The carrier that shouldn't have been there

Studio once produced a diagonal color weave and colored streaks around sprite edges. At some scales they passed for a richer analog look. At others they were clearly interference. We had to question texture we had already grown used to.

Two sampling rules that did not agree

The older encoder converted optical RGB into luma and chroma, mixed the chroma with sine/cosine carriers into a composite-like signal, and multiplied that signal by carriers again to produce the channels passed to transport. The stored values were therefore already carrier-dependent, not plain Y/I/Q awaiting a later encoder.

The carrier advanced in physical output-pixel coordinates, with both horizontal and vertical phase contributions. Meanwhile, transport tap spacing had been made proportional to presentation width. Linear interpolation between those stored carrier-bearing samples was not equivalent to interpolating the underlying channels and then evaluating a carrier at the requested position.

The rules lined up at the large reference presentation, then diverged at other scales. Sparse taps could pass the carrier and its second harmonic. Point-sampled fractional decoder offsets also caused abrupt changes during resize.

Studio's own signal path introduced the pattern. Detect Grid could not repair it.

Fixing the first explanation was not enough

Controlled comparisons disabled masks or bypassed the signal treatment separately. The baseline diagonal weave remained without masks and disappeared when signal treatment was bypassed. The masks also had a distinct problem: their old fixed-output-pixel structure did not follow the intended treatment scale.

We first scaled and integrated the material pattern, interpolated fractional decoder samples, and added a small rejection filter for the sampled carrier and its second harmonic. The interference faded, but so did useful fine texture.

Next we stored unmodulated Y/I/Q, interpolated the channels, and moved modulation/demodulation to the transport taps in one signal coordinate system. Small changes to the symmetric filter weights rejected constant-color carrier leakage while keeping the channel gains.

Constant-color interiors became nearly uniform. Structured images brought the diagonal interference back. Mask frequencies could mix with the carrier into lower-frequency sidebands inside the retained chroma band. A uniform mask removed that interaction, but sprite-edge fringes remained.

The flat-field test had missed interactions with image structure. More broad suppression risked losing detail again.

Remove the unsupported mechanism, keep the treatment

The decisive control was to remove carrier mixing. Encode kept signed Y/I/Q, and transport filtered the channels directly. We retained the tap positions, adjusted weights, channel gains, fractional decoder interpolation, luma reconstruction and output quantization. The extra rejection filter and unused trigonometry model could go.

Scale-aware masks, pixel-footprint integration, beam response, glow and tone stayed. Studio kept its display structure and differential Y/I/Q filtering while dropping the mechanism responsible for the unwanted texture.

Matched Jump captures at several scales showed the diagonal background weave and sprite-edge streaks disappear. The larger presentation retained soft transitions, color spread, horizontal beam structure and fine vertical material texture with less excessive chromatic texture.

Three meanings of phase

Removing the chroma carrier did not remove periodic display structure. Studio still has a calculated scanline phase and material patterns. Those are also distinct from the detector's estimate of the presented image's grid phase.

“Carrier-free” describes the Y/I/Q transport only. The CRT beam and masks remain periodic and sensitive to sampling phase.

Carriers still have a role in Beamline's different VHS signal path.

The row beam also needs filtering under curvature, independently of chroma-carrier removal.