 |
Description  |
|
|
FIELD OF THE INVENTION
This invention relates to frequency conversion and more particularly to an
apparatus and a method for converting an input signal having frequency
related information sustained over a first length of time into an output
signal having substantially the same perceived frequency related
information sustained over a second length of time.
BACKGROUND OF THE INVENTION
Various types of frequency converter systems have been known to the prior
art. One specific application of such a frequency converter system is the
time compression or expansion of an audio, video or computer signal
normally stored in some sort of archival system, such as tape, magnetic or
optical disk or memory.
If time compression is desired for a stored signal, the stored signal is
produced at an increased speed to reduce the total duration of the
playback time of the stored signal. Unfortunately, when the stored signal
is produced at an increased speed, the signal undergoes an increase in
frequency relative to the stored signal being produced at a normal speed.
Conversely, when the stored signal is produced at a decreased speed to
expand the signal playback time, the signal undergoes a decrease in
frequency relative to the stored signal being produced at a normal speed.
Under either circumstance it would be desirable to provide the ability to
change the frequency of the produced signal by a percentage function based
on the ratio between the lengths of time of production. Alternately one
can change the frequency of a real time signal (i.e., without altering the
time period of production).
For example, it might be desirable to replay a prerecorded thirty (30)
minute television program in a time duration of twenty-eight (28) minutes
in order to fit an allocated time slot without the associated seven
percent (7%) increase in frequency. The replay of a prerecorded thirty
(30) minute television program in twenty-eight (28) minutes would allow
alternately for the insertion of an extra two (2) minutes of commercials.
Unfortunately, a viewer can discern the seven, percent (7%) increase in
pitch resulting from an increased speed in the replay of the signal.
Alternatively, a taped show may run only 45 minutes, with the station
having a 50 minute time slot. The present invention allows the show to be
expanded to fit the time slot.
In another example, entertainment or educational programs or movies could
be presented in shorter time to reduce the operating costs of a movie
theater or to allow more movies to be shown in a evening. A similar
advantage could be realized in the replay of prerecorded music or voice on
a radio station. Messages from an answering machine could be accelerated,
perhaps greatly, for rapid playback while retaining normal voice
frequencies. Again, the present invention removes the pitch shift
artifacts that would otherwise be ascertainable to the consumer.
A further example, some live talk shows have a six (6) second profanity
dump memory (that allows the selective deletion of expletives). Some of
these dumps, however, also can produce an audio gap after the expletive is
deleted due to the need to fill up their memory line with new audio
information. The present invention allows for an effective instantaneous
switch back to live audio, since delay can be gradually re-accumulated
without pitch change.
Another example would be to change the relative pitch of the human voice so
as to allow an individual to sing harmony with themselves in real time.
Another example would be to lower the occupied bandwidth of a signal to be
transmitted over a radio propagation or other transmission medium.
A last example wherein data may be fed in to a memory, perhaps
intermittently, at one speed and fed out at a second (normally slower and
perhaps constant) speed, thus facilitating computer operation or
elementary data operations.
Some in the prior art have attempted to develop a frequency converter
system with varying degrees of success. Although some of these frequency
converter systems functioned properly, many of these frequency converter
systems were excessively complex and costly to manufacture. Accordingly,
the frequency converter systems of the past have not found wide use in the
media art.
Frequency converter systems of the prior art include U.S. Pat. No.
4,829,257 to J. Carl Cooper for an improved device for accurately phase or
frequency shifting an input signal. This invention incorporated a variable
resistor extending between at least two known phase shifted values of the
input signal. U.S. Pat. No. 4,868,428 to J. Carl Cooper discloses an
apparatus and method for accurately shifting the phase or frequency of a
complex signal. U.S. Pat. No. 5,097,218 to J. Carl Cooper discloses an
apparatus and method for accurately multiplying the phase or frequency of
complex time varying signals by a given factor which may be non-integer.
OBJECTS AND SUMMARY OF THE INVENTION
Therefore, it is an object of the present invention to provide an improved
apparatus and method for frequency conversion with reduced complexity of
manufacture and operation.
Another object of the present invention is to provide an improved apparatus
and method for frequency conversion incorporating a linear interpolator
for reducing harmonic and/or other distortion.
Another object of the present invention is to provide an improved apparatus
and method for frequency conversion capable of decreasing or increasing
the time base of a signal without a significant change in frequency.
Another object of the present invention is to provide an improved apparatus
and method for frequency conversion capable of a significant decrease or
increase of the time base of a signal without significant change in
perceptible frequency.
An additional object of the present invention is to allow real time
frequency shifting of an input signal, for example, musical instrument or
human voice.
Another object of the present invention is to provide an improved apparatus
and method for frequency conversion capable of use with a computer based
storage and retrieval system of prerecorded programs or information.
A further object of the present invention is to increase the reproduction
utilization capabilities of video and audio recorders, movies and films,
answering machines, voice mail boxes, and other signal storage systems.
Another object of the present invention is to provide an improved apparatus
and method for frequency conversion with superior overall performance
heretofore unknown.
Other objects and a more complete understanding of the invention may be had
by referring to the following description and drawings in which:
BRIEF DESCRIPTION OF THE DRAWINGS
The structure, operation, and advantages of the presently disclosed
preferred embodiment of the invention will become apparent when
consideration of the following description, taken in conjunction with the
accompanying drawings wherein:
FIG. 1 is a block diagram of the theory of the invention;
FIG. 2 is a block diagram illustrating an improved frequency converter
system of the present invention connected to an example input signal
having a first frequency sustained for a first length of time for
generating an example output signal having substantially the same first
frequency sustained for a second length of time;
FIG. 3 is a graph illustrating a single cycle of all example input analog
signal at a first frequency;
FIG. 4 is a graph illustrating a translation of the single cycle of the
input analog signal of FIG. 3 into digital form;
FIG. 5 is a graph illustrating the selection of a digital sample for the
signal of FIG. 4;
FIG. 6 is a graph illustrating the addition of a duplicate of the selected
digital sample from the samples of FIG. 5 to provide a sample digital form
of a modified output signal;
FIG. 7 is a graph illustrating a type of linear interpolation of the
signals of FIG. 6;
FIG. 8 is a graph illustrating a transformation of the digital samples of
FIG. 7 into an output analog signal;
FIG. 9 is a graph comparing a single cycle of the input signal of FIG. 3
and the output signal of FIG. 8;
FIG. 10 is a graph setting forth the output signal of FIG. 8 as actually
perceived by the consumer due to its production at a higher reproduction
rate than the input signal of FIG. 3;
FIGS. 11-12 are drawings demonstrating the sampling and nature of signals;
FIGS. 13-15 are figures like FIGS. 6-8 showing an example deletion of a
digital sample;
FIG. 16 is a detailed block diagram of a frequency converter system;
FIG. 17 is a graph comparing a first signal and a second signal and
illustrating the reset of the phase angle of the signals at a common zero
cross-over;
FIG. 18 is a figure like FIG. 17 illustrating a reset at the top of a
signal waveform;
FIG. 19 is a graph of the constant, non-reset first signal in FIGS. 17 and
18;
FIGS. 20-21 are circuit diagrams of example frequency converter systems;
FIGS. 22-23 are block circuit diagrams of a MPEG implementation of the
invention;
FIG. 24 is a block circuit diagram of a multiple channel device;
FIG. 25 is a block diagram of a circuit for signal playback;
FIG. 26 is a series of representational block diagrams setting forth an
example resetting of memory lines in an expansion or contraction frequency
conversion device; and,
FIG. 27 shows an adaptive filter network for incorporation into the system
of FIG. 25.
DETAILED DESCRIPTION OF THE INVENTION
The present invention is directed to a conversion system and devices that
incorporate it, which conversion system can convert an input signal having
frequency related information normally sustained over a first length of
time into an output signal having substantially the same perceived
frequency related information, with the information now normally sustained
over a second length of time alternately just frequency, and/or frequency
and length of time can be modified. The theory behind this operation is
shown in the FIGURES, including FIG. 1.
The theory behind the invention involves getting an input signal 10 (Block
I). This signal has frequency based information sustained over a period of
time. This input signal 10 is provided to a signal modification circuit 50
(Block II). The signal modification circuit 50 adds or subtracts samples
15 to or from the input signal 10 according to certain principles,
mathematical principles normally based primarily on the ratio of frequency
and/or time between the input 10 and output 100 signals and the complexity
of the signals. The signal modification circuit 50 then remits an output
signal 100 (Block III), which output signal 100 has a relationship to the
input signal 10 as set by the certain mathematical principles.
One skilled in the art should recognize that the devices disclosed in this
application could alter frequency over the same length of time, alter
frequency and length of time, and otherwise function. The easiest way to
do this would be by altering sample and/or clock rates. For uniformity,
this application will primarily utilize as an example devices producing an
output signal that perceptibly has the same frequency related information
as the input signal 10 and may also be sustained over a different length
of time.
In this operation, both the input 10 and output 100 signals have frequency
related information on them. The output 100 signal can be either expanded
or compressed relative to the. input signal 10. The signals themselves can
be audio, television, computer signals, or other signals having frequency
related information thereon. Further, the devices can be used in singular
form (for example a television video signal), paired form (for example
right and left stereo audio signals), or in other combinations including
synchronizing the output signal to a related signal (for example
synchronizing audio to video). The signals themselves can be in analog or
digital form. A digital form is presently preferred in that technology is
presently more established for digital processing of complex wave forms.
However, with the increasing advances in analog circuitry including the
use of charged coupled devices (CCD's), it is envisioned that soon analog
processors may be able to process the complex signals as well and perhaps
better.
The digital signals may be coded in pulse code modulation (PCM), pulse
width modulation (PWM), pulse length modulation (PLM), pulse density
modulation (PDM), pulse amplitude modulation (PAM), pulse position
modulation (PPM), pulse number modulation (PNM), pulse frequency
modulation (PFM), pulse interval modulation (PIM), or other coding scheme.
Pulse amplitude modulation will be utilized in the explanation of the
invention.
The location of the signal modification circuit 50 in the overall
replication path is not critical. In most instances, the signal
modification circuit 50 would be located after some sort of signal storage
means for modification of the stored signal. This is generally preferred
in that the stored signal would contain the highest quality signal. Such
stored signal could also be otherwise used. However, the signal
modification circuit 50 could be located prior to the storage means or
even within such storage means. The circuit 50 could also operate in real
time. Further, the order of the conversion steps are not critical as long
as all steps are accomplished. For example, the clocking shift and analog
to digital conversion could occur prior to real time signal modification
in the overall frequency conversion of an analog signal. An example of
this would be playing an answering machine at high speeds with subsequent
real time frequency conversion to lower the voice pitch to normal values.
Further example in FIG. 2 the storage means could be located before/after
or between any of the blocks of circuitry at points A-H respectively. The
operation of the invention is thus also not dependent on a storage
location.
FIG. 2 is a block diagram of the signal modification circuit 50 receiving
an input signal 10 sustained over a first length of time 13 at a first
ascertainable frequency. In real time this length of time 13 would be the
period of production of the input signal 10. As the signal 10 utilized as
a uniform example in this specification is an analog signal, a digital
converter 14 converts the input analog signal 10 into a digitally sampled
version 20 of the input analog signal 10 (if the input signal 10 was
itself digital or already a digitally sampled version of an analog signal,
no conversion would normally be necessary; oversampling however, might be
appropriate. The inclusion of the converter 14 in the modification circuit
50 is thus dependent on the nature of the processed signals).
In the particular circuitry example of the figures, the input signal 10 is
an analog signal having an alpha length 13. Here, applicant defines alpha
length as the time duration of a contiguous signal block, exclusive of any
reset operations (reset operations will be addressed in detail below).
This input signal 10 is normally replicated over a certain time period, a
period normally directly related to the alpha length. The input signal 10
normally exists for reproduction over a certain set length of time, a time
length analogous to inverse clock rate including real time.
In FIGS. 2 and 16, input clock refers to the speed of production or
reproduction of the input signal. The input sample rate refers to the rate
at which discrete-time samples are presented to the signal modification
circuit. The input clock and input sample rate may or may not be related.
For example, if the input source is an analog tape player, the input clock
would refer to the speed of the tape. Tape speed might be variable, while
the input sample rate may or may not be variable. As another example, the
input source may be a compact disk player outputting digital samples at a
44.1 kHz rate. In this case, no continuous-time to discrete-time
conversion is necessary. If no sample rate conversion was used, the input
sample rate would here be the same as the input clock rate. If the speed
of playback of the compact disk was varied, then both the input clock and
input sample rate would vary. The output sample rate is the rate at which
discrete-time samples are output from the signal modification circuit. It
may or may not be equal to the input sample rate. The output clock rate
refers to the speed of production of the output signal, and may or may not
be related to the output sample rate.
The input signal 10 is normally preferably fed into a digital converter 14
in order to replicate such input signal 10 in digital samples 15. The
nature and rate of the digital sampling is selected in accord with the
overall circuitry design. Examples of the type of digital sampling that
can be utilized have been previously set forth. For uniformity, the
preferred embodiment of the invention will be set forth with pulse
amplitude modulation (PAM) digital sampling.
It is preferred that the digital coding and/or rate be selected in respect
to the nature and frequencies of both the input and output signals. For
example, according to the sampling theory, a sampling rate of a little
over twice the highest expected frequency will allow for the accurate
reproduction of an analog signal with minimal distortion. An example of
this is the 44.1 kHz sampling rate for common compact disks. In addition
to this, the sampling rate must be selected in order to provide for the
compression/expansion of the signal in an accurate manner. This entails a
review of the signal content. In specific, if a computer on/off binary
signal was involved with a conversion of 3:2, a sampling rate three times
the clock speed of the input signal would provide for completely accurate
conversion (FIG. 11). However, with an audio signal at the same somewhat
extreme example 3:2 reduction, a sampling rate of twice the frequency of
the input audio signal (for example a sampling rate of 44.1 kHz) would
provide a normally unacceptable result due to the distortion on the output
signal 100. The reason for this would be that aliasing would occur if
one-third (1/3) the samples were removed. It is therefore necessary to
sample the audio input signal 10 at a rate much higher than the Nyquist
rate in order to provide for an acceptable output signal for the analog
signal (FIG. 12). Over and above this restriction, it is preferred that
any input signal 10 be sampled at as high a rate as possible, in order
that the addition/deletion of individual samples would have a minimal
effect on the information available on such input signal 10. For example,
the deletion of one out of every ten samples at a 10,000 times over
sampling rate would have less artifacts than the deletion of one out of
ten samples at a ten times oversampling rate although both provide the
same 10 percent (10%) signal compression. The reason for this is that with
higher rate sampling, the many artifacts which would be produced would
occur at an extremely high frequency, with many occurring at a frequency
above that perceptible to the senses of the consumer. The Philip's pulse
amplitude modulation at a standard rate of 256 over sampling
(256.times.44.1 kHz) is a natural sampling technique for the invention in
audio applications.
The difference ratio 51 that is input to the actual modification circuit 52
determines the scope and nature of the relationship between the input 10
and output 100 signals. Examples of these relationships have been
previously given in the BACKGROUND OF THE INVENTION section. The general
concept is that there is an input signal 10 which has frequency related
information, which input signal 10 further has some frequency and/or time
ratio to the output signal 100, normally a ratio based on the times of
expected signal production. If time is the determinant, the difference
ratio is selected such that the output signal 100 when perceived has the
same frequency related content as the input signal 10. Alternately the
output signal 100 may have the same time as the input, but a different
frequency or both may be varied simultaneously.
The difference ratio may be defined as the output frequency-time product,
divided by the input frequency-time product. For example, suppose that the
difference ratio is 0.855. If the input and output times are the same,
then the output frequency is 0.855 the input frequency. If the input and
output frequencies are the same, then the output time is 0.855 the input
time. If the output frequency is 0.95 the input frequency, then the output
time would be 0.9 the input time, since 0.95 multiplied by 0.9 equals
0.855.
The difference ratio 51 can be set manually or automatically. An example of
the former would be having a technician dial in a factor representative of
the input length and then a second factor representative of the output
length. This type of manual setting would be particularly appropriate
where the technician knew that a thirty (30) minute television program
needed to be inserted into a twenty-eight (28) minute time slot. As an
example of the automatic setting, in television signals the horizontal
sync pulses could be utilized to automatically decompress a tape recorded
television movie. This type of automatic functioning would be particularly
appropriate for signals having known, repetitive, determinable attributes
or where the function of the circuitry can be readily determined (for
example a profanity dump).
In the circuitry of FIGS. 2 and 16, the difference ratio is as previously
defined. This ratio has been previously determined by the technician
responsible for conversion. Pitch shift may be obtained by sample
insertion or deletion, if the input and output sample rates are the same.
Alternatively, pitch shift may be obtained by using differing input and
output sample rates, without sample insertion or deletion. Additionally, a
combination of sample insertion/deletion and differing sample rates may be
used. The sample rates preferably are at least greater than the Nyquist
rate for both input and output. Over and above this, distortion
considerations could require that the input signal be sampled at a rate
much higher (for example 20 times) the highest input frequency in order to
insure production of the output signal with minimal distortion. Note,
however, that in non-critical applications the sampling rates can be much
lower, particularly if the signals can be band width limited while
retaining acceptable information content (an example of this would be band
width limiting an audio signal to 5 kHz).
As shown in FIGS. 3-15, the modification circuit 52 selects at least one
sample 22 from the digital sampled version 20 of the input signal 10 and
generates a second plurality of digital samples 120 by altering the number
of the digital sampled version 20 of the input signal 10 by the selected
digital sample(s) 22. This can be by addition to expand (samples 122 in
FIGS. 6 and 7) or by subtraction to compress (samples 22 in FIG. 13) as
appropriate. The location of the added/deleted samples is selected in view
of the signal content so as to minimize artifacts. For very high
oversampling rates, the samples can be spread out over the entire alpha
length of the signal. For lower oversampling rates, locations of least
slope, least differences, signal peaks, or other minimal signal
information points are preferred.
Note that FIGS. 3-15 are given by way of example. Other
sampling/modification methods could also be utilized with the invention.
Note also that for clarity of explanation in these figures that the input
sample is converted to digital by a leading edge sample and hold circuit
(left edge), while the output sample is converted to analog by a trailing
edge conversion circuit (right edge). Alternate conversion circuits could
be utilized if desired. For ease of comprehension, no interpolation is
used in FIGS. 3-15.
The second plurality of digital samples 120 can be interpolated to reduce
distortion caused by replication or deletion of the selected digital
sample 22 if appropriate. A second digital converter 114 then generates an
output signal 100 from the second plurality of digital samples 120 over
the second duration of time 113 (again the inclusion of this convertor is
dependent on the nature of the output signal). In the example shown, this
produces a signal having substantially the same first frequency when
clocked or reproduced at the new speed provided that the rate of
occurrence of altered samples 22 relative to the input sample rate
corresponds to the ratio between the first 13 and second 113 duration
times. compare the analog input signal 10 of FIG. 3 with the output analog
signal 100 of FIG. 10, which output signal 100 is being produced during a
different, shorter, length of time: In real time 80, the perceived
frequencies or alpha length of the signals are the same. In this respect,
it is noted that minor unobjectionable shifts could be accepted by the
overseeing technician, this even though the pitch of the resultant signal
is not absolutely accurate. Alternately the output may have a different
frequency and same duration or a combination of different frequency and
duration.
It should again be appreciated by those skilled in the art that either the
digital converter 14 or the digital converter 114 may be optional in the
event that either the input signal 10 or the output signal 100 is a
digital signal of suitable signal content to allow for acceptable
modification. Otherwise, oversampling would normally be appropriate.
The selected digital sample(s) 22 is added (FIG. 6) to the sampled version
20 of the input signal 10 to provide pitch correction to facilitate the
prior or subsequent decrease of the duration time. The selected digital
sample(s) 22 is removed (FIG. 13) from the sampled version 20 of the input
signal 10 to provide pitch correction to facilitate the prior or
subsequent increase of the duration time. This will be discussed more
extensively later, especially with respect to FIGS. 16-18.
The signal modification circuit 50 adds or subtracts to the apparent alpha
length of the input signal 10 according to the difference ratio in order
to produce the output signal 100.
With present digital technology, this signal modification circuit 50 would
begin with digital signals or a digital sampling replication 15 of an
analog signal. This signal modification circuit 50 then repeats (to add or
expand) or deletes (to subtract or compress) from these samples in order
to alter the input signal 10 to an output signal 100 in accord with the
set difference ratio. (Note again that a digital signal might, like an
analog signal, have to be digitally over sampled to achieve acceptable
distortion performance.)
A more sophisticated signal modification circuit 50 could average or
linearly or otherwise interpolate the modified samples in order to
optimize the functioning of the device in certain applications.
For example, a theoretical analysis of the spectrum resulting from applying
a 5.6 percent expansion to a sine wave was performed. The sine wave period
was 36 samples. The purpose of the analysis was to determine the relative
distortion levels resulting from the insertion of samples at different
phases of the input sine wave. Two samples per period were inserted. The
type of linear interpolation used in the preferred embodiment is
essentially the second-order approximation method used for sample rate
conversion. Proakis, John G., Rader, Charles M., Ling, Fuyun, and Nikias,
Chrysostomos L., Advanced Digital Signal Processing, Macmillen, 1992. This
particular interpolation operates along the entire signal, and weighs two
adjacent samples in proportion to the distance from the last sample
insertion. This type of interpolation gives exact (within quantification
limits) output: in the case of a linear ramp signal with constant slope
(zero second difference). Three insertion points were examined: peaks,
zero crossing, and 30.degree. lagging from zero crossing. Insertion of
samples at the zero crossing resulted in the lowest distortion of the
three. Insertion of samples at the peaks was slightly inferior to the zero
crossing. Insertion of samples at the 30.degree. points resulted in the
highest distortion. A feature of the zero crossing point of a sine wave is
that it is the location of minimum second difference magnitude. A feature
of the positive and negative peaks of a sine wave is that they are the
location of minimum slope, or minimum first difference, magnitude. A
feature of points on the sine wave which are removed from the peaks or
zero crossings, such as the 30.degree. phase point, is that neither the
first or second difference is minimized.
Sample insertion at the peaks and at the zero crossings was also
investigated for the case of no interpolation. In this case, insertion at
the peaks gave lower distortion. Distortion performance in both cases was
significantly worse than that obtained where interpolation was used.
It is envisioned that other interpolation algorithms may be advantageous
from a performance standpoint. Examples of other algorithms would be
polyphase subfiltering, higher order Lagrange polynomial interpolation,
and finite impulse response low-pass filtering. In the case of expansion,
it may also be advantageous for the added sample to be of some value other
than the value of the immediately preceding sample.
Customarily, the signal modification circuit 50 would. contain a delay
length preferably of the variable type, the length of which must at least
allow for the appropriate shifting of the signals to add or subtract whole
cycles. Specifically, the maximum shift between the input signal 10 and
the output signal 100 should be within the effective delay length of the
available memory. Further, to provide for a smooth output signal 100, the
present invention preferably uses a memory delay longer than this period
in order to provide for a seamless operation.
The present invention accomplishes this with less than the amount of memory
otherwise needed by comparing a first and second signal in order to reset
the information in the memory (as later described) to add or delete blocks
or cycles thus to provide for a seamless integration of the signals. In
general, the more memory that is available, the more time can pass before
the device is reset subject to an ascertainable artifact override. For
complex audio signals, up to a point, the quicker the resetting; normally
the less ascertainable the artifacts as will be described in more detail
below. In addition, an even greater memory would allow an operator to
delete or add blocks, cycles, or multiple cycles of signal information
with no pitch change.
The delay could be a single memory if variable taps were available and the
signal was actively followed through such variable taps by the signal
modification circuit 50. Changing over between effectively two separate
memory delay lengths 56, 57 is preferred, which delay lengths are each
long enough to provide for the later described resetting, because of the
way the design evolved. A single memory delay element was originally used.
Later, a second memory delay element was added, since this was, at the
time, the easiest way to accommodate the needed later described cross fade
reset operation. At any given time, exclusive of reset operations, one
memory circuit would be actively utilized in real time by the signal
modification circuit, while the other memory preferably would continue to
be updated with input signal data. An example of this is the two stretch
cell embodiments set forth in FIGS. 20 and 21.
The circuitry of FIGS. 16 and 20-21 include two effective memory lines 56
and 57 together with two modification circuits 60 and 61. FIG. 16 utilizes
two separate RAM memories while FIGS. 20 and 21 utilize a single RAM
memory, with two different address spaces allocated respectively to two
stretch cells to allow resetting.
The memory lines 56, 57 are preferably RAM memory circuits. These circuits
provide for a delay necessary for the processing to occur. The length of
these memory circuits is chosen in order to optimize the performance of
the overall circuitry while at the same time preferably minimizing
expense. The selection of length is normally a compromise between
excessive delay versus the ascertainable artifacts which might occur
during resetting at an earlier than appropriate time or resetting at too
high a rate. In general, changing over more quickly reduces information
loss. However, changing over too quickly results in unacceptable
artifacts, so a compromise is chosen. Also, in general, optimization of
performance is determinative of the preferable length for the memories 56,
57, even though theoretically an infinite memory length could be used.
(Too long a delay can produce noticeable gap and/or echo effects. It also
violates EIA/TIA-250-C.) Due to the use of two memory lines, and the
changing therebetween (as later described), memory lengths of from 38 mS
to 149 mS are more than sufficient for a normal audio signal. This memory
is sufficiently long to allow for the smooth crossover between memories
56, 57 while at the same time storing sufficient samples so as to support
the typically experienced maximum time between crossovers. Note that the
length of the memory is a design choice. For example, there are frequent
times of audio silence in a television talk show. If the memory was
sufficiently long as a function of the expected length between silences in
audio content, and the duration of the silence was at least as long as the
section of memory to be traversed by the reset, the memory could be reset
during these times of silence, thus not dumping or repeating an | | |
