Optical heterodyne detection is a method of extracting information encoded as modulation of the phase, frequency or both of electromagnetic radiation in the wavelength band of visible or infrared light.
[1] The comparison of the two light signals is typically accomplished by combining them in a photodiode detector, which has a response that is linear in energy, and hence quadratic in amplitude of electromagnetic field.
This technique became widely applicable to topographical and velocity-sensitive imaging with the invention in the 1990s of synthetic array heterodyne detection.
[2] The light reflected from a target scene is focussed on a relatively inexpensive photodetector consisting of a single large physical pixel, while a different LO frequency is also tightly focussed on each virtual pixel of this detector, resulting in an electrical signal from the detector carrying a mixture of beat frequencies that can be electronically isolated and distributed spatially to present an image of the scene.
Unlike RF band detection, optical frequencies oscillate too rapidly to directly measure and process the electric field electronically.
Hence the primary purpose of heterodyne mixing is to down shift the signal from the optical band to an electronically tractable frequency range.
Nevertheless, the intuitive pure-frequency heterodyne concept still holds perfectly for the wideband case provided that the signal and LO are mutually coherent.
Consequently, optical heterodyne detection is usually performed as interferometry where the LO and signal share a common origin, rather than, as in radio, a transmitter sending to a remote receiver.
[6] After optical heterodyne became an established technique, consideration was given to the conceptual basis for operation at such low signal light levels that "only a few, or even fractions of, photons enter the receiver in a characteristic time interval".
In optical heterodyne detection, the mixing-gain happens directly in the physics of the initial photon absorption event, making this ideal.
Hence, narrow electronic filtering near the difference frequency is highly effective at removing the remaining, generally broadband, noise sources.
(Of course, this is a highly idealized description; practical limits on the LO intensity matter in real detectors and an impure LO might carry some noise at the difference frequency) Array detection of light, i.e. detecting light in a large number of independent detector pixels, is common in digital camera image sensors.
However, it tends to be quite difficult in heterodyne detection, since the signal of interest is oscillating (also called AC by analogy to circuits), often at millions of cycles per second or more.
At the typical frame rates for image sensors, which are much slower, each pixel would integrate the total light received over many oscillation cycles, and this time-integration would destroy the signal of interest.
Thus a heterodyne array must usually have parallel direct connections from every sensor pixel to separate electrical amplifiers, filters, and processing systems.
[13] The time domain conjugate of this approach is Fourier transform heterodyne detection,[14] which also has the multiplex advantage and also allows a single element detector to act like an imaging array.
The physical position where each photon arrived is encoded in the resulting difference frequency itself, making a virtual 1D array on a single element detector.
[17] In RF detection the antenna is rarely larger than the wavelength so all excited electrons move coherently within the antenna, whereas in optics the detector is usually much larger than the wavelength and thus can intercept a distorted phase front, resulting in destructive interference by out-of-phase photo-generated electrons within the detector.
However, as noted above, scaling physical arrays to large element counts is challenging for heterodyne detection due to the oscillating or even multi-frequency nature of the output signal.
With a virtual array one can then either adaptively select just one of the LO frequencies, track a slowly moving bright speckle, or add them all in post-processing by the electronics.
One can incoherently add the magnitudes of a time series of N independent pulses to obtain a √N improvement in the signal to noise on the amplitude, but at the expense of losing the phase information.
The practical limitation is adjacent pulses from typical lasers have a minute frequency drift that translates to a large random phase shift in any long distance return signal, and thus just like the case for spatially scrambled-phase pixels, destructively interfere when added coherently.