Frequency resolution refers to the ability to separate the two adjacent peaks. In practical applications, it is the minimum resolution interval of two signals of different frequencies. The most effective way to study the digital spectrum is usually discrete Fourier transform.
definition
The first explanation
Frequency resolution is to be understood that when using the DFT, the minimum frequency in a frequency axis can be obtained interval
Wherein, N is the number of sampling points, sampling frequency, sampling interval. So that the length of the sampling time T before the analog signal, the longer the length of the signal, the better the frequency resolution.
How much and how often the requirements related to the resolution of the sampling points.
For example: the engine speed 3000r / min = 50Hz, if the frequency of failure is estimated to be analyzed in the 8 octave or less, the required frequency resolution spectrum ΔF = 1 Hz, the sampling frequency and set points:
maximum frequency of Fm = 3000 Analysis / 8 · 50Hz = 400Hz;
sampling frequency Fs = 2.56 · Fm = 2.56 · 400Hz = 1024Hz;
sampling points N = 2.56 · (Fm / ΔF ) = 2.56 · (400Hz / 1Hz) = 1024
line number M = N / 2.56 = 1024/400 = 2.56
According to the FFT, in fact, the resulting spectrum is 1024 points, but we know that there is a negative frequency on mathematical calculations, is symmetrical, so in practice we focus on the positive frequency part of the spectrum corresponding to, that there are positive frequency line 512, typically why we added 400 line on the road in this case, because usually the influence due to the frequency and time domain aliasing truncated, generally considered the accuracy of the spectral line 401 to the line 512 is high and is not disregarded.
Further, the sampling points are not easily set, i.e. is not the better, and vice versa. For the whole period of rotation must meet the mechanical sampling frequency to eliminate deformities, increase the resolution can not be eliminated simply frequency deformity. In the past, some people think the data as long as possible, or just timing domain signal length, in fact, doing so is not clear on certain concepts, for example, do not know the entire sampling period.
Does not produce the lowest frequency aliasing sampling frequency Fs analysis requires twice the maximum frequency Fm, the reason for using 2.56-fold with computer binary representation of the main relevant. Its main purpose is to avoid confusion signal to ensure a high frequency signal into a low frequency signal is not distorted.
First select sample length T to ensure reflect picture signal, the transient signal should include the entire transient; periodic signals, a periodic signal acquisition theory on it. Second, consider the frequency resolution needed, sample length T at the maximum frequency of Fm analysis determined the frequency resolution △ f is an inverse relationship, i.e. the longer T △ f is smaller the higher the frequency resolution.
General analysis software are provided several lines M, sampling points N = 2.56M. Signal analysis is commonly used 512,1024,2048,4096 sampling points and the like. We often equivalent to the number of lines, say 200,400,800,1600 spectral lines, spectrum analysis, sampling points selected power of two. △ f = Fm / M, the greater the number M of the visible spectrum frequency resolution △ f is smaller the higher the frequency resolution.
In the motor failure diagnosis in order to find the polar side pass frequency band interval (typically less 1Hz) peak, often require high resolution (hereinafter 1Hz), usually selected 210HzFm, 6400 lines.
As for the entire sampling period is difficult to achieve, it will inevitably cut off because the signal generated by leaks in order to avoid these errors, so to take the windowing approach.
A second explanation
research method