It came to my mind that it's possible that the chip I use can/does not update it's center frequency during operation.

So I edited the code to reinitialize the device completly for each step in the for-loop.

Here is the new loop:

for i in np.arange(50, 1000, SAMPLERATE/1e6):
    sdr = RtlSdr()

    # configure device
    sdr.sample_rate = SAMPLERATE
    sdr.center_freq = i*1e6  # Hz
    sdr.freq_correction = 60   # PPM
    sdr.gain = 'auto'  # 4
    samples = sdr.read_samples(8*1024)
    sdr.close()
    sdr = None
    # use matplotlib to estimate and plot the PSD
    power, psd_freq = mlab.psd(samples, NFFT=1024, Fs=SAMPLERATE /
                               1e6)
    psd_freq = psd_freq + i
    powe = np.concatenate((powe, np.array(power)))
    freq = np.concatenate((freq, np.array(psd_freq)))
    print(f"{i}/1000")

The print statement at the end is just to keep track of how long it's going to take. This bit ran for about 7 minutes, because it is very slow to reinitialize like this all the time. There has to be a better way...

I'll keep searching for that. During that time you can enjoy the new output: