将此For cylce替换为numpy矢量化操作？

Question

我正在尝试将这个'For cycle'替换为numpy矢量化操作，其主要思想是获得更好的性能，因为它将应用于大型数据集。我需要知道这是否可能，或者如果这个循环无法矢量化，我怎样才能获得更好的性能？

import numpy as np
import numpy.fft as fft

def compress(x1, x2, nfft):
   return fft.ifft(np.conj(fft.fft(x1, nfft)) * fft.fft(x2, nfft))

def chirp_comp(bw, T, fs, f0=0, phi=0):
    b = -bw / 2 + f0
    a = (bw / 2 + f0 - b) / (2 * T)
    tn = np.arange(0, T, 1 / fs, np.dtype(np.complex64))
    y = np.exp(1j * 2 * np.pi * (a * np.power(tn, 2) + b * tn + phi / (2 * np.pi)))

    return y, tn

(chirp_rg, t) = chirp_comp(250e6, 500e-6, 1500e6)
raw = np.random.rand(5, np.size(chirp_rg,0))
# Compress each range line
data_rg_compr =  np.empty(raw.shape, dtype= np.dtype(np.complex64)) # Initialize matrix

#THIS IS THE CYCLE 
for k in np.arange(np.size(raw, 0)):
    aux = compress(chirp_rg, raw[k, :], np.size(chirp_rg, 0))  # Correlation
    data_rg_compr[k, :] = aux[0:np.size(data_rg_compr, 1)]  # Save result for line k

print "ends"