System and method for constructing a backscatter low-power communication network for wide-area communication

Abstract

The present invention belongs to the field of communication technology, and discloses a system and method for constructing a backscattering low-power communication network for wide-area communication. The radio frequency signal carrier detection and identification module uses a lightweight BP neural network accelerator to detect and identify the frame structure of environmental weak wireless carrier signals; the low-power variable throughput modulation communication module uses frequency shift keying modulation based on a code table to perform low-power variable throughput communication of passive nodes; the scattering transmission module performs kilometer-level backscattering transmission of weak radio frequency signals based on a scattering field beamforming method; the environmental energy conversion module performs environmental energy conversion and application based on efficient energy collection and global energy management circuits on metamaterial surfaces, so that passive nodes can work continuously. The present invention ensures the integrity and real-time transmission of data in a passive sensing network, and promotes the development and widespread deployment of wide-area, low-source, and low-cost Internet of Things.

Description

Technical Field

[0001] The present invention belongs to the field of communication technology, and in particular relates to a system and method for constructing a backscattering low-power communication network for wide-area communication. Background Art

[0002] At present, for thousands of years, human beings have been able to continue to multiply and create brilliant civilizations in the process of understanding and changing the world. After humans entered the information age, the Internet of Things has become a new way for humans to understand the world and communicate with nature. The present invention hopes to use hundreds of millions of low-cost micro nodes as "eyes and ears" to perceive everything, promote the deep integration of natural information and human wisdom, and the harmonious development of man and nature.

[0003] The application of massive nodes has laid the foundation for the Internet of Things to perceive the world, but since most networks are powered by batteries, once the power is exhausted and the nodes fail, the network cannot accurately capture environmental information, and the transmission of key information is blocked; and the use of a large number of batteries seriously pollutes the ecological environment on which the present invention depends. According to international forecasting agencies such as Gartner, the number of global Internet of Things connections will reach 25.1 billion in 2025, of which the energy consumption required for smart lighting alone is as high as 7TW / h, and smart lighting is only one of the applications with the lowest energy consumption in the Internet of Things. Such huge energy consumption seriously threatens global energy security and makes mankind face the threat of energy crisis at all times. High energy consumption and dependence on batteries have become bottleneck problems that need to be solved urgently in the development of the Internet of Things.

[0004] The deployment and long-term manual maintenance of massive nodes are difficult and require a large investment, which seriously affects the popularization of IoT applications. IoT based on technologies such as WiFi, Bluetooth, and Zigbee requires a large number of nodes to be deployed for long-distance transmission due to the small coverage of a single node and the limited distance of information transmission. As the network scale increases, energy consumption will also increase exponentially with the increase in the radius of the perceptible range, and its expenditure may even reach 100 billion yuan. Although the long-distance communication IoT based on technologies such as low-power wide-area IoT (LPWAN) has alleviated the difficulties faced by IoT deployment to a certain extent due to its long-distance transmission advantage, it still requires a large amount of long-term manpower investment to ensure the normal operation of its nodes, such as regular replacement of batteries.

[0005] Therefore, exploring and researching "low-power passive Internet of Things implementation solutions" to get rid of the node's dependence on batteries and solve the bottleneck problems of difficult deployment and maintenance and high investment have great practical significance and application value.

[0006] Passive backscatter low-power communication technology (Battery-free Backscatter Communication Technology) uses energy harvesting technology to capture ambient energy, making nodes based on low-power design self-sufficient and free from dependence on traditional batteries. It has become a hot topic pursued by industry and academia in recent years. Since 2005, well-known domestic and foreign research institutions such as Intel Seattle Research Institute, MIT, Tsinghua University, Peking University, and Alibaba Damo Academy have successively intervened in this research, and it has become a hot topic tracked by top conferences and high-quality journals such as SIGCOMM, MOBICOM, and ISSCC in the past five years. SIGCOMM 2020 has opened a special topic for this research direction for paper lectures and discussions. The boom in IoT research continues to rise.

[0007] The essence of passive scattering communication technology is that nodes obtain weak energy from the environment (such as ubiquitous electromagnetic waves, sunlight, temperature differences) to support their perception, computing and wireless communication. The key to its low communication energy consumption lies in: (1) The RF carrier of the information is obtained by the environmental electromagnetic waves captured by the node, or by the small signal excitation of special semiconductor devices. (2) The loading of information is achieved by optimizing the antenna impedance matching coefficient and changing the antenna's reflection or absorption state of the environmental RF signal. Through continuous tracking of this technology, the present invention has found that the existing passive backscattering system cannot be widely deployed and applied in real indoor and outdoor environments, as shown in the following:

[0008] First, the existing passive backscatter communication system has low ability to detect and identify environmental signals, which is specifically manifested as "the detection distance is limited to a hundred meters, and the frame structure of the target signal protocol cannot be accurately identified and parsed"; second, the passive backscatter node communication capability is insufficient, which is specifically manifested as "the backscatter communication radius is limited and the throughput is low". The backscatter communication system using existing low-cost commercial IoT devices as the carrier excitation source can only obtain a kilometer-level backscatter distance when it is within 1m from the excitation source. If the distance from the excitation source is extended to 10m, the backscatter range will drop sharply to tens of meters, which seriously hinders its applicability outdoors and makes it impossible for gateways, especially those in rural areas, to act as both excitation sources and receivers at the same time; third, the collection and conversion efficiency based on traditional antennas and energy harvesting circuits is low (actually about 30% to 50%), and it is difficult to ensure the reliability of data transmission under low power consumption constraints.

[0009] Therefore, the existing passive backscatter systems still face challenges in long distance, high throughput and real-time communication in real complex environments.

[0010] Existing backscatter communication systems (Backscatter) are based on acoustic signals, light and wireless signals to transmit information. The backscatter technology based on acoustic signals loads information onto the ultrasonic signal emitted by the excitation device for transmission, and is used in applications such as acoustic gesture recognition, close-range positioning and liquid recognition. In 2019, MIT's Fadel achieved underwater communication for the first time through a sound-based passive backscatter system, and optimized this work again in 2020. At present, the communication range of the sound-based backscatter system is still tens of meters, which cannot be widely deployed and applied.

[0011] The existing technology's 1-centimeter reflection distance and active power supply method limit the communication range and application scenarios of existing sound backscatter tags, and periodic battery updates also introduce high deployment costs.

[0012] Prior art 2 Passive backscatter communication based on optical signals collects solar energy for power supply and uses optical signals emitted by commercial equipment as excitation source signals. Although it has achieved a large transmission throughput, the ambient light as background noise has a great impact on optical backscatter communication, limiting its wide deployment and application, especially outdoors during the day, and its transmission reliability cannot be guaranteed.

[0013] Prior Art 3 Backscattering technology based on commercial wireless radio frequency signals has flourished in recent years. It uses the signals emitted by existing radio frequency sources such as RFID, Wi-Fi, LoRa signals in the environment as environmental carriers, modulates its own information for communication, and has received great attention from academia and industry.

[0014] The earliest backscattering system, the passive backscattering tag excited by the commercial RFID reader signal, provides a strong research foundation for low-power contactless automatic identification. In recent years, backscattering systems using TV signals, Wi-Fi, Bluetooth and FM signals as energy sources and excitation sources have also been born. Due to their universality and passive low power consumption, they have attracted much attention in short-range backscattering applications.

[0015] In 2017, the team of Shyamnath Golakota at the University of Washington first increased the backscatter transmission distance of passive nodes to the kilometer level. They modulated passive backscatter into signals that can be recognized by the LoRaWan Internet of Things. In the same year, the team of Varshney Ambuj also proposed a kilometer-level long-distance solution, modulating signals that can be recognized by RFID readers. Both use signals generated by dedicated signal generating equipment rather than environmental excitation signals as carriers. Compared with the application of commercial equipment, it is difficult to deploy and use them widely in real scenarios. In 2018, the applicant and the team proposed a passive scattering communication solution based on commercial LoRa signals, and reported and demonstrated it at the SIGCOMM 2018 conference. For the first time, commercial signals were used as excitation sources to extend the backscatter transmission distance of passive nodes to 1.1km. Although this work has achieved kilometer-level transmission of passive backscattering under the stimulation of commercial environment signals, it requires that the passive device must be close to the stimulation source, and the backscattering transmission distance decreases rapidly as the stimulation source moves away; that is, there is a large gap between the "receiving" distance of commercial node information and the "sending" distance of the passive node's own information, and the communication radius is unbalanced, so passive nodes cannot be truly applied to wide-area communications. The research on passive backscatter systems continues to heat up, and passive backscatter systems based on network signals such as LTE Net and OFDMA have sprung up, but the problem of passive backscatter nodes being difficult to apply to wide-area communications still exists. From the perspective of communication radius and deployment topology, the communication radius of passive backscatter systems based on RF signals is compared.

[0016] It can be seen that the backscattering system that uses special equipment signal excitation has high carrier generation cost and high deployment cost; while the passive backscattering system that does not require special equipment and directly uses existing signals in the environment as excitation faces the challenge of uneven communication radius, which means that it must rely on the excitation source node to work and cannot be deployed independently. It cannot solve the problem of exponential increase in the number of nodes and the sharp increase in cost and energy consumption when the scope of IoT application expands. The main reason for this situation is that when away from the excitation source, the environmental carrier signal from commercial equipment becomes very weak. Due to the scarce energy reserves of the passive system, the existing passive backscattering system is difficult to effectively identify and efficiently modulate it, and it is even more impossible to transmit the weak RF carrier signal loaded with passive node information over long distances through scattering.

[0017] The main reason why the existing backscatter communication schemes that use commercial equipment RF signals as carriers are limited in distance is that the communication distance of the excitation signals they choose is limited to hundreds of meters. After the signal is scattered, the energy is greatly attenuated, and the propagation distance is far less than the original excitation signal, and it is even more difficult to exceed a hundred meters.

[0018] Through the above analysis, the problems and defects of the existing technology are: the existing environmentally stimulated passive backscatter communication system has uneven communication radius, limited communication range, and discontinuous energy supply, and cannot be truly applied to wide area communication.

[0019] The present invention can solve the following defects of the prior art:

[0020] (1) Commonly used carrier sources are divided into dedicated equipment and commercial equipment that has been deployed in the environment. The cost of generating carriers for dedicated equipment is high, the utilization rate of network resources is low, and the advantages of backscattering technology cannot be brought into play. Using the signals of commercial equipment that have been deployed in the environment as carrier sources, due to the time-varying nature of commercial signals and the quota of transmission power, the existing low-power environmental excitation signal detection method has low sensitivity and the detection distance is limited to 100 meters. The node deployment cost and energy consumption are huge. The environmental excitation signal (such as LoRa Chirp) emitted by commercial equipment is a time-varying signal. Its three characteristic elements of intensity, phase and frequency are interrelated and change continuously over time. This makes it impossible for the existing environmental excitation signal detection method that only targets one signal feature such as intensity to identify the LoRa frame structure information; and when the carrier signal weakens as the detection distance increases, the detection difficulty doubles. Commercial LoRa devices first use high-precision ADC to sample the incoming signal of the LoRa RF front end, and then use complex signal conjugate multiplication and FFT transformation algorithm to finally realize packet detection. The complex operation makes commercial packet detection impossible to be applied in passive backscattering systems under extremely strong low power constraints.

[0021] (2) Existing passive nodes for ambient energy harvesting mostly use switch-based amplitude shift keying (OOK) modulation and coarse-grained frequency shift keying (FSK) modulation. The former is directly dependent on signal strength and has low anti-interference ability, making it impossible to transmit weak backscattered signals over long distances; the latter has strong anti-interference ability, but it is difficult to break through the current situation of single transmission rate and low throughput caused by insufficient energy of passive nodes, and it cannot adapt to the burst transmission of the network.

[0022] (3) The excitation signals of the existing backscattering systems based on commercial RF signals can be divided into general network signals (such as WiFi, Bluetooth, ZigBee, etc.) and. The former type of signal is used as the excitation source for backscattering, and the transmission distance is mostly limited to hundreds of meters; based on LPWAN signals for backscattering transmission, the transmission distance can reach kilometers when the distance between the excitation source and the backscattering node is very close, but once the distance between the excitation source and the backscattering node is expanded to tens of meters, the transmission distance will quickly drop to within hundreds of meters.

[0023] (4) Although existing energy collection modules can collect multiple energies, the collection and conversion efficiency still cannot meet the real-time operation of the nodes. In addition, passive nodes lack optimized energy management modules, which often causes ongoing work to be interrupted due to insufficient energy supply. Summary of the invention

[0024] In view of the problems existing in the prior art, the present invention provides a system for constructing a backscattering low-power communication network for wide-area communication.

[0025] The present invention is implemented in this way: a system for constructing a backscattering low-power communication network for wide area communication, the system for constructing a backscattering low-power communication network for wide area communication comprising:

[0026] Radio frequency signal carrier detection and identification module, used to detect and identify frame structures of weak wireless carrier signals in the environment using a lightweight BP neural network accelerator;

[0027] A low-power variable throughput modulation communication module for low-power variable throughput communication of passive nodes using code table-based frequency shift keying modulation;

[0028] Scattering transmission module, used for kilometer-level backscattering transmission of weak RF signals based on the scattered field beamforming method;

[0029] The environmental energy conversion module is used to convert and apply environmental energy based on efficient energy collection and global energy management circuits on the metamaterial surface, allowing passive nodes to work continuously.

[0030] Further, the radio frequency signal carrier detection and identification module includes:

[0031] A radio frequency signal acquisition unit, used for collecting and converting radio frequency signals;

[0032] The sample library building unit is used to collect LoRa signals in real environments at different distances outdoors and build a sample library;

[0033] Dataset generation unit, used to build and enhance datasets;

[0034] A model building unit, which builds a multi-layer perceptron neural network model as a training network and trains the built training network;

[0035] The accelerator construction unit is used to build a sample library and train a training network, and obtain a lightweight neural network accelerator using an FPGA chip;

[0036] The test unit is used to perform simulation analysis and testing on the obtained lightweight neural network accelerator.

[0037] Furthermore, the radio frequency signal acquisition unit includes:

[0038] A collection subunit, used for collecting radio frequency signals;

[0039] An impedance matching subunit, used for impedance matching of the collected radio frequency signal;

[0040] A filter subunit, used for filtering the signal using a passive bandpass filter;

[0041] The conversion subunit is used to convert the filtered radio frequency signal into a digital signal by using a driver, a filter and an analog-to-digital converter.

[0042] Further, the data set generating unit includes:

[0043] The meta-dataset construction sub-unit is used to identify and classify the collected samples using a standard commercial LoRa packet recognition algorithm, taking 1 / 4 of the LoRa Chirp signal as a sample, distinguishing the real LoRa Chirp and forming a data set; then the sampling function module in MATLAB is used to process the LoRa Chirp data set and the non-LoRa Chirp data set respectively, converting them into string form, adding labels for data calibration, converting the primary samples into positive and negative sample sets, and obtaining a small training set of about 1000 entries, namely the meta-training set;

[0044] The enhancement subunit is used to collect different types of wireless signals, work with LoRa2 at the same frequency and different link distances, mix the interference signal with the LoRa signal and environmental noise, generate synthetic data, and resample the synthetic data using a packet detection circuit, use the resampled data together with the synthetic data as a data set, and divide the data set into a training set and a test set.

[0045] Furthermore, the lightweight neural network accelerator includes a data input part, a serial-to-parallel converter, a normalization part, a first layer of neurons, an activation part, a second layer of neurons and a parallel-to-serial converter; used for converting digital signals into modulated signals;

[0046] The data input part is composed of an ADC and a cascaded D flip-flop group, which is used to convert analog signals into digital signals using the ADC, and use the D flip-flops to buffer data between different clock domains to eliminate metastable states during transmission;

[0047] A serial-to-parallel converter is used to convert the 12-bit serial signal output by the ADC into a set of parallel signals;

[0048] The normalization part is used to normalize the voltage value in a small range to the voltage value in a large range to enhance the signal strength;

[0049] The first layer of neurons is used to calculate the normalized output data. When resetting, each weight is assigned a value. After each calculation, the count is automatically increased by 1. When the count reaches 128 times, the calculation result is output.

[0050] The activation part is used to use the ReLU activation function;

[0051] The second layer of neurons is used to input weights during reset, multiply the calculation results of the first layer of neurons by the weights, and add them to the bias;

[0052] The parallel-to-serial converter is used to convert the 45-bit data output by the secondary neuron into serial data output.

[0053] Further, the low power consumption variable throughput modulation communication module includes:

[0054] An intermediate frequency signal synthesis unit, used for performing frequency synthesis using a low-power intermediate frequency signal generator based on wavetable query, i.e., a wavetable synthesizer;

[0055] The signal modulation unit is used to load the collected data using FSK modulation, take the original carrier LoRa Chirp as the reference signal, and perform different frequency shift modulation on the reference signal to obtain different data.

[0056] Further, the scattering transmission module includes:

[0057] Scattering field antenna array construction unit; used to determine the antenna array optimization algorithm and obtain the optimal nine-element antenna array model;

[0058] The beamforming unit is used to expand the 2D antenna array into 3D space and supports multiple beamforming angles.

[0059] Furthermore, the scattered field antenna array construction unit includes:

[0060] The antenna array optimization algorithm selects sub-units to optimize the spacing in the antenna array using the algorithm;

[0061] The spacing calculation subunit is used to establish a scattering communication system antenna array model using HFSS, and set the element antenna spacing of the scattering communication system antenna array model as a changeable variable; and is also used to obtain the optimal element antenna spacing by using MATLAB through a joint calculation using a differential evolution algorithm and HFSS;

[0062] The optimization subunit is used to iteratively determine the optimal nine-element antenna array model in HFSS;

[0063] Furthermore, the environmental energy conversion module includes:

[0064] An energy harvesting unit for harvesting energy using a metamaterial efficient energy harvesting surface based on a complementary open resonant ring of an electromagnetic metamaterial unit;

[0065] Energy optimization distribution unit is used for energy storage, release and optimization management.

[0066] Furthermore, the energy optimization distribution unit includes:

[0067] An energy storage and release subunit, for storing and releasing energy using a multi-capacitor storage structure;

[0068] A global energy management subunit, used for performing global energy management based on a finite state machine using a global energy management circuit;

[0069] Furthermore, the global energy management circuit is composed of three logic units: a charger, an energy controller and a voltage output device;

[0070] chargers to harvest energy from ambient RF signals and solar energy;

[0071] an energy controller for managing a capacitor for powering a passive backscatter tag;

[0072] Voltage output module, used to regulate the stored power to a constant voltage.

[0073] The present invention also provides a method for constructing a backscattering low-power communication network for wide area communication, comprising:

[0074] Use lightweight BP neural network accelerator to detect weak wireless carrier signals and identify frame structures;

[0075] Low-power variable-throughput communication for passive nodes using code table-based frequency-shift keying modulation;

[0076] Kilometer-level backscatter transmission of weak RF signals based on scattered field beamforming method;

[0077] Based on efficient energy collection on the metamaterial surface and global energy management circuit, environmental energy is converted and applied to enable passive nodes to work continuously.

[0078] In combination with all the above-mentioned technical solutions, the advantages and positive effects of the present invention are as follows: In view of the problem that the existing environmentally stimulated passive backscatter communication system cannot be truly applied to wide-area communication due to uneven communication radius, limited communication range, and discontinuous energy supply, the present invention starts from four aspects: high-sensitivity detection and identification of environmental signals, extremely low power modulation of passive nodes, long-distance backscatter transmission of weak signals, and efficient conversion and utilization of environmental energy. In conjunction with experimental verification and deployment testing based on real environments, the present invention explores a method for implementing a system for constructing a backscatter low-power communication network for wide-area communication, ensures the integrity and real-time transmission of data in a passive sensing network, and promotes the development and widespread deployment of a wide-area, low-source, and low-cost Internet of Things.

[0079] The present invention introduces the idea of ​​machine learning to detect weak RF signals in the environment, establishes a target carrier feature library based on a real indoor and outdoor RF signal sample library, and trains to obtain a lightweight BP neural network architecture and network parameters with low computational complexity and burden; then, by designing an ultra-low power consumption neural network accelerator model, a neural network accelerator that can be used for passive systems is realized, which effectively identifies and captures the target modulated carrier signal emitted by an excitation source hundreds of meters away from a real space environment where multiple RF signals coexist, and realizes accurate signal symbol alignment modulation based on a data frame structure. While extending the detection distance of passive backscatter nodes to more than 100 meters (signals emitted by commercial excitation sources hundreds of meters away can be detected), the modulated information is guaranteed to be compatible with commercial networks (such as LoRaWAN).

[0080] The present invention introduces a design method for a fine-grained multi-frequency intermediate frequency signal generator based on a wave table synthesis algorithm. By storing a sine wave signal with a low sampling period on a passive backscattering node, and obtaining a reasonable sampling step length through basic digital operations, and then sampling the stored sine wave signal according to the step length, a specific frequency shift signal for FSK modulation is obtained by fitting. The generated frequency shift signal, in conjunction with a fine-grained FSK modulation and channel conversion method based on spatial signal stitching, improves the communication rate of the passive backscattering node while making it have a certain degree of anti-interference, and finally proposes a design method for a low-power passive backscattering communication system with variable throughput.

[0081] The present invention proposes that the scattering transmission distance of the backscattering system can be enhanced by using scattering field antenna array and beamforming technology. By introducing the scattering field antenna array theory, a scattering field antenna array EDA design platform based on MATLAB and HFSS joint simulation is built, and combined with the particle swarm optimization algorithm, a scattering field antenna array that can support multi-angle beamforming in three-dimensional space is obtained. Ultimately, the backscattering communication node can still transmit the information it perceives over an ultra-long distance with optional transmission direction when it is far away from the excitation source, and the transmission has a certain spatial anti-interference ability thanks to the beamforming technology.

[0082] The present invention introduces the improvement of energy collection efficiency based on metamaterial surface and energy optimization management based on heterogeneous multi-capacitor structure. The unique physical properties of metamaterials are used to improve the collection efficiency of electromagnetic wave energy by metamaterial periodic surface, and at the same time, multi-element energy collection is used to realize the open source energy collection of passive backscattering nodes; based on the energy storage and supply scheme of heterogeneous multi-capacitors, the difference in charging and discharging time between heterogeneous capacitors is used, and the global energy consumption management module based on the state machine is used to reduce the system power consumption, so that the passive LoRa backscattering system can work on demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1It is a schematic diagram of the structure of a system for constructing a backscattering low-power communication network for wide-area communication provided by an embodiment of the present invention;

[0084] In the figure: 1. RF signal carrier detection and identification module; 2. Low-power variable throughput modulation communication module; 3. Scattering transmission module; 4. Environmental energy conversion module.

[0085] Figure 2 It is a system implementation and deployment diagram provided by an embodiment of the present invention.

[0086] Figure 3 It is a block diagram of an environmental carrier identification system based on a lightweight neural network accelerator provided in an embodiment of the present invention.

[0087] Figure 4 It is a circuit diagram of a radio frequency signal acquisition module provided by an embodiment of the present invention.

[0088] Figure 5 It is a schematic diagram that the LoRa preamble provided by an embodiment of the present invention is composed of ten identical Chirp signals.

[0089] Figure 6 It is a diagram of the number of neural network layers provided by an embodiment of the present invention.

[0090] Figure 7 It is a block diagram of a neural network accelerator based on a low-power FPGA chip provided in an embodiment of the present invention.

[0091] Figure 8 It is a schematic diagram of LoRa Chirp signals "0" and "1" based on CSS modulation provided by an embodiment of the present invention.

[0092] Fig. 9 It is a schematic diagram of generating a step length provided by an embodiment of the present invention.

[0093] Fig.10 It is a schematic diagram of a wavetable synthesizer for generating a desired shifted frequency provided by an embodiment of the present invention.

[0094] Fig.11 This is a chirp signal coding diagram based on stitching technology provided by an embodiment of the present invention.

[0095] Fig.12 It is a schematic diagram of scattering analysis provided by an embodiment of the present invention.

[0096] Fig.13 It is a flow chart of establishing the HFSS joint simulation-EDA platform provided by an embodiment of the present invention.

[0097] Fig.14 It is a schematic diagram comparing the gain of the nine-array antenna model provided in an embodiment of the present invention with that of a single antenna and a nine-element antenna.

[0098] Fig.15 It is a schematic diagram of an antenna array obtained with different antenna angles provided in an embodiment of the present invention.

[0099] Fig.16 It is a schematic diagram of beam pattern simulation of a 3D antenna array provided in an embodiment of the present invention.

[0100] Fig.17 It is a schematic diagram of an SRR energy harvesting array and a CSRR energy harvesting array provided in an embodiment of the present invention.

[0101] Fig.18 It is a schematic diagram of a CRSS unit and a CSRR array provided in an embodiment of the present invention.

[0102] Fig.19 It is a schematic diagram comparing the charging and discharging of a single capacitor and a dual capacitor provided in an embodiment of the present invention.

[0103] Fig. 20 Schematic diagram of an energy management circuit provided by an embodiment of the present invention.

[0104] Fig.21 It is a channel topology diagram for information transmission between a gateway and a LoRa tag provided in an embodiment of the present invention.

[0105] Fig. 22 It is a schematic diagram of a multi-node deployment topology provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0106] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0107] In view of the problems existing in the prior art, the present invention provides a system for constructing a backscattering low-power communication network for wide area communication. The present invention is described in detail below in conjunction with the accompanying drawings.

[0108] like Figure 1 As shown, the system for constructing a backscattering low-power communication network for wide area communication provided by an embodiment of the present invention includes:

[0109] The radio frequency signal carrier detection and identification module 1 is used to detect and identify the frame structure of the weak wireless carrier signal in the environment using a lightweight BP neural network accelerator;

[0110] A low-power variable throughput modulation communication module 2, used for low-power variable throughput communication of passive nodes using code table-based frequency shift keying modulation;

[0111] Scattering transmission module 3, used for performing kilometer-level backscattering transmission of weak radio frequency signals based on a scattering field beamforming method;

[0112] The environmental energy conversion module 4 is used to convert and apply environmental energy based on efficient energy collection and global energy management circuits on the metamaterial surface, so that passive nodes can work continuously.

[0113] The radio frequency signal carrier detection and identification module provided in the embodiment of the present invention includes:

[0114] A radio frequency signal acquisition unit, used for collecting and converting radio frequency signals;

[0115] The sample library building unit is used to collect LoRa signals in real environments at different distances outdoors and build a sample library;

[0116] Dataset generation unit, used to build and enhance datasets;

[0117] A model building unit, which builds a multi-layer perceptron neural network model as a training network and trains the built training network;

[0118] The accelerator construction unit is used to build a sample library and train a training network, and obtain a lightweight neural network accelerator using an FPGA chip;

[0119] The test unit is used to perform simulation analysis and testing on the obtained lightweight neural network accelerator.

[0120] The radio frequency signal acquisition unit provided in the embodiment of the present invention includes:

[0121] A collection subunit, used for collecting radio frequency signals;

[0122] An impedance matching subunit, used for impedance matching of the collected radio frequency signal;

[0123] A filter subunit, used for filtering the signal using a passive bandpass filter;

[0124] The conversion subunit is used to convert the filtered radio frequency signal into a digital signal by using a driver, a filter and an analog-to-digital converter.

[0125] The data set generation unit provided in the embodiment of the present invention includes:

[0126] The meta-dataset construction sub-unit is used to identify and classify the collected samples using a standard commercial LoRa packet recognition algorithm, taking 1 / 4 of the LoRa Chirp signal as a sample, distinguishing the real LoRa Chirp and forming a data set; then the sampling function module in MATLAB is used to process the LoRa Chirp data set and the non-LoRa Chirp data set respectively, converting them into string form, adding labels for data calibration, converting the primary samples into positive and negative sample sets, and obtaining a small training set of about 1000 entries, namely the meta-training set;

[0127] The enhancement subunit is used to collect different types of wireless signals, work with LoRa2 at the same frequency and different link distances, mix the interference signal with the LoRa signal and environmental noise, generate synthetic data, and resample the synthetic data using a packet detection circuit, use the resampled data together with the synthetic data as a data set, and divide the data set into a training set and a test set.

[0128] The lightweight neural network accelerator provided by the embodiment of the present invention includes a data input part, a serial-to-parallel converter, a normalization part, a first layer of neurons, an activation part, a second layer of neurons and a parallel-to-serial converter; and is used to convert a digital signal into a modulated signal;

[0129] The data input part is composed of an ADC and a cascaded D flip-flop group, which is used to convert analog signals into digital signals using the ADC, and use the D flip-flops to buffer data between different clock domains to eliminate metastable states during transmission;

[0130] A serial-to-parallel converter is used to convert the 12-bit serial signal output by the ADC into a set of parallel signals;

[0131] The normalization part is used to normalize the voltage value in a small range to the voltage value in a large range to enhance the signal strength;

[0132] The first layer of neurons is used to calculate the normalized output data. When resetting, each weight is assigned a value. After each calculation, the count is automatically increased by 1. When the count reaches 128 times, the calculation result is output.

[0133] The activation part is used to use the ReLU activation function;

[0134] The second layer of neurons is used to input weights during reset, multiply the calculation results of the first layer of neurons by the weights, and add them to the bias;

[0135] The parallel-to-serial converter is used to convert the 45-bit data output by the secondary neuron into serial data output.

[0136] The low-power consumption variable throughput modulation communication module provided by the embodiment of the present invention includes:

[0137] An intermediate frequency signal synthesis unit, used for performing frequency synthesis using a low-power intermediate frequency signal generator based on wavetable query, i.e., a wavetable synthesizer;

[0138] The signal modulation unit is used to load the collected data using FSK modulation, take the original carrier LoRa Chirp as the reference signal, and perform different frequency shift modulation on the reference signal to obtain different data.

[0139] The scattering transmission module provided in the embodiment of the present invention includes:

[0140] Scattering field antenna array construction unit; used to determine the antenna array optimization algorithm and obtain the optimal nine-element antenna array model;

[0141] The beamforming unit is used to expand the 2D antenna array into 3D space and supports multiple beamforming angles.

[0142] The scattered field antenna array construction unit provided in the embodiment of the present invention includes:

[0143] The antenna array optimization algorithm selects sub-units to optimize the spacing in the antenna array using the algorithm;

[0144] The spacing calculation subunit is used to establish a scattering communication system antenna array model using HFSS, and set the element antenna spacing of the scattering communication system antenna array model as a changeable variable; and is also used to obtain the optimal element antenna spacing by using MATLAB through a joint calculation using a differential evolution algorithm and HFSS;

[0145] The optimization subunit is used to iteratively determine the optimal nine-element antenna array model in HFSS;

[0146] The environmental energy conversion module provided by the embodiment of the present invention includes:

[0147] An energy harvesting unit for harvesting energy using a metamaterial efficient energy harvesting surface based on a complementary open resonant ring of an electromagnetic metamaterial unit;

[0148] Energy optimization distribution unit is used for energy storage, release and optimization management.

[0149] The energy optimization distribution unit provided in the embodiment of the present invention includes:

[0150] An energy storage and release subunit, for storing and releasing energy using a multi-capacitor storage structure;

[0151] A global energy management subunit, used for performing global energy management based on a finite state machine using a global energy management circuit;

[0152] The global energy management circuit provided in the embodiment of the present invention is composed of three logic units: a charger, an energy controller and a voltage output device;

[0153] chargers to harvest energy from ambient RF signals and solar energy;

[0154] an energy controller for managing a capacitor for powering a passive backscatter tag;

[0155] Voltage output module, used to regulate the stored power to a constant voltage.

[0156] The technical solution of the present invention is further described below in conjunction with specific embodiments.

[0157] Embodiment 1:

[0158] 1. The complete technical solution provided by the present invention (inventive solution)

[0159] The present invention realizes the detection and frame structure recognition of environmental weak wireless carrier signals by designing a lightweight BP neural network accelerator and its related low-power radio frequency modules; realizes low-power variable throughput communication of passive nodes by frequency shift keying modulation based on a code table; realizes kilometer-level backscatter transmission of weak radio frequency signals by scattering field beamforming technology; effectively transforms and applies environmental energy by efficient energy collection on the metamaterial surface and global energy management circuit, and realizes continuous operation of passive nodes; finally, the proposed passive backscatter communication theory and design method are verified, and the realized software and hardware systems are deployed in real indoor and outdoor environments to evaluate and test the degree to which the system achieves the goals.

[0160] The present invention establishes a passive backscatter wide-area communication system for low-cost commercial IoT devices such as LoRa. Four aspects of research are required: research on environmental carrier detection frame structure recognition based on lightweight neural network accelerators, low-power multi-granularity frequency shift keying modulation technology, long-distance backscatter transmission based on scattered field beamforming, efficient energy collection metamaterial surfaces and global energy consumption optimization management methods, as well as system-level verification and testing methods in real environments and typical scenarios. The first four items are necessary technologies for establishing a passive backscatter communication system for wide-area communications, and the fifth item is an important means to verify the effectiveness of the first four studies, which provides a guarantee for the application and promotion of the entire system in a real environment. Figure 2 This is the overall framework diagram of the system, including impedance matching, energy harvesting metamaterial surface, energy management, ADC, and modulation modules.

[0161] 1.1 Research on environmental carrier detection and frame structure recognition methods based on lightweight neural network accelerator

[0162] The RF signal carrier detection and identification module based on machine learning includes: RF signal acquisition module, carrier sample library and neural network design module of LoRa commercial signal source, and low-power BP neural network module based on FPGA platform. The RF signal enters from the antenna, passes through the RF front-end detection module, is converted into a digital signal and enters the BP neural network module for data processing, and is converted into a serial output after network discrimination by FPGA and enters the signal modulation part. Figure 3 Block diagram of the environmental carrier identification system based on a lightweight neural network accelerator.

[0163] (1) Design of RF signal acquisition module.

[0164] Step 1 Impedance matching and filtering design

[0165] In the air, after the signal passes through the antenna, it needs to pass through the impedance matching circuit, the purpose of which is to optimize the impedance matching coefficient to reduce the loss of radio energy and enable the back-end circuit to obtain greater transmission power. The signal is filtered by a passive bandpass filter, and the ADC signal acquisition circuit will obtain a signal with a pure frequency.

[0166] The specific circuit design diagram is shown in Figure 4. The antenna that receives LoRa signals and RFID signals is usually implemented with an antenna array (here, the 915M frequency band). The impedance matching module is designed by obtaining the Smith original image through ADS simulation. The ambient RF LoRa signal enters the RF front end through the corresponding frequency band antenna, flows through the impedance matching network (LC circuit implementation) that matches its frequency, and then passes to the back-end ADC circuit through the filtering design.

[0167] Step 2 ADC Design

[0168] The FPGA-based neural network accelerator can only perform digital operations. The present invention needs to convert the RF signal into an analog signal through impedance matching and filtering, and then convert it into a digital signal through ADC sampling, and perform signal feature comparison and calculation to determine whether the RF signal in the environment is the carrier of the target LoRa Chirp, so as to achieve detection. Due to the strong energy constraints of the passive system, the present invention needs to use a low-power and high-precision ADC for sampling. According to research, ADS7042 is a 12-bit, 1MSPS analog-to-digital converter with a small size and extremely low power consumption, which meets the energy consumption constraints of the ADC module in the passive node system.

[0169] The ADC signal acquisition and digital-to-analog conversion circuit includes a driver, a filter, and an analog-to-digital converter. The impedance-matched and filtered signal is sent to the driver OPA835, then passes through the high-pass filter circuit, and finally sent to the analog-to-digital converter ADS7042 to output 12-bit valid data. The specific circuit design is as follows Figure 4 shown.

[0170] (2) Establishment of carrier sample library and neural network design based on Lo Ra commercial signal source.

[0171] Traditional signal detection technologies mainly include energy detection and matched filtering detection. Energy detection is a detection method based on signal energy. Although this method is easy to implement, it has many defects: 1. The detection effect is greatly affected by environmental noise; 2. It cannot distinguish different types of signals. The matched filtering method is to add a matched filter with prior knowledge to maximize the input signal-to-noise ratio, thereby improving the detection effect and reducing the detection time. Its main defect is that it takes a certain amount of time to obtain the prior knowledge of the signal to be tested, and it is greatly affected by the prior knowledge. In addition, the matched filtering algorithm is complex and consumes a lot of power, so it is not suitable for low-power systems.

[0172] According to the universal approximation theorem, a wireless signal can be described as a complex signal consisting of a string of real and imaginary parts, which can be defined as a function F(x)∈C. According to the universal approximation theorem of the feedforward neural network, there should be a function G(x)∈C that can approximate any F(x) such that |F(x)-G(x)|<ε, and for any x∈C, ε>0. Therefore, if the present invention can find an appropriate mapping function and establish a rich sample library, the deep learning network can become G(x) by learning the features of signals based on any wireless protocol. These features carry unique typical information of the sample signal, such as frequency, time, and intensity. Therefore, the present invention is expected to use machine learning methods to identify target wireless signals.

[0173] The present invention intends to use a radio frequency signal detection method and system based on machine learning (ML) to achieve a more accurate signal detection function while using fewer hardware modules. The method and system use the voltage of the signal to be detected and the fluctuation of the signal as the input of the deep neural network. The device does not need to obtain additional instantaneous features and stable periodic features, and data processing is simple and easy.

[0174] Step 1: Establishment of sample library and generation of training set

[0175] After determining the communication environment of the signal to be detected (LoRa chirp), including collecting the frequency band (907-927MHz), bandwidth, modulation mode, transmitter and receiver models, communication channel status information, and noise conditions of the detection signal, the present invention intends to use the commercial Dragino LoRa gateway (equipped with SX1276 LoRa transceiver) as a transmitter and the software-defined radio USRP 2954 as a receiver to collect LoRa signals in real environments (different weather conditions, different temperatures, different scenes, and different strengths) as samples at different distances outdoors.

[0176] Generation of Meta-Training Set

[0177] Model training is the process of learning a reasonable approximation of the unknown potential mapping function from input to output. The training set is a data set used to train the parameters within the model; large amounts of data training usually result in good approximations. However, in practice, it is difficult to generate a wide range of training sets because enumerating all kinds of channel conditions (e.g., different distances, interferences, noise floors, etc.) requires intensive field data collection and labeling. The present invention first constructs a small training set (called a meta-training set) to reduce the workload of model training, and facilitates changing model parameters and adjusting model structure when the training effect is not good.

[0178] The present invention intends to use 1 / 4 of the LoRa Chirp signal as a sample. The reason is that the selected ADC operating frequency is 1MHz, and each LoRa chirp generates 102,000 samples. For a passive system, it is difficult to calculate all samples as input data. The LoRa signal preamble consists of 10 identical upchirps with an initial frequency offset of zero. Therefore, the present invention focuses on the 1 / 4 of each LoRa chirp and subsamples the first 1 / 4 of each LoRa chirp at different rates. Through the analysis of LoRa modulation theory and data packet structure, and a large number of early monitoring experiments and theoretical analysis of standard LoRa packets based on the USRP system, the real LoRa data packet structure is obtained as follows Figure 5 shown.

[0179] The present invention intends to use a standard commercial LoRa packet recognition algorithm to identify and classify the collected information, distinguish the real LoRa Chirp, and form a data set. Then use the sampling function module in MATLAB to process the LoRaChirp data set and the non-LoRa Chirp data set separately, convert them into string form, add labels for data calibration, so as to convert the primary samples into positive and negative sample sets, and obtain a small training set of about 1000 items, which is called the meta-training set.

[0180] Enhanced training set

[0181] As can be seen from the above, the present invention first establishes a meta-training set, but the amount of data is too small to achieve a better approximation. Therefore, the training set can be enhanced by using the meta-training set data expansion method. Through data expansion, the present invention can list various channel conditions and establish a huge high-fidelity training set. This synthetic training set can be automatically labeled because the basic facts are known a priori.

[0182] In order to enhance the meta-training set, the present invention first collects different types of wireless signals and uses lora2 to work at the same frequency and different link distances. Interference signals (UHF RFID, Sigfox, GSM, and narrowband IoT) are mixed with LoRa signals and environmental noise to generate synthetic data. In order to reduce the domain gap caused by different hardware platforms (USRP and passive backscatter communication systems use different ADCs and impedance matching circuits), the present invention intends to resample these synthetic data using packet detection circuits. The resampled data is calculated as a data set together with the synthetic data. After data enhancement, a large data set containing 10,000 positive examples and 10,000 negative examples will be obtained, which are further divided equally into training sets and test sets.

[0183] Step 2: Build a neural network architecture and establish a model

[0184] Commonly used neural networks mainly include: BP neural network, convolutional neural network (CNN), recurrent neural network (RNN). Among them, BP (Back Propagation) neural network is a hierarchical neural network composed of input layer, middle layer and output layer. Each neuron between adjacent layers is fully connected, and there is no connection between neurons in each layer. When a pair of learning patterns is provided to the network, each neuron obtains the input response of the network to generate a connection weight (Weight). Then, in the direction of reducing the error between the desired output and the actual output, the connection weights are corrected layer by layer from the output layer through each middle layer, and then returned to the input layer, and repeated alternation is performed until the global error of the network tends to the given minimum value, that is, the learning process is completed. The main advantage of BP neural network is that it has extremely strong nonlinear mapping ability.

[0185] CNN includes one-dimensional convolutional neural network, two-dimensional convolutional neural network and three-dimensional convolutional neural network. It is mainly used to identify two-dimensional graphics that are invariant to displacement, scaling and other forms of distortion. RNN is a special neural network structure, which is proposed based on the view that "human cognition is based on past experience and memory". RNN neural network is generally used in problems that consider time sequence. Common applications include: speech recognition, image description generation, text similarity calculation, music recommendation and other new application areas.

[0186] As the core module of low-power backscatter tags, the packet detection based on neural network should be lightweight, because the network is denser and requires more computing and storage resources. At the same time, since weaker LoRa signals are easily buried by other interference signals, the neural network should also be responsive. Combining these two criteria and the application fields of the above three neural networks, the present invention intends to use the simplest BP neural network to train the data set, permanently excluding the mainstream deep neural network architectures—CNN and RNN, because they both involve a lot of computing and memory overhead. The following describes how to generate a BP neural network.

[0187] Generation of training network

[0188] It is proposed to use a multilayer perceptron (MLP) to implement the neural network. The MLP model uses digital samples of the ADC incoming signal as input and outputs a binary 1 or 0 to indicate whether a leading chirp signal exists. The computational and memory overhead of the MLP model is proportional to its network size (i.e., the number of neurons). The larger the network size, the better the detection performance. Therefore, it is necessary to find an appropriate network size to balance the power budget and detection accuracy. In a passive backscatter communication system, starting from a dense network, the number of units in the system is reduced, and the neural network model is continuously optimized until a minimized topological structure and a simplified computing unit are obtained, achieving a balance between minimizing the network structure and maximizing the detection effect. After training, the present invention intends to obtain a neural network input layer M size of 128, a hidden layer N of 2, and an activation function selected as the ReLU function.

[0189] The present invention intends to use the positive correlation between the accuracy of the network model and the size of the input data, design a sorted list of input data from small to large, select the maximum value of the input sorted list for the first training, and after testing the feasibility of the network, use the binary search method to test the neural network until the minimum neural network input M is found. The model is trained with the previous sampling signal and the detection accuracy is tested. The results are shown in Table 1-1. Based on this benchmark test result, M is set to 128, which will empirically balance the accuracy and computational overhead. Figure 6 The number of neural network layers is shown.

[0190] Table 1-1: Impact of input unit M on package measurement accuracy

[0191] Input layer size M 64 128 256 12 Detection accuracy (%) 94 97.8 99.75 00

[0192] The present invention presets a list of hidden layer numbers, a list of neuron numbers, and a list of batchsize sizes; when used, the model is trained from the minimum value in the list, and each parameter is trained dozens of times. After the network is trained, the neural network effect is tested until a model that meets the test criteria appears, which is the minimized neural network model, thereby determining the hidden layer size N. The present invention intends to use 20 to initialize N, and gradually reduce it until the detection performance drops significantly. According to this criterion, the present invention sets N to 2, and can maintain a detection accuracy of more than 97% in the test set. The activation function is a function running on the neurons of the artificial neural network, which is responsible for mapping the input of the neuron to the output end. The activation function is introduced to increase the nonlinearity of the neural network model. The present invention intends to use a rectified linear unit (ReLU) function as an activation function, because its implementation is simple, it can reduce the overhead of implementing nonlinear operations in embedded systems; and its performance in various prediction tasks can be recognized.

[0193] Neural network training

[0194] Training a neural network means using existing data to train network weights and biases suitable for the target classification algorithm based on the characteristics of the existing sample library. The pre-generated neural network structure is: the input data uses the voltage value received by the ADC, the hidden layer is a certain number of neurons, each neuron multiplies and accumulates the weight and output, and the output of each layer is used as the input of the next layer after the nonlinear operation of the activation model, until the output of the output layer is used as the output data.

[0195] The detailed training steps are: according to the performance of the generated neural network, the training parameters of the neural network are adjusted. The technical process of adjusting the parameter weights and biases includes: by expanding and rounding the weights as a whole, removing the decimal storage in the weights, thereby reducing the difficulty of decimal calculations; modifying the weights according to the reserved calculation bit width and calculation accuracy in the embedded system, and converting the weights into numerical values ​​that can be represented in the embedded system. The present invention intends to implement the MLP model on TensorFlow and use 4 NVIDIA Titan RTX GPUs for training. The training process of each parameter: first import the data set labels (including positive and negative 20,000 samples) pre-processed in MATLAB on the TensorFlow platform, use the keras framework design, select the ReLU function as the activation function, and select the mean square error as the loss function. When it iterates 1000 times or the loss function no longer changes significantly, the present invention will consider the training to be established, and the program will generate 128 weights and 2 biases at the end.

[0196] (3) Lightweight BP neural network accelerator based on low-power FPGA chip

[0197] Based on the established sample library and the designed prototype BP neural network, the present invention intends to rely on ultra-low power FPGA chips to realize a lightweight neural network accelerator and integrate it into a low-power passive backscattering LoRa communication system, thereby achieving the effect of detecting the target environment carrier.

[0198] Step 1 FPGA module design of neural network signal detection circuit

[0199] Figure 7 The block diagram of the neural network accelerator based on the low-power FPGA chip to be implemented in the present invention includes: a data input part, a serial-to-parallel converter, a normalization part, a first layer of neurons, an activation part, a second layer of neurons and a parallel-to-serial converter.

[0200] The design of the neural network FPGA module starts with the ADC data acquisition module outputting digital signals into the neural network, and ends with the second-layer neurons outputting the calculation results and converting them into serial output data. Finally, a signal is sent to enable the next-level modulation module, which controls the modulation module to perform passive backscattering and loads the collected information onto the environmental carrier. The data input part consists of an ADC and a cascaded D-type flip-flop group. The ADC converts analog signals into digital signals, and the D-type flip-flops buffer data between different clock domains to eliminate metastable states during transmission. Serial-to-parallel converter: converts the 12-bit serial signal output by the ADC into a group of parallel signals (12 bits) for subsequent neurons to calculate. Normalization: normalizes the voltage value in a small range to the voltage value in a large range to enhance the signal strength. The first-layer neurons: calculate the data output after normalization, assign values ​​to each weight at reset, and automatically increase the count by one after each calculation. When the count reaches 128 times, the calculation result will be output. Activation function: select the ReLU function. When the output of the first-layer neurons is negative, the output is changed to 0, and when the output of the neurons is positive, the original value remains unchanged. Second layer neurons: Input weights at reset, multiply the calculation results of the first layer neurons by the weights, and finally add the bias. Parallel-to-serial converter: Convert the 45-bit data output by the second-level neurons into serial data output.

[0201] Step 2: Test of low-power BP neural network accelerator based on FPGA platform

[0202] The present invention intends to select the low-power IGLOO nano 600 chip as a platform for realizing lightweight neural networks. The IGLOO nano 600 chip has 1Kb of FlashROM, consumes 5μW of power in flash freeze mode, and is reprogrammable.

[0203] After implementing the design of the FPGA module, the peripheral circuit design of the implemented system is carried out, and its signal detection capability is tested. During the test, a LORA transmitting node, several connecting wires, several attenuators, a USRP, and a desktop computer are planned to be used. The LORA transmitting node, attenuator, and USRP are connected in series, and the LORA data under different attenuations are collected as pure LORA data. Finally, the Verilog program in the Libero simulation software is burned to the FPGA board, sent through the Gateway, the USRP receives the data, and the oscilloscope displays the waveform for analysis.

[0204] 2 Study on low-power variable throughput modulation scheme based on wavetable synthesis algorithm

[0205] (1) Design of low-power intermediate frequency signal generator based on wave table query

[0206] Step 1 Synthesize signal generator based on wave table

[0207] Wavetable synthesizer is a digital frequency synthesis technology, which is a new frequency synthesis technology that directly synthesizes the required waveform based on the concept of phase. Compared with traditional frequency synthesizers, wavetable synthesizers have the advantages of low cost, low power consumption, high resolution and fast conversion time. They are widely used in the fields of telecommunications and electronic instruments and are a key technology for achieving full digitalization of equipment. This technology is often used in audio processing and has been widely adopted by digital music synthesizers to produce natural, tone-like sounds. Wavetable synthesizers can control the output signal frequency by setting the step size.

[0208] Wave table theory and step length calculation. The present invention uses frequency f s One cycle of the sine wave is sampled and these signal samples are stored in the read-only memory of the passive backscatter communication system tag, called a waveform table. Let L be the length of this wave table, the fundamental frequency f of the stored waveform 0 It can be expressed as: 0 =f s / L, and then the present invention generates a target frequency f=f by sampling / subsampling the wave table with a step size step 0 *step, where f 0 =f s / L=CLK / 2 32 ;CLK / 2 32 The calculation formula of step length is: step = f*2 32 / CLK

[0209] Step 2: Implementation of Wavetable Synthesized Signal Generator

[0210] The wavetable synthesizer module consists of two parts: 1) step accumulator; 2) wavetable register. First, according to the Nyquist sampling theorem, the present invention samples, encodes and quantizes a complete period of sinusoidal signal from the phase of the continuous signal to form a sinusoidal function table stored in the wavetable register. A sinusoidal wave of one period in the wavetable register is divided into 2 M (M=16) parts, and the amplitude is amplified to improve the accuracy of the output frequency. The step accumulator samples the sine function table in the wave table register. When synthesizing, the phase increment is changed by changing the frequency control word of the step accumulator. The different phase increments lead to different number of sampling points in one cycle. When the sampling frequency remains unchanged, the synthesized frequency is also different due to different phase increments.

[0211] There are two parameters related to the generation of the wavetable synthesizer: 1) FPGA onboard crystal oscillator clock CLK. 2) Frequency control word defined by STEP. CLK is used as a clock signal to drive the subsequent phase accumulator to calculate the address value, that is, the sampling position. The frequency control word STEP determines the synthesized frequency. The step accumulator used in the present invention is a 32-bit accumulator. In order to reduce the storage pressure of the wavetable register, the present invention takes the first 16 bits of 32 bits as the sampling address and outputs it to the wavetable register. This operation can reduce the complexity of the algorithm and improve efficiency. In order to generate a fine shift frequency in a low-power manner, the present invention intends to use wavetable synthesis. For greater clarity, the present invention assumes that the base frequency of the stored waveform is 100KHz. In order to generate a 50KHz waveform, read the value of each stored sample twice, and then read the next one (i.e., index order: 1, 1, 2, 2, 3, 3, ...). In order to generate a 200KHz waveform, the backscatter tag only needs to read the values ​​with odd indices (i.e., 1, 3, 5, 7, ...). When the cursor exceeds the table length L, it needs to wrap around to resample again. By changing the increment (step size), the present invention can obtain a high-precision frequency shift plan, sufficient to support various data rates (up to 8 bits per LoRa chirp). Fig.10 shows the workflow of this waveform table synthesizer. In a passive backscatter communication system, L can be set to 2 32 , you will find that the issues of storage overhead and waveform granularity are balanced.

[0212] (2) Fine-grained FSK modulation and spatial signal stitching technology

[0213] Step 1 FSK modulation and signal stitching

[0214] The present invention intends to use FSK modulation to load the data collected by the passive scattering node. The original carrier LoRa Chirp is used as the reference signal, and different frequency shifts are performed on the reference signal to modulate different data.

[0215] The chirp in the Lora data packet is CSS-encoded. Within the entire bandwidth BW, the frequency of the chirp increases linearly. Different chirps have different starting frequencies. The present invention can shift the frequency of the existing chirp, control the starting frequency of the chirp, and encode the chirp. Then, different antennas are used to simultaneously move the FSK-modulated signal to another channel for spatial stitching of the signal. The stitched signal can still occupy the entire LoRa channel and maintain the long-distance transmission characteristics of the LoRa Chirp. After the encoder acquires the sensor data, it generates different frequencies by controlling the code table synthesizer. The starting frequency of the chirp is shifted by frequency, and different data is represented according to the different starting frequencies.

[0216] The decoding is based on the linear frequency modulation characteristics of the LoRa signal. After multiplying the upchirp in the LoRa signal by a standard downchirp, a fast Fourier transform is performed, and an energy concentration point will appear in the frequency domain. This point is at the center frequency of the chirp. Unlike the traditional LoRa decoding method, the traditional method directly intercepts the downchirp of the same frequency from the received signal and multiplies it with the encoded chirp, and then performs subsequent processing. The present invention has made some improvements. The receiving end first pre-stores a downchirp of a reference signal, and after receiving the signal, it intercepts the chirp encoded by the LoRa packet after backscattering according to time. After multiplying with the reference downchirp, an FFT transform is performed to obtain the energy concentration point in the frequency domain.

[0217] Step 2 Hardware Implementation

[0218] The hardware implementation part includes: clock division, clock counting module, sensor data encoding module, wavetable synthesizer frequency generation, and frequency shifting frequency selection.

[0219] Step 1: The frequency division or multiplication module performs clock matching processing on the onboard crystal oscillator output signal (such as 20MHz divided by 2) and then provides it to the clock counting module.

[0220] Step 2: The clock counting module starts working after the system detects the LoRa signal. It calculates the duration of the LoRa signal preamble according to the LoRa packet structure and accumulates the time to the data load of the LoRa signal. At this time, it can choose to locate the starting position of the chirp used for encoding in the backscattered LoRa signal, and provide the encoding clock trigger signal to the sensor data encoding module based on the duration of the chirp.

[0221] Step 3: After the sensor data encoding module obtains the sensor data such as temperature and humidity, the data-shift frequency correspondence table provided by the backscatter communication coding modulation scheme based on the LoRa signal is used. For example, see Table 1.2. Through the step conversion formula, the expected shift frequency corresponding to the information to be encoded is converted into the corresponding frequency selection step, and then the shift frequency corresponding to the encoded data is obtained through the fine-grained frequency generation module. At this time, the difference between the key frequency of the reference signal energy set and the encoded data corresponds one by one, and decoding can be performed by comparing the coding table, see Table 1.2.

[0222] Table 1.2 Backscatter communication coding and modulation scheme based on LoRa signal

[0223] serial number Sensor / Decimal Encoded information / binary Shift frequency / Hz 1 0 00000000 0KHz 2 1 00000001 2KHz 3 2 00000010 4KHz …… 250 250 11111010 500KHz

[0224] Step 4: The wavetable synthesizer generates frequency and selects the frequency shift, which is composed of a phase accumulator module and a discrete sine lookup table. The accumulator counts and accumulates on the rising edge of the clock, and provides the accumulator variable value as the signal sampling coordinate to the sine signal register in real time. After the sine signal register is sampled, the expected frequency shift is generated in the fine-grained frequency generation module, but the signal is not a square wave signal used for frequency shift. Finally, it is binarized and output as a square wave frequency shift. The following is the pseudo code for the implementation of the phase accumulator on the FPGA.

[0225]

[0226]

[0227] 3 Research on beamforming technology based on scattered fields

[0228] (1) Scattering signal enhancement antenna array design method based on scattering field theory

[0229] Step 1 Based on the scattering field theory, use an antenna array instead of a single antenna for backscattering transmission

[0230] Antennas, as a special scattering target, not only affect electromagnetic wave scattering as a physical entity, but also affect electromagnetic waves due to their structure as antennas. By using the scattering matrix to derive the basic theory of antennas as scattering targets, the basic theoretical formula of antenna scattering field is obtained as follows:

[0231]

[0232] In the formula, the scattering of the antenna It consists of two parts. Usually called structural mode scattering σ S , that is, the antenna rear end load (Z l =Zc ) when the energy entering the antenna is completely absorbed. The antenna scattering is only related to the physical parameters such as the shape, size and material of the antenna. It is called the antenna mode scattering σ a It is a unique scattering in which the antenna, as a radiator, radiates the energy brought by the incident wave for a second time. It is also the difference between antenna scattering and general object scattering.

[0233] In passive backscattering, single antenna scattering often has throughput and direction limitations, so it is proposed to use the design of scattered field antenna array to achieve more efficient scattering.

[0234] Step 2 Scattering Field Antenna Array Design Theory

[0235] The function of the antenna array is to enhance the directivity of the antenna, improve the gain coefficient of the antenna, or to obtain the desired directional characteristics. For the N-element uniform linear array antenna, the present invention uses an equivalent network model to analyze the field in its external space. In the radiation space of the antenna, a sphere S containing the antenna is selected, and the field in the space outside the antenna can be expanded using a vector spherical wave function:

[0236]

[0237] Through the joint action of N units in the array, the directivity of the antenna can be enhanced and the gain coefficient of the antenna can be improved. Therefore, it is proposed to design an antenna array for use in tags of passive backscatter communication systems to form a more efficient scattering field antenna array to meet the requirements of gain and direction.

[0238] (2) Scattering field antenna array optimization algorithm and EDA design platform construction

[0239] Step 1 Selection of antenna array optimization algorithm

[0240] In antenna array design, the first thing to solve is the spacing problem of each unit in the array, so that the antenna array can achieve high scattering efficiency. For the optimization problem of the spacing in the scattering field antenna array, the commonly used methods are particle swarm optimization (POS algorithm), genetic algorithm (GA algorithm) and differential evolution algorithm (DE algorithm). The POS algorithm lacks dynamic adjustment of speed and is prone to fall into local optimality, resulting in low convergence accuracy and difficulty in convergence. The GA algorithm requires a large number of individual calculations, and the processing time of complex problems is relatively slow.

[0241]

[0242] Therefore, the DE algorithm is selected to optimize the spacing in the antenna array. The DE algorithm starts from a randomly generated initial population, generates new individuals by summing the vector difference between any two individuals in the population and the third individual, and then compares the new individuals with the corresponding individuals in the contemporary population. Through continuous evolution, good individuals are retained, and inferior individuals are eliminated, guiding the search to approach the optimal solution. It is suitable for spacing optimization in scattered field antenna arrays.

[0243] Step 2 HFSS joint simulation--Establishment of EDA platform

[0244] HFSS is a full-wave three-dimensional electromagnetic simulation software based on the electromagnetic field finite element method. It provides a comprehensive solution for the design of antennas and antenna systems, and can accurately simulate and calculate many antenna parameters such as antenna radiation field pattern, antenna gain and S parameters. However, in HFSS simulation, any change in small data requires resetting the parameters. Moreover, the DE algorithm requires dozens or even hundreds of iterations in iterative calculations, so each iteration requires resetting the parameters in HFSS, which seriously affects the simulation efficiency and accuracy. Therefore, the present invention proposes an HFSS joint simulation-EDA platform based on the DE algorithm. A design method for an anti-interference scattering communication system antenna array is proposed, characterized in that the design method specifically adopts MATLAB and HFSS for design, HFSS establishes a scattering communication system antenna array model, and sets the array element antenna spacing of the scattering communication system antenna array model as a changeable variable, and MATLAB jointly calculates the optimal array element antenna spacing by using a differential evolution algorithm and HFSS.

[0245] In the present invention, by using the differential evolution algorithm and HFSS for joint calculation on MATLAB, the antenna scattering direction in the antenna array of the scattering communication system can be concentrated, which not only improves the distance of traditional backscattering communication; but also because the direction can be concentrated, directional transmission of data can be achieved without affecting other channels.

[0246] Step 3 Design and optimization of nine-element antenna array

[0247] Unlike active beamforming that follows the theory of radiation propagation, the backscattered signal reflected by the antenna array establishes a scattered electromagnetic field characterized by the antenna characteristics and the array structure. Since the antenna characteristics mainly depend on the impedance and form factor of the antenna itself, the present invention does not consider antenna design, but focuses on array layout optimization. Array layout design inevitably involves a trade-off between the number of antennas N and the required beam width θ. The larger the antenna array (i.e., the more antennas), the narrower the beam width. However, it increases the form factor and imposes additional costs on the backscatter tag. In order to balance the array size and beam width, the present invention presets the requirements of the target beam width, beamforming gain, and sidelobe variance, and gradually increases the number of antennas to approach the target beam width. Design. The design of the present invention runs in an iterative manner. In each iteration, the algorithm takes the number of antennas as input. Then, it iteratively searches for the optimal antenna position that minimizes the following optimization function;

[0248] L=α*ΔW+β*ΔG+γ*ΔV

[0249] Where L is the loss; α, β, and γ are different weights to weigh different components of the optimization objective. ΔW, ΔG, and ΔV are the main lobe width, maximum gain, and sidelobe variance, respectively.

[0250] The present invention performs multiple iterative calculations and sets α, β and γ to 5, 7 and 0.6, which can obtain higher beamforming gain and narrower beamwidth. In each array layout setting, the present invention sends the physical parameters of the antenna position and antenna elements (such as dielectric constant, size, radiation and scattering characteristics, etc.) in HFSS to simulate the scattering pattern of the antenna array. Then, HFSS outputs the beamforming gain, beamwidth and sidelobe gain variance of this specific antenna array layout, which the present invention uses to calculate the loss L. After algorithm iteration, until the array layout with the minimum loss L is found. However, it is computationally difficult to find the best array layout in each antenna setting number because the present invention needs to list all possible antenna positions. In order to speed up the search speed, three measures are proposed:

[0251] Set the minimum antenna spacing to λ / 4;

[0252] Set the fixed search step size to λ / 8;

[0253] The DE algorithm is used to intelligently search for near-optimal solutions.

[0254] Then, the optimized nine-element antenna array model was obtained in HFSS. The results of the optimized nine-element antenna array were compared with the simulation results of the single antenna design. The gain of the optimized antenna was increased by 20dB.

[0255] (3) Beamforming technology that can be used in three-dimensional space

[0256] In the actual application of antennas, since the deployment locations of network communication gateways and nodes are three-dimensional, the target beamforming direction needs to be able to cover every angle in the 3D space to support deployment applications in real environments. In order to solve the limitations of 2D antenna arrays in real environment deployment, based on the scattering field theory, the present invention extends the 2D antenna array to 3D space to support multiple beamforming angles. Therefore, a 3D antenna array is designed. Fig.18 This is the spatial model diagram of the 3D antenna array.

[0257] Specifically, multiple antennas are placed on each axis of the 3D coordinate system, and the passive backscatter tag switches the beamforming angle by selecting different antenna groups to achieve the directivity of the antenna in three-dimensional space. Fig.15 As shown, the antenna arrays obtained at different angles.

[0258] According to the designed 3D antenna array, the present invention obtains the 3D antenna array pattern simulation result in HFSS. Fig.16 shown.

[0259] The present invention observes that the main lobe is aligned with the target angle under all three angle settings. The experimental results meet the expected requirements of the present invention for antenna directivity. At the same time, similar to radar, the beamforming gain of the antenna array of the present invention is proportional to its effective array size. Although the current design of the present invention uses λ / 4 as the minimum antenna spacing, it is possible to miniaturize the antenna array by changing the antenna shape factor, reducing the minimum antenna spacing and optimizing the manufacturing process.

[0260] 4. Research on efficient energy harvesting surfaces and global energy optimization supply solutions based on metamaterials

[0261] Passive backscattering collects energy from the surrounding environment to support computing and communication tasks. Without an effective energy management module, the backscatter tag will quickly run out of energy and stop working. Therefore, the present invention introduces a low-power energy management module based on a metamaterial substrate to solve this problem.

[0262] Research on efficient energy harvesting surfaces based on metamaterials

[0263] Backscatter tags collect energy from the surrounding environment to support computing and communication tasks. There are many mature environmental energy collection technologies today, such as solar energy collection, thermoelectric energy collection, kinetic energy collection, etc. Obviously, solar energy is the most widely used and most mature energy collection method today, but its disadvantage is that energy can only be collected in places with sufficient light. For kinetic or mechanical energy collection, piezoelectric capacitors can be used to collect energy, which can usually be collected from bridges, roads, railways, etc., but the energy that can be obtained is usually small and the usage scenarios are limited.

[0264] Step 1 Determination of high-efficiency energy collection surface materials

[0265] The present invention intends to use the complementary split ring resonator (RSRR) of electromagnetic metamaterial units to design the energy collection surface. In the entire electromagnetic energy collection system, the receiving antenna directly determines the electromagnetic power that can be collected. Traditional receiving antennas generally use microstrip antennas, but the microstrip antenna itself has a narrow band, which is not conducive to wide-band energy collection. In order to increase the working bandwidth of the rectenna, the scientific community has been making various attempts. In 2012, Omar M. Ramahi and others used split ring resonators (SRR) to form an array as a receiving antenna in a WPT system for the first time, and found that its energy absorption efficiency was 70%. Subsequently, it was discovered that the efficiency of the complementary split ring resonator array in energy collection was significantly enhanced compared to the SRR energy collection array, and the RF energy collection efficiency reached 92%.

[0266] Therefore, compared with traditional rectennas, the energy harvesting system based on metamaterials has the advantages of miniaturization and high efficiency, and is suitable for the application of passive backscattering systems. Electromagnetic metamaterials are defined as artificial composite structures or composite materials that are different from conventional materials. By designing the topological structure and size of metamaterials, the known laws of nature are broken through to give them some special physical properties. The study of metamaterials started with the theory and experiment of left-handed materials. The present invention intends to use SRR or CSRR left-handed materials to study the energy harvesting of metamaterials.

[0267] Step 2 Theory and simulation based on metamaterial surface design

[0268] In a passive space, the current density J and the charge density ρ are 0 everywhere. In an isotropic linear homogeneous medium, E and H satisfy Maxwell's equations:

[0269]

[0270]

[0271]

[0272]

[0273] Will Substituting into the above equation, we can get the basic equation of time-harmonic electromagnetic waves, namely the Helmholtz equation:

[0274]

[0275]

[0276] The propagation wave vector k of the electromagnetic wave is:

[0277] k 2 =ω 2 εμ

[0278] In left-handed materials, ε<0, μ<0, k2>0, electromagnetic waves can still propagate in the medium, and the wave vectors k, E and H conform to the left-handed spiral relationship. The Poynting vector S gives the direction of energy flow:

[0279]

[0280] Therefore, left-handed materials have many physical properties different from traditional right-handed materials, such as backward wave characteristics, inverse Doppler effect, inverse Cerenkov radiation effect, perfect refraction effect and negative refraction phenomenon, etc. Scientists have developed a left-handed metamaterial structure in which the incident plane wave vector is perpendicular to the metamaterial surface, making the planar left-handed metamaterial have the advantages of smaller unit size and wide application, and have a wider application prospect.

[0281] We plan to use HFSS for simulation. First, we design a CSRR unit such as Fig.17 In the left figure, RF4B substrate is used, the dielectric constant is 2.65, the electrical loss is 0.001, the substrate thickness is t = 0.5mm, the parameters are set to L = 8mm, s = 4mm, w = 0.55mm, and the copper layer thickness is h = 35μmm. Then, a 4×4 CSRR unit array is formed according to the designed CSRR unit. Fig.18 Right picture.

[0282] State-transfer global energy optimization allocation based on heterogeneous multi-capacitor storage structure.

[0283] Step 1 Energy storage and release based on heterogeneous multi-capacitors

[0284] Conventional backscatter designs often use a single supercapacitor to store the collected energy. As a result, passive backscatter tags can only work intermittently, alternating between charging and discharging states. At the same time, because the tag stops working when the supercapacitor is charging, the working time is greatly shortened.

[0285] The present invention intends to use a multi-capacitor storage structure to solve this problem. Taking the heterogeneous dual-capacitor storage structure as an example, a supercapacitor Ccap with a larger capacity (such as 220mf) and an electrolytic capacitor C with a smaller capacity (such as 10μF) are selected to construct a power storage reservoir. Electrolytic capacitors have the characteristics of high mechanical strength, small size, small capacity, fast charging speed, and low price, but short life; supercapacitors have high power density, long cycle life, small characteristic changes, high charging and discharging efficiency and green environmental protection, but slow charging speed. According to the above capacitance characteristics, a side-by-side capacitor structure is adopted. The electrolytic capacitor has a small capacity and a fast charging speed. When the supercapacitor is still charging, it is discharged first, which can solve the problem of slow charging of the supercapacitor and affecting efficiency. When the output voltage of the electrolytic capacitor exceeds the trigger threshold of the tag, the electrolytic capacitor will start to discharge and drive the tag to work. When the output voltage drops below the trigger threshold, the electrolytic capacitor will switch to the charging state. At this time, the supercapacitor is awakened and starts to discharge. Compared with a single supercapacitor, heterogeneous dual capacitors have the following advantages:

[0286] By using fast-charging electrolytic capacitors, cold-start delays can be shortened.

[0287] By alternating between the two capacitors, the system can operate continuously.

[0288] Step 2 Global energy management based on finite state machine

[0289] Pre-designed global energy management circuits such as Fig.19 As shown in the figure. The circuit consists of three logic units: charger, energy controller and voltage output. The charger collects energy from ambient RF signals and solar energy to improve charging efficiency. The collected energy is then stored in a capacitor bank. The energy controller manages the capacitors to power the passive backscatter tag. The voltage output module regulates the stored power to a constant voltage to achieve maximum low power consumption.

[0290] The working principle of the circuit can be described by a finite state machine, such as Fig. 20 As shown. By constructing a finite state machine, using different circuit states to control the charging and discharging of the overall circuit, and designing the current gate module, leakage is effectively prevented, and the data transmission integrity and reliability of the system are guaranteed. Among them, V c Represents the electrolytic capacitor voltage, V cap Represents the supercapacitor voltage, V thFor the diode array and switch tube G 0 The combined on-state voltage, V a Represents the NOT gate input voltage, V b is the NOT gate output voltage, V G Indicates switch tube G 3 The transition between each state is as follows: Fig.21 .

[0291] 5. Verification and testing experimental design

[0292] In order to verify the long-distance communication effect of the system and the effectiveness of the energy supply module in utilizing environmental energy, the present invention will detect and evaluate the system indicators through indoor, indoor and outdoor experiments and typical scene experiments in real environments, and continuously optimize the system through analysis of multiple experimental results to achieve continuous optimization of system indicators. The indicators that need to be detected and evaluated for the verification and testing of the "passive backscattering long-distance, low-power communication system stimulated by environmental information" are: signal strength, carrier detection distance, backscattering transmission distance, communication system packet loss rate and bit error rate, energy collection speed, system continuous working time, etc.

[0293] (1) Experimental design

[0294] In the actual experiment, the following three wireless communication experiments with different topologies were implemented, based on the premise that the LoRa tag does not have a battery and obtains energy from RF signals and environmental energy. Fig. 22 As shown in the figure: the downlink LoRa Chirp signal is used as an excitation signal and modulated into a standard uplink LoRa Chirp signal; the uplink LoRa Chirp signals are detected and used as excitation to modulate the sensing data; the downlink LoRa Chirp signals are detected and used as excitation to modulate the sensing data.

[0295] 1) Taking the above topology as a reference for system deployment, the present invention will verify the effectiveness of the kilometer-level backscatter communication system using the commercial LoRa Chirp as the excitation source through experiments. The design of the experiment needs to answer the following questions:

[0296] ① In the communication platform built with dedicated equipment (USRP 2954R) and commercial equipment (Dragino LoRaGateway), what are the delay, packet loss rate, bit error rate, and throughput of the backscattered signal of the system proposed in this invention? What is the communication coverage and robustness of the system? Compared with the existing mainstream system design, how is the accuracy and effectiveness of information transmission?

[0297] ② What is the impact of the "passive backscattering" method of the system proposed in the present invention on the communication effect when changing parameters (such as the transmission rate, transmission power, transmission frequency band of the original LoRaWAN network data and the spreading factor of the LoRa Chirp signal, etc.)?

[0298] ③What is the impact of 3D antenna array and source nodes, number of scattering tags, antenna spacing, tag spacing, geometric topology of antennas and tags, and device deployment density on the system communication distance, packet loss rate, and bit error rate?

[0299] ④ What is the maximum distance that the environmental carrier detection method based on the lightweight neural network accelerator provided by the present invention can detect, and how does the detection sensitivity change when the source signal strength changes?

[0300] ⑤ When the signal source changes dynamically, what is the impact of the moving speed and trajectory on the maximum scattering distance? What is the effect of multi-node multi-frequency random frequency hopping?

[0301] ⑥ How is the concentration and effectiveness of beamforming in the three-dimensional antenna array design? Compared with the existing single dipole antenna, can the reflected signal strength in directions other than the target beamforming direction be reduced?

[0302] ⑦ When the passive backscatter tag is in an environment with different temperatures, humidity, different air pollution levels (PM2.5 / PM10), and different weather conditions (such as rain, snow, sunny, foggy, windy, etc.), what are the differences in the system's communication distance and signal strength? Is the system robust enough to adapt to different working environments?

[0303] ⑧ What is the effect of the passive backscatter long-distance communication system proposed in the present invention in different obstacle environments, such as human, animal, and plant environments, or when the number of obstacles is different, and does its application have limitations?

[0304] ⑨ When the environment is interfered by wireless signals of the same frequency band (such as GSM network signals, RFID signals), what is the impact of the strength, distance, and working frequency of the interference source on the communication indicators of the proposed system? Does the system have anti-interference capabilities?

[0305] ⑩ The module power consumption of the passive backscatter system, compared with other passive backscatter systems, is the energy efficiency greatly improved? What is the system leakage situation, and how is the energy storage effect?

[0306] 2) For the system's efficient multi-energy collection and energy storage optimization, the following issues need to be verified:

[0307] ①How does the charging efficiency of the energy harvesting module change under different environmental conditions? What factors are related to the charging time?

[0308] ② Is it still possible to charge effectively when the ambient energy is very limited?

[0309] ③ What is the relationship between capacitor energy storage and working time? What is the working time curve of the energy storage module when the system load is different?

[0310] ④ Can the current gate be effectively opened and closed when the passive reflective tag is working and idle to achieve system energy supply optimization?

[0311] (2) Simulation experiment:

[0312] The simulation experiment starts from the following aspects:

[0313] ① Experiments were conducted based on the USRP 2954R software radio system, the Xlinx FPGA verification platform, and a deployment environment built with different types of commercial devices (Dragino LoRa Gateway and nodes). According to the different information characteristics of the provided devices, the simulation platform was used to simulate various devices for experiments to examine whether different devices, as passive backward transmission targets, have any impact on the expected indicators of the communication system.

[0314] ②Through simulation, the parameters such as capacitance, inductance, resistance and switch in the hardware design are adjusted, and combined with the theoretical calculation model, the key design parameters and topological structure of the proposed passive backscattering system are found to achieve low power consumption.

[0315] ③ Use simulation software to simulate the different receiving and transmitting states of various devices. Under different deployment topologies, obtain multiple sets of data through multiple experiments, and continuously enrich and improve the theoretical system of the proposed method from the analysis report of the experimental data.

[0316] (3) Indoor experiment design and deployment diagram

[0317] The real environment experiment is mainly to test the difference between the effect of the proposed method in the actual environment and the simulation effect. The indoor real environment experiment plan is as follows.

[0318] ① The impact of indoor multipath on the system. Different devices analyze the collected signals in the environment with general and rich multipath indoors, record the results and compare them. Analyze the effectiveness of carrier detection, node information loading, frequency shift modulation and randomized multi-frequency hopping of the passive backscatter communication system under different indoor environmental interference, and measure and compare the system transmission distance, throughput, bit error rate, energy consumption and other parameters in different channels under different environmental conditions.

[0319] ② Changes in indoor reflected signal strength and the furthest indoor reflected distance. When there is no other signal interference indoors, change the distance between the signal source and the passive backscatter tag, measure the furthest reflected distance of the signal every 5 meters, and the corresponding reflected signal strength, packet loss rate, and bit error rate. Record the changes in reflected signal strength.

[0320] ③ The impact of different channel combinations on backscattering effects. In an indoor environment without other signal interference, the reflection effects from the LoRa downlink channel to the uplink channel, the uplink channel to the uplink channel, and the downlink to the downlink channel are measured to obtain the impact of the three different reflection modes on the farthest reflection distance of the reflected signal.

[0321] ④ The impact of walls on signal strength. In an indoor environment without other signal interference, measure the number of walls that the signal can pass through, as well as the signal strength attenuation after passing through each wall, and the changes in packet loss rate and bit error rate.

[0322] ⑤ Indoor charging efficiency experiment. Measure the energy harvesting charging efficiency under indoor fluorescent lighting conditions. Test the charging time corresponding to charging the energy harvesting module to 1.9V, 2.1V, 2.3V, 2.5V, and 2.7V in a well-lit room. Follow the same experimental steps, replace the fluorescent light source with outdoor natural light conditions, and complete the comparative experiment.

[0323] (4) Outdoor experiment design and deployment diagram

[0324] Outdoor tests are mainly conducted in open outdoor areas to verify the impact of different hardware devices, device deployment plans, environmental conditions and other variables on the experimental results in outdoor scenarios:

[0325] ① The impact of dedicated equipment and commercial equipment on the system. Experiments were conducted using a test platform (USRP2954R, USRP 2921) and commercial equipment (Dragino LoRa Gateway) under the same outdoor conditions, and the impact of environmental factors, equipment deployment schemes, etc. on the experimental results were analyzed.

[0326] ② Random multi-frequency hopping experiment with multi-node deployment in different topologies. Multiple reflection nodes are randomly deployed around the source node, for example, some are 10m, 50m, 100m, 150m, etc. away from the carrier transmission source node. Each reflection node reflects the carrier signal to a different frequency band, and the signal strength, throughput, bit error rate, and packet loss rate of these reflected signals are measured every 20m.

[0327] ③ The impact of natural environmental factors on system robustness. The system is evaluated under different weather conditions (cloudy / sunny / rainy / snowy), temperature, humidity, air pollutants (PM10), light, smoke, alcohol concentration, etc. The distance between the transmitting source node and the backscattering node is fixed, the farthest scattering distance is measured, and the intensity of the scattered signal is measured every 20m, and a curve of the scattered signal intensity attenuation with distance is drawn. Similarly, the throughput, packet loss rate and bit error rate of the communication system are measured and analyzed.

[0328] ④Verification of communication effect in interference environment. In outdoor environment, change the transmission frequency of RFID signals in the same frequency band, and measure the impact of interference signals on the backscatter signal propagation distance, signal strength, packet loss rate and bit error rate in different situations in the LoRa network band and out of band.

[0329] In addition, in the outdoor passive sensing backscattering experiment, there are single-transmit single-receive, multiple-transmit single-receive, single-transmit multiple-receive, and multiple-receive multiple-transmit communication situations. It is necessary to adjust the equipment deployment plan to verify the impact of different topology situations on the experimental results.

[0330] The outdoor random multi-frequency hopping experiment was carried out in a square with a line of sight of 500m×100m, and the other ranging experiments were carried out in a venue with a line of sight of 2km×1.6km.

[0331] (5) Typical field scene experiments

[0332] Experimental location: Qinling Golden Monkey Research Base, Yuhuang Temple, Zhouzhi National Nature Reserve, Shaanxi Province.

[0333] Estimated equipment models: two sets of USRP 2945R, one set of USRP 2921, four sets of Dragino LoRa Gateway, ten sets of passive backscatter tag nodes, ten sets of LoRa SX1276 nodes, four sets of solar charging panels, four sets of Intel 5300NIC (CSI) devices, four sets of Wi-Fi devices, four sets of RFID devices, ten sets of various sensors such as temperature, humidity, and light. Indicators to be evaluated for verification effect: charging and discharging efficiency, energy consumption, communication range, information transmission quality, manpower consumption, and robustness. The present invention group intends to select a 1.2km×1km area in the Yuhuang Temple golden monkey feeding observation area to test the passive backscatter communication range, energy collection module charging efficiency, information collection, transmission accuracy and real-time performance, and evaluate the system energy consumption and manpower cost. The present invention plans to deploy ten sets of LoRa SX1276 signal generating nodes, ten sets of passive backscatter tags (including sensors and energy management modules), two sets of USRP 2945R signal receiving devices, four sets of Dragino LoRa Gateway signal receiving devices, four sets of Intel 5300NIC transceiver devices, four sets of Wi-Fi devices, and four sets of R1.FID devices in the experimental area according to the experimental plan. The equipment is fixed at a position of 1.70m on the solar road pole, and a JJ-75D solar panel is installed on the top of the pole to provide power. At the same time, a 500×600×200mm outdoor stainless steel waterproof box is hung on the pole.

[0334] (6) Real application scenario deployment and prospects

[0335] Experimental location: Inside the office building,

[0336] Estimated equipment models: one set of USRP 2954R, one set of LoRa SX1276 signal transmitting nodes, one set of passive backscattering tags (taking the indoor non-line-of-sight communication distance of backscattering as an example for experimental explanation), LoRaSX1276 signal transmitting nodes and reflection nodes are placed in the room at a distance of 50m, 100m, and 150m, and the USRP receiving device changes the receiving position in several adjacent rooms. When the target is in the sensing area, the carrier is received and modulated by the antenna, and then scattered through the node to achieve indoor non-line-of-sight long-distance communication. The energy collection efficiency, information transmission accuracy and real-time performance of the system in this environment are measured. At the same time, the relationship between the scattering distance of the passive backscattering tag signal and the distance between the transmitting node and the scattering node is explored, and the communication quality, robustness, system energy consumption, and human cost are evaluated.

[0337] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A system for constructing a backscatter low-power communication network for wide-area communication, characterized in that: The system for constructing a backscattering low-power communication network for wide area communication includes: Radio frequency signal carrier detection and identification module, used to detect and identify frame structures of weak wireless carrier signals in the environment using a lightweight neural network accelerator; A low-power variable throughput modulation communication module for low-power variable throughput communication of passive nodes using code table-based frequency shift keying modulation; Scattering transmission module, used for kilometer-level backscattering transmission of weak RF signals based on the scattered field beamforming method; Environmental energy conversion module, which is used to convert and apply environmental energy based on efficient energy collection and global energy management circuit on metamaterial surface, so that passive nodes can work continuously; The lightweight neural network accelerator includes a data input part, a serial-to-parallel converter, a normalization part, a first layer of neurons, an activation part, a second layer of neurons and a parallel-to-serial converter; and is used to convert digital signals into modulated signals; The data input part is composed of an ADC and a cascaded D flip-flop group, which is used to convert analog signals into digital signals using the ADC, and use the D flip-flops to buffer data between different clock domains to eliminate metastable states during transmission; A serial-to-parallel converter is used to convert the 12-bit serial signal output by the ADC into a set of parallel signals; The normalization part is used to normalize the voltage value in a small range to the voltage value in a large range to enhance the signal strength; The first layer of neurons is used to calculate the normalized output data. When resetting, each weight is assigned a value. After each calculation, the count is automatically increased by one. When the count reaches 128 times, the calculation result is output. The activation part is used to use the ReLU activation function; The second layer of neurons is used to input weights during reset, multiply the calculation results of the first layer of neurons by the weights, and add them to the bias; A parallel-to-serial converter, used to convert the 45-bit data output by the secondary neuron into serial data output; The environmental energy conversion module includes: An energy harvesting unit for harvesting energy using a metamaterial efficient energy harvesting surface based on a complementary open resonant ring of an electromagnetic metamaterial unit; Energy optimization distribution unit, used for energy storage, release and optimization management; The energy optimization distribution unit comprises: An energy storage and release subunit, for storing and releasing energy using a multi-capacitor storage structure; A global energy management subunit, used for performing global energy management based on a finite state machine using a global energy management circuit; The global energy management circuit is composed of three logic units: a charger, an energy controller and a voltage output device; chargers to harvest energy from ambient RF signals and solar energy; an energy controller for managing a capacitor for powering a passive backscatter tag; Voltage output module, used to regulate the stored power to a constant voltage.

2. The system for constructing a backscatter low-power communication network for wide area communication as claimed in claim 1, characterized in that: The radio frequency signal carrier detection and identification module comprises: A radio frequency signal acquisition unit, used for collecting and converting radio frequency signals; The sample library building unit is used to collect LoRa signals in real environments at different distances outdoors and build a sample library; Dataset generation unit, used to build and enhance datasets; A model building unit, which builds a multi-layer perceptron neural network model as a training network and trains the built training network; The accelerator construction unit is used to build a sample library and train a training network, and obtain a lightweight neural network accelerator using an FPGA chip; The test unit is used to perform simulation analysis and testing on the obtained lightweight neural network accelerator.

3. The system for constructing a backscatter low-power communication network for wide area communication as claimed in claim 2, characterized in that: The radio frequency signal acquisition unit comprises: A collection subunit, used for collecting radio frequency signals; An impedance matching subunit, used for impedance matching of the collected radio frequency signal; A filter subunit, used for filtering the signal using a passive bandpass filter; The conversion subunit is used to convert the filtered radio frequency signal into a digital signal by using a driver, a filter and an analog-to-digital converter.

4. The system for constructing a backscatter low-power communication network for wide-area communication as claimed in claim 2, characterized in that The data set generating unit comprises: The meta-dataset construction sub-unit is used to identify and classify the collected samples using a standard commercial LoRa packet recognition algorithm, taking 1 / 4 of the LoRa Chirp signal as a sample, distinguishing the real LoRa Chirp and forming a data set; then the sampling function module in MATLAB is used to process the LoRa Chirp data set and the non-LoRa Chirp data set respectively, converting them into string form, adding labels for data calibration, converting the primary samples into positive and negative sample sets, and obtaining a small training set of 1000 entries, namely the meta-training set; The enhancement subunit is used to collect different types of wireless signals, work with LoRa2 at the same frequency and different link distances, mix the interference signal with the LoRa signal and environmental noise, generate synthetic data, and resample the synthetic data using a packet detection circuit, use the resampled data together with the synthetic data as a data set, and divide the data set into a training set and a test set.

5. The system for constructing a backscatter low-power communication network for wide area communication as claimed in claim 1, characterized in that: The low power consumption variable throughput modulation communication module comprises: An intermediate frequency signal synthesis unit, used for performing frequency synthesis using a low-power intermediate frequency signal generator based on wavetable query, i.e., a wavetable synthesizer; The signal modulation unit is used to load the collected data using FSK modulation, take the original carrier LoRa Chirp as the reference signal, and perform different frequency shift modulation on the reference signal to obtain different data.

6. The system for constructing a backscatter low-power communication network for wide area communication as claimed in claim 1, characterized in that: The scatter transmission module comprises: A scattered field antenna array construction unit is used to determine the antenna array optimization algorithm and obtain the optimal nine-element antenna array model; The beamforming unit is used to expand the 2D antenna array into 3D space and supports multiple beamforming angles.

7. The system for constructing a backscatter low-power communication network for wide area communication as claimed in claim 6, characterized in that: The scattered field antenna array construction unit comprises: The antenna array optimization algorithm selects sub-units to optimize the spacing in the antenna array using the algorithm; The spacing calculation subunit is used to establish a scattering communication system antenna array model using HFSS, and set the element antenna spacing of the scattering communication system antenna array model as a changeable variable; and is also used to obtain the optimal element antenna spacing by using MATLAB through a joint calculation using a differential evolution algorithm and HFSS; The optimization subunit is used to iteratively determine the optimal nine-element antenna array model in HFSS.

Images