 |
Description  |
|
|
BACKGROUND OF THE INVENTION
This invention relates generally to improvements in signal
encoding/decoding methods and apparatus and, more particularly, to a new
and improved digital encoding and decoding system for lower distortion,
higher resolution, and increased dynamic range reproduction of analog
signals while remaining compatible with industry standardized signal
playback apparatus and standards not incorporating the decoding features
of the present invention. In addition, recordings lacking the encoding
process features of the invention are likewise compatible with playback
decoders which do embody the invention, and are provided some enhancement.
Quite often a recording or communications system is standardized and its
format cannot be readily altered without affecting a substantial quantity
of equipment already in existence. Hence, adding information with
supplemental codes may not always be practical unless provisions have been
standardized for such insertions. Unfortunately, modern digital systems
are not very expandable since data bandwidth, resolution, error
correction, synchronization, ancillary data and other "housekeeping"
information essentially occupy the entire digital capacity of the storage
or transmission medium.
However, electronic equipment manufacturers and users of such devices
continue to seek enhanced performance and more features from such
standardized systems. An important example is the need to make a
compatible recording well suited simultaneously for portable, automotive,
television and audiophile markets. Today, many recordings are made for the
most profitable market while other users suffer compromised sonics. The
obvious conflicting performance requirements of different listening
environments and the need for sonic improvement should desirably be
implemented by a new system which is compatible with older systems and
recordings.
Automobile and portable equipment are usually low cost and must operate in
noisy environments. Hence, in such situations, a slightly restricted
dynamic range playback is beneficial. Audiophile systems require utmost
accuracy, dynamic range, and resolution beyond that which is available in
the current standards. Thus, in any new compatible system, as provided by
the present invention, encoded dynamics and slew rate modifications which
achieve lowest distortion and best resolution for the audiophile when
decided, should also provide improved sonics for portable and automotive
playback when not decoded.
Compact Disc pulse code modulation and other digital audio encoding schemes
are good examples of highly developed and standardized systems which push
signal conditioning and digital information limits. Most such digital
systems originally evolved around then practical 2.5 to 3.5 mHz rotary
head video recorder bandwidths. In such standards, the data bits with
error correction and housekeeping entirely fill the available bandwidth.
Accordingly, the need for a "smart" optimization technique, which does not
rely upon increased bandwidth for its implementation, becomes apparent.
By way of background, let us consider a typical digital audio record-play
system, its most frequently encountered components, operation, and
difficulties. In its simplest form, the recorder includes a sampling
switch and an analog to digital converter. The switch breaks the
continuous analog signal into a series of voltage steps, each of which is
converted to number groups or digital words. Digital level meters and
simple communication systems often operate with just these functions in a
single IC chip. Practical high performance record and playback systems
require many added operations to prevent undesired internal and external
analog-digital signal interactions, as well as beats and non-linear
feed-through between digital and analog frequencies. Well-known
technologies to deal with these problems include sharp cut-off or "brick
wall" low-pass filters, fast sample and hold circuits, and high common
mode rejection amplifiers. Unfortunately, although these components and
subsystems solve many problems, they also create others.
Briefly, in typical digital recording systems, low-pass filters ring, and
if of analog construction, have pre-echo, are subject to sudden phase
shifts near band edge, and have capacitors which often cause troublesome
dielectric hysteresis effects. Sample and hold circuits have unpredictable
timing and capture errors for different signal slew rates and also suffer
from capacitor problems. Fast digital signals and the high speed
amplifiers needed to handle them often create and are sensitive to ground
currents which can cause audible strobe-beat effects. Digital reproducing
systems have similar problems, along with spike or glitch generation
caused by digital to analog conversion, and digital filter word length
round off problems. Usually the recorder is designed to have state of the
art performance while that of the reproducer degrades depending on the
economies of "consumer" construction. These and other problems continue to
plague modern high performance digital audio systems.
Unfortunately, such technical difficulties usually create jarring
non-harmonic distortions, typically centered in the most sensitive and
perceptive human hearing range. Often these distortions are caused by the
highest, almost inaudible frequencies contained within the program
material. Taking the ratio of high and low frequency hearing acuity into
account, and the fact that sounds unrelated to the program material stand
out, the presence of even an extraordinarily small amount of these
distortions can be quite objectionable to the listener. Fortunately, often
only very small corrections are needed to minimize some of these
distortions. However, left as is, these distortion errors can combine to
yield the equivalent of 13 to 14 bit performance accuracy from systems
originally designed for 16 bit resolution. In practice, while some feel
the advantages of current digital recordings outweigh the disadvantages of
their distortion errors, many sophisticated listeners and audiophiles are
not so tolerant.
Accordingly, those concerned with the development and use of digital signal
encoding and decoding systems for analog signals have long recognized the
need for a higher quality, lower distortion digital system for
reproduction of such analog signals, which for all practical purposes is
also compatible with existing equipment standards. The present invention
fulfills all of these needs.
SUMMARY OF THE INVENTION
Briefly, and in general terms, the present invention provides new and
improved digital encoding/decoding methods and apparatus for ultra low
distortion reproduction of analog signals which are also compatible with
industry standardized signal playback apparatus not incorporating the
decoding features of the present invention. In addition, signals lacking
the encoding process features of the invention are likewise compatible
with playback decoders which do embody the invention, and are provided
some overall enhancement.
Basically, the present invention is directed to various aspects of an
improved encode/decode system for providing a predetermined balance or
interplay of gain structures, filter characteristics, various slew rate
modifications, and wave synthesis operations to reduce signal distortions
and improve apparent resolution. During the encoding process, an analysis
of the signal to be encoded is made over time and the results of this
analysis are subsequently utilized in the encoding and decoding process to
more accurately reconstruct the original waveform upon playback. This is
accomplished while minimizing the deleterious effects normally encountered
in sampling and converting analog signals to digital signals and
subsequently reconverting the digital signals back to an accurate
simulation of the original analog waveform.
In accordance with the invention, control information developed during the
aforedescribed waveform analysis is concealed within a standard digital
code and this information is subsequently used to dynamically change and
control the reproduction process for best performance. These concealed
control codes trigger appropriate decoding signal reconstruction
compensation complementing the encoding process selected as a result of
the aforementioned signal analysis. Since the control code is silent and
the overall digital information rate is normally fixed, the process can
operate compatibly with existing equipment and industry standards. In
addition, and as previously indicated, signals lacking the encoding
process features of the invention are likewise compatible with playback
decoders which do embody the invention, to provide some beneficial
enhancement.
To achieve higher performance with a fixed information rate, an on-going
trade-off is made between dynamic range, to achieve improved small signal
resolution, and peak level and/or slew rate, to achieve fast signal
response accuracy. These small change and fast change aspects of a signal,
as well as large and small amplitude aspects, each have their own digital
distortion or system compromise mechanisms. Since both large and small
aspects will not occur at the same time, an optimum encoding process or
mix of processes favoring each signal condition can be chosen dynamically,
in accordance with the invention, to achieve an improved signal
reproduction within a fixed digital information rate. A silent or hidden
control code documents these changes from time to time in the signal
encoding process and is used to create the complementary level, slew rate,
filter character, and waveform synthesis necessary to restore the original
signal during the decoding process.
In a presently preferred embodiment of the invention, the encoder system
has much higher resolution and speed than the industry standard or encoded
product, and is set up as an acquisition system with sufficient look
forward and look behind memory to compute the optimum processing of the
signal and its corresponding reconstruction control code. As previously
noted, the processing of the signal is determined based on a consideration
of which trade-offs of resolution, speed, and level are most appropriate
for the signal conditions over time and how the reproducer can best be
programmed to allow the most accurate reproduction of the original analog
signal.
To be inaudible, the computed reconstruction control signal is encoded or
encrypted to a random number sequence which may be inserted continuously
or dynamically when needed into the least significant digital bit or bits.
The processed audio or signal becomes encoded to the remaining bits.
Conventional decoding by a simple digital to analog converter of all bits
of a recording encoded in accordance with the invention, yields a signal
with slightly less dynamic range and only slightly higher background
noise. However, the signal will have lower quantization and slew induced
distortions and, hence, the processed encoded product, when reproduced on
non-decoding standard equipment, will sound equal to or better than an
unencoded product.
A fully decoding player, in accordance with the invention, retrieves the
control signal and uses it to set up, operate and dynamically change a
complementary process to recover the pre-computed high accuracy
information and provide low distortion reproduction of the original analog
signal. Operations to do this include fast peak expansion, averaged low
level gain reductions, selecting complementary interpolation filters,
waveform synthesis, and others. When these are selected according to
ongoing trade-offs, optimum for a particular set of signal conditions, an
apparent increase of bandwidth and resolution occurs.
An improved digital system, in accordance with the invention, uses groups
of dynamically changing pre-determined performance trade-offs made when
signal conditions of the recorded program would create distortion. Since
digital distortions occur at extremes of high level, slew rate, and high
frequencies, on one hand, and with quiet signals and short small
transients on the other hand, a best encode/decode strategy is chosen for
that extreme without the process compromise hurting the opposite aspects
of the program. To achieve this, the program is delayed long enough so
that a most likely distortion mechanism is identified prior to its
emergence from the time delay, thereby allowing a best encoding strategy
and complementary decoding method to be determined and encoded.
Performance is improved because any distortion compromise made occurs for
opposite signal conditions, which are essentially nonexistent at that
time.
In the simplest form of the system, an encoded dynamic range compression
and complementary reproduce expansion will improve performance.
Furthermore, improvements are had by using averaged levels of small
signals independent of their lower frequency and near supersonic frequency
spectral components to control processing providing improved complementary
restored resolution. In a similar manner, the strongest signals receive
processing having DC to maximum bandwidth for instantaneous peak
conditions, which also yields best complementary restoration. Only one
correction need operate at a time and, hence, digital information is
saved, or conversely, more apparent performance is obtained from an
unchanged digital information rate.
In addition, a further reduction of known and predictable digital
distortions occurs by selecting a best low pass filter with the least
compromise for program conditions during encoding and using a
complementary interpolation or low pass filter during reproduction. Also,
other improvements are had from the reduction of known recurrent
distortions, such as transient errors, by synthesizing these components
from lookup table curves of these distortions or missing information, and
then scaling these to the signal at hand.
All of the aforedescribed improvements can also operate with varying
degrees of success in a default or "open loop" mode at the reproducer by
detecting information about the encoded signal and then varying these
processes from the detected signal.
Digital systems typically have a very high signal to noise ratio, but have
a restricted working dynamic range of levels and restricted frequency
response. The improved system of the present invention reduces distortions
and, as such, uses signal character dependent gain changes, filter
optimization, slew rate processing, and waveform reconstruction or
synthesis to do this. The improved system computes, within memory and
process time limits, a continuously changing best compromise strategy of
available processes to give the best signal reconstruction. This obviously
complex task yields a restoration control signal silently encrypted or
noise disguised in a least significant bit code. By comparison, the
reproducer system is simple, since its decoding and complementary signal
restoration can occur with conventional multiplying converters, digital
signal processors and other analog and digital devices similar to or
already used in consumer electronics.
A conventional recording and reproducing digital system appears relatively
simple and potentially accurate for all the data bits encoded. In
practice, however, using a very near to theoretical minimum sampling rate
and the least acceptable number of data bits substantially aggravates
speed and accuracy limitations from even the best state-of-the-art
circuits and components. In this regard, the worst offenders are items
such as filters, sample and hold circuits, analog-to-digital converters,
digital-to-analog converters, and system grounding, timing and various
process interactions and crosstalk.
The aforedescribed practical technological difficulties and their potential
distortions can be greatly minimized by using higher sampling rates and
more data bits than current standards allow. In fact, current
technological capability permits the reduction of cross-talk, time jitter
and other noise interaction problems which, along with digital bandwidth
limitations, prevented the practical implementation of higher data rates
when current digital standards were first envisioned and established. With
today's high speed converters operating much faster with more data bits,
filters can become less severe and the greater difference between highest
audio frequencies and the digital sampling rate then reduces beats,
sideband foldovers, aliasing, as well as loss of small signal information.
The present invention uses these capabilities by employing a high speed
conversion process. The digital information rate, though now much higher,
can be computed, as an ongoing acquisition process, to an "error free"
mathematically filtered lower sampling rate 16 bit code compatible with
current standards. Most decimation oversampling encoders work like this.
However, in addition, the invention anticipates alias, aperture,
interpolation and amplitude resolution distortions from an "ideal"
standard reproducer and computes them during the encoding process for
correction during reproduction. When the full process of the invention is
used, even certain frequencies above the audio range or Nyquist limit of
industry standard equipment can be sent through the system without
creating sub-harmonic or foldover distortions. Hence, a closer to perfect
record/playback system is provided with minimal problems from filters,
converters, and other components or subsystems while remaining compatible
with industry standards.
For a Compact Disc system, "perfect" reproduction to 16 bit industry
standards will have a maximum of 65,536 well defined equally spaced
resolution steps, each about 150 microvolts in amplitude when scaled to
normal professional audio levels (10 volts peak-to-peak maximum). This
number, when stepped consecutively at the industry standard 44.1 kHz
sampling rate, provides a slew of less than 7 volts per second. Faster
rates will skip numbers until, for a 10 kHz triangle segment, only 2.2
sample points remain to define that waveshape as it would be filtered to
its 20 kHz bandwidth. In this regard, more than a 1 giga Hertz sample rate
would be required to include all 65,536 resolution points to create that
wave segment. Fortunately, an ideal interpolation filter will fill in all
of these points provided the 2.2 samples have been timed accurately
enough. To do this to achieve a one half bit RMS averaged accurate sample
of a fast changing signal the sample timing must occur within:
##EQU1##
This sample, accurate in time and amplitude, must be held long enough for
conversion to digital code. Usually, a charge on a capacitor represents
this information. However, most dielectrics and insulators used to
fabricate capacitors have complex losses as well as past history memory
which create a complex delayed voltage change, field re-distribution
errors and leakage. When abrupt changes in level from sample to sample
occur, as they do with sampled high frequency audio signals, these errors
often become much greater than when signal levels don't change. To have
less than a half LSB of RMS averaged error the hold accuracy becomes:
##EQU2##
or about 2.3 u Volt per u sec.
Such performance is well beyond simple applications of most modern
electrical passive components, much less integrated circuits. Obviously,
practical consumer playback equipment will not do better, and the
resulting errors can produce slew rate related transient intermodulation
distortion components, which are among the most audibly objectionable.
Specifically, these result from acquisition time uncertainty or jitter,
slew rate related non-linear switching offsets, various types of
dielectric hysteresis causing previous event related errors, polarity
dependent sample discrepancies, and unpredictable hysteresis within
converters as well as other factors. Thus, practical systems often have
complex signal related errors as high as twenty times more than
theoretical resolution limits of the current 16 bit standard. Hence, a
process providing more sample points per second with the least voltage
change per sample will yield a signal with lower transient intermodulation
distortion.
A second distortion mechanism occurs with very small signal amplitude
changes of about 5 to 20 millivolts represented by digital activities of
less than about 8 bits in a typical 16 bit system. These levels seldom
occur by themselves yet can still be a small but audible part of a larger
low frequency dominated signal. Hence, these small signals can occur
averaged at many different voltage levels or digital numbers of a larger
slow waveform. A practical example of this would be midband hall
reverberation decay and bass sounds combined. The reverberation signal
attenuates and sometimes completely disappears as it becomes chopped or
broken segment parts of the bass waveform. As previously indicated, these
breaks represent the 150 uV resolution limits of a "perfect" 16 bit
| | |
