基本上，解决方案是将freqz函数（我不需要完全重写freqz，只是部分逻辑）重写为C#沿着依赖代码（使用Numpy）：

public static (NDarray w, NDarray h) freqz(NDarray b,
                                           NDarray a,
                                           NDarray? worN,
                                           float fs = (float)Math.PI * 2)
{
    using var b_ = np.atleast_1d(b);
    using var a_ = np.atleast_1d(a);

    NDarray w = np.atleast_1d(worN);

    w = 2 * np.pi * w / fs;

    using var i = np.array(new Complex[] { new(real: 0, imaginary: -1) });
    var zm1 = (i * w).exp();

    var nppPolyval_b = npp_polyval(zm1, b_);
    var nppPolyval_a = npp_polyval(zm1, a_);
    NDarray h = nppPolyval_b / nppPolyval_a;

    w = w * fs / (2 * np.pi);

    return (w, h);
}

private static NDarray npp_polyval(NDarray x, NDarray c)
{
    using var c_ = np.array(c, ndmin: 1, copy: false);

    var c0 = c_[":, -1"] + x * 0;

    for (int i = 2; i <= c_.size; i++)
    {
        c0 = c_[$":, -{i}"] + (c0 * x);
    }
    return c0;
}