我正在尝试使用带有F＃的低通滤波器使用MathNet.Filter，当我绘制时，结果不起作用。 我在C＃中发现了一些类似的问题，但是当我在F＃中重现代码时，结果是错误的。

这是我使用MathNet.Filter的方式：

let fs ,fcut, order = 5000., 2.5, 5  // set the filter parameters
let lowPass2 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut,order)
let disp_beam_f = disp_beam |> lowPass2.ProcessSamples // apply the filter

// I print max/min to do a quick check
(disp_beam_f |> Array.min, disp_beam |> Array.min )

然后，当我绘制一些数据时，我的结果是： 当我只绘制过滤后的数据时，绘图看起来像原始数据，但具有不同的比例： 我已经尝试更改过滤器的顺序，而不是使用过滤器的顺序，结果更好，但它不是过滤后的数据。

• 我做错了什么？ 我知道fs是频率采样（以Hz为单位）， fcut切割高频的频率（以Hz为单位）， 顺序表示fcut以上频率的信号衰减多少（或快）。

有关更多信息，请参阅以下脚本： https ： //gist.github.com/josesoyo/f45b59a035c3e2ec2cdf0fe1f95cc84b
数据样本位于： https ： //drive.google.com/open？id = 1rA4B21i-S5g2wrB4eK7xk3_A6Rs0tLHf
结果应该是这个文件： https ： //drive.google.com/open？id = 1ukmWtzFm_I4ou3zuQcLWCrFIOJs2eV41

编辑基于Gene Belitski的答案我相信澄清案例：

运行下一个脚本会更好地显示我遇到的问题：

#I __SOURCE_DIRECTORY__
#r @"..\packages\MathNet.Numerics.3.8.0\lib\net40\MathNet.Numerics.dll"
#r @"..\packages\MathNet.Filtering.0.4.0\lib\net40\MathNet.Filtering.dll"
#load @"..\packages\FSharp.Charting.0.91.1\lib\net45\FSharp.Charting.fsx"

let length, samplingRate,  amplitude = 750, 44100.,20.
let flow, fhigh = 200., 15000.
let lo = Generate.Sinusoidal(length,samplingRate,flow,amplitude)
let hi = Generate.Sinusoidal(length,samplingRate,fhigh,amplitude/4.)
let sumLoHi = lo |> Array.map2 (+) hi
let fs ,fcut, order = float(samplingRate), 5000., 10 

// first low pass filter
let lowPass2 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut,order)
let filtered = sumLoHi |> lowPass2.ProcessSamples

// second low pass filter
let lowPass3 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut/5.,order)
let filtered_2 = sumLoHi |> lowPass3.ProcessSamples

// plot
Chart.Combine([Chart.Line(lo, Name="lo", Color=Color.DarkBlue);
               Chart.Line(sumLoHi, Name="sumLoHi", Color=Color.LightSkyBlue);
               Chart.Line(filtered, Name="filtered ratio fcut/SamplingRatio = 0.1", Color=Color.Purple);
               Chart.Line(filtered_2, Name="filtered ratio fcut/SamplingRatio = 0.02", Color=Color.Fuchsia)])
|> Chart.WithLegend(Title="Combined",Docking=ChartTypes.Docking.Bottom)
|> Chart.WithTitle("OnlineFilter")

结果是：

我已经指定了samplingRate和过滤频率之间的比率，因为我认为问题问题与此参数有关。 是不是我不应该以这种方式使用OnlineFilter，还是其他的东西？

我也试过使用IIR和FIR滤波器，结果是一样的。 相反，python / scipy的类似情况效果很好。

I am trying to use a low-pass filter with F# using MathNet.Filter and when I plot the results is not working. I have found some similar questions in C#, but when I reproduce the code in F# The result is wrong.

Here it is the way I am using MathNet.Filter:

let fs ,fcut, order = 5000., 2.5, 5  // set the filter parameters
let lowPass2 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut,order)
let disp_beam_f = disp_beam |> lowPass2.ProcessSamples // apply the filter

// I print max/min to do a quick check
(disp_beam_f |> Array.min, disp_beam |> Array.min )

Then, when I plot some data my result is: When I plot just the filtered data, the plot looks like the original data, but with a different scale: I have already tried changing the order of the filter, and not using the order of the filter, the result is better, but it is not the filtered data.

• What I am doing wrong? I understand that fs is the frequency sampling (in Hz), fcut the frequency at which high frequencies are cut (in Hz) and order represents how much (or fast) the signal for frequencies above fcut are attenuated.

For more information, here there is a script: https://gist.github.com/josesoyo/f45b59a035c3e2ec2cdf0fe1f95cc84b
and a sample of data is in : https://drive.google.com/open?id=1rA4B21i-S5g2wrB4eK7xk3_A6Rs0tLHf
The result should be this file: https://drive.google.com/open?id=1ukmWtzFm_I4ou3zuQcLWCrFIOJs2eV41

Edit Based on Gene Belitski answer which I believes clarifies the case:

Running the next script shows better what is the problem I have:

#I __SOURCE_DIRECTORY__
#r @"..\packages\MathNet.Numerics.3.8.0\lib\net40\MathNet.Numerics.dll"
#r @"..\packages\MathNet.Filtering.0.4.0\lib\net40\MathNet.Filtering.dll"
#load @"..\packages\FSharp.Charting.0.91.1\lib\net45\FSharp.Charting.fsx"

let length, samplingRate,  amplitude = 750, 44100.,20.
let flow, fhigh = 200., 15000.
let lo = Generate.Sinusoidal(length,samplingRate,flow,amplitude)
let hi = Generate.Sinusoidal(length,samplingRate,fhigh,amplitude/4.)
let sumLoHi = lo |> Array.map2 (+) hi
let fs ,fcut, order = float(samplingRate), 5000., 10 

// first low pass filter
let lowPass2 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut,order)
let filtered = sumLoHi |> lowPass2.ProcessSamples

// second low pass filter
let lowPass3 = MathNet.Filtering.OnlineFilter.CreateLowpass(MathNet.Filtering.ImpulseResponse.Finite,fs,fcut/5.,order)
let filtered_2 = sumLoHi |> lowPass3.ProcessSamples

// plot
Chart.Combine([Chart.Line(lo, Name="lo", Color=Color.DarkBlue);
               Chart.Line(sumLoHi, Name="sumLoHi", Color=Color.LightSkyBlue);
               Chart.Line(filtered, Name="filtered ratio fcut/SamplingRatio = 0.1", Color=Color.Purple);
               Chart.Line(filtered_2, Name="filtered ratio fcut/SamplingRatio = 0.02", Color=Color.Fuchsia)])
|> Chart.WithLegend(Title="Combined",Docking=ChartTypes.Docking.Bottom)
|> Chart.WithTitle("OnlineFilter")

And the result is:

I have specified the ratio between the samplingRate and the filtering frequency because I believe that the problem problem is related with this parameter. Is it that I shouldn't use OnlineFilter in this way or is it something else?

I have also tried to use IIR and FIR filters and the result is the same. Instead, a similar situation with python/scipy works well.