CN116955996B - Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network - Google Patents
Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network Download PDFInfo
- Publication number
- CN116955996B CN116955996B CN202311188066.3A CN202311188066A CN116955996B CN 116955996 B CN116955996 B CN 116955996B CN 202311188066 A CN202311188066 A CN 202311188066A CN 116955996 B CN116955996 B CN 116955996B
- Authority
- CN
- China
- Prior art keywords
- edge
- edge device
- remote radio
- noise
- radio unit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 239000013598 vector Substances 0.000 claims abstract description 194
- 230000007613 environmental effect Effects 0.000 claims abstract description 49
- 230000006870 function Effects 0.000 claims abstract description 44
- 238000013139 quantization Methods 0.000 claims abstract description 29
- 230000002776 aggregation Effects 0.000 claims abstract description 23
- 238000004220 aggregation Methods 0.000 claims abstract description 23
- 238000012847 principal component analysis method Methods 0.000 claims abstract description 17
- 230000005540 biological transmission Effects 0.000 claims description 52
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000009826 distribution Methods 0.000 claims description 18
- 238000013461 design Methods 0.000 claims description 16
- 238000004422 calculation algorithm Methods 0.000 claims description 15
- 239000000654 additive Substances 0.000 claims description 12
- 230000000996 additive effect Effects 0.000 claims description 12
- 238000000605 extraction Methods 0.000 claims description 11
- 238000005286 illumination Methods 0.000 claims description 7
- 238000005457 optimization Methods 0.000 claims description 6
- 238000005265 energy consumption Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000005562 fading Methods 0.000 claims description 4
- 206010021403 Illusion Diseases 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 230000001953 sensory effect Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 11
- 239000000306 component Substances 0.000 description 11
- 238000004891 communication Methods 0.000 description 9
- 238000012545 processing Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 230000004931 aggregating effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009365 direct transmission Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The specification discloses a cloud edge collaborative intelligent reasoning method and a cloud edge collaborative intelligent reasoning system based on a cloud wireless access network, wherein the method is suitable for being configured on edge equipment, a remote radio unit and a central controller and executed, and comprises the following steps: collecting environmental data around the edge equipment through various sensors, and extracting the characteristics of the environmental data at the edge equipment by using a principal component analysis method to obtain characteristic vectors; transmitting the feature vectors of all edge devices to a radio remote unit at the same frequency through an air computing technology, carrying out uplink feature aggregation and quantization on the feature vectors at the radio remote unit, and then transmitting the quantized aggregated feature vectors to a central controller; the central controller receives the aggregated feature vectors, analyzes probability density functions of the aggregated feature vectors, constructs classification discrimination gains, identifies the aggregated feature vectors based on the constructed classification discrimination gains, obtains identification results and makes decision of reasoning tasks.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to a cloud edge cooperative intelligent reasoning method and system based on a cloud wireless access network.
Background
Wireless communication networks are coming from the evolution of "people's alliance", "things's alliance" towards "everything's alliance, the need for supporting diverse intelligent services (e.g. autopilot, smart city, telemedicine) is also increasing, and in this context, edge reasoning techniques have emerged. Edge reasoning refers to the step of downloading a well-trained artificial intelligent model to the edge of a communication network, and making various intelligent decisions by utilizing the reasoning capability of the model. At present, three implementation schemes are adopted for edge reasoning: 1. and executing the reasoning task II at the equipment end, and executing the reasoning task III at the cloud end, so as to realize cloud-edge collaborative reasoning. However, the former two schemes are either difficult to bear intensive computing tasks due to hardware limitations, or limited to sending data to the cloud end would introduce serious communication overhead, and practical deployment faces certain difficulties. In contrast, the cloud-edge collaborative reasoning scheme is adopted to better weigh the contradiction between the computing efficiency and the communication overhead. Specifically, cloud-edge collaborative reasoning divides an artificial intelligent model into two parts: a small portion is deployed at the device end for extracting features from the raw data, which typically involves less computation; the computationally intensive portion is deployed at the cloud end to receive the features extracted from the edge device end and complete the remaining reasoning tasks. In addition, cloud-edge collaborative reasoning is additionally beneficial to the advantage of data privacy protection due to the adoption of a strategy for avoiding direct transmission of high-dimensional original data.
However, current research on Yun Bian collaborative reasoning still focuses mainly on a scheme in which a single edge device participates in reasoning, and this scheme cannot guarantee reasoning performance. Because the features extracted by a single edge device are either concentrated in a narrow view, insufficient information is provided to accomplish the inference task; or the data itself from which the features are extracted may suffer serious distortion defects. To overcome this problem of limited perceptibility of a single device, deploying multiple edge devices to improve reasoning performance becomes a necessary choice. However, the implementation of the present industry edge reasoning system does not consider the actual radio access network architecture supporting multi-device deployment, especially for multi-cell oriented service scenarios supporting massive user connections. Therefore, a flexible wireless access network architecture is needed to solve the problems of a great increase in energy consumption and mutual interference between access nodes caused by densely deploying access nodes in the actual deployment of the edge reasoning system.
In summary, the present invention aims to provide a new method for solving the existing problems in the edge reasoning technology, which combines the flexible wireless access network architecture supporting multi-device deployment with cloud-edge collaborative reasoning to realize efficient intelligent decision services and further improve the application potential of the edge reasoning system in various intelligent service scenarios.
Disclosure of Invention
The specification provides a cloud edge collaborative intelligent reasoning method and a cloud edge collaborative intelligent reasoning system based on a cloud wireless access network, which are used for overcoming at least one technical problem in the related art.
According to a first aspect of embodiments of the present disclosure, a cloud edge collaborative intelligent reasoning method based on a cloud radio access network is provided, which is adapted to be configured on an edge device, a remote radio unit, and a central controller, and includes:
collecting environmental data around the edge equipment through various sensors, and carrying out feature extraction on the environmental data at the edge equipment by using a principal component analysis method to obtain feature vectors for reasoning tasks; the environment data comprise measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data;
transmitting the feature vectors of all edge devices to a remote radio unit at the same frequency through an air computing technology so as to perform uplink feature aggregation on the feature vectors at the remote radio unit and obtain aggregated feature vectors;
quantizing the aggregated feature vector at the remote radio unit, and then transmitting the quantized aggregated feature vector to a central controller;
Analyzing probability density functions of the aggregated feature vectors and constructing classification discrimination gains according to the aggregated feature vectors received by the central controller from the remote radio unit, identifying the aggregated feature vectors based on the constructed classification discrimination gains to obtain identification results and making decisions of reasoning tasks.
Optionally, the step of collecting environmental data around the edge device by using multiple sensors, and performing feature extraction on the environmental data at the edge device by using a principal component analysis method to obtain a feature vector for an inference task includes:
collecting environmental data around the edge device by a plurality of sensors;
and extracting the characteristics of the environmental data by using the principal component matrix to obtain characteristic vectors, and further obtaining the distribution of the characteristic vectors according to the characteristic vectors.
Optionally, the environmental data around the edge device is collected by various sensors; extracting features of the environmental data by using the principal component matrix to obtain feature vectors, and further obtaining distribution of the feature vectors according to the feature vectors, wherein the method comprises the following steps:
the environmental data around the edge equipment is collected through various sensors, and the serial number isThe acquired environmental data around the edge device is
；
Wherein,is the sensing data of the true value,is a perceptual distortion, i.e. a noise vector, of the same dimension as the sensed data, whereinIs a diagonal covariance matrix;
using principal component matricesExtracting the characteristics of the environment data to obtain characteristic vectors：
Wherein the noise vector distribution after projection is still unchanged, i.e；
The sensed data assuming true values isA mixture of gaussian-like distributions, the probability density function of the truth-sensing data is expressed as:
；
wherein,is the firstCentroid of classes, and each class carries the same covariance matrix；
Based on the truth-sensing data and the distribution of noise vectors, the distribution of feature vectors extracted by each edge device is derived as
。
Optionally, the step of transmitting the feature vectors of all edge devices to the remote radio unit simultaneously in the same frequency by using an air computing technology to perform uplink feature aggregation on the feature vectors at the remote radio unit, so as to obtain aggregated feature vectors includes:
and simultaneously transmitting the feature vectors of all edge devices to a remote radio unit at the same frequency by utilizing the waveform superposition property of the wireless multiple access channel so as to perform uplink feature aggregation on the feature vectors and obtain aggregated feature vectors.
Optionally, the step of simultaneously transmitting the feature vectors of all edge devices to the remote radio unit at the same frequency by using the waveform superposition property of the wireless multiple access channel to perform uplink feature aggregation on the feature vectors to obtain aggregated feature vectors includes:
transmitting the characteristic vectors of all edge devices to a remote radio unit at the same time and same frequency by utilizing the waveform superposition property of the wireless multiple access channel, and transmitting the characteristic vectors to the remote radio unit at the first frequencyTime slot, atThe aggregated eigenvectors received at the remote units are
；
Wherein,is an edge deviceAnd a remote radio unitChannel coefficients in between, whereinIs the number of antennas on each remote unit,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is an apparatusThe symbol to be transmitted is a symbol,represented in remote radio unitThe additive white gaussian noise received on the upper side,noise power representing gaussian white noise;
during each edge device transmission, the following transmission power constraints have to be met:
wherein,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,representation deviceIs set to be a maximum transmission power limit of (c),is an apparatusSymbols to be transmitted;
Since the variance of the transmitted signal is estimated from the offline data samples, the optimized transmit power constraint is:
furthermore, the energy consumption of each edge device at all time slots obeys an energy constraint term:
；
wherein,representing the total energy constraint of the device,the duration of the aggregation is calculated every time over the air,is to be designed at the edge equipment endThe transmission of the scalar, i.e. the transmission of the pre-code,is an apparatusSymbols to be transmitted.
Optionally, the step of quantizing the aggregated feature vector at the remote radio unit and then sending the quantized aggregated feature vector to the central controller includes:
after each remote radio unit receives the aggregated feature vector elements, the aggregated feature vector elements are quantized and then transmitted to the central controller, and the quantization operation is characterized as
Wherein,representing quantization noise, whereThe diagonal covariance matrix of quantization errors introduced by the independent quantization scheme is used as the quantity to be optimized of the algorithm;
based on the rate distortion theory, in the firstTransmission rate of all remote units to central controller in each time slotSatisfy the following requirements
；
Wherein,is the channel state information of the concatenation,is the quantized covariance matrix of the uplink, Is the transmission scalar to be designed at the edge device side, i.e. send and encode,representing the noise power of the additive white gaussian noise received at the remote radio unit m,representing the maximum transmit power among all edge device transmit power constraints,representing a forward capacity limit between the remote radio unit and the central controller.
Optionally, the step of analyzing the probability density function of the aggregated feature vector and constructing a classification discrimination gain according to the aggregated feature vector received by the central controller from the remote radio unit, identifying the aggregated feature vector based on the constructed classification discrimination gain, obtaining an identification result and making a decision of an inference task includes:
stacking the quantized eigenvectors from all the remote radio units by a central controller, and designing an eigenvector for decoding and aggregating the received beamforming vector;
analyzing the classification probability function of the aggregated feature vector, constructing classification discrimination gain based on the classification probability function, and decoding the aggregated feature vector from the user according to the classification discrimination gain.
Optionally, the central controller stacks the quantized feature vectors from all remote radio units, and designs the feature vectors for decoding and aggregating the received beamforming vectors; analyzing the classification probability function of the aggregated feature vector, constructing a classification discrimination gain based on the classification probability function, and decoding the aggregated feature vector from the user based on the classification discrimination gain, comprising:
Stacking quantized feature vectors from all remote units by a central controller,
；
wherein,is an edge deviceIs used for the channel coefficients of the (c) signal,is the transmission scalar to be designed at the edge device side, i.e. send and encode,represented in remote radio unitThe additive white gaussian noise received on the upper side,representing quantization noise;
the central controller designs the received beam forming vector and takes the real part to decode the aggregated eigenvector, and the estimated eigenvector is
；
Wherein,is in time slotIs used to determine the received beamforming vector of (a),for equivalent uplink noise, beamforming vectors of a given design equivalent uplink noiseThe variance is，Is an edge deviceIs used for the channel coefficients of the (c) signal,the transmission scalar to be designed is the edge equipment end, namely the transmission and coding;
to compensate for channel fading, the transmission scalar of the edge device is encoded as zero-breaking code
；
Wherein,is in time slotIs used to determine the received beamforming vector of (a),is an edge deviceIs used for the channel coefficients of the (c) signal,representing edge devicesTo be optimized;
substituting the estimated feature vector expression to obtain
；
Wherein,representing edge devicesIs used for the transmission signal strength of the (c) signal,representing the equivalent uplink noise and, Is a noise vector;
resolving probability density functions of the aggregated feature vectors:
；
the mean and variance are respectively:
；
furthermore, by environmental categoryClass of environmentBuilding an environment class pair, two class pair elements based on the environment class pairAnd (3) withAnd matching from the feature vector element classification probability functions to obtain element classification functions in one-to-one correspondence, and constructing classification function pairs, wherein the pair-by-pair discrimination gain of each classification function pair is expressed as:
；
applying the mode of the discrimination gain of the constructed class pair to all class pairs and feature subsets of all time slotsThe total discrimination gain is expressed as:
；
the optimization problem of the overall cloud-edge collaborative reasoning can be modeled as:
；
classification discrimination gainAnd solving the corresponding enhancement analytic expression to enhance the classified discrimination gain so as to obtain the optimized discrimination gain.
According to a second aspect of embodiments of the present disclosure, a cloud-edge collaborative intelligent reasoning system based on a cloud wireless access network is provided, including a plurality of edge devices, a plurality of remote radio units, and a central controller, where
The edge equipment is configured to collect environmental data around the edge equipment through various sensors, and the edge equipment utilizes a principal component analysis method to conduct feature extraction on the environmental data to obtain feature vectors for reasoning tasks; the environment data comprise measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data; transmitting the feature vectors of all edge devices to a remote radio unit at the same frequency through an air computing technology so as to perform uplink feature aggregation on the feature vectors at the remote radio unit;
The remote radio unit is configured to receive the aggregated feature vector transmitted by the edge equipment, quantize the aggregated feature vector, and then send the quantized aggregated feature vector to the central controller;
the central controller is configured to analyze the probability density function of the aggregated feature vector according to the aggregated feature vector received from the remote radio unit, construct a classification discrimination gain, identify the aggregated feature vector based on the constructed classification discrimination gain, obtain an identification result and make a decision of reasoning tasks.
The beneficial effects of the embodiment of the specification are as follows:
according to the cloud edge collaborative intelligent reasoning method and system based on the cloud wireless access network, intelligent reasoning tasks are completed by the coordination edge equipment, the remote radio unit and the central processing unit, and reasoning performance is improved based on design criteria of discrimination gain. Specifically, the edge device captures the perceived data samples of real-time noise interference from the environment, extracts noisy local feature vectors that are aggregated at each remote radio unit to suppress perceived noise, and employs air computing techniques to accomplish the downlink reasoning task by allowing each remote radio unit to simultaneously receive local feature vectors from all edge devices on the same resource block, which are quantized and transmitted to the central processor over the forward link for further aggregation. Compared with the prior art, the method and the corresponding system can support connection of mass equipment to access a network, can adaptively allocate resources, simultaneously inhibit coupled perceived noise, distortion caused by air calculation and quantization errors caused by limited capacity of a forward link, and ensure that the most important characteristic elements can be accurately received by a central processing unit by maximizing discrimination gain, thereby improving the reasoning accuracy of the whole system.
The innovation points of the embodiment of the specification comprise:
1. in the present specification, an air computing technology is adopted, so that by allowing each remote radio unit to simultaneously receive local feature vectors from all edge devices on the same resource block, fast and efficient uplink feature aggregation is achieved, which is one of innovation points in the embodiments of the present specification.
2. In the specification, the cloud edge collaborative reasoning system supports mass equipment to perform connection cooperation, improves reasoning performance based on a design criterion of a discrimination gain, and is one of innovation points of the embodiment of the specification.
3. In the present specification, the performance is improved by adopting the transmit precoding (step S122), the receive beamforming (step S144) and the quantization error control (step S130), and by maximizing the discrimination gain, the resources can be adaptively allocated to ensure that the most important feature elements can be accurately received in the central processing unit, thereby improving the overall inference accuracy, and realizing the simultaneous suppression of coupled perceived noise, distortion caused by air computation and quantization error caused by limited capacity of the forward link, which is one of the innovation points of the embodiments of the present specification.
Drawings
In order to more clearly illustrate the embodiments of the present description or the technical solutions in the related art, the drawings that are required to be used in the embodiments or the related technical descriptions will be briefly described, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort for those skilled in the art.
Fig. 1 is a schematic flow chart of a cloud edge collaborative intelligent reasoning method based on a cloud wireless access network according to an embodiment of the present disclosure;
FIGS. 2 and 3 are graphs comparing the performance of the reasoning method provided by an embodiment of the present disclosure with that of an unoptimized baseline algorithm in terms of the change in reasoning performance with the capacity of the forward link under a perceived dataset;
FIGS. 4 and 5 are graphs comparing the performance of the reasoning method provided by one embodiment of the present disclosure with the performance of the unoptimized baseline algorithm in terms of system energy under a perceived dataset;
fig. 6 is a schematic structural diagram of a cloud-edge collaborative intelligent reasoning system based on a cloud wireless access network according to an embodiment of the present disclosure;
fig. 7 is an application schematic diagram of a cloud-edge collaborative intelligent reasoning system based on a cloud wireless access network according to an embodiment of the present disclosure.
Detailed Description
The technical solutions of the embodiments of the present specification will be clearly and completely described below with reference to the drawings of the embodiments of the present specification, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without any inventive effort, are intended to be within the scope of the invention.
It should be noted that the terms "comprising" and "having" and any variations thereof in the embodiments and figures herein are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.
The embodiment of the specification discloses a cloud edge cooperative intelligent reasoning method and a cloud edge cooperative intelligent reasoning system based on a cloud wireless access network, and the method and the system are respectively described in detail below.
The cloud edge collaborative intelligent reasoning system in the embodiment of the specification comprises three execution bodies: the system comprises edge equipment, a remote radio unit and a central controller. The Edge Device (Edge Device) has basic sensing and communication functions, and acquires and processes information of the target environment in real time through the environment physical quantity sensor. A remote radio unit (Remote Radio Head) which is a low cost, low power remote radio head. Unlike conventional base stations, remote radio units have no basic signal coding and decoding functions, and play a role in cloud radio access networks mainly as relays for signal transmission, and transmit features from edge devices to a central controller. A central controller (Central Precessor), which is one of the core components in the cloud radio access network, has high computing and storage capabilities. The central controller is responsible for centrally processing data from multiple remote units and performing some critical network functions including task allocation, data aggregation, decision making reasoning, etc. The central controller has strong computing capacity and storage capacity, and can realize efficient data processing, resource allocation and network management. Through centralized processing and management, the central processor can provide more advanced functions and services, such as intelligent scheduling, dynamic spectrum allocation, network optimization, and the like. At the same time, the central controller may also provide greater security and reliability to protect the data and communication connections of the wireless communication system.
Fig. 1 is a schematic flow chart of a cloud edge collaborative intelligent reasoning method based on a cloud wireless access network according to an embodiment of the present disclosure. As shown in fig. 1, a cloud edge collaborative intelligent reasoning method based on a cloud wireless access network is suitable for being configured on an edge device, a remote radio unit and a central controller and executed, and comprises the following steps:
s110, acquiring environmental data around the edge equipment through various sensors, and performing feature extraction on the environmental data at the edge equipment by using a principal component analysis method to obtain feature vectors for reasoning tasks; the environmental data includes measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data.
And sensing the surrounding environment by adopting edge equipment, and collecting the interested environment information. The interesting environment information is different according to the reasoning task, and can be realized by carrying various sensors on the edge equipment side. The environmental information can be measured values of physical parameters such as temperature, humidity, illumination and the like; may be video or image data capturing visual information in an environment; may be sound or audio data that captures sound signals in the environment; the geographic position information of the user can be acquired by utilizing GPS or other positioning technologies; motion data provided by sensors such as acceleration, angular velocity, etc.; the method can also monitor the network connection state, transmission rate and other network related information; and other environmental related parameters such as humidity, air pressure, wind speed, etc., in this embodiment taking into account the human motion data set perceived by the WiFi signal.
In a specific embodiment, the step S110 of collecting environmental data around the edge device through a plurality of sensors, and performing feature extraction on the environmental data at the edge device by using a principal component analysis method to obtain a feature vector for an inference task includes:
s112, collecting environmental data around the edge equipment through various sensors.
And S114, carrying out feature extraction on the environmental data by using the principal component matrix to obtain feature vectors, and further obtaining the distribution of the feature vectors according to the feature vectors.
The edge device is not limited to extracting the feature vector from the perception data by a specific method, and can be simple physical statistical features such as mean value, median and the like; may be a neural network sub-model.
In this embodiment, the principal component analysis (Principal Components Analysis, PCA) is considered to extract feature vectors, and the idea of dimension reduction is used to convert multiple indices into a few comprehensive indices.
Principal component analysis is a commonly used statistical method for reducing the data dimensions and extracting the main features of the data. It converts the original data into a set of uncorrelated principal components by linear transformation, where the first principal component represents the largest variance in the data, the second principal component represents the second largest variance in the data, and so on. By selecting the first few principal components, we can retain most of the information in the data while significantly reducing the dimensionality of the data, thereby speeding up the subsequent reasoning and decision process.
In the present specification, the reason why the principal component analysis method is selected to extract the feature vector is as follows. First, edge devices are often limited by computational and memory resources, and therefore an efficient method is needed to process large amounts of perceived data. PCA is used as a linear transformation method, has lower computational complexity and storage requirement, and is suitable for being implemented on edge equipment. Second, PCA is able to automatically learn the main features from the data without manually selecting the features, thus being more flexible and intelligent. In addition, PCA can also eliminate redundant information in the original data, retain the most important information in the data, and contribute to improving the accuracy and efficiency of the follow-up reasoning task. By adopting PCA as the feature extraction method, the cloud edge collaborative reasoning system can fully utilize data and realize rapid and accurate reasoning and decision in the environment with limited resources.
The edge device extracts features from the collected environmental information by using an artificial intelligent model deployed at the edge device end, and a classification probability function of the features is obtained.
In specific implementation, the environmental data around the edge equipment is collected through various sensors; extracting features of the environmental data by using the principal component matrix to obtain feature vectors, and further obtaining distribution of the feature vectors according to the feature vectors, wherein the method comprises the following steps:
The environmental data around the edge equipment is collected through various sensors, and the serial number isThe acquired environmental data around the edge device is
；
Wherein,is the sensing data of the true value,is a perceptual distortion, i.e. a noise vector, of the same dimension as the sensed data, whereinIs a diagonal covariance matrix.
Using principal component matricesExtracting the characteristics of the environment data to obtain characteristic vectors：
Wherein the noise vector distribution after projection is still unchanged, i.e。
The sensed data assuming true values isA mixture of gaussian-like distributions, the probability density function of the truth-sensing data is expressed as:
；
wherein,is the firstCentroid of classes, and each class carries the same covariance matrix。
Based on the truth-sensing data and the distribution of noise vectors, the distribution of feature vectors extracted by each edge device is derived as
。
And S120, transmitting the feature vectors of all the edge devices to a remote radio unit at the same frequency through an air computing technology so as to perform uplink feature aggregation on the feature vectors at the remote radio unit and obtain aggregated feature vectors.
Specifically, the step S120 of transmitting the feature vectors of all edge devices to the remote radio unit through the air computing technology at the same time, so as to perform uplink feature aggregation on the feature vectors at the remote radio unit, and the step of obtaining the aggregated feature vectors includes:
S122, utilizing waveform superposition property of the wireless multiple access channels to simultaneously and co-frequency transmit the feature vectors of all edge devices to a remote radio unit so as to perform uplink feature aggregation on the feature vectors and obtain aggregated feature vectors.
Feature air aggregation is realized at the remote radio unit, namely, by utilizing the waveform superposition property of a wireless multiple access channel, all edge devices simultaneously and simultaneously transmit the feature vectors of the edge devices at the same frequency through an air computing technology, so that feature aggregation is realized at the remote radio unit, and precoding of the feature vectors is realized at the remote radio unit.
In a specific embodiment, the step S122 of simultaneously transmitting the feature vectors of all edge devices to the remote radio unit at the same frequency by using the waveform superposition property of the wireless multiple access channel to perform uplink feature aggregation on the feature vectors to obtain aggregated feature vectors includes:
transmitting the characteristic vectors of all edge devices to a remote radio unit at the same time and same frequency by utilizing the waveform superposition property of the wireless multiple access channel, and transmitting the characteristic vectors to the remote radio unit at the first frequencyTime slot, atThe aggregated eigenvectors received at the remote units are
；
Wherein,is an edge device And a remote radio unitChannel coefficients in between, whereinIs the number of antennas on each remote unit,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is an apparatusThe symbol to be transmitted is a symbol,represented in remote radio unitThe additive white gaussian noise received on the upper side,representing the noise power of gaussian white noise.
During each edge device transmission, the following transmission power constraints have to be met:
wherein,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is an apparatusThe symbol to be transmitted is a symbol,representation deviceMaximum transmission power limit of (a).
Since the variance of the transmitted signal is estimated from the offline data samples, the optimized transmit power constraint is:
furthermore, the energy consumption of each edge device at all time slots obeys an energy constraint term:
；
wherein,representing the total energy constraint of the device,the duration of the aggregation is calculated every time over the air,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is an apparatusSymbols to be transmitted.
S130, quantizing the aggregated feature vector at the remote radio unit, and then sending the quantized aggregated feature vector to the central controller.
The feature vectors aggregated at the remote units need to be quantized to meet the link capacity constraint between the remote units and the central controller, and therefore, each remote unit needs to perform quantization after receiving the aggregated intermediate feature vector elements and then transmit the quantized feature vector elements to the central controller.
In a specific embodiment, the step of quantizing the aggregated feature vector at the remote radio unit and then sending the quantized aggregated feature vector to the central controller includes:
after each remote radio unit receives the aggregated feature vector elements, the aggregated feature vector elements are quantized and then transmitted to the central controller, and the quantization operation is characterized as
Wherein,representing quantization noise, whereThe diagonal covariance matrix of quantization errors introduced by the independent quantization scheme is used as the quantity to be optimized of the algorithm;
based on the rate distortion theory, in the firstTransmission rate of all remote units to central controller in each time slotSatisfy the following requirements
；
Wherein,is the channel state information of the concatenation,is the quantized covariance matrix of the uplink,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding, Noise power representing additive white gaussian noise received at remote radio unit m，Representing the maximum transmit power among all edge device transmit power constraints,representing a forward capacity limit between the remote radio unit and the central controller.
In the step, the characteristic vector aggregated at the remote radio unit is quantized to meet the link capacity constraint between the remote radio unit and the central controller, so that the stability of link transmission is ensured, the performance optimization is realized through quantization error control, and the self-adaptive allocation of resources is facilitated.
S140, analyzing probability density functions of the aggregated feature vectors and constructing classification discrimination gains according to the aggregated feature vectors received by the central controller from the remote radio unit, identifying the aggregated feature vectors based on the constructed classification discrimination gains, obtaining identification results and making decisions of reasoning tasks.
Global feature aggregation is implemented at a central controller: the central controller aggregates the feature vectors from all the remote radio units, analyzes the classification probability function of the aggregated feature vectors, and constructs classification discrimination gain to guide the design of the received beam forming vector to decode so as to obtain the aggregated feature vectors from the user.
In a specific embodiment, the step S140 of analyzing a probability density function of the aggregated feature vector and constructing a classification discrimination gain according to the aggregated feature vector received by the central controller from the remote radio unit, and identifying the aggregated feature vector based on the constructed classification discrimination gain to obtain an identification result and making a decision of an inference task includes:
s142, stacking the quantized eigenvectors from all the remote radio units by the central controller, and designing the eigenvectors for decoding and aggregation of the received beamforming vectors.
S144, analyzing the classification probability function of the aggregated feature vector, constructing classification discrimination gain according to the classification probability function, and decoding the aggregated feature vector from the user according to the classification discrimination gain.
In one implementation, the central controller stacks the quantized eigenvectors from all remote radio units, and designs the eigenvectors of the received beamforming vector decoding aggregate; analyzing the classification probability function of the aggregated feature vector, constructing a classification discrimination gain based on the classification probability function, and decoding the aggregated feature vector from the user based on the classification discrimination gain, comprising:
stacking quantized feature vectors from all remote units by a central controller,
；
Wherein,is an edge deviceIs used for the channel coefficients of the (c) signal,is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,represented in remote radio unitThe additive white gaussian noise received on the upper side,representing quantization noise.
The central controller designs the received beam forming vector and takes the real part to decode the aggregated eigenvector, and the estimated eigenvector is
；
Wherein,is in time slotIs used to determine the received beamforming vector of (a),for equivalent uplink noise, beamforming vectors of a given design equivalent uplink noiseThe variance is，Is an edge deviceIs used for the channel coefficients of the (c) signal,is the transmission scalar to be designed at the edge device end, i.e. the transmit precoding.
To compensate for channel fading, the transmission scalar of the edge device is encoded as zero-breaking code
；
Wherein,is in time slotIs used to determine the received beamforming vector of (a),is an edge deviceIs used for the channel coefficients of the (c) signal,representing edge devicesTo be optimized;
substituting the estimated feature vector expression to obtain
；
Wherein,representing edge devicesIs used for the transmission signal strength of the (c) signal,representing the equivalent uplink noise and,is a noise vector;
resolving probability density functions of the aggregated feature vectors:
；
the mean and variance are respectively:
；
Furthermore, by environmental categoryClass of environmentBuilding an environment class pair, two class pair elements based on the environment class pairAnd (3) withAnd matching from the feature vector element classification probability functions to obtain element classification functions in one-to-one correspondence, and constructing classification function pairs, wherein the pair-by-pair discrimination gain of each classification function pair is expressed as:
applying the mode of the discrimination gain of the constructed class pair to all class pairs and feature subsets of all time slotsThe total discrimination gain is expressed as:
；
the optimization problem of the overall cloud-edge collaborative reasoning can be modeled as:
；
classification discrimination gainAnd solving the corresponding enhancement analytic expression to enhance the classified discrimination gain so as to obtain the optimized discrimination gain. By maximizing the discrimination gain, resources are adaptively allocated to ensure that the most important feature elements can be accurately received at the central processing unit, thereby improving overall inference accuracy while suppressing coupled perceived noise, distortion due to air computation, and quantization errors due to limited capacity of the forward link.
In the embodiment, through the design of the cloud edge collaborative reasoning system, collaborative work among the edge equipment, the remote radio unit and the central controller is realized, so that the system can be flexibly deployed and managed in different application scenes, and efficient and real-time data processing and decision support are provided. The flexibility and the expandability of the system architecture enable the system architecture to be suitable for intelligent application in various fields, and provide important technical support for future intelligent development.
Fig. 2 and 3 are graphs comparing the performance of the reasoning method provided in an embodiment of the present disclosure with that of the unoptimized baseline algorithm in terms of the change in reasoning performance with the capacity of the forward link under the perceived data set. In the figure, the higher the forward link capacity between the remote radio unit and the central controller is, the better the reasoning performance effect of the method in the specification is, compared with the baseline algorithm, the design method provided by the embodiment of the specification is obviously superior to the comparison method 1 and the comparison method 2 of the non-optimized baseline algorithm in terms of classification accuracy, the comparison method 1 is an algorithm which is not optimized by a quantization noise matrix, the comparison method 2 is an algorithm which is not beamformed, the algorithm in the embodiment of the specification is obviously superior to the comparison algorithm in terms of reasoning performance, and the feasibility and effectiveness of implementing deployment of the system are verified.
Fig. 4 and 5 are graphs comparing the performance of the reasoning method provided in an embodiment of the present disclosure with the performance of the unoptimized baseline algorithm in terms of system energy under a perceived dataset. The method has the advantages that the method is similar to the effect of the capacity of a forward link, the higher the total allowable energy of the whole system is, the better the effect of reasoning performance is, and the design method provided by the embodiment of the specification is obviously superior to a baseline algorithm in terms of classification accuracy, and the feasibility and effectiveness of implementing deployment of the system are also proved.
These embodiments illustrate the potential of the intelligent edge reasoning system of the embodiments of the present specification to be applied in a variety of intelligent service scenarios, such as autopilot, smart city, telemedicine, etc., which would bring significant improvements and advances to the prior art.
Fig. 6 is a schematic structural diagram of a cloud-edge collaborative intelligent reasoning system based on a cloud wireless access network according to an embodiment of the present disclosure. As shown in fig. 6, a cloud-edge collaborative intelligent reasoning system 600 based on a cloud wireless access network includes a plurality of edge devices 610, a plurality of remote radio units 620 and a central controller 630, wherein
The edge device 610 is configured to collect environmental data around the edge device through various sensors, and perform feature extraction on the environmental data at the edge device by using a principal component analysis method to obtain feature vectors for reasoning tasks; the environment data comprise measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data; and transmitting the feature vectors of all the edge devices to the remote radio unit at the same time in the same frequency through an air computing technology so as to carry out uplink feature aggregation on the feature vectors at the remote radio unit.
The remote radio unit 620 is configured to receive the aggregated feature vector transmitted by the edge device, quantize the aggregated feature vector, and then send the quantized aggregated feature vector to the central controller.
The central controller 630 is configured to parse the probability density function of the aggregated feature vector and construct a classification discrimination gain according to the aggregated feature vector received from the remote radio unit, identify the aggregated feature vector based on the constructed classification discrimination gain, obtain the identification result and make a decision of reasoning task.
Fig. 7 is an application schematic diagram of a cloud-edge collaborative intelligent reasoning system based on a cloud wireless access network according to an embodiment of the present disclosure. As shown in fig. 7, the edge device 610 perceives the environment S1, performs waveform superposition S3 at the remote radio unit 620 via the cloud radio access network S2 to implement feature aggregation, and performs a forward link communication S4 between the remote radio unit and the central processor, and the central processor 630 finally completes the reasoning task, in a specific embodiment, identifies the human activity S5.
In summary, the embodiments of the present disclosure provide a cloud edge collaborative intelligent reasoning method and system based on a cloud wireless access network, which implement efficient intelligent decision service and promote application potential of an edge reasoning system in various intelligent service scenarios.
Those of ordinary skill in the art will appreciate that: the drawing is a schematic diagram of one embodiment and the modules or flows in the drawing are not necessarily required to practice the invention.
Those of ordinary skill in the art will appreciate that: the modules in the apparatus of the embodiments may be distributed in the apparatus of the embodiments according to the description of the embodiments, or may be located in one or more apparatuses different from the present embodiments with corresponding changes. The modules of the above embodiments may be combined into one module, or may be further split into a plurality of sub-modules.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (4)
1. The cloud edge cooperative intelligent reasoning method based on the cloud wireless access network is characterized by being suitable for being configured on edge equipment, a remote radio unit and a central controller to execute, and comprises the following steps:
Collecting environmental data around the edge equipment through various sensors, and carrying out feature extraction on the environmental data at the edge equipment by using a principal component analysis method to obtain feature vectors for reasoning tasks; the environment data comprise measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data;
transmitting the characteristic vectors of all edge devices to a remote radio unit at the same time and same frequency by utilizing the waveform superposition property of the wireless multiple access channel, and transmitting the characteristic vectors to the remote radio unit at the first frequencyTime slot, in->The aggregated eigenvectors received at the remote units are
；
Wherein,is an edge device->And a remote radio unit->Channel coefficient between, wherein->Is the number of antennas on each remote radio unit, ">Is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is a device->Symbol to be transmitted, ">Is shown in remote radio unit->Additive white gaussian noise of upper reception, +.>Noise power representing gaussian white noise;
during each edge device transmission, the following transmission power constraints have to be met:
；
wherein,is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o >Indicating device->Maximum transmission power limit of>Is a device->Symbols to be transmitted;
since the variance of the transmitted signal is estimated from the offline data samples, the optimized transmit power constraint is:；
furthermore, the energy consumption of each edge device at all time slots obeys an energy constraint term:
；
wherein,representing the total energy constraint, +.>Is the duration of each over-the-air calculation aggregation, +.>Is the transmission to be designed at the edge equipment endScalar input, i.e. transmit pre-coding, +.>Is a device->Symbols to be transmitted;
after each remote radio unit receives the aggregated feature vector elements, the aggregated feature vector elements are quantized and then transmitted to the central controller, and the quantization operation is characterized as
；
Wherein,representing quantization noise, wherein->The diagonal covariance matrix of quantization errors introduced by the independent quantization scheme is used as the quantity to be optimized of the algorithm;
based on the rate distortion theory, in the firstTransmission rate of all remote units to central controller in each time slotSatisfy the following requirements
；
Wherein,is concatenated channel state information, +.>Is the quantized covariance matrix of the uplink, < +.>Is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o >Noise power representing additive white gaussian noise received at remote radio unit m,/->Representing the maximum transmission power in all edge device transmission power constraints, +.>Representing a forward capacity limit between the remote radio unit and the central controller;
stacking quantized feature vectors from all remote units by a central controller,
；
wherein,is an edge device->Channel coefficient of>Is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o>Is shown in remote radio unit->Additive white gaussian noise of upper reception, +.>Representing quantization noise to be designed;
the central controller designs the received beam forming vector and takes the real part to decode the aggregated eigenvector, and the estimated eigenvector is
；
Wherein,is in the slot->Is used to determine the received beamforming vector of (a),for equivalent uplink noise, the beamforming vector of a given design is equivalent to the uplink noise +.>The variance is->，/>Is an edge device->Is used for the channel coefficients of the (c) signal,the method is characterized in that the method is a transmission scalar to be designed at an edge device end, namely, a precoding is sent;
to compensate for channel fading, the transmission scalar of the edge device is encoded as zero-breaking code
；
Wherein,is in the slot- >Is a receive beamforming vector of>Is an edge device->Channel coefficient of>Representing edge devices->To be optimized;
substituting the estimated feature vector expression to obtain
；
Wherein,representing edge devices->Transmission signal strength,/,>representing equivalent uplink noise, +.>Is a noise vector;
resolving probability density functions of the aggregated feature vectors:
；
the mean and variance are respectively:
；
furthermore, by environmental categoryAnd environmental category->Building an environment class pair, based on two class pair elements of the environment class pair>And->And matching from the feature vector element classification probability functions to obtain element classification functions in one-to-one correspondence, and constructing classification function pairs, wherein the pair-by-pair discrimination gain of each classification function pair is expressed as:
；
applying the mode of the discrimination gain of the constructed class pair to all class pairs and feature subsets of all time slotsThe total discrimination gain is expressed as:
；
the optimization problem of the overall cloud-edge collaborative reasoning can be modeled as:
；
classification discrimination gainSolving the corresponding enhancement analysis to enhance the classified discrimination gain so as to obtain the optimized discrimination gain;
and identifying the aggregated feature vector based on the constructed classification discrimination gain to obtain an identification result and making a decision of an reasoning task.
2. The method of claim 1, wherein the step of collecting environmental data around the edge device by a plurality of sensors, performing feature extraction on the environmental data at the edge device by using a principal component analysis method, and obtaining feature vectors for the inference task includes:
collecting environmental data around the edge device by a plurality of sensors;
and extracting the characteristics of the environmental data by using the principal component matrix to obtain characteristic vectors, and further obtaining the distribution of the characteristic vectors according to the characteristic vectors.
3. The method of claim 2, wherein the environmental data surrounding the edge device is collected by a plurality of sensors; extracting features of the environmental data by using the principal component matrix to obtain feature vectors, and further obtaining distribution of the feature vectors according to the feature vectors, wherein the method comprises the following steps:
the environmental data around the edge equipment is collected through various sensors, and the serial number isThe acquired environmental data around the edge device is
；
Wherein,sensory data which is true value, +.>Is a perceptual distortion, i.e. a noise vector, of the same dimension as the sensed data, wherein +.>Is a diagonal covariance matrix;
using principal component matrices Extracting the characteristics of the environmental data to obtain a characteristic vector +.>：
；
Wherein the noise vector distribution after projection is still unchanged, i.e；
The sensed data assuming true values isA mixture of gaussian-like distributions, the probability density function of the truth-sensing data is expressed as:
；
wherein,is->Centroid of class and each class carries the same covariance matrix +.>；
Based on the truth-sensing data and the distribution of noise vectors, the distribution of feature vectors extracted by each edge device is derived as
。
4. Cloud edge cooperative intelligent reasoning system based on cloud wireless access network is characterized by comprising a plurality of edge devices, a plurality of remote radio units and a central controller, wherein
The edge equipment is configured to collect environmental data around the edge equipment through various sensors, and the edge equipment utilizes a principal component analysis method to conduct feature extraction on the environmental data to obtain feature vectors for reasoning tasks; the environment data comprise measured values of physical parameters of temperature, humidity and illumination, and the principal component analysis method is used for reducing the dimension of the data and extracting main characteristics of the data;
transmitting the characteristic vectors of all edge devices to a remote radio unit at the same time and same frequency by utilizing the waveform superposition property of the wireless multiple access channel, and transmitting the characteristic vectors to the remote radio unit at the first frequency Time slot, in->The aggregated eigenvectors received at the remote units are
；
Wherein,is an edge device->And a remote radio unit->Channel coefficient between, wherein->Is the number of antennas on each remote radio unit, ">Is the transmission scalar to be designed at the edge device side, i.e. the transmit precoding,is a device->Symbol to be transmitted, ">Is shown in remote radio unit->Additive white gaussian noise of upper reception, +.>Noise power representing gaussian white noise;
during each edge device transmission, the following transmission power constraints have to be met:
；
wherein,is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o>Indicating device->Maximum transmission power limit of>Is a device->Symbols to be transmitted;
since the variance of the transmitted signal is estimated from the offline data samples, the optimized transmit power constraint is:
；
furthermore, the energy consumption of each edge device at all time slots obeys an energy constraint term:
；
wherein,representing the total energy constraint, +.>Is the duration of each over-the-air calculation aggregation, +.>Is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o>Is a device->Symbols to be transmitted;
The remote radio units are configured to execute quantization operation on the aggregated feature vectors after receiving the aggregated feature vector elements, and then transmit the aggregated feature vectors to the central controller, wherein the quantization operation is characterized by
；
Wherein,representing quantization noise, wherein->The diagonal covariance matrix of quantization errors introduced by the independent quantization scheme is used as the quantity to be optimized of the algorithm;
based on the rate distortion theory, in the firstTransmission rate of all remote units to central controller in each time slotSatisfy the following requirements
；
Wherein,is concatenated channel state information, +.>Is the quantized covariance matrix of the uplink, < +.>Is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o>Noise power representing additive white gaussian noise received at remote radio unit m,/->Representing the maximum transmission power in all edge device transmission power constraints, +.>Representing a forward capacity limit between the remote radio unit and the central controller;
a central controller configured to stack quantized feature vectors from all remote units,
；
wherein,is an edge device->Channel coefficient of>Is the transmission scalar to be designed at the edge device side, i.e. transmit precoding,/o >Is shown in remote radio unit->Additive white gaussian noise of upper reception, +.>Representing quantization noise to be designed;
the central controller designs the received beam forming vector and takes the real part to decode the aggregated eigenvector, and the estimated eigenvector is
；
Wherein,is in the slot->Is used to determine the received beamforming vector of (a),for equivalent uplink noise, the beamforming vector of a given design is equivalent to the uplink noise +.>The variance is->，/>Is an edge device->Is used for the channel coefficients of the (c) signal,the method is characterized in that the method is a transmission scalar to be designed at an edge device end, namely, a precoding is sent;
to compensate for channel fading, the transmission scalar of the edge device is encoded as zero-breaking code
；
Wherein,is in the slot->Is a receive beamforming vector of>Is an edge device->Channel coefficient of>Representing edge devices->To be optimized;
substituting the estimated feature vector expression to obtain
；
Wherein,representing edge devices->Transmission signal strength,/,>representing equivalent uplink noise, +.>Is a noise vector;
resolving probability density functions of the aggregated feature vectors:
；
the mean and variance are respectively:
；
furthermore, by environmental categoryAnd environmental category->Building an environment class pair, based on two class pair elements of the environment class pair >And->And matching from the feature vector element classification probability functions to obtain element classification functions in one-to-one correspondence, and constructing classification function pairs, wherein the pair-by-pair discrimination gain of each classification function pair is expressed as:
；
applying the mode of the discrimination gain of the constructed class pair to all class pairs and feature subsets of all time slotsThe total discrimination gain is expressed as:
；
the optimization problem of the overall cloud-edge collaborative reasoning can be modeled as:
；
classification discrimination gainSolving the corresponding enhancement analysis to enhance the classified discrimination gain so as to obtain the optimized discrimination gain; and identifying the aggregated feature vector based on the constructed classification discrimination gain to obtain an identification result and making a decision of an reasoning task.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311188066.3A CN116955996B (en) | 2023-09-15 | 2023-09-15 | Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311188066.3A CN116955996B (en) | 2023-09-15 | 2023-09-15 | Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116955996A CN116955996A (en) | 2023-10-27 |
| CN116955996B true CN116955996B (en) | 2023-12-05 |
Family
ID=88462286
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311188066.3A Active CN116955996B (en) | 2023-09-15 | 2023-09-15 | Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116955996B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108737027A (en) * | 2018-05-09 | 2018-11-02 | 浙江工业大学 | A kind of cloud access is surfed the net no-rate codes number of degrees distribution optimization method |
| CN113301575A (en) * | 2021-05-08 | 2021-08-24 | 三维通信股份有限公司 | Resource allocation method for multiple base stations and radio remote unit |
| WO2022252496A1 (en) * | 2021-06-01 | 2022-12-08 | 苏州路之遥科技股份有限公司 | Software-defined cloud-side collaborative network energy consumption optimisation method and system |
| CN115983390A (en) * | 2022-12-02 | 2023-04-18 | 上海科技大学 | Edge intelligent reasoning method and system based on multi-antenna aerial calculation |
-
2023
- 2023-09-15 CN CN202311188066.3A patent/CN116955996B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108737027A (en) * | 2018-05-09 | 2018-11-02 | 浙江工业大学 | A kind of cloud access is surfed the net no-rate codes number of degrees distribution optimization method |
| CN113301575A (en) * | 2021-05-08 | 2021-08-24 | 三维通信股份有限公司 | Resource allocation method for multiple base stations and radio remote unit |
| WO2022252496A1 (en) * | 2021-06-01 | 2022-12-08 | 苏州路之遥科技股份有限公司 | Software-defined cloud-side collaborative network energy consumption optimisation method and system |
| CN115983390A (en) * | 2022-12-02 | 2023-04-18 | 上海科技大学 | Edge intelligent reasoning method and system based on multi-antenna aerial calculation |
Non-Patent Citations (2)
| Title |
|---|
| A Cloud-Edge-Device Collaborative Inference Framework;Xiang Jia et al.;《IEEE Xplore》;全文 * |
| 面向边缘计算的制造资源感知接入与智能网关技术研究;邹萍 等;计算机集成制造系统(第01期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116955996A (en) | 2023-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113938232A (en) | Communication method and communication device | |
| US9425876B2 (en) | Apparatus and method for channel feedback in multiple input multiple output system | |
| CN114079493A (en) | Channel state information measurement feedback method and related device | |
| TW201806427A (en) | Radio scene recognizing device and method and radio communication apparatus and system | |
| Elbir et al. | Federated learning for physical layer design | |
| EP4262121A1 (en) | Neural network training method and related apparatus | |
| KR102510513B1 (en) | Deep learning based beamforming method and apparatus for the same | |
| CN114219354A (en) | Resource allocation optimization method and system based on federal learning | |
| Ozer et al. | Offloading deep learning powered vision tasks from UAV to 5G edge server with denoising | |
| CN116955996B (en) | Cloud edge collaborative intelligent reasoning method and system based on cloud wireless access network | |
| KR20230091093A (en) | Data-Based Probabilistic Modeling of Wireless Channels Using Conditional Variational Auto-Encoders | |
| Naeem et al. | Efficient scheduling of video camera sensor networks for IoT systems in smart cities | |
| CN113825236A (en) | Method for integrating sensing, computing and communication in wireless network | |
| CN117669699A (en) | Digital twinning-oriented semantic information federation learning method and system in industrial Internet of things scene | |
| WO2022037390A1 (en) | Data transmission method and apparatus | |
| Wen et al. | Task-oriented over-the-air computation for multi-device edge ai | |
| CN116419257A (en) | Communication method and device | |
| Gong et al. | A Scalable Multi-Device Semantic Communication System for Multi-Task Execution | |
| CN116633399B (en) | Semantic communication waveform design method, device, equipment and storage medium | |
| WO2023036323A1 (en) | Communication method and apparatus | |
| CN111914923B (en) | Target distributed identification method based on clustering feature extraction | |
| Rajput et al. | Optimal precoder designs for the estimation of a temporally correlated parameter in OFDM-based WSNs | |
| WO2023179675A1 (en) | Information processing method and communication apparatus | |
| CN117692026B (en) | Link sensing method and device for power line communication | |
| EP4373006A1 (en) | Communication method and apparatus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |