HK40080127A - Adaptive radio configuration in wireless networks - Google Patents

Description

Adaptive radio configuration in a wireless network

Background

Low-power consumption remote wireless networks such as LoRa (remote) have become the mainstream of deployment of internet of things. These protocols support a variety of data rates and bandwidths in view of the versatility of the applications they support. However, for a given network deployment that can span miles, the network operator needs to specify the same configuration or a small subset of configurations for all devices in the network in order to communicate with each other. This all-round approach is extremely inefficient in large networks with hundreds of devices spanning miles, as it typically results in many, if not most, wireless devices connected to a base station (gateway) on a low-power remote network having a data transfer rate that is below the optimal data transfer rate.

Disclosure of Invention

There is provided a wireless network system comprising a base station device comprising processing circuitry configured to detect a transmission rate from a portion of a preamble of an incoming packet transmission signal and adapt a radio configuration to receive a remaining portion of the incoming packet transmission signal at the transmission rate.

This summary introduces a number of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Furthermore, the claimed subject matter is not limited to implementations that solve any or all disadvantages noted in any part of this disclosure.

Drawings

Fig. 1 shows a schematic diagram of a wireless network system according to an embodiment of the present disclosure.

Fig. 2 shows a schematic diagram of analysis of a transmission packet by the wireless network system of fig. 1.

Fig. 3 shows a schematic diagram of a base station apparatus reading the transmission packet of fig. 2.

Fig. 4A and 4B show diagrams depicting data rates and preamble structures of transmission packets such as those in fig. 1.

Fig. 5 shows a schematic diagram of a base station apparatus (such as the base station apparatus of fig. 3) configured with a software designed radio.

Fig. 6A-6C show diagrams illustrating a sampling method for sampling transport packets such as those in fig. 2.

Fig. 7 shows a diagram illustrating adaptive sampling of transport packets such as those in fig. 2.

Fig. 8A-8C show spectral diagrams of a chirp on LoRa included in a transmission packet such as those of fig. 2.

Fig. 9A-9C show diagrams illustrating data characteristics of a transmission packet used by the wireless network system of fig. 1.

Fig. 10 shows a coverage diagram illustrating combinations of propagation factors and bandwidths supported by the base station apparatus of fig. 3.

FIG. 11 illustrates a schematic diagram of a multi-stage artificial intelligence model used in the wireless network system of FIG. 1.

12A-12B show schematic diagrams of neural networks used in the multi-stage artificial intelligence model of FIG. 11.

Fig. 13 shows a graph depicting the accuracy of the wireless network system of fig. 1 and the accuracy of another system.

Fig. 14A-14D show graphs depicting the accuracy of the wireless network system of fig. 1 across different bandwidths, propagation factors, and locations.

Fig. 15A-15C show graphs depicting the accuracy of the wireless network system of fig. 1 across different locations and times.

16A-16B illustrate a flow diagram of a method according to an embodiment of the present disclosure.

FIG. 17 illustrates an exemplary computer environment in which the system of FIG. 1 may be implemented.

Detailed Description

To address the above-described issues, fig. 1 illustrates an example wireless network system 100 configured to allow network devices to transmit at any data rate. The wireless network system 100 includes a base station device 102 that uses the first few symbols in the preamble 107A1 of the packet transmission signal 107A to classify the correct data rate, switch the base station radio configuration, and then decode the data. The design of the present disclosure takes advantage of the asymmetry inherent in outdoor IoT deployments, where client power is insufficient and resources are limited, while base station device 102 (i.e., wireless gateway) is not. (the terms base station and wireless gateway are used interchangeably herein.) the wireless network system 100 disclosed herein is backward compatible with existing LoRa protocols and accurately identifies the correct configuration with an accuracy of over 97% in indoor and outdoor deployments.

Section 1: brief introduction to the drawings

Low Power Wide Area Networks (LPWANs), such as LoRaWAN, are becoming increasingly popular in large-scale IoT (IoT) deployments. Although LoRaWAN has just started, over 1 hundred million devices are in deployment, and this number is expected to exceed 7.3 million by 2023. LPWAN is lower power, longer communication distance, and lower cost than other mainstream solutions. These characteristics make devices equipped with such radios well suited for low throughput large scale networks in cities, agriculture, forests and many other industries.

To support remote and diverse device requirements, the LoRaWAN may operate at a variety of different data rates. As shown in fig. 4A, the data rate is configured using two parameters: bandwidth (BW) of chirp and propagation factor (SF) used in LoRa transmission. Fig. 4B shows a spectral diagram of a LoRa preamble with eight upper chirps and two lower chirps. The actual data rate also depends on the code rate used to ensure error correction. A fixed code rate is assumed. As expected, higher bandwidths may enable higher data rates. The propagation factor defines the time required to transmit a chirp, i.e. a higher propagation factor means that the time required to transmit the signal is longer and therefore the data rate is lower. Popular LoRa implementations can support bandwidths of 7.8kHz to 500kHz and propagation factors of 7 to 12 (on a logarithmic scale). Thus, a device transmitting at 7.8kHz and a propagation factor of 12 will achieve a data rate that is approximately 1189 times lower than a device transmitting at 500kHz and a propagation factor of 7.

Despite such a wide range of possibilities for devices in a network, the current paradigm requires that the system designer configure a single configuration setting (or a small subset of compatible configuration settings) of bandwidth and propagation factor for the entire network, i.e., the bandwidth and propagation factor of all devices are the same. Although LoRaWAN Automatic Data Rate (ADR) algorithms have been proposed, they may take hours to days to converge, have significant control overhead, and cannot handle multiple bandwidths. Thus, for example, in a farm network, the configuration of the network is typically such that the furthest away located device, such as a tractor, can be connected, even though most network connected sensors on the farm and even tractors are typically close to the closest wireless base station device of the network. This design choice stems from the need to limit network complexity and reduce control overhead to coordinate frequent data rate changes. However, this design option has three serious drawbacks, as described below.

Network throughput

LPWAN devices operate over large areas. The design coverage of a single LoRa gateway (LoRaWAN uses the gateway client mode of operation) is approximately 10 kilometers, and may cover up to thousands of devices. In such large-scale deployments, devices at the end of the range are almost unable to support lower data rates. Thus, this "all-round" design forces even devices that support high data rates to operate at very low data rates. This reduces the overall network throughput and reduces the number of devices that the network can support by as much as two orders of magnitude.

Deployment overhead

The optimal configuration of the gateway needs to be set by the network operator. Typically, this is accomplished by testing multiple configurations and selecting the configuration that is applicable to all client devices. This process requires technical labor and is not always available; for example, when deploying such devices for agricultural monitoring in remote rural areas. Second, the configuration selection needs to be dynamic. This configuration will stop working for a subset of devices over time and require frequent updates due to changes in the environment or incremental deployment of the devices.

Fluidity of the resin

IoT devices may be mounted on mobile vehicles, such as tractors, buses, or pick-up trucks. The optimal configuration varies with vehicle movement and is difficult to predict prior to movement. The lowest data rate configuration may be selected, but this may significantly reduce network capacity.

Shown here is a new wireless network system 100 that can support wireless devices 101 transmitting at different data rates. Each wireless device 101 transmits data for itself at the best possible data rate, which may depend on signal quality and application requirements, without the wireless device 101 having to inform the base station device 102 of the configuration of the wireless device 101 in advance, i.e. before starting wireless communication. The methods described herein do not require the wireless device 101 to transmit any control packets, do not require modification of the LoRa protocol, and are backwards compatible with existing devices (i.e., do not require any hardware modifications to the IoT wireless device 101 using the LoRa protocol).

As shown in fig. 5, wireless network system 100 uses a Software Defined Radio (SDR) as a wireless gateway in front of the LoRa transceiver of wireless base station 104. The SDR detects the preamble, identifies the bandwidth and propagation factor of the preamble signal, and tunes (i.e., adjusts) the radio configuration of the LoRa radio to the correct setting to receive the packet. This allows the wireless gateway to successfully receive packets from clients operating in any configuration. Since this method operates on a per packet level, it supports data rate changes due to client mobility and dynamic changes in the environment. In the wireless network system 100, there is a set of neural networks that use a small number of samples in the LoRa chirp to classify the correct radio configuration of each incoming packet transmission signal 107A at the base station.

The configuration of the wireless network system 100 disclosed herein addresses the following three technical goals and the challenges associated with achieving these technical goals in a practical deployment.

Sensitivity of the probe

To maintain the long range of the LoRa deployment, a first potential technical goal of the wireless network system 100 is to be able to operate at low signal-to-noise ratio (SNR).

Real-time operation

A second potential technical goal of the wireless network system 100 is to be able to detect packets and reconfigure the LoRa radio in real time, with enough time to correctly receive the rest of the packet signal, thus ensuring that packets are not lost.

Compatibility with existing deployments

A third potential technical goal is that existing deployments do not require protocol or client hardware changes to existing LoRa devices, although future generations of devices are not so limited.

The challenges associated with achieving these technical goals are described below and the system of the present disclosure is summarized in section 2, which is discussed subsequently.

The wireless network system 100 of the present disclosure employs a novel approach to address the basic rate adaptation problem in mobile networks. It does not require the client device and gateway to agree on the rate beforehand. One might want to be able to borrow an existing rate adaptation protocol, such as Wi-Fi, where the preamble containing the data rate configuration is sent using the lowest data rate. Such approaches are not suitable for LPWANs because they are primarily designed for large-scale deployment, where each client needs to send a small amount of data. Furthermore, since the data rate variation in LPWAN is higher than Wi-Fi, this results in very high overhead for packets sent at high data rates, but low data volume (one symbol at the minimum data rate is 1189 times longer than one at the highest data rate). Furthermore, this increases the hardware complexity of the client device design and does not directly account for the different bandwidths used by the client in LoRa.

The gateway of the wireless network system 100 is implemented using a Universal Software Radio Peripheral (USRP) SDR platform, which may be implemented in one implementation using an off-the-shelf LoRa chipset as a client. The wireless network system 100 has been evaluated in a variety of environments: desktop experiments with different signal strengths, indoor deployments across multiple rooms, and outdoor deployments. The results are summarized below.

In testing, the wireless network system 100 was configured with a detection algorithm that can detect the correct encoding parameters of incoming packet transmission signals in indoor, outdoor and desktop experiments with 99.8%, 95% and 98.2% accuracy, respectively. In contrast, the accuracy of the autocorrelation baseline reached 67.4%, 67%, and 78%, respectively.

The wireless network system 100 continues to operate effectively at low signal-to-noise ratios: an accuracy of 94% can be achieved even if the signal attenuation exceeds 140 dB.

The algorithm of the wireless network system 100 can be effectively generalized to a new environment and continuously run in a dynamic environment. In a five day experiment, the accuracy of the wireless network system 100 was consistently over 99%, with little change per day.

Finally, it should be understood that the wireless network system 100 disclosed herein may be applied to future generations of devices. With the development of neural networks and the development of faster hardware implementations, the algorithms described herein may be applied to shift only the rate adaptation burden to the power connected base station infrastructure (i.e., to the gateway/base station device 102, rather than requiring coordinated configuration of both the base station device 102 and the mobile wireless device 101 as is the current case), thereby mitigating the rate configuration overhead for battery powered mobile devices. Thus, the methods described herein are not limited to application to low power wide area networks, but may be applied to various other types of wireless networks, including high speed networks, such as the so-called sixth generation (6G) wireless networks currently being developed.

Section 2: challenge(s)

As noted above, the wireless network system 100 disclosed herein is intended to achieve the triple goals of sensitivity, real-time operation, and compatibility. However, these goals are inherently challenging.

First, the sensitivity will be discussed. The sensitivity of the LPWAN protocol is directly related to bandwidth. The lower bandwidth signal is less noisy and can therefore be received at a lower signal strength. Conversely, higher bandwidth signals require higher signal strength at the receiver for proper decoding. Thus, if the wireless network system 100 configures its SDR to operate at a low bandwidth, it will meet the sensitivity requirements, but will miss signals received at higher bandwidths. On the other hand, if the wireless network system 100 sets its bandwidth too high, it may miss signals from a greater distance (and thus reduce signal strength) at a lower bandwidth.

Secondly, to ensure real-time operation, the SDR preferably uses only a few symbols to identify the correct configuration of the packet. However, the length of the symbol itself depends on the configuration used by the transmitter. The symbols transmitted using the propagation factor 12 are 64 times longer than the symbols transmitted using the propagation factor 6. If the signal sampling time is too long, there is a risk of losing the entire packet for the sender with the highest data rate. On the other hand, if the signal sampling duration is too short, there may not be enough information to identify the correct coding parameter configuration for the low data rate transmitter.

Finally, to ensure backward compatibility, it is desirable for the wireless network system 100 to receive the entire packet after the correct configuration is set up at the gateway. However, this requires identifying the configuration before the signal reaches the gateway-a task that seems unlikely. These challenges are illustrated in fig. 6. This figure illustrates the challenges associated with configuring the receiving SDR itself. The figure shows the chirping of three different configurations, which are relatively close to each other. Although there are more serious differences, it is difficult to visually represent such more serious differences on these figures because of the large scale differences. As shown in fig. 6A, sampling one symbol of maximum bandwidth captures only a small portion of the high bandwidth signal and reduces sensitivity. On the other hand, if one symbol length is sampled for the low data rate configuration in fig. 6B, high sensitivity is maintained, but it may introduce significant delay for the high data rate symbol(s). Finally, one might want why all possible configurations do not use the minimum of frequency bandwidth and symbol duration. This will ensure sensitivity to low signal strengths and real-time operation. However, such a configuration eventually misses some configurations, as shown in fig. 6C.

To resolve the conflict between sensitivity and real-time operation, the wireless network system 100 employs an adaptive approach. It samples a small block of bandwidth in a short time using a bank of bandpass filters in the digital domain. It uses the bandwidth and frequency of these tiles to decide whether it has captured long enough to decide on the configuration, or whether a longer sampling time is needed. In any case, the wireless network system 100 does not use more than two symbols for any configuration to make this decision. This idea is illustrated in fig. 7.

Finally, to be compatible with existing hardware, the gateway needs to receive the entire packet after configuration. Since the SDR uses at least a portion of the preamble to identify the gateway, this goal is at first glance difficult, if not impossible, to achieve. One way to address this problem is to buffer the time samples at the transmission rate determination gateway 105 of the wireless network system 100 and then replay them at the radio 106 of the base station 014. However, this complicates the circuitry of the wireless network system 10 and increases its cost. Instead, to address this challenge, the structure of the preamble in the LoRa protocol is utilized. Important to the operation of the system is an operating principle that can dynamically configure the preamble length of a packet. One necessary operating principle is that the dynamically configured preamble can be larger than the preamble length required for the base station radio to detect the packet. The wireless network system 100 may use the remaining symbols to determine configuration parameters and set these parameters on the base station radio. For example, the base station may be configured to expect a preamble of 8 symbols, but the client may be configured to use 10 symbols. The base stations of wireless network system 100 may be assigned these two additional symbols to predict the coding parameters of the incoming packet transmission signal and reconfigure the LoRa base station to correctly receive the signal according to the coding parameters. The gateway can then use the remaining signal to decode the packet. Note that since the number of up-chirps is variable, the gateway can still see a complete preamble with a sequence of up-chirps followed by two down-chirps and can successfully decode the packet.

Section 3: loRa

LoRa is a physical layer implementation of LPWAN based on Chirp Spread Spectrum (CSS) technology. In LoRa modulation, a chirp signal is generated to encoded data symbols. As shown in fig. 8, the frequency of the chirp varies linearly with time. Two parameters define the effective data rate: bandwidth and propagation factor. The bandwidth controls the total span of chirp in the frequency domain. The propagation factor defines the length of time in the time domain for each chirp. In particular, for chirps with a propagation factor SF, the time taken for transmission is equal to 2 ^SF Is in direct proportion.

Thus, the time T required for transmitting the chirp _s ByWhere BW is chirp and SF is the propagation factor. Thus, a higher bandwidth will shorten the duration of each chirp, while a higher propagation factor will exponentially increase the duration of each chirp.

To transmit a bit of information, the transmitter modifies the initial frequency f of the chirp. Specifically, to send the symbol value S, the transmitter sets the start frequency to:

LoRa allows S to be taken as {0,2,, 2 } ^SF A value within the range. Thus, a chirp conveys a symbol that conveys SF bits. Thus, the effective data rate R for the LoRa transmission is:

as shown in equation 2, an increase in bandwidth increases the data rate. A decrease in the propagation factor increases the data rate. One might want to reduce the rate if the SF is higher, why it is used. This is because a higher SF also increases the duration of the symbol, making it easier to decode correctly.

Finally, the terminology of the remainder of this document is reiterated. The symbol is the unit of data transmitted by each chirp. The duration of the symbol is the same as the duration of the chirp. Each symbol or chirp consists of a number of samples depending on the sampling rate and the sampling duration. For example, for a sampling rate of 106 samples per second, a symbol duration of 2 milliseconds would correspond to 2000 samples.

Section 4: wireless network system

The wireless network system 100 disclosed herein is a new gateway design for LoRa that supports dynamic link configuration. Using the wireless network system 100, a client can optimize its data rate without notifying the wireless base station 104 of its updated configuration. This, in turn, allows a single wireless base station 104 to support hundreds of wireless devices 101 on a large scale without impacting performance. For example, a LoRa network deployment may include client devices dispersed within a radius of several miles from base station device 102. The achievable throughput varies with distance and different channel conditions across this coverage area. Wireless network system 100 enables a LoRa network to support multiple configurations that would otherwise have to sacrifice performance to support all devices in a wide coverage area.

To better understand the performance of LoRa, a range test is performed to determine the maximum achievable data rate associated with the range of base station apparatus 102. Fig. 10 shows an overlay of the best configuration settings that can be supported while maintaining a reliable communication link between the LoRa base station and the client. In an industrial campus setting, the base stations are at fixed locations and client locations are changed throughout the campus. The client wireless device continuously transmits the LoRa packets at 20dB transmission power and changes the encoding parameters at each location to test the limitations of the system. Fig. 10 shows the maximum supported data rates across all locations and corresponding BWs and SFs. The key point is that there is a large difference between the supported coding parameters, which justifies the desire to support more dynamic networks.

The wireless network system 100 accomplishes this by employing a neural network approach to predict the bandwidth and propagation factors for data transmission by any given client. In turn, the base station device's radio will reconfigure accordingly to properly receive and decode the incoming packets. One view of the architecture of the base station apparatus 102 of the wireless network system 100 is shown in fig. 5. As shown, the base station apparatus 102 includes a transmission rate determination gateway 105. The transmission rate decision gateway 105 is SDR as described above and has three components: a packet detection module 110 (packet detector) for detecting incoming LoRa packet transmissions, a classifier 112 (which may be a neural network processing unit as described below) for classifying the coding configuration, and a last radio configuration module 114 that communicates with the LoRaWAN wireless base station 104 to update the coding parameters. Although "base station" is included in the name of the LoRaWAN wireless base station 104 and "gateway" is included in the name of the transmission rate determination gateway 105, it is understood that both are included in a single device as the base station device 102 and also serve as a gateway of the WAN when connected thereto.

Fig. 1 depicts at 100 a general depiction of a wireless network system thus described, in which the base station apparatus 102 depicted in fig. 5 may be deployed. As shown, wireless network system 100 includes a base station device 102, which base station device 102 is configured to communicate with a plurality of wireless devices 101 (e.g., loran configured devices) via a wireless network 108 (e.g., loran network) using signals 107. The base station device 102 is configured to operate as a gateway device for a Wide Area Network (WAN), such as the Internet, through which the base station may communicate with remote devices, such as remote servers and remote clients.

The base station apparatus 102 comprises a processing circuit 103, the processing circuit 103 being configured to detect a transmission rate from a portion of the preamble 107A1 of the incoming packet transmission signal 107A and cause its radio 106 to receive a remaining portion 107A2 of the incoming packet transmission signal 107A at the transmission rate. Base station device 102 is configured to implement a low-power wide area network and to transmit an incoming packet transmission signal 107A from wireless device 101 to base station device 102 according to the LoRaWAN communication protocol. Thus, in this example, the incoming packet transmission signals 107A are transmitted from the plurality of wireless devices 101 using the LoRaWAN network protocol, although other network protocols may be used. For example, other low power remote protocols may be used, or high speed network protocols (such as 6G), or other suitable network protocols may be used. In this example, three wireless devices 101 are shown communicating with the base station device 102, but it will be understood that up to thousands of wireless devices 102 may communicate with the base station device 101.

Continuing with fig. 1, the transmission rate determination gateway 105 (SDR, as described above) of the base station device 102 also includes a packet detection module 110 that implements an adaptive sampling algorithm to collect samples of the preamble 107A1 of the incoming packet transmission signal 107A, the incoming packet transmission signal 107A received by the receiver 115 of the base station device 102 from one of the plurality of wireless devices 101. The transmission rate determination gateway 105 of the base station device 102 further comprises a classifier 112, which may take the form of a CNN, configured to receive the samples and output a classification 117 indicative of one or more encoding parameters of the incoming packet transmission signal. In this example, the coding factors are bandwidth and propagation factors, however, in other examples, other coding factors may be used. Transmission rate determination gateway 105 also includes radio configuration module 114 that sends configuration commands to configure radio 106 of wireless base station 104 to receive remaining portion 107A2 of incoming packet transmission signal 107A according to one or more encoding parameters (such as bandwidth and propagation factor) indicated by classification 117. The process described in this paragraph is also illustrated in fig. 3, which shows that the preamble 107A1 is processed by the adaptive sampling algorithm to produce samples corresponding to the initial symbols in the preamble, and then processed by the classifier 112 to produce a classification 117 indicative of the encoding parameters, which in turn is used to configure the radio 106 to correctly receive the remaining portion 107A2 of the incoming packet transmission signal 107A.

There are three technical challenges to the implementation of the wireless network system 100. First, wireless network system 10 presents a challenge to determine configuration parameters for received packets in near real-time. Second, the wireless network system 100 faces the challenge of backward compatibility with existing LoRa solutions. Third, the challenge faced by the wireless base station device 102 of the wireless network system 100 is how to achieve high prediction accuracy through the various possible coding parameters selected by the wireless device 101. The following sections detail how wireless network system 100 addresses each challenge and describe the architecture of a neural network that can be used to implement classifier 112.

4.1 real-time prediction

In order to successfully decode the incoming packet transmission signal 107A, the base station device 102 needs to configure its radio 106 with parameters that match the incoming packet transmission signal 107A. This reconfiguration needs to be done quickly enough so that the radio 106 still has time to detect the incoming packet transmission signal 107A. To detect the incoming packet transmission signal 107A, the radio 106 requires the preamble 107A1 of the incoming packet transmission signal 107A.

As described in section 2, the wireless network system 100 uses the overhead symbols added to the LoRa packet preamble 107A to determine the configuration parameters and sets these parameters at the radio 106 of the base station device 102. To validate this approach, two Semtech SX1276LoRa chips were configured as the base station apparatus 102 and the wireless apparatus 101, respectively. The API of the popular LoRa chipset (Semtech SX 1262/1276) is used to configure a 6-65535 symbol LoRa packet preamble, where packet detection requires at least 6 symbols. Preamble 107A1 at base station device 102 is set to 8 symbols while changing the wireless device 101 preamble length. The wireless device 101 transmits the packet over the air using different preamble lengths and then verifies receipt of the packet at the base station device 102. The result shows that additional five symbols can be added to the preamble 107A1 of the wireless device 101 while still maintaining reliable reception by the base station device 102. The wireless network system 100 requires a maximum of two symbols depending on the encoding parameters used for the input data. This variation is because the input shape of the data is consistent for any neural network. The number of symbols used for any given input will also vary depending on the number of data samples that are classified by the incoming network, since the duration of a symbol is a function of BW and SF.

4.2 inferring SF and BW

As described above, the encoding parameters are not pre-negotiated between the wireless device 101 and the base station device 102 until the base station device 102 receives the incoming packet transmission signal 107A. It will be appreciated that in the wireless network system 100, the wireless device 101 is configured to set the encoding parameters (such as bandwidth and propagation factor) to values selected from a plurality of preset values of the encoding parameters at the wireless device 101. These preset values typically include all possible values that the network protocol (e.g., loRaWAN) defines as being available, and are typically not a subset of such possible encoding parameters that the network administrator sets in the configuration step. Once the wireless device 101 autonomously selects the encoding parameters, the wireless device 101 is configured to begin transmitting the incoming packet transmission signal 107A in accordance with the encoding parameters to pre-negotiate the encoding parameters without any prior communication with the base station device 102.

The wireless network system 100 is intended to predict the propagation factor and bandwidth of the LoRa packet transmission using a neural network method. Before delving into the network architecture, we will first describe why BW and SF can be inferred. By comparing the number of samples per symbol, differences between particular combinations of BW and SF can be easily distinguished. However, in some cases, the total number of samples is the same (e.g., BW =125khz, sf =8 and BW =500khz, sf = 10).

One way to distinguish between encoding configurations is to first compare the frequency increase of any given chirp with respect to time. This will provide insight into the propagation factor. Second, the start and stop frequencies of the chirp can be used to determine the bandwidth. Referring to fig. 8A and 8B, it is shown that the rate of change of frequency varies with the propagation factor, and the difference between the start frequency and the stop frequency results in a bandwidth for chirp. This technique is sufficient if the entire symbol duration is used to predict the parameters, but doing so can add significantly to the delay, since the duration of a single symbol can be up to 525ms. Thus, the number of samples used to determine the encoding parameters will be minimized.

The previously described method can still be used to infer the propagation factor and bandwidth using a subset of samples of the LoRa preamble symbols, but the trade-off in this case is accuracy. It may become more difficult to distinguish between different coding parameters in view of the changes in RSSI and SNR that a signal may experience when transmitted over the air. The wireless network system 100 takes into account the characteristics of the LoRa chirp to train the Convolutional Neural Network (CNN) to classify a number of different combinations of propagation factors and bandwidths. Specifically, classification is performed using three features extracted from the symbols of the LoRa preamble 107A1.

It should therefore be appreciated that as shown in fig. 2, an analog-to-digital converter 111 may be included in the wireless network system 100 that is configured to sample incoming transmission signals at different rates under the control of an adaptive sampling algorithm 113 implemented by the packet detection module 110. As described in the following section, samples taken from two or fewer symbols are typically used by an artificial intelligence model of the classifier 112 to output a classification 117. Thus, in particular, the samples comprise samples extracted from two symbols (e.g., symbol (0) and symbol (l)) in the preamble 107A1 of the incoming packet transmission signal 107A, and the artificial intelligence model of the classifier 112 determines the classification 117 using a plurality of extracted features of the samples, the plurality of features comprising real parts of the samples, imaginary parts of the samples, and a fast fourier transform of the samples.

The first two features are the real and imaginary parts of the signal, and the last is the Fast Fourier Transform (FFT). Using data in both the signal time and frequency domains is crucial to achieving high prediction accuracy. For example, if only the FFT of each signal is used, it is almost impossible to distinguish very low BW settings. As shown in fig. 9, the lower kHz range starts to look very similar when evaluating FFTs of different bandwidth and propagation factor settings. Supplementing this with time domain characteristics helps to capture the variation in the preamble symbol oscillation frequency, while the FFT provides insight into the bandwidth variation.

4.3 adaptive sampling

The wireless network system 100 uses an adaptive sampling method to optimize sensitivity, delay, and classification accuracy. Returning to fig. 7, an adaptive sampling method is illustrated. A bank of digital bandpass filters is used to create a subset of the bandwidth in a short time.

The adaptive sampling algorithm 113 shown in fig. 2 is configured to filter the incoming packet transmission signal 107A using one or more band pass filters, thereby generating a plurality of filtered incoming packet transmission signal components, and determine that the captured signal is sufficient to determine one or more encoding parameters for one of the filtered incoming packet transmission signal components.

These subsets are used to determine whether the acquired signal is long enough to provide accurate information about the radio configuration, or whether sampling should continue. In particular, the wireless network system 100 uses a total of 12808 samples (65 ms) for the first six classes representing the two low bandwidths and 800 samples (4 ms) for the last nine classes representing the higher bandwidth radio configuration. Intuitively, it makes sense to use a larger set of samples for lower bandwidth settings, since symbol duration increases with decreasing BW.

4.4 classifier architecture

The classifier 112 shown in fig. 1 can be implemented as an artificial intelligence model that includes at least one convolutional neural network, and can use a hierarchical neural network architecture that includes multiple (e.g., two) stages. 11-12B, the artificial intelligence model can be a multi-level model and thus can include a first level and a second level. The first stage may include a bandwidth classifier that includes a first convolutional neural network that classifies the incoming packet transmission signal into one of a plurality of bandwidth range classifications (e.g., high-range and low-range). A mid-range between the upper and lower ranges may also be defined. At a second stage, for signals with bandwidths below a predetermined threshold, the signals are classified into one of a plurality of low bandwidth code classes by a low bandwidth code classifier comprising a second convolutional neural network, and for information above the predetermined threshold, the signals are classified into one of a plurality of high bandwidth code classes by a high bandwidth code classifier comprising a third convolutional neural network.

Continuing with the embodiment shown in FIG. 11, a binary classifier is first used to distinguish between lower and higher bandwidths. Based on the prediction, BW and SF radio configurations will next be predicted using either a six-class or nine-class classifier. Fig. 12 illustrates the neural network architecture used by the wireless network system at each stage, where the key difference is the number of classes, features and samples input per classifier.

The low bandwidth classifier relies on the three features described previously. Binary and nine-class classifiers use 30 features to predict the radio configuration. The features include a real part, an imaginary part and an FFT; however, the samples are divided into ten blocks of 20kHz. As discussed previously in section 2, the variation in the number of samples and features of each classifier is selected according to the type of signal that needs to be classified. For example, a low bandwidth classifier uses more than 16 times more samples, since the symbol duration may be tens of milliseconds, requiring more samples to have meaningful features. On the other hand, even for low bandwidths, the binary classifier uses only 800 samples, but this is sufficient since it does not need to distinguish between individual bandwidths.

The neural network for each classifier starts with four convolutional layers, each layer having a filter size of 128. These layers convolute the input and are activated by a rectifying linear element (ReLu) function. The ReLu activation function outputs the maximum value zero and the input data and provides the output in the form of a feature map. The following is a max pooling layer to reduce the size of the generated feature map and retain the most meaningful information. In this network, the maximum pooling size is 2. Next are six convolutional layers, each layer having a filter size of 128-32. These layers also use the ReLu activation function. After which a global average pooling layer is added that calculates the average output of each feature map in the previous convolution layer. The size of the final dense layer is equal to the total number of possible classifications. The dense layer uses a sigmoid activation function that provides output probabilities for all classes between a value of 0 and a value of 1. To retrieve the prediction classes, the maximum probability of the final output layer is taken.

To evaluate the effect of neural networks on modeling data sets, a categorical cross-entropy loss function was used,

where N is the number of classes, p is the prediction probability of the current sample, and t is a binary indicator of whether class c is correct. The loss function evaluates the performance of the classification model with an output probability between 0 and 1. In other words, if the model prediction deviates from the actual value, the cross entropy will increase, thereby providing an error measure. In order to obtain an accurate prediction, it is also necessary to minimize the error using an optimization function. At a higher level, the optimization function computes the partial derivatives of the loss function with respect to the weights used in the model. These weights are modified until a minimum value of the loss function is reached. The 100 network architecture of the wireless network system uses an Adam optimizer to perform this task.

Three batch normalization layers and one discard layer are also added to the network. The batch normalization layer normalizes the output of the previous layer by subtracting the batch mean and dividing by the batch standard deviation, where the batch is part of the data passed into the model for training. Batch standardization improves the stability of the network and helps reduce the time required to train the network. Finally, for normalization, a 0.5 step is used before the final dense layer to reduce overfitting.

Section 5: to realize

The implementation of the wireless network system and the setup for experimental evaluation are described in detail below.

5.1 hardware

A hardware prototype of the wireless network system 100 gateway was designed using the Universal Software Radio (USRP) platform. The operating frequency of the wireless network system 100 gateway is 915MHz, which is the frequency used by most LoRa deployments in the united states. The USRP is co-located with the LoRa receiver and needs to be configured to the correct configuration to successfully receive the packet.

The client uses a 1276Semtech chipset design. The chipset allows a propagation factor of 7 to 12 and a bandwidth of 7.8kHz to 500kHz. Bandwidths of 10.4kHz, 15.6kHz, 125kHz, 250kHz and 500kHz were selected for experiments to cover the extreme ends of the spectrum. Note that by selecting the two lowest bandwidths possible, the smallest difference between the bandwidths can be used. Finally, a spreading factor of 10 to 12 was used in the experiment.

The client chip is embedded in the PCB for setting the propagation factor and bandwidth and allowing transmission of data bits. The chip is controlled using an ARM STM32L151 microcontroller. Custom firmware has been written for this microcontroller. The wireless network system 100 may use any such implementation on the client side without any modification.

5.2 software

The wireless network system 100 gateway is controlled using GNU Radio software. The software was run on a computer with 32GB RAM and samples were collected at a center frequency of 915MHz and a sampling rate of 200 ksps. This is the minimum sampling rate that USRP can achieve and results in a 200kHz bandwidth at receiver 115. Each packet record is passed through a band pass filter to further reduce the receiver bandwidth to 20kHz. Additional filtering is performed to increase the sensitivity of the receiver 115. The samples are then sliced into individual symbols using a packet detection algorithm that uses a combination of power threshold over a sliding window and autocorrelation.

CNN was implemented using the tensoflow 2.0 framework in Python. It runs on Microsoft Surface 2 for 16GB RAM and NVIDIA GeForce GTX 1050GPU for 2GB memory. The CNN was trained using Adam optimizer, except that the learning rate was set to 0.0001 and the parameters were default. 20% of the training set is set as the validation set. All experiments performed 20 cycles of training on the models and the best model was selected based on the validation set performance. Unless otherwise noted, each experiment was performed on three different training test segments. The next section specifies the number of training points for each experiment.

Section 6: results

The empirical evaluation of the wireless network system 100 is as follows.

6.1 Experimental setup

To evaluate the wireless network system 100, a data set was first generated to represent 15 possible classifications of propagation factors ranging from 10-12 with bandwidths of 10.4, 15.6, 125, 250, and 500kHz. Since the preamble of the LoRa packet is a series of chirps, a data set is created that consists of a single chirp in the form of a complex baseband signal extracted from the preamble of each packet. The radio described in section 5.1 is used to transmit the LoRa packets and receive using USRP. With this setup, data is collected in controlled, indoor and outdoor environments.

Indoor data collection: indoor experiments were performed in the office space. The experiment spanned six different rooms, with a total area of 1000 square feet. A transmitting device (e.g., wireless device 101) and a receiving device (e.g., base station device 102) are randomly placed in different rooms. In each setting, data for each class is collected. For each position, 800 symbols of data are collected on average per class.

Collecting outdoor data: to simulate an outdoor deployment, data is collected using a campus-scale deployment. The receiving device is placed in a fixed position on the ground level. The transmitting device may be moved manually or on top of a car to a different location within the campus area of 0.02 square miles. For each location, a random propagation factor and a random bandwidth are selected to transmit data. The position and the GPS coordinates of the used configuration are recorded manually. A total of 16 campus location data were collected.

Desktop data collection: to replicate the remote outdoor experiments, a bench-top experimental setup was used to create controlled data sets with different RSSIs (receiver signal strength indicators). In this arrangement, the transmitting device and the receiving device are directly connected by a wire. For each symbol class, a variable attenuator is used to attenuate the transmitted signal by 40-140 dB.

Baseline: a baseline based cross-correlation operation is used. An example set is used that contains one example signal (bandwidth and propagation factor pair) for each class. For a given signal input S,is example E in S and class i _i Is then calculated as the similarity score of class i

Finally, the highest scoring class is assigned to the input. Notably, this is a computationally intensive process. Cross-correlation is an O (N log (N)) operation, where N is the length of the signal and needs to be performed for each class.

6.2 accuracy assessment

First, the accuracy of the correct configuration of CNN identification packets of the wireless network system 100 is evaluated. As previously described, for CNN of wireless network systems, the raw signal is captured for 4ms and used as input to the binary classifier. If the received packet belongs to the low bandwidth class, the signal acquisition will increase to 65ms, otherwise for high bandwidth the signal acquisition remains unchanged. This corresponds to the two chirp (or symbol) durations (bandwidth 500kHz and propagation factor 10) of the highest data rate in the experiment, and about 1/6 chirp duration of the lowest data rate. The performance of the neural network was evaluated by analyzing the accuracy of all three scenarios described above. Due to limitations, the analysis uses a mix of indoor and desktop data to train the network. 30% of the data collected was used for training and all other data was used for testing.

Turning now to fig. 13, the accuracy of the wireless network system 100 will be described. As shown, the 100CNN of the wireless network system achieves very high overall accuracy of 99.8%, 95%, and 98.2% in indoor, outdoor, and desktop evaluations, respectively. This high accuracy demonstrates the feasibility of the core idea of the wireless network system 100 that the correct configuration of packets on the gateway can be identified with high accuracy. In contrast, baseline performance was significantly poorer. For these three settings, the baseline accuracy was 67.5%, 67%, and 78%, respectively. One reason for poor baseline performance is the difficulty in identifying small differences in frequency bandwidth, such as 10.4kHz and 15.6kHz. Unlike the higher bandwidths of 125kHz and 250kHz, these bandwidths are relatively close and there is noise and multipath, so it is difficult to distinguish them.

A change across the environment.

FIG. 13 also demonstrates changes across environments. The system performs better outdoors than indoors. This is mainly because outdoor environments contain more free space and less multipath fading than indoor environments. On the other hand, indoor environments have more multipath reflections, making them more challenging.

A change across bandwidth.

Fig. 14A illustrates the performance of the wireless network system 100 across different bandwidths. For this experiment, the call was reported as it is more meaningful. The call is the number of points correctly classified as bandwidth B divided by the number of points actually transmitted at bandwidth B. As shown, the call rate for all bandwidths remains around 99%, with a minimum of 98.4% (15.6 kHz), and a maximum of 10.4kHz and 125kHz, approaching 100%.

Variation across propagation factors.

Fig. 14B depicts performance variation of the wireless network system 100 over different propagation factors. As shown, the modulation rate remains around 99% for all three propagation factors. The call rate is slightly lower because the propagation factor is highest. This is mainly because the highest propagation factor corresponds to the maximum time of each chirp. This means that if the sampling occurs at a fixed duration, as it is for the duration of the input, then the minimum chirp fraction of the highest propagation factor is obtained. This makes the classification problem more challenging as the propagation factor increases. However, even at the highest propagation factor used by LoRa, wireless network systems can achieve greater than 95% accuracy, using less than half of a single chirp duration. This demonstrates the powerful performance of CNN designs for wireless network systems.

A change in cross-position.

Fig. 14C depicts performance variations of the wireless network system 100 in different physical spaces. L0 to L4 represent four different positions. The accuracy of the wireless network system 100 is always maintained around 99-100% in each of these locations.

Change over time.

Fig. 14D plots the performance change over time for all 15 types of wireless network systems 100. In this experiment, data was collected over the air for 30 minutes for 5 consecutive days. As shown, the accuracy remains almost 100% on all days. Compared with the baseline method, the accuracy is remarkably reduced to about 88%.

One important finding of accuracy analysis is that the wireless network system 100 is able to correctly identify radio configurations in different scenarios with high accuracy. The overall accuracy of the wireless network system reaches 97.7%, namely the packet loss rate is less than 1/20. This loss becomes insignificant considering the overall packet loss for LoRa. In an outdoor urban scenario, with a bandwidth of 125kHz and a propagation factor of 12, the packet loss for a distance of 0-15km may be between 12% and 74%. It is believed that the parasitic losses incurred by the wireless network system 100 are ultimately a reasonable tradeoff for achieving automatic radio configuration.

6.3 overview

One problem that arises with most machine learning frameworks is that they can generalize to new environments that are not seen in the training set. This problem is addressed in wireless network systems by two empirical evaluations.

First, the model is trained while excluding two locations (different rooms in the indoor environment) from the training data. Specifically, the data obtained from L5 and L6 are not included in the training set. The data from these two locations is split for the test set. This allows testing the versatility of the new environment. The results of this experiment are plotted in fig. 15A. As shown, the positioning accuracy decreased slightly, from 98.9% to 94.5%.

Second, the model is tested for generalization across time. Test data was collected on a day not included in the training set (one week apart). The model maintained performance for the previous days (97% accuracy). This indicates that there was some positional difference in accuracy, but no temporal difference was observed. A key conclusion of this result is that the wireless network system can achieve high accuracy even for input signals of scenarios not encountered by CNN. This indicates that the CNN of the wireless network system can be used for a variety of LoRa networks.

6.4 sensitivity

Depending on the SF and BW settings used, the LoRa can operate in a sensitivity range of-149 to-118 dBm. In order for wireless network system 100 to be valuable for LoRa network deployment, it must also be able to achieve high accuracy within the same sensitivity range. To evaluate the accuracy of CNN of wireless network system 100 for low power signals, a data set was generated with a 40-140dB attenuation and the accuracy of the model was analyzed. Fig. 15B shows model accuracy as a function of wireless network system 100 and baseline method decay. The average accuracy of the wireless network system 100 is 96.7% up to 99%, independent of fading. This is consistent with the accuracy obtained for the entire bench-top experiment reported in fig. 13. On the other hand, the accuracy of the baseline method fluctuates and decreases for signals exposed to high attenuation amounts. The overall results show that the wireless network system 100 is robust to changes in signal strength, and thus should be able to maintain the accuracy of the prediction of the signal conditions that the LoRa may be exposed to.

6.5 delay

Minimizing the delay of the wireless network system 100 is critical to maintaining real-time predictions. As described above, it is determined that five additional symbols may be added to the preamble 107A1 of the LoRa packet transmission, which may be allocated to the wireless network system 100 to detect, classify, and update the radio configuration at the base station apparatus 102. This means that the duration varies from 0.01s to 1.92 s. The wireless network system 100 uses a maximum of two symbols per class (most classes are less than one symbol) and the remaining time is available for classification and parameter configuration. CNN delay of wireless network system 100 is evaluated and compared to a baseline method. Fig. 13C shows a comparison of the delay between the two methods. It is worth mentioning that the wireless network system 100 using the CPU for classification needs about 60ms per sample, and the wireless network system 100 using the notebook version NVIDIA GTX 1050GPU for classification needs about 3ms per sample, and the delay can be increased by 20 times. The baseline method has a computation time of 140ms per sample, and thus cannot classify most LoRa coding parameter configurations in real time.

Section 7: overview of the System

The present disclosure describes a new gateway design that allows wireless devices 101 using LoRa and other protocols to transmit at their selected data rates. This enables, for example, base station device 102 to support remote large-scale mobile wireless devices 101 without sacrificing overall network performance. Wireless network system 100 uses CNN to predict the bandwidth and propagation factor of packets transmitted by wireless device 101 and enables base station device 102 to decode the packets across different signal coding parameter settings, thereby quickly configuring, in one example, radio 106 of base station device 102 to correctly receive the remaining portion 107A2 of incoming packet transmission signal 107A based only on information from the first two symbols of preamble 107A1.

The test implementation of the wireless network system 100 includes the following component features.

Classifier for LoRa radio configuration

According to the test results, a neural network was implemented that could classify 15 different LoRa radio configurations with 99.8% and 95% accuracy for indoor and outdoor scenarios, respectively.

Real-time classification

Tests have shown that by using dynamic preamble settings for the LoRa packet, radio configuration can be automated and performed in real time. The wireless network system 100 performs high-accuracy classification in different scenarios depending on at most two preamble symbols.

Adaptive sampling

Adaptive sampling is implemented to optimize the trade-off between sensitivity, accuracy and delay of the network. Wireless network system 100 adjusts bandwidth and acquisition duration to classify the large number of radio configurations supported by LoRa.

Although a particular application of the disclosed wireless network system 100 is described herein, it should be understood that the wireless network system may be used in other applications. Examples of such applications are described below.

Rate adaptation

Wireless network system 100 may be used to improve the rate adaptation technique of LoRa. Many typical overheads may be avoided because the client can configure its own encoding parameters and the wireless network system 100 may automatically configure the base station device 102 to meet the requirements. For example, control messaging between the base and client may be minimized. Developing new rate adaptation protocols based on the wireless network system 100 may further improve the performance and efficiency of LPWANs.

Field programmable gate array implementation

Although not shown in the drawings, the wireless network system 100 may be implemented on a Field Programmable Gate Array (FPGA). Compared to other hardware computing platforms, FPGAs provide faster performance and also provide flexibility to support different algorithms, logic and memory resources. Such an implementation may help improve latency of the wireless network system 100 by minimizing the time required to detect, classify, and update radio parameters.

Alternative hardware

While the system was developed as a gateway and equipped with software defined radios, some off-the-shelf gateways (e.g., SX 1257) support accessing raw IQ samples of a signal and are compatible with design.

Network pruning

Pruning the networks used by wireless network system 100 is a promising approach for improving latency. The idea behind network pruning is that since there are many parameters in the network, there must be some parameters that are redundant and do not contribute much. This minimizes the size of the network and thus optimizes the time required to perform classification.

With the ultimate definition of the 5G standard, there is an increasing interest in defining 6G networks with the goal of an order of magnitude improvement over 5G in bandwidth and delay. One promising approach being explored is machine learning, and how devices automatically reconfigure to communicate with devices, including devices using other standards. This may significantly reduce control overhead, thereby increasing network capacity. The wireless network system 100 architecture is a step towards the vision that full interoperability is achieved while still maintaining backward compatibility with legacy devices. Accordingly, the systems and methods described herein are considered applicable to future protocols, including future high-speed wireless communication protocols, such as the emerging 6G protocol.

Turning now to fig. 16A, a wireless network method will be described. A wireless network method 1600 is provided. As shown, at 1602, the method in one embodiment includes detecting a transmission rate from a portion of a preamble of an incoming packet transmission signal, and at 1614, the method further includes adapting a radio to receive a remaining portion of the incoming packet transmission signal at the transmission rate. More details of this method are provided below.

The method also includes implementing, via the processing circuit, an adaptive sampling algorithm to collect samples of a preamble of the incoming packet transmission signal at 1604. The processing circuitry may be included in a base station equipped with a radio configured to receive and transmit wireless signals. In this embodiment, the wireless signals are received and transmitted according to the LoRa network protocol, although other network protocols may be used in other embodiments. For example, other suitable low power consumption or remote network protocols in which the length of the data symbols in the transmitted signal varies widely may benefit from the application of this method. An incoming packet transmission signal is received from a wireless device. Although the method of the present embodiment describes an incoming packet transmission signal received from one wireless device, it should be understood that the method is applicable to accepting incoming packet transmission signals from multiple wireless devices. For example, tens, hundreds, or even thousands of wireless devices may be used.

At 1606, the method further includes receiving the sample at a classifier and outputting a classification indicative of one or more encoding parameters of the incoming packet transmission signal. The encoding parameters are not pre-negotiated between the wireless device and the base station prior to receiving the incoming packet transmission signal. The benefit of not pre-negotiating encoding parameters is that the client device sending the incoming packet transmission signal can be used as is. In other words, the methods described herein do not require modification of the client device. In the method, the one or more encoding parameters include a bandwidth and/or a propagation factor, although other suitable encoding parameters may be used.

At 1608, in one example configuration of the method, the classifier is an artificial intelligence model that includes at least one convolutional neural network. Details of the artificial intelligence model are shown in FIG. 16B and described below.

At 1610, the samples include samples extracted from two symbols in a preamble of the packet signal, an artificial intelligence model of the classifier determines the classification using a plurality of features of the samples, the plurality of features including a real part of the samples, an imaginary part of the samples, and a Fast Fourier Transform (FFT) of the samples. Using data of the signal time and frequency domains is crucial to achieve high prediction accuracy. For example, if only FFT per signal is used, it is difficult to distinguish very low BW settings. Supplementing the FFT with time domain characteristics helps to capture the variation in the oscillating frequency of the preamble symbol, while the FFT helps to gain insight into the bandwidth variation. By using both time and frequency domain features of the signal, samples extracted from two or fewer symbols are used by the artificial intelligence model to output a classification.

At 1616, the method includes sending a configuration command to configure the radio to receive a remaining portion of the incoming packet transmission signal according to the one or more encoding parameters indicated by the classification. In this way, the radio may receive the remainder of the incoming packet transmission signal.

Turning now to fig. 16B, further details of 1608 are provided. At 1618, the artificial intelligence model is a multi-level model including a first level, wherein the bandwidth classifier includes a first convolutional neural network that classifies the incoming packet transmission signal into one of a plurality of bandwidth range classifications. In this example, a bandwidth classifier comprising a first convolutional neural network uses the real, imaginary and FFT of the incoming packet transmission signal, each divided into ten 20kHz blocks. However, in other examples, two, four, six, eight, or any suitable number of blocks may be used. In the first stage of the present example, the incoming packet transmission signal is classified as one of two bandwidth range classifications, although three, four, or any other suitable number may be used.

At 1620, at a second stage, wherein for signals having bandwidths below the predetermined threshold, the signal is classified by a low bandwidth encoding classifier comprising a second convolutional neural network into one of a plurality of low bandwidth encoding classes.

At 1622, for signals above a predetermined threshold, the signal is classified by a high bandwidth encoding classifier comprising a third convolutional neural network into one of a plurality of high bandwidth encoding classes.

In some embodiments, the methods and processes described herein may be associated with a computing system of one or more computing devices. In particular, the methods and processes may be implemented as a computer application or service, an Application Programming Interface (API), a library, and/or other computer program product.

FIG. 17 schematically illustrates a non-limiting embodiment of a computing system 1700 that can implement one or more of the methods and processes described above. Computing system 1700 is shown in simplified form. The computing system 1701 may include the wireless device 101, the base station device 102, and/or the remote devices described above and shown in fig. 1. The computing system 170 may take the form of: one or more personal computers, server computers, tablet computers, home entertainment computers, network computing devices, gaming devices, mobile computing devices, mobile communication devices (e.g., smartphones), ioT devices, remote sensor devices, and/or other computing devices.

Computing system 1700 includes a logical processor 1702, volatile memory 1704, and non-volatile storage 1706. Computing system 1701 may optionally include a display subsystem 1708, an input subsystem 1710, a communication subsystem 1712, and/or other components not shown in fig. 17.

Logical processors 1702 include one or more physical devices configured to execute instructions. For example, a logical processor may be configured to execute instructions that are part of one or more applications, programs, routines, libraries, objects, components, data structures, or other logical structures. Such instructions may be used to perform a task, implement a data type, transform the state of one or more components, achieve a technical effect, or otherwise achieve a desired result.

Logical processors may include one or more physical processors (hardware) configured to execute software instructions. Additionally or alternatively, a logical processor may include one or more hardware logic circuits or firmware devices configured to execute hardware-implemented logic or firmware instructions. The processors of the logical processor 1702 may be single-core or multi-core, and the instructions executed thereon may be configured for sequential, parallel, and/or distributed processing. Various components of a logical processor may be selectively distributed among two or more separate devices, which may be remotely located and/or configured to coordinate processing. Various aspects of the logical processor may be virtualized and executed by remotely accessible network computing devices configured in a cloud computing configuration. In this case, it will be appreciated that these virtualization aspects run on different physical logical processors of different machines.

The non-volatile storage device 1706 includes one or more physical devices configured to hold instructions executable by a logical processor to implement the methods and processes described herein. When the methods and processes are implemented, the state of the non-volatile storage device 1706 may be transformed, for example, to hold different data.

The non-volatile storage device 1706 may include a removable and/or built-in physical device. The non-volatile memory device 1706 may include optical memory (e.g., CD, DVD, HD-DVD, blu-Ray disc, etc.), semiconductor memory (e.g., ROM, EPROM, EEPROM, FLASH memory, etc.) and/or magnetic memory (e.g., hard disk drive, floppy disk drive, tape drive, MRAM, etc.), or other mass storage device technology. The non-volatile storage devices 1706 may include non-volatile, dynamic, static, read/write, read-only, sequential access, location addressable, file addressable, and/or content addressable devices. Notably, the non-volatile storage device 1706 is configured to save instructions even when the non-volatile storage device 1707 is powered down.

Volatile memory 1704 may include physical devices including random access memory. The logical processor 1702 typically uses volatile memory 1704 to temporarily store information during processing of the software instructions. It will be appreciated that when the volatile memory 1704 is powered down, the volatile memory 1704 will typically not continue to store instructions.

Various aspects of the logical processor 1702, volatile memory 1704, and non-volatile storage 1706 may be integrated into one or more hardware logic components. For example, such hardware logic components may include Field Programmable Gate Arrays (FPGAs), program and application specific integrated circuits (PASIC/ASIC), program and application specific standard products (PSSP/ASSP), system on a chip (SOC), and Complex Programmable Logic Devices (CPLDs).

The terms "module," "program," and "engine" may be used to describe an aspect of computing system 1700 that is typically implemented in software by a processor to perform a particular function using portions of volatile memory, which relates to a conversion process that specially configures the processor to perform that function. Thus, a module, program, or engine can be instantiated via the logical processor 1702 executing instructions held by the non-volatile storage device 1706 using portions of the volatile memory 1704. It is to be appreciated that different modules, programs, and/or engines can be instantiated from the same application, service, code block, object, library, routine, API, function, etc. Likewise, the same module, program, and/or engine may be instantiated by different applications, services, code blocks, objects, routines, APIs, functions, etc. The terms "module," "program," and "engine" may include a single or group of executable files, data files, libraries, drivers, scripts, database records, and the like.

When included, display subsystem 1708 may be used to present a visual representation of data held by non-volatile storage device 1706. The visual representation may take the form of a Graphical User Interface (GUI). As the herein described methods and processes change the data held by the non-volatile storage device, and thus change the state of the non-volatile storage, the state of display subsystem 1708 may likewise change to visually represent changes in the underlying data. Display subsystem 1708 may include one or more display devices using virtually any type of technology. Such display devices may be combined with the logical processor 1702, volatile memory 1704, and/or nonvolatile storage 1706 in a shared enclosure, or such display devices may be peripheral display devices.

When included, input subsystem 1710 may include or interface with one or more user input devices, such as a keyboard, a mouse, a camera, a microphone, a touchpad, a finger-operable pointing device, a touch screen, or a game controller.

When included, the communication subsystem 1712 may be configured to communicatively couple various computing devices described herein with each other and with other devices. The communication subsystem 1712 may include wired and/or wireless communication devices compatible with one or more different communication protocols, including low-power long-range wireless protocols, such as LoRaWAN described above. As a non-limiting example, the communication subsystem may be configured to communicate via a wireless telephone network or a wired or wireless local or wide area network. In some embodiments, the communication subsystem may allow computing system 1700 to send and/or receive messages to and/or from other devices via a network (such as the internet).

The following paragraphs supplement the description of the presently disclosed subject matter. According to one aspect, there is provided a wireless network system comprising a base station device comprising processing circuitry configured to detect a transmission rate from a portion of a preamble of an incoming packet transmission signal and to adjust a radio configuration to receive a remaining portion of the incoming packet transmission signal at the transmission rate.

In this regard, the base station device may further include a packet detection module that implements an adaptive sampling algorithm to collect samples of a preamble of an incoming packet transmission signal. The receiver of the base station device has received an incoming packet transmission signal from the wireless device. The base station device may also include a classifier configured to receive the samples and output a classification indicative of one or more encoding parameters of the incoming packet transmission signal. The base station device may also include a radio configuration module that sends a configuration command to configure a radio of the base station device to receive the remaining portion of the incoming packet transmission signal according to the one or more encoding parameters of the classification indication.

In this regard, the encoding parameters may not be pre-negotiated between the wireless device and the base station device prior to receipt of the incoming packet transmission signal.

In this regard, the wireless device may be further configured to set the encoding parameter to a value selected at the wireless device from a plurality of preset values of the encoding parameter, and to begin transmitting the incoming packet transmission signal in accordance with the encoding parameter to pre-negotiate the encoding parameter without any prior communication with the base station device.

In this aspect, the samples may include samples extracted from two symbols in a preamble of the packet signal, and the artificial intelligence model of the classifier determines the classification using a plurality of features of the samples, the plurality of features including a real part of the sample, an imaginary part of the sample, and a fast fourier transform of the sample.

In this regard, samples extracted from two or fewer symbols may be used by the artificial intelligence model to output a classification.

In this regard, the one or more encoding parameters may include a bandwidth and/or a propagation factor.

In this regard, the adaptive sampling algorithm may be further configured to filter the incoming packet transmission signal using one or more band pass filters, thereby generating a plurality of filtered incoming packet transmission signal components, and determine that the captured signal is sufficient to determine one or more encoding parameters for one of the filtered incoming packet transmission signal components.

In this regard, the classifier can include an artificial intelligence model including at least one convolutional neural network.

In this aspect, the artificial intelligence model may be a multi-stage model and include a first stage in which the bandwidth classifier includes a first convolutional neural network that classifies the incoming packet transmission signal into one of a plurality of bandwidth range classes, and a second stage in which, for signals having a bandwidth below a predetermined threshold, the signal is classified into one of a plurality of low bandwidth coding classes by a low bandwidth coding classifier that includes a second convolutional neural network, and for signals above the predetermined threshold, the signal is classified into one of a plurality of high bandwidth coding classes by a high bandwidth coding classifier that includes a third convolutional neural network.

In this aspect, the base station device may be configured to implement a low-power wide area network and transmit an incoming packet transmission signal from the wireless device to the base station device according to a LoRaWAN communication protocol.

According to another aspect, there is provided a wireless networking method comprising: the transmission rate is detected from a portion of a preamble of the incoming packet transmission signal and the radio is caused to receive a remaining portion of the incoming packet transmission signal at the transmission rate.

In this aspect, the method may further include, via the processing circuitry, implementing an adaptive sampling algorithm to collect samples of a preamble of an incoming packet transmission signal, the incoming packet transmission signal received from the wireless device, receiving the samples at a classifier, and outputting a classification indicative of one or more encoding parameters of the incoming packet transmission signal, and sending a configuration command to configure the radio to receive a remaining portion of the incoming packet transmission signal in accordance with the one or more encoding parameters indicated by the classification.

In this regard, the encoding parameters may not be pre-negotiated between the wireless device and the base station device prior to receiving the incoming packet transmission signal.

In this aspect, the samples may include samples extracted from two symbols in a preamble of the packet signal, and the artificial intelligence model of the classifier determines the classification using a plurality of features of the samples, the plurality of features including a real part of the samples, an imaginary part of the samples, and a fast fourier transform of the samples.

In this regard, samples taken from two or fewer symbols may be used by the artificial intelligence model to output a classification.

In this regard, the one or more encoding parameters may include a bandwidth and/or a propagation factor.

In this regard, the classifier can be an artificial intelligence model that includes at least one convolutional neural network.

In this regard, the artificial intelligence model may be a multi-stage model including a first stage in which the bandwidth classifier includes a first convolutional neural network that classifies the incoming packet transmission signal into one of a plurality of bandwidth range classes, and a second stage in which, for signals having a bandwidth below a predetermined threshold, the signal is classified into one of a plurality of low bandwidth coding classes by a low bandwidth coding classifier that includes a second convolutional neural network, and for signals above the predetermined threshold, the signal is classified into one of a plurality of high bandwidth coding classes by a high bandwidth coding classifier that includes a third convolutional neural network.

According to another aspect, a wireless network system is provided that includes processing circuitry configured to execute a packet detection module that implements an adaptive sampling algorithm to collect samples of a preamble of an incoming packet transmission signal received by a receiver from a wireless device. The wireless network system may also be configured to execute a classifier comprising a neural network configured to receive the samples and output a classification indicative of one or more encoding parameters of the incoming packet transmission signal. The wireless network system may also be configured to execute a radio configuration module that sends a configuration command to configure the associated radio to receive the remaining portion of the incoming packet transmission signal according to the one or more encoding parameters indicated by the classification.

It is to be understood that the configurations and/or approaches described herein are exemplary in nature, and that these specific embodiments or examples are not to be considered in a limiting sense, because numerous variations are possible. The specific routines or methods described herein may represent one or more of any number of processing strategies. As such, various acts illustrated and/or described may be performed in the sequence illustrated and/or described, in other sequences, in parallel, or with omissions. Also, the order of the above processes may be changed.

The subject matter of the present disclosure includes all novel and nonobvious combinations and subcombinations of the various processes, systems and configurations, and other features, functions, acts, and/or properties disclosed herein, as well as any and all equivalents thereof.

Claims (15)

1. A wireless network system, comprising:

a base station device comprising processing circuitry configured to detect a transmission rate from a portion of a preamble of an incoming packet transmission signal and adapt a radio configuration to receive a remaining portion of the incoming packet transmission signal at the transmission rate.

2. The wireless network system according to claim 1, wherein the base station apparatus further includes:

a packet detection module that implements an adaptive sampling algorithm to collect samples of the preamble of the incoming packet transmission signal that is received by a receiver of the base station device from a wireless device;

a classifier configured to receive the samples and output a classification indicative of one or more encoding parameters of the incoming packet transmission signal; and

a radio configuration module that sends a configuration command to configure a radio of the base station device to receive a remaining portion of the incoming packet transmission signal in accordance with the one or more encoding parameters indicated by the classification.

3. The wireless network system of claim 2, wherein the encoding parameters are not pre-negotiated between the wireless device and the base station device prior to receiving the incoming packet transmission signal.

4. The wireless network system of claim 2,

the wireless device is configured to:

setting the encoding parameter to a value selected at the wireless device from a plurality of preset values for the encoding parameter, an

Starting transmission of the incoming packet transmission signal in accordance with the encoding parameters to pre-negotiate the encoding parameters without any prior communication with the base station apparatus.

5. The wireless network system of claim 2, wherein the samples comprise samples extracted from two symbols in a preamble of the packet signal, and an artificial intelligence model of the classifier determines the classification using a plurality of features of the samples, the plurality of features comprising a real part of the samples, an imaginary part of the samples, and a fast fourier transform of the samples.

6. The wireless network system of claim 5, wherein samples extracted from two or fewer symbols are used by the artificial intelligence model to output the classification.

7. The wireless network system of claim 2, wherein the one or more encoding parameters include a bandwidth and/or a propagation factor.

8. The wireless network system of claim 2, wherein the adaptive sampling algorithm is configured to:

filtering the incoming packet transmission signal using one or more band pass filters, thereby generating a plurality of filtered incoming packet transmission signal components;

determining that the captured signal is sufficient to determine the one or more encoding parameters for one of the filtered incoming packet transmission signal components.

9. The wireless network system of claim 2, wherein the classifier is an artificial intelligence model comprising at least one convolutional neural network.

10. The wireless network system of claim 5, wherein the artificial intelligence model is a multi-level model and comprises:

a first stage in which a bandwidth classifier includes a first convolutional neural network that classifies the incoming packet transmission signal into one of a plurality of bandwidth range classes; and

a second stage in which, for signals having a bandwidth below the predetermined threshold, the signal is classified by a low bandwidth encoding classifier comprising a second convolutional neural network into one of a plurality of low bandwidth encoding classes, and for signals above the predetermined threshold, the signal is classified by a high bandwidth encoding classifier comprising a third convolutional neural network into one of a plurality of high bandwidth encoding classes.

11. The wireless network system of claim 1, wherein the base station device is configured to implement a low-power wide area network and to transmit the incoming packet transmission signal from the wireless device to the base station device according to a LoRaWAN communication protocol.

12. A wireless network method, comprising:

detecting a transmission rate from a portion of a preamble of an incoming packet transmission signal; and

adapting a radio to receive a remaining portion of the incoming packet transmission signal at the transmission rate.

13. The method of claim 12, further comprising:

via a processing circuit:

implementing an adaptive sampling algorithm to collect samples of the preamble of the incoming packet transmission signal, the incoming packet transmission signal received from a wireless device;

receiving the samples at a classifier and outputting a classification indicative of one or more encoding parameters of the incoming packet transmission signal; and

sending a configuration command to configure the radio to receive a remaining portion of the incoming packet transmission signal according to the one or more encoding parameters indicated by the classification.

14. The method of claim 13, wherein the encoding parameters are not pre-negotiated between the wireless device and the base station device prior to receiving the incoming packet transmission signal.

15. The method of claim 13, wherein the samples comprise samples extracted from two symbols in a preamble of the packet signal, and an artificial intelligence model of the classifier determines the classification using a plurality of features of the samples, the plurality of features comprising a real part of the samples, an imaginary part of the samples, and a fast fourier transform of the samples.