Symmetrical “Ghost” Signals: IMD, I/Q Image or Overload?

My first reaction to a beautiful pair of spectrum “ghosts” is not to name the mechanism. I freeze the receiver state, write down the absolute frequencies and ask what operation maps the real signals onto the suspicious ones. Symmetry is a clue only after we say what it is symmetric around.

There Is More Than One Kind of Symmetry

Suppose two real tones are at f₁ and f₂, with f₂ > f₁ and spacing Δf = f₂ − f₁. A weak memoryless nonlinearity can generate products of the form |m f₁ ± n f₂|. The order is |m| + |n|.

Those two products sit one parent spacing outside the two tones. Their midpoint is the midpoint of the parents. They are not, in general, mirror copies around the SDR’s tuned centre. Second-order products at f₁ + f₂ and |f₁ − f₂| have a different geometry and often fall far from the parents.

An I/Q image uses another symmetry. In a quadrature receiver, amplitude or phase imbalance can produce a conjugate copy across the complex LO or tuned centre:

A signal at +Δ relative to the LO then has an image at −Δ. That image follows the receiver centre when the LO moves; a genuine two-tone IM3 relationship follows its parent frequencies. A dc or LO-leakage spur at the exact centre is yet another case and should not be renamed an I/Q image.

Map the Candidate Mechanisms

The analogue and digital categories can overlap. A front-end amplifier or mixer may create a clean-looking product before conversion; the ADC may clip the entire composite waveform later; and any out-of-band harmonic or product that reaches a sampled stage can alias into the visible band. One trace cannot locate the guilty stage by itself.

Freeze the Receiver State Before Testing

Capture a baseline before touching a control. Record:

• absolute parent and candidate frequencies, tuned/LO centre, span and sample rate;
• FFT size or bin width, window, resolution bandwidth, detector and averaging;
• amplitude reference—dBm, dBc or dBFS—and whether the display is calibrated at the antenna connector;
• RF attenuator, preamp/LNA, analogue gain, AGC or manual-gain state, preselector and filter bandwidth;
• antenna or test source, cable/filter path and the physical reference plane; and
• ADC peak, headroom, clipping counter or overload indication, if the receiver exposes one.

Keep these settings fixed during each comparison. Changing FFT length, resolution bandwidth, window, averaging or detector can change the displayed noise and peak values without changing the RF. Likewise, a reference-level control may be only a plot scale in one SDR and may command analogue attenuation in another. Verify what the control actually changes.

dBFS is referenced to converter full scale, not automatically to watts at the antenna port. Treat dBFS as an internal receiver quantity unless a declared calibration and signal path translate it to dBm.

Run an Attenuation Sweep, Not a Magic Attenuator Test

Place a calibrated external step attenuator at the receiver input, or use a receiver attenuator known to sit ahead of the stage under investigation. Step it through several values while leaving AGC, gain, FFT and display settings unchanged. For every step, record both parent levels and the suspicious product.

In the small-signal power-series region, if both equal parents fall by A dB, an ideal second-order product falls by about 2A dB and an ideal third-order product by about 3A dB. Relative to one parent, that is an improvement of about A or 2A dB respectively. This slope is a diagnostic clue, not a universal promise: switched gain, unequal parent changes, AGC, filtering, multiple nonlinear stages, converter limits and the noise floor can all break the simple law.

If a candidate drops roughly with the wanted signal and their ratio stays similar, receiver overload has not been proved. The artifact may be an I/Q image, an on-air signal, or a product created before the chosen attenuator. If it changes abruptly when an analogue gain state changes, that strongly localises the problem—but only after the control’s position in the signal chain is known.

Use Preselection to Remove a Parent

A band-pass, high-pass, low-pass or notch filter can be more informative than simply reducing everything. Choose a filter that suppresses one suspected parent before the nonlinear stage while passing the candidate frequency. If the candidate disappears while a real signal at that same frequency would pass, the result supports a mixing relationship involving the rejected parent.

Account for filter insertion loss, impedance and any receiver gain-state change. A tunable preselector can also reveal a strong out-of-band parent that was invisible in the displayed span. Preselection helps only when it is physically ahead of the stage making the product.

Move the LO and Sample Rate Separately

Retune the LO while keeping the external signal fixed. An I/Q image reflects through the new centre; a true external signal stays at its absolute RF frequency. Then change sample rate, if the receiver permits it. A component that moves according to a new Nyquist fold is an alias candidate. A clock spur may move with sample rate, while a dc spur stays at the complex centre.

For an I/Q check, use one clean, safely levelled tone at a known offset and measure the wanted-to-image power ratio with fixed bandwidth and gain. Repeat at more than one offset and level because quadrature correction can vary with frequency, temperature and receiver state. Do not connect a transmitter or generator directly unless rated attenuation, isolation and the receiver’s maximum-input limit are all satisfied.

Separate Analogue Overload from ADC Overload

Turning off a preamp, reducing RF gain or adding input attenuation may restore linear operation in an LNA, mixer, IF amplifier or ADC. The useful question is where the level changed. If an attenuator lies after an overloaded amplifier, it can make the display smaller while leaving the distortion ratio unchanged.

An ADC clips when its input exceeds the converter range. Hard clipping creates strong distortion; it is not well represented by the smooth amplifier IP3 extrapolation used below compression. Analogue attenuation ahead of the converter can recover headroom. Digital gain, waterfall contrast or a lower visual reference line applied after conversion cannot undo clipped samples.

Converter harmonics and intermodulation products can themselves alias. Check both the unaliased formula and the folded location before labelling a visible peak. This matters especially when tones or harmonics sit near Nyquist-zone boundaries.

Confirm the Result with an Independent Path

A second receiver with a genuinely independent front end is an excellent witness. If the candidate appears at the same absolute RF frequency and comparable time in both paths, it may be on air or generated elsewhere in the site. If only one receiver creates it, repeat the attenuation, filtering and LO tests on that unit.

With the antenna disconnected and the port terminated in a suitable load, receiver-generated clock, dc and computer-interface spurs can become easier to map. That test does not prove that every disappearing signal came through the antenna normally; changing the termination also changes the conducted and common-mode environment. Use it as one observation in an A/B/A sequence.

Let the Diagnosis Choose the Remedy

• Analogue overload: reduce gain before the nonlinear stage, disable an unnecessary preamp, add suitable input attenuation, or reject unwanted bands before they reach that stage.
• ADC clipping: reduce the composite level at the converter, preserve analogue headroom and use the receiver’s overload indication instead of trusting waterfall appearance.
• I/Q image: verify calibration and correction for the receiver, frequency, bandwidth, temperature and gain state; attenuation alone does not repair quadrature imbalance.
• Alias: apply appropriate analogue anti-alias or band-selection filtering and choose a sampling plan that keeps unwanted zones out of the result.
• Clock or dc spur: trace power, grounding, shielding, interfaces and receiver configuration while observing whether the spur is fixed in RF or display coordinates.
• External signal or mixing: investigate the antenna system and site with independent instruments; do not “fix” a genuine emission by hiding it on one display.

Questions, antenna-factor records or height trials to share? Contact RF.Guru.