Spend an evening with an SDR and you will "find" stations that are not there: a strong broadcaster appearing in two places at once, a signal that vanishes when you turn down the gain, a spike glued to the centre of the screen wherever you tune. Every one is an artifact of the receiver, not the air, and each kind has a distinct cause and a distinct tell. Learning to tell a phantom from a real signal is most of what separates reading a waterfall from guessing at it.
//Four Phantoms, Four Mechanisms
The confusion in most explanations is that these all get lumped together as "images" or "noise." They are four different things happening at four different points in the receive chain, and keeping them apart is the whole skill.
Mixer image - analog, before the digitiser
A superheterodyne receiver multiplies the incoming signal by a local oscillator to shift it to a fixed intermediate frequency. The catch is arithmetic: for a given LO and IF, two input frequencies - one above the LO and one below - produce the same IF. The one you want, and its image, sitting twice the IF away. Classic receivers suppress the image with a tracking preselector filter or an image-reject mixer before the mixing stage. On a direct-conversion SDR the related artifact is the mirror-image reflection about the centre caused by imperfect I/Q balance. The shared tell: when you retune, a mixer image slides the opposite direction to a real signal.
Alias - digital, in the sampling
Aliasing is a different beast entirely: it happens in the analog-to-digital converter. Any energy above the Nyquist frequency - half the sample rate - folds back in and appears at a false, lower frequency, indistinguishable after the fact from a real signal. This is why an anti-alias filter sits before the digitiser, as the how-an-SDR-works piece shows. An image is analog frequency-mixing; an alias is sampling fold-back. The tell: change the sample rate and an alias jumps, because its position is tied to the clock, not to any real frequency.
Intermodulation - analog, from overload
Push the front end too hard and it stops behaving linearly. Two strong signals then mix inside the receiver to produce false ones, the troublesome pair being the third-order products at 2f1 minus f2 and 2f2 minus f1, which land close to the real signals where no filter can remove them. Their signature is their slope: a third-order product rises (and falls) about 3 dB for every 1 dB change at the input, so it collapses roughly three times faster than a real signal when you add attenuation - and disappears entirely once you stop overloading the receiver. This is the same overload behaviour behind the gain-staging sweet spot.
Birdies and the DC spike - the receiver talking to itself
Some phantoms never came off the air at all. Birdies are internal spurs - harmonics of the dongle's own clock (the RTL-SDR's 28.8 MHz oscillator is a common culprit), plus switching-supply and USB noise. And the fat spike pinned to the centre of the display on a direct-conversion SDR is a DC offset in the front end, not a signal. Birdies shift when you change the sample rate because they track the clock; the centre DC spike is the giveaway that ignores everything - it stays at centre, does not move when you retune, and does not drop when you attenuate.
//The Three Tests, and Why They Work
You do not need to recognise a phantom on sight. Three cheap tests separate all four kinds from a real signal:
Add attenuation. A real signal drops about 1 dB for every 1 dB of attenuation. An intermod product drops about 3 dB per dB - three times faster - and can be driven into the noise while the real signal merely dips. The DC spike barely moves at all, because it is internal.
Retune the centre. Real signals hold their true frequency. A mixer image slides the opposite way. The DC spike stays locked to the centre of the display.
Change the sample rate. Aliases and clock birdies jump to a new position, because they are tied to the sampling and the oscillator rather than to any real frequency; a real signal and a mixer image keep their RF position.
//Sourcing and Framing
The mixer-image mechanism (LO plus or minus IF landing on the same IF, image at twice the IF, preselector / image-reject suppression) is standard superheterodyne theory; the opposite-direction-on-retune behaviour and the I/Q-balance mirror on direct-conversion SDRs are documented in SDR community measurements. The alias-versus-image distinction is standard sampling theory. The third-order intermod products at 2f1-f2 / 2f2-f1 and the 3:1 slope are from RF intercept-point references, and the RTL-SDR overload, clock-harmonic birdies and zero-IF DC-spike behaviour are from rtl-sdr.com's bench measurements paired with a direct-conversion DC-offset reference. The three discrimination tests are the practical consensus of those sources. The piece stays, like the rest of the signals cluster, at the textbook receiver-theory layer; the interpretation is that the four phantoms are routinely conflated, and keeping their mechanisms separate is what makes the display legible.