MXPA97001907A - Blind equalizing apparatus - Google Patents

Abstract

A demodulator for demodulating a signal modulated with digital information symbols so as to extract the information symbols is disclosed. A receiver receives a signal over a communications channel and samplers and digitizers produce a sequence of numerical sample values representative of the received signal. Memories are provided each having a number of state memories each associated with a hypothesized symbol string. A controller selectively retrieves values from the memory means and controls the timing of operations thereupon. A metric computer computes candidate metrics using a hypothesis of a next of the information symbols to be demodulated made by the controller, one of the numerical sample values, path metric values, B-matrices, and U-vectors and the candidate metrics associated state number selected by the controller from the memory means. A best predecessor computer determines the best of the candidate metrics to be selected to be written back into the memory means along with a successor B-matrix, U-vector and path history. The successor B-matrices, U-vectors and path history are then updated using corresponding values associated with the best predecessor and one of numerical sample values.

Description

"BLIND COMPENSATION APPARATUS" FIELD OF THE INVENTION The present invention relates to a method and apparatus for demodulating a digitally modulated radio signal that has passed through a channel suffering from variable constant time dispersion, such as a cellular radiotelephone signal. The present invention eliminates the need for a channel model by substituting a minimum-quadratic solution for the channel coefficients, which can be obtained in closed form.

BACKGROUND OF THE INVENTION The term "Compensator" is a generic term for a signal processing device that can demodulate or decode a signal while compensating certain channel imperfections. The channel imperfections that are usually corrected by a compensator are attenuation or uneven phase in the channel at different frequencies covered by the signal. Echoes are a manifestation in radio propagation that can cause variations in attenuation and phase through a frequency band. When digital radio transmission is employed, the echoes sometimes result in intersymbol interference (ISI), where a sample of the received signal depends on more than one adjacent symbol that have been "mixed" together by echo paths or delayed propagation with different delays. One type of compensator known in the prior art is the Infinite Impulse Response (FIR) compensator, or Transversal compensator. An FIR compensator tries to build an inverse of the channel imperfections in order to correct the signal. The disadvantage with this type of compensator becomes evident when the compensator tries to replace the components of the signal frequency that have been totally suppressed by a notch in the channel. In this situation, the compensator tries to create an infinite gain at that frequency which unduly accentuates the noise. Another type of known compensator is the Decision Feedback compensator (DF). The Decision Feedback compensator subtracts a weighted version of the already decoded symbol from the sample of the next signal to be decoded, thus pretending to cancel the echo of the just decoded symbol. The disadvantage with this type of compensators is that in a cell radio propagation environment, a direct wave can temporarily vanish leaving the echo as the component that carries the main signal. In this case, the echo should not be discarded such as the technique used in the Decision Feedback compensator, but rather used. When it can be identified that the main path has attenuation larger than a delayed echo path, it is possible to sample and store the signal in a memory as a sequence of samples, and then retrospectively process the signal in the inverted sample sequence in time of so that the echo is decoded and the weaker main path is deleted. An adaptive change in the direction of demodulation is disclosed in U.S. Patent Application Number 07 / 965,848 and in U.S. Patent Application No. 08 / 218,236, filed on March 28, 199. The Viterbi compensator is also shown in the prior art. The Viterbi compensator avoids the deficiencies of both FIR and DF compensators by not trying to undo the channel distortions and being insensitive to whether the shorter or delayed paths are dominant. Instead, the Viterbi compensator uses a channel or propagation trajectory model that is applied to hypothesized symbol sequences to predict what should be received. The hypothesis of the one that most closely matches the received signal is actually retained later. The Viterbi method can be considered as looking forward, where the current symbol is decided separately with all possible hypotheses for a limited number of future symbols. These multiple decisions are then gradually trimmed as equivalent decisions are made in future symbols. In the Viterbi compensator of the prior art, if the channel changes during a decoding sequence of multiple symbols, the channel model must be updated accordingly. U.S. Patent No. 5,164,961 discloses a Viterbi compensator having a channel model updated separately for each of the future hypothecated symbol combinations, wherein the selection of which channel model survives to be updated and used is brought to out in relation to equivalent decisions in future symbols. US Patent Application Number 07 / 894,933 describes a version of the adaptive Viterbi compensator called a "channel-by-state model" that does not use an explicit channel model, but rather uses direct predictions of signal samples for different symbol hypotheses that are they update directly after the symbol hypotheses are cut off in number, without having to go through the intermediate stage of a channel model. The aforementioned adaptive Viterbi compensator requires an initial channel calculation. In the prior art, an initial channel calculation is formed with the aid of groups of known symbols included in the transmitted data. These groups of symbols are called synchronization words or compensator training patterns. When the channel is not expected to change between training patterns, the initial calculation can be used without updating between the training patterns. This can lead to a loss of performance as a complexity against change, when the initial calculation is based on only a few known training symbols. Another known compensator is the compensator called "Blind". Blind compensators are supposed to work without the benefit of an initial calculation based on known symbols. Many blind compensators of the prior art have been granted to decode continuous data transmissions through the main telephone lines, for example. However, in these systems, there is no consequence if the systems lose a few hundred or a thousand symbols while they acquire initial convergence.

COMPENDIUM OF THE INVENTION The present invention relates to a device for demodulating a modulated signal with digital information symbols to extract the symbols. A receiving means receives a signal through a communication channel and a sampling and digitizing means produces a sequence of numerical values of the sample representative of the received signal. A memory medium that can have a number of status memories is disclosed where each state memory is associated with a string of hypothesized symbols and each comprises a path metric memory, a matrix memory B, a memory of vector U, and a trajectory history memory. The control means selectively recovers the values of the memory means and controls the synchronization operations therein. The metric calculation means calculates the candidate metric using a hypothesis of the information symbols to be demodulated next produced by the control means, one of the numerical sample values, path metric values, B matrices, U vectors and their associated state number that is selected by the control means from the memory medium. The optimal predecessor calculation means determines the optical candidate metric to be selected so that it is written again in the memory medium together with the successor matrix B, the vector U and the history of the trajectory. Finally, an updating means calculates the successor matrices B, vectors U and history of the trajectory using corresponding values associated with the optimal predecessor and one of the numerical sample values. Another embodiment of the present invention is capable of distinguishing itself through the prior art by being able to blindly compensate for short bursts of data that do not contain, in principle, known symbols, providing a channel calculation without losing any data. All symbols, in principle, are demodulated with the same precision as the last symbol, where the maximum information received and used to extract clues as to channel distortions. This behavior is obtained by mathematically eliminating the channel model using calculations based on the hypothesis of the entire demodulated symbol sequence, using respective calculations to test the possibility of their associated hypotheses.

BRIEF DESCRIPTION OF THE DRAWING These and other features and advantages of the present invention will be readily apparent to a person skilled in the art of the following written description, used in conjunction with the drawings, in which: Figure 1 illustrates a functional diagram of an embodiment of the present invention; Figure 2 illustrates an arithmetic unit of Matrix and Vector according to one embodiment of the present invention; and Figure 3 illustrates a modification of the present invention for Exponential Forgetting.

DETAILED DESCRIPTION OF THE EXHIBITION The present invention is primarily intended for use in cellular communication systems even though it will be understood by those skilled in the art that the present invention can be used in other different communication applications. A sequence of transmitted symbols is represented by sl, s2, s3 These symbols adopt binary values, such as +1, quaternary values such as + l / + j, or higher order modulation values.

- - The complex received samples rl, r2, r3 taken at a separation symbol are assumed to be linearly dependent on the symbols transmitted through a set of channel echo coefficients cl, c2, c3 ... cL according to the equation: This equation can be abbreviated as Rn = Sn-C where the subscript n is the first received n samples included in R and also the matrix S has n rows and L columns. The task of the receiver is to find the Sn sequence that best explains the received Rn waveform. further, the channel coefficients may not be known by any other means that the received signal has not been observed. Some restrictions have to be placed on how quickly the channel can vary. For example, the channel can not be allowed to switch completely between one symbol and the next, otherwise for any sequence of hypothesized symbols, a set of variable channel coefficients could be found that would explain the received waveform. Therefore, the channel should be assumed to vary to a regime that is slower than the regime of the symbol. A solution for the case of static channel will now be described below. The errors between the waveform expected for the sequence Sn of symbols and the samples Rn received are: En = Sn-C - Rn The error of sums of square In '-En = C'Sn'SnC - C'Sn'Rn -Rn'SnC + Rn'Rn where "'" is a transposition of conjugate. For a given Sn sequence, this sum-of-squares error can be minimized with respect to C by differentiating with respect to each C value and graduating equal to 0. It results in the set of simultaneous equations Sn'SnC = Sn'Rn, which can be rewrite as C = (Sn'Sn) -1Sn'Rn When this value for C is substituted in the equation of the sum-of-squares error, several terms are canceled leaving: En'En = Rn'Rn - Rn 'Sn (Sn' Sn) -1Sn 'Rn Since Rn'Rn does not depend on the selection of the sequence of Sn, the best sequence is simply that which maximizes the metric Rn' Sn (Sn 'Sn) ~ 1 Sn'Rn. In principle, a large number of Sn sequences can be tested and the sequence that provides the largest value of the aforementioned expression can be selected. However, the effort of the calculation increases exponentially with the number of symbols n that is included in the optimization, just as in the calculation of the maximum possibility sequence before the discovery of the Viterbi algorithm, or in problems of least squared adjustment before the discovery of quadratic minimum sequence algorithms such as Kalman. Therefore, a sequence algorithm to progressively extend the demodulation by means of a further symbol needs to be constructed. By presenting a Viterbi-like machine that maintains a number of candidate sequences with associated scores or trajectory metrics, one object is to find ways to extend these sequences by means of an extra symbol. Now we will explain the way in which one of these sequences is extended and its metric is updated. In this example, let Bn = (Sn'Sn) ~ l, where Bn is a square matrix whose elements depend only on the sequence Sn of hypothecated symbols. The term Bn refers to the values of the matrix after the symbols n have been hypothesized. An expression for B (n + 1) in terms of Bn will now develop. Sn is extended to S (n + 1) by the addition of a new row Zn -. { s (n + l), s (n), s (n-l) s (n-L + 2)} Therefore S '(n + 1) • S (n + 1) = Sn'Sn + Zn'Zn. Applying the matrix investment slogan to this, we provide: BnZn'ZnBn B (n + 1) = [Sn'Sn + Zn'Zn] -l = B - 1 + ZnBnZn ' In addition, the new product R '(n + 1) S (n + 1) = Rn'Sn + R * (n + l) Zn where r (n + l) is the sample received last. Substituting for B (n + 1) and R '(n + 1) S (n + 1) the following expression for the metric is obtained: [Rn'Sn + r '(n + l) Zn] Bn - [Sn'Rn + r (n + 1) Zn'] Multiplying this metric provides eight terms that will be listed below.

Term 1 Rn'SnBnSn'Rn, which is the previous Mn metric; Term 2 r * (n + 1) ZnBnSn'Rn; Term 3 Rn 'SnBnZn' r (n + 1), note that terms 2 and 3 are complex conjugates of one another; Term 4 r (n + 1) | ZnBnZn ' Term 5: -Rn 'SnBnZn' ZnBnSn 'Rn, 1 + ZnBnZn' where the numerator Rn 'SnBnZn' is a scalar and the other factors are its complex conjugate, therefore Term 5 can be written as: - | Rn 'SnBnZn' \ 1 1 + ZnBnZn 'Term 6: -r (n + l) (ZnBnZn') (ZnBnSn'Rn), where 1 + ZnBnZn ' the first of the numerator terms in parentheses is also needed in the denominator, and the second numerator term in parentheses is the same as the one that was calculated for terms 2, 3 and 5; Term 7: - (Rn 'SnBnZn') (ZnBnZn ') r (n + 1 j 1 + ZnBnZn which is the conjugate of Term 6; Term 8: r (n + l) (ZnBnZn) (ZnBnZn) r (n + 1) = - | r (n + 1) I í_ (ZnBnZn) 1 + ZnBnZn '1 + ZnBnZn' A calculation strategy for the 8 terms will be described below. First, Un '= Rn'Sn is available from the aforementioned iteration. The term Un is a vector with a length equal to the symbol time dispersion number, L. Then Zn is formed using the newest symbol s (n + l) and Vn = Bn Zn'is calculated to be also a vector of the L-element. The inverse Bn is an L through the matrix L available from the aforementioned iteration. The complex scalar a = Un'Vn and the real scalar b = Zn Vn are then calculated. As a result, Mn + 2 (l-b) Re. { a "r (n + 1).}. + | r (n + l) | 2 - | a | 2 M (n + 1) = 1 + b provides the new metric in terms of the old one. The matrix entities involved in these calculations is equal to the dispersion number L of symbol time that has to be handled and does not grow with the number of symbols processed.Therefore, the metric of the trajectory can be updated with a constant effort by means of the symbol processed, that is, not exponentially increasing the stress as successive symbols are progressively decoded In other words, the effort to decode 100 bits is just 10 times the effort to decode 10 bits.

U '(n + 1) = A' + r '(n + l) Zn, and VnVn 'B (n + 1) = Bn - 1 + b to be used in the next iteration. The aforementioned metric update has to be carried out for each candidate sequence Sn and for each possible value of the newest symbol s (n + l).

This results in a tree that expands exponentially from possible sequences that do not have to be trimmed. The Viterbi algorithm selects to retain the best of all the sequences that suit its last symbols, thus restricting the number of states retained to 2 in a binary modulation system. An alternative would be to retain only the best K states. It can be understood that the sequences differ only in sign, or in a symbol phase rotation provides the same result and therefore corresponds to the ambiguity of the demodulation when a reference is not provided. absolute phase. This ambiguity has to be solved either by differential decoding after the demodulation or by including symbols of known polarity in the transmitted data to restrict the permissible sequences. Unlike the methods of the prior art, the present method does not place any limit on the number of sequences that can be usefully retained for the possibility of additional testing, since the result of each interaction depends not only on the number of symbols in the response length of the finite channel pulse but also in the channel calculation involved throughout the sequence to date. The structure of an apparatus for carrying out the aforementioned demodulation algorithm will be described below. The present example uses a modulation system of 4-aryl phases such as QPSK or Pi / 4-QPSK which is used in e). digital cellular system of the United States. The time dispersion of the period of a symbol must be allowed so that each channel has two terms and therefore L = 2. The matrices of Bn are therefore 2 x 2 and the multiplication of the Zn consists of an old symbol and a new symbol. Therefore, a 4-state machine is proposed in which all four possibilities of the preceding symbol are retained. The internal values in the Viterbi processor can therefore be structured in the following way: STATE HISTORY matrices vector SYMBOL METRIC B U TRAJECTORIES 013201032001321 Bo Uo Mo 132103201032023 Bl Ul MI 320103201032321 B2 U2 M2 221302102201132 B3 U3 M3 The operation of the apparatus will now be described. With state 0 as the predecessor and a new one - 1! symbol of 0, the sample r (n + l) just received is used to calculate the new metric of Mo. This is repeated for states 1, 2 and 3 as predecessors and the predecessor that provides the largest new Mo metric is selected to be the predecessor of the new 0 state. The associated B matrix and the U vector are updated to be the new Bo and Uo and the history of the associated symbol is copied to the new state 0, shifting the status number of the selected predecessor to the left. This is repeated using the new symbol postulates of 1, 2 and then 3 to generate new states 1, 2 and 3. When the channel varies in time, it is not valid to replace the minimum-quadratic calculation of the calculated C channel through the Total symbol history. One solution would be to calculate the channel through a movable block of symbols and replace this one. In accordance with one embodiment of the present invention, the same algorithm described above can be used if the matrix B is updated by adding the influence of the last symbol while subtracting the influence of the oldest symbol in the block. The U vectors are updated in the same way. When the channel is expected to vary so rapidly that the number of symbols in this block window is small, for example 5, there are only 32 possible B matrices (in the binary case) and all these matrices can be pre-calculated and stored, avoiding in this way the need to update the matrices B. A second solution to vary the channels time is to calculate the channel with exponential de-emphasis of the oldest errors. A solution for the channel that minimizes the sum of squares of errors Ei is provided by: Therefore, the last error (i = n) is not weighted downwards, while the errors that occur in the symbols q are weighted beforehand by EXP (-qT). Taking the exponential weights inside the parentheses and applying them to the rows of S and the values r, ee gets a modified S matrix where the last row is weighted by 1, and in this way it is not changed, but where the Previous rows are weighed progressively downwards by EXP (-qT). The new vector R also has an element r (n) last not changed and previous elements that are weighed down exponentially. With this modified vector matrix S, the solution for the channel that minimizes the error of the square sums through an i-1 to n is still provided by: Cn = (Sn'Sn) _1Sn'Rn The vector R now however is updated by means of: where d - EXP (-T) That is, the vector R is extended to the elements n + 1 by not weighing the previous elements by d and attaching the new received sample r (n + l). The matrix S is updated by means of In other words, the matrix S is expanded by a new row previously represented by Zn, while the previous rows are weighed down by d. In this way: S (n + 1) 'S (n + 1) = Exp (-2T) Sn'Sn + Zn'Zn provided the recursion formula for B (n + 1) as: Bn - Bn | d2 Then BnZn'ZnBn B (n + 1) = Bn 1 + ZnBnZn ' The recursion formula for the vector previously defined as Un becomes: A '< = d2-A 'U' (n + 1) = A '+ r "(n + 1) • Zn Other than the aforementioned modifications to update Un and Bn, no other changes to the algorithm are needed to deal with channels that vary in time. Note that the increase of Bn by EXP (2T) and the reduction of A by EXP (-2T) should be carried out as a first step, so that the algorithm for the varying channels becomes: i) calculate A = d «One where d = EXP (-2T) ii) calculate Bn = Bn / d iii) form Zn using the newest symbol s (n + l) and Calculate Vn = Bn Zn iv) calculate a = Un'Vn yb = ZnVn Then Mn + 2 (l-b (Re { A-r / n + l).}. + | R (n + l) | 2 - the M8n + 1 '1 + b v) U '(n + 1) = A' + r '(n + l) Zn vi) B (n + 1) = Bn - VnVn 'T + b The d factor of exponential forgetting in the aforementioned equations must be selected according to how quickly the channel can vary. The optimal value for d for a given channel change rate can be established by simulation. One embodiment of the present invention is illustrated in Figures 1 to 3. Figure 1 shows a radio receiver 11 that receives a signal from antenna 10.

Alternatively, the radio receiver can be connected to a telephone line. In addition, the radio receiver can be placed in a subscriber unit of the cellular radiotelephone or a cellular radio network base station. The radio receiver filters amplify and convert the received signal to an appropriate frequency and a form for digitization in a sampling and digitization unit 12. The numerical samples representing the value of the complex number of the instantaneous radio signal are damped in a sample buffer 13. As a result, the samples can be selectively recovered under the control of a control unit 16. A collection of state memories 15 includes, for each state, a path metric memory element, a matrix memory element B, a vector memory element U, and a path history memory. The trajectory history memory for each state contains a string of symbols, of previously hypothesized symbols that will become the output of the device, if the associated trajectory metric is at a certain point that is judged to be the largest. The control unit 16 may select the content of each state memory in turn to be applied to the metric computer 14 together with a selected signal sample recovered from the intermediate memory 13. The control unit 16 also informs the metric computer of the selected status number corresponding to each unresolved hypothesis of combinations of symbols in sequence not yet achieved in the history memory of the state path. For example, if only one unresolved symbol hypothesis is allowed in the device (that is, the device is configured to deal with only delayed echoes in a symbol period), and the symbol is a binary bit, there will be only two status memories and the control unit will just inform the metric computer if the status "1" or status "0" is currently selected. If the symbol is quaternary, all four states will exist and the control unit 16 will inform the 14 metric computer if the symbol 0, 1, 2 or 3 is selected. If two unresolved quaternary symbols are allowed that correspond to the devices that are configured to deal with up to delayed echoes over a period of two symbols, then the control unit 16 informs the computer 14 metric which of the 16 states 00.01,02,03,10,11,12,13,20,21 , 22, 23, 30, 31, 32 or 33 is currently selected. The 14 metric computer actually requires the complex values that the transmitter has transmitted for these symbols. It is important if the control unit 16 supplies the complex values directly to the metric computer, or if the metric computer converts the symbols into complex values using a model of the transmission modulation process. For example, if the transmission modulator process is QPSK, where the symbols are represented by a complex vector of constant length and phase angle of 45 °, 135 °, -45 ° or -135 °, then the complex values that correspond to the quaternary symbols 0,1,2 and 3 are (l + j) / V2 ~, (-l + i) / y2, (l-j) TV and (-l-j 27 respectively. These are the S values used in the aforementioned calculations. Matrix and vector arithmetic unit 14 calculates a new metric for a given new symbol hypothesis and each of the previous selected states in turn. The largest of the new values are selected as being the new metric value of the state, the hypothesis of the new symbol will be associated with a new unresolved symbol. When the greater of the new metric values is selected, the oldest hypothesis of the previously unresolved symbol is solved and this resolved symbol is then moved to the memory of the history of the associated trajectory. In this way, a new unresolved symbol is created during each iteration of the device while the old hypothesis is resolved, so that the number of unresolved symbols remains constant and the number of states remains constant. In addition to calculating a new metric value for each of the new states, the matrix and vector arithmetic unit 14 calculates the new matrix B and vector U for each new state. The new elements B and U are calculated using the elements of corresponding entities of the old state that gave rise to the largest metric that was selected for the new state. The signal flow within the matrix and vector arithmetic unit 14 is shown in Figure 2. A concatenator 20 forms a Zn vector of the complex values associated with the unsolved symbols at present (Sn, S (n-1 ) ...) of the selected state of the control unit, plus a new unresolved symbol S (n + 1). The vector Zn is simply (S (n + 1), Sn, S (n-1) ....) The complex conjugate of Zn is multiplied by the matrix B selected from the control unit in a multiplier 21 to generate the new vector Vn. The vector Vn is then multiplied by the vector Zn again in the multiplier 22 to yield the numerical "b" signal in a 24 metric computer. The vector Vn is also multiplied by a vector U selected from the decimal unit in a multiplier 23 to yield a second scalar "a" quantity that also applies to the 24 metric computer. The other values used in the metric computer are also scalar, namely the sample of the signal withdrawn from the control signal r (n + l) of the intermediate memory 13 and the metric value M of the selected path of the control unit. control. The 24 metric computer combines the values to produce the new metric with the control unit by selecting the entities M, B and U of each state of the state memory. The results are temporarily stored in a temporary storage so that they can be compared in a comparator 26 to find the largest value. The state that gives rise to the largest value and the largest value is determined by the comparator 26. This is called an optimal predecessor state. This resolves the oldest hypothesis of unresolved symbols that will be associated with the state within which the hypothesis will be associated. new symbol S (n + 1). This is called the successor state. The resolved symbol is appended to the other symbols in the trajectory history of the optimal predecessor to become the history of the successor's trajectory. In addition, entities B and U of the optimal predecessor are updated to become entities B and U of the successor state. The update is carried out by blocks (27-31) of Figure 2. Block 27 carries out the external multiplication of vector Vn with its own conjugate transposition to form a square matrix. This is scaled by l / (l + b) in a scalar 28 using the value b calculated by the multiplier 22, and then subtracted from the matrix B of the predecessor state in order to obtain the matrix B of the successor state. The vector U of the successor state is calculated by a multiplier 30 and an adder 31. The vector Zn of the concatenator 20 is multiplied by the complex conjugate of the sample of the selected received signal r (n + l) in the multiplier 30. The result then it is added to vector 1 of the predecessor state to form the vector U of the successor state in the adder 31. When the control unit has completed the aforementioned processing for all the new hypotheses of the symbol, the play of the content of the new state memory (successor states) becomes the content of the state memory to start a new processing cycle in order to decode the sample of the signal r (n + 2). The process described above exhibits the always diminished matrices B and the vectors U always increased. This is appropriate if the channel imperfections (eg, echoes or time dispersion) are not variable. However, if the channel is variable, the exponential downward weight of the past history is more appropriate, and this can only be applied by modifying the matrices and the U vectors before use, as illustrated in Figure 3. Figure 3 shows two scaling units 32 and 33 that modify the B arrays and the selected U vectors of the control unit respectively before being used in Figure 2. The B matrices are augmented by the d factor, while the U vectors are decreased by the same one. factor. This inhibits the permanently diminished compartment of the B matrices and the permanently increased behavior of the U vectors that occurs without exponential forgetting. As a result, the present invention tolerates channel variations and resolves the decoded symbols based more on the samples of the newly processed signal than on the samples of the processed signal for some time. It can also be understood that "most recently processed" is not necessarily the same as "received more recently", since the use of a buffer allows the received samples to be processed in an inverted order in time or even in half in inverted order in time and half in order not invested in time from a central starting point.

In addition, other means for adapting the device to the case of variable channel can be used, which correspond to the Kalman formulation of the minimum-quadratic sequence processing compared to the processes in minimal-quadratic sequence with the exponential forgetting. These differ only in the manner in which the matrices B are continuously prevented from decreasing. In the Kalman approach, a constant Q matrix is added to matrices B as a pre-modification before use. All of these mini-quadratic formulations can be implemented by the present invention and are selected according to the manner in which the channel spreads to vary. Operation simulation can be used to determine which approach to use and to determine optimal d values or Q matrix values for a specific type of channel. The arithmetic units illustrated in the blocks of Figures 1 to 3 should preferably be simplified to the maximum degree by observing whether the values are purely real rather than complex and that they do not generate an imaginary part if the values are real.

Also, the values of the symbol for the binary modulation are just real and + or -1, allowing further simplification of the multipliers 21, 22 and 30. Even if the quaternary symbols of QPSK are used as mentioned above, the factor common 1 / V2 between the real and imaginary parts can be removed leaving only the values of +1 and -1, which again allows large simplifications of the multipliers 21, 22 and 30. The construction of the present invention, however, is applicable to decode any representation of the digital data by modulating the amplitude or phase values of a radio signal or both. It can also be appreciated that the present invention can be applied to the decoding of digital data signals that have been received over a telephone line. It will be appreciated that a person skilled in the art can devise many variations in the detailed implementation of the invention, including the use of programmable signal processors or stored program arithmetic devices to carry out the necessary arithmetic. All these variations are considered to be within the scope of the invention, as presented in the following claims.

Claims (19)

CLAIMS;

1. An apparatus for demodulating a modulated signal with digital information symbols in order to extract the information symbols, comprising: a receiver means for receiving a signal through a communication channel; a sampling and digitizing means for producing a sequence of numerical sample values representative of the received signal; a memory medium having a number of status memories each associated with a string of hypothesized symbols and each comprising: a path metric memory, a matrix memory B, a vector memory U, and a memory history of trajectory; a control means for selectively recovering the values of the memory medium and controlling the synchronization of the operations therein; a metric calculation means for calculating the candidate metrics using a hypothesis of a following symbol of the information symbols to be demodulated by the control means, one of the numerical sample values, trajectory metric values, B matrices and vectors U and candidate metrics associated with the state number selected by the control means from the memory medium; a means of calculating the optimal predecessor to determine the best of the candidate metrics to be selected to be written back to the memory medium together with the successor matrix B, the vector U and the trajectory history; and an update means for calculating the successor B matrices, the U vectors and the history of the trajectory using corresponding values associated with the optimal predecessor and one of the values of the numerical sample.

The demodulation apparatus according to claim 1, wherein the receiving means is a radio receiving means equipped with an antenna.

3. The demodulation apparatus according to claim 1, wherein the receiving means is connected to a telephone line.

The demodulation apparatus according to claim 2, wherein the receiving means is used in a cellular radiotelephone subscriber unit.

The demodulation apparatus according to claim 2, wherein the receiving means is used in a cellular radio network base station.

6. The demodulation apparatus according to claim 1, wherein the matrices B are modified before being used in the metric calculation means and the updating means.

The demodulation apparatus according to claim 1, wherein the U vectors are modified before being used in the metric calculation means and the updating means.

The demodulation apparatus according to claim 6, wherein the matrices B are modified by ascending the arrays B by a predetermined factor.

9. The demodulation apparatus according to claim 7, wherein the U vectors are modified by scaling down the U vectors by a predetermined factor.

The demodulation apparatus according to claim 6, wherein the matrices B are modified by adding a constant matrix to the matrices B.

11. The demodulation apparatus according to claim 10, wherein the constant matrix has only elements diagonals that are not zero.

The demodulation apparatus according to claim 1, wherein the signal is modulated by bits of binary information.

13. The demodulation apparatus according to claim 12, wherein the method of modulation is Inverse Keypad in Phase.

The demodulation apparatus according to claim 12, wherein the modulation changes the phase of the signal.

15. The demodulation apparatus according to claim 1, wherein the signal is modulated with quaternary symbols.

16. The demodulation apparatus according to claim 15, wherein the modulation method is the Quadrature Phase Displacement Keypad.

17. The demodulation apparatus according to claim 12, wherein the modulation method is the off-center quadrature phase shift keyboard.

18. A blind compensation apparatus for demodulating the signals carrying an information symbol received from a communications channel with unknown echoes or time dispersion, comprising: a processor with maximum trend to the sequence having a number of status memories each equipped with: a path history memory, a path metric memory, and a first memory for a plurality of adaptive values; a path metric calculation means for calculating the new path metric values using old path metric values, adaptive values, and a latent sample of the received signal; a means of adaptation to adapt the adaptive values in such a way that each of the new metric values is essentially equal in value to one of the old metric values plus an increment based on the received signal sample to the last one that had the old metric values that were skeletal with the benefit of the information contained with the signal sample received last.

19. The compensation apparatus according to claim 18, wherein the process of maximum trend to the sequence is a Viterbi processor. SUMMARY OF THE INVENTION A demodulator is disclosed to demodulate a modulated signal with digital information symbols in order to extract the information symbols. A receiver receives a signal through a communication channel and the extracted samples and digitizers produce a sequence of numerical sample values representative of the received signal. Memories are provided each having a number of status memories each associated with a string of hypothesized symbols. A controller selectively retrieves the values from the memory medium and controls the synchronization of the operations therein. A metric computer calculates the candidate metrics using a hypothesis of a next symbol of the information symbols to be demodulated produced by the controller, one of the numerical sample values, the path metric values, the B matrices and the U vectors and the candidate metrics associated with the number selected by the controller from the memory medium. An optimal predecessor computer determines the best of the candidate metrics to be selected to be written back to the memory medium together with the successor matrix B, the vector U, and the history of the trajectory. Successor B matrices, U vectors and trajectory history are then updated using correding values associated with the optimal predecessor and one of the values of the numerical sample.