Image Rejection

Image rejection is the problem a receiver's frequency-conversion mixer creates by folding two entirely different input frequencies onto the identical output frequency — the actual desired signal, and an unwanted "image" frequency sitting the same distance from the local oscillator but on the opposite side — and an image-reject mixer is the specific architecture built to tell them apart before they become permanently indistinguishable.

The image problem is a direct, unavoidable consequence of how frequency mixing itself works mathematically, not a flaw in any particular mixer implementation. Mixing an RF input against a local oscillator produces output components at both the sum and the difference of the two input frequencies, and a receiver's intermediate frequency stage is built around that difference component. But the difference frequency is identical whether the original RF signal sat above the LO by exactly the IF amount, or below the LO by that same IF amount — both cases produce a mixer output at precisely the same intermediate frequency, and a simple, real-valued mixer has no mathematical way to distinguish which side of the LO the original signal actually came from.

This matters because the image frequency is a real part of the radio spectrum that can carry real interference, and once folded down to the IF it corrupts the desired signal exactly as if it belonged there. A receiver tuned to pick up a signal at one specific frequency has no inherent interest in whatever happens to be transmitting at the image frequency on the opposite side of the LO — but the mixer doesn't know that, and any signal, noise, or interference present at the image frequency gets downconverted and summed directly with the wanted signal at the IF, with no way to separate the two after the fact. This is precisely why image rejection has to be addressed before or during the mixing process itself, never afterward.

The traditional brute-force fix is a sharp RF filter ahead of the mixer, physically removing the image frequency before it ever reaches the mixing stage — straightforward in principle, but increasingly impractical as the chosen intermediate frequency gets smaller. A lower IF places the image frequency closer to the desired RF signal, which is often desirable for other receiver design reasons, but it also means the filter separating them has to be correspondingly sharper, bulkier, and more expensive to achieve adequate rejection — a filtering approach that works well for a high-IF design becomes impractical exactly where low-IF and direct-conversion architectures want to operate.

Image-reject mixer architectures, most commonly built around the Hartley or Weaver topology, solve the same problem through phase cancellation instead of physical filtering. Both architectures split the signal path into two parallel mixer branches driven by local oscillator signals maintained ninety degrees apart in phase — an in-phase and quadrature pair — each branch independently producing its own mixed output. When those two outputs are recombined with the correct additional phase relationship, the desired signal's two copies add constructively while the image's two copies add destructively, canceling to a residual that depends entirely on how precisely the two parallel paths actually matched each other, rather than on any physical filter's frequency selectivity.

This is the same underlying strategy — moving or canceling specific spectral content through a deliberate phase or timing relationship, rather than removing it with a physical filter — that chopper stabilization already applies in an entirely different application domain. Chopper stabilization relocates an amplifier's own offset and flicker noise to a frequency a filter can remove cleanly, specifically because that offset started out sitting at the same frequency as the signal it corrupted. Image-reject mixing instead cancels an unwanted frequency component through destructive interference between two deliberately phase-shifted paths, rather than relocating it — a different specific mechanism, but the same broader philosophy of using a circuit's own timing or phase structure to solve a problem physical filtering struggles with.

Real achievable image rejection depends almost entirely on how precisely the two quadrature paths actually match in both gain and phase, which ties this architecture directly to the device-matching discipline this series has already covered. A perfect ninety-degree phase relationship and perfectly equal gain between the two mixer branches would, in principle, cancel the image completely — but real circuits never achieve perfect matching, and any residual gain or phase error between the two paths leaves a corresponding fraction of the image uncancelled. The same common-centroid layout and device-matching techniques already covered for differential pairs and capacitor arrays apply here too, just governing a pair of parallel signal-processing paths instead of a pair of matched transistors or capacitors; real image-reject mixers typically achieve thirty to fifty decibels of rejection, bounded directly by how well that matching discipline was executed.

Image rejection is a reminder that the same frequency-domain thinking this series has already applied to precision analog design — modulating, demodulating, and canceling specific spectral content through deliberate phase relationships — extends directly into RF receiver design, solving a structurally different problem with a recognizably similar toolkit. Where the problem originates and what gets canceled differ completely between a chopper-stabilized amplifier and an image-reject mixer, but the underlying insight, that phase and timing can do what filtering alone cannot, carries across both domains unchanged.