KR102645886B1 - Method and apparatus of signal detection in massive machine type communication based on deep neural network - Google Patents

Abstract

A signal detection method performed by a signal detection device includes receiving signal information into a learning model for signal detection; And detecting a transmission signal of an active device or an inactive device from the signal information through a learning model for signal detection, wherein the learning model is learnable in a massive machine type communication (mMTC) system. It may be learned to detect the transmission signal of an active device using consensus ADMM (Learnable Consensus-ADMM; cADMM).

Description

Signal detection method and device in large-scale object communication using deep neural network {METHOD AND APPARATUS OF SIGNAL DETECTION IN MASSIVE MACHINE TYPE COMMUNICATION BASED ON DEEP NEURAL NETWORK}

The explanation below relates to signal detection technology.

Recently, hundreds to thousands of IoT (Internet of Things) devices transmit signals simultaneously in next-generation communication systems in various fields such as factory automation, autonomous driving, and transportation. In next-generation communication, communication in situations where the number of IoT devices is very large is considered a major issue, and massive machine type communication (mMTC) is being used in the 5G system. mMTC refers to a system in which a large number of devices connect and communicate simultaneously, and unlike human wireless communication, uplink accounts for most of the traffic rather than downlink. Not all devices transmit signals at the same time, but some devices transmit signals and some do not transmit signals. Devices that transmit signals are called active devices, and devices that do not transmit signals are called inactive devices. It is called an inactive device. The fact that only active devices transmit signals and inactive devices do not transmit signals can be mathematically modeled as a sparse vector (a vector in which most of the values of the vector are 0). Detection of sparse symbol vectors can be converted into a multi-user detection problem based on compressive sensing and expressed as a transmission signal detection problem of an active device.

However, the problem of detecting signals in the mMTC system is an NP-hard problem, so as more devices are connected simultaneously, complexity increases, making it difficult to optimally detect signals. The compressed sensing-based multi-user detection method to solve this problem did not have good performance, and the signal detection technique using consensus-ADMM (cADMM), which can find a solution with low complexity, did not have sufficient performance to distinguish inactive devices. Not good.

It can provide a learning model for signal detection in a massive machine type communication (mMTC) system.

A method and apparatus for detecting signals of an active device or an inactive device can be provided using a consensus ADMM (consensus-ADMM; cADMM) learning model for signal detection.

A signal detection method performed by a signal detection device includes receiving signal information into a learning model for signal detection; And detecting a transmission signal of an active device or an inactive device from the signal information through a learning model for signal detection, wherein the learning model is learnable in a massive machine type communication (mMTC) system. It may be learned to detect the transmission signal of an active device using consensus ADMM (Learnable Consensus-ADMM; cADMM).

The learning model can learn the hyperparameters of the consensus ADMM by applying the majorization-minimization (MM) algorithm.

The loss function of the learning model uses the output value of the final layer of the learning model and the mean square error (MSE) of the transmission signal of the active device, and the learning model uses training data for signal detection to detect the active device. Hyper parameters can be discovered through learning to reduce the mean square error of the device's transmission signal, and signals can be detected using a learning model composed of the discovered hyper parameters.

The learning model acquires the received signal of the base station and the channel information between the base station and the device, adjusts the threshold according to the size of the obtained received signal, and learns the hyper parameters of the consensus ADMM through updating each variable. there is.

The learning model may repeat the minimization process for each variable as many times as the number of layers configured in the learning model.

A signal detection device includes a signal input unit that receives signal information into a learning model for signal detection; and a signal detection unit that detects a transmission signal of an active device or an inactive device from the signal information through a learning model for signal detection, wherein the learning model is learned in a massive machine type communication (mMTC) system. It may be learned to detect the transmitted signal of an active device using a possible consensus ADMM (Learnable Consensus-ADMM; cADMM).

Signal detection performance and classification accuracy of active/inactive devices can be improved through a learning model for signal detection in a massive machine type communication (mMTC) system.

1 is a block diagram for explaining the configuration of a signal detection device according to an embodiment.
Figure 2 is a flowchart for explaining a signal detection method in a signal detection device according to an embodiment.
FIG. 3 is a diagram illustrating an mMTC detection algorithm based on consensus ADMM through learning, in one embodiment.
Figure 4 is a diagram showing a symbol error rate graph according to SNR, according to one embodiment.
Figure 5 is a diagram showing an activation/deactivation classification accuracy graph according to SNR, according to one embodiment.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram for explaining the configuration of a signal detection device according to an embodiment, and FIG. 2 is a flowchart for explaining a signal detection method in the signal detection device according to an embodiment.

The processor of the signal detection device 100 may include a signal input unit 110 and a signal detection unit 120. These processor components may be expressions of different functions performed by the processor according to control instructions provided by program codes stored in the signal detection device. The processor and its components may control the signal detection device to perform steps 210 to 220 included in the signal detection method of FIG. 2. At this time, the processor and its components may be implemented to execute instructions according to the code of an operating system included in the memory and the code of at least one program.

The processor may load program code stored in a program file for a signal detection method into memory. For example, when a program is executed in a signal detection device, the processor may control the signal detection device to load the program code from the program file into the memory under the control of the operating system. At this time, the signal input unit 110 and the signal detection unit 120 may each be different functional expressions of a processor for executing the subsequent steps 210 to 220 by executing instructions of the corresponding portion of the program code loaded in the memory. there is.

In step 210, the signal input unit 110 may receive signal information as input to a learning model for signal detection. At this time, the learning model may be learned to detect the transmission signal of the active device using a learnable consensus-ADMM (cADMM) in a massive machine type communication (mMTC) system. The learning model can learn the hyperparameters of the consensus ADMM by applying the majorization-minimization (MM) algorithm. The learning model searches for hyperparameters through learning to reduce the mean square error of the transmission signal of the active device using learning data for signal detection, and detects signals using a learning model composed of the discovered hyperparameters. You can. The learning model acquires the received signal of the base station and the channel information between the base station and the device, adjusts the threshold according to the size of the obtained received signal, and learns the hyper parameters of the consensus ADMM through updating each variable. . The learning model can repeat the minimization process for each variable as many times as the number of layers configured in the learning model.

In step 220, the signal detection unit 120 may detect a transmission signal of an active device or an inactive device from signal information through a learning model for signal detection.

FIG. 3 is a diagram illustrating an mMTC detection algorithm based on consensus ADMM through learning, in one embodiment.

The signal detection device can obtain the received signal and channel information (310). In an mMTC system with N devices each having a single antenna (N is a natural number) and one base station having M antennas (M is a natural number), the signal from the base station can be expressed as Equation 1.

Equation 1:

At this time, is the received signal vector of the base station, is a vector that collects signals transmitted by M devices, has a mean of 0 and a variance of is a white noise (AWGN) vector. channel is the channel matrix between the device and the base station and is composed of a Gaussian distribution, and it is assumed that the receiver knows the channel. In an embodiment, in the mMTC system, only m devices out of M devices are activated, so the transmission signal vector is a sparse vector. The signal sent by each device is QPSK modulated. was used, and considering that inactive devices do not send signals, It becomes. At this time, ego, am. means a set in which a zero vector is added to the QPSK signal constellation.

The results of sparsity-aware maximum a posteriori (S-MAP) are maximizing can be transformed into Equation 2 by Bayes' rule.

Equation 2:

The probability that each device transmits a signal is and has a value less than 1. Therefore, the signal generated from the device If you say has sparse characteristics. If each device generates signals independently, Equation 3 becomes:

Equation 3:

of each element by If it is not 0, it returns 1, and if it is 0, it returns 0.

The result of combining Equation 2 and Equation 3 can be expressed as Equation 4.

Equation 4:

here, is a regularization parameter, Since it is inversely proportional to It serves to emphasize the sparse nature of . However, because Equation 4 is not a convex optimization problem, The terms must be converted into a convex optimization problem using a different convex norm. If So in equation 4 second If converted to, it becomes Equation 5.

Equation 5:

is an indicator function. If is satisfied, it is 0, otherwise it is ∞. This issue If written in separable form using , it becomes Equation 6.

Equation 6:

At this time, each function is , , am. The augmented Lagrangian for Equation 6 can be expressed as Equation 7.

Equation 7:

Is It is a dual variable corresponding to is a penalty parameter. In the above equation and By combining , Equation 7 can be easily solved using scaled dual variables. If defined as Equation 8, is a scaled dual variable.

Equation 8:

Equation 9:

If equation 7 is rewritten, it becomes equation 9.

Consensus ADMM (cADMM) is for each variable , , Minimization is repeated alternately. The kth repetition can be expressed as Equation 10, Equation 11, and Equation 12.

Equation 10:

Equation 11:

Equation 12:

Since the mean of the dual variable is 0, the above minimization process can be simplified.

In other words, Because of, It can be expressed as Equation 13 and Equation 14.

Equation 13:

Equation 14:

Newly proposed auxiliary variable Through Equation 13 and Equation 14, Equation 13 and Equation 14 can be expressed simply as Equation 15 and Equation 16.

Equation 15:

Equation 16:

Next, each The process of minimizing is as follows. first, The equation for updating is shown in Equation 17.

Equation 17:

Computational complexity can be reduced by using proximal terms. first, Since is a smooth convex function, it is equivalent to Equation 18.

Equation 18:

ego, Is is the largest eigenvalue of . Then, the majorization-minimization (MM) technique was applied to approximate The update for is the same as Equation 19 (320).

Equation 19:

From here, It is defined as

next This is the update 330 process. The update for is Equation 20.

Equation 20:

The problem of minimizing the value of each element can be solved as follows. silver It is the mth element of the vector.

Equation 21:

In equation 21, am. silver It can be expressed as Equation 22 by applying the proximal operator of the function.

Equation 22:

From here, am. Equation 22 is element-wise soft thresholding. is the threshold Proportional to the probability that the mth device is activated The smaller it is becomes larger than the threshold Because it gets bigger increases the possibility of having 0. However, since Equation 22 only reduces the size of the signal, a problem may occur where it becomes sparse as repetition progresses. To solve this Accordingly, all m are updated as shown in Equation 23 so that the threshold can be changed and sparsity can be reduced (350).

Equation 23:

At this time, , are all critical values It plays a role in regulating see The sign of the hyperbolic tangent changes depending on whether is large or small. becomes the threshold for whether to increase or decrease scarcity as shown in Equation 22. The smaller the The value of the hyperbolic tangent changes slightly depending on The larger the value, the more it changes, so it plays a role in determining how much the threshold is adjusted.

finally, The minimization process for is the same as Equation 24 (340).

Equation 24:

Is Each element of This means projection to the nearest constellation point. Since this projective operator cannot be differentiated, a function such as Equation 25 is used using the hyperbolic tangent.

Equation 25:

From here, when, If you say , It is defined as follows. The smaller the hyperbolic tangent, the softer the decision. The larger the hyperbolic tangent, the harder the decision.

As shown in Figure 3, the algorithm proposed in the embodiment repeats Equation 19, Equation 23, Equation 24, and Equation 16 until convergence (360). Each iteration corresponds to one layer of the neural network and hyperparameters are derived from the training data. As an algorithm for learning (370), Learnable Consensus-ADMM (Learnable cADMM) is proposed. The loss function uses the output value of the final layer and the mean square error (MSE) of the transmitted signal. In order to reduce the error obtained in this way, appropriate hyperparameters are found through learning and the signal is detected.

To explain the technical effectiveness of the method proposed in the embodiment, the symbol error rate of the ideal algorithm Oracle-ZF, OMP using the existing orthogonal matching pursuit (OMP), the existing cADMM without learning, and the Learnable cADMM proposed in the embodiment ( SER), performance on active/inactive classification accuracy can be compared. Oracle-ZF has the best performance because it knows all active devices and detects signals, but it is an algorithm that cannot be used in real situations because active devices are not known in advance. OMP is a greedy algorithm that finds active devices at each iteration and uses zero forcing. In the simulation, the parameters are: number of base station antennas (N) = 64, number of devices (M) = 96, device activation probability ( )=0.1, QPSK can be used as the modulation method. Figures 4 and 5 show the average results of experiments on 300,000 independent channels.

Figure 4 is a diagram showing a symbol error rate graph according to SNR, according to one embodiment.

Oracle-ZF has the best performance because it knows the active devices and is not an underdetermined system. Learnable cADMM achieves better performance than OMP and cADMM as SNR increases. It can be seen that Learnable cADMM has a performance improvement of more than 1 dB over cADMM at a symbol error rate of 10 ^-4 .

Figure 5 is a diagram showing an activation/deactivation classification accuracy graph according to SNR, according to one embodiment.

Figure 5 is a graph showing classification accuracy according to signal-to-noise ratio. Classification accuracy is the number of devices that are active or not divided by the total number of devices, indicating how accurately the device was classified as active or not. Oracle-ZF knows whether all devices are active and has 100% performance. It can be seen that the Learnable cADMM proposed in the example is superior to other methods and achieves more than 90% accuracy at 4dB.

The device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), etc. , may be implemented using one or more general-purpose or special-purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may execute an operating system (OS) and one or more software applications that run on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used on any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. It can be embodied in . Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

As described above, although the embodiments have been described with limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

Therefore, other implementations, other embodiments, and equivalents of the claims also fall within the scope of the claims described below.

Claims (6)

In a signal detection method performed by a signal detection device,
Receiving signal information including a received signal and channel information into a learning model for signal detection; and
Detecting a transmission signal of an active device or an inactive device from the signal information through the learning model for signal detection.
Including,
The learning model uses a learnable consensus ADMM (Learnable Consensus-ADMM; cADMM) in a massive machine type communication (mMTC) system, and the hyperparameters of the learnable consensus ADMM are learned to detect the transmission signal of an active device. As it has been done,
Auxiliary variables in signal detection problem to detect transmitted signal using signal information to apply consensus ADMM Each variable through and repeating the minimization process for each variable,
variable The process of minimizing is , , Including updating , and applying the majorization-minimization (MM) technique through Equation 1 to update and adjust the threshold according to the signal size. Through Equation 2 for each element of Update , and through Equation 3 to update
Equation 1 above is:
Equation 2 is,

Equation 3 is,

A signal detection method characterized by delete According to paragraph 1,
The loss function of the learning model uses the output value of the final layer of the learning model and the mean square error (MSE) of the transmission signal of the active device,
The learning model is,
Using learning data for signal detection, hyper-parameters are searched through learning to reduce the mean square error of the transmission signal of the active device, and signals are detected using a learning model composed of the discovered hyper-parameters.
A signal detection method characterized in that. delete According to paragraph 1,
The learning model is,
A signal detection method characterized in that the minimization process for each variable is repeated as many times as the number of layers configured in the learning model. In the signal detection device,
A signal input unit that receives signal information including a received signal and channel information into a learning model for signal detection; and
A signal detection unit that detects a transmission signal of an active device or an inactive device from the signal information through the learning model for signal detection.
Including,
The learning model uses a learnable consensus ADMM (Learnable Consensus-ADMM; cADMM) in a massive machine type communication (mMTC) system, and the hyperparameters of the learnable consensus ADMM are learned to detect the transmission signal of an active device. As learned,
Auxiliary variables in signal detection problem to detect transmitted signals using signal information to apply consensus ADMM Each variable through and repeating the minimization process for each variable,
variable The process of minimizing is , , Including updating , and applying the majorization-minimization (MM) technique through Equation 1 to update and adjust the threshold according to the signal size. Through Equation 2 for each element of Update , and through Equation 3 to update
Equation 1 above is:

Equation 2 is,

Equation 3 is,

A signal detection device characterized by a.

Images