THE SPECTRAL RECORDING PROCESS 2413 (C-6)

Ray Dolby
Dolby Laboratories Inc.
San Francisco, California
London, England

Presented at ^ u DIO
the 81st Convention
1986 November 12-16
Los Angeles, California
Thispreprinthasbeenreproducedfrom theauthor'sadvance
manuscript,withoutediting,correctionsor consideration
by
theReviewBoard.TheAES takesno responsibility for the
contents.

Additionalpreprintsmayha obtainedbysendingrequest
and remittanceto the AudioEngineeringSociety,60 East
42ndStreet,New York,New York10165USA.

All rightsreserved.Reproductionof thispreprint,or any

portionthereof,isnot permittedwithoutdirectpermission
from theJournalof theAudioEngineeringSociety.

AN AUDIO ENGINEERING SOCIETY PREPRINT

 The Spectral Recording Process

Ray Dolby

DolbyLaboratorles Inc., San Francisco and London

Abstract

A complementary audlo signal encodingand decoding

format, designated spectral recording (SR), for use in
professional magnetic recording and similar applications ls
described. The processing algorithm is highly responsive to the
spectral properties of the signal; a further characteristic used
during encodingde-emphasizes high level signal components in
the frequency regions usually subject to channel overload. The
processresults in a significant reduction of noise and
distortion arising in the channel.

0 INTRODUCTION

In 1980, some 14 years after the introduction of A-type noise reduction

( I ), the author beganwork on the development of the next generation system
for general purposeprofessional recording and transmission. A configuration
that would employ the A-type characteristics as part of the new system, with
switchable compatibility, was considered initially. However, this would not
take full advantageof the new technology embodiedin the C-type system (4),
nor would it readily allow the incorporation of some further new concepts.
Therefore, the particular parameters of the A-type system were abandonedas
a starting point for the new development. However, the basic principles,
which appear to be as valid as when they were first introduced, were retained:
the use of a maln signal path without any dynamicprocessing to pass hlgh
level signals, coupledwith a Iow-level side chain compressor to provide
dynamic action.

The design goals of the new system were set high. The new technology,
called spectral recording (SR), should provide master recordings of the very
highest quality, especially with regard to signal purity, and yet the system
should be practical and economical for routine applications. It should be
suitable for easy and trouble-free use in a wide variety of professional
recording and transmission environments. Certain new techniques, to be
described, provide the required signal quality and practicality but result tn
circuit complexity. Reliance has been placed on improved circuit
implementation and manufacturing techniques to overcomethe problems of
complexity and to ensureeconomical production of the new system.

I BRIEF OUTLINE

Thegoal of the spectral recording process is to modify the various

componentsof the incoming signal in such a way as to load an imperfect
recording or transmlss$onmedium in the most ratlonal way. Generally, high
level signal components at both endsof the spectrum are attenuated, whereby
a better match with the overload characteristic of the medium Is provided. At
the same time, Iow level componentsof the signal are ampi'Ifled substantially,
in a highly frequency selective way. These effects are reversed during
reproduction, restoring the original signal. Theresult Is a significant
reduction of distortion and noise, both in the absenceand presence of signals.

The process has a number of'layout and operating characteristics in

commonwith the A-type ( I ), B-type (2, 3), and C-type (4) noise reduction
systems. The SEprocess takes these developmentsconsiderably further in the
samegeneral direction. Regardinggeneral principles, reference should be
madeto the technical papers on these systems, the C-type paper (Ref. 4) being
particularly relevant.

Referring to Fig. I, which will later be described in detail, a main signal

path is primarily responsible for conveyinghigh level signals. A side chain
signal with the SRprocess characteristic is additively combined with the
main signal in the encodingmode and subtractively tn the decodingmode,
whereby an overall complementary action is obtained.

The SRstage layout resembles that of the C-type system, except that
three levels or stages of action staggering are used: high-level, mid-level, and
low-level (HLS,MLS,LL$). There are various advantagesarising from the use
of multi-level stages, including accuracy and reproducibility, iow distortion,
low overshoot, and action compoundingfor good spectral discrimination. For
the high-level and mid-level stages both high frequency and Iow frequency
circuits are used, wtth a crossover frequency of 800 Hz. The Iow-level stage
is high frequency only, with an 800 Hz high pass characteristic.

Eachstage abovehas a Iow-level gain of somewhat over 8 dB, whereby a

total dynamic effect of about 16 dB is obtained at Iow frequencies, 24 dB at
high frequencies. A further dynamic action of about I dB takes place above the
reference level.

The spectral skewing network has the same purpose and function as in the
C-type system, except that a spectral skewing action is provided at Iow
frequencies as well. Thespectral skewing networks de-sensitize the $R

- 2 -
process to the Influence of signal componentsat the extreme endsof the audio
frequency band. This effect is particularly helpful if the recording or
transmission system has an uncertain frequency response In these regions.
Thenetworks are also important in attenuating sub-sonic and super-sonic
Interferences of all kinds. The spectral skewing action is compensatedin the
decoder,resulting in an overall flat frequency response.

Both high frequency and Iow frequency antlsaturatlon networks are

provided in the main signal path, again operating in substantially the same
way as in the C-type system. There is an effective compoundingof the
antlsaturatlon effects producedby the antisaturation networks and the
spectral skewing networks. In this way the 5R process achieves a significant
increase in high and Iow frequency headroom.

2 6ENERAL PRINCIPLES

2.1 Least Treatment Principle

A design philosphy used In the development of the new system Is that the
best treatment of the signal is the least treatment. The design goal for the
encoder is to provide a predetermined, fixed gain for all sub-threshold signal
components. If a large signal componentappearsat a particular frequency or
frequencies, then the galn should be reducedat those frequencies only, In
accordancewith a predetermined compression law so that It Is possible to
restore the signal during decoding. In Other words, the compressor tries to
keep all signal components fully boosted at all times; when the boosting must
be cut back at a particular frequency the effect should not be extended to
Iow-level signal components at other frequencies.

The audible effect of this type of compression is that the signal appears
to be enhancedand brighter but without any apparent dynamic compression
effects (the ear detects dynamic action primarily by the effect of a gain
changedue to a signal componentat one frequency on a signal componentat
some other frequency, somewhat removed). If the ear cannot detect dynamic
effects in the compressedsignal then a) it is unlikely that noise modulation
effects will be evident in the decodedsignal, and b) it is unlikely that signal
modulation effects will be evident in the decodedsignal if there should be a
galn or frequency response error in the recording or transmission channel.

In the SRprocess two new methods are used that greatly reduce the
circuitry required to achieve the design goal of a full spectrally responsive
system. In particular, both fixed and sliding bandsaceused in a unique
combination, called action substitution, that draws on the best features of
both types of circuits. A further technique, called modulation control, greatly
improves the performance of both the fixed and sliding bands in resisting any

- 3 -
modulation of signal componentsunless necessary.

The use of the new methods reduces the basic encoderto two frequency
bandsonly (high frequency and Iow frequency), eachwith a fixed band circuit
and a sliding bandcircuit (this combination being referred to as a stage).
Whenthe three-level action-staggering layout is taken into account, five fixed
bandsand five sliding bandsare employed in the spectral recording process.

2.2 Action Substitution

A new type of compression and expansion action that is highly responsive

to spectral changescan be achieved by superposing or overlaying the individual
characteristics of different types of dynamlc action circuits. Onecircuit may
provide a quiescent characteristic or defining umbrella; a further
characteristic is hidden until signal componentsappear that cause the hidden
characteristic to be revealed and become active.

For discussion purposes let the gains in a compressor system be arranged

such that sub-threshold signals pass without attenuation. That is, the
maximum possible action ls that of providing a gain of unity. Somehow to
achieve this gain over as broad a range of frequencies as possible, in the
presence of hlgher level (dominant) signals, is the task of the system.

Thus, in a superposedaction compressor circuit, represented in Fig. 2, a

signalisfedintoa first compressor circuit.Theoutputfromthiscircuit
represents thecompleted partofthetotal potential action.Theuncompleted
partistherefore theinputsignal minusthecompleted part;thisisso derived
andfedintothenextcompressor circuit,
whichhassome different
characteristic. Theoutputofthesecondcircuit isthenaddedtothatofthe
first,
augmenting theactionofthefirst. Inanextremecondition, inwhichthe
outputof thefirst circuit
may benegligibleataparticular frequency, the
actionofthesecondcircuit iseffectively substituted forthatofthefirst.

Theoperation
oftheactionsubstitution
compressor
canbecharacterized
directly
fromtheabovedescription.
Withan Input
signal
V landan output
signalVo,a first
compressor
transfer
function
F I,anda secondcompressor
transfer
function
F2,we have:

Vo = Vi[Fi(V I ) + F2(V!) - Fi(Vi)F2(Vi)] eq.(I)

This equation shows that the overall transfer function is the sum of the
individual transfer functions minus their product. In other words, to the
extent that the transfer functions may overlap, a factor Is subtracted from the

-4 -
sum ofthetransfer
functions,

Theabovetypeofactioncanbe achieved withvarious circuittopologies.

the one used in the present implementation of the SRsystem being shown in
Fig. 3. In this arrangement, the compressor circuits are arranged in a stack.
Both circuits are fed in parallel and the output is taken from the top circuit,
which is configured as a three terminal network with terminals (a), (b), and
(c). The output of the first circuit is fed to, the reference termlnal (c) of the
top circuit. It can be shown that the signal components at the output terminal
(b) are those specified by eq.( 1).

The usefulness of the superposltlon technique can be appreciated by

consideration of Figs. 4(a) and 4(b). The advantagesof fixed band compressor
circuits, Fig, 4(a), arise from the fact that all signal frequencies within the
bandare treated equally, In contrast with sliding band action, Fig. 4(b). Thus
the appearanceof a dominant signal componentactuating the compressor
results in a loss of noise reduction effect that manifests itself in a uniform
mannerthroughout the band, 2 dB in the example shown. The loss is not
concentrated in any particular frequency region as It is tn sliding band
circuits; note the 5 db loss shown in the example of Fig. 4(b). Themain
significance of this is that If the recorder or transmission channel has an
error in gain and/or frequency response, there is no amplification of the error
at other, non-dominant, signal frequencies; in sliding bandcircuits the
amplification may be significant (the mid-band modulation effect, discussed
In Ref. 4).

In contrast, the advantagesof sliding band compressionand expansion

circuits derive from the fact that all signal frequencies are DgZ,treated
equally. In particular, compression, expansion, and noise reduction action are
well maintained abovethe frequency of the dominant S_gnalcomponent in high
frequency circuits, and below the frequency of the dominant signal component
in Iow frequency circuits. This action maintenance effect, except on a one to
one basis, is absent In fixed band circuits.

Clearly It would be desirable to have the benefit of fixed bandoperation on

the stop-band side of the dominant signal frequency and sliding band operation
on the pass-band side. The action substitution technique provides this useful
combination. In Fig. 4(c) the response of an action substitution compressor to
the signal conditions of the two previous figures is shown. As is seen, the
output is primarily from the fixed bandfor frequencies up to the dominant
signal componentand from the sliding bandabovethat frequency. Conversely,
for a Iow frequency stage the output is from the fixed band for frequencies
down to a Iow frequency dominant componentand from the sliding bandbelow
that frequency. This cooperative effect is particularly useful in the level
region from the circuit threshold up to some 20 dB thereabn,,,

- 5 -
 In the 5Rprocess, actlon subsitution operation is used in both the high and
Iow frequency circuits. Thus both fixed band andsliding band dynamic actions
are used In eachof the five stages, a total of ten compressor circuits. While
there is an effective interaction of the fixed and sliding bands in any
particular stage, all of the stages operate independently. Dependingon the
levels and spectral conditions in each stage, fixed bandoperation is used
whenever it providesbest performance; sliding bandoperation Is substituted
whenever it has an advantage. The substitution is effective on a continuous
and frequency by frequency basis.

Eventhough the frequency division of the stages is nominally 800 Hz, the
use of what are effectively single-pole banddefining filters results tn a
significant overlap region between the high and iow frequency stages; the high
frequency stages extend their effects down to about 200 Hz, the Iow frequency
stages extending their effects up to about 3 kHz. This overlap, together with
the use of action substitution, contributes to the achievement of a very good
spectral tracking effect under all frequency and level conditions. The
practical significance is that an excel lent noise reduction effect is obtained
in the presenceof signals, and that the system has a remarkable tolerance to
gain and frequency response errors in the signal channel.

A further aspect of action substitution relates to the transient recovery

characteristics of the system. A fixed band compressor circuit has a recovery
time that is essentially independentof frequency, at least in the pass band. A
sliding band circuit has a fast recovery time for non-dominant signals at the
pass band end of the spectrum, and a slow recovery time for non-dominant
signals at the stop band end of the spectrum. Thechoice of integrator
recovery times is therefore a matter of compromise between this recovery
time situation and the amount of steady state and modulation distortion
obtained. The compromise is mademuch easier, however, by the use of the
action substitution technique. In particular, the fixed band provides a definite
and rapid recovery time for the overall system, so that the sliding band can
employ longer time constants than would otherwise be desirable. This results
both In Iow modulation distortion and a fast recovery time.

Thus the action substitution technique provides the advantagesof fixed

and sliding band circults while avoiding their disadvantages. In other words,
there ls a significantly improved adherenceto the ideal of least signal
treatment; tn the level region somewhat above the circuit threshold the signal
more closely approaches fully boosted conditions in the encodingmode, with a
consequently improved noise reduction effect in the decoding mode. For
signals at higher levels the technique of modulation control, described below,
is employed.

-6 -
 2.3 ModulationControl

In the A-type, B-type, and C-type systems the signal from the side chain
is highly Ilmited under high-level signal conditions. This high degree of
limiting, beginning at a Iow-level threshold, is responsible for the low
distortion, Iow overshoot, and Iow modulation distortion which characterize
these systems.

A closer examination shows that it Is unnecessary to utilize such a Iow

threshold and such a strong Ilmltlng characteristic under certain signal
conditions. In particular, whenever the side chain signal departs from an
in-phase condition with respect to the maln path signal, then the threshold can
be raised. Moreover, after an appropriate degreeof limiting has taken place at
a given frequency (in order to create the desired overall compression law),
then it is unnecessary to continue the limiting when the signal level rises
even further. Rather, the level of the side chain signal can be allowed to rise
as a function of a further increase in signal level, whereby it stabilizes at
some significant fraction of the main path signal level.

In the fixed band portions of the spectral recording process the above
arrangement results in conventional performance in the pass-band(in-phase)
frequency region. However, in the stop-band region the modulation control
schemecauses the limiting threshold to rise and the degreeof limiting to be
reduced. The possibility of doing this can be appreciated by consideration of
the phasor diagrams of the two conditions shown In Figs. 5(a) and 5(b). In the
pass-band (In-phase) condition the slde chain signal and the maln path signal
add directly; therefore a relatively Iow threshold must be maintained at all
pass-band frequencies (Fig. 5(a)). However, in the stoP-band the effective
amplitude contribution of the side chain signal may be mlnlmal due to the
phasedifference between it and the main path signal; becauseof this it is
possible to raise the threshold significantly and to reducethe limiting
strength once the desired amount of attenuation has beenobtained at a given
frequency (Fig. 5(b)). Theresult is that large signals in the stop-band do not
cause signal modulation in the pass-band and consequently create an
impairment of the noise reduction effect achieved during decoding.

Similar considerations apply in the SRsliding bandcircuits. By way of

introduction, in the B-type and C-type sliding bandcircuits a variable filter
follows a fixed filter, which has proved to be an efficient and reproducible
arrangement. At frequencies outside the pass-band a pure two-pole filter
would result in overall amplitude subtraction from the main path signal
becauseof the large phase angles created. Therefore, the type of filter which
has beenemployed is only quasi-two-pole (a single pole fixed filter plus a
variable shelf characteristic).
 The same type of arrangement is used in the spectral recording process,
with a one-octave difference (in the stop-band direction) between the variable
filter turnover frequency (under quiescent conditions) and the fixed fllter
cutoff frequency. Abovethe threshold at a particular frequency the variable
filter slides to the turnover frequency neededto create the overall (main path
plus side chain slgnal) compression law. As the input level rises, and oncean
overall gain of about unity is obtained - when the variable filter cutoff
frequency is about two to three octaves abovethe dominant signal frequency -
there is no reason for further sliding of the variable filter. At this point the
modulation control arrangement counteracts further sliding of the filter; as
with the fixed bandcircuits, this arrangement prevents unnecessary
modulation of the signal.

Theaboveeffects inboththefixedandsliding bandsarecreated by

circuits called modulation control circuits. Suitably filtered or frequency
weighted signalsfrom the main signal path are rectified,and insome cases
smoothed, andarefedinopposition tothecontrol signals generated bythethe
control circuits of the various stages. The result at higher signal levels,
relatively (beginning at about 20 dB above the threshold of the relevant
compressor circuit), is to tend to create a balance or equilibrium between the
compressor circuit control signals and the modulation control signals. Under
these conditions there is a significantly reduced gain reduction or sliding of
the variable filters as a function of increasing input signal levels.

Figs. 6(al and 6(b) illustrate the action of modulation control with a high
frequency fixed band compressor circuit. The circuit has a low level gain of
about 8 dB andan 800 Hz high pass characteristic. Fig. 6(al shows the
response of the circuit in the absence of modulation control. Ideally there
should be no attenuation in response to a I00 Hz signal because the overall
shape of the envelope is such that there is negligible signal boosting at I00
Hz. Nevertheless, with a conventional compressor circuit as shown here, when
the I O0 Hz signal increases in level there is a reduction of Iow level signal
boosting over the whole frequency band. The unnecessary attenuation has two
effects: (al substantial noise reduction action is lost during expansion,and (b)
when the amplitude of the 100 Hz signal varies it can modulate low level
signal componentsat higher frequencies, resulting in possible incorrect
restoration of the signal by the expander if the recording channel has an
irregular frequency response in the vicinity of 1O0 Hz.

Fig. 6(b) shows the operation of the same circuit with modulation control.
A greatlyreduced attenuation occurs when the IO0 Hz signal isvaried over the
same range of levels as in Fig.6(al. Thus a significantimmunity to strong
signals inthestopbandfrequency regionisachieved, theeffectdecreasing as
the dominant signal frequency approaches the pass band frequency region of
the circuit.
 Referring to Fig. 7(a), the operation of a sliding bandcircuit under
comparable conditions Is shown. As with the fixed bandcircuit, Ideally there
should be no sliding In response to a strong I00 Hz signal. Nonetheless, as the
I00 Hz signal increases in level the band slides upwards. As with the fixed
band circuit, the unnecessarysliding results In a loss of noise reduction
action and the modulation of signals at higher frequencies when the sliding
band varies under the control control of the I O0 Hz signal.

Fig. 7(b) shows the operation of the same circuit with modulation control.
Hinimal sliding occurs when the 100 Hz signal is varied over the same range of
levels as in Fig. 7(a). Thus the sliding band compressor is also made
essentially Immune to strong signals outside its pass band.

Theeffect of modulation control is further Illustrated by Figs. 7(c) and

7(d), except that the frequency of the dominant signal is changedto 800 Hz, a
frequency within the pass band of the circuit. Ideally, sliding is required to go
only so far as not to boost the 800 Hz signal abovethe 0 dB reference level.
Thus, in the Fig. 7(c) response, without modulation control, the sliding
producedby the 800 Hz signal at levels above-lO dB is excessive. Fig. 7(d)
shows the response of the circuit with modulation control: sliding at and
abovethe 0 dB level is greatly reduced. The effect is progressively reduced
for low signal levels but is observable to some extent at the -I0 dB level.

The useof modulation control techniques also has advantagesunder

transient conditions, both from their use in the steady state circuits and also
becauseof their use in the overshoot suppressions circuits. Modulation
control generally prevents any further fixed band attenuation or sliding of the
variable filter than is required to respond to a given signal situation.
Therefore, a) signal modulation is reduced, b) the SRprocess is rendered very
tolerant of channel errors, c) subsequentnoise modulation during decoding is
reduced, and d) recovery from transient signal conditions is faster. The
electronic reality in both the steady state circuits and transient control
circuits is that the integrator capacitors are preventedfrom charging to
voltages as high as they normally would in the absenceof modulation control;
with lower fully chargedvoltages, recovery Is faster.

The modulation control aspects of the SRprocess result in an encoding

action which is remarkably free of noticeable signal related modulation
effects. Working together with action substitution, modulation control
contributes to the goal of least treatment, in providing a highly boosted,
audibly stable encodedsignal.

-9
 2.4 Overshoot SuDDresslon
A side effect of the modulation control scheme is that at high signal
levels the amplitudes of the signals from the several stages are relatively
high in comparison with the situation in the A-type, B-type, and C-type
systems. Becauseof this it is not possible to employ simple overshoot
suppression diodes as in these previous systems. A more flexible but
necessarily more complex scheme is used that operates directly upon the
control signals.

In commonwith the A-type, B-type, and C-type systems, the SRprocess

features overshoot suppression thresholds that are significantly higher than
the steady state thresholds; this results in Iow modulation distortion. The
overshoot suppression thresholds are set about I0 dB abovethe relevant
steady state thresholds. The net result ls that for most musical signals the
overshoot suppressors rarely operate; the compressors are controlled by well
smoothedsignals. Whenthe the suppressors do operate, the effect is so
controlled that modulation distortion Is minimal.

Underextreme transient conditions, e.g.from a sub-threshold signal

situation, the overshoot suppression threshold is set at its lowest point. The
overshoot suppression effects are then phasedin gradually as a function of
increasing impulse level.

Underrelatively steady state but nonetheless changing signal conditions

the overshoot suppression effects are gradually phasedout as a function of
increasing signal levels; this action further ensures Iow overall modulation
distortion from the system. The phasingout effect is achieved by increasing
the overshoot suppression thresholds; the thresholds are controlled by signals
that are the same as or derived from the modulation control signals usedto
control the steady state characteristics, whereby a tracking action between
the transient and steady state behavior is obtained. This arrangement results
in both well controlled overshoots and low modulation distortion.

Both primary and secondaryovershoot suppression circuits are employed,

the latter acting as fall-back or long-stop suppressors. In the high frequency
circuits the secondaryovershoot suppressors Improve the performance just
inside the stop band - that Is, In the 400 Hz to 800 Hz reglon. In the Iow
frequency circuits these additional suppressors Improve the performance
under extreme complex signal conditions (e.g. high-level Iow and mid
frequency transient signals in combination with high-level high frequency
signals). In the Iow frequency clrcults a further overshoot suppressor (LF O/S)
is used for very Iow frequency signals; this is a very gentle, slow acting
circuit which reduces Iow frequency transient distortion.

I0-
 2.5 Staoaered
vv Action Multi-Level Format
The principles discussed aboveare incorporated into each stage of the
multi-level staggered action encoderand decoder. See reference (4) for a
detailed discussion of staggered action circuits. In the SRsystem two stages
are employed at iow frequencies, three at high frequencies. The thresholds
used are approximately -30 dB, -48 dB, and -62 dB below reference level (20
dB below SE peak signal level). In the series connected staggered actlon
format there is a compoundingof the actions of the Individual stages; the
transfer functions of the several stages are multiplied, whereby the dB
characteristics add. In this way a large total dynamic action can be achieved
with very Iow modulation distortion, Iow overshoot, and goodmanufacturing
reproducibility. An Important additional result Is that there ls an overall
steepness enhancementof the frequency discrimination abilities of the
clrcuTt, further Inhibiting signal modulation and noise modulation effects.

2.6 Spectral Skewino

The spectral skewing networks employed in the SRprocess comprises both

high frequency and Iow frequency sections, with the same rationale and mode
of operation as discussed in Ref. 4 The spectral distributions of the signals
processed by the encoderare altered or skewed, well within the pass band,
such that the encoder, or compressor, action is significantly less susceptible
to the influence of signals beyondthe abrupt roll-off frequencies of the
spectral skewing networks.

The high frequency network is a Iow pass filter with an attenuation

characteristic similar to that of a 12 kHz two-pole Butterworth filter but
with a limiting attenuation of about 35 d8 (i.e. a shelf). The iow frequency
network is a 40 Hz high pass filter, connected in series wlth the high
frequency network, also with a two-pole Butterworth-llke characteristic but
with about a 25 dB limiting attenuation. Theseshelves do not interfere with
the attenuation within the audio band but provide phasecharacteristics that
are essential In the decodingmode.

2.7 Antlsaturatlon

The general principle of antlsaturation was described In Ref. 4. Drlefly, by

placlng a fixed attenuation network, usually a shelf, in the main path of a dual
path compressor, it Is possible to create a very effective antlsaturatlon
characteristic at the extremes of the audio band without any undueadverse
effect on the noise reduction effect achleved durlng decoding. In the SR
process high and low frequency networks are operative aboveabout 5 kHz and
below about I00 Hz,respectively. In addition, the spectral skewing networks
have a secondary but very useful antlsaturatlon effect, especially at very low
and very high frequencies.

.-II-
:3 Block Diagrams

3.1 Basic Block Diagram

AS mentioned previously, Fig. i shows the basic layout of an SRprocessor.

While the whole system comprises an encoderand a complementary decoder,
the figure as drawn shows a switchable configuration, which generally Is the
most useful one. The main signal path transmits high level slgnal components.
To this is addedin the encodingmode, andsubtracted in the decodingmode, the
output of the side chain circuitry, designated the SRsignal, point C. The stage
circuits, as well as the spectral skewing networks and anttsaturation
networks above,are driven from point A. See Ref. i for a mathematical
explanation of these arrangments.

A secondary main path which doesnot include any antlsaturatfon is

employed as the basis of the side chain, to which the outputs of the high level
stages and mid level stages are added In the first stage adder and secondstage
adder, respectively. The Iow level stage and modulation control circuits 1 - 7
are driven directly from the output of the secondstage adder. Hodulation
control circuit 8 is driven from the output of the spectral skewing network, as
will be discussed.

The antisaturatlon effects are created in the dashedblock labelled Stage

Signal Combiner and Antlsaturatlon. The arrangement shown provides a high
frequency de-emphasis effect on the secondarymain path signal, which
includes the output of the high level stages, and on the high frequency mld
level stage signal. This de-emphasis ls effective not only on the steady state
aspects of these signals but on all transient effects as well. Theoutput of the
Iow level stage ls then addeddirectly. For Iow frequency antlsaturatlon the
Iow frequency de-emphasis is effective on the secondarymain path signal,
Including the high level stage outputs. The Iow frequency mid level signal is
then addeddirectly. Ref. 4 provides the mathematical basis for these
arrangements.

With the final combination of signals In the last adder, an SE encoded

signal appears at point B. The encodedsignal can be considered to comprise an
unmodified componentfrom the input plus an SRsignal which carries ail of the
SRcharacteristics. Thus the SRcomponentcan be derived by subtracting the
unmodified input signal at point A from the 5R encoderoutput at point B. This
provides an $R signal at point C that can be handled and switched in same way
as in the A-type system. This simplifies practical use of the system.

3.2 Modulation Control Circuits

Fig. 1 shows how the inputs of the various modulation control circuits are

12-
connectedand how thc r_ultant _ignal ar_ distributed. Modulation control
signals MC1- MC7are derived from the output of the second stage adder. In
this way the modulation control signals begin to have a significant Influence
at relatively Iow levels, such as at -30 dB (becauseof the contributions of the
HLSand MLSstages); the phaserelationships between the modulation control
signals andthesignals inthecontrol circuitsoftheseveral stagesarealso
optimized. Inthegeneration ofMC8,whichisusedforlow-frequency stage
overshoot-suppression Inhibition underhigh-frequency transient signal
conditions, theinfluences of thenoisereduction stagesareundesirable. MC8
istherefore derivedfromthefirst feedpointofthestages, justfollowing the
spectral skewingnetworks.

Fig.Ialsoshows thedistribution
schemeofthemodulationcontrol
signals.
MC 1- MC3 areusedforthehighfrequency
stages;
MC5 - MC8 areused
fortheIowfrequency stages.

InFig. 8,thebasiclayout of themodulation control circuits lsshown.MCI

controls thehighfrequency sliding bandcircuits. Thesignal fromthetakeoff
point ls fed through a 3 kHz single-pole high pass filter, full wave rectified
(all rectifiers In the system are full wave), and fed In opposition to the
control signals generated by the high frequency stages. An all-pass phase
.shift network Is used to optimize the phaseof the MCsignal in relation to the
stage control slgnal at low frequencies; this reduces control signal ripple.
MCI is also smoothed by a two-stage 1 ms integrator and is employed, as MC2,
to opposethe operation of the high frequency sliding bandovershoot
suppression circuits; the overshoot suppression thresholds thereby track the
steady state thresholds. MC2must be smoothedbecause the phase
relationships of MCI and the signals in the stages vary (because of the sliding
band action) throughout the audio band,being a function of frequency and level.

MC3controls the high frequency fixed band circuits. Thesignal from the
takeoff point is weighted by cascaded400 Hz and 800 Hz single-pole Iow pass
filters, rectified, and fed in opposition to both the steady state and transient
control clrcults of the high frequency fixed band circuits. There is no need to
provide a smoothed MCsignal for the overshoot suppressors of the high
frequency fixed band stages becausea fixed phaserelationship exists between
the stage signals and the control signals throughout the audio band.

MC4controls the sllding band circuits of the iow frequency stages. The
signal from the takeoff point is fed through a 200 Hz single-pole Iow pass
filter, rectified, and fed in opposition to the sliding bandcontrol signals
generated in the stages. Thephase relationship of the modulation control
signal is optimized by the use of an all-pass phaseshifter, as with MCI;
Iow-frequency control signal ripple is thereby reduced. MC4is smoothed by a
two-stage 2 ms integrator to form MC5;this signal is used to control the Iow

13-
frequency sliding bandovershoot suppressors.

MC6controls the Iow frequency fixed band circuits. The signal from the
takeoff point is weighted by cascaded800 Hz and 1.6kHzsingle-pole high pass
filters, rectified, and usedto opposethe steady-state fixed band control
signals. MC6is also smoothed in a two-stage 2 ms integrator, forming MC7,
which is used to control the Iow frequency fixed-band overshoot suppressors.
This smoothing is necessary in the low frequency fixed-band stages because,
unlike the §ltuatton in the high frequency fixed-band stages, there is no fixed
phaserelationship between the stage signals and the overshoot suppression
signals. MC7is also used in a supplemental way to control the Iow frequency
sliding-band overshoot suppressors.

MC8is used to control the overshoot suppression circuits of both the fixed
and sliding band Iow frequency circuits. MC8compensates for the fact that no
frequency weighting is used in the generation of the Iow frequency primary
overshoot suppression signals.Highfrequency transient signalcomponents
aredetected andusedtoopposetheoperation oftheLF primaryovershoot
suppression circuits. Thesignal fromtheMC8 takeoff pointisfedthrough a 3
kHzhighpassfilter, rectified,doubledifferentiated with15 Its
time-constants, andpeak-held withabouta 30 ms time-constant. The
resultant hf-translent modulation-control signal MC8 isthenemployedto
opposetheIowfrequency overshoot suppression action.

3.3HighFreQuency
Stage

Fig.
g shows boththesteady-state andtransient
control
aspectsofthe
highfrequency stages.Thediagramshowsonlythebasicparameter
determining elements; thepractical
circuits
ofcoursecontainotherdetails
suchasbuffering, amplification,
andattenuation.
Thehigh-level,
mid-level,
andlow-level stageshavethesame basicblockdiagramsandschematics.
The
maindistinctions arethattheacanddc circuitgainsareincreased
forthe
mid- and low-level stages.

Referring to the block diagram, each stage comprises a fixed band section
on the bottom and a sliding bandsection on the top, each with Its own control
circuits. The fixed and sliding bandcircuits are fed in parallel and the output
signal is taken from the sliding bandcircuit. The sliding bandvariable filter
is referenced to the output of the fixed band;that is, the fixed bandoutput Is
fed directly to the bottom end of the sliding bandvariable resisltance RVs.
This connection results in the action substitution operation as discussed
previously. At any given frequency the overall output will be the larger of, or
some combination of, the fixed and sliding bandcontributions. If there lsa
signal situation in which the fixed bandoutput is negligible, then the sliding
band predominates. Conversely, if there Is little or no sliding band

-14-
contribution, the output from the fixed band will still feed through to the
output through RVs. In this way the action of one circuit augments that of the
other, and, as the occasion requires, may be substituted for that of the other.

Theincomingsignal isfedthrough an 800 Hz single-polebanddefining

filter.
Thisisfollowed bya 400Hz single-pole filterwhichreducestheIow
frequencysignallevels
handledbyboththefixedandsliding bandcircuits,
therebyreducingwaveformdistortion athighsignal levels.Thefilter also
formspartofthefixedbandcontrol signalweighting network.Theoutput
signalIstakenfromthesliding bandstageandisfedthrough a 400 Hz
networkhavinga reciprocal
characteristictothatofthe400 Hz highpass
filter
at theInput.Thustheoverall quiescent(sub-threshold)frequency
responseof thecircuit
Isthatofa single-pole 800 Hz highpassnetwork.The
lowfrequency stageshavea complementary 800 Hz single-poleIowpass
characteristic,
whichoverallresultsInoptimal combination ofthesignals
fromthehighandIowfrequency stages.

ThefixedbandoutputfromRVf Isfedtotwo control circuits,

themain
controlsignal
circuit(top)andthepass-band control
circuit
(bottom).Inthe
maincontrolcircuit
thethesignal Isrectifiedandopposedbythemodulation
controlsignal
MC3. Theresultingdc signalissmoothedby an Integrator
circuit
witha 15 ms timeconstant (theoverallsteady-state
controlsignal
characteristic
inthisandallotherstagesisaverageresponding). Thecontrol
signalisthenfedtooneinput ofa maximum selectorcircuit,whichpassesto
itsoutputthelargerof twosignalsappliedtotheinput.

Thefixedbandoutputisalsofedtothepass-band control
circuit,
which
comprises
a 1.6kHzsingle-polehighpassfilter,
a rectifier,
anda smoothing
circuit
(15ms).Thepass-band controlsignal
isappliedtotheotherinputof
themaximum selectorcircuit.
Theoutputofthemaximum selectorcircuitis
further
smoothedby a 160ms timeconstantandisusedtocontrolthefixed
bandvariable
resistanceRVforothervariable
gainmeans.

Thedual control circuit arrangement described aboveis employed to

obtain optimal performance under both simple signal (single dominant signal)
and complex signal (more than one dominant signal) situations. The
modulation control signal MC3 is optimized in frequency weighting and amount
for simple signal conditions, in which the modulation control action Is most
useful. Undercomplexsignal conditions,however,themodulation control
signal developed becomeslarger, andthesubsequent modulation control action
isthengreater thannecessary; thatis,thedc control signal outputfromthe
maincontrol circuitislessthanrequired. Underthiscondition thecontrol
signal fromthepass-band circuit isphasedIn,viathemaximum selector
circuit,tocontrol theoverall actionofthefixed bandcompressor circuit.

-15-
 Theoutputofthefixedbandisfedthrough a bufferwithanoverall gainof
unity to provide the reference for the slidingband filter;this is the only
signal outputofthefixedbandcompressor circuit.

The sliding bandcontrol signal is derived from the stage output. The
signal is fed through a single-pole high pass weighting network (about I0 kHz,
different for each stage) and is rectified. Therectified signal is opposedby
modulation control signal MC1;since MCI also has a single-pole high pass
characteristic, the ratio between the rectified control signal and MCI
monitors the signal attenuation (this ratio creates an end-stop effect on the
sliding band actlon). The result Is smoothedfirst by a tlme constant of about
5 ms (different for eachstage), and finally by a time constant of 80 ms. The
smoothedcontrol signal Is then used to control the sliding bandvariable
resistance RVs. A single control' circuit suffices in the sliding bandcircuit
becausethe I0 kHz high pass control weighting network tends to offset the
effect of complex signals on the modulation control voltage developed(MCI).

A modification ismade inthesliding bandcontrolcharacteristic

at Iow
levels. Signals fromthefixedbandcircuit arecombinedinoppositionwith
thesliding bandoutputsignal (seecombiningcircuitatrightofFig.
9).The
effectisinthedirection ofsimulating thederivationofthesliding
band
control voltage fromthevoltage acrossthesliding
bandvariablefilter
only
(i.e.
fromthevoltage acrossRVs).Thistendstoraisethesliding band
threshold athighfrequencies, whichreducesunnecessary sliding
oftheband.
(ThelOkHzcontrol weighting networkprovidesthecorrectamountofcontrol
signalforthevariable filteratmediumandhighlevels, butitproduces
the
undesirable sideeffect of loweringthethresholdathighfrequencies.
The
differential control signalderivationmethodcounteractsthisthreshold
lowering effect.)

The overshoot suppression (O/S) arrangments are also shown In Fig. 9. In

the high frequency circuits a general feature is that unsmoothed rectified
signals from the control circuit rectifiers are opposed by appropriate
modulation control signals and are fed via diodes to the final integrator
circuits. The Iow frequency arrangments follow the same pattern, with some
modifications.

Referringtothediagram, Inthehighfrequencyfixed
bandcircuit
the
overshootsuppression signal
isderived
fromtherectifierofthemaincontrol
circuit.
As withthesteadystatecontrol signal,
therectified
signal
is
opposedbyMC3,so thattheovershoot suppressionthreshold
isappropriate
for
conditionsinthesteadystateregime.Theresultantovershootsuppression
signaliscoupledby a diodetothefinal
integrator
circuit.

Inthesliding
bandcircuit
twoovershoot
suppression
signals
areused,

-16-
primary and secondary,.Thc primary _crohootsuppression signal lo derived
from the control circuit rectifier and opposedby MC2,a smoothedversion of
MC1 (MC1 controls the steady state characteristics). Thesmoothing is
necessary because,unlike the situation in the fixed bandcircuit, there is no
constant and favorable phaserelationship between the the signal In the
control circuit andMC1(because of the sliding band);the smoothing enables
tellable bucking action to take place.

The secondaryovershoot suppressor supplements the action of the primary

overshoot suppressor under certain conditions. The primary overshoot
suppression signal is derived from the same rectifier used in the steady state
control circuit, with the consequencenot only of economybut of a favorable
phaserelationship between the overshoot suppression impulse and the signal
to be controlled; this results in low transient distortion. However, the control
circuit frequency weighting responsible for this situation also causes a
reduction of control signal amplitude with falling frequency. A dc bias is used
in the overshoot suppression circuit to create the required suppression
threshold; when the signal amplitude in the overshoot suppression clrcuit
decreases, the bias results in the overshoot suppression effect falling away
faster than the signal amplitude. For frequencies below about 400 Hz a
reducedovershoot suppression effect is appropriate becauseof the
attenuation and phasesh!ft of the stage input filter (see Fig. 5). However,in
the 400 Hz to 800 Hz region there is a slight overshoot suppression deficiency;
this can be compensated,or trimmed, very simply by feeding a small amount of
overshoot suppression signal from the fixed bandcircuit lnto the sliding band
circuit. Thls supplemental signal is called the secondaryovershoot
suppression signal.

Regardingrecovery times, the use of action substitution and modulation

control both contribute to rapid action, as already mentioned. Nonetheless,
reverse biased recovery speed-up diodes are used in a fairly gentle way
(series resistors) to provide a further increase in speed.

3.4 Low FreauencyStaoe: Ste_-State ^snects

Fig. I0 shows only the steady-state layout of the Iow frequency stages.
As with the high frequency stages, only the basic parameter determining
elements are shown. The high level andmid level Iow frequency stages have
the same block diagrams and circuits, but the ac and dc gains are increased for
the mid level stage; there are also some other minor differences.

Referring to the block diagram, certain similarities and differences may

be noted with respect to the high frequency diagrams. The dual-layer
arrangement of the fixed band on the bottom and the sliding band on the top is
similar. However, the sliding band acts downwardly, using a simulated

-17-
inductance (gyrator circuit). As with the high frequency stages, the fixed and
sllding band circuits are fed In parallel, and the output signal is taken from
the sliding bandcircult. The fixed band output is coupled to bottom of the
slldlng band to provide the action substitution operation described previously.

A notable difference from the high frequency circuit is that the fixed 800
Hz band determining filter follows, rather than precedes,the variable filter.
This arrangment has several advantages: a) overshoot suppression signals can
be generated without the delay Inherent In a Iow pass filter, resulting in lower
transient distortion, b) any transient distortion producedby the circuit Is
attenuated by the 800 Hz low pass filter, and c) noise generatedby the
gyrator circuit is attenuated by the filter. Theprice to be paid for these
advantages Is the resulting high signal levels that the variable resistances
RVf and RVs must be capable of handling at high frequencies (there is no
active attenuation at all at very high frequencies, since sufficient passive
attenuation and phaseshift are provided by the 800 Hz Iow pass filter).
Special control arrangements, called high mode,comprising complementary
bootstrapping and control circuit gain-boosting, enable the fixed and sliding
bandsignal clrcuits to handle the required levels with low distortion and Iow
noise.

Referring to the fixed band section, the incoming signal is applied directly
to the variable attenuator circuit. Control circuit frequency weighting is
provided by cascadedsingle-pole 800 Hz and 1.6kHz low pass filters (the
corresponding filters in the high frequency stages are the 800 Hz and 400 Hz
filters at the inputs of the circuits). Themain control circuit rectifies the
filtered signal; the resulting dc signal is buckedby modulation control signal
MC6,smoothed by a 15 ms integrator, and fed to one input of the maximum
selector circuit. The maximum selector clrcuit has the same purpose and
modeof operation as in the high frequency circuits.

The 800 Hz and 1.6kHz frequency weighted output of the fixed band circuit
is also fed to the pass band control circuit. Herethe control signal is further
weighted by a 400 Hz single-pole low pass filter, rectified, smoothedby a 15
ms Integrator, and fed to the other input of the maximum selector. The larger
of the two signals is passed to the final integrator (300 ms) to become the
fixed 'bandcontrol signal applied to RVf (or some other variable gain device);
in this way, both simple and complex signals are accomodated,as in the high
frequency stages.

As in the high frequency circuits, the sliding band control signal is derived
from the stage output - that is, from a point following both the fixed 800 Hz
band determining filter and the variable filter. The signal is frequency
weighted by an 80 Hz single-pole Iow pass filter, rectified, and bucked by
modulation control signal MC4(which also has a single-pole low pass

-18-
characteristic, with the same type of sliding band end-stop effect ss in the
high frequency circuits). The result Is smoothed by a 7.5 ms integrator and
flnally smoothedby a 150 ms integrator to become the sliding bandcontrol
signal applied to RVs. As In the high frequency stages, a single control circuit
suffices for the sliding band.

The same type of Iow level control characteristic modification is made In

the Iow frequency circuits as in the high frequency circuits. Namely, a signal
from the fixed band is combined In opposition with the sliding bandoutput
signal (see combining circuit at right of Fig. 10). The modification raises the
sliding bandthreshold aL low frequencies.

· ,5 Low Frequency_t,aoe: Transient Control Aspects

The transient control (Including steady-state) aspects of the low

frequency stages are shown In Fig. I I. In a mannergenerally similar to that of
the high frequency circuits, unsmoothedrectified signals from the outputs of
the variable elements are opposedby appropriate modulation control signals
and are fed via diodes to the final Integrator circuits.

Both the fixed and sliding bandseach haveprimary and secondary

overshoot suppressors, which operate at frequencies aboveabout I00 Hz.
Additionally, each has a gentle and slow-acting low frequency overshoot
suppressor, operating at frequencies below about 200 Hz; there is a crossover
effect between the two types of overshoot suppression In the I00 - 200 Hz
region. The primary overshoot suppressors provide the earliest and strongest
suppression effect in simple transient situations. With more complex signals
the primary overshoot suppression thresholds rise and eventually the
secondaryovershoot suppression circuitry takes control.

In contrast with the high frequency situation, the Iow frequency general
strategy Is to derive the primary overshoot suppression signals from signal
points that do not Include any control circuit frequency weighting. This is
becausethe required control circuit weighting networks of the low frequency
stages are Iow-pass In character, resulting In delays (the high-pass networks
usedfor control circuit weighting in the high frequency stages do not
Introduce delays). However, becauseof the lack of a weighting factor In the
primary overshoot suppression signal, there Is no Inherent tracking between
the steady-state and overshoot suppression thresholds of the circults
Involved, particularly In the sLop-bands. Therefore further modulation control
techniques are employed to obtain the requlred tracking; secondaryovershoot
suppression signals are derived from a point in the fixed band circuitry that
provides adequate tracking In both the fixed and sllding bands.

Referring to Fig. 1I, the fixed band primary overshoot suppression signal

-19-
ls generatedby passing the vaMable attenuator output through a 200 Hz
single-pole high pass filter. Thls filter reduces the Influence of the primary
overshoot suppressor at Iow frequencies, allowing the more gentle LF
overshoot suppressor to take over the transient control function. The signal Is
rectified and then opposedby modulation control signal MC7,a 2 ms smoothed
version of MC6(the fixed band steady-state modulation control signal); the
effect is In the direction of Improving the steady-state and overshoot
suppression threshold tracking on a steady-state basis. However, the
thresholds must also track on a transient basis. This Is the function of the HF
transient modulation control signal MC8,which lsa high-frequency weighted,
peak-detected signal that opposesthe primary overshoot suppression signal In
the time interval before MC7becomeseffective. The overshoot suppression
signal ls then diode coupled to the final Integrator circuit of the fixed band
circuit.

In the generation of the sliding band primary overshoot suppression signal

the output of the variable filter Is fed through a 200 Hz single-pole filter tn
order to reduce the effect of the circuit at Iow frequencies (as in the fixed
band circuit). The signal is rectified and then opposedby modulation control
signals MC5 andMC7to provide an adequatedegree of tracking between the
steady-state threshold and the overshoot suppression threshold on a
steady-state basis. As in the fixed band circuit, MC8provides the required
degreeof tracking on a transient basis. The resultant overshoot suppression
signal is diode coupledto the sliding band final integrator circuit.

In both the fixed and sliding band circuits, the effects of the primary
overshoot suppression circuits are maximized for the most significant
transient signal situation - that is, a single impulse or toneburst starting
froma sub-threshold signallevel. A sideeffectoftheuseofsmoothedMC5
andMC7 signals isthattheovershoot suppression levels forIowandmedium
frequency transient signals areraisedundercertain complexsignal
conditions, especially thoseInwhichrelatively steady-state highfrequency
signals athighlevels arealsopresent. To compensateforthiseffect,
secondaryovershoot suppression signals are derived from the fixed band
overshoot suppression signal and are diode coupled to the fixed band and
sliding band final Integrator circuits; the secondary overshoot suppressors
have higher thresholds than the primary suppressors and operate only rarely
becauseof the unusualcircumstances for which they are designed.

The secondary overshoot suppression signals are generated from the

frequency weighted point (800 Hz Iow pass and 1.6kHz Iow pass) In the fixed
bandsteady-state control circuit. To prevent Interference with the LF
overshoot suppressioncircuit at Iow frequencies, the signal is further filtered
by a 200 Hz single-pole high pass network, as in the the primary overshoot
suppression circuits; the filtered signal is then rectified. (Note that in Fig. 10

- 20 -
this filter is not shown, for clarity. On a steady-state basis the pass-band
controt circuit controls the circuit at very Iow frequencies, via the maximum
selector circuit; this arrangement allows the main control circuit rectifier to
serve a double function.) The dc signal is opposedby MC6in order to phaseout
the secondaryovershoot suppression effect at hlgh frequencies; an optimal
phase relationship Is obtained between the rectified signal and MC6,apart
from the effect of the 200 Hz filter (which is negligible). An Ideal tracking
effect Is achieved between the steady state and secondaryovershoot
suppression thresholds.

The effect of the 800 Hz and 1.6kHz frequency weighting networks ls to

introduce a time delay Into the secondary overshoot suppression signal. The
effective delay Is significantly reducedby increasing the gain used In the
secondary overshoot suppressor circuit and applying limiting. The resultant
overshoot suppression signal is more in the nature of a nearly fixed amplitude
impulse, applied In the rare circumstances when necessary, than It is a
proportional response. The signal is coupled through a diode to the fixed b; _ld
final integrator circuit and is also used, suitably biased, for secondary
overshoot suppression in the sliding bandcircuit, also coupled through a diode.

TheLF overshoot suppression signal is developedby tapping the rectified,

but unsmoothed,output of the pass-band control circuit of the flxed band
circuit. The signal is opposedDy MC6to desensitize the circuit to high-level,
high frequency components. The signal is further opposedby the resulting
fixed bandsmoothed control signal, in a negative feedback fashion (when the
fixed bandcontrol signal has risen to a sufficient level, there ls no further
need for any LF overshoot suppressor actlon). The stgnal is then highly
amplified and limited, peakrectified, and smoothedby an Integrator with
about a 20 ms decay time constant. The resulting high amplitude pulses are
fed through a differentiating network, with a time constant of the same order
as the Integration time constant, to provide LF overshoot suppression Impulses
of defined strength for distribution to the fixed bandand sliding band final
Integrators, via high value resistors and series diodes. Theresult is a
decaying "constant current" charging of the capacitors of the final integrator
circuits. This Is In contrast with the higher peak currents and correspondingly
more abrupt control voltage changesproducedby the relatively Iow-impedance
primary and secondaryovershoot suppressors. Theuse of the LF overshoot
suppression method results In Iow waveform distortion of relatively slowly
changing Iow-frequency signal impulses applied to the system.

Thecontrol signal recovery characteristlcs are similar to those of the

high frequency circuits, although being about'half as fast becauseof the longer
time constants employed. As In the high frequency stages, the recovery time
is favorably affected by the use of action substitution and modulation control,
but is further augmentedby the use of speed-up diodes.

-21 -
4 Operating Characteristics

The objective of the SRprocess ls to provide a very clean and accurate

soundingreplica of the input signal. This result is achieved by suitable
actions on both the encoding and decodingsides of the process. By meansof
the antisaturatlon technique, the incoming signal is first encodedin such a
way as to reduce harmonic and intermodulation distortion generated in the
recording or transmission channel. Then the decodingprocess reduces any
noise and/or distortion introduced by the channel.

4.1 Dynamic (compressslon) action foe steady-state dominant signals

Referring to the single-tone compression curves shown in Fig. 12,

several features may be noted.

4.1.1 Low freauencies

Dynamicactionoccursintherangefromabout-48dB to- 5 dB (with

respecttoreferencelevel).Thatis,thereisno action(butfull
boosting)
in
thelower35 - 40 dB of thedynamicrange.Similarly
thereisnoactioninthe
top25 dB of thetotaldynamicrange.A linear
dynamiccharacteristic
prevails
inthesetwo regions(abi-linearcharacteristic).

4.1.2 High freauencies

Dynamicaction occurs in the range from about -62 dB to -5 dB. That is,
there is no action in the lower 20 - 25 dB (but full boosting), or the top 25 dB
of the dynamic range (a linear dynamic characteristic in these regions).

In the intermediate level regions of dynamic action the effects of the

multi-level stages are Joined together to create a compression ratio of about
2:I.

Referring to the high-level Iow and high frequency portions of the curves,
the effective antisaturation of the system can be seen, with the combined
effects of the spectral skewing and antlsaturation networks, the SRstages,
and about I dB of wideband level compensation built into the coefficients of
the stage signal combiner (Fig. I). The overall result is an antisaturation
effect of about 2 dB at 5 kHz, 6 dB at 10 kHz and I0 dB at 25 Hz and 15 kHz. At
high frequencies this amount of antisaturation significantly reduces
distortion, reduces signal compression effects, and, with tape recording,
improves the long term stability of the recording. The high frequency
improvments are especially significant with 35 ps CCIRrecordings. The
antisaturation effect at Iow frequencies usefully counteracts tape overload,

- 22 -
p_rticularl¥ wtth 31O0 ps NABrecordings.

4.2 Quiescent (sub-threshold) signal characteristic

Thevery low level, or sub-threshold, characteristics of the SRprocess are

shown In Fig. 13. The general shapeof this characteristic was determined in a
way that takes good advantageof the properties of humanhearing. First, there
is less of a noise generation and perception problem at moderately Iow
frequencies (e.g. 200 Hz) than at moderately high frequencies (e.g.3 kHz);
therefore two Iow frequency stages are employed,but three high frequency
stages are used.

Second,at very Iow and very high frequencies even less noise reduction is
needed(below SOHz and above I0 kHz). Strong spectral skewing actions can
therefore be Usedin these regions, resulting in accurate decodingeven when
the recording medium has response irregularities. Additionally, the spectral
skewing networks provide for good immunity to high and low frequency
interference (supersonic audio components, tape recorder bias; subsonic noise
components arising from wind, traffic, or other rumble sources).

Note that the overall shapeof the low-level SRdecodecharacteristic

resembles the the low level Fletcher-Hunson and Robinson-Dadsoncurves; the
encodecharacteristic resembles the'subsequently derived CCIRnoise
weighting curve.

Thus the SRsystem is designedto reduce only those noises that can be
heard. The prevention of action In inaudible slgnal regions promotes accuracy
In the audible region.

4.3 Treatment of steady-state non-dominant signal components:

The behavior of the system with low-level non-dominant signal

components in the presence of higher level dominant signal componentscan be
simulated by the use of probe tones. Refer to the probe tone curves shown in
Figs. 14 - 16. These curves were obtained by adding a swept frequency probe
tone at levels between -60dB and -80 dB into the encoder input signal and
detecting the tone at the output with a tracking wave analyzer.

Non-dominant signals are boosted over and abovethat of the dominant

signal towards the two spectrum ends by high and low frequency sliding band
actions. If there are two domlnant signals, a fixed bandcompression effect
prevails for the non-dominant signal components between the frequencies of
the dominant signal components.

Thus, non-dominant signal components are boosted by an amount at least

- 23 -
equal to the amount of boost of the dominant signal. The boosting of the
non-dominant signals is maintained towards the spectrum endseven though
the level of the dominant signal is relatively high (e.g. in the range -30 dB to 0
dB). This boosting action spectrally tracks the dominant signal frequency or
frequencies.

It is advantageousto have a steeply rising boosting effect away from the

frequency of the dominant signal component. In this connection the SRcircuit
profits from the steepness enhancingeffect of cascadedstages. The iow
frequencies have two stages of steepness compounding;the use of three
stages at high frequencies further improves the effect. These effects are
particularly evident in the high level areas of the probe tone curves.

The curves show that the compressor circuit tends towards keepingall
low level signal components boosted at all times. Only those components
abovethe threshold are subject to a reduction of boosting. Wlth regard to the
overall system the advantagesof this type of characteristic are:
a) A powerful noise reduction effect in the presence of slgnals.
b) A relative tolerance of level and frequency response errors in the
channelbetween the encoderand decoder.

The audible encodingeffect of the system ls to create a dense, bright

soundingsignal, but with little or no apparent dynamic action. Harmonics,
overtones, and small scale componentsof the sound, including noise, are all
enhanced. The antisaturation characteristics tend to compensate for this
brightness at very high signal levels.

The decodingeffect of the system is to create a very clean sounding

signal; the decoder significantly reduces tape bias noise and modulation noise.
An audible reduction of intermodulation distortion is achieved by the
low-frequency noise reduction capabilities of the process.

The decodercan also reduce audible and measurable harmonic distortion

producedby the recording medium. Steady-state thlrd harmonic distortion is
typically reduced to less than one-half; fifth harmonic distortion is reduced to
less than one-quarter. Higher order harmonics are even further reduced. Thus,
especially if the medium has a hard clipping characteristic, the subjective
cleanliness of the signal at high recording levels Is significantly improved.

5 Calibration

The spectral recording calibration procedures are conceptually similar to

those of the A-type, B-type and C-type systems. That is, signal levels in the
decoder circuit ideally should match those in the encoder circuit, even though
the SRprocess has been designed to be more tolerant of gain and frequency

- 24 -
response errorsthantheseprevious systems.Fortapeinterchange
standardization itisalsopreferable if,atleastwithina givenorganization,
the reference level of the encoderand decoder correspondsto a known and
fixed flux level. Whether or not a standardized flux ts used for this, the
matching of the decoder to the encoder is accomplished by a calibration signal
generated in the encoderand recorded on the tape; this allows the tape replay
gain to be set correctly, using the meter in the decoder unit.

Host problems in the studio use of noise reduction, and Indeedanalogue

recording In general, can be traced to Incorrect level settlngs and/or
frequency response errors In the recorder. This may be becausechecking these
factors is a time consumingandboring process. A faster and more interesting
method of accomplishing these checks would be more likely to produce reliable
and consistent results. For this reason, practical embodiments of the SR
process include pink noise generators which are used for both level and
frequency response calibration, Instead of single-tone slne wave oscillators.
For identification, the pink noise is Interrupted with 20 ms "nicks" every 2
seconds. During recording this signal is fed to the tape at a level of 15 dB
below reference level, a level low enoughnot to cause saturation problems
with Iow speedtape recording or highly equalized transmission channels.

During playback the tape signal is automatically alternated with Internally

generated reference pink noise (uninterrupted) in 4 secondsegments (8 second
total cycle time) and passed to the monitor output. An audlble comparison can
thus be madebetween the reference pink noise and the calib?ation noise
coming from the tape. Any discrepancies In level and/or spectral balance are
immediately noticeable andcan be corrected or at least taken note of. If
desired, the signal can also be fed to a spectrum analyzer.

In using the new calibration method it is important to be able to tell when

the 4 second tape segments are being passed to the monitor and when the
signal heard is from the reference pink noise generator. Differentiation of the
tape segments from the reference segments Is accomplished In two ways.
First, the reference segments are 4 secondsof continuous pink noise, and the
tape segments begin with a nick, havea nick In the middle, and end with a
nick; this tlme sequenceIs easily identified wlth a little practice. Second,
colored lights identify the two different slgnals.

The new calibration facility gives the recording and production personnel a
useful control of the recording process. At any time a check of the recorder
can be made; the result can be heard immediately and conclusions drawn about
whether adjustments might be necessary.

With tape and signal interchanges it is possible to tell qulckly whether

there is any error or misunderstanding about levels, equalization, azimuth and

- 25 -
the like. If the original recording of calibration noise stays with the tape, the
quality of the ultimate playback, even after copying, can be retained. Thus the
comparison function serves to ensure that the recorder and spectral recording
process provide on a routine basis the signal quality and reliability of which
they are capable.

6 Conclusion

A new professional recording format, designated spectral recording, has

been described. The objective of the new encodingand decodingprocess is to
record and reproduce audio signals with a hlgh degree of audible signal purity.

The system employs a dual path multi-level, staggered action arrangment

of two low frequency compressor stages and three high frequency compressor
stages, each with a fixed bandand a sliding band. The outputs of the bandsare
combined in a unique way, called action substitution, which results in an
unusually responsive treatment of the signal with respect to both frequency
and level; a technique referred to as modulation control augmentsthe spectral
tracking abilities of the system. Spectral skewing contributes to a tolerance
of channel errors, and the employment of both high and Iow frequency
antisaturatlon techniques results in a significantly improved channel overload
characteristic.

7 Acknowledgement

The author is indebted to DouglasGreenfield for the printed circuit

layouts, mechanical design, and general assistance with the first practical
embodiments of the new system. Thanksare also due to many other
individuals, particularly Henry Bottino for the design of testers, to Brad
Teaguefor pre-production testing, and to Michael Smith, Christopher James,
and Martin Fried for production engineering.

8 References

I ) R.M. Dolby, "An Audio Noise Reduction System," dl Audio Eng.Soc.,vol.

1S, pp 383-388 ( 1967 Oct.).

2) R.M.Dolby, "A Noise Reduction System for ConsumerApplications,"

presented at the 39th Convention of the Audio Engineering Society,
J_Audio Eng.Soc.(Abstracts), vol. 18, P.704 ( I970 Dec.).

3) R.Berkovltz & K.J.Gundry,"Dolby B-type Noise Reduction System,"

Audio, ( 1973 Sept. & Oct.).

4) R.M. Dolby, 'A 20 dB Audio Noise Reduction System for Consumer

Applications," d_ttzc[J9_?._l__ vol. 31, pp 98-113 (IgS_ March).

- 26 -
 ?
_, co.
_: _°
_m
E_ _- _
-- , __._-
O'J Cr t"_
_'_._ _ _
u. m T.
_._ § -'_
_ 7:Emm
>..'"" _r_ ._
_'__
c_
 __l CompressorF22

v,I , F1 I
!
Fig. 2 Layout of action substitution compressor.

a b Output
: _ Compressor 2 .

c_

Input_ _I Compressor1

Fig, 3 Action substitution configuration usedin SRprocess.

 LEVEL
+10 dB2 i

I
I
I

o /f , T
I
I
FREQUENCYOF
B,
f
DOMINANT SIGNAL

a. Fixed band compressor characteristic.

LEVEL
+10 dB _ _

2dB/_ -

5dB/T

0 ,
FREQUENCYOF f
DOMINANT SIGNAL

b. Sliding band compressor characteristic.

LEVEL
-I-10 dB _._

J' OUTPUT

!
FREQUENCYOF
DOMINANT SIGNAL
f

c, Action substitution compressor characteristic.

Flg. 4 Basic types of compressor characteristics,

 900

Signal Total Signal

I Side Chain
!r F'- r'

/
Main Path
Signal

a) Inthepass-band,
a low threshold
andstronglimiting
characteristic
arerequired.

90°

Main Path
Signal

b) In the stop-band, phaseshift results In the side chain

signal having negligible Influence on the total signal
amplitude; therefore a higher threshold and weaker
Ilmltlng can be used.

Fig. 5 Phasordiagrams - dual path compressor.

 '{- 10 dB

No signal

0 dB

+ 10

-I- 20 0
20 200 2000 20000

FREQUENCY (Hz)

a) Probetonecurveswith 100Hzsignalat the levels indicated- no

modulationcontrol.

+10 dB

+ 5 dB and below
_-_ +10

_ _'"'- +20
/
20 200 2000 20000

FREQUENCY (Hz)

b) Sameas6(a), with modulationcontrol.

Fig.6 Effect of modulationcontrol onfixed bandcompressorcircuit,

 +lOdB

O
20 200 2000 20000

FREQUENCY (Hz)

a) Probe tone curves with I O0Hz signal at the levels indicated - no

modulation control.

-I- 10 dB

No _12/ r'
/i/"4 +'°
j/ -+2o

0
20 200 2000 20000

FREQUENCY (Hz)

b) Same as 7(a), with modulation control.

Fig. 7 Effect of modulation control on sliding band compressor circuit.

 +10 dB

// / / ?
20 200 2000 20000
FREQUENCY (Hz)

c) Probetone curves with 800 Hz signal at the levels Indicated - no

modulation control.

+lOdB

,o 4/y
/ /// ',_+,o
1// '+20

0
20 200 2000 20000
FREQUENCY (Hz)

d) Sameas 7(c), with modulation control.

Fig. 7 Effect of modulation control on sliding bandcompressorcircuit.

Fig. B Basic layout of modulation control circuits. Thesecircuits reduce
modulationsof the _ains andfrequencyresponsecharacteristiCS
use(tin the system, especially at s_gnal_eve_ssignificantly above
the thresholdS.
 : O'BQQQ i

S
dngand _ l
In Out

(~) (-)
MC4

Ii,ed Band

Fig. I0 Low frequencystage block diagram - steady state aspects.

 t°
iT
v
r_
r_
..C
O_
o_
0
o
.Q
_c
 + 20 -t- 20 dB

-10
+0 lO
dB ____,_
___,l__. 4-10
0 dB
-10

_ _09

__"_ 4__._ _
_--- 50 --2030

__ - 60 dB
20 200 2000 20000

FREQUENCY (Hz)

Fig. 12 Single-tone SRencodercharacteristics.

 ,i,

ENCO_-- _ +20dB

+10
,.J
Lu
:>
LU

O dB _-
0.
I-

--10

DECO_ - 20dB

20 200 2000 20000

FREQUENCY (Hz)

Fig. 13 Low-level (sub-threshold)frequencyresponse of SRencoder.

 ,°"___ ,2od_

20 200 2000 20000

FREQUENCY (Hz)

Fig. 14 Low-level encoding characteris[ic in presence 200 Hz signal at

the levels indicated.

j -40 _ J /,2OdB
// I -I-10

' 1,
V 0

20 200 2000 20000

FREQUENCY (Hz)

Fig. 15 Low-level encodingcharacteristic in presence 800 Hz signal at

the levels indicated.
 + 20 dB

,_ _o
20 200 2000 20000

FREQUENCY (Hz)

Fig. 16 Low-level encodingcharacteristic in presence 3 kHz signal at.

the levels indicated.