为什么R的汉明函数与Matlab的不同?

7kqas0il  于 2023-10-13  发布在  Matlab
关注(0)|答案(1)|浏览(115)

为什么R中的hamming函数会给予Matlab中同名函数不同的值?
Matlab:

hamming(76)

    0.0800
    0.0816
    0.0864
    0.0945
    0.1056
    0.1198

R:

library(gsignal)
hamming (76)

0.08695652 0.08851577 0.09318288 0.10092597 0.11169213 0.1254078
pw9qyyiw

pw9qyyiw1#

算法

他们使用不同的公式来计算汉明窗

  • R中,

(25/46)-(21/46)* cos(2 * pi *(0:n)/N)
输入hamming时可以查看

> hamming
function (n, method = c("symmetric", "periodic")) 
{
    if (!isPosscal(n) || !isWhole(n) || n <= 0)
        stop("n must be an integer strictly positive")
    method <- match.arg(method)
    if (method == "periodic") {
        N <- n
    }
    else if (method == "symmetric") {
        N <- n - 1
    }
    else {
        stop("method must be either 'periodic' or 'symmetric'")
    }
    if (n == 1) {
        w <- 1
    }
    else {
        n <- n - 1
        w <- (25/46) - (21/46) * cos(2 * pi * (0:n)/N)
    }
    w
}

和 * 两个系数 * 有一定的精度差异

> 25/46
[1] 0.5434783
> 21/46
[1] 0.4565217

分析

造成差异的两个原因
1.在gsignals::hamming中,如果默认选择symmetric,则将有N <- n-1n <- n-1,这意味着与Matlab的hamming等效的应该是hamming(76+1)

  1. Matlab中的近似0.540.46导致了与R版本的偏差。

手工实现测试

我们可以分别根据MATLABR中的公式实现hamming函数,例如,

hamMatalb <- function(n) 0.54 - 0.46 * cos(2 * pi * (0:n) / n)
hamR <- function(n) 25 / 46 - 21 / 46 * cos(2 * pi * (0:n) / n)

然后运行

> hamMatalb(75) # the same as obtained by hamming(76) in MATLAB
 [1] 0.08000000 0.08161328 0.08644182 0.09445175 0.10558687 0.11976909
 [7] 0.13689893 0.15685623 0.17950101 0.20467443 0.23219992 0.26188441
[13] 0.29351967 0.32688382 0.36174283 0.39785218 0.43495860 0.47280181
[19] 0.51111636 0.54963351 0.58808309 0.62619540 0.66370312 0.70034314
[25] 0.73585847 0.77000000 0.80252824 0.83321504 0.86184514 0.88821773
[31] 0.91214782 0.93346756 0.95202741 0.96769718 0.98036697 0.98994790
[37] 0.99637276 0.99959650 0.99959650 0.99637276 0.98994790 0.98036697
[43] 0.96769718 0.95202741 0.93346756 0.91214782 0.88821773 0.86184514
[49] 0.83321504 0.80252824 0.77000000 0.73585847 0.70034314 0.66370312
[55] 0.62619540 0.58808309 0.54963351 0.51111636 0.47280181 0.43495860
[61] 0.39785218 0.36174283 0.32688382 0.29351967 0.26188441 0.23219992
[67] 0.20467443 0.17950101 0.15685623 0.13689893 0.11976909 0.10558687
[73] 0.09445175 0.08644182 0.08161328 0.08000000

> hamR(75) # the same as obtained by gsignal::hamming(76) in R
 [1] 0.08695652 0.08855761 0.09334964 0.10129899 0.11234992 0.12642490
 [7] 0.14342521 0.16323161 0.18570516 0.21068824 0.23800559 0.26746562
[13] 0.29886168 0.33197355 0.36656897 0.40240529 0.43923112 0.47678818
[19] 0.51481302 0.55303893 0.59119778 0.62902190 0.66624601 0.70260898
[25] 0.73785576 0.77173913 0.80402141 0.83447617 0.86288979 0.88906296
[31] 0.91281211 0.93397064 0.95239015 0.96794144 0.98051542 0.99002390
[37] 0.99640019 0.99959955 0.99959955 0.99640019 0.99002390 0.98051542
[43] 0.96794144 0.95239015 0.93397064 0.91281211 0.88906296 0.86288979
[49] 0.83447617 0.80402141 0.77173913 0.73785576 0.70260898 0.66624601
[55] 0.62902190 0.59119778 0.55303893 0.51481302 0.47678818 0.43923112
[61] 0.40240529 0.36656897 0.33197355 0.29886168 0.26746562 0.23800559
[67] 0.21068824 0.18570516 0.16323161 0.14342521 0.12642490 0.11234992
[73] 0.10129899 0.09334964 0.08855761 0.08695652

我们可以检查结果窗口的长度

> length(hamMatalb(75))
[1] 76

> length(hamR(75))
[1] 76

频谱性能

给定n <- 75,我们可以使用freqz来查看频域中的差异。我们可以看到,差异主要在于阻带,但仍然很小,这并不影响滤波性能太多。

  • freqz(hamR(n))

  • freqz(hamMatlab(n))

相关问题