 |
Description  |
|
|
FIELD OF INVENTION
This invention relates to spread spectrum receivers and more particularly
to a method and apparatus for improving the signal acquisition and
tracking of a spread spectrum receiver.
BACKGROUND OF THE INVENTION
Present communication systems currently employ pseudo-noise (PN) spread
spectrum modulation. The purpose of the spread spectrum modulation depends
on the particular application. Such communication systems can be utilized
for purposes of security, anti-jam protection, multiple access capability,
or a safe level of power flux density radiated back to earth. Regardless
of the application, the problems of acquiring and tracking the PN code are
paramount in all these systems. In order to optimize a system, the code is
to be acquired in as short a time as possible.
In communications systems utilizing pseudo-noise sequences or codes in
their signals, it is a requisite that the phase of the receiver's code
sequence be synchronized to that of the transmitted sequence. It is
therefore the purpose of a code tracker within a spread spectrum receiver
to seek out this phase and follow it by providing the appropriately phased
replica of the code to the decoder unit within the spread spectrum
receiver. When the incoming coded signal is correlated with the
appropriately phased replica at the receiver, the information on the
transmitted signal may be read out.
As used herein, a spread spectrum system is one in which the transmitted
signal is spread over a wide frequency band, much wider in fact than the
minimum band width required to transmit the information being sent. Three
general types of modulations produce spread spectrum signals, e.g. (1)
Modulation of a carrier by a digital code sequence whose bit rate is much
higher than the information signal bandwidth. Such systems are called
"direct sequence" modulated systems; (2) Carrier frequency shifting in
discrete increments in a pattern dictated by a code sequence. These are
called "frequency hoppers." The transmitter jumps from frequency to
frequency within some predetermined set, with the order of frequency usage
being determined by a code sequence; and (3) Pulsed-FM or "chirp"
modulation in which a carrier is swept over a wide band during a given
pulse interval.
The term "pseudo-noise" mentioned above refers to a predetermined
apparently random pulse sequence having a recurring period or cycle which
is long compared with a prevailing information or message duration. This
pseudo-random pulse sequence is usually used in a direct sequence system
which, in the biphase phase shift keyed embodiment, involves a carrier
which is 180.degree. phase-shifted in accordance with the output of a
pseudo-random-number code generator. Thus the incoming signal consists of
a sequence of phase transitions from one constant value to another. In one
system, these transitions occur at a fixed frequency, with the period of
the transitions being referred to as one "chip." Of particular interest in
many applications is the binary, pseudo-random-number code which consists
of a pseudo-randomly generated sequence of numbers having a value of +1 or
-1. A pseudo-random-number code is one which is derived from a sequence
which can be generated systematically but which has some of the properties
of a random-number sequence. Pseudo-random codes are well known and are of
practical interest since a receiver which is capable of generating the
pseudo-random-number code sequence can lock onto a pseudo-random-number
code signal which looks to other receivers like noise. At the same time,
spurious signals which may accompany the incoming pseudo-random-number
code signal (as, for example, spurious signals generated by thermal noise
or external interfering signals) will appear as noise to the receiver and
may be rejected by proper filtering techniques. Since pseudo- randomly
coded spread spectrum signals look like noise when received by a
conventional receiver, the class of receivers which detect spread spectrum
pseudo-random-number codes are called pseudo-noise (PN) spread-spectrum
receivers.
U.S. Pat. Nos. 3,305,636; 3,350,644; 3,402,265; 3,439,279; 3,629,505;
3,666,889; 3,852,354; 4,007,330; 4,017,798; 4,039,749; 4,048,563;
4,092,601; 4,122,393; 4,203,071; 4,214,209; and 4,221,005 described
various correlator systems for providing the appropriately phased replica
of the PN code to be used in the decoders of the spread spectrum
receivers.
For parallel-correlator systems, a bank of correlators is fed with the
incoming signal, with each correlator channel being provided with
progressively advanced and retarded versions of a local PN code generator
sequence. This means that each correlator channel is provided with a
differently phased replica of the predetermined pseudo-noise code. Were it
not for noise in the system, the correlator channel having the highest
correlation output would indicate which of the differently phased replicas
is the one which matches the phase of the transmitted code. However,
Doppler shifts, propagation disturbances and interference result in more
than one correlator channel having a high correlation value. In order to
determine which correlator channel is the one identifying the
appropriately phased replica, various delay-locked loop and integration or
averaging techniques are utilized. While all of the prior art signal
acquisition and tracking systems can acquire and lock up to the
appropriately phased replica, excessive lock-up time precludes the use of
these systems where the phase of the incoming signal rapidly changes. This
problem is particularly severe in communication with fast-moving vehicles
such as jet aircraft, rockets, and nonsynchronous satellites.
In the past, fixed weighting systems have been utilized in which each
individual correlator channel is provided with a predetermined weight
depending upon certain a priori considerations, such as slant range to the
transmitting satellite, Doppler shift, and known atmospheric effects. From
this a priori information, it can be ascertained which channel or channels
have a high probability of being those channel or channels associated with
the correctly phased replica. Once having ascertained these channels with
a priori knowledge, their outputs may be given increased weights, whereas
other channels are given decreased weights. However, all fixed weighting
systems suffer from non-adaptive assigning of weights.
In an attempt to decrease signal acquisition time of the fixed weighting
systems, the sharpness of the detector characteristic is adjusted
depending on the noise conditions and the probability distribution of
signal phase. If relatively little knowledge is available with respect to
which probabilities can be calculated, the detector characteristic is
relatively flat to permit signal acquisition by a large number of
correlator channels. This is equivalent to extending the acquisition range
of the receiver by extending the detector response. As more information
becomes available as to the incoming signal, the detector characteristic
is narrowed so as to effectively reduce the number of correlator channels
and thus the receiver range. Change of detector range can be accomplished
by adding or decreasing shift register bits for the shift register used to
generate the phased replicas. Alternatively this can be accomplished by
changing the correlation channel weights making up the nonlinear detector
characteristic. This latter type of system is described in detail in the
aforementioned U.S. Pat. No. 4,203,071 issued May 13, 1980 to W. M.
Bowles, D. B. Cox, Jr., and W. J. Guinon, assigned to the assignee thereof
and incorporated herein by reference. This patent describes a detection
and tracking system which permits rapid acquisition but does not involve
adaptive tracking or combined adaptive acquisition and adaptive tracking
since it relies solely on statistical methods of computing error. In
short, no signal sampling is used to automatically adjust weights.
Moreover, the weights are not adjusted to enhance the weight of a channel
which is established as having a high correlation value.
One of the aforementioned patents, U.S. Pat. No. 4,007,330, describes a
system for accommodating Doppler shifts by a correction system which
selects among three correlator channels according to correlation. Here the
incoming signal is delayed by different amounts and then correlated with a
predetermined signal, with the correlator channel having the peak
correlation identifying the Doppler shift. It will however be appreciated
that this system does not utilize adaptive weighting.
SUMMARY OF THE INVENTION
In order to decrease the signal acquisition time and to simultaneously
decrease the time it takes for a spread spectrum receiver to lock on or
track the incoming signal, a pseudo-noise code-tracking spread spectrum,
receiver, having faster tracking dynamics (or increased noise rejection),
includes an adaptive weighting system for the outputs of parallel-fed
correlator channels. The degree of correlation in each channel is sensed,
and the weighting is automatically set to increase the weight for a
correlator channel exhibiting a high degree of correlation, and to
decrease toward zero the weights on all other channels.
More particularly, this portion of the spread spectrum receiver can be
considered to have a detector characteristic that is determined by the
weights for the individual correlator channels. The weight distribution
for the channels is shaped and determined from a sampled portion of the
input signal which exists after the input signal is correlated with an
advanced or retarded form of the predetermined code. The advanced or
retarded signals constitute phase-shifted replicas.
In order to determine the degree of correlation in each channel and thus
the weights for each of the correlator channels, an additional correlator
is used. In this additional correlation, the output of the prime
correlator for the channel is correlated with a specialized error signal
to provide a signal which when integrated by an integrator in the channel
is used to determine channel weight. The specialized error signal is
composed of a number of error components, each of which is uniquely
associated with the channel from which it is derived. Each additional
correlator thus recognizes and measures only the component in the
specalized error signal that is due to the output of its prime correlator.
If there is such a component, this indicates that the weight is wrong, and
the associated integrator is driven up or down accordingly to produce the
proper weight for the channel.
The specialized error signal fed back to each additional correlator is
generated as the sum of the weighted correlator channel outputs subtracted
from a training signal, in this case the number "+1" representing perfect
correlation. For a "one" correlation in any one of the correlator
channels, and with zeros in the other channels, the mean error signal is
zero, and all weights are in the steady state.
Lock-up is achieved with near-zero outputs for all but one channel, and a
maximum value for the in-phase channel. At regular intervals during
acquisition and during tracking, a determination of signal phase is made.
This is accomplished by matched filtering the array of correlator channel
weights to determine which channel is aligned with the incoming signal. As
successive determinations are made, the weights and replicas are
simultaneously adjusted to center the detector characteristic about the
identified channel. Once an in-phase correlator channel has been
identified, the associated replica of the PN code used in this correlation
channel is supplied to decode the incoming signal.
The use of the adaptive weighting system obviates the need for the
relatively slow delay-locked loop utilized in the prior art and results in
faster acquisition and faster tracking, with a given amount of noise
immunity. This is because the detector characteristic is at first
relatively flat until such time as correlation exists between an incoming
signal and the replica supplied to one of the correlator channels. Upon
such correlation, this channel is given greater weight, thereby
effectively narrowing the detector characteristic. This characteristic is
even further narrowed in a dynamic fashion until one correlator channel is
identified as that having the maximum correlation to the incoming signal.
Upon identification of a particular correlator channel having a high degree
of correlation with the input signal, the associated replica is shifted to
the center channel by phase-delaying the replicas to each of the
correlator channels and shifting the previously adaptively set weights to
the appropriate channels. The tracking thereafter proceeds from a center
correlation channel having equal numbers of advanced and retarded
phase-replica channels to either side.
While the subject adaptive code tracker will be described in conjunction
with direct sequence spread spectrum systems, it will be appreciated that
the type of tracking is adaptable to frequency hopping and chirping
systems. Moreover, while the direct sequence system described hereinafter
is a nonsynchronous system, the subject system can also be utilized to
detect synchronizing signals transmitted in a synchronous system, thereby
to adjust the phase of the local receiver to the phase of the incoming
sync signal.
The advantages of the adaptive code tracker are first that the detector
characteristic can be made quite broad, such that the range of the
receiver is quite wide. Upon acquisition of a signal, the adaptive
weighting system automatically narrows the detector characteristic by
increasing the weight on the channels having the highest correlation
values. This in turn causes a sharpening of the detector characteristic
until one channel is isolated as having a maximum correlation value. At
this point, the tracking loop is locked up, and the replica associated
with the locked-up correlation channel is then supplied to decode the
spread sprectrum signal. Because of the adaptive nature of the subject
code tracker, actual signal statistics need not be computed, which allows
the subject system to operate robustly in a highly variable signal
environment.
The speed of the subject system is established by the relatively short
integration time of the integrators which integrate the outputs of the
additional correlators.
BRIEF DESCRIPTION OF THE DRAWINGS
These and other features of the subject invention will be better understood
in connection with the detailed description taken in conjunction with the
drawings of which:
FIG. 1 is a block diagram of a spread spectrum receiver utilizing the
subject code tracking system;
FIG. 2 is a schematic and waveform diagram illustrating a fixed weighted
delay-locked loop tracking system in which the weights are set by a priori
information;
FIG. 3 is a schematic diagram of the subject code tracking system for
utilization in the spread spectrum receiver of FIG. 1, illustrating
adaptive weight setting through the utilization of an additional
correlation and integration step for each correlator channel;
FIG. 4 is a waveform diagram illustrating the amount of phase shifting
utilized in phase shifting the code replicas for use in the code tracking
system of FIG. 3;
FIG. 5 is a schematic and waveform diagram illustrating a matched filter
for use in the system of FIG. 3; and
FIG. 6 is a schematic diagram of a replacement circuit for use in the
system of FIG. 3 for noncoherent operation of the subject system.
DETAILED DESCRIPTION
Previous methods for achieving synchronization have included so-called
sliding correlators for phase acquisition and either a tau-dither loop or
a delay-locked loop for tracking. The sliding correlator steps the phase
of the local replica of the transmitted code and measures the correlation
at each step between the replica and incoming signal. Acquisition is
indicated when the above measurements exceed some threshold, after which
the tracking loop takes control. A disadvantage of the sliding correlator
is that only one correlator is used to search out the proper code phase.
Thus this is not a system in which correlator channels are fed in
parallel. If a code has a repetition rate of N clock periods and the time
required for each correlation measurement is T, it takes at worst NT
seconds to acquire the code, assuming that no false indications occur due
to excessive noise. Such a false indication requires additional time to
dismiss the false alarm and resume code search.
With respect to systems involving parallel-fed correlator channels, the
combined output of two or more parallel correlators drives a delay-locked
loop, with the correlators being driven by eariler and later versions of
the code sequence. The sum of the outputs of the correlators is used as an
error signal which is used to change the phase of the code replicas to
eventually drive the difference between the replica and the received code
to zero. Other delay-locked loop systems include a so-called extended
range detector of U.S. Pat. No. 4,203,071 which operates much like the
delay-locked loop in the sense that the outputs of a number of correlators
are driven by delayed replicas. However, the extended range detection-type
systems utilize fixed weights in which the outputs of the correlator
channels are weighted according to a priori information depending on the
projected noise variance of the incoming signal, and signal dynamics
induced by line of sight motions.
The tau-dither loop uses a single correlator and derives an error signal by
phase-modulating the replica over a small range which is a fraction of the
code clock period or chip. The correlator output is then demodulated and
averaged. The averaging is in the form of an integration which takes a
considerable amount of time.
The delay-locked loop and tau-dither loop systems have the further drawback
of requiring the initial code phase error to be within one clock period or
chip for subsequent tracking to take place. Also, if tracking errors
become greater than one chip, no error signal obtains and tracking ceases.
The principal disadvantage of the extended range detector is the need to
predict the noise variance statistic from the signal before the weights
can be chosen. It should be noted that in both the delay-locked loop and
the extendedrange detectors the weighting system is tailored through the
use of a priori knowledge to give the detector a predetermined nonlinear
characteristic as opposed to being adaptively set based on incoming
samples of the signal plus noise.
A system for rapidly acquiring and tracking the spread-spectrum signal by
using adaptive weighting is now discussed. Referring now to FIG. 1, the
receiving portion of a spread spectrum system includes an antenna 10 which
receives the spread spectrum signals, the output of which is coupled to a
conventional receiver 12, which amplifies the incoming signal. The output
of receiver 12 is coupled to a correlation unit 14 which is supplied with
a number of phase-delayed replicas of a predetermined code generated by a
code generator 16, the initial phase of which is set by a code set unit
18. A phase-shifter 20 is utilized to shift the phase of the replicas
generated by code generator 16. Initially the phase shift unit 20 is set
to zero.
The output of the code generator 16 is also applied to a replica selection
unit 22 which selects which of the replicas generated by code generator 16
is to be applied to a balanced modulator 24 and thence to a mixer 26, the
input of which is coupled to the output of receiver 12. The other input to
balanced modulator 24 is a signal generator 28 which generates
f.sub.carrier .+-.f.sub.IF. The output of mixer 26 is applied to an IF
bandpass filter 30 which is in turn coupled to a conventional demodulator
32 which demodulates the information on the received signal.
In operation, an incoming signal is simultaneously correlated with each of
the phase-delayed replicas from code generator 16, and as will be
described in connection with FIG. 3, the weights on the correlator channel
outputs are adaptively changed such that increased weights are given to
the outputs of correlator channels having higher degrees of signal
correlation, with reduced weights being given to those channels having
lower correlation values, the weighting for each of the parallel-fed
correlation channels being adaptively determined. Lock-up is achieved when
a particular channel is identified as having a maximum correlation value,
in one embodiment in terms of the weight for a channel exceeding a
predetermined threshold, in another embodiment by matched filtering the
array output.
A matched filter is that filter whose impulse response is the time reverse
of the expected signal waveform. That is, the filter is "matched" to the
incoming signal. The matched filter is the optimal linear filter in that
it maximizes the output signal-to-noise ratio at the arrival time of the
expected signal.
In practice, the sequence of weights across the bank of correlators may be
filtered by a matched filter whose impulse response is the
auto-correlation of the incoming signal. This auto-correlation is usually
known a priori as a property of the incoming signal. As discussed in
connection with FIG. 5, in the case of direct sequence coding the
auto-correlation is triangular.
The occurrence of the matched filter's output maximum identifies the
correlator most likely to be aligned with the desired signal. Lock-up
therefore establishes the in-phase replica, and this identified replica is
selected by replica selection circuit 22 to be applied to balanced
modulator 24, such that when the output of the balanced modulator is
applied to mixer 26, the received signal is decoded, with the information
of the decoded signal then being passed to the IF bandpass filter and
demodulator for demodulating the information on the decoded signal.
It is advantageous to use a matched filter when the signal auto-correlation
is broad enough so that more than one correlator is temporally aligned
with the incoming signal. This would be the case for instance if the local
reference waveforms which feed the correlators are related by small time
shifts, i.e. they are close together with respect to the auto-correlation
of the incoming waveform.
For a PN code, the width of the auto-correlation is the time of one chip of
the PN code. Thus if the required tracking precision is a fractional chip,
it is necessary to space the reference waveforms closer than one chip,
leading to partial correlations in correlators adjacent to the main peak.
In this sense, the matched filter serves to take this signal energy
(partial correlations in adjacent correlators) and compress it into a
single peak.
After lock-up, the phase of the replicas generated by code generator 16 is
shifted such that the channel having been identified as that carrying the
in-phase replica is shifted to the center such that equal numbers of
channels carrying advanced and retarded replicas are on either side
thereof. Concomitantly therewith, the weights existing at lock-up are
shifted to the corresponding channels as the channels are phase-shifted,
such that the detector characteristic existing at lock-up is preserved
during the centering of the in-phase replica channel.
It will be appreciated that if the replicas alone were shifted to center
the in-phase replica, immediately after the moment of shift, the in-phase
replica would be in the center of the bank of correlators, but with
nominally zero weight. This would necessitate another acquisition cycle
where the weight previously associated with the in-phase replica relaxed
to zero, and the center weight rose to its maximum. This transient effect
is avoided if the weights associated with each replica are shifted between
channels along with that replica.
Referring now to FIG. 2, a prior art delay-locked loop system is
illustrated in which the weights for the correlation channels are set by a
fixed weighting system generally indicated at dotted box 40 to include
weighting amplifiers or attenuators 41. A PN code plus noise is fed in
parallel to the inputs of correlation channels defined by correlators 42,
44, 46, and 48. The channels are supplied respectively with phase-shifted
code replicas C.sub.0, C.sub.1, C.sub.2, . . . C.sub.N. These code
replicas are generated by code generator 50 which is controlled by a
voltage-controlled oscillator 52 which is in turn fed from the output of a
loop filter 54, the input of which is the sum of the outputs of the
weighted correlator channels provided by summer 56.
It will be appreciated that the outputs of the indicated channels include a
correlation value multiplied by the associated weight plus the noise in
the channel. As will be seen, the larger the number of channels, the lower
is the signal-to-noise ratio for the system, or the longer it takes the
system to lock up. The output of summer 56 being applied to loop filter 54
integrates out the noise and provides that the voltage-controlled
oscillator (VCO) be tuned in accordance with the error signal provided at
the output of the loop filter. Thus the VCO is tuned to bring the center
correlator into closer phase alignment with the PN coded signal. It will
be appreciated that the loop filter has an exceptionally long integration
time in order to eliminate the effects of noise from the system. The more
correlation channels utilized, the higher the noise value, and therefore
the longer the integration time necessary in order to provide lock-up, or
tracking to a given degree of accuracy.
As illustrated to the right of the diagram of FIG. 2, various fixed
weighting schemes are utilized in tailoring the weights provided by
amplifiers 41. As illustrated by waveform A, a linear weighting system may
be utilized which gives increased weight to the correlation channels
associated with the most advanced or retarded phase-shifted replicas. This
results in an increased weighting signal for correlations at the most
advanced and retarded phase replica positions, which in turn causes the
VCO to shift the system either upwardly or downwardly so that the channel
aligned with the PN code is finally centered in the sense that it is the
center correlation channel or C.sub.N/2. The linear characteristic is not
an optimal characteristic, and a discontinuous linear characteristic may
be more useful, such as the one shown as waveform B. Although a linear
shape is more appealing from a control theory standpoint, in the presence
of radio noise it has a higher mean squared error than certain other
shapes. This is due to the large multipliers associated with extreme
weights.
The distribution of waveform C is a preferred weighting system for the
correlator system of FIG. 2. As illustrated by dotted waveform 58, the
correlator may originally be given a relatively flat characteristic so as
to accommodate large initial misalignment. As the system approaches
alignment, the fixed weightings may again be refixed to give the sharper
characteristic illustrated by solid lines 60, which gives increased
weighting to the central channels of the correlator while giving decreased
weighting to the outlying channels. The change of detector characteristic
from that shown by dotted line 58 to that shown by dotted line 60 is
governed by a priori information, and is not adaptive in the present sense
in which signal sampling results in automatic adjustment of the weight
values. Thus all of the weighting techniques described by waveforms A, B,
and C are fixed weighting systems, whether or not they are adjusted after
some time to change the characteristic. In all cases of the prior art
delay-locked or tau-dither systems, the weighting system is fixed by a
priori information and not by any adaptive techniques.
Referring now to FIG. 3 in which like reference characters are used with
respect to FIGS. 1 and 3, correlator 14 is provided with a number of
correlation channels 70, each provided with a primary correlator 72 and an
additional correlator 74. It is the purpose of the primary correlators to
correlate the incoming signal with an appropriately delayed replica of a
predetermined code, with the replicas herein being labeled C.sub.0,
C.sub.1, C.sub.2, . . . C.sub.N. The output of each of correlators 72 is
provided as one input to the associated additional correlator 74, the
other input thereto being an error signal E, the derivation of which will
be described hereinafter. The output of each additional correlator 74 is
integrated by an integration circuit 76, the output of which is the weight
for the associated channel. This weight is multiplied by the correlation
value from the primary correlators at multiplier 78, and the output of the
channels as defined thereby is supplied to a summer 80. The output of the
summer is applied to a differential summer 82 which differentially sums
the output of summer 80 with a training signal, here indicated by the
number "1." The number "1" corresponds to the highest weight in the
system. The output of the differential summer 82 is multiplied by the
constant m which determines the loop gain and thus the rapidity of the
adaptive response. The output of multiplier 84 is the specialized error
signal E which is applied to each of the individual additional
correlators. The PN code is illustrated to the left of the correlator 14
and phase-delayed replicas C.sub.0, C.sub.1, and C.sub.2 are illustrated
in FIG. 4 as being delayed by one chip or the smallest time period between
changes in the code.
In one embodiment, initially with zero weights in all of the channels, the
error signal is high, because the output of summer 80 will be zero. When
this is subtracted from the training signal, i.e., the number 1, E will be
equal to 1, ignoring the gain factor, m. Assuming the outputs of all
correlators 72 have zero average value, and assuming E equals 1, then the
correlation at each additional correlator is zero and the integrators do
not have a signal to integrate. Upon signal correlation in one of the
channels, the correlation values from the corresponding additional
correlators increase. This value is integrated at the associated
integrator, and the weight for this particular channel is increased. With
an increase in weight for the channel, the output of summer 80 increases
which decreases this channel's component of the error signal E. However,
the decrease in correlation value from these correlators does not affect
the other associated integrators, especially the one which has already
integrated up to a predetermined weight value based on a high correlation
in the respective channel. The reason that the other channels are
unaffected is because of the additional correlation in each channel which
causes the channel to respond only to that channel's component in the
error signal.
The weights produced by this system are driven to a maximum in the
correlation channel having the greatest degree of phase alignment with the
incoming PN code, while the weights of the other channels tend toward
zero. This situation is sensed by a detector 90 which establishes the
channel associated with the closest in-phase replica. The identity of this
channel is ascertained in one embodiment by a matched filter to be
described hereinafter, with lock-up being achieved when the weights of all
but one channel are at a minimum, and the output from the in-phase channel
(or channels) is at some maximum as determined by the weight for this
channel.
The overall operation of the adaptive weight-forming network attached to
each prime correlator output is to make the weighted sum of the correlator
outputs as close as possible to the training signal, in a least mean
square error sense. The specialized error signal is a measure of the error
between the weighted sum of the prime correlators and the training signal.
The additional correlators associated with each channel measure any
component in the specialized error that is due to the output of that
correlator. Each weight is adjusted by closed-loop action to minimize the
contribution of that channel to the specialized error signal.
To summarize the evolution of the correlator weights, assume all weights
are initially zero at the time the signal plus noise are applied to the
system. Initially the weighted sum of the correlator outputs is zero and
after subtraction from the training signal, the specialized error signal
is +1. When this error signal is multiplied by the primary correlator
outputs in the additional correlators, a zero average value results in all
channels but the one aligned with the signal. The aligned channel,
however, has a constant component in its output due to correlation with a
phase-aligned reference. When this is multiplied by the specialized error
signal, the average value is positive. This input to the integrator
increases the output value for the integrator and thus the weight
associated with the aligned channel. The other channels which are
unaligned with the signal have an output which is uncorrelated with the
specialized error signal and so the outputs of the associated integrators
stay near zero.
In an alternative embodiment, the initial weights may be all set to +1 at
the moment the signal plus noise are applied to the system. In this case,
the specialized error signal has a component proportional to the negative
of the noise coming out of each correlator. When this component is
multiplied by the noise coming out of the correlator, a negative average
value results which drives the output of the integrator down, reducing the
weight for that channel and thus reducing the noise component in the
specialized error signal due to that channel.
The channel which is aligned with the signal also has noise associated with
it, and this tends to depress the weight. However, there is also a steady
component due to signal correlation, and this component in the specialized
error signal tends to raise the weight for the aligned channel.
More specifically, for initial weights equal to "1" for each channel, for
correlation channels not aligned with the signal, the initial prime
correlator value is predominantly noise, e.g. a random fluctuating signal.
This noise (N) is multiplied by the initial channel weight (e.g. "1") and
since 1.times.N=N, N is subtracted from the training signal (e.g. 1-N).
Thus a sign-reversed or negative version (-N) of the correlator output
noise exists as a component of the specialized error signal. When the
prime correlator output N is multiplied at the additional correlator by
the error signal, the result (N.times.-N) is a negative number that tends
to drive the integrator output and channel weight to a lower value. Note
each additional correlator recognizes its own noise component since the
additional correlator mulpilies that negative component (-N) by the
original positive version (N) to obtain a negative average value (-N) to
drive the associated integrator. The further off the weight in the
channel, the higher the additional correlation and the higher the negative
average value; thus the quicker the reduction of the initial "1" weight
to zero.
For a correlator channel aligned with the signal, the initial prime
correlator output is a "one" plus additive noise (1+N). This is multiplied
by the initial channel weight (e.g. "1") and 1+N is subtracted from the
training signal ("1") to leave a negative noise (-N) which initially
depresses this channel weight. Now the weight is no longer "1." When this
lower weight is subtracted from "1," the training signal, the result is a
net positive signal. Thus a steady value equal to the aligned channel
weight plus additive noise exists as a component of the error signal. When
the prime correlator output is multiplied by this error signal, the result
(e.g. 1+noise.times.1-noise) is a positive number that tends to increase
the integrator output and channel weight to a larger number.
The system described is an adaptive system in that the noise in each
channel is measured by virtue of the combination of the additional
correlator and the associated integrator. The integration time of the
integrators is shorter than that associated with delay-locked loops such
that both acquisition and tracking lock-up is established in a very short
period of time without a priori knowledge of the characteristics of the
incoming signal.
Having identified the channel corresponding to that having a replica which
is in phase with the incoming signal, in one embodiment the entire
detector portion of the system is reconfigured, such that the identified
channel becomes the center channel with equal numbers of advanced and
delayed replicas on either side. This is easily accomplished by utilizing
a shift register for the replica generation and by altering the phase of
the clock driving the shift register. In this manner, the in-phase replica
may be shifted so that it appears at the center tap for the register, in
which case equal numbers of advanced and retarded replicas will be
available from the other taps of the shift register. However once the
detector characteristic is centered on the in-phase channel, it is
incumbent upon the system to provide the previously ascertained weights to
the appropriate channels. These previously ascertained weights are those
which prevailed at the time of lock-up and, if digital integrators are
utilized, it is merely a bookkeeping chore in which the addresses of the
weights are altered so that the appropriate weights appear in the
appropriate channels as the channels are being shifted so as to center the
in-phase replica channel. This rearrangement of the weights may be
conventionally accomplished by an N.times.N switch 34 such that any
weight, W.sub.OUT, can be switched into any other weight. In short, all
possible permutations of the weights can be realized by this switch. Thus
the output of the matched filter identifying the in-phase channel not only
controls phase shi | | |
