Spectral (Frequency) Inversion
This description of the phenomenon of spectral inversion was written for the
RF Cascade Workbook 2005
user's manual, and provides a unique pictorial representation of what is happening when a frequency mixer multiplies
two frequencies together.
Click here to use the Frequency Conversion
All four combinations of the mixing equation shown next to the mixer above are covered. Figures
A and B, below, are equivalent results for both the LO-RF and the RF-LO cases. In case A, the subtraction results
in negative frequencies, but the negative terms are said to be “reflected” or “folded” about the 0 Hz (DC) axis.
The subtraction in case B does not produce any negative results, so there is no need to mathematically reflect it
about the 0 Hz axis. There is no real physical reflection mechanism occurring, it is merely a mathematical convenience.
The mixer has no way of determining which frequency is being subtracted from the other, regardless of how you happen
to write the equation.
No spectral inversion occurs in either the upper sideband (USB) or the lower sideband
(LSB) with a low-side LO injection (cases A and B). However, with a high-side LO injection, the LSB will exhibit
spectral inversion while the USB does not (cases C, D and E). An explanation of spectral inversion is provided in
the next paragraph. First, though, note that with a high-side LO there is an equivalent subtraction/reflection pair
similar to the low-side LO case. Case E is a special case where the LO functions as both a low-side and a high-side
signal. There, a portion of the band experiences spectral inversion while the other portion does not (termed "frequency
Spectral inversion occurs when the lower sideband of a high-side injected LO is used for the output, like examples
C, D and E. The result is the lower frequencies of the input band being translated into the upper frequencies of
the output band, and vice versa. Shading and arrow-tipped lines are used to track the relative band edges. A second
inversion will cancel out the original, but a third will reintroduce the inversion, and so on. Digital systems can
un-invert the spectral inversion in software or firmware if it knows to expect it. However, spectral inversion of
analog voice information results in unintelligible garble. If multiple conversions are required, then an even number
of inversions will result in no net inversion at the output while an odd number of inversions will always result
in a net inversion at the output.