的示例。
表1
在另一个示例中,R大于M,此时,R个量化部分中有一些量化部分可以对应M个子字典中的同一个子字典,如表2的示例。
表2
如上文所述,第一量化字典的字典使用方式指示了M个子字典和待量化的信息的R个量化部分的对应关系,M个子字典中的每个子字典用于对所对应的量化部分进行量化。以M个子字典中包括第一子字典为例,第一子字典和R个量化部分中的第一量化部分对应,则第一子字典用于对第一量化部分进行量化。
例如,在表1中,第一量化字典包括3个子字典,分别为Y1,Y2和Y3。待量化的信息包括3个量化部分,分别为量化部分1,量化部分2和量化部分3。由此,Y1和该3个量化部分中的量化部分1对应。Y2和该3个量化部分中的量化部分2对应。Y3和该3个量化部分中的量化部分3对应。可见,字典使用方式用于指示一个子字典对应的量化部分。
又例如,在表2中,第一量化字典包括2个子字典,分别为Y1和Y2,待量化的信息包括3个量化部分。子字典Y1和该3个量化部分中的量化部分1和量化部分3对应,由此,子字典Y1用于对待量化的信息的量化部分1和量化部分3进行量化。
M个子字典中的每个子字典的字典内容包括至少一个字典元素,以第一子字典为例,第一子字典的字典内容所包括的至少一个字典元素为第一量化部分的量化信息的候选。换句话说,第一子字典的字典内容中包括的字典元素,是第一子字典所对应的量化部分的量化信息的候选。例如,在表1中,Y1的字典内容中包含的每个字典元素是量化部分1的量化信息的候选,Y2的字典内容中包含的每个字典元素是量化部分2的量化信息的候选。又例如,在表2中,由于Y1对应量化部分1和量化部分3,因此,Y1的字典内容中的每个字典元素既是量化部分1的量化信息的候选,同时也是量化部分3的量化信息的候选。
第一量化字典包括M个子字典,M个字典中的任意一个子字典的字典内容均包含至少一个字典元素。进一步地,每个字典元素对应一个量化信息和一个索引。可选地,每个字典元素对应的“量化信息”也可以替换表达为“量化表示”。因此,也可以说,如果待量化的信息的第一量化部分对应的M个子字典中的第一子字典,则第一子字典中的每个字典元素对应的量化表示为第一量化部分的量化表示的候选,如表3的示例。
表3
如表3,子字典Y1的字典内容包括4个字典元素,分别为a1,a2,a3,a4;子字典Y2的字典内容包括8个字典元素,分别为b1,b2,…,b8;子字典Y3的字典内容包括2个字典元素,分别c1,c2。根据
上文对于字典元素的介绍可知,字典元素a1,a2,a3,a4均是量化部分1的量化信息的候选。或者说,当需要将待量化的信息的量化部分1量化时,可以从Y1中的字典元素a1,a2,a3,a4中选择一个字典元素作为量化部分1的量化信息。在选择一个量化部分的量化信息时,可以按照一定的规则来选择。示例性地,可以从Y1中选择和量化部分1的欧式距离最接近的一个字典元素作为量化部分1的量化信息。假设如果字典元素a3与量化部分1的欧式距离最接近,则字典元素a3即为量化部分1的量化信息,也或者说,字典元素a3是量化部分1的量化表示。
下面再给出一个具体的示例。
假设待量化的信息是长度为N的一个实数向量,如果将该实数向量表示划分为互相不重复也不遗漏的多个量化部分,可以对该多个量化部分分别进行量化。作为一个示例,一个量化部分的表示方式可以为[x1,x2),其中,x1表示该量化部分的起点的索引,x2表示紧接着的下一个量化部分的起点的索引,例如,x1表示量化部分1的起点的索引,x2表示量化部分2的起点的索引。由此,该量化字典的字典使用方式可以如表4所示。如表4,编码器的输出信息的量化部分[x1,x2)和子字典Y1对应,则Y1用于对量化部分[x1,x2)进行量化。量化部分[x2,x3)和子字典Y2对应,则Y2用于对量化部分[x2,x3)以此类推。
表4
作为一个示例,假设待量化的信息为N=8的实数向量,将该实数向量划分为4个等长的不重不漏的量化部分,则针对该实数向量的量化,字典使用方式的一种可能如表5所示。
表5
则表5表示,针对N=8长的实数向量所划分为4个等长的量化部分,采用标识为Y1的子字典分别对第一个量化部分(该实数向量中的第0个值和第1个值)和第三个量化部分(该实数向量中的第4个值和第5个值)进行量化,采用标识为Y2的子字典对第二个量化部分(该实数向量的第2个值和第3个值)和第四个量化部分(该实数向量中的第6个值和第7个值)进行量化。
作为一个示例,假设待量化的信息为长度N=8的实数向量,具体为[0.6,0.2,-0.7,-0.1,0.3,0.8,-0.1,-0.2],将该实数向量按表5划分量化部分并针对每个量化部分分别进行量化,即用量化字典Y1,对第一量化部分(第0个值到第1个值,即[0.6,0.2])和第三量化部分(第4个值到第5个值,即[0.3,0.8])进行量化;用量化字典Y2,对第二量化部分(即第2个值到第3个值,即[-0.7,-0.1])和第四量化部分(即第6个值到第7个值,即[-0.1,-0.2])进行量化。
假设子字典Y1包括4个字典元素,该4个字典元素各自对应一个量化信息,则该4个字典元素对应的4个量化信息为a1=[0.1,0.1],a2=[0.1,0.6],a3=[0.6,0.1],a4=[0.6,0.6],并且,该4个字典元素在子字典Y1中对应的索引依次为(0,1,2,3)。
假设子字典Y2包括4个字典元素,该4个字典元素各自对应的量化信息依次为b1=[-0.1,-0.1],b2=[-0.1,-0.6],b3=[-0.6,-0.1],b4=[-0.6,-0.6]。该4个字典元素在子字典Y2中对应的索引依次为(0,1,2,3)。
则根据表5和子字典Y1、Y2的字典内容,可得到该实数向量,即[0.6,0.2,-0.7,-0.1,0.3,0.8,-0.1,
-0.2],的4个量化部分的量化信息为[0.6,0.1],[-0.6,-0.1],[0.1,0.6],[-0.1,-0.1],该4个量化信息分别对应Y1中的索引2,Y2中的索引2,Y1中的索引1,和Y2中的索引0。索引2,2,1,0会反馈给基站,进而基站可根据表5查找对应的子字典,得到4个量化部分的量化信息分别为[0.6,0.1],[-0.6,-0.1],[0.1,0.6],[-0.1,-0.1]。
对于这一示例,量化误差为量化信息[0.6,0.1],[-0.6,-0.1],[0.1,0.6],[-0.1,-0.1]和量化之前的信息[0.6,0.2,-0.7,-0.1,0.3,0.8,-0.1,-0.2]之间的差值。作为一个示例,量化误差的一种计算方法为两者的均方误差,即量化误差=((0.6-0.6)2+(0.1-0.2)2+(-0.6+0.7)2+(-0.1+0.1)2+(0.1-0.3)2+(0.6-0.8)2+(-0.1+0.1)2+(-0.1+0.2)2))/8。
对于每一个待量化的信息,都可以按照上述步骤计算量化误差,多个量化误差的均值可以作为量化字典的量化性能的评估准则。当平均量化误差超过一个阈值,可以认为量化字典无法很好地对待量化的信息进行量化。反之,如果平均量化误差不超过(小于或等于)该阈值,则可以认为量化字典可以很好地对待量化的信息进行量化,量化性能可以满足量化需求。
在本申请中,第一量化字典包括M个子字典,M个子字典中的每个子字典可以由两个参数(K,B)确定,其中,K表示该子字典用于长度为K,即包括K个向量,的一个量化部分的量化,B表示对于该子字典所对应的量化部分的每一个输入采用B个比特进行量化。该B个比特的不同取值,可以对应不同的候选量化信息。也就是说,该B个比特的不同取值,为不同的候选量化信息的索引。可以理解的是,对于参数为(K,B)的子字典,最多可以包括2B个字典元素。需要注意的是,对于不同的子字典,K和B的取值可能相同或不同,例如,子字典Y1可以由(K1,B1)确定,子字典Y2可以由(K2,B2)确定。
表a
若一个子字典用于实现矢量量化,则对于参数为(K,B)的一个子字典,作为一个示例,该子字典最多可以包括2B个字典元素(或者简称为元素),其中每个元素对应一个长度为K的实数向量和一个索引,每个元素对应的实数向量均为该子字典对应的量化部分的量化信息的候选。其中,子字典中不同元素所对应的索引不同。该索引用于在该子字典中唯一标识该元素。例如,以待量化的信息为UE的编码器的输出信息为例,假设该编码器的输出信息的一个长度为K的量化部分使用参数为(K,B)的一个子字典进行量化,假设编码器的一次输出中,该长度为K的量化部分的取值记作o1,则从该子字典中寻找一个与o1距离(可以指欧式距离)最近的元素作为o1的近似,则该元素对应的长度为K的实数向量为o1的量化信息,并且该元素对应的索引可以用于在子字典中唯一标识该元素,从而也可以唯一对应o1的量化信息。基站根据该量化信息的索引,可以从同一个子字典中找到对应的长度为K的实数向量作为o1的近似。
可选地,若一个字典用于实现其它量化方法,则对于参数为(K,B)的一个子字典,可采用代码或者模型的方式存储,该代码或模型包含一个量化函数。量化函数的输入为长度为K的待量化实数向量,量化函数的输出为B个比特的二进制比特串。该量化函数对应一个反量化函数,反量化函数的输入为B个比特的二进制比特串,反量化函数的输出为长度为K的实数向量。本申请中,反量化和解量化可以互相替换。
以上对本申请中涉及的量化字典的表示格式作了详细说明。
可以理解的是,如上所涉及的表格均为一些对应关系的集合,这些对应关系还可以采用其他形式进行表示,比如,字符串,函数等,在此不予限定。
可选地,在步骤310中,第一设备获取第一量化字典,可以包括多种可能的实现方式。
实现方式1
第一设备对第二量化字典进行更新,获得第一量化字典。
可以理解的是,第二量化字典可以是指在第一设备在获取更新后的第一量化字典之前,第一设备和第二设备所使用的,用于对第一设备和第二设备之间交互的信息进行量化或解量化的量化字典。
如上文所述,由于第一设备和第二设备使用同一个量化字典,因此,如果量化字典是由第一设备更新获得,第一设备需要将更新获得的第一量化字典发送给第二设备。
可选地,第一设备和第二设备可以分别存储有多个子字典用于不同应用场景(或称为量化场景)的量化,或者用于对待量化的信息中的不同量化部分进行量化。作为一个示例,第一设备可以保存有一个量化字典表,量化字典表中包括有一个或多个候选子字典。第一设备对第二量化字典进行更新,获得第一量化字典。可选地,第一量化字典和第二量化字典相比,可以是字典使用方式的更新,或者,可以是待量化的信息的一个或多个量化部分所对应的子字典的字典内容的更新,或者是上述两个方面的更新均涉及。需要理解的是,第二量化字典更新为第一量化字典,可能涉及到待量化的信息的量化部分的数目的变化,例如,量化部分的数目增多,或者量化部分的数目减少。或者,量化部分的数目没有变化,不作限定。其中,量化部分的数目的更新也包含在上述字典使用方式的更新或字典内容的更新的范畴内。
示例性地,第一设备可以是从量化字典表中的候选子字典中选取一个或多个候选子字典,作为更新后的第一量化子字典所包括的M个子字典中的部分子字典。或者,如果量化字典表中包括的候选子字典都不适合当前的应用场景,第一设备也可以进行量化字典训练,从而获得新的子字典。在本申请,第一设备通过量化字典训练获得的子字典,默认为是不包含在第一设备存储的量化字典表中的。或者说,第一设备通过训练获得的,是一个新的子字典,第一设备后续可以将该新的子字典加入到量化字典表中,可选地,第一设备可以为该新的子字典分配一个新的标识,以和量化字典表中已有的子字典进行区分,例如,量化字典表中保存有子字典Y1,Y2和Y3,第一设备为训练获得的新的子字典分配标识Y4,并将Y4增加到量化字典表中,此时量化字典表中包含子字典Y1,Y2,Y3和Y4。可选地,第一设备也可以为该新的子字典分配已有标识,或者说,可以用新的子字典替换已有的某个子字典。例如,第一设备为训练获得的新的子字典分配标识Y3,由此,量化字典表中保存的子字典为已有的子字典Y1,Y2和训练获得的新的子字典(标识为Y3)。
可选地,第一设备获得的第一量化字典中包括M个子字典,M个子字典包括P个子字典和Q和子字典。其中,P个子字典是第一设备从第一设备存储的量化字典表中选取的,其中,量化字典表中包括至少P个子字典;Q个子字典是第一设备通过进行量化字典训练获得的。0≤P≤M,0≤Q≤M,P+Q=M,P和Q均为整数。
应理解,当P=0时,表示第一量化字典中包含的M个子字典均是由第一设备通过量化字典训练获得的;当Q=0时,表示第一量化字典中包含的M个子字典均是第一设备从量化字典表中选取的。当P>0且Q>0时,表示第一量化字典中包含的M个子字典中的P个是从量化字典表中的候选子字典中选取的,剩余的Q个子字典是第一设备通过训练获得的。
第一设备更新获得第一量化字典之后,需要向第二设备发送第一量化字典。根据上文对于量化字典的格式的说明,一个完整的量化字典的传输包括该量化字典的字典使用方式的传输和字典内容的传输。由于字典内容通常较大,所需的传输开销也很大,因此,在字典更新后,应尽量减少完整字典的传输的次数。
根据本文中定义的量化字典的表示格式,量化字典的更新可以分为如下几种场景,在不同的场景下,可以采用不同的传输更新后的量化字典的方式,以降低传输开销。
(1)量化字典的字典使用方式发生更新,但是量化字典的字典内容未更新。
例如,在AI模型部署阶段,第一设备和第二设备可部署多个子字典,这多个子字典可用于适应不同场景的量化需求,或者适应待量化的信息的不同量化部分的量化需求。当场景变化需要进行AI模型/量化字典的更新时,第一设备可以从已有的多个子字典中选择更适合当前场景的子字典用于量化或解量化的处理。例如,第一设备和第二设备部署了多个子字典以分别应对低速场景和高速场景的量化需求。当用户速度发生变化时,只需要更新量化字典的字典使用方式即可。在此场景下,第一设备不需要传输更新后的量化字典的字典内容,只需要传输更新后的量化字典的字典使用方式中涉及更新的部分。
作为一个示例,第一设备和第二设备部署有子字典Y1,Y2,Y3和Y4。假设更新之前的第二量化字
典如表6,更新之后的第一量化字典如表7。
表6
表7
可见,在该场景下,量化字典的更新,可以仅涉及字典使用方式的更新,或者说,待量化的信息的量化部分和子字典的对应关系发生了更新。可选地,和第二量化字典的字典使用方式所指示的对应关系相比,第一量化字典的字典使用方式所指示的对应关系,可以是部分发生变化,也可以是全部发生变化,不作限定。表7中以对应关系部分发生变化为例,例如,量化部分1和子字典Y1的对应关系未变化,只有量化部分2和量化部分3和子字典的对应关系发生变化,但是由于Y3和Y4是量化字典表中存储的已有子字典,因此,量化字典的更新可以仅涉及到字典使用方式的更新。第一设备可以向第二设备指示量化字典的更新部分。作为一个示例,第一设备向第二设备发送第一消息,第一消息指示第一量化字典相对于第二量化字典的更新部分。以表6和表7为例,第一消息指示量化部分2和子字典Y3的对应关系,以及量化部分3和子字典Y4的对应关系。例如,第一消息包含子字典Y3的标识和量化部分2,以及子字典Y4的标识和量化部分3。其中,Y3的标识和量化部分2对应,Y4的标识和量化部分3对应。
在再一个示例中,第一设备和第二设备的量化字典表中存储有子字典Y1,Y2,Y3,Y4和Y5。假设更新之前的第二量化字典如表6所示,更新后的第一量化字典如表8所示。
表8
在此情况下,量化字典的字典使用方式发生更新,具体地,由第二量化字典所指示的3个量化部分(量化部分1,量化部分2和量化部分3)和3个子字典(Y1,Y2,Y3)之间的对应关系,更新为4个量化部分(量化部分1~量化部分4)和4个子字典(Y1,Y2,Y3,Y4)之间的对应关系。可以看出,在字典使用方式更新时,涉及到待量化的信息重新划分为多个量化部分,量化部分的数目增多。在该示例中,第一消息可以包含Y1,Y3,Y4和Y5各自的标识,以及Y1,Y3,Y4和Y5各自对应的量化部分。
(2)量化字典的字典使用方式不发生更新,量化字典的字典内容更新。
在一些情况下,当应用场景发生变化,需要进行AI模型更新或量化字典更新时,如果量化字典表中存储的已有的子字典无法适应当前的应用场景,则第一设备需要重新训练量化字典。应理解,已有的子字典无法适应当前的应用场景,表示更新前所使用的子字典以及量化字典表中的子字典均无法适用当前的应用场景,第一设备需要通过重新训练获得新的量化字典。
作为一个示例,一种量化字典的训练方式或者更新方式为采用聚类法的方式更新量化字典。参见图4,图4为采用聚类法更新量化字典的示意图。以表5中所示的量化字典为例,该量化字典包括4个子字典,针对长度K=8的实数向量(待量化的信息的一个示例)所划分的4个等长的量化部分,对于量化部分[0,2)而言,假设待量化的信息中,量化部分[0,2)的取值分布如图4的左图所示,如果采用
2个比特(B=2)对该量化部分进行量化,则可以将该量化部分的取值空间分为2B=4块,取每一块中取值分布的中心位置为一个字典元素,得到图4的右图,则子字典的字典元素的取值可以较好地近似待量化的信息。这里仅以待量化的信息的一个量化部分作为示例进行说明,剩余三个量化部分对应的子字典的更新也是类似的,不予赘述。
因此,在量化字典的字典使用方式不变的情况下,可以根据待量化的信息的分布更新字典元素。这里,根据待量化的信息的分布更新字典元素,仅是作为更新字典内容的一种实现方式。如果待量化的信息中只有一些量化部分的取值分布发生变化,则可以仅更新取值分布发生变化的量化部分对应的子字典的字典内容。
(3)量化字典的字典使用方式发生更新,且字典内容也发生更新。
当应用场景发生变化,进行AI模型和/或量化字典的更新时,如果量化字典表中已有的子字典均无法适应新的应用场景,则可以重新训练量化字典。
在此情况下,在对量化字典进行更新时,可以同时考虑量化字典的字典使用方式的更新和字典内容的更新。可选地,字典使用方式的更新可以包括量化部分的重新划分,以及每个量化字典对应的子字典标识的更新。
在实现方式1中,第一设备通过从量化字典表中选取子字典,和/或,第一设备重新进行字典训练获得新的子字典,以获得更新后的第一量化字典所包含的M个子字典,实现对量化字典进行更新。
在第一设备对量化字典进行更新的情况下,可选地,第一设备在更新之前,向第二设备发送第一信息,第一信息指示由第一设备对量化字典进行更新。在更新获得第一量化字典之后,第一设备将第一量化字典发送给第二设备。可选地,作为另一种实现,第一设备和第二设备可以约定量化字典的更新由第一设备或第二设备来执行,在此实现中,在对量化字典进行更新之前,第一设备和第二设备之间不需要交互由谁来进行量化字典更新的信息,而是在约定的一方完成量化字典更新之后,直接将更新后的量化字典发送给对方。
上文介绍了更新量化字典的多种可能,在完成更新之后,第一设备向第二设备发送第一量化字典时,第一设备向第二设备发送第一量化字典相对于第二量化字典的更新部分。具体地,这些更新部分是以字典使用方式或者字典内容的方式发送的。例如,第一设备向第二设备发送第一消息,第一消息指示该更新部分。可选地,第一消息可以是一个或多个,不作限定。
以第一量化字典中包含M个子字典为例,若量化字典的更新仅涉及字典使用方式的更新,则更新部分包括R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数。第一设备在向第二设备通知更新部分时,示例性地,第一设备向第二设备发送第一消息,第一消息指示该L个量化部分以及L个量化部分和子字典之间的对应关系。应理解,L个量化部分是指和子字典的对应关系发生变化的量化部分。
若量化字典的更新仅涉及字典内容的更新,则更新部分包括R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。第一设备向第二设备通知更新部分时,第一设备向第二设备发送第一消息,第一消息指示Z个子字典的标识,以及Z个子字典各自涉及更新的字典元素以及索引。其中,Z个量化部分是指所对应的子字典的字典内容发生变化的量化部分。
因此,第二设备在从第一设备接收第一量化字典时,具体可以是接收第一量化字典的字典使用方式和字典内容,并且是涉及更新的字典使用方式和字典内容,从而在第二量化字典的基础上,将第二量化字典同步更新为第一量化字典。
可选地,第二设备在获得第一量化字典之后,向第一设备发送第二信息,第二信息用于请求启用第一量化字典。第一设备在接收到第二信息之后,向第二设备发送第三信息,第三信息用于指示启用第一量化字典。可选地,第三信息中可以携带一个时间间隔T。如果忽略第一设备发送第三信息的时刻,和第二设备接收第三信息的时刻之间的时间误差,第一设备从发送第三信息的时刻开始,经过时间间隔T启用第一量化字典,第二设备从接收到第三信息的时刻开始,经过时间间隔T,开始启用新的量化字典,即第一量化字典。可选地,时间间隔T也可以由协议约定。第二设备从接收到第三信息的时刻开始,默认为经过时间间隔T,双方都启用新的量化字典。
此外,可选地,第二设备在从第一设备接收新的量化字典的过程中,根据量化字典的表示格式来解析接收到的第一量化字典,并判断接收到的第一量化字典是否完整,也即第一量化字典的字典内容
和字典使用方式均无缺损。如果确定所接收到的第一量化字典有缺损部分,第二设备可以向第一设备发送第四信息,第四信息用于指示第一设备重新发送缺损部分。在接收到第一设备重新发送的缺损部分之后,第二设备可以获得完整的第一量化字典。
实现方式2
第一设备从第二设备接收第一量化字典。
可选地,在实现方式2中,第一量化字典可以是由第二设备初始生成,或者对第二量化字典进行更新获得的。第二设备获得第一量化字典之后,第二设备向第一设备发送第一量化字典。
上文对第一设备更新量化字典的过程,以及更新获得新的量化字典之后,第一设备如何向第二设备发送新的量化字典的过程均作了详细说明。可以理解的是,在由第二设备更新量化字典的实现方式2中,第二设备所作的处理或操作与实现方式1中的第一设备所作的处理或操作可以认为是相同的,为了避免赘述,不予赘述。
可选地,在实现方式2中,在更新量化字典之前,第二设备向第一设备发送第五信息,第五信息指示由第二设备对量化字典进行更新。类似地,第一设备在从第二设备获得新的量化字典之后,第一设备向第二设备发送第六信息,第六信息指示请求启用新的量化字典。后续,基于第六信息,第二设备向第一设备发送第七信息,第七信息指示启用新的量化字典。可选地,如果将第二设备发送第七信息的时刻,和第一设备接收第七信息的时刻之间的时间误差考虑在内,则第一设备从接收到第七信息的时刻开始,默认为经过时间间隔T1,启用新的量化字典;第二设备从发送第七信息的时刻开始,默认为经过时间间隔T2,启用新的量化字典。应理解,考虑到第二设备发送第七信息的时刻,和第一设备接收第七信息的时刻有一个时间差,为了保证第一设备和第二设备之间同步启用新的量化字典,第一设备和第二设备分别经过不同的时间间隔之后,开启新的量化字典。作为一个示例,T1和T2的设计可以将上述时间差考虑在内,以使第一设备和第二设备开启新的量化字典的时刻相同。或者,第二设备在第七信息中指示时间间隔T1,以便于第一设备根据接收到第七信息的时刻和时间间隔T1,确定启用新的量化字典的时刻。
可选地,在第一设备从第二设备接收新的量化字典的实现方式2中,第一设备在接收到第一量化字典之后,到第一设备发送用于请求启用第一量化字典的第六信息之前,第一设备可以对第一量化字典的量化性能进行评估。
作为一个示例,第一设备将预先保存的一个测试集的信道信息输入到编码器,并采用第一量化字典对编码器的输出信息进行量化,并计算量化误差,从而评估第一量化字典的量化性能。若量化性能符合量化需求,例如,量化误差小于或等于一个阈值,则认为符合量化需求,第一设备向第二设备发送上述第六信息,以请求启用第一量化字典。
可选地,作为一个示例,在实现方式2中,第一设备还可以根据新的量化字典对编码器(第一设备为UE,第二设备为接入网设备)或解码器(第一设备为接入网设备,第二设备为UE)进行更新。以第一设备为接入网设备为例,若接入网设备确定解码器与从UE获取的第一量化字典不匹配,接入网设备根据第一量化字典对解码器进行更新,例如,更新解码器的参数。例如,接入网设备将量化字典中的量化信息候补输入解码器,得到输出信息,计算输出信息的概率分布和训练集中信道信息的概率分布的差,例如互信息,以此为损失函数更新解码器。以第一设备为UE为例,若UE确定编码器的输出与从接入网设备获取的第一量化字典不匹配,UE根据第一量化字典对编码器进行更新,例如,更新编码器的参数。UE采用一个训练集,将训练集中的信道数据输入编码器得到输出,并采用新字典对编码器输出进行量化,进而可计算得到训练集上的平均量化误差,作为损失函数,该损失函数是编码器这一AI模型的权重的一个函数。如果损失函数小于一个阈值,停止训练,否则更新编码器权重以降低损失函数,例如计算量化误差这一损失函数相对于编码器权重的梯度,并通过梯度下降法更新编码器。其中,编码器权重可以包括编码器由神经网络实现时,神经网络中各神经元的权重。
320、第一设备使用第一量化字典对待量化的信息进行量化,或者,第一设备使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化。
通过步骤310,第一设备获得第一量化字典。之后,第一设备使用第一量化字典进行量化,或者使用第一量化字典进行解量化。示例性地,第一设备接收到的R个量化部分经量化获得的信息可以为R个量化索引。
作为一个示例,在上述UE向网络侧上报CSI的场景中,对于向网络侧上报CSI的终端设备而言,第一量化字典用于对CSI的压缩信息进行量化,之后,UE向网络侧上报经量化后的信息。具体地,经量化后的信息为压缩信息所对应的量化信息在对应的子字典中的索引(例如,简称为量化索引)。对于网络侧而言,网络侧接收到终端设备上报的量化信息的索引(即量化索引),根据第一量化字典,确定出该索引对应的信息,进而恢复出与CSI的压缩信息所对应的量化信息,并经过解码压缩信息,最终获得信道信息。因此,在一个示例中,若第一设备为UE,UE使用第一量化字典对信道信息的压缩信息进行量化。若第一设备为接入网设备,接入网设备使用第一量化字典接收到的经量化后的信息,如量化信息所对应的索引,进行解量化,以恢复出信道信息的压缩信息。
此外,可以理解的是,在更新量化字典的过程中以及启用新的第一量化字典之前,第一设备继续使用第二量化字典对待量化的信息进行量化,第二设备继续使用第二量化字典对接收到的经量化后的信息,如量化信息所对应的索引,进行解量化。
例如,若第一设备为接入网设备,在启用第一量化字典之前,接入网设备接收到来自于UE的第一经量化获得的信息,接入网设备使用第二量化字典,对第一经量化获得的信息进行解量化。在启用第一量化字典之后,若接入网设备接收到来自于UE的第二经量化获得的信息,接入网设备使用第一量化字典,对第二经量化获得的信息进行解量化。
若第一设备为UE,在启用第一量化字典之前,若UE需要向接入网设备反馈第一信道信息,UE使用第二量化字典对第一信道信息进行量化,并发送第一信道信息的经量化后的信息。在启用第一量化字典之后,若UE需要向接入网设备反馈第二信道信息,UE使用第一量化字典对第二信道信息进行量化,并发送第二信道信息的经量化后的信息。
以上对本申请提供的量化字典的表示格式,以及一个设备获取量化字典的过程进行了详细说明。下面结合图5和图6给出更新量化字典的流程的两个示例。需要说明的是,图5和图6旨在对更新量化字典的流程或机制给出一个完整的示例说明,其中涉及到的信息或步骤的详细说明可参考上文相关说明,该示例中不予赘述。
在如下更新量化字典的示例中,假设基站和UE初始使用或之前使用的量化字典为第二量化字典。
参见图5,图5为本申请提供的量化方法的一个示例。
501、可选地,基站向UE发送第一信息,第一信息指示由基站进行量化字典的更新。
502、基站对第二量化字典进行更新,获得第一量化字典。
503、基站向UE发送第一消息,第一消息用于指示量化字典的更新部分。
504、UE接收量化字典的更新部分。
505、UE解析量化字典的更新部分,获得第一量化字典。
506、可选地,根据量化字典的表示格式,如果UE确定所接收到的第一量化字典有缺损部分,UE向基站发送第四信息,第四信息用于请求基站重发第一量化字典的缺损部分。
507、可选地,基站向UE重新发送缺损部分。
UE接收到缺损部分之后,获得完整的第一量化字典。
应理解,步骤506-507是相互关联的步骤,方法500同时包含步骤506-507,或者,不包含步骤506-507。
508、可选地,UE评估第一量化字典的量化性能。
UE完成第一量化字典的量化性能的评估之后,若满足量化需求,可以请求启用第一量化字典,如步骤510。
可选地,UE在请求启用第一量化字典之前,也可以测试第一量化字典和编码器的匹配程度,并根据第一量化字典对编码器进行更新,如步骤509。
509、可选地,若确定编码器的输出与第一量化字典不匹配,UE根据第一量化字典更新编码器。
510、可选地,UE向基站发送第二信息,第二信息用于请求启用第一量化字典。
基站接收来自于UE的第二信息。
511、基站向UE发送第三信息,第三信息用于指示启用第一量化字典。
UE接收来自于基站的第三信息。
512、启用第一量化字典之后,基站采用第一量化字典对接收到的经量化后的信息进行解量化,UE采用第一量化字典对待量化的信息进行量化。
可选地,作为一个示例,UE侧的需要待量化的信息可以为CSI,或者其它需要向基站上报的信息,不作限定。
图5所示流程是由基站更新量化字典的一个示例。下面再给出一个由UE更新量化字典的流程的一个示例,如图6。
参见图6,图6为本申请提供的量化方法的一个示例。
601、可选地,UE向基站发送第一信息,第一信息指示由UE进行量化字典的更新。
602、UE对第二量化字典进行更新,获得第一量化字典。
603、UE向基站发送第一消息,第一消息用于指示第一量化字典相对于第二量化字典的更新部分。
604、基站接收该更新部分。
605、基站解析该更新部分,获得第一量化字典。
可选地,若基站确定第一量化字典有缺损部分,基站可以请求UE重新发送缺损部分。基站接收到缺损部分之后,获得完整的第一量化字典。
可选地,基站可以评估第一量化字典的量化性能,并在评估完成,符合量化需求的情况下,请求UE启用第一量化字典,如步骤606。
可选地,基站在请求启用第一量化字典之前,还可以判断解码器和第一量化字典的匹配程度。若解码器和第一量化字典不匹配,基站根据第一量化字典,更新解码器,例如,更新解码器的参数。
606、可选地,基站向UE发送第二信息,第二信息用于请求启用第一量化字典。
UE接收第二信息。响应于第二信息,UE向基站发送第三信息。
607、可选地,基站接收来自于UE的第三信息,第三信息用于指示启用第一量化字典。
608、启用第一量化字典之后,基站采用第一量化字典对接收到的经量化后的信息进行解量化,UE采用第一量化字典对待量化的信息进行量化。
以上均以量化设备(如第一设备)和解量化设备(如第二设备)中的一项进行量化字典的初始生成和/或更新为例进行说明。可以理解的是,初始生成的量化字典或更新的量化字典也可以由量化设备或解量化设备之外的第三设备提供,比如,量化设备和解量化设备均从第三设备接收该更新的量化字典中的全部或部分或该初始生成的量化字典,或者,量化设备从第三设备接收该更新的量化字典中的全部或部分或该初始生成的量化字典,并将该更新的量化字典中的全部或部分或该初始生成的量化字典发送给解量化设备;或者,解量化设备从第三设备接收该更新的量化字典中的全部或部分或该初始生成的量化字典,并将该更新的量化字典中的全部或部分或该初始生成的量化字典发送给量化设备。示例性地,第三设备可以为OTT服务器(over-the-top server)。
以上对本申请提供的量化方法作了详细说明,下面介绍本申请提供的用于量化或解量化的装置。
参见图7,图7为本申请提供的用于量化或解量化的装置的示意性框图。如图7,装置1000包括处理单元1100、存储单元1200和收发单元1300。
处理单元1100,用于
获取第一量化字典;
使用第一量化字典对待量化的信息进行量化,其中,所述第一量化字典包括M个子字典,所述M个子字典用于对所述待量化的信息的R个量化部分进行量化,M和R均为大于1的正整数;或者,
使用第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,解量化后的信息用于作为解码器的输入,其中,所述第一量化字典包括M个子字典,所述M个子字典用于对所接收到的R个量化部分经量化后的信息进行解量化,M和R均为大于1的正整数;
所述第一量化字典包括所述字典使用方式和所述M个子字典各自的字典内容,所述字典使用方式指示所述M个子字典和所述R个量化部分的对应关系,所述M个子字典中的第一子字典和所述R个量化部分中的第一量化部分对应,所述第一子字典用于对所述第一量化部分进行量化,或者所述第一子字典用于对所述第一量化部分经量化后的信息进行解量化;所述M个子字典中的每个子字典的字典内容包括至少一个字典元素,所述M个子字典中的所述第一子字典所包括的每个字典元素为所述第一量化部分的量化信息的候选,M为正整数;
使用所述第一量化字典进行信道信息的量化或恢复的处理。
可选地,在一个实施例中,所述第一子字典所包括的每个字典元素对应一个量化信息和一个索引;
所述第一子字典所包括的每个字典元素为所述第一量化部分的量化信息的候选,包括:
所述第一子字典所包含的每个字典元素所对应的量化信息为所述第一量化部分的量化信息的候选。
可选地,在一个实施例中,所述处理单元1100,用于对第二量化字典进行更新,获得所述第一量化字典;
以及,所述收发单元1300,还用于向第二设备发送所述第一量化字典的全部或部分。
可选地,在一个实施例中,所述M个子字典包括P个子字典和Q个子字典,其中,
所述P个子字典是所述第一设备从所述第一设备存储的量化字典表中选取的,所述量化字典表中包括至少P个候选子字典;
所述Q个子字典是所述第一设备通过进行字典训练获得的;
0≤P≤M,0≤Q≤M,P+Q=M,P和Q均为整数。
可选地,在一个实施例中,所述收发单元1300,用于:
向所述第二设备发送第一消息,所述第一消息指示所述第一量化字典相对于所述第二量化字典的更新部分。
可选地,在一个实施例中,所述R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;和/或,
所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
可选地,在一个实施例中,若所述更新部分包括所述R个量化部分中的L个量化部分和子字典之间的对应关系,则所述第一消息指示所述L个量化部分,以及所述L个量化部分各自对应的子字典的标识;和/或,
若所述更新部分包括所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,则所述第一消息指示所述Z个子字典的标识,以及所述Z个子字典各自涉及更新的字典元素以及索引。
可选地,在一个实施例中,所述收发单元1300,还用于:
向所述第二设备发送第一信息,所述第一信息用于指示由所述第一设备对所述第二量化字典进行更新。
可选地,在一个实施例中,所述收发单元1300,还用于:
从所述第二设备接收第二信息,所述第二信息用于请求启用所述第一量化字典。
可选地,在一个实施例中,所述收发单元1300,还用于:
向所述第二设备发送第三信息,所述第三信息用于指示启用所述第一量化字典。
可选地,在一个实施例中,所述处理单元1100,用于:
从发送所述第三信息的时刻开始,经过时间间隔T,启用所述第一量化字典对待量化的信息进行量化,其中,所述时间间隔T是基于协议约定或由所述第三信息指示的;
或者,
从发送所述第三信息的时刻开始,经过时间间隔T,启用所述第一量化字典对接收到的所述R个量化部分经量化获得的信息进行解量化,其中,所述时间间隔T是基于协议约定或由所述第三信息指示的。
可选地,在一个实施例中,所述收发单元1300,用于从所述第二设备接收所述第一量化字典的字典使用方式和字典内容;
所述处理单元1100,用于从所述第二设备接收所述第一量化字典的全部或部分。
可选地,在一个实施例中,所述收发单元1300,用于:
从所述第二设备接收第二消息,所述第二消息指示所述第一量化字典相对于第二量化字典的更新部分,所述第一量化字典是由所述第二设备对所述第二量化字典进行更新获得的。
可选地,在一个实施例中,所述更新部分包括:
所述R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;
和/或,
所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
可选地,在一个实施例中,若所述更新部分包括所述R个量化部分中的L个量化部分和子字典之间的对应关系,则所述第二消息指示所述L个量化部分,以及所述L个量化部分各自对应的子字典的
标识;和/或,
若所述更新部分包括所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,则所述第二消息指示所述Z个子字典的标识,以及所述Z个子字典各自涉及更新的字典元素以及索引。
可选地,在一个实施例中,所述处理单元1100,用于确定所获得的第一量化字典有缺损;
以及,所述收发单元1300,用于:
向所述第二设备发送第四信息,所述第四信息用于指示所述第二设备发送所述第一量化字典的缺损部分,所述缺损部分包括所述第一量化字典的所述字典使用方式的缺损和/或所述字典内容的缺损;
从所述第二设备接收所述缺损部分。
可选地,在一个实施例中,所述收发单元1300,用于接收来自于所述第二设备的第五信息,所述第五信息用于指示由所述第二设备对量化字典进行更新。
可选地,在一个实施例中,所述收发单元1300,还用于:
向所述第二设备发送第六信息,所述第六信息用于请求启用所述第一量化字典。
可选地,在一个实施例中,所述收发单元1300,还用于:
接收来自于所述第二设备的第七信息,所述第七信息用于指示启用所述第一量化字典。
可选地,在一个实施例中,所述处理单元1100,用于:
从所述收发单元1300接收到所述第七信息的时刻开始,经过时间间隔T,启用所述第一量化字典对所述待量化的信息进行量化,其中,所述时间间隔T是基于协议约定或由所述第七信息指示的;
或者,
从所述收发单元1300接收到所述第七信息的时刻开始,经过时间间隔T,启用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,其中,所述时间间隔T是基于协议约定或由所述第七信息指示的。
可选地,在一个实施例中,所述装置为接入网设备或配置在接入网设备中的芯片(或芯片系统),所述第二设备为终端设备,所述待量化的信息为编码器的输出信息,所述第一量化字典用于所述装置对接收到的所述R个量化部分经量化获得的信息进行解量化,以获得解码器的输入;
以及,所述处理单元1100,用于:
评估所述第一量化字典的量化性能;
若确定所述装置的解码器与所述第一量化字典不匹配,根据所述第一量化字典对所述解码器进行更新。
可选地,在一个实施例中,所述装置为接入网设备或配置在接入网设备中的芯片(或芯片系统);
所述收发单元1300,用于接收来自于所述第二设备的第一经量化获得的信息;
所述处理单元1100,用于根据所述第一经量化获得的信息和所述第二量化字典,对所述第一经量化获得的信息进行解量化。
可选地,在一个实施例中,所述装置为接入网设备或配置在接入网设备中的芯片(或芯片系统);
所述收发单元1300,用于接收来自于所述第二设备的第二经量化获得的信息;
所述处理单元1100,用于根据所述第二经量化获得的信息和所述第一量化字典,对所述第二经量化获得的信息进行解量化。
可选地,在一个实施例中,所述装置为终端设备或配置在终端设备中的芯片(或芯片系统),所述第二设备为接入网设备,所述待量化的信息为编码器的输出信息,所述第一量化字典用于所述装置对所述编码器的输出信息进行量化;
所述处理单元1100,用于:
评估所述第一量化字典的量化性能;
若确定所述装置的编码器的输出与所述第一量化字典不匹配,根据所述第一量化字典对所述编码器进行更新。
可选地,在一个实施例中,所述装置为终端设备或配置在终端设备中的芯片(或芯片系统);
所述处理单元1100,用于根据所述第二量化字典,确定待反馈的第一信道信息的经量化后的信息;
以及,所述收发单元1300,用于向所述第二设备发送所述第一信道信息的经量化后的信息。
可选地,在一个实施例中,所述装置为终端设备或配置在终端设备中的芯片(或芯片系统);
所述处理单元1100,用于根据所述第一量化字典,确定待反馈的第二信道信息的经量化后的信息;
以及,所述收发单元1300,用于向所述第二设备发送所述第二信道信息的经量化后的信息。
可选地,在一个实施例中,若所述装置为终端设备或配置在所述终端设备上的芯片或芯片系统,所述待量化的信息为编码器的输出信息;所述处理单元1100,具体用于使用所述第一量化字典对所述编码器的输出信息进行量化;或者,
若所述装置为接入网设备或配置在所述接入网设备上的芯片或芯片系统,所述处理单元1100,具体用于使用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化。
在以上各装置实施例中,处理单元1100用于执行除了发送和接收的动作之外由第一设备内部实现的处理和/或操作。收发单元1300用于执行第一设备的接收或发送的动作。存储单元1200用于存储量化字典,以及量化字典表等信息。
例如,在图3中,处理单元1100用于执行量化字典更新的处理,以获得第一量化字典;或者,收发单元1300用于执行从第二设备接收第一量化字典的操作,获得第一量化字典。处理单元1100,还用于执行步骤320。
又例如,在图5中,收发单元1300用于执行由基站执行的步骤501,步骤503,步骤506-507,步骤510-511;处理单元1100用于执行步骤502和由基站执行的步骤512等。在图6中,收发单元1300用于执行由UE执行的步骤601,步骤603,步骤606-607;处理单元1100用于执行步骤602和由UE执行的步骤608。
本申请中对单元(或者也可以为模块)的划分是示意性的,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,另外,在本公开各个示例中的各功能单元可以集成在一个处理器中,也可以是单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。
基于相同的技术构思,本申请还提供一种用于量化或解量化的装置,如图8。
参见图8,图8为本申请提供的用于量化或解量化的装置的示意性结构图。如图8,装置10包括:一个或多个处理器11,一个或多个存储器12以及一个或多个通信接口13。处理器11用于控制通信接口13收发信号,存储器12用于存储计算机程序,处理器11用于从存储器12中调用并运行该计算机程序,以使得装置10执行本申请各方法实施例中由第一设备执行的处理。
例如,处理器11可以具有图7中所示的处理单元1100的功能,存储器12可以具有存储单元1200的功能,通信接口13可以具有收发单元1300的功能。具体地,处理器11可以用于执行由装置内部执行的处理或操作,存储器可以用于执行缓存以及从缓存中删除信息单元的操作,通信接口13用于执行装置的发送和/或接收的操作。
在一种可能的实施方式中,该装置10可以应用于接入网设备,具体装置10可以是接入网设备,也可以是能够支持接入网设备实现上述涉及的任一示例中接入网设备的功能的装置。
示例性地,装置10可以为方法实施例中的第一设备。在这种实现方式中,通信接口13可以为第一设备的收发器。收发器可以包括接收器和/或发射器。可选地,处理器11可以为第一设备的基带装置,通信接口13可以为射频装置。
示例性地,装置10可以为安装在第一设备中的芯片(或芯片系统)。在这种实现方式中,通信接口13可以为接口电路或者输入/输出接口。
其中,图8中器件(例如,处理器、存储器或通信接口)后面的虚线框表示该器件可以为一个以上。
可选地,上述各装置实施例中的存储器与处理器可以是物理上相互独立的单元,或者,存储器也可以和处理器集成在一起,本文不作限定。
此外,本申请还提供一种计算机可读存储介质,所述计算机可读存储介质中存储有计算机指令,当计算机指令在计算机上运行时,使得本申请各方法实施例中由第一设备执行的操作和/或处理被执行。
此外,本申请还提供一种计算机程序产品,计算机程序产品包括计算机程序代码或指令,当计算机程序代码或指令在计算机上运行时,使得本申请各方法实施例中由第一设备执行的操作和/或处理被执行。
此外,本申请还提供一种芯片,所述芯片包括处理器,用于存储计算机程序的存储器独立于芯片
而设置,处理器用于执行存储器中存储的计算机程序,使得安装有所述芯片的装置执行任意一个方法实施例中由第一设备执行的操作和/或处理。
进一步地,所述芯片还可以包括通信接口。所述通信接口可以是输入/输出接口,也可以为接口电路等。进一步地,所述芯片还可以包括所述存储器。
此外,本申请还提供一种通信装置(例如,可以为芯片或芯片系统),包括存储器、处理器和通信接口,所述通信接口用于接收第一信息,并将第一信息发送所述处理器;所述处理器用于根据所述第一信息对所述存储器中缓存的第一信息单元进行处理。
此外,本申请还提供一种通信装置,包括至少一个处理器,所述至少一个处理器与至少一个存储器耦合,所述至少一个处理器用于执行所述至少一个存储器中存储的计算机程序或指令,使得所述通信装置执行任意一个方法实施例中由第一设备执行的操作和/或处理。
此外,本申请还提供一种无线通信系统,包括本申请各方式实施例中的第一设备和第二设备。
本申请实施例中涉及的至少一个(项),指示一个(项)或多个(项)。多个(项),是指两个(项)或两个(项)以上。“和/或”,描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。另外,应当理解,尽管在本公开中可能采用术语第一、第二等来描述各对象、但这些对象不应限于这些术语。这些术语仅用来将各对象彼此区分开。
本申请实施例中所提到的术语“包括”及其任何变形,意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备,没有限定于已列出的步骤或单元,而是可选地还包括其他没有列出的步骤或单元,或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。需要说明的是,本申请中,“示例性地”或者“例如”等词用于表示作例子、例证或说明。本申请中被描述为“示例性地”或者“例如”的任何方法或设计方案不应被解释为比其它方法或设计方案更优选或更具优势。确切而言,使用“示例性的”或者“例如”等词旨在以具体方式呈现相关概念。
本申请实施例中提及的处理器可以是中央处理单元(central processing unit,CPU),还可以是其他通用处理器、数字信号处理器(digital signal processor,DSP)、专用集成电路(application specific integrated circuit,ASIC)、现成可编程门阵列(field programmable gate array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
本申请实施例中的存储器可以是易失性存储器或非易失性存储器,或可包括易失性和非易失性存储器两者。其中,非易失性存储器可以是只读存储器(read-only memory,ROM)、可编程只读存储器(programmable ROM,PROM)、可擦除可编程只读存储器(erasable PROM,EPROM)、电可擦除可编程只读存储器(electrically EPROM,EEPROM)或闪存。易失性存储器可以是随机存取存储器(random access memory,RAM),其用作外部高速缓存。通过示例性但不是限制性说明,许多形式的RAM可用,例如静态随机存取存储器(static RAM,SRAM)、动态随机存取存储器(dynamic RAM,DRAM)、同步动态随机存取存储器(synchronous DRAM,SDRAM)、双倍数据速率同步动态随机存取存储器(double data rate SDRAM,DDR SDRAM)、增强型同步动态随机存取存储器(enhanced SDRAM,ESDRAM)、同步连接动态随机存取存储器(synchlink DRAM,SLDRAM)和直接内存总线随机存取存储器(direct rambus RAM,DRRAM)。应注意,本文描述的系统和方法的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
本领域普通技术人员可以意识到,结合本文中所公开的实施例描述的各示例的单元及算法步骤,能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行,取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能,但是这种实现不应认为超出本申请的范围。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统、装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统、装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如多个单元或组件可以结合或者可以集成到另一个
系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到变化或替换,都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以所述权利要求的保护范围为准。
Claims (37)
- 一种量化方法,其特征在于,由第一设备或用于所述第一设备的芯片执行,所述方法包括:获取第一量化字典;使用所述第一量化字典对待量化的信息进行量化,其中,所述第一量化字典包括M个子字典,所述M个子字典用于对所述待量化的信息的R个量化部分进行量化,M和R均为大于1的正整数;或者,使用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,其中,所述第一量化字典包括M个子字典,所述M个子字典用于对接收到的所述R个量化部分经量化获得的信息进行解量化,M和R均为大于1的正整数;所述第一量化字典包括字典使用方式和所述M个子字典各自的字典内容,所述字典使用方式指示所述M个子字典和所述R个量化部分的对应关系,所述M个子字典中的第一子字典和所述R个量化部分中的第一量化部分对应,所述第一子字典用于对所述第一量化部分进行量化,或者所述第一子字典用于对所述第一量化部分经量化获得的信息进行解量化;所述M个子字典中的每个子字典的字典内容包括至少一个字典元素,所述M个子字典中的所述第一子字典所包括的每个字典元素为所述第一量化部分的量化信息的候选。
- 根据权利要求1所述的方法,其特征在于,所述第一子字典所包括的每个字典元素对应一个量化信息和一个索引;所述第一子字典所包括的每个字典元素为所述第一量化部分的量化信息的候选,包括:所述第一子字典所包含的每个字典元素所对应的量化信息为所述第一量化部分的量化信息的候选。
- 根据权利要求1或2所述的方法,其特征在于,所述获取第一量化字典,包括:对第二量化字典进行更新,获得所述第一量化字典;以及,所述方法还包括:向第二设备发送所述第一量化字典的全部或部分。
- 根据权利要求3所述的方法,其特征在于,所述M个子字典包括P个子字典和Q个子字典,其中,所述P个子字典是从存储的量化字典表中选取的,所述量化字典表中包括至少P个候选子字典;所述Q个子字典是通过进行字典训练获得的;0≤P≤M,0≤Q≤M,P+Q=M,P和Q均为整数。
- 根据权利要求3或4所述的方法,其特征在于,所述向第二设备发送所述第一量化字典的部分,包括:向所述第二设备发送第一消息,所述第一消息指示所述第一量化字典相对于所述第二量化字典的更新部分,所述第一量化字典是基于对所述第二量化字典进行更新获得的。
- 根据权利要求5所述的方法,其特征在于,所述更新部分包括:所述R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;和/或,所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
- 根据权利要求6所述的方法,其特征在于,若所述更新部分包括所述R个量化部分中的L个量化部分和子字典之间的对应关系,则所述第一消息指示所述L个量化部分,以及所述L个量化部分各自对应的子字典的标识;和/或,若所述更新部分包括所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,则所述第一消息指示所述Z个子字典的标识,以及所述Z个子字典各自涉及更新的字典元素以及索引。
- 根据权利要求3至7中任一项所述的方法,其特征在于,所述获取第一量化字典之前,所述方法还包括:向所述第二设备发送第一信息,所述第一信息用于指示由所述对所述第二量化字典进行更新。
- 根据权利要求3至8中任一项所述的方法,其特征在于,所述向第二设备发送所述第一量化字典的全部或部分之后,所述方法还包括:从所述第二设备接收第二信息,所述第二信息用于请求启用所述第一量化字典。
- 根据权利要求3至9中任一项所述的方法,其特征在于,所述方法还包括:向所述第二设备发送第三信息,所述第三信息用于指示启用所述第一量化字典。
- 根据权利要求10所述的方法,其特征在于,所述使用所述第一量化字典对待量化的信息进行量化,包括:从发送所述第三信息的时刻开始,经过时间间隔T,启用所述第一量化字典对所述待量化的信息进行量化,其中,所述时间间隔T是基于协议约定或由所述第三信息指示的;或者,所述使用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,包括:从发送所述第三信息的时刻开始,经过时间间隔T,启用所述第一量化字典对接收到的所述R个量化部分经量化获得的信息进行解量化,其中,所述时间间隔T是基于协议约定或通过所述第三信息指示给所述第二设备。
- 根据权利要求1或2所述的方法,其特征在于,所述获取所述第一量化字典,包括:从第二设备接收所述第一量化字典的全部或部分;解析所述第一量化字典的字典使用方式和/或字典内容,以获得所述第一量化字典。
- 根据权利要求12所述的方法,其特征在于,所述从第二设备接收所述第一量化字典的部分,包括:从所述第二设备接收第二消息,所述第二消息指示所述第一量化字典相对于第二量化字典的更新部分,所述第一量化字典是基于对所述第二量化字典进行更新获得的。
- 根据权利要求13所述的方法,其特征在于,所述更新部分包括:所述R个量化部分中的L个量化部分和子字典之间的对应关系,1≤L≤R,L为整数;和/或,所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,1≤Z≤R,Z为整数。
- 根据权利要求14所述的方法,其特征在于,若所述更新部分包括所述R个量化部分中的L个量化部分和子字典之间的对应关系,则所述第二消息指示所述L个量化部分,以及所述L个量化部分各自对应的子字典的标识;和/或,若所述更新部分包括所述R个量化部分中的Z个量化部分所对应的子字典的字典内容,则所述第二消息指示所述Z个子字典的标识,以及所述Z个子字典各自涉及更新的字典元素以及索引。
- 根据权利要求12至15中任一项所述的方法,其特征在于,所述方法还包括:若确定所获得的第一量化字典有缺损,向所述第二设备发送第四信息,所述第四信息用于指示所述第二设备发送所述第一量化字典的缺损部分,所述缺损部分包括所述第一量化字典的所述字典使用方式的缺损和/或所述字典内容的缺损;从所述第二设备接收所述缺损部分。
- 根据权利要求12至16中任一项所述的方法,其特征在于,所述获取所述第一量化字典的全部或部分之前,所述方法还包括:接收来自于所述第二设备的第五信息,所述第五信息用于指示由所述第二设备对量化字典进行更新。
- 根据权利要求12至17中任一项所述的方法,其特征在于,所述获取到所述第一量化字典的全部或部分之后,所述方法还包括:向所述第二设备发送第六信息,所述第六信息用于请求启用所述第一量化字典。
- 根据权利要求12至18中任一项所述的方法,其特征在于,所述方法还包括:接收来自于所述第二设备的第七信息,所述第七信息用于指示启用所述第一量化字典。
- 根据权利要求19所述的方法,其特征在于,所述使用所述第一量化字典对待量化的信息进行量化,包括:从接收到所述第七信息的时刻开始,经过时间间隔T,启用所述第一量化字典对所述待量化的信息进行量化,其中,所述时间间隔T是基于协议约定或由所述第七信息指示的;或者,所述使用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,包括:从接收到所述第七信息的时刻开始,经过时间间隔T,启用所述第一量化字典对接收到的所述R 个量化部分经量化获得的信息进行解量化,其中,所述时间间隔T是基于协议约定或通过所述第七信息指示给所述第二设备。
- 根据权利要求12至20中任一项所述的方法,其特征在于,所述第一设备为接入网设备,所述第二设备为终端设备,所述待量化的信息为编码器的输出信息,所述第一量化字典用于所述第一设备对接收到的所述R个量化部分经量化获得的信息进行解量化,以获得解码器的输入;所述获取到所述第一量化字典之后,所述方法还包括:评估所述第一量化字典的量化性能;若确定所述第一设备的解码器与所述第一量化字典不匹配,根据所述第一量化字典对所述解码器进行更新。
- 根据权利要求21所述的方法,其特征在于,所述在启用所述第一量化字典之前,所述方法还包括:接收来自于所述第二设备的第一经量化获得的信息;根据所述第一经量化获得的信息和第二量化字典,对所述第一经量化获得的信息进行解量化,所述第一量化字典基于所述第二量化字典更新。
- 根据权利要求21或22所述的方法,其特征在于,所述在启用所述第一量化字典之后,所述方法还包括:接收来自于所述第二设备的第二经量化获得的信息;根据所述第二经量化获得的信息和所述第一量化字典,对所述第二经量化获得的信息进行解量化。
- 根据权利要求12至20中任一项所述的方法,其特征在于,所述第一设备为终端设备,所述第二设备为接入网设备,所述待量化的信息为编码器的输出信息,所述第一量化字典用于所述第一设备对所述编码器的输出信息进行量化;所述获得所述第一量化字典之后,所述方法还包括:评估所述第一量化字典的量化性能;若确定所述第一设备的编码器的输出与所述第一量化字典不匹配,根据所述第一量化字典对所述编码器进行更新。
- 根据权利要求24所述的方法,其特征在于,所述在启用所述第一量化字典之前,所述方法还包括:根据所述第二量化字典,确定待反馈的第一信道信息的经量化后的信息;向所述第二设备发送所述第一信道信息的经量化后的信息。
- 根据权利要求24或25所述的方法,其特征在于,所述在启用所述第一量化字典之后,所述方法还包括:根据所述第一量化字典,确定待反馈的第二信道信息的经量化后的信息;向所述第二设备发送所述第二信道信息的经量化后的信息。
- 根据权利要求1-20中任一项所述的方法,其特征在于,所述使用所述第一量化字典对待量化的信息进行量化,包括:若所述第一设备为终端设备,所述待量化的信息为编码器的输出信息,所述第一设备使用所述第一量化字典对所述编码器的输出信息进行量化;或者,所述使用所述第一量化字典对接收到的R个量化部分经量化获得的信息进行解量化,包括:若所述第一设备为接入网设备,所述第一设备使用所述第一量化字典对接收到的所述R个量化部分经量化获得的信息进行解量化。
- 一种量化方法,其特征在于,由第一设备或用于所述第一设备的芯片执行,所述方法包括:获取第一量化字典;使用第一量化字典对第一待量化的信息进行量化,或者,使用第一量化字典对接收到的第一经量化获得的信息进行解量化;获取第二量化字典;使用第二量化字典对第二待量化的信息进行量化,或者,使用第二量化字典对接收到的第二经量化获得的信息进行解量化,其中,第一量化字典包括多个字典元素,所述第一量化字典所包括的每个字典元素为第一量化场景中产生的待量化的信息的候选;所述第二量化字典包括多个字典元素,所述第二量化字典所包括的每个字典元素为第二量化场景中产生的待量化信息的候选;所述第一量化字典和所述第二量化字典不同。
- 根据权利要求28所述的方法,其特征在于,所述第一量化字典对应第一量化场景,所述第二量化字典对应第二量化场景。
- 根据权利要求28或29所述的方法,其特征在于,所述第一待量化的信息的取值分布与所述第二待量化的信息的取值分布不同。
- 根据权利要求28至30中任一项所述的方法,其特征在于,所述第一量化字典包括M个子字典,所述M个子字典用于对待量化的信息的R个量化部分进行分别量化或者用于对R个量化部分分别经量化获得的信息进行解量化,M和R均为正整数;所述第二量化字典包括C个子字典,所述C个子字典用于对待量化的信息的S个量化部分进行分别量化或者用于对S个量化部分分别经量化获得的信息进行解量化,S和C均为正整数;所述第一量化字典包括字典使用方式和M个子字典各自的字典内容,所述字典使用方式指示M个子字典和R个量化部分的对应关系,所述M个子字典中的第一子字典和所述R个量化部分中的第一量化部分对应,所述第一子字典用于对所述第一量化部分进行量化,或者,所述第一子字典用于对所述第一量化部分经量化获得的信息进行解量化;所述M个子字典中的每个子字典的字典内容包括至少一个字典元素,所述M个子字典中的第一子字典所包括的每个字典元素为所述第一量化部分的量化信息的候选;所述第二量化字典包括字典使用方式和C个子字典各自的字典内容,所述字典使用方式指示C个子字典和S个量化部分的对应关系,所述C个子字典中的第一子字典和所述S个量化部分中的第一量化部分对应,第一子字典用于对第一量化部分进行量化,或者,第一子字典用于对第一量化部分经量化获得的信息进行解量化;C个子字典中的每个子字典的字典内容包括至少一个字典元素,C个子字典中的第一子字典所包括的每个字典元素为第一量化部分的量化信息的候选。
- 一种通信装置,其特征在于,用于实现如权利要求1-31中任一项所述的方法。
- 一种通信装置,其特征在于,包括至少一个处理器,所述至少一个处理器与至少一个存储器耦合,所述至少一个处理器用于执行所述至少一个存储器中存储的计算机程序或指令,以使所述通信装置执行如权利要求1-31中任一项所述的方法。
- 一种量化或解量化的装置,其特征在于,包括处理器和通信接口,所述通信接口用于接收数据和/或信息,并将接收到的数据和/或信息传输至所述处理器,所述处理器处理所述数据和/或信息,以及,通信接口还用于输出经处理器处理之后的数据和/或信息,以使得所述装置执行如权利要求1-31中任一项所述的方法。
- 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有计算机指令,当计算机指令在计算机上运行时,使得如权利要求1-31中任一项所述的方法被实现。
- 一种计算机程序产品,其特征在于,所述计算机程序产品包括计算机程序代码,当所述计算机程序代码在计算机上运行时,使得如权利要求1-31中任一项所述的方法被执行。
- 一种无线通信系统,其特征在于,包括如权利要求32或33所述的通信装置,或者如权利要求34所述的装置。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210971983.8 | 2022-08-12 | ||
CN202210971983.8A CN117639864A (zh) | 2022-08-12 | 2022-08-12 | 量化方法和装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2024032775A1 true WO2024032775A1 (zh) | 2024-02-15 |
Family
ID=89851024
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2023/112591 WO2024032775A1 (zh) | 2022-08-12 | 2023-08-11 | 量化方法和装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN117639864A (zh) |
WO (1) | WO2024032775A1 (zh) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107170020A (zh) * | 2017-06-06 | 2017-09-15 | 西北工业大学 | 基于最小量化误差准则的字典学习静态图像有损压缩方法 |
CN107733569A (zh) * | 2017-09-27 | 2018-02-23 | 中国电子科技集团公司第五十四研究所 | 一种星上载荷多波束采样数据压缩方法 |
CN109075851A (zh) * | 2016-03-24 | 2018-12-21 | 华为技术有限公司 | 大规模多输入多输出(mimo)中的下行信道估计系统和方法 |
US20190312664A1 (en) * | 2016-12-05 | 2019-10-10 | Telefonaktiebolaget Lm Ericsson (Publ) | Methods, encoder and decoder for handling a data stream for transmission between a remote unit and a base unit of a base station system |
-
2022
- 2022-08-12 CN CN202210971983.8A patent/CN117639864A/zh active Pending
-
2023
- 2023-08-11 WO PCT/CN2023/112591 patent/WO2024032775A1/zh unknown
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109075851A (zh) * | 2016-03-24 | 2018-12-21 | 华为技术有限公司 | 大规模多输入多输出(mimo)中的下行信道估计系统和方法 |
US20190312664A1 (en) * | 2016-12-05 | 2019-10-10 | Telefonaktiebolaget Lm Ericsson (Publ) | Methods, encoder and decoder for handling a data stream for transmission between a remote unit and a base unit of a base station system |
CN107170020A (zh) * | 2017-06-06 | 2017-09-15 | 西北工业大学 | 基于最小量化误差准则的字典学习静态图像有损压缩方法 |
CN107733569A (zh) * | 2017-09-27 | 2018-02-23 | 中国电子科技集团公司第五十四研究所 | 一种星上载荷多波束采样数据压缩方法 |
Non-Patent Citations (2)
Title |
---|
MAN CHEN, ZHONG YONG, LI ZHEN-DONG: "Multi--focus image fusion based on latent low• •rank representation combining low• •rank representation", JOURNAL OF JILIN UNIVERSITY(ENGINEERING AND TECHNOLOGY EDITION)., vol. 50, no. 1, 1 January 2020 (2020-01-01), XP093137905, DOI: :10.13229/j.cnki.jdxbgxb20180805 * |
ZHANG, TING: "Quantization for Approximate Nearest Neighbor Search", MASTER’S THESES , 1 May 2017 (2017-05-01), China, pages 1 - 125, XP009552520 * |
Also Published As
Publication number | Publication date |
---|---|
CN117639864A (zh) | 2024-03-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN114762276B (zh) | 信道状态信息反馈 | |
WO2021217519A1 (zh) | 用于调整神经网络的方法和装置 | |
WO2021253937A1 (zh) | 无线通信系统的终端、基站以及由终端和基站执行的方法 | |
WO2018196529A1 (zh) | 一种资源信息确定方法及终端设备、网络设备 | |
US20240030980A1 (en) | Data transmission method and apparatus | |
WO2019051328A1 (en) | REDUCTION OF HIGHER LAYER SIGNALING OVERHEAD IN MULTI-ANTENNA WIRELESS COMMUNICATION SYSTEMS | |
JP2023082207A (ja) | チャネル状態情報を送受信するための方法、端末装置およびネットワーク装置 | |
WO2022001822A1 (zh) | 获取神经网络的方法和装置 | |
CN114492784A (zh) | 神经网络的测试方法和装置 | |
WO2024032775A1 (zh) | 量化方法和装置 | |
WO2022012256A1 (zh) | 通信的方法及通信装置 | |
CN112840697B (zh) | 关于csi开销减少的装置、方法和计算机程序 | |
WO2023115254A1 (zh) | 处理数据的方法及装置 | |
WO2023236143A1 (zh) | 信息收发方法与装置 | |
WO2023231620A1 (zh) | 通信方法及装置 | |
WO2024031456A1 (zh) | 通信方法、装置、设备、存储介质、芯片及程序产品 | |
WO2024046215A1 (zh) | 一种通信方法及装置 | |
WO2024008004A1 (zh) | 一种通信方法及装置 | |
WO2024031701A1 (zh) | 数据获取方法与装置 | |
WO2024046288A1 (zh) | 通信方法和装置 | |
WO2023006096A1 (zh) | 一种通信方法及装置 | |
WO2024026792A1 (zh) | 通信方法、装置、设备、存储介质、芯片及程序产品 | |
US20240080788A1 (en) | Information transmission method, lightweight processing method, and related communication apparatus | |
WO2024149387A1 (zh) | 通信方法及通信装置 | |
WO2024125509A1 (zh) | 模型更新方法、装置、设备及存储介质 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23851995 Country of ref document: EP Kind code of ref document: A1 |