2002 May 28
13
Philips Semiconductors
Product specification
Sample rate converter with embedded high quality
dynamic noise reduction and expansion port
SAA4979H
7.1.4.1
Band-splitting
The frequencies of the difference signals of luminance
(delta Y) and chrominance (delta U/V) can be split
optionally into an upper band (HF) and a lower band (LF)
with a low-pass filter in both signal paths. The lower
frequency band signals (LF delta Y and LF delta U/V) are
used as input for the noise reduction function.
The lower frequency band of the difference signals can
also be used for the motion detection. If, for example, only
the lower frequency band contains information, the
specific picture content does not move or is moving slowly.
Optionally it is possible to bridge the band-splitting (control
input: unfiltered = 1).
7.1.4.2
Motion detection
The same signals (the noise reduction is applied to) are
also used to detect the amount of motion in the difference
signals. Therefore, the absolute values of the difference
signals are generated and limited to a maximum value.
TheabsolutevaluesofthedifferencesignalofU and Vare
then averaged. The signals are low-pass filtered for
smoothingthesesignals.Thefilteredsignalsareamplified,
depending on the setting of the control inputs:
Yadapt_gain and Cadapt_gain respectively.
The amplified signals, which correlate to the amount of
movement in the chrominance or luminance signal path,
are transferred into 1 out of 9 possible K factor values via
look-up tables. The look-up tables consist of 9 intervals,
each related to one K factor. The boundaries between the
9 intervals are defined by 8 programmable steps (control
inputs:Kstep0 to Kstep7).Thestepvaluesarevalidforthe
look-uptablesforboththechrominanceandtheluminance
path. For example, signal values between Kstep2 and
Kstep3 result in a K factor of K =
3
/
8
.
7.1.4.3
K factor
The amount of noise reduction (field averaging) is
described my means of the K factor. When K = 1 no
averaging is applied and the new field information is used.
When K = 0 no averaging is applied and thus only the old
field information is used like in a still picture mode. All
values inbetween mean that a weighted averaging is
applied. It is possible to use fixed K factor values if the
control inputs lumafix or chromafix are set to logic 1. The
possible fixed K factor values of the control inputs
Klumafix and Kchromafix are given in Table 6.
7.1.4.4
Noise shape
Possible shadow picture information in the chrominance
and luminance path, resulting from a low K factor value,
will be eliminated if the noise shaping is activated. The
noise shaping function can be switched off via the
microcontroller (control input: noiseshape).
7.1.5
N
OISE ESTIMATOR
The noise level of the luminance signal can be measured
within a programmable window (control inputs: ne_hstart,
ne_hstop, ne_vstart and ne_vstop). The correlation in flat
areas is used to estimate the noise in the video signal.
A large number of estimates of the noise is calculated for
every video field. Such an estimate is obtained by
summing absolute differences between current pixel
values and delayed pixel values within blocks of 4 pixels.
Within the lower part of the total range of possible
estimates 15 intervals are defined. Each interval is defined
by a lower boundary and an upper boundary. The lower
boundary is equal to the number of the interval, whereas
the upper boundary has a fixed relationship to the lower
boundary (control input: gain_upbnd).
The lower boundary is increased or decreased by 1 in
each field until an interval is found which contains at least
a predefined number of estimates, and is at the same time
lowestintherange.Thevalueofthelowerboundaryofthis
interval determines the current noise figure output. The
predefined number of estimates can be set via the
microcontroller (control input: wanted_value), and good
results were obtained with a value which is approximately
0.27% of the total number of blocks.
For video fields with a lot of noise the number of small
differences is very low, that means the number of noise
estimates in the lower intervals is close to 0. Contrary to
this, for clean sequences this number is very high. This
means that for clean sequences the noise estimate figure
will be close to 0, and for sequences with a lot of noise the
noise estimate figure (control output: nest) will reach 15.
To improve the performance of the noise estimator,
severalfunctionsareimplementedwhichcanbecontrolled
by the microcontroller. To increase the sensitivity of the
noise measurement a prefilter with different gain settings
is available (control input: Ypscale). Since the video
content, e.g. sequences with a lot of high frequencies, can
influence the noise estimate figure, a detail-counter is
built-in.