Re: [ARSCLIST] Analog: A Race Against Time

[seva <seva@xxxxxxxxxxxxxxxx>]

here is an article, in full, which was written by michael quite a number of
years ago, maybe in the 80's (sorry i'm not sure). it covers many other issues
and possible solutions. pardon the long posting, but i think some of you will
find this very helpful and interesting.



Don't Destroy The Archives!
by Michael Gerzon
Abstract

This report explains the value of not destroying original recordings once they
have been transcribed to digital media. It is shown that there is information in
the original master recordings, whether in disc or analogue tape formats, that
cannot be recovered with present day technology which would allow a future
technology to recover improved quality. By destroying original recordings, the
possibility of such future improvements is permanently lost.
1. INTRODUCTION
1a. The problem

This report is prompted by alarming reports from a number of quarters that,
after being archived in a modern digital format, original master recordings in
the form of disc or analogue tape are being destroyed. What this article seeks
to show is that there is in the original recording information which cannot be
recovered with present day technology that will, at a future time with much more
powerful signal recovery, storage and processing technology than is presently
available, be usable to get much improved quality from the original recording
than is currently possible. The very process of transcribing into a current
digital format throws away over 99% of this information, which is thereby not
available to future technologies. The conclusion is: never throw away a master
recording, no matter how low quality it appears to be.
1b. Reasons for destroying masters.

The reason why master recordings are destroyed is a combination of two factors:
1) The storage of master recordings is often expensive, both in terms of cost of
space and manpower, and of the cost of maintaining optimum storage conditions
(temperature, humidity, etc.). 2) There is a widespread belief, which this
article will show to be wholly unfounded, that modern digital transcription
technology is practically perfect, so that it is wrongly believed that the
digital transcription is virtually as good as the original master.

In fact, as we shall see, there is every reason to believe that future
technologies will be able to recover a much more accurate sound from the
original masters than can current transcription technologies, and we shall go
into some detail about what this information in the master is and how it will be
used in the future.
1c. An awful warning

This belief that modern transcription technology is practically perfect is not a
new one, and there are past awful warnings about how historical material can be
lost. By way of a typical example, during the 1960's and 1970's, whenever
historical 78 rpm material was remastered onto analogue tape for commercial
release by some major record companies, the original master parts were then
destroyed, unknown at the time to the remastering engineers involved - who of
course were aware of how compromised the transcription was with available
technology. The result is that it is not now possible to remaster to digital
from the original masters, meaning that this material has the extra drop-out,
modulation noise, flutter and other losses inherent in the analogue tape
technology used but not inherent in the original 78 rpm medium.

Also, as recent revelations about the chemical self-destructing properties of
certain master-tape media used around 10 years ago has made clear, newer storage
media cannot be guaranteed to have good archival storage properties, and this is
very true of most known digital storage media, where the risk of an
unrecoverable loss of digits beyond the powers of error correction is very real,
especially for media such as DAT that are pushing storage technology to its limits.
1d. Alternatives

There are alternatives to destroying the original masters if the cost of storage
is too high. One is to deposit master recordings into national or charitably
funded archive organisations, such as the National Sound Archives in the UK. The
National Sound Archives, in particular, has a rigorous policy for preventing
breaches of copyright of material in its hands.

There is no need to store master recordings that would otherwise be destroyed in
expensive urban locations, especially once transcribed, as access to such
recordings may be wanted only every ten or twenty years (as transcription
technology improves), and a joint archive funded by the recording industry as a
whole in cheap non-urban locations may be an appropriate means of storage both
for reducing costs and for having a higher expertise in storage technology. Such
an organisation can draw on charitable, governmental arts and foundation funding
and can also take advantage in some countries of tax breaks for charitable or
cultural activities.

If all else fails, once a master recording has been transcribed for release, one
can appeal publicly for people or organisations willing to store the original
recordings - an appropriately enthusiastic organisation or even specialist
individual is more likely to take care of such recordings than others - although
it is appreciated that such a proposal may at first sight not integrate well
with the corporate philosophy of larger record companies! If the reason for
wishing to destroy masters is that a digital transcription has been made for
release, then the problem of breach of copyright from a master recording is in
any case no more serious than that from piracy from the released CD
2. INFORMATION ON ANALOGUE TAPE

There is a great deal more information on analogue tape recordings, both
reel-to-reel and even cassette, than is recovered on current playback machines.
Various technical faults can be identified and removed using this extra
information, and we list different aspects that we have been able to identify.
2.1 Wow and Flutter

All analogue tape recordings have wow and flutter due to speed irregularities in
the original recording process, as well as alterations in the physical
dimensions of the tape base and imperfection in the playback machine. It is
surprising to note that most master tape recordings made during the era of AC
bias (i.e. the vast majority of archive tapes) have on them information that
allows almost complete removal of wow and flutter!

The high frequency AC bias, usually at ultrasonic frequencies between 50 and 200
kHz, used in recording was intended to linearise the tape medium at audio
frequencies, but had the side effect of recording the bias frequency itself on
the tape. This bias frequency is subject to a degree of self-erasure, and can
only be played back via a playback head with a very fine gap and via electronics
with a suitably extended frequency response, and accurately adjusted azimuth.
The recorded AC bias component is not usually recovered in conventional playback
machines, and even if it were, would be outside the bandwidth of any digital
audio recorder used to transcribe the signal.

If the AC bias signal is played back and recorded by using appropriate playback
heads and electronics, and a digital recorder with a bandwidth of say 250 kHz or
more, then it provides a reference for the original speed of the tape recorder
used for recording, assuming that the bias oscillator had a stable frequency. By
retiming the signal so as to remove any phase modulation of the ultrasonic AC
bias component, wow and flutter could be removed. The algorithms involved are
fairly complicated, due to the need to interpolate for arbitrary instants of
time, being similar to algorithms used for sampling rate conversion, except that
they also have to work at much higher sampling rates say around 500 kHz.

This alone, plus the need for very high audio-quality grade recording with a
wide bandwidth around 250 kHz, makes such transcription difficult to do
currently, but should become possible in time.

In practice, algorithms to remove wow and flutter will need to be more
complicated, since the phase shift on the recorded AC bias is also dependent on
the recorded signal level, and means of compensating for this effect will also
need to be incorporated. Additionally, stratagems to cope with temporary
drop-outs or loss of the AC bias signal, to keep the playback speed steady when
phase-lock is lost, will have to be incorporated, since it cannot be expected
that there will be no drop-outs in the recovery of the AC bias signal.
2.2 Print-through .

Another problem plaguing analogue tape is print-through, i.e. magnetisation of a
layer of tape by adjacent layers during storage. Again surprisingly, there are
methods of greatly reducing such print-through, which work only if the original
tape is available, in this case making use of the magnetic properties of the
tape. It is generally found that most print-through is due to the existence in
the tape coating of a low-coercivity population of magnetic particles among the
generally much higher coercivity particles that make up most of the coating.
This low-coercivity population is much more easily magnetised (hence its
importance in causing print-through), but is also much more easily erased.

The basic principle of using a weak erase current through the erase head to
erase the low-coercivity print-though particles without much affecting the
high-coercivity particles has long been known, but this has generally not been
used because if the erase current has enough effect to markedly reduce
print-through, it also has enough effect to selectively erase wanted high
frequencies on the tape, thereby damaging the wanted recording through loss of
high frequencies. Moreover, this damage is irreversible, so that if done wrong,
it can never be undone.

However, there is a greatly improved version of the erasure procedure that has
very much lower risk of damaging the wanted recording. This improvement was
originally discovered in the 1950's at the BBC Research Department, but was
generally forgotten until independently rediscovered by the writer.

This is based on an understanding of what parts of the magnetised tape layer are
most affected by print-through and by erasure. An erase head produces the
strongest erasure effect at that "near" side of the magnetic layer nearest the
head, with the erasure effect diminishing on the "far" side of the magnetic
layer. High frequencies (i.e. short wavelengths) are, contrary to popular myth,
recorded throughout the depth of the magnetic layer, but the positive and
negative cycles of the high frequency components at some distance from the
playback surface tend to cancel out, meaning that high frequencies are played
back mainly from the surface nearest the playback head. Therefore an erase head
on the same side of the tape will tend selectively to erase signals at the near
surface, and so will cause a loss of high frequencies on playback. However,
placing an erase head on the opposite side of the tape to erase print-through
will have least effect at the playback surface, and so cause least loss of high
frequencies.

A print-through erasure based on a weak erase current through an erase head on
the "wrong" side of the tape will therefore be much more effective at erasing
print-through while leaving the wanted signal unaffected. Nevertheless,
experiments with "unimportant" recordings on similar tape stock is advisable
before risking damaging master recordings. There is also the danger that
print-through erasure may damage information contained in the ultrasonic
recorded bias signal (see sections 2.1 and 2.6).

Sections 2.4 and 2.5 mention other methods of reducing print-through from a
master tape by future playback technology that will not involve the risks of
erasure.
2.3 Track splitting.

Conventional analogue tape playback uses a single head to read the whole width
of a tape track, but this averages information across the width of the tape
track, thereby losing potentially useful information obtainable from looking at
the way the magnetisation varies across the width of the track.

For archiving purposes, ideally this information should be retained, and this is
possible by using a multitrack head to subdivide the original track into a large
number of subtracks. Even splitting a track into two halves gives a worthwhile
improvement in the recovered information, although ideally splitting a track
into 10 or 30 or 100 subtracks would be much better. One of the practical
problems here is designing a tape head that splits the track without losing
significant amounts of information between the subtracks (i.e. making sure that
the subtracks are truly contiguous) and also of ensuring that each subtrack
individually is recovered with good signal to noise ratio (S/N). Improvements in
playback head technology are making such a goal closer. The availability of the
individual subtracks gives a profile of the magnetisation across the width of
the tape track which may be used to analyse and to compensate for irregularities
in this profile.

The main technical problems with the use of track splittings currently are (i)
the inadequate performance for this application of multitrack heads, (ii) the
amount of extra information that needs to be recorded - an increase of data rate
by a factor of up to 50 or 100, (iii) the formidable amounts of digital signal
processing required to process this large amount of data. We expect the future
to solve all three problems.

We give some examples of improvements possible by such width profile analysis:
2.3.1 Dynamic azimuth correction

Phase differences between the subtracks are evidence of azimuth errors, which
can thereby be measured and compensated by digital time correction of the
individual subtracks before adding them to get the original sound. Additionally,
it is possible to compensate for departures from a straight line profile for the
original record head (e.g. due to head wear or contamination). Such azimuth
correction can be dynamic and fast in operation, so can compensate for example
for the common problem of rapid periodic azimuth variations due for example to
warping of the tape due to pressure on the tape spool.
2.3.2 Dynamic drop-out correction

Tape dropouts in general will not be uniform across the width of a track, and by
comparing levels on the subtracks, and finding that subtrack with the highest
wanted signal level and the relative gains of the other subtracks, it is
possible to determine (if desired even as a function of frequency) the degree of
gain loss individually on each of the subtracks. In the absence of dropout, the
optimum sound would simply be the sum of the subtracks, but a weighted sum
(using a so-called "Weiner filtering" weighting), with overall gain determined
by the level from the highest subtrack level, will allow not only compensation
for dropout, but the best possible S/N in the recovered signal on a moment by
moment basis.
2.3.3 Dynamic lateral head adjustment

If many subtracks are used, the process described in the last paragraph also has
the automatic effect of dynamically compensating for misadjustments of track
placement across the width of the tape, thereby ensuring optimum adjustment of
the effective positioning of the playback head on a moment by moment basis.
2.3.4 Crosstalk reduction

Moreover, by appropriate "deconvolution" of the magnetic profile across the
width of the tape, such effects as "bleed" between adjacent tape tracks can be
reduced, thereby reducing crosstalk both between the tracks and between a track
and the unwanted noise from guardbands between the tracks. In some cases, such
guardbands may contain spurious signals due to imperfect tape erasure or to
magnetisation of the tape before recording.
2.3.5 Modulation noise reduction

A particular benefit of the track splitting approach is that the difference
signal between subtracks contains information about the tape noise without the
signal. This allows recovery not only of the steady background noise spectrum
from the track differences, but also of modulation noise, i.e. variations of
noise with the wanted signal.

One of the problems with existing noise removal systems such as CEDAR or
NO-NOISE is that they rely for effective operation on reasonable estimates of
the noise spectrum, and it is currently not possible to estimate this noise
spectrum on a dynamic basis to optimally remove modulation noise effects.
However, using differences between subtracks, the modulation noise spectrum can
be determined dynamically on a moment-by-moment basis, thereby providing
improved information for noise reduction processes. The subjective effect of
modulation noise is one of the most serious problems with analogue magnetic
tape, and this improvement will help reduce the problem.
2.4 Double-sided playback

In section 2.2 above, we suggested the use of an erase head on the opposite side
of the tape from the playback head. It is also possible to recover additional
information from a master tape by using a separate playback head on both sides
of the tape. A playback head on the "wrong" side of the tape is of course spaced
away from the wanted tape layer, causing a severe loss of high frequencies, but
future head technology may reduce the resulting h.f. S/N penalties, at least up
to middle audio frequencies.

The benefit of such double-sided playback, if the two head outputs are
synchronised in time (this itself may require the use of DSP) is that the two
heads are respectively more sensitive to magnetisation on their side of the
thickness of the magnetic layer, thereby giving some indication of the profile
of magnetisation across the thickness of the layer. This helps distinguish the
wanted signal (which has strongest magnetisation near the surface) and
print-through from the back of the tape, and an appropriate linear combination
of the two head outputs can cancel out this component of print-through.
2.5 Reading non-transverse magnetisation

Conventional tape playback heads only "read" magnetisation in one direction
relative to the motion of the tape, essentially a transversal component along
tape surface. However, if in addition, magnetisation perpendicular to the tape
surface is read, this provides additional means of separating the wanted signal
from spurious signals, and also of separating out various modulation noise and
distortion components. In particular, it is known that print-through from each
side of the tape has a distinctive magnetisation angle relative to the tape
surface, and again this can be used to reduce print-through by taking
appropriate linear combinations of the transversal and the perpendicular
magnetisation component. The required head technologies may, for example, be
based on the Hall effect..
2.6 Analysing the bias signal

The AC bias signal recorded on the tape (see section 2.1) may contain extra
information allowing playback with reduced distortion. The recorded AC bias
frequency, and its harmonics, will be modulated, both in amplitude and phase by
information related to the original signal. With further understanding of the
distortion mechanisms in AC biasing, it may prove possible to use this
additional information, along with nonlinear signal processing, to recover a
less distorted version of the wanted signal than can be recovered by direct
playback on its own. As in section 2.1, this requires a much wider bandwidth
than the normal audio band.

Additionally, aliasing distortion due to sidebands of the modulated bias
frequency overlapping the audio baseband, are a significant cause of quality
loss in recordings using bias frequencies below 100 kHz, and recovery of the
bias signal may allow such aliasing to be computed and removed.
2.7 Combined methods

The above methods can be combined, although the practical problems of devising a
double-sided head divided into say 100 subtracks, each responsive to transversal
and perpendicular magnetisation, responding up to say 500 kHz with audio-grade
quality and good S/N are formidable by present-day standards!

Even if these technical problems are solved, the data rate involved will be
thousands of times larger than for conventional digital audio, requiring the use
of recorders with data rates comparable to or exceeding that of a digital HDTV
recorder for data storage. The digital signal processing involved to use this
information effectively is also formidable - many thousands of times more
complex than what is feasible economically at the moment.

At present, only very partial implementations are feasible, although we urge at
least the splitting of tracks into two subtracks, especially for mono material,
so that the difference signal (or, more technically, the smaller of the
principal eigenvalues of the 2-channel spectral matrix) can be used to determine
the spectrum of the noise signal, and so that dynamic azimuth correction becomes
possible.
3. GROOVED RECORDS

The problem of recovering all information from recordings mastered, for example,
in a 78 rpm grooved format is in its own way as formidable as the tape case
above, although as we shall note, some past signal recovery technologies have
already used information thrown away by digital transcription!

We here consider a number of aspects that are difficult or impossible to recover
from a digital transcription.
3.1 Impulse noise

78 rpm records typically have serious problems from what is often termed
"scratch" noise, which is a noise consisting of a large number of added
impulses. The first process to detect and suppress such impulses was implemented
at EMI in the late 1940's, and various analogue processes were subsequently
developed by the present writer with Peter Craven in the early 1970's and
commercially by Packburn, and currently, digital processes using more
sophisticated predictor-type impulse detection and interpolation type
replacement have been devised by CEDAR and by NO-NOISE.
3.1a Ultrasonic components

However, such processes inevitably damage the integrity of the original signal,
resulting in a less clean sound, and it is an important aim to minimise such
damage. In separating the impulses from the music signal, it is found important
to preserve a very wide bandwidth, preferably more than 40 kHz, since the
duration of an impulse is inversely proportional to the bandwidth, so that short
impulses will only occur for wide bandwidths. Also, there is a relative lack of
music-related signal energy above say 15 kHz, so that the location of impulses
can much more easily be determined in the ultrasonic region above 15 kHz -
indeed the EMI process in the 1940's remarkably used the ultrasonic components
to localise the impulses in time.

However, current digital recording technology, being bandlimited to 20 kHz,
throws away this ultrasonic information and smears out the duration of impulses,
giving a signal in which separating out the music from the impulses is much
harder to do well. Ideally, a bandwidth in the 50 to 100 kHz region would
capture information for removing impulses much better.
3.1b Stereo pick-up

A second piece of information that allows improved separation of impulses from
the wanted signal, used both by the author's early work with Craven, and by
Packburn, used the fact that record grooves have two walls, and the signals for
each wall can be separately recovered using a stereo pick-up. It is generally
found that most impulses occur either on one groove wall or the other, so that
on mono records. the impulse alone can be recovered by taking the difference of
the two wall signals. In practice, this "vertical" difference signal is
contaminated by various distortions, but the use of two channels of recovered
information nevertheless allows much more reliable detection of impulses.

Additionally, the fact that most impulses occur on only one groove wall means
that it is possible to detect which groove wall and to switch recovery of the
wanted signal to the other groove wall - a technique known as groove-wall
switching. There are various problems with this technique, ranging from residual
cross-talk of impulse noise to the effects of stylus tracing distortion on the
individual groove wall signals.
3.1c Combined information

Nevertheless, if from a mono 78 rpm record one can recover both groove walls
separately with a bandwidth of 50 kHz or more, the resulting information can be
processed much more reliably to remove impulse noise with minimal distortion of
the wanted signal. This relies on using specialised digital recording
technology, as the standard stereo digital formats have inadequate bandwidth.
The DSP processing to make optimum use of the extra information has not yet been
developed, but could be near-future technology were there to be a demand for it
(e.g. by a major record company commissioning the development of such a system
from one of the existing specialist suppliers such as CEDAR or NO-NOISE).
3.2 Tracing Distortion

The ideal playback stylus would recover information from a point or line contact
with each groove wall, but actual styli have a finite radius of contact in the
direction of travel of the groove. This radius of contact causes what is termed
"tracing distortion", whose theory was well developed by Shiga, Cooper and
others in the 1960's. As first noted in 1975 by the writer, tracing distortion
has the effect not only of adding nonlinear distortion to the wanted signal, but
it also has the effect of prolonging the duration of unwanted noise impulses,
thereby increasing impulse noise. Therefore, optimum recovery of the signal to
reduce impulse noise requires the use of a stylus with as small a contact minor
radius as possible.

However, there is a practical limit to the size of minor radius that can be used
with feasible playback styli, and electronic correction of tracing distortion,
as in Cooper's elegant "skew sampling" technology of the 1960's, is the only
feasible way of reducing tracing distortion further. However, such correction
technology itself imposes three demanding requirements on information recovery:
1) that the two groove walls be recovered separately, since each has to be
processed independently to reduce tracing distortion, (2) that the recovered
bandwidth be much wider than the audio bandwidth, as well as being phase-linear,
since bandlimiting before tracing distortion correction itself introduces
errors, and 3) That an accurate record be taken not only of the stylus radius,
but also of the groove velocity at each point in the playback, since without a
knowledge of both these parameters, tracing distortion correction cannot be done.

Existing digital transcription throws away all this information, making both
tracing distortion reduction and proper impulse length reduction impossible.
Future technology will allow both proper recovery and storage of this
information and signal processing to recover a clean tracing-distortion
corrected signal of wide bandwidth for each groove wall. Such signals form a
better basis for the impulse-reduction processing methods described in section
3.1 above than the "raw" outputs of a pick-up cartridge.
3.3 Groove wall profile

There is far more information that in principle can be recovered from an
original disc or metal parts. Ideally, one would aim not simply to recover a
kind of "average" of the signal on each groove wall over the contact area of a
stylus, but to record separately the signal at each different height up the
groove wall, so as to recover the complete profile of the cross section of the
groove at each point along the groove.

Such a process divides the signal on each groove wall into a large number of
parallel "subtracks", one for each different height, analogous to those
suggested above for tape playback. Analysis of the differences between these
subtracks can be used not only to analyse noise as in section 2.3.5, but also to
analyse at which heights the effects of record wear and noises from the "land"
or the groove bottom are likely to be least serious. By this means, one would be
able in effect to vary stylus height and contact profile adaptively by signal
processing to optimise playback moment by moment.

As in the tape track splitting case, both the storage and subsequent signal
processing demands of a groove wall profile approach are extremely high, and
generally this technology is still in the future.

The actual playback technology required also does not yet exist. One might
consider using an optical playback technology, but this has numerous problems
both due to the size of the wavelength of light (too large!) to the fact that
optical playback, unlike mechanical styli, does not push unwanted contamination
out of the way. Probably the best way to recover profile information is to play
back with a number of styli with different sizes, and then to use DSP to
synchronise the recordings, to remove tracing distortion from each, and then to
process the signals to recover a wall profile.

One of the uses of groove profile information is to correct for different
effective angles of cutting stylus rake, including dynamic correction similar to
the dynamic azimuth correction described above for tape in section 2.3.1.
3.4 Other parameters

Besides the obvious geometric parameters for the groove surface, both metal
parts and actual records have numerous other mechanical, chemical and physical
parameters such as stress, elasticity, coefficient of friction and so forth.
Each of these may provide additional information allowing deduction of
distortions in the reproduced sound and permitting correction of the
distortions. It is difficult to predict what parameters may be found useful in
future, but they can clearly only be recovered from the original records or masters.

Aspects of elasticity may be recovered as signals by tracking the same recording
with the same stylus at different stylus pressures, and synchronising the
different recordings digitally. The difference signal between the recordings
will (apart from noise signals due to contamination) contain information about
the physical properties of the record.
3.5 Other information

In addition to measuring information about the groove walls and the playback
velocity, other information that may allow improved signal recovery includes: 1)
playback of the "land" between the grooves, since this may correlate with noises
in the groove itself 2) playback of the bottom of the groove, for similar
reasons, and 3) measurement of the precise physical relationship between
adjacent grooves (including distance and timing relationships), both to help
reduce pre- and post-echo effects (the disc equivalent of print-though) and to
detect periodic disturbances that may be filtered out by appropriate long-term
comb-filter averaging.
3.6 Information content

The information recoverable from original disc masters involves data rates of
the order of 50 times greater than that of a conventional digital audio channel,
due to the extra audio bandwidth, use of stereo channels, and the use of extra
subchannels to record groove profile and (where relevant) elasticity
information. As in the analogue tape case, this requires more powerful recording
media (here a digital video recorder would have an adequate data rate), much
more DSP power than is currently used to process the data to recover a signal,
and finally the use of multiple playback of the source master with careful
measurement of all relevant physical parameters.

While not technologically as extreme as the requirements for the analogue tape
case, it will be not less than several more years before the appropriate
technology is fully developed, and there is always the possibility of new
unexpected data from section 3.4 above that may require new technology to be
developed to improve wanted-signal recovery further.
4. COPIES FROM MASTERS

While the above processing possibilities are most apt to masters of the
recordings, similar techniques could be applied to first generation copies when
these are all that are available. While the loss of information in the original
copying process cannot be undone, at least one will in future be able to reduce
the effect of imperfections in the medium onto which the copy is made. The
improvements that this will give will still be very worthwhile from a quality
viewpoint.

This applies to copies from the master, whether on analogue tape or in the form
of parts or discs cut from an original master tape. Often, the first release of
an LP, especially in the 'home' country of the recording, were cut from master
tapes, so that the original parts or mint unplayed copies of such releases
should be considered an archive resource.

In many cases, either master tapes have already been lost or mislaid or they
have significantly deteriorated - for example acquiring drop-outs. In this case,
direct transcriptions made early on onto disc may be the best available source.

It should also be noted that poor or incorrect tape box labelling means that
often one cannot be sure that tapes labelled as masters or copies are what they
claim to be - it is not unknown for a so-called "copy" to be a master - and only
attentive listening and investigations can decide the issue.

For these reasons, the greatest care should be taken to avoid disposing of "copy
tapes" until one is absolutely sure that they do not provide a useful access to
the original recording.
5. DIGITAL CONVERTER IMPERFECTIONS

Besides the problems of recovering all relevant information from the original
master, there is also the problem of imperfections in the transcription medium.
The naivete of the early days of digital audio when it was thought to be
essentially "perfect" has recently been replaced by an understanding of many of
the mechanisms by which the ears hear faults that, according to traditional
audio measurements were negligibly small. This work, based on the researches of
Louis Fielder at Dolby Labs and Bob Stuart of Meridian Audio in modelling of
masking in auditory perception, shows that there are still significant audible
faults especially in available analogue-to-digital converters (ADCs). This
should not be a surprise, since such faults are heard not only by many audio
professionals, but even by many lay listeners who hear a distinctive loss in
digital transcriptions.

One of the known sources of audible faults is "jitter", i.e. tiny variations in
the timing of the digits. All digital recording media and signal interconnects
introduce audibly significant amounts of jitter, and most DACs currently do not
have adequate de-jittering, resulting in each digital player having its own
"sound". However, the problems of designing DACs to remove jitter have recently
been solved, and the choice of digital transcription medium itself should no
longer be a serious quality problem provided that there is no loss of corrected
digits.

However, jitter in the original ADC cannot subsequently be removed, and neither
can faults due to non-linearities, modulation noise and limit cycles in the ADC.
Currently, the quality of most ADCs used to transcribe material in digital form
still leave a great deal to be desired, and these faults are clearly audible
even on old archive disc and tape material. A great deal of listening is still
required before the ADCs used are selected, and they should also be used in a
way that minimises jitter effects in the conversion process itself.
6. CONCLUSIONS

This report has shown that current digital transcription technology cannot yet
recover most of the information in master disc or analogue tape recordings, and
that future technologies will allow the recovery of extra information from the
original master that cannot be recovered from a digital transcription. We have
described improvements that we expect will become possible with future
transcription technologies, although some of these may still be some time away
due to limitations in current technologies.

It was also noted that even the sound quality of current transcription
technology still leaves quite a lot to be desired.

It is therefore recommended in the strongest terms that original master
recordings, or the closest available copies to these, should be preserved, since
the digital transcriptions are no substitute for the potential quality
recoverable in the future from the masters. Some of these quality gains may be
very substantial - e.g. virtual removal of print-through, modulation noise and
wow and flutter from analogue master tapes.

Insofar as a digital transcription is required for archive purposes (e.g. for
safety back-up or because of physical deterioration of the master), it is
recommended that each mono track of the original be split into two subtracks as
described above for disc (two groove walls) or tape (dividing the playback track
into two), so that the extra information can be used by future signal processing
should the original master be lost. It is also recommended that such safety
copies should use the best available ADCs in their preparation, since currently,
they are a quality bottleneck.