Approximately APS(i) = APS(n/2-i) + 10^(5-20*i/n)
Signals aren't "mirrored" but noise is. I first noticed this using a sampling frequency of 8K, and sample size of 512, running the data through Hanning, AutoPowerSpectrum and plotting out 10*log(APS). The plot looks fine on a logarithmic frequency scale but when I switched to a linear frequency scale, it was obvious that all of the small noise peaks were mirrored at 2KHz. I tried this again with different noise files, windows, higher sampling frequencies and larger sample sizes but they all have a "noise mirror" at exactly half the Nyquist frequency or one quarter of the sampling frequency. I'm thinking that I must have
made a stupid error somewhere.