Description
Walking noise is the most characteristic artifact of imperfect calibration combined with poorly configured dithering.
When a sensor retains residual hot pixels after calibration, and the telescope drifts slowly from frame to frame in a given direction -- whether deliberately through overly unidirectional dithering, or involuntarily through mount drift, flexure, or differential atmospheric refraction -- the defective pixels leave a diagonal trail after stacking.
Each frame adds the hot pixel at a slightly shifted position, and alignment reprojects all those positions along a single axis: the result is a pattern of fine streaks oriented along the cumulative drift direction.
This defect is particularly visible against a dark, uniform sky background, and immediately flags an acquisition workflow that needs fixing.
Visual signature
Diagonal streaks, parallel to one another, crossing the entire image in a single direction. The orientation is typically 30-60 degrees from the sensor's horizontal axis, corresponding to the RA/Dec drift direction during the session.
The streaks can be bright (residual hot pixels) or dark (residual cold pixels). They become especially glaring after an aggressive stretch of the sky background, giving the image a "striped" or "banded" appearance that is pathognomonic of this defect.
On a poorly processed master, one can count 50 to 200 individual streaks depending on the number of defective pixels and the number of stacked frames.
Differential diagnosis
Distinguish from non-dithered residual hot/cold pixels (same defective pixels, but stationary: they form isolated points rather than streaks -- it is the absence of dithering that changes the signature).
Do not confuse with a diagonal gradient (smooth luminosity variation, no fine linear structure).
Different from satellite trails, which are few in number (1-3 per image) and broader, whereas walking noise is massive and systematic.
If the streaks are perfectly horizontal or vertical, it is more likely a sensor readout defect (banding) rather than walking noise.
Verify that the streaks are aligned with the drift direction visible when blinking the subs before alignment: this is the primary diagnostic test.
Probable causes
- Dithering disabled at acquisition, combined with natural mount drift
- Dithering enabled but with too small an amplitude (<2 pixels) to decorrelate defective pixels
- Overly unidirectional dithering (degenerated spiral or linear pattern instead of random 2D)
- Failing dark calibration leaving too many residual pixels (see dedicated defect)
- Uncorrected mount drift (approximate polar alignment, atmospheric refraction)
- Differential flexure between the imaging tube and the guide tube
- Very short individual exposures (<60s) with many frames, amplifying the contribution of each residual pixel
- Stacking by simple mean without sigma-clipping rejection
- Dark library outdated relative to sensor aging
Course of action
- Enable inter-frame dithering in NINA/SGP/APT/Voyager, amplitude >= 3 pixels, random 2D pattern
- Set the dithering frequency to every 1 to 3 frames depending on individual exposure length
- Redo adapted dark frames (same temperature, gain, duration as lights) to minimize residual pixels
- Enable CosmeticCorrection (PixInsight) or automatic defective-pixel detection (Siril) before alignment
- Stack with statistical rejection: Winsorized Sigma Clipping, kappa low/high 2.5-3.0
- Verify that PHD2 confirms proper execution of dither commands (settle distance and time)
- On an already-affected image, a MultiscaleMedianTransform targeted at high frequencies can reduce (but never eliminate) the streaks
- As a last resort on the final master, NoiseXTerminator or directional MMT aligned with the streak axis
- Re-acquire with correct dithering if the image is critical
The Doc's advice
Walking noise is a double penalty: your calibration is flawed AND your dithering is flawed. The good news is that both are fixed in five minutes for the next session. The bad news is that on the existing image, no matter how many filters you run, the streaks will never disappear completely. Walking noise is prevented at acquisition; it cannot be fixed in processing. Random 2D dithering of at least 3 pixels every 2 frames: non-negotiable.
Think you can see this defect in your image?
Run a diagnosisFrequently asked questions
Why must dithering be 2D and not linear?
Because linear dithering (for example, only in RA) displaces hot pixels along a single axis: at stacking, they form streaks in that direction instead of averaging out statistically. Random 2D dithering (both RA and Dec, with variable amplitude) distributes the positions of defective pixels in all directions, allowing sigma clipping to reject them pixel by pixel. In PHD2 and most sequencers, the 2D "random" mode is the default; verify that it has not been disabled.
What dithering amplitude is appropriate for a given image scale?
The empirical rule is to exceed the spatial correlation of defects, meaning a minimum of 2-3 pixels on the main sensor. For a standard image scale (1-2 arcsec/pixel), a dither of 3-5 pixels is sufficient. On a high-resolution setup (<1 arcsec/pixel), increase to 5-10 pixels to properly decorrelate. Beware of excessive amplitudes, which reduce the useful overlap area between frames and complicate post-dither guiding settle.
Is walking noise worse in narrowband than in RGB?
Often, yes, for two reasons. First, narrowband exposures are generally longer (300-600s compared to 60-180s in RGB), which amplifies the signal from residual hot pixels per frame. Second, narrowband images undergo very aggressive stretching to reveal faint nebulosity, making residual defects far more visible than in moderately stretched RGB images. Calibration and dithering discipline is therefore even more critical in narrowband.
My master shows walking noise even though dithering was enabled. Why?
Several causes to check in order: (1) is the dithering amplitude actually being applied -- verify the RA/Dec coordinates in the FITS headers of successive frames; (2) is the settle time sufficient for the mount to stabilize after the jump, otherwise the telescope continues drifting in the same direction; (3) does calibration leave an abnormally high number of residual pixels (redo recent darks); (4) is sigma clipping at stacking strict enough (kappa low <= 3.0). If all four points are nominal and walking noise persists, suspect a systematic mount drift that exceeds the dither amplitude.