Errors Resulting from Channel Filters and Adjacent

Channel Crosstalk in DSB/SC Telemetry Systems

Item Type text; Proceedings

Authors Salter, W. E.; Frost, W. O.

Publisher International Foundation for Telemetering

Journal International Telemetering Conference Proceedings

Rights Copyright © International Foundation for Telemetering

Download date 29/06/2020 20:00:23

Link to Item http://hdl.handle.net/10150/606548

 ERRORS RESULTING FROM CHANNEL FILTERS AND
ADJACENT CHANNEL CROSSTALK IN DSB/SC TELEMETRY
SYSTEMS

W. E. SALTER
Sperry-Rand Corporation

W. O. FROST
Marshall Space Flight Center

Summary The waveform distortion resulting from adjacent channel crosstalk and
from amplitude and phase nonlinearity in channel filters limits the minimum channel
spacing, and hence the bandwidth utilization efficiency of a double sideband/suppressed
carrier (DSB/SC) telemetry link. The paper presents results of an analysis defining the
minimum achievable mean-square error when Butterworth filters are used in the DSB
demodulator/demultiplexer. With data inputs consisting of band-limited random signals,
solutions are given for various combinations of data order, filter order, channel spacing,
and filter cut-off. The trade-off between waveform distortion and channel spacing is
illustrated, and optimum locations for the filter cut-off are defined. The irremovable error
based on Weiner optimum filter theory is presented as an interesting basis for
comparison.

Introduction The search for efficient transmission techniques for wideband data such
as shock, vibration, and acoustical measurements has directed attention to frequency
division arrangements of AM suppressed carrier subcarriers. Single sideband/suppressed
carrier (SSB/SC) techniques have been widely used in several large space vehicle
programs since 1962.1 More recently, use of the DSB/SC technique has been proposed
for applications where the practical limitations of SSB/SC offset its advantage in
bandwidth efficiency.

Two essential attributes of a transmission link used for wideband data are radio
frequency (RF) signal-to-noise (SIN) performance and RF bandwidth efficiency. With
FM carrier modulation, both are very sensitive to the spacing of channels in the
frequency baseband, and each is improved by packing the channels more closely
together.

Although waveform fidelity is not a stringent requirement for many wideband data
applications, it is an important attribute for measurements such as shock. Good
waveform fidelity involves attention not only to noise originating from RF sources,
intermodulation, and crosstalk, but also to distortion resulting from amplitude and phase
nonlinearities in channel filters. The channel filters associated with the
demodulator/demultiplexer should eliminate as much of the adjacent channel signals as
possible, while committing minimum distortion on the channel signal of interest. This
involves careful selection of filter order and cut-off frequency.

In principle, errors resulting from crosstalk and channel filters can be made negligible by
increasing channel spacing; but this alternative leads to degradation in RF bandwidth
utilization and S/N performance. Therefore the system user must seek a suitable
compromise between these sources of error and the RF link errors to achieve optimum
performance. This paper presents an analysis of the waveform distortion arising from
adjacent channel crosstalk and channel filter errors in a DSB/SC link where the channel
filters are the familiar Butterworth type. The results are presented in a form which
defines the optimum selection of filter parameters for a specified channel spacing and
identifies the trade-off relation between signal distortion and channel spacing.

System Description The simplified block diagram in Figure i shows the data now path
and identifies the individual elements of a DSB/SC transmission link. The data channel
of a typical ground system includes the channel demodulator and, depending on the
method of implementation, either one or two channel filters. If sinusoidal reference
carriers and ideal product demodulators are used, channel separation can be
accomplished with only a lowpass filter following the demodulator output. In practice,
however, a bandpass filter is sometimes needed at the demodulator input for proper
operation. In either case, for the purpose of analysis, channel filtering can be represented
by an equivalent lowpass filter following the demodulator output.

Since the magnitude of errors resulting from channel filters depends on characteristics of
the input signal, it is difficult to correlate results, using a simple test signal such as a
single sinusoid, to performance with actual data signals. Therefore, a band-limited
random signal with uniform spectral density was selected to simulate the input data. As
shown in Figure 2(a) the single-sided spectrum of the signal is defined as flat from dc to
fo and attenuated at a rate of 6m db/octave above fo. All channel signals are assumed to
be identical in spectral shape, to occupy equal bandwidths, and to be equally spaced in
the baseband.

The signal at the output of the telemetry channel, shown in Figure 2(b) , is assumed to be
a perfect DSB/SC image of the input signal s(t) together with equally spaced images of
adjacent channels. Hence the analysis does not consider distortion caused by channel
filters at the transmitter or errors arising from RF and intermodulation noise. The
demodulation reference carrier phase is assumed identical with that of the carrier at the
modulator, so that the two sidebands add directly. Thus the product modulation and
demodulation are reversing operations on the signal s(t).

The signal si(t) , which results from the second product operation in Figure 1, consists of
s(t) plus superimposed images of the two adjacent channels centered about a frequency
equal to the channel spacing. The one-sided spectrum of si(t) is shown in Figure 2(c) .

interest by a factor of %2. This results because the sidebands of the DSB subcarriers add
Note that the adjacent channel signals have a spectral amplitude less than the signal of

directly while the two adjacent channels (assumed noncoherent) add rms-wise.

Error Calculations A meaningful evaluation of a systems performance requires that a

criterion be established for judging its merits or comparing it with other systems.
Although a criterion based on S/N ratio or dynamic range presents a useful measure of
performance for certain types of analyses, it is not suitable for the measurement of
waveform distortion because it neglects errors caused by nonideal amplitude and phase
characteristics of the channel filters.

A realistic and useful representation for these errors is the mean of the squared difference
between the output signal so(t) and s(t -"), which is the input signal delayed by the
constant time a that minimizes the mean-square difference. This so-called mean-square
error criterion 3 provides results that may be readily summed with mean-square errors
from other sources to produce overall system error.

A mathematical model for the analysis is shown in Figure 3. The input signal s(t) is
summed with n(t), the translated and superimposed images of the two adjacent channels,
to form si(t) with spectrum indicated in Figure 2(c). This signal is passed through a filter
with characteristics H(T) which is the equivalent of the demodulator output lowpass
filter in series with the input bandpass filter (if used) translated to zero frequency. The
result is then compared with s (t - ") , the delayed replica of the input signal; the
difference represents the instantaneous system error.

In accordance with the model depicted in Figure 3, signal distortion introduced into the
data channel is classified as either adjacent channel crosstalk error or channel filter error.
Crosstalk error occurs because portions of the translated adjacent channel spectra are
passed by the channel filter. Channel filter error results from distortion of the data signal
caused by nonideal amplitude and phase characteristics of the channel filter.

To obtain an expression for crosstalk error, we multiply the power density spectrum of
n(t) by *H(T)*2 and integrate over all frequencies. The phase of the interfering
components may be neglected since they are uncorrelated with the desired signal.
Expressed in closed form with H(T) corresponding to the familiar Butterworth filter
function, this gives for the mean-square error resulting from adjacent channel crosstalk,

(1)

where
z = f/fco
fo = data cut-off frequency
fco = filter cut-off frequency
m = data order
n = filter order
$s = channel separation factor

The amplitude response characteristic of the filter is down 3 db at the cut-off frequency
fco and is attenuated at a rate of 6n db/octave outside the passband. The channel
separation factor $s is equal to the channel spacing divided by the data cut-off frequency.

To determine the channel filter error, the random input signal s(t) is represented as an
infinite sum of sinusoids spaced equally in frequency and having relative magnitudes
corresponding to the envelope of the voltage spectrum.4 This produces the following
expression for mean-square error caused by channel filters:

where
x = f/fco
A(x) = amplitude characteristic of the channel filter
N(x) = phase characteristic of the channel filter
"o = time delay factor
Equations (1) and (2) as shown are normalized to the mean-square value of the input
signal s(t). Since the two error waveforms are uncorrelated, their mean-square values
may be surnmed to give the total mean-square error,

(3)

Using a digital computer, solutions to Equations (1) and (2) were obtained for various
combinations of the variables: fco/fo, m, n, and $s. Using reiteration procedures, a value
was selected for the time delay factor "o to minimize the mean-square error, thereby
assuring a true representation of waveform distortion. 5

Results Figure 4 shows how crosstalk error, filter error, and total waveform distortion
varies with filter cut-off frequency for a specific choice of data order, filter order, and
channel spacing. Note that crosstalk error Fct rises steadily as the filter cut-off frequency
is increased while filter error Ff has the reverse relationship. The curve for Fe, the total
waveform distortion obtained by rms summation of Fct and Ff, clearly identifies an
optimum ratio between the filter and data cut-off frequencies. For the parameters
depicted in Figure 4 (m = 3, n = 6, $s = 4) , we see that a filter with cut-off frequency
slightly less than 2 times the data cut-off frequency provides minimum distortion.

The values for rms error shown on the graphs are normalized to rms values of the input
signal. To obtain rms error normalized to peak-to-peak signal which is sometimes
preferred, one should divide the ordinate values by two times the crest factor (peak/rms
ratio) of the input signal. For example, if we assume a crest factor of 3 as is often used
for vibration data, the ordinate values would be divided by 6.

Figures 5, 6, and 7 depict families of curves similar to Figure 4 where each curve
represents a specific value for the channel separation factor $s. From these curves the
minimum achievable distortion for each selection of $s and the filter cut-off frequency
necessary to achieve this minimum is readily identified. One should notice that the
selection of filter cut-off frequency becomes more critical as channel spacing is
decreased. The three figures illustrate results for data order m equal to 2, 3, and 6,
respectively. In each case the filter order n = 6 was used, a selection which may be
justified from the next set of curves.

Figures 8, 9, and 10 illustrate the trade-off between error, channel spacing, and filter
order. Here each point is plotted using the ratio of fco/fo that yields minimum distortion,
thus providing a convenient form for determining $s for a specified rms error. Although
error decreases with higher filter order, it is apparent that negligible improvement results
for n greater than 6.
Irremovable Error The dashed curves shown in Figures 8, 9, and 10 represent the
waveform distortion resulting from a system implemented with an optimum infinite lag
filter. If we apply Parseval’s theorem to the expression for the minimum mean-square
error of an optimum system and use the property of zero covariance between channels,
the irremovable error is given by6

(4)

where

Mmm(T) = power density spectrum of the input data signal

Mnn(T) = power density spectrum of the adjacent channel signals
Fd = rms value of the input data signal

The irremovable error represents a lower bound on system distortion and thus presents
an interesting basis for comparison.

Experimental Results Channel filter distortion resulting from the nonlinear amplitude
and phase characteristics of a sixth-order Butterworth filter was measured using shaped
random noise to simulate third and sixth-order data signals. As shown in Figure 10, the
measured values agree very closely with the corresponding calculated values of Ff.

Conclusions Optimum performance from a DSB/SC transmission link, where the

criterion is minimum mean-square error, requires careful selection of channel
spacing/data bandwidth ratio and channel filter cut-off frequency. A logical design
choice for most applications should be to make the filter and crosstalk error
approximately equal to the expected error originating from RF and intermodulation
noise. Although filters other than the Butterworth design used in this analysis may give
better results, the relationship shown in Figures 8, 9, and 10 between the curve for n = 6
and the dashed curves representing the performance of an optimum infinite lag filter
indicates that one should not expect phenomenal improvement from other filter designs.

REFERENCES

1. H. L. Stiltz, ed., Aerospace Telemetry (Englewood Cliffs, New Jersey, Prentice-

Hall, Inc. , 1966) , pp. 105-166.

2. W. O. Frost, F. H. Emens, and R. Williams, “A Compatible Double

Sideband/Single Sideband/Constant Bandwidth FM Telemetry System for
Wideband Data,” ITC 1966 Proceedings.
3. J. C. Hancock, An Introduction to the Principles of Communication Theory (New
York, McGraw-Hill Book Company, Inc. , 1961), pp. 138-145.

4. M. Schwartz, Information Transmission, Modulation, and Noise (New York,

McGraw-Hill Book Company, Inc., 1959), pp. 294-295.

5. R. S. Simpson and R. C. Houts, “A Definition of Average Time Delay for a Linear

System,” Proceedings of the IEEE, Vol. 55, p. 1733, October 1967.

6. Y. W. Lee, Statistical Theory of Communication.(New York, John Wiley and

Sons, Inc. , 1960), p. 433.

Figure 1 - DSB/SC Transmission Link

Figure 2 - Signal Characteristics

 Figure 3 - System Model

Figure 4 - Illustrative Example of Waveform Distortion

Figure 5 - Waveform Distortion for Second-Order Figure 6 - Waveform Distortion for Sixth-Order
Data and Sixth-Order Filtering Data and Sixth-Order Filtering
Figure 7 - Waveform Distortion for Sixth-Order Figure 8 - Minimum Waveform Distortion
Data and Sixth-Order Filtering for Second-Order Data
Figure 9 - Minimum Waveform Distortion Figure 10 - Minimum Waveform Distortion
for Third-Order Data for Sixth-Order Data
Figure 11 - Channel Filter Distortion for Sixth-Order Filtering