Description
Banding refers to a pattern of light or dark lines strictly parallel to the sensor axes, horizontal or vertical depending on the readout architecture.
The phenomenon originates in the sensor's own electronics: each row (or column) of pixels is read sequentially by a dedicated amplifier, and any fluctuation in the reference signal during that readout introduces a systematic offset across all pixels in the read row.
On older CCD sensors, banding is rare because readout is serial and slow.
On modern CMOS sensors, where readout is parallel row by row, the slightest noise in the power supply, ground plane, or thermal regulation generates a line pattern.
Banding can be fine (individual lines 1-2 pixels thick) or wide (bands several dozen rows thick forming brightness steps), and is especially visible on uniform sky backgrounds after aggressive stretching.
It almost always signals a calibration defect or a power supply problem, and remains subtle on well-regulated modern sensors.
Visual signature
Parallel lines, strictly aligned with one sensor axis (horizontal or vertical, never diagonal or curved).
Line thickness varies with origin: fine 1-2 pixel banding typical of high-frequency readout noise, wide 10-100 pixel bands characteristic of low-frequency oscillation in the power supply.
On a single stretched dark, banding appears as a regular pattern of lines that are lighter or darker than the mean background.
On a poorly calibrated master, bands may cross the entire field or affect only a zone of the sensor (at the beginning or end of readout).
The pattern is fixed relative to the sensor: it does not rotate with the target, does not shift with dithering, does not change with telescope orientation.
On an OSC after debayering, banding can produce colored bands if only one of the two Bayer pattern rows is affected.
Differential diagnosis
To be distinguished from walking noise (diagonal streaks oriented along the direction of cumulative drift, not aligned with the sensor axes).
Not to be confused with a column defect or row defect (single line or restricted group, permanent sensor defect at a fixed position).
Different from rolling shutter on a CMOS sensor, which produces temporal distortion visible only on fast-moving objects (not applicable to long-exposure astrophotography).
Not to be mixed with linear atmospheric gradients (light-pollution gradient from a horizon, but with a soft transition, not sharp lines).
Different from posterization (luminance steps in gradients, not aligned with the sensor axes).
Check on a single dark: if banding appears there with the same intensity, the defect is linked to sensor readout; if it appears only in lights, suspect an external source (power interference, electromagnetic interference).
To be distinguished from fixed pattern noise (FPN): banding draws lines aligned with one sensor axis, whereas FPN is a fine, repeating pixel-level pattern. Both are fixed sensor noise types that calibration and dithering attenuate.
Probable causes
- Unstable or noisy 12 V power supply (cheap adapter, aging battery)
- Power cable too long or undersized, causing voltage drops
- Electromagnetic interference from a nearby device (poorly shielded dew heater, PWM fan, mains transformer)
- Poor electrical grounding between camera, mount, and power supply
- Inadequate bias library or no bias applied when the pipeline requires it
- In PixInsight, ImageCalibration with "Optimize" checked on a CMOS sensor (introduces artificial banding)
- HCG/LCG readout mode inconsistent between lights, darks, and bias frames
- Older sensor type such as the ASI1600, known for residual banding even when well calibrated
- Thermal regulation oscillating (Peltier cooler cycling on/off with poor regulation)
- Degraded or excessively long USB cable (usually causes corrupt frames or missing blocks rather than regular banding)
Course of action
- Check power supply quality: use a dedicated linear-regulated 12 V/5 A minimum supply, never a budget switching supply
- Measure actual voltage at the camera terminals: must remain stable at 12.0 V within 0.2 V under load
- Use short power cables (<2 m) of adequate gauge (2.5 mm2 minimum for 5 A)
- Build a rigorous bias and dark library: same readout mode (HCG/LCG), same gain, same offset as the lights
- In PixInsight: ImageCalibration with "Optimize" unchecked for modern CMOS sensors
- For OSC sensors, calibrate before debayering to avoid amplifying banding through interpolation
- Test electrical isolation: disconnect the dew heater and other active accessories, take new darks to check whether banding appears or disappears
- For persistent banding, use dedicated tools: CanonBandingReduction (Photoshop), the BandingReduction script (PixInsight), or Siril's linear suppression tool
- As a last resort on an affected image, a directional MultiscaleMedianTransform oriented perpendicular to the bands can attenuate it (never eliminate it perfectly)
- For older sensors with chronic banding (ASI1600), consider dedicated HCG readout mode and thorough calibration
The Doc's advice
Banding is almost always an electrical problem, not an optical one. Cheap power supply, cable too long, poor ground connection: that is where to look before suspecting your sensor. On modern CMOS sensors like the IMX571 or IMX455 that are properly powered and calibrated, you get zero visible banding. If you see banding, check your cables before crying in PixInsight. And above all, in ImageCalibration: "Optimize" unchecked for CMOS, make it a reflex, the option introduces more banding than it removes.
Think you can see this defect in your image?
Run a diagnosisFrequently asked questions
Why does my banding appear in the lights but not in the darks?
This indicates that the source of the banding is external to the pure readout chain: it only appears when additional current is flowing (sensor under live exposure, active accessories such as a dew heater or fan). Typical causes: electromagnetic interference from an accessory running during imaging, a poorly bonded ground that is only stressed during exposures, or current draw variation during a sub that creates ripple on the power supply. Discriminating test: reproduce the exact conditions of a light (sensor exposing, accessories connected, mount tracking) but with the shutter closed: if banding appears, it is external; if not, it is likely a very faint parasitic light source picking up noise from something nearby.
Does the "Optimize" option in PixInsight's ImageCalibration improve or worsen banding?
On an older CCD sensor, "Optimize" generally improves calibration by compensating for bias variations between frames. On a modern CMOS sensor (post-2015), "Optimize" degrades calibration and can introduce artificial banding: standard practice on CMOS is to disable it. The algorithm attempts to scale the master dark to each light by finding the optimal factor, but this optimization is not suited to the nature of CMOS bias and disrupts subtraction. For a standard CMOS workflow in PixInsight: ImageCalibration with master dark only, "Calibrate" checked, "Optimize" unchecked, no master bias.
Is banding more frequent in narrowband than in broadband RGB?
Not more frequent but more visible. Narrowband subs, which are longer and carry less useful signal per frame, are stretched far more aggressively at integration. Any residual calibration imperfection, including banding, comes out proportionally more. In moderate RGB, a 0.5 ADU band is invisible; in a 600 s Ha sub stretched aggressively, the same band becomes visible. Calibration rigor and power supply quality are therefore especially critical in narrowband work.
Can banding already present in the master be definitively removed?
Partially. Banding reduction tools (dedicated PixInsight scripts, Photoshop plugins, Siril function) apply a per-row or per-column mean subtraction: they remove most of the pattern but introduce a slight residual in high-signal areas (bright stars, dense nebulosity) where the per-row mean is biased. On fine, weak banding, the result is generally satisfactory. On pronounced banding, the suppression leaves visible residuals in bright zones and usually leads to redoing the session with corrected electrical setup. The rule: prevention through clean power and rigorous calibration, not recovery in post-processing.