A method for accurate signal frequency measurement based on machine learning and measurement range compensation

Abstract

This invention discloses a method for accurate signal frequency measurement based on machine learning and measurement range compensation. First, utilizing the characteristics of the LSTM model, adjacent I / Q signals are smoothed and preprocessed using the LSTM model to obtain smoother and more accurate I / Q signals, thus improving the accuracy of the measurement results. Then, the smoothed I / Q signals are compensated in the complex signal domain. Finally, the Cordic frequency measurement algorithm is applied to the compensated complex signal to obtain the frequency fr of the compensated complex signal. n_n+1 Finally, subtract the signal sampling rate f. s To eliminate pre-compensation, the signal frequency f is obtained from the two adjacent I / Q signals. n_n+1 This allows for the compensation of the measurement range of the Cordic algorithm, providing reverse compensation and correction for missing ranges in the data characteristics, thereby expanding the signal measurement range.

Description

Technical Field

[0001] This invention belongs to the field of frequency measurement technology, and more specifically, relates to a method for accurate measurement of signal frequency based on machine learning and measurement range compensation, for accurate measurement of I / Q signal frequency based on zero intermediate frequency architecture. Background Technology

[0002] In the development of the communications field, signal quality and measurement have always played a crucial role, with signal frequency measurement being a particularly important part. Currently, most signal frequency measurement methods utilize the Fourier Transform (FFT) principle. Based on this, complex algorithms or specific mathematical properties are used to deduce and calculate the frequency. For example, the Cordic algorithm cleverly utilizes coordinate rotation algorithms in mathematics to perform complex trigonometric function operations with minimal computation, thereby obtaining the frequency value of the signal under test.

[0003] However, the Cordic algorithm relies too heavily on the accuracy of random data, and while the use of rotation of mathematical coordinates greatly simplifies the complexity of mathematical calculations, it also leads to a lack of measurement range.

[0004] The following is a brief explanation of the problems with the Cordic algorithm:

[0005] 1. Measurement errors caused by over-reliance on random data. Since the algorithm's measurement principle involves arbitrarily selecting I / Q signals at any given moment for calculation, in other words, the data source is randomly obtained. For a perfect standard signal, random selection wouldn't affect the measurement results. However, in actual measurement and operation, a standard signal doesn't exist. Therefore, if the randomly obtained data deviates significantly from the standard data, the measurement results become unreliable. Current Cordic algorithms use optimization algorithms such as multi-channel averaging and Kalman filtering to reduce measurement errors caused by data source errors. These optimization methods have some effect, but they only optimize relatively regular signals; they have no significant corrective effect on signals with large deviations, significant noise, or irregularities.

[0006] 2. By cleverly utilizing the properties of trigonometric identities, the computational complexity is indeed greatly reduced. However, due to its own angle rotation, angle compensation, and π calculations, the frequency measurement range of the Cordic algorithm is no longer standard. And became Here, T refers to the original signal measurement range, which is generally determined by the sampling frequency, the number of multiple signals, and the characteristics of the hardware itself. Clearly, the frequency measurement range, like the signal range, changes due to the rotational characteristics. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing Cordic frequency measurement algorithms and propose a precise signal frequency measurement method based on machine learning and measurement range compensation. This method makes the measured data smoother and more accurate, thereby eliminating undesirable measurements caused by abnormal noise and random values, resulting in more accurate measurement results. At the same time, the Cordic algorithm performs measurement range compensation to reverse compensate and correct the missing range in the data characteristics, thus expanding the signal measurement range.

[0008] To achieve the above-mentioned objectives, the present invention provides a method for accurate signal frequency measurement based on machine learning and measurement range compensation, characterized by comprising the following steps:

[0009] (1) Smoothing preprocessing based on machine learning

[0010] For two adjacent I / Q signals S n S n+1 The signals are fed into a corresponding LSTM (Long Short Term Memory) model for smoothing preprocessing to obtain the corresponding signals S′. n S′ n+1 Where n represents the sequence number of the I / Q signal;

[0011] (2) Perform pre-compensation

[0012] The signal S′ obtained by smoothing preprocessing n S′ n+1 Increasing the sampling rate by 1 / 4 transforms the direct frequency compensation into a process of adjusting the sampling rate of the measured signal, i.e., signal S′. n S′ n+1 To provide compensation, that is:

[0013]

[0014] in, For I / Q signal S′ n The I signal in For I / Q signal S′ n The Q signal, f s Let x be the signal sampling rate, t represent time, and x be the signal sampling rate. n For the compensated complex signal of the nth I / Q signal, x n+1 The complex signal of the (n+1)th I / Q signal after compensation;

[0015] (3) The frequency of the compensated complex signal is obtained by using the Cordic frequency measurement algorithm.

[0016] The complex signal x nPerform a conjugate transformation, then combine it with the complex signal x. n+1 Perform a dot product to obtain a new complex signal R. n_n+1 And calculate its arctangent angle θ n_n+1 :

[0017]

[0018] Then perform the π operation: if the complex signal R n_n+1 real(R) n_n+1 ) greater than or equal to 0, arctangent angle θ n_n+1 If the complex signal R remains unchanged... n_n+1 real(R) n_n+1 If θ is less than 0, then the arctangent angle θ n_n+1 Then add π, that is, θ n_n+1 =θ n_n+1 +π;

[0019] Finally, the complex signal R is obtained. n_n+1 frequency fr n_n+1 :

[0020]

[0021] (4) Eliminate pre-compensation

[0022] Based on the two adjacent I / Q signals S n S n+1 Calculated signal frequency f n_n+1 for:

[0023] f n_n+1 =fr n_n+1 -f s .

[0024] The objective of this invention is achieved as follows:

[0025] This invention presents a precise signal frequency measurement method based on machine learning and measurement range compensation. First, it utilizes the characteristics of the LSTM model to perform smoothing preprocessing on two adjacent I / Q signals, resulting in smoother and more accurate I / Q signals, thus improving measurement accuracy. Then, it compensates the two smoothed I / Q signals in the complex signal domain. Finally, it applies the Cordic frequency measurement algorithm to the compensated complex signal to obtain the frequency fr of the compensated complex signal. n_n+1 Finally, subtract the signal sampling rate f. s To eliminate pre-compensation, the signal frequency f is obtained from the two adjacent I / Q signals. n_n+1 This allows for the compensation of the measurement range of the Cordic algorithm, providing reverse compensation and correction for missing ranges in the data characteristics, thereby expanding the signal measurement range. Attached Figure Description

[0026] Figure 1 This is a flowchart of a specific implementation of the signal frequency precision measurement method based on machine learning and measurement range compensation of the present invention;

[0027] Figure 2 This is a schematic diagram of the signal processing process of a specific implementation of the signal frequency precision measurement method based on machine learning and measurement range compensation of the present invention;

[0028] Figure 3 This is a schematic diagram illustrating the principle of the LSTM model;

[0029] Figure 4 This is a time-domain comparison diagram of the original I / Q signal and the calibrated I / Q signal, where (a) is the original I / Q signal and (b) is the calibrated I / Q signal. Detailed Implementation

[0030] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be particularly noted that in the following description, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0031] In this embodiment, the measuring instrument can be improved by adding a range calibration mode. In practice, it still includes a conventional measurement mode and a precision measurement mode, while the newly added range calibration mode optimizes the measurement range data, thereby better meeting the user's measurement needs. Furthermore, the newly added machine learning component can further improve data accuracy.

[0032] In this embodiment, as Figure 1 , 2 As shown, the signal frequency accurate measurement method based on machine learning and measurement range compensation of the present invention includes the following steps:

[0033] Step S1: Smoothing preprocessing based on machine learning

[0034] like Figure 2 As shown, for two adjacent I / Q signals S n S n+1 The signals are fed into a corresponding LSTM (Long Short-Term Memory) model for smoothing preprocessing to obtain the corresponding signals S′. n S′ n+1 , where n represents the sequence number of the I / Q signal.

[0035] Since the Cordic frequency measurement algorithm is based on arbitrary values, a large error in the chosen value will inevitably lead to measurement errors. To eliminate this random error, a smoothing preprocessing of the signal to be measured is required. In this invention, machine learning is used for signal smoothing preprocessing. Utilizing the specificity of the signal, this invention uses an LSTM model for signal training and prediction, i.e., smoothing preprocessing. The principle of the LSTM model is as follows: Figure 3 As shown, its principle is existing technology and will not be elaborated here.

[0036] By comparing the signal predicted by the model with the signal before prediction, such as Figure 4 This is a comparison graph of the time-domain signals of the original I / Q signal and the calibrated I / Q signal. By analyzing the signal in the graph, it is easy to find that the calibrated signal is more stable and smoother, and can obtain better measurement signal results.

[0037] Step S2: Perform pre-compensation

[0038] Due to the characteristics of the data used in the Cordic frequency measurement algorithm, the frequency measurement range of this algorithm is limited by... And became If a full-range measurement is required, pre-compensation processing is needed on the signal obtained after the above smoothing process. This invention provides a method for supplementing the signal frequency to perform signal pre-compensation processing, specifically for the smoothed pre-processed signal S′. n S′ n+1 Increasing the sampling rate by 1 / 4 transforms the direct frequency compensation into a process of adjusting the sampling rate of the measured signal, i.e., signal S′. n S′ n+1 To provide compensation, that is:

[0039]

[0040] in, For I / Q signal S′ n The I signal in For I / Q signal S′ n The Q signal, f s Let x be the signal sampling rate, t represent time, and x be the signal sampling rate. n For the compensated complex signal of the nth I / Q signal, x n+1 This is the complex signal of the (n+1)th I / Q signal after compensation.

[0041] Step S3: Use the Cordic frequency measurement algorithm to obtain the frequency of the compensated complex signal.

[0042] Step S3.1: Convert the complex signal x n Perform a conjugate transformation, then combine it with the complex signal x. n+1Perform a dot product to obtain a new complex signal R. n_n+1 And calculate its arctangent angle θ n_n+1 :

[0043]

[0044] Step S3.2: Perform phase compensation

[0045] The above algorithm already satisfies the condition of measuring half the sampling frequency of each channel. However, in practice, we want to measure the sampling condition of the total sampling rate. Therefore, we need to perform an algorithm compensation, i.e., phase compensation of π. If the real part Re of the complex signal R obtained above is negative, it means that the Cordic rotation has exceeded the phase limit. To obtain the true value, it needs to be supplemented by π, i.e., a π operation needs to be performed: if the complex signal R... n_n+1 real(R) n_n+1 ) greater than or equal to 0, arctangent angle θ n_n+1 If the complex signal R remains unchanged... n_n+1 real(R) n_n+1 If θ is less than 0, then the arctangent angle θ n_n+1 Then add π, that is, θ n_n+1 =θ n_n+1 +π.

[0046] Step S3.3: Obtain the frequency of the compensated complex signal

[0047] The complex signal R is obtained according to the following formula. n_n+1 frequency fr n_n+1 :

[0048]

[0049] Step S4: Eliminate pre-compensation

[0050] Since pre-compensation is performed in step S2, the real part Re and imaginary part IM in step S2 will also be updated after the signal is compensated. Therefore, it is necessary to recalculate the real and imaginary parts of the compensated signal. Alternatively, the measurement results can be directly compensated. This invention adopts direct compensation of the measurement results, that is, based on the two adjacent I / Q signals S... n S n+1 Calculated signal frequency f n_n+1 for:

[0051] f n_n+1 =fr n_n+1 -f s .

[0052] This yields the final measurement result. Of course, in the actual implementation process, the precise measurement mode is still supported, so users can still enable the function of averaging multiple measurements and the Kalman filtering function. Enabling these functions does not affect the method and new functions provided by this invention.

[0053] In the specific implementation process, for multiple signals, the signal frequency is calculated for every pair of signals, and then the frequencies f of all the obtained signals are... n_n+1 Calculate the mean and use it as the output.

[0054] This invention presents a precise signal frequency measurement method based on machine learning and measurement range compensation. Building upon the Cordic frequency measurement algorithm, it addresses the lack of frequency measurement range inherent in the original algorithm due to the limitations of Cordic itself. Simultaneously, it employs a machine learning model LSTM for signal smoothing, thereby obtaining a more accurate frequency value for the signal under test. This invention effectively improves the accuracy of frequency measurement by correcting the signal under test using the LSTM machine learning model, demonstrating good practicality.

[0055] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A method for accurate signal frequency measurement based on machine learning and measurement range compensation, characterized in that, Includes the following steps: (1) Smoothing preprocessing based on machine learning For two adjacent I / Q signals S n S n+1 The signals are fed into a corresponding LSTM (Long Short Term Memory) model for smoothing preprocessing to obtain the corresponding signals S′. n S′ n+1 Where n represents the sequence number of the I / Q signal; (2) Perform pre-compensation The signal S′ obtained by smoothing preprocessing n S′ n+1 Increasing the sampling rate by 1 / 4 transforms the direct frequency compensation into a process of adjusting the sampling rate of the measured signal, i.e., signal S′. n S′ n+1 To provide compensation, that is: in, For I / Q signal S′ n The I signal in For I / Q signal S′ n The Q signal, f s Let x be the signal sampling rate, t represent time, and x be the signal sampling rate. n For the compensated complex signal of the nth I / Q signal, x n+1 The complex signal of the (n+1)th I / Q signal after compensation; (3) The frequency of the compensated complex signal is obtained by using the Cordic frequency measurement algorithm. The complex signal x n Perform a conjugate transformation, then combine it with the complex signal x. n+1 Perform a dot product to obtain a new complex signal R. n_n+1 And calculate its arctangent angle θ n_n+1 : Then perform the π operation: if the complex signal R n_n+1 real(R) n_n+1 ) greater than or equal to 0, arctangent angle θ n_n+1 If the complex signal R remains unchanged... n_n+1 real(R) n_n+1 If θ is less than 0, then the arctangent angle θ n_n+1 Then add π, that is, θ n_n+1 =θ n_n+1 +π; Finally, the complex signal R is obtained. n_n+1 frequency fr n_n+1 : (4) Eliminate pre-compensation Based on the two adjacent I / Q signals S n S n+1 Calculated signal frequency f n_n+1 for: f n_n+1 *fr n_n+1 -f s ...

2. The method for accurate signal frequency measurement based on machine learning and measurement range compensation according to claim 1, characterized in that, For multiple signals, the signal frequency is calculated for every pair of signals, and then the frequencies f of all the obtained signals are... n_n+1 Calculate the mean and use it as the output.

Images