Document text
A MULTIPLEXED FM DATA SYSTEM
WITH IDENTICAL CHANNEL CHARACTERISTICS
William P. McGarry
Data-Control Systems, Inc.
Danbury, C onne cticut
Summary: Seven systems each with ten FM constant bandwidth
channels are discussed. The required bandwidth for the seven
systems and the respective maximum deviation are 800 kc, 400 kc,
200 kc, 100 kc, 50 kc, 25 kc, and 12.5 kc; ±16 kc, ±8 kc, ±4 kc,
±2 kc, ±1 kc, ±0.5 kc, and ±0. 125 kc. The adjacent channel inter¬
ference problem is examined as are the problems associated with
maintaining time correlation between the channels of any one system
The results of tests on a record/reproduce sequence for a typical
system are given.
Introduction
The characteristics of the standard IRIG FM subcarrier system
are well known to the industry. In brief, the system consists of
18 FM subcarriers which can be deviated ±7. 5% and the channels
are separated by guard bands which vary from 0. 62 to 0. 95 of band¬
width. This system has been very successful and some measure of
that success can be traced to the inherent good features of a constant
percentage deviation system. From the data handler*s point of view,
the system provides a set of channels which range from 6 cps to
2100 cps. The constant percentage system also has its advantages
from the equipment manufacturer’s point of view. Since the band¬
width bears the same relation to the center frequency regardless of
the value of the center frequency, all designs for oscillators and
frequency determining networks can be normalized. This greatly
aids standardization of designs and confines the development prob¬
lem. In short, this feature permits a manufacturer to come out with
a new line of subcarrier oscillators or discriminators without a chan¬
nel by channel development.
It is doubted that the FM subcarrier telemetry art could be where it
is today had it not been for the ease of manufacturing standardization
that is inherent in the constant percentage deviation system. On the
other hand, the limitations of the constant percentage system are
also straightforward. The foremost limitations arise because every
channel has a different bandwidth and a different delay of intelligence.
This is very restricting for in today’s work, there are many prob¬
lems where it is necessary to telemeter the outputs of many identical
transducers and also it is required to have time correlation between
the data channels. Such data requirements call for systems com¬
prised of channels of constant bandwidth per channel rather than
constant percentage deviation per channel. In constant bandwidth
systems, the percentage deviation varies from a large percent¬
age deviation at the lower frequency channels to a progressively
smaller percentage deviation as the channels go higheT in fre¬
quency. The range of permissible percentage deviation is limited
by the characteristics of the subcarrier oscillators and subcarrier
discriminators. As the channel bandwidth becomes a progressively
smaller percentage deviation of the subcarrier frequency, the drift
and noise of the oscillators and discriminators become a progress¬
ively larger percentage of the channel bandwidth. As the deviation
becomes a larger percentage of the subcarrier, it becomes pro¬
gressively more difficult to maintain modulator and demodulator
linearity and more difficult to filter the data from the subcarrier
at the output of the discriminator. The limits on percentage devia¬
tion are, of course, arbitrary, but at present, DCS runs over the
range from ±4.5% to ±40%. These limits, of course, can be ex¬
ceeded but at a further compromise with either drift, nonlinearity
or subcarrier in the output.
When a broad analogue data bandwidth is available such as with the
modern tape machines which run beyond one megacycle and one has
a requirement for many channels of identical bandwidth and time
correlation, then it is obvious that neither the constant percentage
deviation system nor the constant deviation (bandwidth) system
will suffice. The constant percentage system is not appropriate be¬
cause every channel has a different bandwidth and the constant band¬
width system soon runs into the modulator and/or demodulator
limitations.
To illustrate this problem, consider a requirement for as many
identical data channels as possible, each with 1 kc data bandwidth,
all channels time-correlated and all to be telemetered in 250 kc
bandwidth. Let the requirement for channel interference noise be
such as to necessitate a modulation index of 2.0 for each channel.
This requires a maximum carrier deviation of ±2 kc. Recalling that
the range of percentage deviation permitted by the modulators or de¬
modulators runs from 4. 5% to 40%, then the minimum percentage
deviation of 4. 5% will give the highest permissible channel at 45 kc
±2.0 kc while the largest permitted percentage deviation will set the
lowest channel at 5. 0 kc ±2. 0 kc. From this it is seen that the
spectrum between the 45 kc channel and the upper limit of 250 kc is
precluded by the minimum permitted deviation.
To overcome this basic problem, the technique of Frequency Trans¬
lation ^ has been used successfully. In this technique, groups of
constant bandwidth channels have been heterodyned against a refer¬
ence carrier so as to relocate the channels as the sidebands about
the carrier. For instance, in the example under consideration, the
channels are in the range from 5.0 kc to 45 kc. This spectrum can
be shifted with acceptable degradation by heterodyning the group
against, say, a 100 kc reference carrier. The output from such a
heterodyne process would have the 5.0 to 45 kc spectrum translated
to the bands of 100 kc ±(5. 0 to 45 kc). Since all the required chan¬
nel signals are contained in either sideband, only one need be re¬
corded or transmitted. After translation, the data is located in the
spectrum as a very small percentage deviation, but prior to demodu¬
lation by a subcarrier discriminator, it is detranslated to its original
location in the spectrum. By this technique of translating up in fre¬
quency then detranslating down, the limitations of the subcarrier
- 2 -
oscillators and discriminators have been circumvented.
Ten Channels of Constant Bandwidth-Time Correlated Systems:
The ease of design standardization inherent when the bandwidth of
each channel of a system is a certain percentage of the carrier has
been emphasized. It has also been pointed out that in a given sys¬
tem of constant bandwidth, the normalization feature is not present.
However, closer examination of the general problem shows that all
of this normalization feature need not be lost.
Figure 1 is a chart of standard DCS constant bandwidth data systems.
Each system is alike in that it contains ten constant bandwidth chan¬
nels plus a reference channel. However, each system has a different
maximum deviation starting from ±250 cps and increasing by octaves
up to ±16 kc. The systems are composed of two sets of five channels
which DCS has arbitrarily designated A thru E (not to be confused
with IRIG lettered channels). It should be noted that the percentage
deviation ranges from ±4. 7% to ±16%. This difference in percentage
deviation causes an extra development burden. However, it should
be noted that to go from one system to another is merely a scale
change. This degree of standardization is very important for if
special problems arise which require some range of frequencies not
given by Figure 1, it is not a major design change to scale the entire
set of A thru E channels. A system requiring a scale change is a
minor effort while a system made up of channels with percentage de¬
viations other than those of A thru E requires engineering design and
evaluation. This is normally inefficient for both the manufacturer
and the user.
To describe the make-up of the ten channel system, consider the 2 kc
deviation-100 kc bandwidth system. It is seen that the lowest chan¬
nel of the untranslated set is at 12. 5 kc and the next four are spaced
7.5 kc between the band centers. Since the maximum deviation is
±2 kc, the guard band is 3. 5 kc which, in turn, is 7/8 of the band¬
width. Note in translating that the highest translated channel is ob¬
tained by subtracting the 12. 5 kc from 100 kc. If the lowest channel
had been chosen lower than 12. 5 kc, then the highest channel would
be correspondingly closer to the 100 kc. This is a significant point
and will be discussed further when the problem of recovering the
reference is considered. The highest subcarrier channel of the un¬
translated set is 42.5 kc ±2 kc. This is a ±4.7% deviation channel
which approaches the minimum permitted percentage deviation
value.
Next, consider the translated set. As indicated in Figure 5, the five
subcarrier oscillators are summed to form a multiplex; then the
composite signal is translated to the lower sideband of a reference
carrier. To locate any of the translated A thru E bands, subtract
the original center frequency from the reference carrier value. Note
that the spacing between the lowest of the translated channels and the
highest of the untranslated channels is twice the spacing between the
other channels.
To recover the data after transmission or storage on tape, the upper
set must be returned to its original location in the spectrum prior to
-3-
A
demodulation. This is true for two reasons: As mentioned, the
drift and noise from a discriminator is correspondingly higher
as the band becomes a smaller percentage of the center frequen¬
cy but when using a tape machine, the consideration of wow and
flutter is even more important. Wow and flutter on a machine is
equivalent to a tape speed change which has the effect of changing
all recorded frequencies by the percent change in speed. When the
band is a small percent of the carrier, then the percentage wow and
flutter can become a very large percentage of the total channel band¬
width.
To illustrate, take the 100 kc bandwidth system and consider the two
12.5 kc ±2 kc channels -- one untranslated, one translated. Let
both these channels be recorded on a machine with a ±1% wow and flut¬
ter variation. The center frequency at 12. 5 kc will vary ±0. 125 kc
which is ±3. 1% of the 4 kc bandwidth. However, the channel trans¬
lated from 12. 5 kc to 87.5 kc will vary .875 kc or ±22% of bandwidth.
If one is looking for a ±1% system, then it is obvious that careful
consideration must be given to the compensation of the wow and
flutter. If the translated channel is detranslated against a fixed
reference, the wow and flutter in the channel remains at the unaccept¬
able ±22%. If, however, the channel is detranslated against a refer¬
ence obtained from the tape, then the wow and flutter problem is re¬
duced to reasonable proportions. This is very evident from series
of simple relationships.
Untranslated Channel = 12.5 kc
Translated Channel = (100 -12.5) = 87.5 kc
Reference = 100 kc
Let ±a% wow and flutter speed variation be introduced.
Untranslated Channel = 12.5 (1 ±a)
Translated Channel = (100 -12.5) (1 ±a)
Reference = (100) (1 ±a)
Let the translated channel be detranslated against the reference with
wow and flutter.
Detranslated Channel = (100 -12.5) (1 ±a)-100 (l±a)
= 12.5 (1 ±a)
Since this is identical with the expression for the wow and flutter on
the 12.5 kc channel recorded untranslated, it is possible to use con¬
ventional wow and flutter compensation from a reference discrimin¬
ator to further reduce the effect.
The Characteristics of the Ten Channel S ystem:
The main concern of multiplexing FM subcarriers in a linear system
is adjacent channel interference. To assure oneself that a new FM
subcarrier system has adequate guard bands, one can do three things:
compare the guard band structure to those of a known system, say the
IRIG; compute the interference assuming sine waves modulating the
adjacent channel and knowing the filter structure; and, finally, one
can set up a channel and measure the interfering noise at various re¬
gions across the band. Comparing the ten channel systems' guard
-4-
band structure with IRIG, one finds that the ten channel system
has a guard band which is 0. 87 5 the channel bandwidth, while the
lower IRIG guard band ranges from 0.6 to 1.0 times the channel
bandwidth.
Figures 2, 3, and 4 give the various aspects of measuring adja¬
cent channel interference. Figure 2 shows the RMS noise in a
modulation index =1.0 linear phase channel when the adjacent chan¬
nels were deviated band edge to band edge by various sine waves
whose frequencies ranged from a modulation index of 1.0 to a modu¬
lation index of 4.0. It can be seen that this worst case noise is less
than 2% of a full band edge to band edge signal. Figure 3 shows how
much the noise can be reduced by changing from a linear phase to a
constant amplitude low pass. Figure 4 gives similar data for a
modulation index = 2.0 system. This system is the maximum rec¬
ommended only for the problems when the higher noise can be
tolerated. Such a problem occurs in vibration data studies which
require large bandwidth and accurate time correlation between
channels.
Figure 5 is a ten channel record/playback configuration which
applies to any of the systems shown in Figure 1. The top half of
the figure diagrams the record portion while the bottom half gives
the playback portion. Examining the record portion, the two like
sets of five VCO's form two separate multiplexes -- one designated
sub L, the other sub H. The sub L group is recorded in the lower
sector of the recorder bandwidth and the s\xb H group is translated
to the high end. The reference carrier is provided by a crystal
controlled oscillator which can be part of the system or can be sup¬
plied to the system from the reference of a time code generator.
The translator block in addition to containing the heterodyning modu¬
lator contains the necessary summing amplifiers, plug-in filter and
line driver amplifier. On playback, the ten channels can be played
back simultaneously using ten discriminators or it can be played
back, a set of five channels at a time. The latter method affords a
great saving in that only five discriminators are required.
The playback system is essentially the inverse of the record. The
higher set is detranslated to its original location in the spectrum
from which it is a conventional subcarrier demodulation problem.
Figure 6 is given for a closer examination of the playback system.
The two units of consideration are the reference discriminator and
the detranslator. The reference is filtered by the band pass of the
reference tuning unit. Since there is a time delay associated with
filtering, the wow and flutter components are accordingly delayed.
This delay is equalized by a plug-in delay line inserted between the
tape output and the detranslator. The output of the unit to the dis¬
criminator's input bus can be switched from the front panel of the
detranslator.
Figures 2 through 5 aid in evaluating the adjacent channel interfer¬
ence as function modulation index of the system. These curves de¬
scribe the geometrical design of the system; i.e., the channel to
channel spacing and guard band to data band ratio. In other words,
the curves are really quite independent of the experimental appa¬
ratus and are solely a frequency spacing evaluation.
-5-
?»
*
Figure 7 is a different breed altogether. It is a complete system
test wherein the resultant data embodies many effects any of which
could be improved or worsened by the characteristics of the vari¬
ous black boxes in the system.
Among the parameters of consideration are: harmonic output of
VCO's; generation of intermodulation terms in the translating
modulators; mismatch of time delay of the two-path detranslation
system; poor wow and flutter compensation; the noise of the tape re¬
corder; intermodulation caused by direct record and direct reproduce
amplifiers; and numerous problems controlled by filter design.
The combination of Figures 2 and 7 give a measure of what can be
expected of a ten channel system using today's equipment. It remains,
however, a system problem and almost beyond the art of specifying
each component black box. To emphasize this point, consider the
problem of assigning a signal level to each channel plus the reference.
If too high a signal is assigned to each channel, the total peak signal
will exceed the linear range of the recorder and cause intermodula¬
tion noise. If the signal is too low, the recorder will give a non¬
optimum signal-to-noise ratio from the record/playback sequence.
Figure 7 then is given as a guide of the type of performance that can
be expected from a ten channel with reference recorded on a standard
quality data tape machine.
The combination of this series of data systems and tape recorders
is a natural marriage because just as the tape machine changes speed
by factors of 2, so does the available tuning units required for data
reduction. Consider the problem of recording fast transient data
which is beyond the speed of practical galvanometer oscillographs.
With the Figure 1 array of channels, these signals could modulate
wide band, high frequency subcarrier oscillators --be recorded at
high tape speed --be played back at reduced speed -- then demodu¬
lated by subcarrier discriminators whose output frequency range
would match the frequency range of the oscillograph equipment. For
real time applications, the wide band demodulation equipment is also
available.
Figure 8 is given to show the physical size of the ten channel system,
record and playback. Each occupies 3-1/2" panel height of a stand¬
ard 19" relay rack. A typical configuration of the system sub units
are spelled out at the bottom of Figure 6.
Time Correlation Between Channels:
Systems composed of channels of identical bandwidth are inherently
systems with time-correlated channels. At first glance, this is not
an obvious statement of fact but a short reflection on the definition
of time delay and the characteristics of linear phase filters will bear
it out. Assuming linear phase filters throughout the system and
assuming the same total number of filters in each channel, and since
there is a specific phase shift at filter cut-off, then it follows that the
phase shift per unit bandwidth will be the same for all channels. The
situation where each linear phase channel has the same total phase
shift at the same cut-off is a restatement of equal envelope time de¬
lay defined as the slope of the phase versus frequency curve. The
problem of time correlation then is one of specifying tolerances for
the various filters in the system. The tolerance on the inter-channel
-6-
correlation will be the sum of the accumulated tolerances from the
filters in each channel. We have found from design and production
experience that when 1% components are used, the attenuation at the
nominal cut-off of three-pole and five-pole passive filters can be
mass produced and tested to a ±1/4 db tolerance. This is reflected
as a ±5% variation in true cut-off, which means a ±5% tolerance of
the phase shift at the nominal cut-off. The main sources of time de¬
lay variation are in the VCO output band pass filter and the discrim¬
inator output low pass. The other system filters are either broad
band which have a short delay or are active filters with a tight toler¬
ance such as the discriminator input bandpass filters.
From this information, one could set mass production tolerances
on the phase shift at channel cut-off: The phase shift at channel cut¬
off for a modulation index of 1.0 is ±12%; ±7% for a modulation index
of 2. 0 system and ±5% for a modulation index of 4. 0 system. In
different words, for a system made up of channels with a data cut¬
off of 1 kc, the approximate tolerance on the time correlation between
channels would be:
±25 usees for a Modulation Index = 4 system
±35 usees for a Modulation Index = 2.0 system
±60 usees for a Modulation Index = 1.0 system
A typical calculation is based on a phase shift of 140° at cut-off for
a five-pole linear phase low pass filter. This represents a delay of
x or approximately one-half period of the cut-off frequency,
c
Then with a cut-off of 1 kc, the period is 1000 usee. Then 1/2 x 5%
of 1000 usees = 25 usee.
A similar time-correlation tolerance table could be constructed for
any other data bandwidth. These tolerances are obtainable in pro¬
duction; if tighter tolerances are required, further effort could be
made to equalize the channels. The engineering, evaluation, and
production test effort figuratively goes up exponentially as the toler¬
ances are tightened.
The foregoing discussion concerned itself with the general problem
of time delay. There is a specific problem that was mentioned early
in the paper that needs close examination. That is the problem of
filtering the reference from the played back multiplex. When a car¬
rier is run through a band pass filter, the envelope is delayed. The
delay is inversely proportional to the bandwidth. In a five-pole linear
phase band pass filter, the phase slope is approximately 300°/band-
width. If the adjacent channel is close to the reference, the bandwidth
of the reference filter must be made accordingly narrower. As the
bandwidth is made narrower, the phase slope is progressively steeper
and the time delay progressively longer. To equalize the delay in the
data path with the delay in the reference path, it is necessary to in¬
sert a delay line. As the delay is increased, tolerance problems are
encountered for the uncertainty in delay is given as a percentage of
the total. Then, as the total rises, the uncertainty in the absolute
rises proportionally. For this very practical reason, the lowest
frequency channel of the set to be translated never is permitted to be
as low as the maximum percentage deviation would have it. To
illuminate -- with a 2 kc deviation system, the restriction of no
greater than ±40% deviation would permit a channel at 5 kc ±40%.
This would put the translated channel at (100 -5) kc which would require
-7-
a narrow 100 kc reference filter which would, in turn, require a
very large value of equalizing time delay.
0
*
References:
1. R. A. Runyan, "Frequency Translation of FM
Subcarriers as a Means of Multiplexing Many
High Frequency Data Channels"
i. O. J. Ott and W. P. McGarry, "Frequency
Translation of FM Subcarriers, " National Tele¬
metering Conference, May I960
-8-
CENTER FREQUENCIES OF FM SUBCARRIERS OF 10 CHANNEL
CONSTANT BANDWIDTH SYSTEMS
<u
>
• H
GO
r—H
v
C
c
(TJ
X
U
T3
0>
4->
(0
C
>h
PQ
W
o o o o o o o
"-• N 00 vO N Tf
m x£>
vD
rM
°
f\J
m
u
•
cr
•
00
•
n£)
•
ro
m
o
o
m
r-
rr
oo
c—H
<\3
m
(NJ
H N lO
• • •
00 vO f\] in O O O
CO \U CO vO (M
_ rvj m
oo r-
m r- m
• • • •
I s - ^ oo r- m o o
^ N in ^ rO vO
00
a
G
G
«J
o
A) TJ
> «
•5 15
00
C
(Ti
4-»
g
D
U
«
m vO N in
W I m o h (M m o o
»-« <nj ^ oo r- rf
——_ ZU2.
r- m
oo r- m
• • •
^ oo I s - m o o o
m r>» rf< oo
—< fvj
rf
^ oo r- in
• • • •
m vo m r- m o o
^ n in n
' (NJ
m
(nj m o o o o o
—< (\J Tf 00 vO
^ n m
m <\j m
m vO nJ in o o
^ n m o
c
o
o
o
•
•r -4
m
o
<XJ
4-J
rvj
m o
o
o
o
o
2
• H
>
o
OsJ
00
Q
TJ
I «*
g
rt a;
a )
u
c
u
a;
m
csj m o o o o o
—« rsj m o o o o
^ N ^ 00
u
g
(V
3
cr
<u
ti
0
4->
6
V V
4-» H
QQ a)
>N U
CO CO
o
0)
150 r<
tfS 6
4J O
*> 15
o .2
* >
o a>
a, q
(M
d
►H
fa
.£
LB BAND UB LBE BC UBE
LBE BC UBE EDGE CENTER EDGE
a
6
<
00
"0
C
C
£
0
O
CTJ
PQ
4 )
4-»
o
4 )
rd
* &
>s
pH
0
(tf
CO
0*
Q
o
00
r-
II
• •
m
m
ii
ctJ
X
(X
i
*
«
W
T3
c
0
J
ctJ
PQ
•
4-»
3
"0
Q
a
A
O
4-»
3
ctJ
3
2
0
0
w
•*
h
W
m
w
j
w
2
2
<
X
2
k
O
H
2
U
a
§
2 °
g£
H 9
w
00
C
•H
ctJ
>
4 )
Q
pH
<t*
C
oc
•r-»
CO
00
TJ
C
rtf
PQ
■*->
<u
u
0}
••n
T)
<
II
00
X)
C -
CTJ
CQ W
w
PQ
i*
X
o
H
Z
w
o
»->
p
z
o
co
A
£ „
<
X
PU
CO
eg
it
W
o
&
to
ii
8
2
s
H
£
s
w w
H H
>* p
co h
O CO
• CO
fM <?
II
X
w
X
H
Pi 0
< Z
P
z
<
m
A
W
z
z
w
z
3
o
00
II
8
L.B. BAND U. B. L. B. BAND U. B. L. B. BAND U. B
EDGE CENTER EDGE EDGE CENTER EDGE EDGE CENTER EDGE
ff
1. Record
VCO--Voltage Controlled Oscillator
XLTR- -T ranslator
XO--Crystal Oscillator
DISCR. - -Subcarrier Discriminator
DXLTR- -Detranslator
Modular Assembly-
Input
Data
2. Playback
front panel.
2. Delay line matching Ref.
Bandpass delay is in
Detranslator.
another modular assembly (in
event all ten channels are to be
played at once)
FIGURE 5 - TYPICAL 10 CHANNEL RECORD-PLAYBACK
CONFIGURATION
FIGURE 6
Reference Discriminator
1
Output 8 %p
Discrim¬
inators
From
Tape
Record
Discrim.
Reference
Band-Pa 8 8
Discrim.
Section
Reference Carrier
Det ranslator
Balanced
Modulator
19
i ?
Low-
Pass
Output
Amplifier
Delay
Line
Output
Amplifier
Wow and
Flutter Com
pensation
Signal
Output
Switchable
to either
Lower or
Higher Multi
plex
Lower
Multiplex
A
B
C
D
E
X
X
A
B
C
D
E
L
L
L
L
L
O
L
T
H
H
H
H
H
R
T
3-1/2
1
Equipment
11
10 V. C. O. 's
1 Xltr
1 Xtal Osc.
Playback
19
t I
H
1»
5 Data Discrim.
1 Ref. Discrim.
1 Detranslator
Plus Tuning Units
o
o
s
2
O
l
r-
w
Oi
D
O
Ui
CO
aJ
V
CO
• H
o
Z
0)
3
H
«
oJ
bO
wO
o"^ CO
o
cJ
d>
in
\h
o
m
CM
•
H
o
o
o
in
r-
m
ro
o
o