Note: Descriptions are shown in the official language in which they were submitted.
8~
1--
IN THE RECEIVING OFFICE OF THE
UNITED STATES PATENT AND TRADEMARK OFFICE
INTERNATIONAL APPLICATION
UNDER THE
PATENT COOPERATION TREATY
NOISE REDUCTION APPARATUS
DESCRIPTION
Technical Field
The present invention relates generally to noise
reduction, in particular, to noise reduction in systems
having a frequency-modulated data stream with undesir-
able noise and a reference stream of predetermined fre-
quency that has been subjected to essentially the same
noise. Such systems are commonly formed in frequency
modulation (FM~ tape recording systems.
Background Art
Analysis of noise problems in typical prior art
for tape recording systems has followed the pattern of
noise analysis in communication theory generally. That
is, cornmunication theory often views noise reduction as
the problem of subtracting a noise component from a
stream that is the sum of a signal component and a
noise component. See, for example, Y.W. Lee, Statis-
tical Theory of Communication (John Wiley & Sons, Inc.,
N.Y. 1960), p. 396.
~3~47~5Z
--2--
Prior art FM tape recording systems commonly have
provided two adjacent channels, one channel for recor-
ding and playback of the FM data stream and the other
channel for recording and playback of a reference
stream. To a great extent both channels are subjected
to the same noise, of which tape flutter is usually
the major component. If no noise were present in
either channel, the data stream would be a pure fre-
quency-modulated carrier wave and the reference stream
would be a carrier wave itself. The presence of
similar noise on both the data and reference streams
led to the prior art approach of separately demodula-
ting each stream and subtracting the demodulated
reference stream from the demodulated data stream.
Thus the prior art technique is to subtract any demo- i
dulated noise of the reference stream from the demodu-
lated data-plus-noise of the data stream. This tech-
nique has caused a substantial improvement over perfor-
mance of FM tape recording systems that lack a reference
channel. Nevertheless, prior art FM tape recording
systems utilizing a reference channel do not remove
all of the flutter-induced noise that is common to
both the data and reference channels. The flutter-
induced noise that remains in such systems can be
troublesome in many instances. Further reduction in
flutter-reduced noise generally can be accomplished
only with expensive tape transport mechanisms or
other expensive or complicated approaches.
Disclosure of Invention
It is an object of the present invention to provide
an improved apparatus for reducing noise in FM tape-
recording systems. r
It is also an object of the present invention to
provide an apparatus for improving dramatically the per-
formance of FM tape-recording systems for a yiven cost.
7~5Z
--3--
It is a further object of the present invention to
provide an improved apparatus, for reducing noise in an
FM data stream, of the type utilizing a reference stream
having a predetermined frequency which has been subjec-
ted to essentially the same noise.
It is a further object of the present invention to
provide an improved apparatus for reducing flutter-
induced noise in FM tape-recording systems.
It is a further object of t:he present inver~ion to
provide a relatively lcw-cost instrumentation FM tape-
recording system.
These and other objects of the invention are achie-
ved by providing an improved apparatus, for reducing
noise in an FM data stream, of the type utilizing a re-
ference stream having a predetermined frequency which
has been subjected to essentially the same noise, wherein
the improvement includes a data stream frequency-discri-
minator to transform the data stream into a first signal
of value proportional to the data stream frequency; a
reference stream frequency-discriminator to form a second
signal of value proportional to the product of the refe-
rence stream frequency times a feedback value; a combiner
to combine the first and second signals to generate an
output value proportional to the feedback value; and a
feedback loop to return the feedback value to the refe-
rence stream frequency-discriminator. Some embodiments
of the invention utilize ratiometric techniques to per-
form some of the functions of the preceding embodiment
without using a feedback loop. Other embodiments of the
invention use a feedback technique such as that just de-
scribed to produce a signal which varies as the ratio of
the frequency of a first, data stream to a second, refe-
rence stream.
Brief Description of the Drawings r
These and other objects and features of the inven- ~
~47852
--4--
tion will be more readily understood by consideration of
the following detailed description taken with the accompan-
ying drawings, in which: -
FIG. l is a simplified block diagram of a noise-
reduction system embodying the principles of the present
invention;
FIG. 2 is a mathematical analysis o~ a model of the
system shown in FIG. 1,
FIG. 3 is a diagrammatic representation of an FM
discrimination section in an embodiment of the present
invention;
FIG. 4 is a diagrammatic representation of a data
channel discriminator used in an embodiment of the present
invention;
FIG. 5 is a diagrammatic representation of a refer- ~r
ence channel discriminator used in an embodiment of the
present invention;
FIG. 6 is a detailed block diagram of a noise-reduction
system embodying the principles of the present invention;
FIG. 7 is a mathematical analysis of a model of the
~ystem of ~'IG. 6;
FIG. 8 is a portion of a schematic diagram of a
preferréd embodiment of the noise-reduction system of
the present invention, the diagram being connected to
FIG. 9 at leads 37 and 38,
FIG. 9 is the remainder of the schematic diagram of
FIG. 8;
. FIG. 10 is a view of an oscilloscope screen showing
the operation of the present invention on a full-scale
saw-tooth signal, recorded on an inexpensive tape trans-
port;
FIG. 11 i~ a view of an oscilloscope face showing
the effect of the present invention on a one percent of r
full-scale, saw-tooth signal, recorded on an inexpensive
tape transport; and
FIG. 12 is a view of an oscilloscope face showing
~478~2
--5--
the effect of the present invention on the same one per-
cent of full-scale, saw-tooth signal played back with
gross amounts of artificially induced flutter noise.
Description of Specific Embodiments
Data transmission or recording using a frequency-
modulated carrier is accomplished by deviating the carrier
frequency in response to the amplitude of a data signal,
and by transmitting or recording this modulated carrier
frequency. In a typical FM system, the magnitude and
polarity of the data signal determine, respectively,
the amount and direction of the frequency
r~
.
'
.
,
, .
~147852
--6--
deviation of the carrier. -A "dc" (n~n-time~
Yarying) si~al, depending on its polarity, either
increases or decreaSes the carrier frequency, while an
- "ac" (time varying) signal alternately increases and
decreases the carrier ab~ve and below its "center", or
unmodulated, frequency at a rate equal to that of the
data frequency. In the typical (nlinearn) FM system,
where
fc= the unmodulated carrier frequency, and
fD= the carrier frequency modulated by the data,
the amplitude of the data signal is expressed by the
relative frequency deviation of the carrier. The
instantaneous value of the data, AD, is given by
( ) AD K(fD fc) ~ K = a constant
fc
where the ratio K/fC is the scaling factor for the
channel, i.e., in a system where K/fC = ~0.01 Volt/Hz,
a positive 100 ~z deviation of the carrier would be
equivalent to a data signal of +l volt.
(+0.01 V/Hz x ~fc + 100 ~ fc) Hz = ~1.0 Volt)
Because the data information is imparted to the
carrier by deviating the carrier frequency, the demodu-
lation (data recovery) process must determine, from
whatever deviations are present in the carrier, theoriginal form of the data signal. Any external factors
which are capable of altering the transmitted and/or
received frequencies, hence, will be demodulated and
exist as undesirable, erroneous "noise" components in
the demodulated data signal. If these noise-inducing
factors are not precisely known, then the exact value
of the original data signal can never be determined.
For this reason, many FM systems employ a
reference frequency, a known frequency which is
transmitted or recorded simultaneouslY with the data-
modulated carrier frequency. It may be assumed that
whatever system perturbations have caused variations in
7~S2
this reference frequency from its known value will also
have caused instantaneous related variations in the
data-carrier frequency. By accurately knowing how the
reference frequency has been affected, one can theore-
tically, with sufficient knowledge of how these known
influences on the reference signal would have affected
the data-~odulated carrier, in effect go "backwards"
and reconstruct the true value of the original data
signal. Let
fR = the transmitted reference frequency
fD = the transmitted carrier freguency modula-
ted by the data signal
fR = the received reference frequency (system
perturbations have changed fR so that f~
no longer equals fR)
f~ = the received data-modulated carrier
frequency (the sa~e perturbations have
also changed fn, ln the transmission
process so tha~ f~D ~ fD)
Let (2) ER = f~ ~ fR the known transmission
, error experienced by the
fR and let
(3) ED = fD ~ fD the (unknown) transmission
~ error experienced by the
fD data channel
With sufficient knowledge of the overall system,
one can define a term S which expresses the instan-
taneous relationship between the transmission error in
the data channel, ED, to the error in the reference
channel, ER, i.e., let
(4) S = ED/ER
If one knows both S and ER, ED can now be found by
(5) ED = S x ER
Replacing the now known value for ED into equation
(3) results in
(6) fD = fD the frequency of the data
= carrier with the
1 + ED transmission error removed.
The true frequency deviation and, hence, the true
value of the original data signal can now be found by
demodulating fD
The overall expression for the amplitude of the
data, AD, as a function of the received data and
I
I
_. . .. .
- ~147~52
--8--
- reference frequencies and the system function S can be
found by combining the expressions for S and ER with
equation (1) or
(7) AD = K - _
1 + S(f~ - fR J fc
- ` fc
In any real F~ data-transmission system, it is
- never possible to determine exactly the value of the
original data signal. Digital systems are limited in
accuracy by non-infinite word lenyth which results in
quantizing error, the uncertainty resulting from the
finite size of the smallest incremental change which
the system can express. Analog systems are limited by
residual noise which generates output signals even when
no input signal is present. Because the presence of
noise limits the accuracy of an information system, it
has become common practice to express the accuracy of a
system as a ratio of the maximum amolitude of a
transmitted siqnal to the amPlitude of the svstem noise.
The signal to noise ratio, or S/N, is usually expressed
in decibels, or db.
S/N ~db) = 20 log10 Amplitude of max signal
Amplitude of noise
Thus a S/N of 40 db would indicate a 100:1 ratio, or
a system uncertainty of one percent of the value of a
maximum amplitude signal. When the noise is not uniform,
but varies with the amplitude of the recorded signal,
then a S/N figure would include this "gain" noise or
"percent of signal amplitude" noise as a separate, spe-
cified term. Because of the often complex waveforms
associated with noise signals, care must be taken in
interpreting S/N figures regarding the units of noise
measurement, bandwidths involved, and other related
variables.
The frequency modulation technique is usually
selected for its ability to transmit "dc" information,
and for such a system's relative insensitivity to ampli-
~4~352
g
tude variations in the transmitted and received carrier
waveforms. But because the information is contained in
- the frequency rather than the amplitude of the modulated
carrier, an FM system is extremely sensitive to unwanted
frequency variations. In a magnetic tape recording/
playback system, flutter (unwanted variations in tape
speed) is unavoidable, both during recording and
playback. These velocity variations in the tape speed
frequency modulate the already modulated carrier. The
effect of flutter on the FM carrier is instantaneous
- multiplication of the recorded frequency by the instan-
taneous value of the cumulative record/playback tape
velocity, which equals (1 + F) where F is the cumulative
instantaneous record/playback flutter, or velocity error.
Thus F becomes a mathematical factor common to all
signals simultaneously exposed to the speed variation. A
+1 percent flutter (i.e., F = +0.01) would cause a 3 KHz
carrier to appear as 3030 Hz, a 5 KHz carrier as 5050 Hz,
etc. [Errors such as these in the time-base (absolute
frequency) accuracy of the data can be eliminated only
through use of a buffer system, where data is input at a
rate modulated by the system flutter, but output at a
corrected rate determined by the original sampling rate
or some other reference rate. Fortunately, these
flutter-generated time-base errors are usually insignifi-
cant compared to the corresponding flutter-generated
amDlitude errors in an FM recording system (as evident in
FIGS. 10, 11, 12). All references to nnoise" will refer,
as is customary, to amplitude error rather than to time-
base error, unless otherwise indicated.]
If one lets
F = the cumulative (for both recording and
playback) instantaneous flutter, or speed
error, in the velocity of the ~agnetic
tape,
f = the frequency of the unmodulated data
c carrler,
m = the amount of modulation of the carrier due
to the data signal, i.e., m - +.5 for +50
percent modulatlon ,
; '''^~
' ' 7
~78SZ
-10--
~ then in a typical (~linear") Fr~ system
(8) fD = f (1 + m) = the frequency of the re-
c corded, data-modulated
carrler, and
5(9) fD = f (l + m~ (1 + F) = the instantaneous
c frequency of the
same signal,
fc ~l + m), upon
playbac~, with
- an lnstantaneou~
cumulatlve flut-
ter of value F.
-If the reference carrier, wnich by definition has a
known modulation (usually zero), has been recorded
simultaneously with the data carrier, then if
fR = the frequency of the recorded reference
carrier, then
(lO) f' = f (1 + F) = the instantaneous fre-
R R ~uençy of the reference
carrler, fR, upon play-
back, with an instanta-
neous cumulatlve flutter
equal to F.
Equation (9) demonstrates the non-uniform effect of
the flutter, in that the frequency error caused by the
flutter is multiplied by (1 + m), or by the data modula-
tion itself. Thus a one percent flutter will cause a one
percent change in the carrier only when m = O, or when no
data is present. In a s~stem where the modulation is
symmetrical about the carrier frequency, the absolute
value of m must be less than one to prevent the carrier
frequency from going to zero at m = -l, so that the
effect of the flutter is nearly doubled as m approaches
+1, and approaches O as m approaches -1.
In the most widely used method for reducing flutter
nolse in FM recordings a reference carrier with no modu-
lation is recorded simultaneously with the data-modulated
carrier, either on the same channel (or track), or on a
separate channel. Each channel is demodulated via a
limiter ~to lessen the effects of amplitude modulation),
an FM discriminator such as a phase-locked loop, or the
more common constant-energy pulse generator (a mono-
stable multivibrator), and a low-pass filter. The output
of the reference channel is then subtracted from the out-
.
~1~7852
put of the data channel. As previously stated, in a
typical FM system the output signal or voltage is
linearly proportional to the deviation of the carrier,
Vout = K ~ ~
Substituting the expression for the played-back fre-
quency of the data channel (equa~ion 9) for fD in the
above equation, and letting K = 1 for the sake of simpli-
city, yields
tll) VOUt [Data Channel] = fc(l + m)(l + F) ~ fc
= (1 + m)(l + F~ - 1
= 1 + m + F + mF - 1
= m + F (1 + m)
Substituting in like manner the expression for the
reference channel playback frequency (equation 10)
results in
(12) VOut[Ref. Channel~ fR(l R
fR
= 1 + F - 1 = F
Subtracting the output of the reference channel from
that of the data channel indicates that
(13) V [Data Channel -
out Ref. Channel] = (m I F + mF) - (F)
= m[data] + mF[noise]
or that the output signal consists of the desired data
term, m, and a data-modulated flutter-generated term, mF.
It is lmmediately apparent that thls flutter correc-
tion method is far from perfect, in that a noise term is
present except when m = 0, i.e., the effect of the
flutter is only removed when there is no data beinq
recorded.
Equation 11 shows that without a reference channel,
the ratio of noise to maximum signal output for the data
channel would be
(14) N/Smax[Data Channel] = F(l + m)
Because practical considerations limit the allowable
modulation to less than 100 percent (i.e., mm~X must be
.. ..
!~ .
.
.
7~352
-12-
- less than 1), equation 14 expresses the "flutter
multiplication" effect on S/N, in that a certain percen-
tage of flutter will result in a qreater percentage of
noise relative to the maximum possible signal.
For a modulation equivalent to +Full Scale of +50
percent, the corresponding N/Smax figures would be
m N/SmaX
- nPositive Full Scalen +.5 F(1.5/.5) = 3F
"Baseline" O F(1/.5) = 2F
nNegative Full Scalen -.5 F(.5/.5) = lF
i.e., the effect of the flutter is doubled at m = O,
tripled at +Full Scale, and in a one-to-one ratio at
-Full Scale.
Because subtracting the output of the reference
channel results in an output expressed by m + mF, the
noise to signal ratio is improved, and expressed by
(15) N/Smax[Data Channel - Ref. Channel] = mF/mmax
The corresponding N/S figures for the same +50 per-
cent modulation are
m N/S
- max
+.5 F
O O
-.5 -F
It is apparent that this technique for flutter com-
pensation is only effective for small amplit~de signals
(m near 0), and that the effect of the flutter is still
100 percent for signals of maximum amplitude.
The preceeding calculations express, of course,
theoretical performance. In a real system using this
technique there will always be some flutter noise present
even when no siqnal is present (m = O), primarily because
of slight variations in the gain and phase responses of
the two low-pass filters necessary for the demodulation
of the two channels.
It follows from my preceeding expressions for the
frequency outputs of the data and reference channels that
~4173852 ..
to perform perfect compensation (i.e., remove all
flutter--generated noise for anY modulation value), a
theoretically perfect demodulator would have to perform
division of the data channel playback frequency by the
reference channel playback frequency, i.e.,
(15) f [Data channel] = fc(l ~ m) (1 + F)
f [Ref. channel~ fR(l + F)
= l + m when fc = fR
which results in a constant (l) and the desired data
signal (m). Subtracting the constant leaves only the
data signal with all am~litude noise caused by the
flutter completelY eliminated.
One possible system to accomplish this involves an
analog divider circuit. ~owever, highly accurate and
stable division is a relatively difficult function to
perform with analog circuitry. Devices which are
currently available to perform such functions are not
only expensive, but fall short of the ideal device in
their operation, being prone to such phenomena as tem-
perature instability, "dc" drift, non-linearity, limited
irequency response, and Lnternelly generated noise.
.
,
, .
. '
78S2
-14-
Another possible system attempts flutter removal
utilizing computer-based digital techniques. The system
measures the frequencies of the reference and data chan-
nels and stores this information in the computer, where
the mathematical process of division is performed upon
the frequencies, resulting in the "true" value for
fD which can then be mathematically demodulated to yield
the information transmitted. Besides requiring a com-
puter and associated peripheral equipment, the overall
result falls short of the theoretical values, due pri-
marily to the fact that:
a. the measurement of frequency requires sampling
over a time interval. The more accurate the
desired measurement, the longer must be the time
over which the sample is taken. Because the
frequencies are usually changing during the
sampling period, the figures which are later
operated upon may not be of sufficient accuracy,
and
b. overall accuracy is limited by the finite word
length used in the calculations.
Another possibility is a system where the input data
signal is digitalized via an analog-to-digital converter,
the resulting samples being recorded in digital form.
-~er
"
1~7b~52
-15-
The playback process then consists of passing each sample
through a digital-to-analog converter to reconstruct the
original data signal. The only effect of flutter in such
a system is inaccuracy in the time base in the reproduced
data, i.e., no amplitude noise is added to the output
signal by the flutter. The system is limited in accuracy
only by the number of bits used in the digitizing pro-
cess. The sampling rate must be at least twice the fre-
quency of the highest data frequency, and requires a
fairly sophisticated digital recorder, as well as the
required analog-to-digital and digital-to-analog conver-
ters. Such a system would be inherently quite expensive
and would very likely require higher bandwidth capabili-
ties in the recording process than would a comparable FM
system. Although such a system would be many times more
expensive than a similar FM system, its improved perfor-
mance over currently existing FM systems would justify
the expense in many situations requiring a low-noise,
high-performance recorder.
As previously stated, the effect of flutter (unwanted
variations in tape speed) on a recorded frequency is
instantaneous multiplication of the recorded frequency by
the instantaneous value of the cumulative record-playback
tape velocity, which equals (1 + F) where F is the cumu-
lative instantaneous record-playback flutter, or accumu-
lated velocity error, and exists as a mathematical factor
common to every signal which is simultaneously exposed to
the noise generating operation.
Where
f = the frequency of the unmodulated data
' carrier, and
m = f - f = the amount of modulation imparted
c to the data carrier frequencY bY
fc the data siqnal (i.e., m = +~
for +50 percent modulation)
then in a typical FM information transmission system
f = f (l ~ m) = the frequency of the recorded,
D c data-modulated carrier fre-
quency
7852
-16-
~ If one lets
F = the cumulative instantaneous flutter, or
speed error, in the velocity of the magnetic
tape
then
1 + F = the effective playback velocity of the
magnetic tape
- and, therefore,
f' = f (1 + m)(l + F ) = the instantaneous fre-
D c D quencv of the recorded
signal, f (1 + m), upon
c
playback, with an
lns~antaneous cumula-
tive data channel
flutter equal to FD.
In similar fashion, where
fR = the frequency of the unmodulated recorded
reference carrler
then
f' = f (1 + F ) = the instantaneous frequency of
R R R the reference carrier, fR,
upon playback, with an instan-
taneous cumulative reference
channel flutter equal to FR.
Due to the fact that the data-modulated carrier and
the reference carrier are simultaneously recorded onto
(and played back from) the same magnetic tape, tape speed
variations during the record or playback process are
essentially identical for each channel at any instant in
time. Although such factors as dynamic skewing and non-
uniform dynamic stretching of the tape can result in
slightly different instantaneous flutter values between
two different tracks on the tape, these mechanical
effects are relatively minute and the analysis will make
the excellent assumption that the F values in the
expressions for the playback frequencies are the same,
i.e., that FD = FR = F.
Because the effect of flutter is "common-mode
multiplication" of the two frequencies (i.e., each fre-
quency by the same amount), the circuitry of the inven-
tion reiects as data any frequency modulation which has a
common multiplier between the data and reference chan-
nels, since such frequency modulation could only have
1~47852
-17-
been the result of flutter. This ~ocess is performed
by multiplying the reference channel playback signal by
(1 ~ m) utilizing a feedback configuration, and by
subtracting the resulting value from the data channel
S signal. Mathematically, the simplified expression for
the overall operation performed is
(23) f (1 + m) (1 + F) ~ fc ~ rfR(l + F) fRl(
rc L R
where the m in the (1 + m) term is provided by feeding
bac~ the output to the reference channel discriminator,
where the necessary multiplication by (1 + m) is
performed.
FIG. 1 is a simplified block diagram of an embodi-
ment of the invention, which shows the basic signal flow
and, more importantly, FIG. 2 shows the analogous mathe-
matical model which demonstrates the feedback technique
whereby the output signal is used to generate itself.
As will be shown below, there are no low-pass filters
between the FM discriminators and the summation circuit.
The summation of the data channel signal with the
inverted flutter channel is, therefore, performed essen-
tially instantaneouslY as the freauencies come off the
tape, by summing the voltage pulses representing the
values of each channel without either signal having to
first pass through a low-pass filter. This predemod~la-
tion pulse summation eliminates the gain, phase, offset,
temperature coefficient, and transient-response differen-
ces inherent between two low-pass filters, and renders
the flutter noise cancellation virtually perfect. The
invention in fact eliminates complete demodulation of the
reference channel, in that no low-pass filter is required
for the reference channel but only a single pulse-
generating discriminator which can supply any number of
data channels, with the proper reference channel signal.
The overall system characteristic is one of translating
the accuracy of the flutter removal from the flutter fre-
quency domain to the data frequency domain; i.e., with
1~47852
-18-
- this system the removal of the flutter noise is essen-
tially independent of the flutter frequency.
A figurative representation of the FM discrimination
section of an embodiment of the present invention is
shown in FIG. 3 (in general), FIG. 4 ~specifically for
data channel) and FIG. 5 (specifically for reference
channel). The data channel uses the simple but very
accurate constant-energy-pulse generator (known as a
- monostable multivi~rator) as the FM discriminator. This
device produces a rectangular pulse of constant amplitude
(V) and constant width (r) for every cycle of the input
frequency signal. The average or "DC" value of such a
series of pulses is given by (see FIG. 3)
( ) average ~ f(t) dt = Vr 3 Vrfin
where f(t) = V for t = O to r, and 0 for t = r to T, and
where T = l/input frequency. Usually r is set such that
when the input frequency eauals fc (i.e., no modulation
is present), r = .5T and the waveform has a 50 percent
t Vaverage = U5T = V Variations in the
input frequency change T and, therefore, the average
value. As the frequency approaches 0, the pulses become
so widely spaced that Vaverage approaches 0, and as the
frequency approaches 2 fc the spacing decreases and
Vaverage approaches V. Thus the relative change in
VaVera9e is directly proportional the frequency
deviation of the input signal from fc. The overall
effect of a changing input frequency on the average value
is that the average value is equal to the mid peak-to-
peak voltage plus half the peak-to-peak amplitude times
the ratio of the frequency deviation to the center fre-
quency or,
(25) Vaverage Vl + V2 + V~ V2~ ¦~;A tc~
when r = .5T for fin = fc
This is because the integral in the expression for
' '~
'
1~785Z
- --19--
~ne average value (Equation 24) is equal to the area
under the wave-form from 0 to T, and may be easily deter-
mined as equal to the sum of the shaded areas I and II
indicated in FIG. 3. Since the area of I equals (Vl -
V2)r and that of II equals V2T, then
( ) average T ~ flt, dt = 1 [area I + area II]
= 1 [V2T + (Vl - V2)r] = V2 + (Vl - V2)r
This may also be expressed as
(27) V = V2 + (Vl ~ V2)r T
Tc T
Because we know from the defined terms that
c / c fin
T l/fin fc
equation 27 may be expressed as
( ) Vaverage = V2 + (Vl - V2)r fi
T f
c c
This may be correctly rewritten as
(29) V = V2 + (Vl ~ V2)r fin -1 + (Vl - V2)r
'l'C ~ - 'l'c
Combining and simplifying terms, ~then
( ) average ' V2 ~ V2r + Vlr +rr (Vl - V2~¦ fi ~ f
~ ~ L~ ~ fc
If we set r equal to 5Tc in equation 30, the
expression further simplifies to that indicated in FIG.
3, namely
(31) Vaverage = V2 ~ V22 + 2l + Vk - V2] [~n ~ e.]
= Vl + V2 +[Vl - V2~ o]
For the data channel, V2 = 0 and fin equals fc(l +
m) (1 + F) when flutter is present, so that
(32) VD = Vaverage = V + _ ~c(l + m)(l + F) - f
[Data Channel] c
= V[l + (1 + m)(l + F) - 1]
= V(l + m)(l + F~
1~7852
-20-
Thus, the amplitude of the pulses is multiplied by
the factor which causes spacing deviations in the output
pulses (i.e., the input frequency); thus it might prove
feasible to perform effective multiplication of a fre-
quency by using the amPlitude of the pulse train as oneof the terms to be multiplied. This is precisely how the
FM discriminator in the reference channel performs the
multiplication indicated in the system block diagram. In
the reference discriminator, Vl is the feedback voltage
from the output before the offset voltage is removed,
namely z(l + m), and V2 is this same voltage inverted, or
V2 = ~ V(l + m). The r-width pulses are inverted to
change the sign of the output signal so that the addition
performed is in reality subtraction.
In like manner, therefore,
R average = ~ (V/2)(l + m) - (V/2)~1 ~ m)
[Ref. channel] [(V/21(l + m) 1(-V/2)(1 + m)
rfR(l + F)
L fR J~
- 0 - (V/2) (1 + m) (1 + F ' 1)
= -(V/2)~1 + m) F
which is a signal proportional to the tape flutter
multiplied by the data modulation.
FIG. 6 is a detailed block diagram of a preferred
embodiment of the invention and FIG. 7 is the mathemati-
cal eqùivalent of the operations performed by the cir-
cuitry. The data and reference fre~uencies are dividedby 2, recorded, and then doubled upon playback to both
conserve tape bandwidth an gain an octave of frequency
separation for simpler output filtering. The resistive
adder 44 attenuates the signals by a factor of 2 and is,
therefore, followed by a gain-of-2 amplifier 30, which is
adjustable over a small range and is used to set the
flutter noise precisely to zero. The ndc~ offset of
'
.
.
i
8S~
-21-_
amplitude V/2 is not removed until the output stage
because it is circulated through the feedback network to
provide the proper switching bias voltage level for the
bipolar reference channel switch 46. The entire demodu-
lation system uses only a single positive voltagereference denoted at +V (+V equals +5 vdc in the actual
circuitry) to generate a bipolar output of +2.5 volts for
+50 percent modulation. Changes which might occur in the
+5 v reference have no effect on the output offset
voltage and no effect on the accuracy of the flutter-
noise removal. The only effect is on the gain since the
output signal is Vm.
A schematic diagram of a preferred embodiment of the
invention is shown in FIGS. 8 and 9. Each heavy square
or triangle represents a tiny integrated circuit. In the
diagram, all of the triangular operational amplifiers
have +12 volts connected to their number 7 pin and -12
volts connected to their number 4 pin. Q4, Q5, Q6 and Q7
represent 50 ohm "N" channel field-effect transistors.
All resistors are one percent, 1/8 watt, m~tal film
unless specified. Zl and Z2 represent Fairchild yA709C
operational amplifiers, Z3 is a National Semiconductor
Corp. LM339 Quad Comparator, Z4 is a SN74L86 Quad
Exclusive-Or Gate, and Z5 and Z6 are Signetics 555T
Timer8. Z7 and Z8 are RCA 4007 cos/mos "Dual
Complementry Pair Plus Invertern, and Z9, Z10, Zll, Z12
and Z13 are LM307 Type operational amplifiers. 20T and
S.T. mean "20 turns" and "single turn" respctively, and
, w.w. means wire-wound. The frequency signals for each
channel are recovered from the moving magnetic tape 20 by
the data channel playback head 34 and reference channel
playback head 35 (both physically in the same "playback
head~) at the millivolt level and are boosted by a gain
of approximately 2000 in the playback amplifiers 21 and
22 over a 3 db bandwidth of from 900 Hz to 5000 Hz. The
center fre~uency fc of the data channel was arbitrarily
chosen at 3000 Hz, with a maximum data-modulated
,, .
~,,
,
,
,..
,.
~j
.,
~78S2
deviation of +1500 Hz (+50 percent). The recorded
reference frequency can also be 3000 Hz, or it may be
varied from this value since it isn't modulated. The
sinewave outputs of the playback amplifiers are fed into
S Z3, an "LM339 Quad Comparatorn, a device which contains
four high-gain open-loop amplifiers which serve as zero-
crossing detectors 23 (two are left over and can be uti-
lized for two more data channels). Each of the two
square-wave outputs from Z3 (one at the data frequency,
one at the reference frequency) drives the frequency
doubler and pulse generator 24 including Z4, a
Transistor-Transistor-Logic (T2L) "Quad Exclusive-Or"
circuit, each input driving a gate with a resistor-
capacitor network on the input which causes each gate to
output a single positive pulse for each transition of the
square-wave input. This doubles the frequency of each
channel, with the pulses approximately 5 micro-seconds
wide. These narrow pulses are inverted by the remaining
2 gates in Z4 to provide the proper negative-pulse input
, 20 trigger signals for the monostable multivibrators 25 and
i 26, Z5 and Z6. These are Signetics Model NE555 Timers
which produce a single-positive output pulse (of
adjustable width) for each trigger pulse. ~ach output is
then level-shifted and inverted through a high-speed
25 transistor switch 27 and 28 tQl and Q2) to provide the
optimum input signal for Z7 and Z8, each of which is an
~CA CMOS "Dual Complementary Pair plus ~nverter~. These
devices are normally used for digital logic applications,
but they are utilized in this invention to
' 30 1. provide the optimum voltages to drive the gates
of the field-effect-transistors tF.E.T. 'S) in
the pulse-summation circuit, and
2. to provide the inversion required so that only a
single F.E.T. in a pair is on at any one time-
i.e., to drive them out of phase with each
other.
Z7 drives the gate of Q3 between -12 vdc (Q3 "off~)
,"
~147852
-23-
and +5 vdc (Q3 "on") and the gate of Q4 between zero
volts (Q4 ~onn) and -12 vdc (Q4 "offn). Z8 drives the
~ gates of Q5 between the feedback voltage (V/2) (1 + m)
(Q5 "onn) and -12 vdc (Q5 "offn), and the gate of Q6 be-
tween (-V/2)(1 + m) (Q6 "onn) and -12 vdc (Q6 ~off~).
The net result of this switching is that VD is an
extremely accurate square wave driven between +5 vdc (Q3
on, Q4 off) and zero volts (Q3 off, Q4 on), with rise and
fall times on the order of 100 nanoseconds, and with a
pulsewidth determined by the data channel monostable
multivibrator 2~. As demonstrated previously, the "dc"
value of VD is (V/2)(1 + m)(l + F). The reference output
voltage VR is also an extremely accurate rectangular
pulse which is switched between (V/2)(1 + m) when Q5 is
on and Q6 is off, and (-V/2)(1 + m) when Q5 is off and Q6
is on. The average value of VR has been shown to equal
(-V/2)(1 + m) F. This minus sign is created in Z8 by
reversing the driver arrangement of Z7 so that the pulses
are negative instead of positive. Thus the summation
circuit 44 can perform subtraction by adding the two
signals.
One reason for the simplicity and extreme accuracy of
this invention is the capability to eliminate the two low-
pass filters which would normally be used to demodulate
these data and reference channel pulses prior to their
addition.
Because the process of low-pass filtering is mathemati-
cally a linear process, theoreticallY there should be no
difference between
1. filtering two signals and then adding them to
form the sum, or
2. adding two signals and then filtering the sum.
If one denotes the filtering process by X, then the
equivalent mathematical statement would simply be
KA + KB = K(A + B)
The summation expressed by the KA + KB term need not be
highly linear outside of the bandwidth represented by the
~; ,
.
~:~L47852
-24-
filtering process K. However, to accurately perform the
~(A + B) summation requires that the process be extremely
linear essentially over the frequency spectrum of the
signals to be added. Any non-linearity will generate
intermodulation distortion products which will appear as
unwanted noise signals at the sum and difference frequen-
cies of the added signals, if the sum and difference fre-
quencies fall within the bandwidth of the filter.
Because A and B are in reality ultra-accurate, fast rise-
and-fall time pulse waveforms containing harmonics from
the fundamental frequency of a few thousand ~ertz up into
the megahertz region, the summation of these signals must
be of exceptional linearity over a comparable bandwidth.
Any attempt to perform this summation with active cir-
lS cuitry (such as with operational amplifiers) prior to anyfiltering of the waveforms would fail dismally due to the
gross non-linearities encountered at increasing frequen-
cies. For this reason the actual summation is performed
by two simple resistors, which are inherently highly
linear over an extremely wide bandwidth.
The effective source impedance of such "switch" with
respect to the summing node (denoted as 50) is 15 k ohms
(+l percent) plus the resistance of the particular field-
effect-transistor which is conducting at the time.
Field-effect transistors of maximum-on-resistance equal
to 50 ohms were chosen because variations in their on-
resistances would be small compared to 15 k ohms. The
actual resistances of the transistors is relatively unim-
portant. The only requirement for a high degree of
linearity is that the transistors driving the summing
junction be
1. matched on a "side" ~i.e., Q3 to Q4, QS to Q6),
or
2. symetrically mismatched.
The linearity of the summation network can be
observed by supplying to the frequency inputs a data and
reference frequency separated in frequency by a number of
~ ,., ,.-
~j7852
Hertz less than the bandwidth of the output filter sothat whatever intermodulation noise is generated at the
difference fre~uency may be observed at the output with a
high-gain oscilloscope. For example, a 3000 Hz and a
3010 ~z signal will cause residual circuit non-
linearities to generate a 10 Hz intermodulation noise
signal of very small amplitude, typically about -66 db
peak-to-peak (all "db" figures mentioned are with respect
to a maximum peak-to-peak output signal for + 50 percent
modulation). Although this figure is exceedingly small
- (1 part in 2000) it was desired to obtain a residual-
system-noise figure well below this level. For this
reason, R47 and potentiometer R48 were added. R48 is
adjusted for minimum intermodulation noise at the output,
in effect creating a symmetrical mismatch of the driving
impedances. This simple adjustment reduces the
intermodulation-distortion products generated by residual
non-linearities in the summation circuitry to the
insignificant level of -80 db, or 1 part in 10,000. This
adjustment would be unnecessary in all but the most cri-
tical of applications.
The feedback filter 29 is a 4 pole filter; a 3-pole
Chebyshev active filter ~Z9) followed by a single pole,
R35 and C20. (The single pole permits simple, predic-
table modiication8 of the filter response, if desired,
by merely changing R35.) The primary purpose of this
fllter is to attenuate the frequencies generated at the
summing node. Any residual data carrier or reference
carrier frequencies at the output are circulated in the
feedback network as noise on the two switchi,ng voltages
[(+V/2)(1 + m)l used by the reference channel, and will
be demodulated by the sampling action of the reference
bipolar switch formed of transistors QS and Q6. The net
effect is similar to that produced by non-linearities in
the summing circuitry in that a small residual
difference-frequency signal will exist on the output
signal. The amplitude of this noise signal is approxima-
,
~1478S2
-26-
- tely equal to the amount o~ attenuation provided by the
feedback filter to the data and reference channel fre-
quencies. This filter, therefore, exhibits better than
70 db of attenuation at 6 k~z to maintain the residual
5 system noise at an insignificant level. It is, of
course, possible to filter one or both of the signals to
be added before the adder network. However, to do so
would be unwise from an economic standpoint, since each
signal would require a separate filter, whereas both
signals benefit from the action of the feedback filter.
The effect of phase shift (time delay) in the feed-
back filter is to generate a small instantaneous ampli-
tude error in the switching voltages of the reference
bipolar switch. The overall effect is the generation of
a small qain error in the data output which is a function
of the data frequency and the cumulative record-playback
flutter. Even with a feedback filter 29 bandwidth as low
as the output filter 32 bandwidth (typically 100 Hz),
this effect is small in that it requires a 30 percent
flutter to generate a few percent peak-to-peak gain error
in the amplitude of a +full scale 100 Hz signal. This
effect can be made insignificant by increasing the band-
width of the feedback filter and thereby decreasing its
phase shift for any particular frequency, retaining of
course, the necessary attenuation required for the data
and reference carriers. For this reason, the feedback
filter shown has a wider bandwidth than the output
filter, with appreciable attenuation starting for fre-
quencies approaching lRHz. The Chebyshev section pro-
vides a small peak in the filter response to improve thephase-shift characteristic. This reduced phase-shift
~- (compared to the output filter phase-shift) reduces the
gain error in a +full scale 100 Hz data signal to well
below 1 percent for a flutter of 30 percent, which ls
approximately 30 times the flutter that would be expected
in normal situations. (Phase correction circuitry could
be incorporated here but its use would be unwarranted in
i all but the most critical applications.)
,, .~
's'~
,. . , ~ . .
,
; .
~7852
-27-
The low-pass output filter 32 indicated on the sche-
matic is also a 3-pole Chebyshev filter due to the more
effective filtering provided over a standard 3 pole
Butterworth filter, at the expense of absolute flatness
in the frequency response characteristic. The actual
number of poles, filter flatness, type, and bandwidth is
dependent on the requirements of the application. In
this circuit a filter bandwidth of from 0 to 100 Hz with
an accuracy of f.l db was typical. It must be remembered
that non-linearities in the filter response will, of
course, demodulate flutter-modulated input signals,
resulting in amplitude modulation of the output. This
is a consideration when determining output filter
requirements and overall system gain accuracy.
Z13, the offset subtractor, and X2 data amplifier
33, is an operational amplifier configured for a positive
gain of 2 with respect to the data signal, and a negative
gain of one with respect to the voltage reference V.
This circuit subtracts the "dc" offset of V/2 from the
data signal and multiplies the data by two so that the
final output signal is Vm. An adjustment is provided so
that the ~dc" offset of the output may be precisely set
to 0.000 Yolts. This circuit could be followed by a
variable-gain stage to vary the output amplitude, if such
a feature is desired.
One technique to initially set up the noise-reducing
circuitry for proper operation is as follows:
1. Ground the signal input to the recorder and adjust
the center frequency of the data oscillator and the
reference oscillator for the desired frequencies.
2. Switch the ci~cuitry to a "Monitor" mode which
connects the recording frequencies to the inputs
of the demodulation circuitry, and switch on the
flutter noise reduction system.
3. Using R18, adjust the pulse width of the data
monostable multivibrator so that the Vd test point
equals 2.500 volts as measured with a Digital
: ,. ~,r~
,,
~147852
-28-
- voltmeter. (It is assumed that the +v voltage
- reference equals +5 volts).
4. Similarly adjust R25 so that the voltage at the
VR test point measures 0.000 volts. ~There
adjustments insure that the pulse generators
have about a 50 percent duty cycle at the center
frequencies.)
5. Record 20 seconds of tape leaving the input to
the data channel grounded--this is a "baseline~
signal which should generate a 0.000 volt out-
put.
6. Playback this signal while monitoring the
voltage at the output of the feedback filter
which will be approximately 2.5 V (2.500 V ~ a
few millivolts).
7. Introduce gross amounts of flutter (by either
; physically slowing down the recorderis capstan,
or by turning on and off the power to the tape-
drive motor) and adjust R37, the flutter noise
adjustment, so that the voltage at the loop
filter output does not chanqe while the tape is
experie~cing these speed variations.
8. Adjust the ZE~O potentiometer ~R45) so that the
output signal is 0.000 volts while playing back
the "baseline" signal.
Recorders in use today which are capable of peak-to-
peak signal-to-noise ratios approaching 50 db are in the
several thousand dollar price range, and are sorely
pressed to attain even the 50 db mark, with or without
present forms of flutter compensation. Use of this
invention enables a $50 audio cassette recorder operating
at 1 7/8 inches per second, to reproduce frequency modu-
lated data with a peak-to-peak signal-to-noise ratio of
better than 60 db over a flat bandwidth of from 0 to 100
Hz. This is several times better than can be presently
obtained by any recorder regardless of cost operating at
twice the tape speed.
~785Z
-29-
The invention works so well in eliminating the
effects of tape speed variation, that if one is willing
to accept gross variations in the time base accuracy of
the data, the 60 db figure can be reached even by turning
5 the recorder motor ~y hand.
As previously mentioned, the flutter removal is
essentially independent of the flutter fre~uency, and is
virtually unaffected by temperature variations, etc.,
which might cause small variations in the recorded fre-
quencies or in the width of the monostable pulses. Sucheffects result primarily in minute changes in gain
accuracy, and in small output offset voltages.
The fact that this invention eliminates complete
demodulation of the reference channel results in the
total demodulation system requiring little additional
circuitry over what would have been required to demodu-
late the FM data signal without any flutter-reduction
system.
Variations in the characteristics of the feedback
fllter make little overall difference because the filter
is, in fact, _mmon to both the data and reference chan-
nels, In fact, the output filter can be moved to the
pos~tlon of the feedback filter, and the feedback filter
discarded, with no "baseline" signal-to-noise ratio
degradation and with only a slight degradation of system
gain accuracy, and only at high data frequencies and
amplitudes.
~f interest is the effect generated when the
reference frequency is offset from the data center fre-
quency, say from 3000 Hz to 3500 Hz. In such a situation
the system would be capable of producing extremely high
~baseline~ signal-to-noise ratios even with inadequate
filtering of the summation signal by the feedback filter,
say only 45 or 50 db. This is because the inter-
modulation noise products generated at the difference
frequency would fall outside of the bandwidth of a 100 ~z
output filter, and be further attenuated by the output
filter.
~?
.
, . . .. .
~47852
.
The only flutter noise which this system cannot
remove is non-coherent flutter, i.e., tape speed
variations not common to both the data and reference
channels. Such flutter is generated by two factors,
5 namely,
a. dynamic skewing of the tape,
b. non-uniform dynamic stretching of the
tape.
This system is insensitive to tape skew, so long as
1 it doesn't adversely affect playback voltages, because a
- deviation in the angle of the tape as it moves past the
playback head doesn't affect the frequencies of the
signals. Only dYnamic skewing which is proportional to
the rate of change of the skew angle will generate an
lS instantaneous relative frequency error between two
tracks. This factor accounts for over half the residual
flutter noise and is caused by relatively low frequency
(less than 10 Hz) dynamic skewing of the .150 inch wide
cassette tape as it crosses the playback head, since the
simple cassette recorder that was used had but a single
and rather ineffective metal guide at the side of the
head. This effect could be reduced by providing better
guiding of the tape. A small portion of the residual
flutter noise is due to minute amounts of non-coherent
ctretching of the magnetic tape, since it is a non-
homogeneous elastic medium under varying degrees of ten-
sion, The effect generates higher frequency flutter
noise than does dynamic skewing, but even with the thin
and narrow tapes in Philips' cassettes this effect is
still essentially negligible. Of course, residual noise
due to non-coherent flutter could be eliminated by
multiplexing the reference frequency onto the same physi-
cal track as the data channel. However, it would appear
that the minute relative improvement would rarely justify
the increase in circuit complexity and cost.
The removal of flutter with this system is in fact
so complete that the invention can also be used as a
means to test and evaluate the dynamic skewing charac-
~.....
,~ ,.
1~4785Z-31-
teristics of various tape transports with an extreme
degree of accuracy.
Of special importance is the fact that this inven-
tion requires no special amplitude or frequency-dependent
contouring of the recorded or reproduced signals, i.e.,
that no pre-emphasis/de-emphasis or similar
compression/expansion techniques are employed in the
flutter elimination process. Thus the recorder using
this system is capable of reproducing data with a flat
frequency/amplitude transfer characteristic, and is
thereby classified as an instrumentation recorder and not
j as a special purpose, limited-use device.
For this reason, the technique described in this
invention may be applied to virtually any existing
recorded FM data in the world which has been recorded
with a reference channel, to remove flutter amplitude
' noise from old recordings with a degree of accuracy
thought impossible to obtain until now. This invention
will be of vital importance to applications employing
,~ 20 extremely low tape speeds, as the relative effect of the
' flutter becomes very large as the velocity of the tape
approaches zero.
' It is highly probable that the basic circuitry of
~ the invention will eventually be incorporated into a
;~ 25 single integrated circuit, with pins for the external
connection of the input frequencies and the appropriate
filtering and adjustments.
The efficacy of the present invention in reducing
amplitude flutter noise is shown in FIGS. 10, 11 and 12.
' 30 FIG. 10 shows the face of a cathode ray-tube oscilloscope
~` displaying the output signal generated from an inexpen-
sive tape transport recording of a 4 Hz saw-tooth signal
of full-scale amplitude. To the left, is the noticably
distorted output before the above embodiment of the
invention is switched into the system. To the right, the
flutter noise is eliminated.
More important, however, is the output of a tiny
;~
"
r'
., ~
-32-
...
- :~14785~
saw-tooth signal of only one percent of full-scale ampli-
tude as shown in FIG. 11. To the left, the signal is
virtually obliterated by the normal flutter noise
inherent in the tape moving mechanisms. Switching in the
noise reducer in accordance with the above embodiment
eliminates nearly all of the flutter noise as shown to
the right.
!,' , FIG. 12 shows the output of the same signal as in
FIG. 11 but with massive flutter introduced by physical
, 10 interference with the tape transport. Between the two
dashed lines, the invention is switched in, thereby eli-
minating nearly all of the flutter noise even in this
grotesque situation. From this example it should be
clear that even extreme low quality of the tape transport
will have little effect on the quality of the output
signal, other than the effect gross deviations in tape
speed have on the time base accuracy of the output,
visible in FIG. 12 as variations in the frequency of the
saw-tooth waveform.
Acco~dingly, while the invention has been described
with particular reference to specific embodiments thereof
in the interest of complete definiteness, it will be
understood that it may be emobodied in a variety of forms
diverse from those shown and described without departing
from the spirit and scope of the invention as defined by
the following ~laims.
r
,:,
:,
.,
,
i:
...
, .
