Note: Descriptions are shown in the official language in which they were submitted.
2144111
Field of Invention
This invention relates to digital video transcoding. In
particular, this invention relates to an oversampling rate
converter for transcoding an input digital signal which converts
the sampling rate of an input signal to a signal at a different
sampling rate, using a single sampling rate converter and filters
of substantially reduced complexity and improved performance.
Backaround of the Invention
In many video applications it is desirable to convert a
digital signal from one sampling rate to another, according to the
format requirements of different devices. The conversion of video
signals between composite and component formats requires an
encoder or a decoder, depending upon the direction of the
conversion, and a sampling rate converter.
In such a conversion polyphase filters are used to
calculate data values for the signal at times other than the
initial sampling times, which requires a separate filter for each
sample subphase. However, component and composite sampling rates
do not have simple integer ratios between the sampling
frequencies. Thus, large finite impulse response (FIR) filters,
mainly implemented in polyphase structure, are required to
interpolate and decimate the input signal to achieve the desired
sampling rate.
Conventional sampling rate converters are implemented in
the component domain. This requires a separate complex rate
converter for each of the three components of the signal and very
large and complex low-pass filters.
Figures 1 and 2 illustrate a conventional sampling rate
conversion from 4:2:2 with a sampling frequency FS of 13.5 MHz to
4fsc with a converted sampling frequency F~ of 14.3 MHz. The
conventional transcoder for this conversion is illustrated in
2144111
- 2 -
Figure 2. Figure 1 illustrates the steps through the conversion
process, in which the signal is interpolated at a 35:1 rate and a
large polyphase FIR filter is used to remove unwanted bands in the
component domain. The signal is then clocked out of the polyphase
filter at the 14.3 MHz rate.
This approach presents a considerable problem in terms
of the design of the required linear phase low pass filter to meet
the frequency domain specifications. The minimum length L of a
linear low-pass FIR filter required to meet the frequency domain
specifications is given by the following equation:
L - -201og(~85) ~2-13 +1
14.60F
where ~ and SS are the passband and stopband ripples,
respectively, and OF' is the transition bandwidth.
If the maximum passband ripple is assumed to be 0.02dB
and the maximum stopband attenuation is assumed to be 60dB, ~ and
8S become 0.0023 and 0.001, respectively. For a conventional
sampling rate converter the transition bandwidth OF' can be seen to
be OF = (6.75 -5.75)/(13.5 x 35) - 0.02116, which results in a
filter length L of 1404.
This problem becomes even more acute in the case of
conversion to 4fsc with a sampling frequency of 17.7 MHz in the
case of the PAL television standard, where the minimum filter
length L is greater than 709,000. The complexity and costs
associated with such a filter renders the design extremely
impractical, if not impossible.
Summary of the Invention
The present invention overcomes these problems by
~14~11~
- 3 -
providing a digital signal transcoder having a sampling rate
converter using polyphase filters which produce accurate signal
conversion without requiring inordinately long and complex
filters. The invention enables the conversion process to be
carried out in the composite domain using a single sampling rate
converter, which provides much better performance and far less
complexity than the three converters required by conventional
methods.
According to the invention the digital input signal is
oversampled by an integer amount before interpolation and
decimation by the sampling rate converter, which substantially
reduces the complexity of the required sampling rate converter.
Using the above example of a sample rate conversion from 4:2:2
with a sampling frequency FS of 13.5 MHz to 4fsc with a converted
sampling frequency F~ of 14.3 MHz, using a 1:2 interpolation for
oversampling of the input signal the transition bandwidth becomes
OF = (27 -6.75 -5.75)/(13.5 x 2 x 35) - 0.1534, which results in a
minimum filter length L of 194. The required length of the FIR
filter is thus decreased many times by implementing oversampling
of the input signal before the conversion.
The present invention thus provides a transcoder for
converting a digital input signal having a first sampling rate to
a digital output signal having a target sampling rate, comprising
means for oversampling the input signal to produce an oversampled
signal, means for converting a sampling rate of the oversampled
signal to produce an oversampled output signal, means for
filtering the oversampled output signal to reduce unwanted signal
bands, and means for decimating the oversampled output signal to
produce a transcoded signal at the target sampling rate.
The present invention further provides a method of
transcoding a digital input signal having a first sampling rate to
a digital output signal having a target sampling rate, comprising
X144111
- 4 -
oversampling the input signal to produce an oversampled signal,
converting a sampling rate of the oversampled signal to produce an
oversampled output signal, filtering the oversampled output signal
to reduce unwanted signal bands, and decimating the oversampled
output signal to produce a transcoded signal at the target
sampling rate.
Brief Description of the Drawincts
In drawings which illustrate by way of example only a
preferred embodiment of the invention,
Figure 1 is a graphic representation of the spectrum
progression of a signal conversion according to the conventional
method,
Figure 2 is a block diagram illustrating the
conventional transcoder for the method shown in Figure 1,
Figure 3 is a block diagram illustrating a preferred
embodiment of the signal transcoder of the invention,
Figure 4 is a graphic representation of the spectrum
progression of a signal conversion according to the embodiment of
Figure 3,
Figure 5 is a block diagram illustrating the transcoding
method for the embodiment of Figure 3,
Figure 6 is a block diagram illustrating a further
example of the transcoding method of the invention,
Figure 7 is a block diagram illustrating a transcoder
for the method of Figure 6, and
Figure 8 is a graphic representation of the spectrum
~1~4111
- 5 -
progression of a signal conversion according to the embodiment of
Figure 7.
Detailed Description of the Invention
Figures 1 and 2 respectively illustrate the spectrum
progression and components of a conventional sample rate
converter. As shown in Figure 2, the input signal, in the example
illustrated a 4:2:2 component signal with a sampling frequency FS
of 13.5 MHz, is fed into a component decoder 10 which decodes the
signal into its primary component parts. The three components of
the signal are passed to an initial polyphase FIR filter 20 which
interpolates the signal at 1:35. This signal is then passed to a
second polyphase FIR filter 30 which decimates the signal at 33:1
in order to produce the required output at the target sampling
frequency. The signal is then passed to an encoder 40 which
converts the signal into a composite video signal.
According to the invention, an embodiment of which is
illustrated in Figure 3, the signal to be converted is first
oversampled by being fed into a fixed coefficient interpolator 50
which produces a 1:K interpolation of the input signal, and
outputs an oversampled signal at K x FS, where FS is the sampling
rate of the unprocessed input signal and K is an integer greater
than 1. The oversampled signal is then input to a FIFO buffer 60.
The output of the FIFO buffer 60 is fed into a polyphase
interpolator 70 whose input is clocked at the buffer output
sampling rate of K x F~. The combination of the FIFO buffer 60 and
the polyphase interpolator 70 eliminates the process of 1:M
interpolation, eliminates unwanted frequency bands and provides
the NK:1 decimation. The output is the signal at the target
sampling frequency.
Figures 3 to 8 illustrate examples of a preferred
X144111
- 6 -
embodiment of the invention. Figures 3 to 5 illustrate an
embodiment of the invention converting a 4:2:2 video input signal
to a 4fsc output signal. In order to significantly reduce the
complexity of the converter, the 4:2:2 input signal is first fed
into a fixed coefficient filter 50 which interpolates the input
signal. In this example the filter 50 provides a 1:2
interpolation of the 13.5 MHz input signal to produce a digital
output signal with a sampling frequency of 27 MHz, twice that of
the input signal.
The interpolated output signal from the filter 50 is
then fed into a FIFO buffer 60 having an input clock rate matching
the sampling rate of the interpolated signal output from the
filter 50, in this example 27 MHz. The FIFO buffer 60 has an
output clock rate set at twice the signal sampling rate of the
desired converted output signal, and thus outputs the converted
signal at a sampling rate of 28.6 MHz to the polyphase
interpolator 70.
The output clock for the FIFO buffer 60 is used as an
input clock for the polyphase interpolator 70, which reduces
unwanted bands and decimates the signal at 2:1 to output a signal
at the desired transcoded signal sampling rate, 14.3 MHz.
Figures 6 to 8 illustrate another example of an
embodiment of the invention, which converts a 4fsc video signal to
a 4:2:2 signal, the reverse of the process in the above example.
The input signal with a sampling rate of 14.3 MHz is fed to the
input of a fixed coefficient filter 80 providing a 1:2
interpolation, to output an oversampled signal at 28.6 MHz, twice
the sampling rate of the input signal. The input clock rate of
the polyphase interpolator 90 is is the same clock rate of 28.6
MHz as is the input stage of FIFO buffer 100. After interpolation
and decimation by the polyphase interpolator 90 which also removes
unwanted bands the FIFO buffer 100 produces an output signal at a
sampling rate of 27 MHz, twice that of the desired 4:2:2 output
~1~4I1~
signal. The output clock for the FIFO buffer 100 is used as a
clock for the fixed coefficient filter 110, providing a 2:1
decimation, to output a signal at 13.5 MHz, the desired output
sampling rate.
It can thus be seen that by oversampling the input
signal prior to processing to the converted format the minimum
length L of the required polyphase filter is substantially
reduced. The invention thus makes high quality video
interpolation filters with fixed coefficients available at low
cost.
The above examples utilize a 1:2 oversampling ratio,
however the oversampling rate can be based on an interpolation
with any integer K greater than 1, and the invention is in no way
limited to the specific examples given. However, for video
signals a smaller integer ratio, such as the 1:2 interpolation
used in the example described, is preferred; because digital video
is already sampled over lOMHz, too large an integer K will make
the interpolation filter difficult to implement. A reasonable
upper limit for K in the case of digital video signals is 16.
