Re: Blurry picture

From: Hársfalvi Levente <>
Date: Wed, 16 Nov 2011 17:50:47 +0100
Message-ID: <>

On 2011-11-15 18:28, Gerrit Heitsch wrote:
> I read somewhere that 2 crystals resulted in poor picture quality
> compared to using a PLL and only one crystal. Never understood why.

In short, moire.

A composite signal is a "blend" of two layers (ie. luma and chroma),
generally speaking, each with their own sampling frequencies (if
sampling applies).

Our videochips generally organize the screen in pixels that can all be
of arbitrary colors. (Not without some restrictions, true, the point is
that pixels don't have separate luma and chroma settings but colours).

The screen is "replayed" at the pixel clock, whose frequency is
generally in the 7-8MHz range.

Luma can be derived from the above, say, stream of pixels without

Chroma ultimately needs to be generated at a constant rate of the
respective color subsystems color subcarrier frequency - 3.58MHz in NTSC
and 4.43MHz in PAL.

The frequency of the pixel clock is usually higher than that of the
color subcarrier (which is otherwise equivalent with, say, that the luma
resolution is higher than chroma resolution). In a videochip which
generates a stream of coloured pixels at 7-8MHz, with the resulting
video output to be some sort of composite signal, be separate
luma/chroma or mixed, some downsampling of the chroma "layer" of the
pixel stream is apparent. (Good news is that chroma is quadrature
modulated, which happily tolerates if the modulation factors - the
chroma to be transmitted - change sometime inbetween, or practically
anytime; but some effects of downsampling will be apparent at the point
of demodulation, nevertheless).

The pixel clock (which is also the sampling freq of luma) and the color
subcarrier clock (the sampling freq of chroma) generally don't need to
be in any relation to each other - they're theoretically independent.
Let's suppose (for a theoretical step) that they're actually
independent. What'd we see? Luma generally won't be affected (no
resampling is done). But chroma is modulated onto some carrier, whose
frequency is clearly below the pixel clock, and whose phase is unknown -
most probably constantly shifting, as related to the pixel clock's
phase. Now, at the point of demodulation - ie. in the display - there'll
be interesting artifacts. Luma will be demodulated without problems. But
the chroma layer will be affected. Some particular pixels hue and
saturation will be greatly affected by the phase, as to how the colour
of that pixel "happened" to be modulated to the color subcarrier.

The effect that you can see varies, according to the connection type and
the display. For separate luma-chroma (s-video), colours will appear to
"flow" on top of the luma-picture, mostly to be noticed around the edges
of different coloured fields. For standard composite (and RF), the
effects will yet be slightly worse. In a composite signal, chroma is
added to luma. Then, at the point of demodulation (in the display), some
attempts are made to filter the resulting high-frequency chroma signal
residue from luma, and luma residue from chroma. This process would also
be affected by the constantly changing chroma signal - so, some luma
moire artifacts and some new colour artifacts will also be introduced.

The effect can be observed (generally speaking) on some early systems. A
typical example would be the ZX Spectrum (the ULA generates a YUV stream
at some multiple of the cpu clock, which is then encoded to PAL using a
standard PAL encoder + crystal). That's probably exactly what one could
see if the VIC-II had separate oscs for luma and chroma.

(Sinclair also went on to change their design, even at the expense of
introducing some incompatibility (slightly higher CPU clock and
different screen timing) to fix the artifact, starting with the 128k+

On a related note: the A520 ("RF Modulator") for the Amiga, even though
also built from standard components and a separate PAL oscillator,
features a hack which effectively syncs the osc to the clock of the
Amiga, in an attempt to get rid of the moire and colour artifacts.

Best regards,


       Message was sent through the cbm-hackers mailing list
Received on 2011-11-16 18:00:29

Archive generated by hypermail 2.2.0.