WO1999023779A2

WO1999023779A2 - Composite trellis system and method

Info

Publication number: WO1999023779A2
Application number: PCT/US1998/023124
Authority: WO
Inventors: Ronald J. Huff
Original assignee: Comtier
Priority date: 1997-10-30
Filing date: 1998-10-29
Publication date: 1999-05-14
Also published as: EP0913950B1; AU1293099A; US5974091A; WO1999023779A3; CA2308211C; EP1130787A2; IL135731A; DE69803351T2; BR9813354A; MXPA00004141A; ES2170989T3; EP0913950A3; DE69803351D1; JP2001522194A; EP0913950A2; CA2308211A1; IL135731A0

Abstract

This invention provides means for processing a received digitally-modulated carrier signal to coherently demodulate (701) or, for some applications, coherently demodulate (704, 705) and decode (703) said signal when the signal's phase has an undesired component and/or to generate successive estimated values of said undesired phase of the received signal. The signal processed is partially represented by a composite trellis diagram (1100) by forming multiple component trellis diagrams (1100) which are phase-displaced versions of a root trellis diagram (204). A demodulated data sequence and/or a sequence of estimated values of the received signal's phase are derived from appropriately structured path history values generated for the multiple states in the composite trellis diagram (1100).

Description

COMPOSITE TRELLIS SYSTEM AND METHOD

Inventor: Ronald J Huff

BACKGROUND OF THE INVENTION

Field of the Invention

This invention relates generally to communication systems wherein modulated carrier signals convey data in digital form and more particularly to the implementation of coherent demodulators and signal phase estimators for such systems which must accommodate undesired signal-phase values and temporal variations thereof. 2. Description of the Related Art

In many carrier-based communication systems the transmit signals used to convey information are of the form

(1) *« ('„) = A_*(U- sin[ω_£tt X + _x(t_u

where r„ represents (relative) transmit time, ω „ represents the transmit radian carrier frequency in radians per second, φ_m ,_x(t, ) represents signal phase variations attributable to any phase modulation effected and A,,(t,_x) represents the signal's amplitude. When the information transmitted is digital in form, the time continuum is generally divided into a succession of contiguous signaling (modulation) intervals and, for each signaling interval, log m (uncoded or coded) data bits select one of in pre- specified options for varying the signal's phase and/or amplitude during the signaling interval. For this invention, the signal's phase may be varied (modulated) to convey digital data using any one of many different modulation schemes including m-ary phase shift keying (PSK). m-ary continuous-phase frequency shift keying (CPFSK) and m-ary CPFSK with modulation and convolutional data encoding effected jointly. The signal's amplitude may vary with time as a consequence of implementing an amplitude modulation scheme such as m-ary amplitude shift keying (ASK) or incidentally, e.g.. due to signal filtering effected. For w-ary ASK, modulation phase component φ_{m lx}(t„) has a constant value considered herein to be zero radian. This invention also applies to several modulation schemes for which signal amplitude and signal phase are varied jointly to convey digital data as for m-ary quadrature amplitude modulation (QAM).

For each of several modulation schemes, the transmitted signal can be modeled equivalently as either a phase modulated signal or an amplitude modulated signal where the signal amplitude is represented by either a real or complex-valued function of time. Such schemes are considered to effect phase modulation herein irrespective of the means used to generate the signals transmitted. Further, for communication systems wherein digital data are transmitted via signaling bursts rather than continuously — as in time division multiple access (TDMA) systems - a transmit signal is modeled by assigning a value of zero to the signal's amplitude whenever signal transmission is disabled.

A transmit signal propagates from its point of origin to one or more receiver locations via one or more communication channels comprised of wire-line, wireless or electronic relay means, or any combination thereof. For a transmit signal representable by Equation 1 , a signal received at a receiver location is of the form (2) s_rx(t) = s{t) + n_∑(t) where t represents (relative) receive time, s(t) is a delayed, attenuated and generally distorted version of the transmit signal,

(3) s{t) = A(t) ■ sin[ω_c - t + φ_m (t) + φ_u (t)] , and n_∑it) represents a sum of noise and undesired signals which exacerbate generation of an exact (delayed) replica of the transmit data at the receiver. In accord with common practice, radian carrier frequency ω_c is assumed to be perfectly known at the receiver; frequency uncertainties that result from imperfect frequency synthesis within transmit, relay and receive subsystems and any doppler shift experienced by the signal in propagating to the receiver are considered to affect the value of signal phase φ„(t): a phase variation that is unintended and undesired. Signal parameter φ,„(t) represents the modulation component of the received signal's phase when phase modulation is effected; otherwise, its value is considered herein to equal zero radian.

For a communication system wherein undesired signal-phase φ„(t) can be made to vary slowly relative to and be distinguishable from phase φ,„(t), the most effective use of signal power can generally be achieved by employing a coherent demodulator to process the received signal. Ideally, a portion of the receiver would generate an exact replica of signal phase φ„(t) and subtract this replica from the phase of the received signal to form a signal which exhibits no undesired phase variations; the latter signal would then be demodulated coherently to generate a nominal replica of the transmitted data stream. For many classes of modulation schemes, prior art provides means for coherently demodulating signals which exhibit no undesired phase variations that can be implemented effectively whenever time intervals spanned by received modulation symbols, modulation intervals, can be accurately determined at the receiver's location.

In practice, undesired signal-phase φ„(t) cannot be replicated exactly at a receiver location, but an estimate of φ„(t) can often be generated with sufficient accuracy to allow the implementation of nearly-ideal coherent demodulation - particularly when m-ary PSK is used to generate signal modulation phase φ_m(t). For ideal (unfiltered) m-ary PSK, the signal's amplitude has a constant value and φ_m(t) assumes any one of m phases that are equally spaced in a 2π radians phase range during each modulation interval. As is well known, an m'^h-order nonlinear device having a received m-ary PSK signal applied to its input port and a bandpass filter that rejects undesired components in the nonlinear device's output signal can generate a signal having a radian center frequency of m • ω_c and phase m - φ_u(t) accompanied by a noise signal that derives from the n^jp) component of the received signal. That is, an rri -order nonlinear device provides a means for distinguishing m ^■ φ_u (t) from φ_m(t) when m- ary PSK is implemented. The nonlinear device's output signal is typically filtered by a relatively narrow-band phase-lock loop to generate a sufficiently accurate estimate of phase m- φ_u(t) , notwithstanding the presence of noise in the received signal. An ambiguous estimate of phase φ„(t) can be determined therefrom by effectively dividing m - φ_u(t) by m. An m-fold ambiguity that derives from the divide operation is generally accommodated by either 1) differentially encoding the transmit data stream and differentially decoding the demodulated data stream in a manner appropriated for the value of m implemented or 2) periodically embedding a priori specified symbols within the transmit date stream which replicate properly in the demodulator only when the appropriate one of m ambiguous phase values is selected. Alternatively, phase-lock loops that incorporate other forms of nonlinearities to distinguish φ„(t) from φ_m(t), e.g., a Costas loop, can be used to provide the desired phase estimate when m-ary PSK modulation is employed.

Phase-lock loops which incorporate nonlinearities as described in the preceding paragraph have attributes which limit system performances - particularly when burst signals are used to convey data as in TDMA systems. As is well known, the time required for a phase-lock loop to achieve lock is statistically distributed and occasionally exceeds the mean acquisition time by a considerable amount

(such an event is referred to as a phase-lock loop hang-up), and the noise component of the received signal occasionally causes a locked loop to cycle slip, i.e., to lose and regain lock in a manner whereby the accumulated phase of the loop's output signal differs from the accumulated phase of the (multiplied) signal being tracked by one or more cycles. Further, as the nonlinearity order — the value of m — is increased, the power spectral density of the noise signal at the output of the nonlinearity increases relative to the level of the desired output signal because a larger number of signal by noise products are generated. Correspondingly, the bandwidth of the signal phase tracking loop must be made increasingly narrow relative to the modulation symbol transmission rate to achieve acceptable steady-state performance. Filtering an m-ary PSK transmit signal to reduce spectral sidelobe levels can also cause tracking-loop performance to degrade.

For a TDMA system wherein estimates of carrier frequency, carrier phase and modulation interval timing are generated at the beginning of each received signal burst by processing preamble symbols (before data demodulation is initiated), making the bandwidths of tracking loops sufficiently small to achieve acceptable steady-state performance can necessitate the use of an unacceptably long burst preamble. For a symbol-synchronous TDMA system, i.e., for a TDMA system wherein signal modulation symbol time bases and carrier frequencies are maintained in accurate synchronism with the time base and frequency, respectively, of a (pulsed-envelope) network timing signal, the transmission of burst preambles can be avoided by converting the received signal bursts into digitally represented sample sequences that are stored in digital memory and implementing "multiple-pass" digital processing. Typically, in-phase and quadrature components of the received signal are integrated over each modulation interval and converted into digital form or, equivalently, appropriate digitized signal sample values are summed to generate the sample sequences processed. Paired in-phase and quadrature sample values are inteφreted as vectors from which vectors which have phase angles equal to in times the sample vector angles are generated and processed digitally over a sliding processing window to form estimated values of m - φ_u(t) modulo 2π radians. These values are divided by m to generate estimated values of φ„(t) which exhibit an m-fold ambiguity. In turn, stored (delayed) sample vectors are rotated in phase as appropriate to correct for the estimated undesired signal-phase values and the phase- corrected sample vectors are demodulated to generate a received data sequence. Of course, such multiple-pass digital processing does not circumvent the fundamental limitations of m'Xrder nonlinear phase estimation methods.

Estimation of an undesired signal-phase component can similarly be accomplished when certain combinations of phase and amplitude modulations are used to convey data, e.g., as for selected forms of QAM. However, important modulation methods are incompatible with the use of nonlinear processing to distinguish undesired signal-phase variation φ„(t) from signal phase modulation φ_m(t) - including most forms of m-ary CPFSK. Circumvention of /Xorder nonlinear processor performance limitations is also of interest.

One alternative approach to distinguishing φ_u(t) from φ,„(t) relies on estimation of φ_m(t) and, effectively, subtraction of the estimated modulation phase from the phase of the received signal to generate a signal which, ideally, would exhibit phase variations φ„(t) only. Estimators which incorporate this alternative approach are often referred to as either decision-directed estimators or data- aided estimators. Their utility derives from the premise that sufficiently accurate estimation of φ„(i) can be accomplished using an estimate of φ_m(t) that is not sufficiently accurate for demodulation puφoses - at least whenever the error in the locally generated estimate of φ_u(t) is large. Proper operation of a decision-directed estimator requires that the modulation means and estimator be implemented jointly in a manner whereby errors made in estimating φ_m(t) only temporarily affect estimator operation and contribute only incrementally to errors made in estimating φ„(t) and demodulating the received signal. This requirement substantially limits the scope of applications for which decision-directed (and data- aided) estimators can be implemented effectively.

SUMMARY OF THE INVENTION

Therefore, one object of the present invention is to provide means for coherently demodulating a received digitally-modulated carrier signal that accommodates an undesired signal-phase component and temporal variations thereof rather than requiring separate estimation of said undesired signal-phase component.

Another object of the invention is to provide means for generating a sequence of estimated values of an undesired signal-phase component for a received digitally-modulated carrier signal.

Another object of the invention is to provide means for coherently demodulating a received digitally-modulated carrier signal comprised of a succession of signal bursts that accommodates an undesired signal-phase component and temporal variations thereof and requires, at most, a few preamble and postamble symbols to be transmitted at the beginning and end of each signal burst, respectively.

Another object of the invention is to provide means for generating a sequence of estimated values of an undesired signal-phase component for a received digitally-modulated carrier signal comprised of a succession of signal bursts of the aforesaid type during the on-time of each signal burst received.

Another object of the invention is to provide means for coherently demodulating multiple received digitally-modulated carrier signals which are each comprised of a succession of non- overlapping signal bursts of the aforesaid type as in TDMA systems

Another object of the invention is to provide means for generating a sequence of estimated values of an undesired signal-phase component for multiple received digitally-modulated earner signals which are each compπsed of a succession of non-overlapping signal bursts of the aforesaid type dunng the on-time of each signal burst received. These and other objects of the invention are provided by a Composite-Trellis Processor (CTP) system and method which processes a received digitally-modulated earner signal that has an undesired phase component as appropriate to effect coherent demodulation and/or generate a sequence of estimated values of said undesired signal-phase component Demodulation and forward error correction

(FEC) decoding are effected jointly for some modulation schemes, e g., for convolutionally encoded m- ary continuous phase frequency shift keying (CPFSK). A CTP implements an augmented version of the

Viterbi algoπthm, or any sub-optimum vanant, denvative or counteφart thereof.

For applications in which accurate estimates of the undesired signal-phase and modulation interval timing can be provided, a Maximum Likelihood Sequence Estimator (MLSE) can be implemented using prior art which reconstructs the transmitted data sequence with a minimal number of erroneous decisions that result from additive receiver noise (when the noise is white and Gaussian distnbuted) Such an estimator correlates the received noisy signal with hypothesized signals generated locally for all possible ways that a trellis diagram can be traversed over successive modulation intervals

(as a consequence of different sequences of data symbols being transmitted) and determines the optimum path through the trellis diagram, i e , the path for which the correlation value is largest Given said optimum path, the sequence of symbols most likely to have been transmitted can easily be determined therefrom

The Viterbi maximum likelihood processing algorithm provides a means for implementing a

MLSE in an iterative manner, that is, by executing one processing cycle following each modulation interval (or following each of a specified number of contiguous modulation intervals for some applications) The processing operations performed principally generate values tor what are often referred to as path metrics (or state metrics) and path histories, one each per to state in a trellis diagram. Each path metric is a cumulative measure of the correlation between the received signal and a locally- generated hypothesized signal that traverses the trellis diagram along a surviving path that terminates at its associated to state. A path history for a surviving path can be represented by a sequence of states traversed by the path, the sequence of modulation symbols that would cause said path to be traversed or both sequences, depending on the specific application. The sequence of modulation symbols for a path can often — but not always, as shown below — be determined from the sequence of states traversed by the path.

For a signal processed in accordance with a CTP system and method, allowed values of signal parameters at a succession of time instants which change as a consequence of the modulation effected are describable abstractly by a trellis diagram designated as a root trellis diagram. A root trellis diagram consists of equal integer numbers of from states and to states and information descriptive of allowed transitions between from states and to states in accord with prior art. Depending on the modulation and data encoding means implemented, a trellis state can correspond to a value for a single signal parameter at an instant in time, or a combination of values which collectively characterize multiple signal parameters and the contents of any memory elements that affect from state to to state transitions, e.g., the contents of a data encoding register, at an instant in time. This invention applies whenever the parameter value(s) associated with a state in the root trellis diagram explicitly or implicitly includes a value for the modulation phase of the transmitted signal, including a modulation phase value of zero radian for m-ary amplitude shift keying (ASK), and the phase of the received signal processed includes undesired component φ_u{t).

The processing operations performed by a CTP are partially described abstractly by a trellis diagram herein referred to as a composite trellis diagram. For this invention, a composite trellis diagram is formed by combining multiple component trellis diagrams which are each a phase- displaced version of a root trellis diagram. For some embodiments of this invention, trellis transitions which interconnect states within different component trellis diagrams, trellis coupling transitions, are incoφorated into the composite trellis diagram to accommodate undesired variations in the received signal's phase. Said trellis coupling transitions typically recur periodically over time intervals which each span a specified number of modulation intervals: a trellis coupling period. The trellis coupling period is set equal to a value which provides an acceptable compromise between demodulator bit error probability (BEP) performance provided and the rate of change in φ„(t) that can be accommodated. That is, the trellis coupling period nominally determines the effective normalized tracking bandwidth of the CTP.

The number of phase states for a root trellis diagram ranges from one for m-ary ASK to thirty-two or more for m-ary CPFSK when the denominator of the signal's modulation index has a large value. Since trellis coupling does not introduce any additional states in the composite trellis diagram, the number of (to or from) phase states in a composite trellis diagram is equal to the number of phase states in the root trellis diagram multiplied by the number of component trellis diagrams within the composite trellis diagram. When the signal phase values associated with the phase states in the composite trellis diagram are equally spaced, successive said signal phase values differ by an amount equal to the phase range spanned by the signal phase values divided by the number of phase states in the composite trellis diagram. The performance degradation attributable to such phase quantization can be made negligibly small by making the number of component trellis diagrams sufficiently large. Generally, the number of component trellis diagrams in the composite trellis diagram is specified to equal a value which provides an acceptable tradeoff between loss in performance due to phase quantization error and implementation complexity.

For this invention, allowed trellis coupling transitions are specified by a trellis coupling diagram. Considerable latitude exists in specifying allowed trellis coupling transitions. Said transitions allow migration to occur "gracefully" between component trellis diagrams as appropriate to accommodate both increases and decreases in the value of φ„(t) for all distinguishable values thereof.

Usually, each trellis coupling transition connects a state in a component trellis diagram to a to state in a "phase-adjacent" component trellis diagram so that phase quantization error is minimized. The phase state in the composite trellis diagram for which the signal phase value is largest is considered to be adjacent to the phase state for which the signal phase value is smallest, and trellis coupling of said states is required to accommodate large-scale variations in φ„(t). Additionally, the trellis coupling transitions are specified so that (i) their presence does not result in path merge distances for the composite trellis diagram being unacceptably smaller than path merge distances for the root trellis diagram; and (ii) the intervals between successive related trellis coupling transitions span sufficient numbers of modulation symbols to result in a BEP performance acceptably close to the BEP performance of an ideal coherent demodulator. Given a priori specified variations in signal parameter values attributable to the modulation effected and a specified composite trellis diagram, a CTP can be implemented through appropriate application of the Viterbi algorithm. This invention provides for implementing a CTP either in rigorous accord with a specified composite trellis diagram or in an expedient manner which provides a preferred tradeoff between processor performance and implementation complexity for many applications. Such alternative methods for processing the received signal are categorized as rigorous processing methods and expedient processing methods, respectively. For either type of processing method, path history values within a CTP can be generated either directly or indirectly with the preferred embodiment for a given application being principally determined by relative processor implementation complexity — not by processor performance.

For direct path history generation, each path history value represents the sequence of states and/or the sequence of modulation symbols for the surviving path that terminates in the to state for which the path history value is generated. Each demodulator output symbol generated is derived from the "oldest" information stored in a path history value. Often, the number of successive modulation symbols for which path sequence information is stored is referred to as the trellis traceback (or chainback) depth. Its value is specified to be sufficiently large to achieve an acceptable tradeoff between BEP performance and implementation complexity for the signal modulation and FEC coding schemes employed.

For indirect path history generation, a "path history" value for a to state often designates either completely or partially the value of the from state for the winning/rom state to to state transition rather than a sequence of symbols or trellis states. For such indirect path history generation, the actual path history for a to state is determined through the implementation of well known prior-art traceback (chainback) means. Trellis traceback is not required when path histories are generated directly.

For both rigorous processing methods and expedient processing methods, the Viterbi algorithm is implemented identically for to states not allowed to be transitioned into via trellis coupling transitions during a processing cycle. Herein, to states of said type are designated as non-coupled testates. The most likely of p allowed transitions into each non-coupled to-state is determined and corresponding path metric and path history values are updated accordingly for each iteration of the Viterbi algorithm. For coupled-to to-states, i.e.. for to states which may be transitioned into via trellis coupling transitions for a processing cycle, the processing operations performed to update path metric values depend on whether a rigorous processing method or an expedient processing method is implemented. When the Viterbi algorithm is implemented rigorously, each trellis coupling transition is specified to connect a coupled-from from-state to a coupled-to to-state. Each path metric value for a coupled-to to- state is updated by 1) calculating a candidate path metric value for each allowed transition into the coupled-to to-state, including trellis coupling transitions. 2) comparing all candidate path metric values for the coupled-to to-state to determine the best path metric value thereof and 3) setting the path metric value for the coupled-to to-state equal to the best path metric value so determined. Path history values for coupled-to to-states are updated in accord with winning trellis transitions, i.e., trellis transitions for which path metric values are best, including winning trellis coupling transitions, using either the direct or indirect path history generation methods described above.

For a rigorous processing method, trellis coupling adds substantially to processor complexity since both matched-filter values and branch-metric values appropriate for the trellis coupling transitions must be generated in addition to generating and comparing additional candidate path metric values. The added complexity that derives from trellis coupling can be reduced markedly without significantly degrading processor performance for most applications by implementing an expedient processing method. For an expedient processing method, each trellis coupling transition is specified to connect a coupled-from w-state to a coupled-to to-state and updated path metric and path history values for coupled-to to-states are selected from appropriate provisional path metric and path history values, respectively, calculated as described above for non-coupled to states, i.e., as if no trellis coupling transitions were allowed.

More specifically, each trellis coupling transition in a trellis coupling diagram appropriate for an expedient processing method typically connects a coupled-to to-state to a phase-adjacent coupled- from to-state to minimize performance degradation attributable to phase quantization. The phase value for a coupled-from to-state may be either smaller or larger than the phase value for a to-state coupled thereto. Furthermore, multiple trellis coupling transitions can concurrently terminate at a coupled-to to- state, e.g., trellis coupling transitions may concurrently connect both to states that are phase adjacent to a to state to the to state. As for rigorous processing methods, the aggregate of trellis coupling transitions that comprise a trellis coupling diagram accommodate both incremental increases and incremental decreases in the undesired signal-phase value for arbitrary values thereof.

For an expedient processing method and each coupled-to to-state, the path metric value for the coupled-to to-state is updated by first comparing aforesaid provisional path metric values for the coupled-to to-state and all coupled-from to-states connected thereto by trellis coupling transitions to determine the best of said provisional path metric values. Then, the path metric value for the coupled-to to-state is updated by setting it equal to the best provisional path metric value so determined. Attendantly, provisional path history values for the applicable states are generated and the path history value for the coupled-to to-state is updated by setting it equal to the provisional path history value that corresponds to the best provisional path metric value.

Output signals generated by a CTP include a demodulated data symbol stream and/or successive estimated values of the undesired signal-phase component. Since a distinct hypothesized value of φ_u(t) is associated with each component trellis diagram, ambiguous estimates of φ_u(t) can be derived from either directly generated or "tracebacked" path history values that represent successive winning component trellis diagrams. The phase ambiguity is often "n_p-fold" where n_p equals the number of phase states in a component trellis diagram. An appropriate traceback depth is mechanized for each output signal generated. That is, the traceback depths applicable to data demodulation and undesired signal-phase estimation can have different values - including zero for signal phase estimation. Note also that the traceback depth(s) required to provide near-ideal CTP performance is (are) minimized by deriving output decisions from successive path histories for to states which have the best overall path metric. Deriving output signals from successive path history values for a pre-specified state can result in unacceptable performance since merge distances for a component trellis diagram can be substantially larger than merge distances for its associated root trellis diagram.

This invention also provides effective means for implementing CTPs which each demodulate and/or estimate the undesired phase of a signal burst or a succession of signal bursts which may originate at one or more transmit terminals, e.g., as in a TDMA system. Specifically, means for initializing and "tailing-off ' operation of a CTP which provide for effective processing of burst signals are provided by this invention.

The features and advantages described in the specification are not all-inclusive, and particularly, many additional features and advantages will be apparent to one of ordinary skill in the art in view of the drawings, specification, and claims hereof. Moreover, it should be noted that the language used in the specification has been principally selected for readability and instructional puφoses, and may not have been selected to delineate or circumscribe the inventive subject matter, resort to the claims being necessary to determine such inventive subject matter.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a block diagram that illustrates a telecommunication system in accordance with the present invention.

Figure 2 A illustrates transmit signal modulation phase variations versus time for a 4-ary, 1REC, CPFSK signal for which the signal modulation index equals one-fourth. Figure 2B is a four-state root trellis diagram for the digital phase modulation depicted in Figure

2A which illustrates state transitions for an arbitrary (single) modulation interval.

Figure 2C is a four-state root trellis diagram for the digital phase modulation depicted in Figure 2A which illustrates state transitions for a succession of modulation intervals.

Figure 3 is a trellis diagram that is a composite of eight four-state component trellis diagrams for a succession of modulation intervals: each component trellis diagram corresponds to a distinct hypothesized value of the undesired component of the received signal's phase and is a phase displaced version of the four-state root trellis diagram illustrated in Figure 2C.

Figure 4A illustrates one embodiment of trellis coupling transitions for the trellis diagram shown in Figure 3 which recur periodically. Figure 4B illustrates a second embodiment of trellis coupling transitions for the trellis diagram shown in Figure 3 which recur periodically.

Figure 4C illustrates a third embodiment of trellis coupling transitions for the trellis diagram shown in Figure 3 which recur periodically.

Figure 5 A is a simplified version of Figure 4A that applies when an expedient processing method is used to generate path metric values.

Figure 5B is a simplified version of Figure 4B that applies when an expedient processing method is used to generate path metric values.

Figure 5C is a simplified version of Figure 4C that applies when an expedient processing method is used to generate path metric values. Figure 6 is a block diagram of a Composite-Trellis Processor System according to one embodiment of the present invention.

Figure 7A is a more detailed block diagram of a portion of the Receiver and Vector Matched Filters illustrated in Figure 6 in accordance with the present invention. Figure 7B is a block diagram of a modified version of the Receiver and Vector Matched Filters illustrated in Figure 7A for one application of the present invention to 4-ary CPFSK.

Figure 8 is a block diagram of a Composite-Trellis Processor for one embodiment of the present invention wherein an expedient processing method is used to generate path metric and associated path history values. Figure 9 depicts data structures for data path history and trellis id path history values for one embodiment of the present invention applicable to 4-ary CPFSK.

Figure 10 is a one-state root trellis diagram for QPSK that applies for an arbitrary (single) modulation interval.

Figure 1 1 is an eight-state composite trellis diagram that illustrates one embodiment of trellis coupling transitions for a Composite-Trellis Processor in accordance with the present invention for QPSK.

Figure 12 is a block diagram of a modified version of the Receiver and Vector Matched Filters illustrated in Figure 7A for one application of the present invention to QPSK.

DETAILED DESCRIPTION OF THE DRAWINGS

The figures depict preferred embodiments of the present invention for puφoses of illustration only. One skilled in the art will readily recognize from the following presentation that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the invention described herein. Figure 1 is a block diagram that illustrates the fundamental elements of a telecommunication system 100 in accordance with the present invention. System 100 includes one or more transmit terminals 101, one or more receive terminals 102 and one or more communication channels 103. The communication channel(s) 103 is (are) comprised of any wireless, wire, optical fiber or electronic means — including one or more passive and/or active (electronic) relay devices — or any combination of such means for propagating each transmit signal from a transmit terminal 101 to one or more receive terminals 102. Any element of the telecommunication system 100 may be either stationary or non- stationary (e.g., moving relative to other elements in the system). Each transmit terminal 101 generates at least one transmit signal using a modulation method appropriate for conveying data in digital form. Each transmit signal is transmitted over at least one communication channel 103 and is received by at least one receive terminal 102. Each transmit signal is expressible by Equation 1 and each signal received by a receive terminal 102 is expressible by Equations 2 and 3. At a receive terminal 102, the desired component of a noisy received signal is nominally of the form of the applicable transmit signal, but its phase is the sum of a desired modulation phase, φM), and an undesired signal-phase that normally varies with time, φ„(t). Receive terminal 102 includes a Composite-Trellis Processor (CTP) 602 (described in reference to Figure 6). A CTP 602 processes a received signal in accordance with the present invention to generate a demodulated data stream and/or estimates of undesired signal-phase Φ t).

For applications in which accurate estimates of the undesired signal-phase and modulation interval timing can be provided, a Maximum Likelihood Sequence Estimator (MLSE) can be implemented using prior art which reconstructs the transmitted data sequence with a minimal number of erroneous decisions that result from additive receiver noise (when the noise is white and Gaussian distributed). Such an estimator conelates the received noisy signal with hypothesized signals generated locally for all possible ways that a trellis diagram can be traversed over successive modulation intervals (as a consequence of different sequences of data symbols being transmitted) and determines the optimum path through the trellis diagram, i.e.. the path for which the correlation value is largest. Given said optimum path, the sequence of symbols most likely to have been transmitted can easily be determined therefrom.

The Viterbi maximum likelihood processing algorithm provides a means for implementing a MLSE in an iterative manner, that is, by executing one processing cycle following each modulation interval (or following each of a specified number of contiguous modulation intervals for some applications). The processing operations performed principally generate values for what are often referred to as path metrics (or state metrics) and path histories, one each per to state in a trellis diagram. Each path metric is a cumulative measure of the correlation between the received signal and a locally- generated hypothesized signal that traverses the trellis diagram along a surviving path that terminates at its associated to state. A path history for a surviving path can be represented by a sequence of states traversed by the path, the sequence of modulation symbols that would cause said path to be traversed or both sequences, depending on the specific application. The sequence of modulation symbols for a path can often ~ but not always, as shown below — be determined from the sequence of states traversed by the path. For a signal processed in accordance with a CTP system and method, allowed values of signal parameters at a succession of time instants which change as a consequence of the modulation effected are describable abstractly by a trellis diagram designated as a root trellis diagram. A root trellis diagram consists of equal integer numbers of from states and to states and information descriptive of allowed transitions between from states and to states in accord with prior art. Depending on the modulation and data encoding means implemented, a trellis state can correspond to a value for a single signal parameter at an instant in time, or a combination of values which collectively characterize multiple signal parameters and the contents of any memory elements that affect from state to to state transitions, e.g., the contents of a data encoding register, at an instant in time. For brevity, that portion of a state's representation which corresponds to a value for the signal's phase is referred to as a phase state, with the understanding that a state is often represented completely as a concatenation of a phase state and one or more other partial state representations, e.g., a code state, an amplitude state and so forth. The time instant at which parameter values applicable to a state are valid is often taken to be either just before or just after the beginning of a modulation interval. A transition between from state and to state parameter values can occur near instantaneously, as for modulation phase values when unfiltered m-ary phase shift keying (PSK) is implemented, or throughout the time interval between the time instants at which from state and to state parameter values are valid, e.g., throughout a modulation interval as for m-ary CPFSK signals. This invention applies whenever the parameter value(s) associated with a state in the root trellis diagram explicitly or implicitly includes a value for the modulation phase of the transmitted signal, including a modulation phase value of zero radian for m-ary amplitude shift keying (ASK), and the phase of the received signal processed includes undesired component φ„(t).

The allowed transitions between states in a root trellis diagram are determined by the modulation means and, when used, data encoding means implemented. For many designs, each from state may transition to either of p predetermined to states, each to state may be transitioned into from either of p predetermined from states, and each demodulator decision represents log ₂p bits of received digital data. When demodulation and convolutional decoding are effected jointly by a CTP, as for coded m-ary CPFSK, the value of p is smaller than the value of m; otherwise p and m have the same value. For each transition in a root trellis diagram, a value for the correct demodulator decision and the manner in which each applicable signal parameter varies with time during the modulation interval are pre-specified (known a priori within a CTP).

The processing operations performed by a CTP are partially described abstractly by a trellis diagram herein referred to as a composite trellis diagram. For this invention, a composite trellis diagram is formed by combining multiple component trellis diagrams which are each a phase- displaced version of a root trellis diagram. For some embodiments of this invention, trellis transitions which interconnect states within different component trellis diagrams, trellis coupling transitions, are incoφorated into the composite trellis diagram to accommodate undesired variations in the received signal's phase. Said trellis coupling transitions typically recur periodically over time intervals which each span a specified number of modulation intervals: a trellis coupling period. The trellis coupling period is set equal to a value which provides an acceptable compromise between demodulator bit error probability (BEP) performance provided and the rate of change in φ„{t) that can be accommodated. That is, the trellis coupling period nominally determines the effective normalized tracking bandwidth of the CTP.

The number of phase states for a root trellis diagram ranges from one for m-ary ASK to thirty-two or more for m-ary CPFSK when the denominator of the signal's modulation index has a large value. Since trellis coupling does not introduce any additional states in the composite trellis diagram, the number of (to or from) phase states in a composite trellis diagram is equal to the number of phase states in the root trellis diagram multiplied by the number of component trellis diagrams within the composite trellis diagram. When the signal phase values associated with the phase states in the composite trellis diagram are equally spaced, successive said signal phase values differ by an amount equal to the phase range spanned by the signal phase values divided by the number of phase states in the composite trellis diagram. The performance degradation attributable to such phase quantization can be made negligibly small by making the number of component trellis diagrams sufficiently large. Generally, the number of component trellis diagrams in the composite trellis diagram is specified to equal a value which provides an acceptable tradeoff between loss in performance due to phase quantization error and implementation complexity. For this invention, allowed trellis coupling transitions are specified by a trellis coupling diagram. Considerable latitude exists in specifying allowed trellis coupling transitions. Said transitions allow migration to occur "gracefully" between component trellis diagrams as appropriate to accommodate both increases and decreases in the value of ø„(t) for all distinguishable values thereof. Usually, each trellis coupling transition connects a state in a component trellis diagram to a to state in a "phase-adjacent" component trellis diagram so that phase quantization error is minimized. The phase state in the composite trellis diagram for which the signal phase value is largest is considered to be adjacent to the phase state for which the signal phase value is smallest, and trellis coupling of said states is required to accommodate large-scale variations in φ_u(t). Additionally, the trellis coupling transitions are specified so that (i) their presence does not result in path merge distances for the composite trellis diagram being unacceptably smaller than path merge distances for the root trellis diagram; and (ii) the intervals between successive related trellis coupling transitions span sufficient numbers of modulation symbols to result in a BEP performance acceptably close to the BEP performance of an ideal coherent demodulator. Given a priori specified variations in signal parameter values attributable to the modulation effected and a specified composite trellis diagram, a CTP can be implemented through appropriate application of the Viterbi algorithm. This invention provides for implementing a CTP either in rigorous accord with a specified composite trellis diagram or in an expedient manner which provides a preferred tradeoff between processor performance and implementation complexity for many applications. Such alternative methods for processing the received signal are categorized as rigorous processing methods and expedient processing methods, respectively. For either type of processing method, path history values within a CTP can be generated either directly or indirectly with the preferred embodiment for a given application being principally determined by relative processor implementation complexity — not by processor performance. For direct path history generation, each path history value represents the sequence of states and/or the sequence of modulation symbols for the surviving path that terminates in the to state for which the path history value is generated. Each demodulator output symbol generated is derived from the "oldest" information stored in a path history value. Often, the number of successive modulation symbols for which path sequence information is stored is referred to as the trellis traceback (or chainback) depth. Its value is specified to be sufficiently large to achieve an acceptable tradeoff between BEP performance and implementation complexity for the signal modulation and FEC coding schemes employed. For direct path history generation, the path history value for a to state is formed by appending information descriptive of the winning/rom state to to state transition to the preceding path history value for the from state for said winning transition and discarding the oldest path history information stored in the path history value when overflow of the memory storage capacity provided would otherwise occur.

For indirect path history generation, a "path history" value for a to state often designates either completely or partially the value of the from state for the winning/rom state to to state transition rather than a sequence of symbols or trellis states. A partial representation of the from state value is often stored when convolutional FEC decoding is implemented to reduce the memory storage capacity required. The "complete" value of a/rom state is typically formed by concatenating the partial value stored in memory with selected bits from a digital representation of the appropriate to state value. For such indirect path history generation, the actual path history for a to state is determined through the implementation of well known prior-art traceback (chainback) means. Specifically, the path history for a to state is determined iteratively by successively constructing/rom state values from progressively older path history values for from states traced back through until the oldest from state value for the path is determined. Trellis traceback is not required when path histories are generated directly.

For both rigorous processing methods and expedient processing methods, the Viterbi algorithm is implemented identically for to states not allowed to be transitioned into via trellis coupling transitions during a processing cycle. Herein, to states of said type are designated as non-coupled to- states. The most likely of p allowed transitions into each non-coupled to-state is determined and corresponding path metric and path history values are updated accordingly for each iteration of the Viterbi algorithm. Specifically, for each non-coupled to-state, p correlation values generally referred to as branch metrics are generated for the p allowed transitions into the to state by correlating the signal received during the most recent modulation interval with p hypothesized signals generated (or effectively generated) locally that have parameter values which vary in accord with the p hypothesized (allowed) transitions. Then, for each non-coupled to-state, p candidate (new) path metric values are generated for the p allowed transitions into the to state by adding each of the p branch metric values to an associated from state path metric value generated during the previous iteration of the algorithm. The p candidate path metric values are then compared to determine the best value thereof and the path metric for the to state is set equal to the best value so determined. For each path metric value updated, an associated path history value (or values) is generated and stored.

For coupled-to to-states, i.e., for to states which may be transitioned into via trellis coupling transitions for a processing cycle, the processing operations performed to update path metric values depend on whether a rigorous processing method or an expedient processing method is implemented. When the Viterbi algorithm is implemented rigorously, each trellis coupling transition is specified to connect a coupled-from from-state to a coupled-to to-state. Each path metric value for a coupled-to to- state is updated by 1) calculating a candidate path metric value for each allowed transition into the coupled-to to-state, including trellis coupling transitions, 2) comparing all candidate path metric values for the coupled-to to-state to determine the best path metric value thereof and 3) setting the path metric value for the coupled-to to-state equal to the best path metric value so determined. Path history values for coupled-to to-states are updated in accord with winning trellis transitions, i.e.. trellis transitions for which path metric values are best, including winning trellis coupling transitions, using either the direct or indirect path history generation methods described above. For a rigorous processing method, trellis coupling adds substantially to processor complexity since both matched-filter values and branch-metric values appropriate for the trellis coupling transitions must be generated in addition to generating and comparing additional candidate path metric values. The added complexity that derives from trellis coupling can be reduced markedly without significantly degrading processor performance for most applications by implementing an expedient processing method. For an expedient processing method, each trellis coupling transition is specified to connect a coupled-from testate to a coupled-to to-state and updated path metric and path history values for coupled-to to-states are selected from appropriate provisional path metric and path history values, respectively, calculated as described above for non-coupled to states, i.e., as if no trellis coupling transitions were allowed. More specifically, each trellis coupling transition in a trellis coupling diagram appropriate for an expedient processing method typically connects a coupled-to to-state to a phase-adjacent coupled- from to-state to minimize performance degradation attributable to phase quantization. The phase value for a coupled-from to-state may be either smaller or larger than the phase value for a to-state coupled thereto. Furthermore, multiple trellis coupling transitions can concurrently terminate at a coupled-to to- state, e.g., trellis coupling transitions may concurrently connect both to states that are phase adjacent to a to state to the to state. As for rigorous processing methods, the aggregate of trellis coupling transitions that comprise a trellis coupling diagram accommodate both incremental increases and incremental decreases in the undesired signal-phase value for arbitrary values thereof.

Embodiments of the present invention for (uncoded) 4-ary, rectangular frequency pulse (lREC), continuous-phase frequency shift keying (CPFSK) are now described. For continuous (non- burst) signaling, the desired component of the received signal is expressible as

(4) s(t) = A - sin[ω_c - t + φ_m(t) + φ_u (t)] where signal amplitude A is constant and, during the n^lh modulation interval, φ_m(t) varies linearly with time:

(5) φ_m(t) T_s < t ≤ n - T_s ,

where h represents the modulation index, T_x represents the symbol duration and a„ e {-3, -1, +1 , +3} for 4-ary signaling. Dibit values C„ determine the values of α_ where, for the so-called direct mapping convention, C„ e {00, 01 , 10, 1 1 } in respective set-value order. The values of C„ are typically generated by Gray encoding transmit symbols (dibits) D„ so that the number of erroneous decoded data bits for a given number of demodulated symbol errors is minimized. That is, D„ is typically formed from binary- valued user transmit data sequence d d₂. d<, ... as D„ = d_t2„._n d₂„ and values thereof are Gray encoded to generate values of C„. For one Gray encoding convention, values of D„ e {00, 01, 1 1, 10} correspond to values of C„ in respective set-value order. When data are transmitted via a succession of signal bursts as in TDMA systems, a preamble symbol that has a pre-specified value, C,„,, is transmitted immediately before symbol C, in each signal burst to provide for effective estimation of D,; similarly, a postamble symbol that has a pre-specified value, C_pm„ is transmitted following transmission of the last Gray- encoded transmit symbol in each signal burst to provide for effective estimation of the last corresponding user data symbol conveyed by the burst.

Figure 2A 200 illustrates temporal variations of modulation phase φ_m(t) 201 in accordance with Equation 5 for 4-ary CPFSK, a modulation index of VA and an example sequence of symbols C„ :

1 1,10,01,00,10,01, -. As shown in Figure 2A the values of φ_m(t) at times « ^• T. relative to the "tilted" reference phase 202 equal either 0, π/2, π, or 3π/2 radians, modulo 2π radians, for all integer values of n. Thus, for the n ^h modulation symbol, allowed transitions from relative phase values (states) at time («- 1 ) • T to relative phase values at time n • T, can be represented abstractly by a four-state root trellis diagram 203 for arbitrary modulation symbol n as shown in Figure 2B. The roof trellis diagram 203 includes four from states 210, four to states 211 and state transitions 212. For the root trellis diagram 203, the relative phase value associated with a state equals the state number multiplied by rc/2 radians.

Figure 2C shows a root trellis diagram 204 that applies for multiple contiguous modulation intervals. This root trellis diagram 204 is formed by replicating the roof trellis diagram 203 shown in Figure 2B for multiple values of n and depicts a succession of state transitions 205 which correspond to transmit modulation phase φ_m(t) 201 shown in Figure 2A.

In accordance with the teaching of the invention, a CTP 602 (described in reference to Figure 6) processes received signals in accordance with a composite trellis diagram comprised of multiple phase- displaced versions of a root trellis diagram (collectively, component trellis diagrams) and any allowed trellis coupling transitions. Figure 3 illustrates a non-uncoupled composite trellis diagram 300 including eight 4-ary CPFSK, h=l/4, four-state component trellis diagrams 305 comprising thirty-two phase states 302 numbered 0 through 31 that span a 2π radians phase range in accordance with one embodiment of the invention. Composite trellis diagram 300 is "non-coupled" meaning that no trellis coupling transitions are specified. The signal phase associated with a phase state 302 equals the phase state number multiplied by 2π/32 radian. When five-bit binary numbers 301 are used to characterize the phase states, the three least significant bits of a state number 303 can be inteφreted as a component- trellis identification number and the two most significant bit of said state number 304 designates the state number 206 of the associated roof trellis diagram 204.

One embodiment of a CTP 602 processes in accordance with the non-coupled composite trellis diagram 300 illustrated in Figure 3 using: (i) eight individual four-state Maximum-Likelihood Processors (MLPs) for which path metric values are initialized and normalized using common control signals; (ii) modules for successively determining the best path metric from among the aggregate of thirty-two path metrics generated by the eight four-state MLPs for each modulation interval; and (iii) modules for generating a demodulated data stream from the path history for the state that has the best path metric and/or successive values that designate the sequence of winning four-state MLPs. Each four-state MLP is assigned an identification number equal to the component trellis identification number 303 for its associated four-state component trellis diagram 305, and each four-state MLP is implemented as appropriate for a hypothesized undesired signal-phase value equal to its identification number 303 multiplied by 2π/32 radian. Thus, estimated values of undesired signal-phase φ_u(t) can be generated by multiplying successive winning four-state MLP identification numbers by 2π/32 radian; such estimates are modulo π/2 radians and, thus, exhibit a four-fold ambiguity.

Since no trellis coupling transitions are specified for the composite trellis diagram 300 shown in Figure 3, composite trellis diagram 300 is used primarily in applications for which undesired signal- phase φ„(t) remains nearly constant over the interval spanned by the signal demodulated, e.g., as in a symbol-synchronous TDMA system wherein a relatively small number of symbols are transmitted during each signal burst. For such applications, initializing and "tailing-off ' operations must be performed to provide for effective demodulation of one or a succession of signal bursts. Methods for implementing such operations are described below following descriptions of means for accommodating relatively slowly changing but large scale variations in undesired signal-phase φ„{t). In accordance with the teaching of this invention, both increases and decreases in the value of φpt) for all values thereof are accommodated by incoφorating trellis coupling transitions in a non- coupled composite trellis diagram 300 in a manner whereby migration between component trellis diagrams 305 can occur. Additionally, the trellis coupling transitions are specified so that: (i) their presence does not result in path merge distances for the composite trellis diagram being unacceptably smaller than path merge distances for the root trellis diagram; and (ii) the intervals between successive related trellis coupling transitions span sufficient numbers of modulation symbols to result in a bit error probability (BEP) performance acceptably close to the performance of an ideal coherent demodulator. Composite trellis diagrams 400a-400c are shown in Figures 4A-4C which include exemplary trellis coupling transitions 401 in addition to trellis transitions 406 for the eight non-coupled component trellis diagrams 305 illustrated in Figure 3. These composite trellis diagrams 400a-400c apply for rigorous processing methods since each trellis coupling transition 401 therein connects a coupled-from from-state 402 to a coupled-to to-state 403. To avoid depicting an unmanageably large number of trellis transitions in a single diagram, Figures 4A-4C each show trellis transitions for a generic component trellis diagram designated by n_c,j and component trellis diagrams phase-adjacent thereto. Further simplification is achieved by depicting trellis transitions which comprise the eight non-coupled component trellis diagrams 305 only for modulation symbol numbers n for which trellis coupling is allowed. For Figure 4A, a complete composite trellis diagram is comprised of eight such diagrams — one for each integer value of n_CHι in the range from zero through seven — and all trellis transitions within component trellis diagrams 305 for all values of modulation symbol number n. For Figures 4B and 4C, n_c,_d equals even integer values zero, two, four and six and, thus, for Figures 4B and 4C, four such diagrams and all trellis transitions within component trellis diagrams 305 for all values of modulation symbol number n comprise a complete composite trellis diagram.

In Figures 4A-4C, each trellis coupling transitions 401 connects a coupled-from from-state 402 in a component trellis diagram 305 to a coupled-to to-state 403 in a phase-adjacent component trellis diagram 305 to minimize performance degradation attributable to phase quantization. Herein, two component trellis diagrams 305 are designated as phase-adjacent when the hypothesized undesired signal-phase values associated therewith are adjacent in an ordered list of said hypothesized values for all component trellis diagrams 305. Further, the component trellis diagram 305 for which the hypothesized undesired signal-phase value is largest is considered to be phase-adjacent to the component trellis diagram 305 for which the hypothesized undesired signal-phase value is smallest. The phase state in the non-coupled composite trellis diagram 300 for which the signal phase value is largest (modulo the phase state value range) is considered to be adjacent to the phase state for which the signal phase value is smallest, and trellis coupling of said states is required to accommodate large-scale variations in φ„(t). Thus, for the example embodiments illustrated in Figures 4A through 4C, state 31 is considered to be adjacent to state 0 and trellis coupling transitions provide phase migration paths both from state 0 to state 31 and from state 31 to state 0.

The trellis coupling transitions 401 specified in Figures 4A through 4C occur periodically over a trellis coupling period that spans N_TCP modulation intervals where N_T p generally ranges from eight to sixteen for 4-ary CPFSK but may be as small as one for Figure 4A and as small as two for Figures 4B and 4C if near-ideal coherent demodulation is not required. Figure 4A shows permitted trellis coupling transitions 401 for one embodiment. In this embodiment, the trellis coupling transitions 401 concurrently accommodate either an increase or a decrease in the value of φ_u(t) relative to any value thereof provided the phase changes not more than plus or minus 2π/32 radian per N_TCP modulation intervals. Thus, the frequency offset accommodated, the frequency-tracking bandwidth, can be made progressively larger by making the value of N_TCP progressively smaller. However, nearly-ideal coherent processing is achieved when the value of N_TCP is on the order of or greater than eight during normal operation.

For selected applications, performance advantages can derive from adaptively controlling the value of N_TCP. For example, for applications that require initial determination (acquisition) of the received signal's frequency, the value of N_TCP is made smaller than normal while the frequency is being acquired to provide a larger than normal acquisition bandwidth. Estimated values of the undesired signal-phase generated by a CTP 602 are processed to estimate successive values of the frequency error that causes φ„(t) to change and these error estimates are used to correct the local estimate of the received signal's frequency. For example, an estimated value for the cumulative change in the undesired component of the signal's phase over several trellis coupling periods is generated for each of a succession of frequency estimation intervals and said values are divided by the frequency estimation interval duration to form a sequence of residual radian frequency enor estimates. In turn, filtered versions of these estimates are used to control the frequency of a receiver local oscillator signal as appropriate to maintain said frequency error at an acceptably small value. Referring still to Figure 4A, from states 402 correspond to the end of the previous modulation interval and fo states 403 correspond to the end of the current modulation interval. Figure 4A shows both transitions 406 for the eight non-coupled component trellis diagrams 305 and trellis coupling transitions 401. The trellis coupling transitions 401 include couple-up transitions 404 and couple-down transitions 405. Couple-up transitions 404 permit trellis coupling from a component trellis diagram to a phase-adjacent, higher-phase component trellis diagram, i.e., coupling to the trellis above. Couple- down transitions 405 permit trellis coupling from a component trellis diagram to a phase-adjacent, lower-phase component trellis diagram, i.e., coupling to the trellis below. In Figure 4A, component trellis diagram transitions 406 are represented as dotted lines, couple-up transitions 404 are represented by solid lines and couple-down transitions 405 are represented by dashed lines. In summary, the trellis coupling transitions illustrated in Figure 4A permit migration between component trellis diagrams at times n = K ■ N_TCP for K e { 1 , 2, 3 .... } where couplings to trellises below via couple-down transitions 405 and trellises above via couple-up transitions 404 are permitted concurrently.

Referring now to Figures 4B and 4C, one-half of the trellis coupling transitions 401 allowed during a trellis coupling period occur at the midpoint of a (any) trellis coupling period and the remaining trellis coupling transitions 401 allowed during a trellis coupling period occur at the end of the trellis coupling period.

Referring now to Figure 4B, at the midpoint of each trellis coupling period that spans N_TCP modulation symbols, each odd-numbered component trellis diagram may couple-down to an even- numbered phase-adjacent component trellis diagram and each even-numbered component trellis diagram may couple-up to an odd-numbered phase-adjacent component trellis diagram. At the end of each trellis coupling period, each odd-numbered component trellis diagram may couple-up to an even- numbered phase-adjacent component trellis diagram and each even-numbered component trellis diagram may couple-down to an odd-numbered phase-adjacent component trellis diagram.

Referring now to Figure 4C, at the midpoint of each trellis coupling period, each component trellis diagram is permitted to couple-up to its phase-adjacent, higher-phase component trellis diagram. At the end of each trellis coupling period, each component trellis diagram is permitted to couple-down to its phase-adjacent, lower-phase component trellis diagram.

The system performances for the trellis coupling configurations illustrated in Figures 4A through 4C are generally comparable for a given value of N_TCP. Similarly, many other trellis coupling configurations are effective, and all such configurations are encompassed by this invention. The preferred trellis coupling configuration for a given application will generally depend on second-order performance differences and relative implementation complexities.

For expedient processing methods, trellis coupling transitions each connect a coupled-from to- state to a coupled-to to-state as noted above. Composite trellis diagrams 500a-500c which include exemplary trellis coupling transitions for expedient processing methods in addition to trellis transitions for the eight non-coupled component trellis diagrams 305 illustrated in Figure 3 are shown in Figures 5A-5C. These figures correspond respectively to Figures 4A-4C. The methods used to avoid depicting an unmanageably large number of trellis transitions in a single diagram for Figures 4A-4C also apply for Figures 5A-5C. In Figure 5A, trellis transitions 501 for the eight non-coupled component trellis diagrams 305 are shown as dotted lines (as is also the case for Figures 5B and 5C). At the end of each trellis coupling period which spans N_TCP modulation symbols, trellis coupling transitions 502, 503 concurrently couple coupled-from to-states 504, 506 to coupled-to to-states 505. Trellis coupling transitions 502 are couple-up transitions and, thus, accommodate an increase in the undesired signal-phase, whereas trellis coupling transitions 503 are couple-down transitions and accommodate a decrease in the undesired signal-phase.

Referring now to Figure 5B, at the midpoint of each trellis coupling period spanning N_TCP modulation symbols, bi-directional trellis coupling transitions 507 connect even numbered to-states 508 to phase-adjacent, higher-phase odd numbered to-states 509. That is, to-states 508 and 509 are each concurrently coupled-from to-states and coupled-to to-states and bi-directional trellis coupling transitions 507 concurrently accommodate either an increase or a decrease in the undesired signal- phase. Similarly, at the end of each trellis coupling period, bi-directional trellis coupling transitions 507 connect even numbered to-states 508 to phase-adjacent, lower-phase odd numbered to-states 510. In Figure 5C, at the midpoint of each trellis coupling period, couple-up trellis coupling transitions 502 connect connect-from to-states 504 to phase-adjacent, higher-phase couple-to to-states 505. At the end of each trellis coupling period, couple-down trellis coupling transitions 503 connect couple-from to-states 506 to phase-adjacent, lower-phase couple-to to-states 505.

Many other trellis coupling configurations consistent with the teachings herein are effective for expedient processing methods and are encompassed by this invention.

A simplified block diagram of a CTP System 600 as used in receive terminal 102 for one embodiment of this invention is illustrated in Figure 6. CTP System 600 includes Receiver and Vector Matched Filters (RVMF) module 601 and Composite Trellis Processor (CTP) 602. The signals applied to the CTP System 600 include a received signal and a receive clock signal, and optionally include signal frequency error estimates, receiver gain error estimates and /or a burst processing control signal. The receive clock signal provides accurate estimates of the time intervals spanned by the received modulation symbols (receiver time synchronization). As discussed above, the residual received-signal frequency error that derives from imperfect generation of various local oscillator signals, relative motion between transmit and receive terminals and/or motion of any element within the communication channel must be maintained suitably small. Frequency error estimates are applied to RVMF module 601 whenever a frequency tracking loop is implemented, and estimates of receiver gain error are applied to the RVMF module 601 whenever externally generated estimates of receiver gain error are used to control receiver gain. The burst processing control signal is provided to the CTP 602 and, for some applications, to RVMF module 601 when signal bursts are processed by the CTP System 600 as in a TDMA system. The value of receive radian carrier frequency ω_c is established by a signal or signals generated within the RVMF module 601.

Implementation of the Viterbi (maximum likelihood processor) algorithm within a CTP 602 includes the generation of branch metrics expressible as: (6) rn_{nss nn} (n) = c_m ^■ j^j s_rx (') ^• *_/,„„ (n_st ,t) ^■ dt

where c,„ is a constant when the transmit signal is phase modulated only, s_rx(t) represents the (noisy) received signal (see Equations 2 through 5) and s_{[ n} \n_st ,t) represents hypothesized signals generated

locally for all trellis transitions applicable for modulation interval n. The fo state in the composite trellis diagram 400a-400c, 500a-500c for which processing is effected is represented by n_m and n„ is a number that identifies the trellis transition for which the branch metric value applies. Since trellis coupling transitions 401 (as specified in Figures 4A-4C) are generally not allowed for every modulation interval and are not required to be specified as occurring concurrently for all states in the composite trellis diagram 400a-400c, 500a-500c, the range of values for n„ can depend on the values of both n and /z„. As noted above, branch metric values are not generated for trellis coupling transitions 401 when an expedient processing method is used to generate path metric values. Thus, for expedient path metric generation, the values of trellis transition identifier n„ in Equation 6 can be set equal to corresponding values of hypothesized received symbols.

Signal s_{t n} [n_st ,t) can often be expressed conveniently as a unit amplitude sinusoid:

(⁷> */,«„ (ⁿst O = ^sin[^ωc ^• ' + Φl,n„ (') + Φk{n,n_st )] where phase terms φ_k (n,n_^, ) are constant-valued over modulation interval n and may or may not vary with n depending on the modulation scheme employed and how time-dependent phase terms φ_{i n} (t)

are defined. When Equation 7 applies, it can easily be shown using Equations 6 and 7 that

(8) "*„,, _,„„ (») = X,„ (») ^• ∞{φi_c (n,n_st )] + Y_q (n) ^• n[φ_lc (n,n_st )]

where

(9) ^Y Ytjι„ ( ("n)) ≡^≡ r ^Cm ■^■ £ ^JX J,) _T PX (0 ^{• sin}[«c ^{• t} + Φl,n„ ( ) • dt and fn T.

(10) Y_q,n„ (n) ≡ _m J V _{n r} s_n(t) cos[ω_c t + φ_{l nn} (t) dt

^J(n-l) T

In accord with conventional practice, Y_{ι n} (n) and Y (n) for a given value of n are

inteφreted as x (in-phase) and y (quadrature) components, respectively, of a vector defined in a nght- hand rectangular coordinate reference frame or, alternatively and equivalently, as the real and imaginary values of a complex number, respectively. By appropriately extending the vector mteφretation. Equation 8 can be inteφreted as the scalar product of two vectors.

As shown in Figure 6, the RVMF module 601 g ^β-^βenerates values for Y ι,.n „„ (n) and γ„ e ,π „„ (n) .

The vector matched filters (such as shown in Figure 7A) included in RVMF module 601 generates paired values of γ_{ι n} (n) and Y (_n) for a specific value of n„. The vector matched filter portion of RVMF module 601 may operate to directly implement Equations 9 and 10 or, as is typically the case, the receiver portion of the RVMF module 601 amplifies, bandpass filters and translates the frequency of (down-converts) the received signal (one or more times) to generate a lower-frequency (intermediate frequency) counteφart of the received signal. In that event, the vector matched filter portion of RVMF module 601 indirectly implements Equations 9 and 10 by suitably processing a down-converted counteφart of the received signal When frequency error estimates are applied to the RVMF module

601 attendant frequency error corrections can be effected by either appropriately changing the frequency of a local oscillator signal used to down-convert the received signal or by adding a phase value that changes linearly with time to all signal phases φ, (t) Either of several means may also be provided to vary the ga of the receiver in accord with either internally or externally generated receiver gam error estimates

The RVMF module 601 shown in Figure 6 can be compπsed of analog circuits, digital circuits, digital signal processing (DSP) software (or firmware) or various combinations thereof. For digital implementations, the received signal (or an intermediate frequency version thereof) is converted into a succession of digital values at an appropriate sample conversion rate by an analog to digital (A/D) converter and all other operations required to implement the RVMF module 601 is performed by digital circuits Alternatively, a baseband counteφart of the received signal comprised of in-phase and quadrature components is generated by analog circuits, said components converted into digital values at an appropriate rate by A/D converters and subsequent processing effected by digital circuits A block diagram that illustrates the latter approach is illustrated in Figure 7A Figure 7A is a functional block diagram of an RVMF module 601 in accordance with the invention. RVMF module 601 includes receiver module 701, control module 702 and vector matched filters 703. Receiver module 701 includes analog mixers 704 coupled to local oscillator signal generator 705 and to received signal s_rx(t). The output of analog mixers 704 are coupled to low pass filters 705. The output of lowpass filters 705 are coupled to analog to digital converters 706. Receiver module 701 downconverts and digitizes received signal s_rx(t). Control module 702 receives a system clock signal and, optionally, a burst processing control signal and generates signals for controlling the operation of receiver module 701 and vector matched filters 703.

In reference to Figure 7A, expressions for analog counteφarts of digital "signals" Y- (η and Y (n) are derivable from Equations 9 and 10 through the use of appropriate trigonometric identities (analog and digital versions of these signals are distinguished by the means used to generate them — not by the nomenclature used herein):

(I D Yι,n„ («) = (')]} " t

(12) Y_q,n„ («) = (/)] - *,- (t) ^• sin[ø_ZιΛff (t)]} ^• dt

where

(13) x_l {t)≡ [c_m s_rx{t) n(ω_c - t)]_hh and

(14) x_q(ή≡ [c_m s_rx(ή cos(ω - t)]_hh .

In Equations 13 and 14, the notation Uermλ designates that only the baseband component of term is retained. With reference to Figure 7A, analog signals x,(t) and x_q(t) are each sampled K times per modulation symbol interval and converted into sequences of digital (quantized) values by the A/D converters represented by X,(k,n) and X_q(k,n) respective where

(15) X, (k,n) x, (t_k ) and

(16) X_q(k,n) = x_v(t_k ) for sampling times t_k given by ( 17) _ti = (n - l) - T, + ⁽ k - - y T. - -^ ; k e {l,2, - , K} .

V 2 ) K

The value specified for K is typically in the range from four to sixteen or more depending on the application.

Digital signals Y_{i n} (n) and Y (π) (see Figure 7A) are expressible approximately as (18) ( ]} and

(19) Y_{q ι} (n) - ∑ |x, (*,/ι) ^• (t* )] - X,- (*,/ι) ^• (t* )]}.

These expressions are inexact because digital circuits generate quantized approximations of sine and cosine values therefor rather than exact values as suggested by the notation used therein. Application of the foregoing teachings of this invention is now made for 4-ary CPFSK, the trellis structure illustrated in Figure 5B and an expedient (path metric) processing method. For this case, trellis transition identifier n„ is assigned values equal to corresponding hypothesized values of modulation symbol C„; that is, C„ is effectively substituted for n„ in applicable equations above. Since the phase values associated with trellis states equal n_sl ■ 2π / 32 radian and are measured relative to the tilted phase reference 202 illustrated in Figure 2A, the phases of the hypothesized received signals generated locally at the end of the X modulation interval must equal _ _n . (?_>Tt/4)+ n_st ■ (27£/32 radian. The complexity of each vector matched filter 703 is minimized by specifying the phase variables for the hypothesized received signals as

3π

(20) φ_[fi(t) ≡ o < t < r_c

T_S 4

Φl,θ(t ^{+ n}s ' T_s) ≡ Φι,o(t) . n_s ≡ {ιa,—}

(21) ι_Λ (t) ': 1 0 < t ≤ T_s

ΦlΛ^{t + n}s - Ts) ^≡ ΦlA^t) • n, e {l,2,^» -}

(22) φ ₂(t) = -φ_l (t) ,

(23) φι ₃(t) = - ι ₀(t) and \ 3κ 2π (24) Φι_c\n,n_st ) = -n - — + n_s — .

A block diagram for one embodiment of the RVMF module 601a for this example application is shown in Figure 7B; this block diagram requires fewer multipliers to implement than does the more general embodiment illustrated in Figure 7A as a consequence of defining the phase variables so that Equations 22 and 23 apply. A partitioning of modules by function for one embodiment of the Composite-Trellis Processor

(CTP) 602 indicated in Figure 6 is depicted in Figure 8. The aggregate of modules implement the Viterbi algorithm and trellis coupling functions in accord with foregoing teaching of this invention. With reference to Figure 8, the Clock and Control Signal Generator 801 generates clock signals for and controls processing operations performed by all other modules in the CTP. Digital values of Y_{t n} (n) and Y_{q n} (n) are processed by the Branch Metric Generator 802 for each value of n to generate branch metric values in accordance with Equations 8 and an expression for φ_lc(n,n_st ) that depends on the

specific application. Path metric values for each state in the composite trellis diagram 400a-400c,

500a-500c are stored in the Path Metric Memory 803. These values are updated for each value of n by the Add, Compare, Select Module 804 and the Trellis Coupling Module 805. (Note that trellis coupling is implemented intrinsically when path metric values are generated rigorously and the Trellis Coupling Module 805 depicted in Figure 8 is not implemented separately for this case.) For each state in the composite trellis diagram 400a-400c, 500a-500c, information directly or indirectly descriptive of winning hypothesized sequence values generated in the process of updating the path metric values is input to the Path History Generator 806. This module generates path history values for each value of n and said values are stored in the Path History Memory 807. The Best Path Metric and State Identification Module 808 determines, for each value of n, the best path metric value and the best state: the state for which the path metric is best. The path history for the best state so determined is processed by the Output Signal Generator 809 to generate the Demodulated Data Stream and/or a sequence of Undesired Signal-Phase Estimates. The best path metric value is typically used to control "normalization" of the path metric values generated by a CTP 602 as appropriate to maintain said values within a range of values accommodated by the design; path metric normalization is typically implemented by subtracting a common value from all path metric values generated during an iteration of the processing algorithm whenever the best path metric value exceeds a pre-specified value.

To this point, the best path metric value has been equated to the largest path metric value. However, for many practical implementations of the Viterbi algorithm, signed digital values for branch metrics are generated and then transformed into unsigned digital values for which small values are "best", and "worst" values may be compressed or otherwise restricted to a range convenient for implementation without significantly compromising performance. Such alternative means apply for this invention.

A more detailed description of the operations described in preceding paragraphs is now given for the 4-ary CPFSK example used to illustrate the teachings of this invention. In the interest of specificity, the path history values are considered to be generated explicitly. Both data sequence and trellis identification sequence path history values are generated for each state in the composite trellis diagram 400a-400c, 500a-500c from which the Demodulated Data Stream and a sequence of Undesired Signal-Phase Estimates are generated respectively. Specifically, for each said trellis state, a data (sequence) path history 902 (as shown in Figure 9) is represented by sixteen two-bit digital numbers and a trellis id(entifιcation) (sequence) path history 901 (as shown in Figure 9) is represented by sixteen three-bit digital numbers. A data path history 902 value for a trellis state specifies the sequence of modulation symbol dibit values for the surviving path that terminates at the trellis state which spans the most recent sixteen iterations of the algorithm. Similarly, the sixteen three-bit numbers that comprise a trellis id path history 901 (aggregate) value, specify the three least significant bits (LSBs) 303 of five- bit state numbers 301 for the sixteen most recent trellis states traversed by the surviving path. (As noted previously, the signal phase modulo π/2 radians associated with a trellis state is equal to the three LSBs of the trellis state's number 303 multiplied by 2π/32 radian.) Note that, for this illustration, the traceback depth equals sixteen for both data demodulation and undesired signal-phase estimation; a data demodulation traceback depth of eight will provide acceptable performance for some applications, and an undesired signal-phase estimation traceback depth of eight or less may be preferred to reduce processing delay for phase estimation and implementation complexity. Further, the appropriate value for the data demodulation traceback depth is significantly larger than sixteen for applications in which demodulation and FEC decoding are performed jointly, e.g., as when convolutionally encoded CPFSK signals are processed. For convenience of implementation, the data path history 902 and trellis id path history 901 values for a trellis state are jointly represented as five sixteen-bit numbers 900 as depicted in Figure 9. For the embodiment illustrated therein, the most significant bits (MSBs) of said five sixteen-bit numbers represent the "oldest" path history information and the LSBs of these numbers represent the most recently determined path history values. The values of the five numbers are updated by multiplying each of them by two — by shifting all bits that comprise the five numbers left by one bit position and discarding the five MSBs — and appending an appropriate LSB to each of the five numbers. A 160 word, sixteen bit per word random access memory (RAM) could be used to store the path history values for all thirty-two states in the composite trellis diagram 400a-400c, 500a-500c.

For the subject embodiment, values of Y^- in) , Y ₀(n) , ..., Y_{t 3}(n) and Y ₃(n) are

processed by the Branch Metric Generator 802 depicted in Figure 8 for each value of n to generate branch metric values in accordance with Equations 8 and 24. For expedient path metric processing, candidate path metnc values for received symbol n are calculated for each state in the composite trellis diagram 500a-500c using the expression

(25) C „_!(,Cj («) = (« - l) _{+ ffl}„_ιp (n) where M_n (n) represents the path metric value for state ti t_mst following completion of processing for received symbol n, and ti_βnst represents the from state for the hypothesized trellis transition. (Note that the values of M_n (n) for all thirty-two values of n„ are set equal to zero before normal operation is initiated.) For the present example embodiment, n^, is given by

(26) n_finsl = {n_st - S - C_n + 32)mod 32 ; C„ e {θ, l,2,3} .

Thus, the candidate path metrics are expressible as

(27) CM_{ns Q}(n) = M_n (n - \) + m, _{ι 0}(n) ,

(28) CM _j (n) = M_k+24)mod32(« - l) + _ffl„_ιp] («) ,

(29) CM„_jp2 W = M_{( +16)mod32}(n - l) + m„_{ff ι2}(n) and

(30) CM_{ns 3} n) = M_{{nst+s)mod 32} n - \) + m_{ns 3}(n)

for hypothesized received symbol values C„ of 0 through 3 respectively. For each state in the composite trellis diagram 500a-500c, four candidate path metric values are generated in accord with Equations 27 through 30, these values are compared to determine the largest value thereof, and the value of a provisional path metric PM_n (n) is set equal to the largest candidate path metric value so determined.

Concurrently, a provisional data path history 902 value and a provisional trellis id path history 901 value are generated for each state in the normal manner. Specifically, values of hypothesized received symbol C„ and ιu_πj/ for which the candidate path metric value is largest are determined, the "old" data path history 902 value for said value of n^_nst is read from memory, left-shifted by one bit position and said value of C„ appended thereto as LSBs (see Figure 9) to generate a provisional data path history value. Similarly, the "old" trellis id path history 901 value for said value of n _nsl is read from

memory, left-shifted by one bit position and the three LSBs of the state number appended thereto as

LSBs (see Figure 9) to generate a provisional trellis id path history value.

When trellis coupling configuration 500b shown in Figure 5B is implemented, for example, trellis coupling is allowed to occur when and only when n equals an integral multiple of N_TCP/2. For all other values of /!, M_n (n) is always set equal to PM „ (n) and new data path history 902 and trellis id path history 901 values are set equal to conesponding provisional values for every value of n„. For each value of n that equals an odd multiple of N_TCPI2, trellis coupling is allowed between each even- numbered state n„ and an associated odd-numbered state (n„ +1) (see Figure 5B). For this case, and for each even-numbered state, the values of PM_n si (n) and are compared to determine

which value is largest. When the value of PM_n (n) is greater than or equal to the value of

PM (n)

and the data path history 902 and trellis id path history 901 values for states n_s, and (n_sl + 1 ) are both set equal to corresponding provisional path history values for state n„ . Otherwise, i.e., when the value of PM_n (n) is less than the value of are both

set equal to the value of PMi ₊Xn) and the data path history 902 and trellis id path history 901

values for states n„ and (n_sl + 1) are both set equal to corresponding provisional path history values for state (/J„ + 1). Similarly, when n equals an even multiple of N_TCP/2 trellis coupling is allowed between each even-numbered state «„ and an associated odd-numbered state, [(«„ +31 ) mod 32] ((n„ - 1 ) except for n„ equal to zero), and the preceding description of operation for n equal to an odd multiple of N_τaJ2 is made applicable by substituting [(«„ +31 ) mod 32] for (π„ + 1 ). The appropriate value for N_TCP generally ranges from about four to thirty-two or more depending on the application.

For the subject example application, generation of the Demodulated Data Stream and a sequence of Undesired Signal-Phase Estimates is initiated after processing has been completed for sixteen received modulation symbols. At that time, the value of n„ for which the thirty-two values of M_n (n) is largest, n„„_u, , is determined. In turn, the oldest hypothesized dibit value stored in the data

path history 902 word for state n„„„, is read from memory, Gray decoded, and the Gray decoded dibit

"output" as two successive data bits in the Demodulated Data Stream. Correspondingly, the oldest three-bit value stored in the trellis id path history 901 words for state n„,_mt is read from memory and output as the first Undesired Signal-Phase Estimate. Subsequently, for each successive value of n, a new value for n„_m, is determined, the oldest path history values for the new value of n_aou, are read from memory, and output signals Demodulated Data Stream and Undesired Signal-Phase Estimates are generated therefrom.

When the received signal processed by a CTP 602 is comprised of a signal burst or a succession of signal bursts which may originate at one or more transmit terminals, e.g., as in a TDMA system, means for initializing and "tailing-off ' CTP operation can be implemented whereby data symbols near the beginnings and ends of signal bursts are processed effectively. For m-ary PSK applications, data symbols are typically differentially encoded prior to transmission and demodulated data symbols are differentially decoded to accommodate an -fold ambiguity in estimating the undesired component of the received signal's phase at the receiver. A (one) differential decoding reference symbol is typically transmitted at the beginning of each signal burst when differential encoding and decoding are implemented whereby the first decoded output symbol for each burst is generated properly within the receiver; typically, no other burst preamble and postamble symbols are required to be transmitted for PSK modulation schemes. When CPFSK is used to convey data, the transmission of a differential decoding reference symbol at the beginning of each signal burst is not required. However, since the merge distances for CPFSK modulations are equal to or greater than two, achievement of a bit error probability (BEP) for bits transmitted at the beginnings and ends of signal bursts comparable to the BEP for mid-burst bits generally requires the transmission of at least one preamble symbol and one postamble symbol in each signal burst. The numbers of preamble and postamble symbols transmitted in a (any) signal burst are each nominally equal to the minimum merge distance for the CPFSK modulation scheme implemented minus one. Additional postamble symbols are transmitted in each CPFSK signal bursts when transmitted data bits (or symbols) are convolutionally encoded as appropriate to implement forward error correction (FEC), as discussed subsequently.

When preamble and/or postamble symbols are present in a received signal burst processed by a CTP, best BEP performance is generally achieved by assigning specific values to specified numbers of preamble and postamble symbols and implementing the Viterbi algorithm so that the winning hypothesized symbol values equal corresponding a priori specified symbol values when preamble and postamble symbols are processed. For example, if a preamble symbol is pre-specified to have a value of "0" then the CTP is implemented so that, for each state in the trellis diagram, the updated path metric value is forced to equal the candidate path metric value for a hypothesized received symbol value of "0" when the preamble symbol is processed. The burst processing control signal indicated in Figure 6 is generated so that the time intervals spanned by the preamble, data and postamble symbols can be determined therefrom. Typically, for each signal burst processed by the CTP 602, transitions in the level of the burst processing control signal occur at the beginning of the first symbol in the burst and at the end of the last data symbol in the burst. For example, the burst processing control signal can have a high level during time intervals spanned by preamble and data symbols processed and a low level otherwise.

When CPFSK and convolutional FEC coding are implemented jointly and signal bursts are transmitted, the state of the FEC encoder is normally set to a pre-specified value - typically "0" - at the beginning of each signal burst. For each signal burst, after the last "user" (information) data bit or symbol conveyed by the burst has been clocked into the FEC encoder shift register (or registers) "flush" bits or symbols that have pre-specified values, e.g., "Os", are successively clocked into the FEC encoder's shift register(s) as if they were user data bits or symbols and the symbols generated by the FEC encoder are transmitted as FEC postamble symbols. The generation of FEC postamble symbols is terminated when the last user data bit or symbol clocked into the FEC encoder can no longer affect the value of an FEC postamble symbol. At the end of each signal burst, one or more additional postamble symbols may be transmitted after the last FEC postamble symbol is transmitted as described above.

For CTPs 602 that jointly demodulate and FEC-decode burst signals, best BEP performance at the beginnings of signal bursts is achieved by initializing the path metric values or otherwise implementing the processor so that the winning from state values for the first iteration of the algorithm are consistent with the a priori known initial state of the FEC encoder (and a priori known preamble symbol value or values when one or more preamble symbols are transmitted). When the last user data bit or symbol transmitted in a signal burst is clocked into the FEC encoder the encoder state has an arbitrary value. For each flush bit or symbol subsequently clocked into the FEC encoder, the number of code states that can be traversed by a transmit trellis path is successively reduced by a factor of two until the code state is known precisely. Best BEP performance at the ends of signal bursts is achieved by implementing the CTP 602 so that only paths through the demodulator trellis that correspond to allowed paths through the transmit trellis are permitted to survive when postamble symbols are processed. (Of course, the ending value of the undesired signal-phase is not known a priori at the receiver location and, thus, the phase state portions of state identification numbers have arbitrary values.) Complete utilization of the a priori information regarding allowed trellis transitions at the ends of bursts within the CTP can result in the BEP for bits at and near the ends of bursts being smaller than the BEP for mid-burst bits.

When signal bursts are processed by a CTP 602, each burst terminates before demodulated values for all data bits conveyed by the burst have been generated and output by the CTP 602. After processing for the last symbol in a signal burst is completed, the trellis state (among the trellis states allowed to be traversed at the end of the burst) for which the path metric value is largest, the last winning state, is determined. Subsequently, demodulated data bits are generated from the path history value for the last winning state. If a capability for processing a succession of closely spaced signal bursts is required, additional means for storing path history information at the end of a burst may need to be provided so that "old" path history information is not overwritten with path history information generated for the new signal burst before the last demodulated data bits are generated for the preceding signal burst. For example, if the direct method is used to generate path history values, the path history (aggregate) value for the last winning state can be transferred to a shift register (or registers) and output signals generated after a burst terminates by appropriately shifting the register contents. Similarly, if the indirect method is used to generate path history values, the sequence of states traced back through after the last winning state for a burst is determined can be stored in a register (or registers) and the stored values can be suitably selected to generate output signals following the end of a burst.

Application of the teachings of this invention to the processing of quaternary phase shift keying (QPSK), also called 4-ary PSK, is now described. A signal generated by an ideal QPSK modulator, an ideal QPSK signal, has a constant envelope and a modulation phase that equals one of four equispaced values in a 2π radians phase range during each modulation symbol interval. Its power spectrum exhibits first side-lobes about thirteen decibels lower in level than the main lobe of the spectrum. Typically, to improve spectrum utilization, the signal transmitted over the communication channel in a system which implements QPSK is a filtered version of an ideal QPSK signal, e.g., Nyquist filters can be used to achieve near-ideal spectrum properties. For clarity of presentation, an embodiment a CTP which processes an ideal QPSK signal is described most extensively herein. Other embodiments of CTPs which process filtered QPSK signals and both unfiltered and filtered m-ary PSK signals for values of m other than four are encompassed by this invention. For ideal QPSK modulation, the desired component of the received signal is expressible as

(31) s(t) = A ^■ sin[ω_c ^■ t + φ_m(t) +

where, during the X modulation symbol interval, φ„,(t) equals one of four constant values:

(32) φ_m(t) = a_n - - , a_n e {0,1,2,3} , (n - l) - T₄ < t < n - T_s .

The values of modulation symbol α„ depend on the sequence of data bits transmitted and the data encoding convention implemented. For specificity, it is assumed that differential encoding and decoding are used to resolve a four-fold phase ambiguity associated with estimating the value of undesired signal-phase φ„(t), and that Gray coding is used to minimize the number of user data bit errors for a given number of demodulated symbol errors. The sequence of user data bits transmitted is represented by d,, d₂, d_}, ... where each d„ has a value of 0 or 1. To provide for the generation of estimated values of d, and d at the receiver location (as required in many systems wherein signal bursts are transmitted), a differential coding reference dibit (a pair of bits) considered here to equal 00 is appended to the user data bit sequence at the transmitter to generate a transmit bit sequence 0, 0, d,, d₂, dj, ... . Since two bits are transmitted via each QPSK modulation symbol, a sequence of dibits, D,, D₂, Dj, ..., is formed from the transmit bit sequence where D, = 0 0, D₂ = d, d₂, D₃ = d_s d₄, ... . For n greater than one, D„ = d_l2n._3> d_(2n.₂₎ . All dibit values D„ are Gray encoded to form dibit values C„ given by (33) C_n = d(2n-3)[^d(2n-3) ® ^d{2n-2)j ^for " > ' ^and ' ⁼ ° ° (^or simply 0) where ® designates the Exclusive OR operation and C„ e {00, 01, 10, 1 1 } or, equivalently, C„ e

{0,1,2,3 } . Differential encoding is effected by adding C„ to the value of α_ for the preceding modulation symbol interval modulo four. Thus, (34) a_n = {c_n + (*(„_,₎ ) mod 4 where C„ e { 0, 1 ,2,3 } and α₀ = 0 .

Either of multiple root trellis diagrams can be used as a basis for implementing a CTP 602 that processes a QPSK signal generated as described in the preceding paragraph. The most obvious applicable roof trellis diagram is comprised of four states — one state for each value of c - for which phase state values are multiples of π/2 radians and state transitions connect each from state to all four fo states. For QPSK, the degradation in BEP performance for a CTP 602 that results from phase quantization error equals a few tenths of a decibel when thirty-two phase states span a 2π radians range and about one-tenth a decibel or less when sixty-four phase states span the same range. For applications where thirty-two phase states provides an appropriate tradeoff between BEP performance loss and implementation complexity, a composite trellis diagram can be formed from eight component trellis diagrams that are each a phase displaced version of the aforesaid four-state roof trellis diagram. However, without compromising CTP 602 perfonnance, a nominal four-fold reduction in implementation complexity derives from representing the signal's phase modulo π/2 radians. For such a representation the root trellis diagram 1000 is comprised of a single state as illustrated in Figure 10. A corresponding eight-state composite trellis diagram 1100 for a CTP 602 is shown in Figure 1 1 for multiple symbol intervals and trellis coupling transitions 1101 of the type indicated in Figure 4B. Since the signal's phase is represented modulo π/2 radians, the signal phase associated with a trellis (phase) state equals the state number multiplied by π/16 radian.

For QPSK, the signals (effectively) generated locally for use in generating vector matched filter signals are expressible by Equation 7 where, when an expedient processing method is used to implement CTP 602,

(35) Φl,_n„ (0 = Φl,n„ ^≡ Φl. _n = °X ^ ^and

(36) Φι_c{n,n_sl ) = φι_c(n_st ) = n_st ^■ — where n_st e {0,1,2,3,4,5,6,7}

for each modulation symbol processed. Note that trellis transition identification number n„is defined as being equal to the of value a„ for the trellis transition in Equation 35. Vector matched filters 1201 (shown in Figure 12) generate output signals Y_{i a} (n) and Y_{q a} in) in accord with either Equations

11, 12 and 35 or Equations 18, 19 and 35. A block diagram for one embodiment of circuits 1200 for which the latter equations apply is shown in Figure 12. Circuit 1200 includes a receiver 1202, control module 1203 and vector matched filters 1201. Other embodiments of vector matched filters 1201 obviously apply; for example, when signal filtering is effected, signals Y_{{ 0}(n) and Y_qβ(n) can be

generated by "root raised cosine" Nyquist filters. Irrespective of how the vector matched filters 1201 are implemented, branch metric values are calculated from the vector matched filter output signals using Equations 8 and 36. Alternatively, since φ_{t a} does not vary with time during a (any) modulation symbol interval, branch metric values can be determined from signals ^_/ o(^w) and Y_{q Q} (H) only using the expression π π π π

(37) m _α (n) = l^₀ (/!) ^• cod „ — I- Π_Γ, — " 2 ^st 16 ⁺ ^.o(") - ^sin, " 2 ^st 16

Candidate path metric values are generated in accord with the Viterbi algorithm as described above for 4-ary CPFSK. In turn, values for provisional path metrics are selected from appropriate candidate path metric values and updated path metric values are selected therefrom as appropriate for the trellis coupling transition configuration implemented.

For the QPSK composite trellis diagram 1100 illustrated in Figure 1 1 , either of four values of α- apply for each allowed combination of from state and to state values. Thus, for indirect data path history generation, values of a„ for winning state transitions must be stored as path history data in addition to winning/rom state values (or values which can be used to determine from state values from applicable to state values). For direct data path history generation, winning hypothesized values of a„ , ά_n n_st ) , are determined and stored for all states in the composite trellis diagram 1100. Then, for one embodiment of this invention, estimated values of Gray encoded dibits C„ , C_n (n_st J , are generated for all trellis states by differentially decoding successive values of ά_n n_sl J ; that is (38) C_n (n_sl ) = [ά_n (n_st ) - ά_{n_ ₎(n_st ) + .

For each trellis state, a data path history 902 value is generated that represents either a succession of C„ [n_st j values or Gray decoded versions thereof for nominally eight to sixteen most recently processed modulation symbols. When a data path history 902 value is comprised of Gray decoded dibit values designated by D_n yn_sl ) , values for bits dι_2n_Xn_st ) and dι_2n_Xn_st ) ^^ai comprise D_n (n_st ) are generated by Gray decoding bits C₍2_n-₃₎ \ⁿ _st ) ^and C₍2_n-_{2) \} ⁿ _st ) that comprise

C„ (n_st ) :

(39) d_{2n_₃₎(n_sl ) = d_{2n_₃₎(n_sl ) , n > \ and

(40) d_(2n_₂₎(n_st ) = c_{2n_₃₎(n_st ) @ c_{2n_₂₎(n_sl ) , n > 1 .

Correspondingly, the bit values that comprise the Demodulated Data Stream are taken directly as the oldest dibit values stored in data path history 902 words for successive winning trellis states, i.e., for successive trellis states having the largest path metric values. When a data path history 902 word is comprised of C_n [n_st ) values, the oldest dibit values stored in data path history 902 words for successive winning trellis states are read from memory and Gray decoded in accord with Equations 39 and 40 to generate bit values that comprise the Demodulated Data Stream. The operations performed to update the values of data path history 902 words when trellis coupling transitions are allowed are the same as described previously for 4-ary CPFSK embodiments of this invention.

For QPSK, Undesired Signal-Phase Estimates can be generated as described previously for 4- ary CPFSK embodiments with each trellis id path history 901 value representing multiple state numbers in their entirety rather than "partial" state numbers - although three-bit values are stored in both cases. However, for some applications of this invention, a preferred tradeoff between performance and implementation complexity can be achieved by effectively setting the trellis traceback depth for generating Undesired Signal-Phase Estimates equal to zero; that is, current values for successive winning trellis states can be used to generate Undesired Signal-Phase Estimates rather than "old" state values stored in trellis id path history 901 words. For either case, the value of the Undesired Signal- Phase Estimates equals the appropriate winning state number multiplied by π/16 radian for the example QPSK embodiment described herein.

The foregoing discussion discloses and describes merely exemplary methods and embodiments of the present invention. As will be understood by those familiar with the art, the invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. Accordingly, the disclosure of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.

What is claimed is:

Claims

1. A trellis processing method for processing a digitally modulated signal having an unknown and undesired phase component over a succession of processing intervals, comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the signal processed, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans a plurality of processing intervals; and a plurality of trellis coupling transitions each coupling a "from state" in a first component trellis diagram to a "to state" in a second component trellis diagram indicating that trellis coupling is permitted between said "from state" and said "to state".

2. The method of claim 1 further comprising: determining path history information for states in said composite trellis diagram over the succession of processing intervals for which said path metric values are generated; and deriving a demodulated symbol sequence from the stored path history information.

3. The method of claim 2 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

4. The method of claim 3 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the "to state" as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time concurrent "from states" coupled therewith.

5. The method of claim 4 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

6. The method of claim 3 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

7. The method of claim 6 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

8. The method of claim 3 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

9. The method of claim 8 wherein the "from state" and the "to state" associated with each trellis coupling transition apply at a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

10. The method of claim 8 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

11. The method of claim 1 further comprising: determining path history information for states in said composite trellis diagram over the succession of processing intervals for which said path metric values are generated; and estimating the unknown and undesired phase component from the stored path history information.

12. The method of claim 1 1 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

13. The method of claim 12 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

14. The method of claim 13 further comprising limiting trellis coupling transitions to couple

"from states" and "to states" that are in phase-adjacent component trellis diagrams.

15. The method of claim 12 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

16. The method of claim 15 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

17. The method of claim 12 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

18. The method of claim 17 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

19. The method of claim 17 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

20. The method of claim 2 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

21. The method of claim 20 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

22. The method of claim 21 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

23. The method of claim 20 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

24. The method of claim 23 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

25. The method of claim 1 1 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

26. The method of claim 25 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

27. The method of claim 26 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

28. The method of claim 25 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

29. The method of claim 28 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

30. The method of claim 2 wherein the digitally modulated signal is a burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts.

31. The method of claim 30 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

32. The method of claim 30 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

33. The method of claim 32 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric: and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

34. The method of claim 1 1 wherein the digitally modulated signal is a burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts.

35. The method of claim 34 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

36. The method of claim 34 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

37. The method of claim 36 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and further comprising, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

38. A trellis processing method for processing a digitally modulated burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts, comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a surviving path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the burst signal processed, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans a plurality of processing intervals; and for each signal burst, determining which of the component trellis diagrams corresponds to a most likely path and estimating therefrom the unknown and undesired phase component.

39. A trellis processing method for processing a digitally modulated burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts, comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a surviving path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the burst signal processed, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans a plurality of processing intervals; for each signal burst, determining path history information for states in said composite diagram over the succession of processing intervals for which said path metric values are generated; and for each signal burst, deriving a demodulated symbol sequence from the stored path history information.

40. A system for processing a digitally modulated signal having an unknown and undesired phase component, comprising: a receiver module for receiving the digitally modulated signal and generating a succession of vector matched filtered output signal samples therefrom; a composite trellis processor (CTP) coupled to the receiver module for performing a CTP method comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the succession of signal samples, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans a plurality of processing intervals; and a plurality of trellis coupling transitions each coupling a "from state" in a first component trellis diagram to a "to state" in a second component trellis diagram indicating that trellis coupling is permitted between said "from state" and said "to state".

41. The system of claim 40 wherein the CTP method further comprises: determining path history information for states in said composite trellis diagram over the succession of processing intervals for which said path metric values are generated; and deriving a demodulated symbol sequence from the stored path history information.

42. The system of claim 41 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

43. The system of claim 42 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

44. The system of claim 43 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

45. The system of claim 42 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

46. The system of claim 45 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

47. The system of claim 42 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

48. The system of claim 47 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

49. The system of claim 47 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

50. The system of claim 40 wherein the CTP method further comprises: determining path history information for states in said composite trellis diagram over the succession of processing intervals for which said path metric values are generated; and estimating the unknown and undesired phase component from the stored path history information.

51. The system of claim 50 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

52. The system of claim 51 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith..

53. The system of claim 52 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

54. The system of claim 51 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

55. The system of claim 54 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

56. The system of claim 51 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

57. The system of claim 56 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

58. The system of claim 56 wherein the path metrics are generated in rigorous accord with the composite trellis diagram.

59. The system of claim 41 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

60. The system of claim 59 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

61. The system of claim 60 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

62 The system of claim 59 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

63. The system of claim 62 further comprising limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

64. The system of claim 50 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

65. The system of claim 64 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

66. The system of claim 65 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

67. The system of claim 64 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

68. The system of claim 67 wherein the CTP method further comprises limiting trellis coupling transitions to couple "from states" and "to states" that are in phase-adjacent component trellis diagrams.

69. The system of claim 41 wherein the digitally modulated signal is a burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts.

70. The system of claim 69 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and any time-concurrent "from states" coupled therewith.

71. The system of claim 69 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

72. The system of claim 71 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

73. The system of claim 50 wherein the digitally modulated signal is a burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts.

74. The system of claim 73 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

75. The system of claim 73 wherein each trellis coupling transition recurs periodically not more often than once every two consecutive processing intervals.

76. The system of claim 75 wherein the "from state" and the "to state" associated with each trellis coupling transition apply to a common time state and wherein the CTP method further comprises, for each processing interval, generating updated path metrics in accordance with an expedient processing method comprising: generating a provisional path metric for each "to state", each provisional path metric corresponding to the path metric indicating the highest likelihood path within the component trellis diagram associated with the "to state"; for each trellis "to state" not coupled to by a trellis coupling transition, selecting the provisional path metric for the state as the updated path metric; and for each trellis "to state" coupled to by at least one trellis coupling transition, selecting as the updated path metric the path metric indicating the highest likelihood path from among the provisional path metrics for the "to state" and all time-concurrent "from states" coupled therewith.

77. A system for processing a digitally modulated burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts, comprising: a receiver module for receiving the digitally modulated signal and generating a succession of vector matched filtered output signal samples therefrom; a composite trellis processor (CTP) coupled to the receiver module for performing a CTP method comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the succession of signal samples, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans a plurality of processing intervals; and for each signal burst, determining which of the component trellis diagrams corresponds to a most likely path and estimating therefrom the unknown and undesired phase component.

78. A system for processing a digitally modulated burst signal having an unknown and undesired phase component over a succession of processing intervals that span a plurality of signal bursts, comprising: a receiver module for receiving the digitally modulated signal and generating a succession of vector matched filtered output signal samples therefrom; a composite trellis processor (CTP) coupled to the receiver module for performing a CTP method comprising: generating path metrics in accordance with a composite trellis diagram having a plurality of states, each path metric being indicative of a likelihood associated with a path through the composite trellis diagram to an associated state in the composite trellis diagram, the likelihood being determined responsive to the succession of signal samples, the composite trellis diagram including: a plurality of component trellis diagrams that are phase displaced versions of a root trellis diagram which spans the succession of processing intervals and a plurality of phase displaced versions of the root trellis diagram; for each signal burst, determining path history information for states in said composite trellis diagram over the succession of processing intervals for which said path metric values are generated; and for each signal burst, deriving a demodulated symbol sequence from the stored path history information.