MXPA96005540A - Receiver in diversity for signals with multip trajectory time dispersion - Google Patents

Abstract

The present invention relates to a digital communications receiver providing MLSE equalization and joint diversity combination. A plurality of branches in diversity are processed to produce complex received data samples and synchronization information. The channel calculators then form channel calculations of the data samples and synchronization information. Data samples and channel calculations are then used by pre-processors to produce metric multipliers. Finally, the metric multipliers are combined with the hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm to produce a stream of demodulated data.

Description

"RECEIVER IN DIVERSITY FOR SIGNALS WITH DISPERSION OF TIME OF MULTIPLE TRAJECTORIES" FIELD OF THE EXHIBITION The present invention relates to the combination and equalization in diversity in a receiver for digital wireless communications.

BACKGROUND OF THE EXHIBITION In recent years, digital wireless communication systems have been used to transport a variety of information between multiple locations. With digital communications, the information is translated in digital or binary form, which is referred to as bits, for communication purposes. The transmitter maps this bitstream in a modulated symbol current that is detected in the digital receiver and is mapped back to bit and information. In digital wireless communications, the radio environment presents many difficulties that impede satisfactory communications. One difficulty is that the level of the signal can be attenuated, because the signal can travel in multiple trajectories. As a result, the signal images arrive at the receiver's antenna out of phase. This type of attenuation is commonly referred to as Rayleigh attenuation or rapid attenuation. When the signal is attenuated, the signal-to-noise ratio becomes lower causing a degradation in the quality of the communication link. A second problem occurs when the multiple signal paths are very different in length. In this case, time dispersion occurs, wherein the multiple attenuation signal images arrive at the receiver antenna for different periods of time, resulting in echoes of the signal. This causes an interference between the symbols (ISI), where the echoes of a symbol interfere with the subsequent symbols. Raleigh attenuation can be mitigated using diversity such as antenna diversity in the receiver. The signal is received in a plurality of antennas. Because the antennas have slightly different antenna locations and / or patterns, the attenuation levels at the antennas are different. In the receiver, these multiple antenna signals are combined either before or after the detection of the signal, using such techniques as combination of maximum ratio, equal gain combination and selective combination. These techniques are well known to those skilled in the art and can be found in normal textbooks such as .C.Y. Lee, Mobil and Communicati ons Engineering, New York: McGraw-Hill, 1982. Time dispersion can be mitigated using an equalizer. Common forms of equalization are provided by linear equalizers, decision feedback equalizers, and maximum likelihood sequence calculation (MLSE) equalizers. A linear equalizer tries to undo the effects of the channel by filtering the received signal. A decision feedback equalizer exploits the previous symbol detections to cancel the interference between the symbols of the echoes of these previous symbols. Finally, an MLSE equalizer hypothesizes several sequences of transmitted symbols and, with a model of the dispersive channel, determines which hypothesis best fits the received data. These matching techniques are well known to those skilled in the art and can be found in normal textbooks such as J.G. Proakis, Digi tal Communicantions, Second Edition, New York: McGraw-Hill, 1989. Of the three common equalization techniques, the MLSE equalization is preferred from an operational point of view. In the MLSE equalizer, all possible sequences of transmitted symbol are taken into account. For each hypothetical sequence, samples of the received signal are predicted using a model for the multipath channel. The difference between the samples of the predicted received signal and the samples of the actual received signal, which is referred to as the prediction error, provides an indication of how good a specific hypothesis is. The quadratic magnitude of the prediction error is used as a metric system to evaluate a specific hypothesis. This metric system is accumulated for different hypotheses to be used to determine which of the hypotheses are better. This process is carried out efficiently using a Viterbi algorithm, which is a form of dynamic programming. Ideally, the process of combination in diversity and the matching process should be combined in some optimal way. Recent research has shown that for the MLSE equalization, the diversity combination must be carried out with the equalizer. This investigation could be found in the articles of W.H. Sheen and G.L. Stüber, "MLSE equalization and decoding for multipath-fading channels," IEEE Trans. Commun. , volume 39, pages 1455 to 1464, October 1991; in the article by Q. Liu and Y. Wan "An adaptive maximum-likelihood sequence estimation receiver with dual diversity combining / selection," Intl. Symp. on Personal, Indoor and Mobile Radio Commun. , Boston, MA, pages 245 to 249, from October 19 to 21, 1992; and in the article by Q. Liu and Y. Wan, "A unified MLSE detection technique for TDMA digital cellular radio," 43rd IEEE Vehicular Technology Conference, Seacaucus, NJ, pages 265-268, from May 18 to 20, 1993. In the above-mentioned research, the combination in diversity is carried out by adding together the quadratic prediction errors of magnitude of the different diversity channels when forming the metric systems. A further improvement is obtained by grading the quadratic prediction errors of different diversity branches. A detailed description of this MLSE equalizer is provided in U.S. Patent No. 5,191,598 issued to T.O. Backstrom et al., Which is incorporated herein by reference. Unfortunately, the MLSE equalizer involves calculating many quadratic prediction error terms. This can be expensive in terms of hardware or software complexity. Therefore, there is a need to reduce the complexity of the MLSE equalizer / diversity combiner. For the MLSE equalizer without a diversity combination, the Ungerboeck method uses two steps to reduce complexity, as described in the article by G. Ungerboeck, "Adaptive Maximum Likelihood Receiver for Carrier Modulated Data Transmission Systems," IEEE Trans. Commun. , volume COM-22, number 4, pages 624 to 535, May 1974. The first step is to expand the magnitude table term to eliminate terms that are common for all hypotheses. As a simple example, the term (a-b) 2 can be expanded by a ^ -2ab + b ^. if "a" does not depend on the hypothetical data, then the term a ^ can be suppressed from the metric calculation. The second step used by Ungerboeck is to re-arrange the order of the metric calculations. With normal MLSE equalization, the metric systems are calculated and updated based on successively received data samples. Each iteration of the Viterbi algorithm corresponds to a new sample of data received. Using the second step, each iteration of the Viterbi algorithm corresponds to a newly transmitted symbol. These two steps can also be explained by a single example. Suppose that the transmitter transmits a symbol current s (n), where each s (n) can take one of the possible complex values of S. In the receiver, the received signal is sampled once every T seconds, where T is the period of the symbol to provide a stream of received signals r (n).

Suppose that the intervention channel consists of two attenuation rays, a main beam and an echo, where the echo reaches T seconds later. Then, the received signal can be modeled as: r (n) = c (0) s (n) + c (l) s (n-l) + n (n) where (c (0) and c (l) are complex channel derivation values and n (n) is an additive noise of a certain kind.In the MLSE equalizer, in iteration n, there must be previous "states" other than S , which correspond to the possible values of S for s (nl) Associated with each previous state would be an accumulated metric system, accumulated from previous interactions, There would also be current states S corresponding to the values of S possible for s (n) Each possible pairing of an earlier state, with a current state, corresponds to a hypothetical sequence { Sftín), Sh (nl)} . For each hypothesis, the value of the predicted received signal must be: rpred (n 'h = c (°) sh (n> + c (l) Sh (n-1).

The corresponding metric system of branch or delta must be provided by Mn (n) = | r (n) - rp ec¿ (n, h) | 2 The candidate metric system for a current state would be the sum of the metric branch system and the previously accumulated metric system associated with Sft (n-l). For each current state, there are S possible previous states. For each current state, the previous state that provides the smallest candidate metric system is selected as the predecessor state, and the candidate metric system becomes the cumulative metric system for that current state. In the next iteration, using r (n + 1), the current states of time n become prior states during time n + l. After all data has been received, the state with the smallest cumulative metric system, and all precedents, indicate the highest probability transmitted symbol sequence that becomes the sequence of the detected symbol. Sometimes decisions are made before all data is received, using a depth of decision. The first step of Ungerboeck can be illustrated by expanding the expression for Mn (n). This provides Mh (n) = A (n) + B (n) + C (n) + D (n) where A (n) - | rTn) | J B (n) - 2 Rß. { r (n) c * (0) s ** (n)} + 2 Re { r (n) c * (l) s > * (n-l)} C (n) - | cCO) | a | s > (n) | l + | c (l) | J (n-l) | * D (n) - 2 Re { c (0) c * (l) sh (n) s > * (n-l))} where "*" represents the conjugation of the complex. The Ungerboeck method suppresses the term A (n) that is common to all Mh (n). The second step of Ungerboeck combines the proportional terms to Sn * (n) of the different iterations. In iteration n + l, the terms become: B (n + L) - 2 Re { r (n + l) c * (0) (n + 1)} + 2 Re { r (n + l) c * (l) (".}. C (n + 1) - jc (0) | l | (n + I) |» + | C (1) | '\ n) \ * D (n + 1) - 2 Rβ. { c (0) c * (l) s "(n + l) s" * (n)} Therefore, there are terms proportional to S ^ * (n) in both iterations. These can be recombined by determining a new metric system, M'h (n), co or M'h (n) = B '(n) + C (n) + D (n) where - - B '(n) - 2 Re { f (n) O} f (n) * r (n) c * (0) + r (n + l) c * (l) C '(n) - (| c (0) j> + | c (l) |.} l in) As a result, B '(n) contains f (n), which can be obtained by filtering the data r (n) received with a filter using the derivations c * (O) and c * (l). In this way, the new metric system uses two data samples, r (n) and r (n + l), instead of just one, r (n). Also, unlike B (n) and C (n), B '(n) and C (n) depend only on one hypothesized symbol, S ^ í), instead of two, Sn (n) and Sn (n -1). Therefore, conceptually, iteration n corresponds to the transmitted symbol Sn (n) instead of the value of the received data r (n). When MLSE matching and diversity combining are carried out together, the Ungerboeck form can be used to reduce complexity. The situation in c (0) and c (l) do not change with time, which is referred to as the case of static channel, will be described in the US Patent Number 5, 031,193 issued to Atkinson et al. In the Patent of Atkinson et al., Both steps of Ungerboeck are used to obtain the demodulator shown in Figures 1 and 2 of the North American Patent Number 5,031,193. In Figure 2 of the Atkinson et al. Patent, the term f (n) is obtained with a matching filter in the diversity branch 1, and a matching filter in the diversity branch 2. However, there are disadvantages with the use of the Ungerboeck form. One is the problem that the channel derivation values, c (0) and c (l) in the previous example can change with the sample time n. In the conventional manner, all terms c (0) and c (l) can be replaced with c (0, n) and c (l, n). As a result, the Ungerboeck form contains a mixture of time channel n and time n * l derivations. This will require a storage capacity for multiple sets of channel derivations and may make it more difficult to track the channel. Channel tracking and prediction are well understood and examples can be found in the A.P. Clark and S. Hariharan, "Adaptive channel estimator for an HF radio link." IEEE Trans. Co one. volume 37, pages 918 to 926, Sept. 1989. In US Patent Number 5,031,193, an alternative solution is provided to address the case of variable time channel derivations. However, this solution does not use the Ungerboeck form, and does not optimally combine matching and diversity. Instead, each branch in diversity has a separate equalizer. The channel follow-up is carried out using branch detections and not the detections that have benefited, the combination in diversity. The outputs of these equalizers are then combined using normal diversity combining techniques. In this way, the equalization and diversity combination steps have been carried out separately and not together. Therefore, there is a need for a receiver that jointly carries out the matching of MLSE and the combination in diversity and that lends itself to the case where the channel varies as a function of time.

EXHIBITION DIGEST An object of the present invention is to provide an efficient way to join the MLSE match and the diversity combination for use in a wireless digital communications receiver. This is achieved by expanding the medical expressions and collection terms that correspond to the same hypothetically transmitted symbol. The embodiments of the present invention will be provided for both static channel and time variable cases. One embodiment of the present invention provides a digital communications receiver comprising means for processing signals of a plurality of branches in diversity to produce complex receiver data samples and synchronization information. The channel calculation means then forms channel calculations of the data samples and synchronization information. The data samples and the channel calculations are then used by the pre-processing means to produce metric multipliers. Finally, the combining means combine the metric multipliers with the hypothesized data sequences to generate and accumulate metric systems that use a sequence calculation algorithm to produce a data stream of odulated. In accordance with another embodiment of the present invention, there is provided a digital communication receiver comprising a means for processing signals of a plurality of branches in diversity to produce complex receiver data samples and synchronization information. The channel calculation means then forms channel calculations using synchronization information and initial data detection to produce channel calculations that vary in time. The data samples and the channel calculations are then used by a preprocessing means to produce metric multipliers. Finally, the combining means combine the metric multipliers with the hypothesized sequences to generate and accumulate metric systems using a sequence calculation algorithm to produce a demodulated data stream.

BRIEF DESCRIPTION OF THE DRAWINGS These and other features and advantages of the 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 digital communication receiver in accordance with one embodiment of the present invention. invention; Figure 2 illustrates a branch signal processor according to an embodiment of the present invention; Figure 3 illustrates a metric pre-processor in accordance with one embodiment of the present invention; Figure 4 illustrates a digital communication receiver in accordance with an embodiment of the present invention; Figure 5 illustrates a branch signal processor in accordance with one embodiment of the present invention; and Figure 6 illustrates a metric pre-processor in accordance with the embodiment of the present invention.

DETAILED DESCRIPTION OF THE EXHIBITION In the present invention, the combination of MLSE equalization and diversity are carried out together in a diversity equalizer. In this mode, the term s (nT) is a sequence of the transmitted symbols where the period of the symbol is T seconds. Impulse configuration is used to generate a continuous time signal, t (t). In addition, the term r (t) is the complex received signal of the baseband of the diversity branch d. After synchronization, the received values that are sampled are processed, where there are M samples per symbol T period. This allows both simulation based matching (M = 1) and fractionally separated matching (M> 1). In this way in the period of time (nT, (n + l) T), the samples r¿ (nT), r¿ ((n + l / M) T) ... r¿ ((n + (Ml) / M) T) are received in each branch in diversity, a sample for each sampling phase. The following channel models are used for these received signals: r, ((n + m / M) T) - Cd? (0, n) s (nT) + Cd? (L, n) s ((nl) T ) + + 9? * CM.n) s ((n-J + l) 1) for m - 0,1, ...? Í-i where J is the number of channel derivations in each model and C < M (j n) is the derivation of channel j'th for the branch d in diversity, the sampling phase m, in the interval n of time. The following metric delta system is used in the Viterbi algorithm in iteration n: Mh. { n) -? ? \ rJ ((n * m / M) T) - r ^ Ji (n * m / M) T, h) where r4 ^ in * mlMT, h) '? cJUjQiffl and Sh (j) represents a hypothesized symbol s (jT). This metric system requires a significant amount of calculation. In some applications, it may be advantageous to use a weighted sum when the metric systems are formed. This leads to the new formula where W m (n) are the non-complex weighting factors. For optimal combination, these weights would be the calculations of the reciprocal of the noise powers at the link d in the sampling phase m, that is, W¿m (n) = 1 / N¿ [m (n). The noise powers can be kept constant through all or some iterations, or they can vary with each iteration n. The noise powers could be obtained from calculations of the initial channel derivation and the synchronization data, since noise samples can be obtained by taking the difference between the received signal samples and the predicted signal samples based on the known synchronization sequence. These noise samples can be quadratic in magnitude and averaged to obtain a calculation of Nd m (n) * It may be desirable to combine the calculations of different sampling phases, m, to obtain the weights W¿ = 1 / N, ¿, which they are independent of the time and the sampling phase. If the channel is variable in time, then the internal signals to each channel follower can be used to calculate the noise power that varies in time. Alternatively, this can be done using detected symbols, channel derivation calculations and the received data. The results of the multiple sampling phases can be combined providing in this way Wd (n) = 1 / Nd (n). For semi-optimal combination, these weights would be all unitary and could be ignored in the implementation. For selection combination, these weights would all be zero except for the diversity channel (s) and the selected sampling phase (s). The selection could be based on a number of criteria such as a calculation of the signal to noise ratio in each branch and poly the sampling phase as well as a calculation of the signal or power received in each branch and poly the sampling phase. There are other poilities for these weights. They would be in a certain way related to the branching quality in diversity. To reduce complexity, the first step of Ungerboeck is applied to each branch in diversity and to each sampling phase. The common terms for all hypotheses are eliminated. This provides Mh (n) = Bh (n) + Ch (n) + Dh (n) where At this point, the second step of Ungerboeck, picking up terms from different iterations, is not carried out. Instead of this, in accordance with the present invention, the order of sum is changed within the same iteration. This leads to the novel way: D '' The terms in parentheses [] do not depend on the hypothetical data. Therefore, these terms can be calculated in a metric pre-processor and used for multiple times when calculating the metric systems in a metric processor. In the term between the parentheses [] in B '^ fn), the internal sum through M, provides an elevation towards a derivation FIR filter M in the data received in branch d in diversity. Since this sum is different for each value of j, that is, different rays or different channel inclinations, this implies a group of FIR filters in each branch, one for each value of j. The filter data and the coefficients both would be complex. For the case of the separate equalization in T, when M = 1, each FIR filter would simply be a 1-bypass filter, which is equivalent to a multiplier. The output of each FIR filter does not have to be calculated for each displacement of the input data. In general, only one output is needed for each M displacement of the input data. This output corresponds to the case when the contents of the FIR data are the values r (nT) to r ((n + (M-l) / M) T). In this way, it can be understood equivalently by a register that, during time n, is loaded with these data values. Then, the calculations of the channel derivation for time n is used to multiply the content of the record, producing the results or products in an accumulator.

The external sum through d involves combining the outputs of the FIR filter groups towards a group of combined outputs, each element corresponding to a derivation j of a different channel. Also, for the terms C'h (n) and D'h (n), the terms in parentheses [] only depend on channel derivations. If the channel derivations do not change, or if they are only updated every so many iterations, then these terms in parentheses [] can be reused in multiple iterations. These implementation considerations lead to the following form of the present invention. The metric delta system is calculated as Mh (n) = B'h (n) + C'h (n) + D'h (n) where Y Q, n) -? ? rJUn + mlMjT) c ^ < f, n) H ^ (?) 0-1 M-l gV, n) »? ? w < »C, (f, n) c ^ (k, n) k > j M-0 to? * 0 The terms e (j, n) and g (j, k, n) can be pre-calculated in a metric preprocessor. This is referred to as the metric multipliers, since they are used in multiplications to form the equalizing metric systems. Then, the metric processor would implement the Viterbi algorithm, taking advantage of the pre-calculated quantities. It should be noted that additional techniques of shift memory and processing requirements can be employed when calculating metric multipliers. For example, you can define Xd, m (n) as the square root of Wd / m (n). Then, the terms C m (j / n) can be replaced with Cd m (j, n) = d, m (n) Cd, m (j, n) and replace Rd ((n + m / M) T ) with rd ((n + m / Mt) = X?? a (n) r?? ((n + m / MT) Alternatively, we could define C'd m (j, n) = wd, m ( n) cd, m (J'n) And use a mixture of premium and non-premium channel models, and the factors of 2 can be suppressed from B '? 1 (n) and D' ^. n), and a The factor of 1/2 can be included in C'h (n) Finally, all terms can be denied, providing a metric system that will be maximized It should also be noted that for certain modulation projects, all transmitted symbols have the same amplitude, that is, | Sft (n) | which is the same for all hy n In this case, the terms C '^ fn) and f (j, n) do not need to be calculated. Also, in certain modulation projects, the term S ^ (n-j) S ^ * (n-k) does not need to be calculated J (J-l) times, since it typically acquires possible values in smaller quantities. One possibility is to store these values in a query box whose Index is determined by the hypothetical values of the symbol. Finally, for certain modulations, the values of the hypothetical symbol are simple values so that the multiplication operation is not necessary. For example, BPSK provides the hypothetical symbol values of +1 and -1, so that multiplication can be replaced by a possible sign change. Similar properties are maintained for QPSK modulations or QPSK based modulations. First we must take into account the case where the channel model is assumed to be static, that is, it does not vary in time to demodulate the data associated with a specific synchronization field. This can happen in a digital TDMA system that uses a short burst duration as in the GSM system. This implies that cd, m (3 n) = Cd Itl (j), e (j, n) = e (j), f (j, n) = f (j), and g (j, k, n) = g (j, k), independently of n. Therefore, a single set of channel calculations is needed for each branch in diversity and sampling phase. Also, if the optimal combination is used, the weights Wd m (n) become Wd ^ m. The concepts described above are implemented in a receiver according to an embodiment of the present invention which is illustrated in Figure 1. The radio signals are received by a plurality of antennas 100. Each antenna signal is processed by a processor 101 of branching signal, which produces sample of baseband complex data as well as synchronization information. This information includes time or synchronization information and possibly initial channel derivation calculations that can be obtained from the correlations of the known synchronization sequence with the received data. The branch signal processor may also include a buffer for storing the data samples. A channel calculator 104 uses the synchronization information and the sampled data corresponding to the synchronization field to determine the calculations of the channel derivation. The channel calculator 104 may use a variety of methods to determine the calculations of the channel derivation. One of these methods is described in U.S. Patent Number 5, 031, 193, where the synchronization correlation values are simply maintained as the channel derivation calculations. An alternative is to find channel derivation calculations that, in at least the sense of frames, can better predict the received data that corresponds to the synchronization field. Because these channel derivation calculations can be noisy, it may be useful to further process these channel derivation calculations. The data samples and the channel derivation calculations are processed by a metric preprocessor 102 which essentially calculates the metric multipliers e (j), g (j, k), and f (j), as required. These multipliers are supplied to a metric processor 103 which carries out the Viterbi equalization process. This results in a detected sequence of symbols that can be converted into an information bit stream. The data stream or bit may be in soft form or passed off with soft information and may be used in subsequent decoding. In making the conversion to data values, the metric processor 103 effectively demodulates the data that may have been modulated using any form of modulation, such as BPSK, GMSK, QPSK, DBPSK, DQPSK, or 7/4-offset DQPSK. Because the channel is static, the terms f (j) / and g (j / k) need only be calculated once, or once per predetermined demodulation interval. Also, in the metric processor 103, the terms C'h and D '^ can be pre-calculated for all possible hypotheses and stored in a frame to be used in each iteration. Yes | Sh (n) | it is the same for all hypotheses, as is the case commonly used for digital modulation projects, so the terms f (j) and C'n do not need to be calculated. One embodiment of the branch signal processor, one by branch diversity, is described in greater detail in Figure 2. The radio receiver 200 converts the radio signal into a received signal from the complex baseband sampled. There are many ways to carry out this conversion, even though most involve some form of filtering, mixed with local oscillator signals and amplification. A well-known approach is shown in Figure 2 of U.S. Patent No. 5,031,193. Another approach involves the use of polar-record quantification, followed subsequently by conversion into complex samples as shown in U.S. Patent No. 5,048,059, which is incorporated herein by reference. The received signal samples are stored in a buffer 201. If the digital cellular system is TDM or TDMA, the buffer memory allows the storage of at least one time interval of the data. In an FDM or FDMA system, the buffer 201 may be omitted. The synchronizer 202 determines which of the data samples must be maintained for further processing by selecting one or more of the sampling phases, wherein each sampling phase corresponds to maintaining 1 sample every T seconds. For fractionally separated equalization, two or more sample phases are maintained (the phases of sample M are maintained). The synchronization branch is well known. Typically, the synchronizer correlates the received data samples to one or more synchronization sequences. Sometimes only a portion or sub-sequence of the known synchronization sequence is used. The correlation values are typically combined in a certain way and then compared with each other to determine the position of the phase or phases of synchronization and best sampling. This sampling information is provided to a decimator 203 which, for each selected sampling phase, keeps only one sample of the data every T seconds. The synchronizer also supplies synchronization information, such as frame synchronization information and channel derivation information. The channel derivation information can be directly related to the correlation values calculated during synchronization. An alternate modality of the diversity branch signal processor is shown in Figure 5. The radio receiver 500 converts the radio signal into a received signal of complex sampled baseband which can be stored in a buffer 501 as described in FIG. the foregoing with respect to Figure 2. Unlike the embodiment shown in Figure 2, the radio receiver 500 does not oversample the received data. Instead, the radio receiver provides MT data samples per symbol period T. Therefore there is no need for decimation of the complex data stream. However, the synchronizer 502 only provides synchronization information. The metric pre-processor is described in more detail in Figure 3. This pre-processor is used to calculate the metric multipliers e (j), g (j, k) and f (j) if necessary. For each diversity branch, there is a FIR filter group 301 consisting of a plurality of 300 FIR filters. There are J filters 300 FIR in each filter group 301, that is, a FIR filter per beam in the channel models. The outputs of these FIR filters are added in such a way that the outputs of the first FIR filter are summed, the outputs of the second FIR filter are added, and so on. These sums are carried out by the adders 302. This provides J metric multipliers for the term B, j1 (p). The metric pre-processor also calculates the metric multipliers for C'n (n), and D'n (n). Each of these terms consists of a sum of the channel derivation products. These can be calculated using the channel derivation processor 303, and need only be calculated once if the channel is static. All pre-processor results are then stored in a buffer 304. The embodiment illustrated in Figure 1 is applicable for the case where the channel does not change appreciably for the data associated with a specific synchronization field. However, when this is not the case, channel derivations must be calculated adaptively or predicted during each branch in diversity. A receiver to be used in this situation is illustrated in Figure 4, where the equal elements correspond to the same elements in Figure 1. In this embodiment, the calculations of the channel derivation can be obtained and the synchronization process and these calculations can also be refined by training a channel follower through the synchronization field. Finally, tentative symbol detections may be fed back to the branch diversified channel trackers to allow for the refinement of the calculations or channel predictions. In Figure 4, each diversity branch has an associated channel follower 404. This follower calculates or predicts the channel derivations corresponding to the sampled data that it provides to the metric pre-processor 402. The tracker can be started using synchronization information that is provided by the diversity branching signal processor 401. A method for both initiation and follow-up is provided in U.S. Patent Application Number 07/942, 270 entitled "A Method of Forming a Channel Estimate for a Time-varying Radio Channel" by Larsson et al., Filed September 9, 1992, which is incorporated herein by reference. The channel follower can have two modes: a training mode and a decision-driven mode. During the training mode, the channel follower uses knowledge of what was transmitted to the channel derivation calculation train. This knowledge may correspond to the synchronization sequence or other known sequences within the stream of transmitted symbols. During the decision-driven mode, the channel tracker takes the tentatively detected data from the metric processor 403 and assumes that it is correct. This allows the channel tracker to update the channel derivation calculations. Those skilled in the art will be aware that there are many forms of channel tracking and prediction. The metric pre-processor 402 calculates the necessary metric multipliers of the sampled data and the calculations or predictions of channel derivation. The metric 403 processor uses the metric multipliers with the sequence calculation algorithm to provide both tentative and final detected data. Tentative data is used for channel tracking purposes. During time n, the tentative data corresponds to the decisions in the transmitted symbols s (n-upd) to s-upd-J + 1), where upd is an update delay design parameter, which is true integer not negative. Knowledge of these symbol values allows channel trackers to predict certain received data values and compare predictions '' with actual values. The difference can be used to update the channel derivation calculations that are used to predict channel values for time n + l.

In accordance with a preferred embodiment of the present invention, the present invention is used in a receiver for digital cellular signals of TDMA IS54. The antenna diversity of two branches (D = 2) and separate equalization T / 2 (M = 2) would be used. The number of channel (J) derivations would be 2. Even though the modulation is / 7"* / 4 displacement of DQPSK, the 77" / 4 shift can be removed from the received data. For a data sample (n + m / D) T, this is achieved by multiplication with exp (-77 ~ (n + m / M) 4). The resulting data can be treated as a stream of QPSK symbols. The mode shown in Figure 4 will be used to demodulate the received signal. In iteration n, the metric multipliers are provided by: < ) ^ 0?) C (l ^) r0 (n-0 + ww (n) c (l ^) r0 ((n + - / 2) 7) + wu () lV) rl? 2) + wu (?) Cy (U) rl ((«+ l / 2) 2 The terms f (j, n) are omitted since I S ^ (n) 12 = i for all h. When the metric system is formed, the term C'h is also omitted. One mode of the metric pre-processor is provided in Figure 6. The received data values are stored in a data buffer 600, the channel derivation calculations are stored in a channel bypass buffer 601, and the of weighting are stored in a weighting buffer 602. During time n, the metric multipliers are calculated one at a time. A selection device 603 selects two complex values, either a data value and a channel derivation value or two channel derivation values. These selected values are multiplied into a complex / complex multiplier 604 that multiplies one of the values with the conjugate of the other value to form a first product or result. A selection device 605 selects a scalar weighting factor that is multiplied by the first product or result in a complex / scalar multiplier 606, which multiplies the complex and scalar values to produce a second product or result. This second product or result is accumulated in an accumulator 607. This process is repeated four times to produce a specific metric multiplier. Then, the contents of the accumulator are readjusted and the process is repeated until all three metric multipliers are produced. In the metric processor, during time n, there are four previous states that correspond to the hypothetical symbols (n-l) th and four current states that correspond to the values of the hypothetical n'th symbol. This provides 16 possible hypotheses that must be taken into account. The metric delta system associated with a given hypothesis h is provided by: Metric multipliers are not multiplied explicitly with hypothetical symbol values, since these hypothetical values for Sh * (n), Sh * (n-1), and Sh * (n) Sh (nl) are all +1, -1 , + i, and -i, where i represents the imaginary unit number. Therefore, the multiplication of a + ib by these values is equivalent to the formation of a + ib, -a-ib, -b + ia, and b-ia, respectively. Therefore, the 16 metric branching systems can be formed by a set of adders and inverters that add up and deny the real and imaginary parts of the metric premultipliers. It is well known to those skilled in the art that other sequence calculation algorithms in addition to the Viterbi algorithm can be used to exploit the metric systems to find a detection decision. for example, sequencing decoding and other non-exhaustive research methods may also be used. It will be known to those skilled in the art. that the aforementioned invention can be applied to other forms of diversity such as frequency diversity, diversity in time and diversity in polarization. Finally, the aforementioned invention can be used to efficiently implement a combination of MLSE equalization and decision feedback as described in the articles of M.V. Eyuboglu and S.U.H. Qureshi, "Reduce-state sequence estimation and set partitioning and decision feedback", IEEE Trans. Com un., Volume 36, pages 13 to 20, January 1988. It will be appreciated by those skilled in the art that the present invention may be encompassed in other specific forms without deviating from the spirit or essential character thereof. The modalities disclosed at present are therefore considered in all respects as being illustrative and not restrictive. The scope of the invention is indicated by the appended claims rather than the foregoing description, and all changes that fall within the meaning and scale of equivalence thereof are intended to be encompassed herein.

Claims (14)

CLAIMS:

1. A digital communications receiver comprising: means for processing signals of a plurality of branches in diversity to produce complex received data samples and synchronization information of the received signals; a means for forming channel derivation calculations of the data samples and synchronization information; a means for forming weighting factors using the data samples, the channel derivation calculations, and the synchronization information; a means to pre-process the data samples, the channel derivation calculations and the weighting factors to produce metric multipliers; and a means for combining the metric multipliers with hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm producing a demodulated data stream. A receiver according to claim 1, wherein the signal processing means comprises: means for processing radio signals to produce complex data samples; and means for synchronizing the receiver using the complex data samples in order to produce synchronization information and initial channel derivation calculations. A receiver according to claim 2, wherein the signal processing means further comprises: a means for storing the complex data samples. A receiver according to claim 1, wherein the signal processing means comprises: means for processing the radio signals to produce complex data samples; means for synchronizing the receiver using the complex data samples to produce time information and channel model information; and a means to decimate complex data samples using sampling phases that are determined by synchronization. A receiver according to claim 1, wherein the pre-processing means comprises: means for filtering each stream of branch data samples using a group of FIR filters and filter coefficient based on the derivation calculations of channel; and a means to calculate the products and their sums using the channel derivation calculations. 6. A digital communications receiver comprising: means for processing the signals of a plurality of branches in diversity to produce data samples to receive the complexes and synchronization formation of the received signals; means for forming channel derivation calculations that vary in time from the data samples, the synchronization information and the tentative data detections; a means for forming weighting factors that vary in time using the data samples, the channel derivation calculations in synchronization information and tentative data detections; a means to pre-process the data samples, the channel derivation calculations and the weighting factors to produce metric multipliers; and a means for combining the metric multipliers with hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm producing a stream of demodulated data. A receiver according to claim 6, wherein the signal processing means comprises: means for processing the radio signals in order to produce complex data samples; and means for synchronizing the receiver using the complex data samples to produce synchronization information and initial channel model information. A receiver according to claim 7, wherein the signal processing means further comprises: a means for storing the complex data samples. 9. A receiver according to claim 6, wherein the signal processing means comprises: a means for processing radio signals to produce complex data samples; means for synchronizing the receiver using the complex data samples in order to produce synchronization information initial channel model information; and a means to decimate complex data samples using sampling phases determined by synchronization. A receiver according to claim 6, wherein the pre-processing means comprises: a means for filtering each stream of branch data samples using a group of FIR filters and the filter coefficients based on the channel calculations; and a means to calculate the products and their sums using channel calculations. 11. A method for pooling and diversity combining in a digital communications receiver comprising the steps of: processing signals of a plurality of branches in diversity to produce complex received data samples and signal synchronization information received; forming channel derivation calculations from the data samples and synchronization information; forming weighting factors using the data samples, channel derivation calculations and synchronization information; pre-process the data samples, channel derivation calculations and weighting factors to produce metric multipliers and combine the metric multipliers with hypothesized data sequences to generate and accumulate the metric systems using a sequence calculation algorithm producing a current of demodulated data. 1

2. A method for joint equalization and diversity combining in a digital communications receiver comprising the steps of: processing the signal of a plurality of branches in diversity to produce complex received data samples and the synchronization information of the received signals; forming channel derivation calculations that vary in time from each of the data samples, synchronization information and tentative data detections; form weighting factors that vary in time using data samples, channel derivation calculations, synchronization information and tentative data detections; pre-process the data samples, channel derivation calculations and weighting factors to produce metric multipliers, and combine the metric multipliers with hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm producing a current of demodulated data. 1

3. In a digital communications system of TDMA based on IS54, a receiver comprising: means for processing the signals of a plurality of branches in diversity to produce complex received data samples and synchronization information of the received signals; means for forming channel derivation calculations that vary in time from the data samples, the synchronization information and the tentative data detections; a means for forming weighting factors that vary in time using the data samples, channel derivation calculations, synchronization information and tentative data detections; a means to pre-process the data samples, the channel derivation calculations and the weighting factors to produce metric multipliers; and a means for combining the metric multipliers with hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm, producing a stream of demodulated data. 1

4. In a TDMA digital communication system based on IS54, a method for joint equalization and diversity combining in a digital communications receiver comprising the steps of: processing the signals of a plurality of branches in diversity to produce samples of complex received data and synchronization information of the received signals; form channel derivation calculations that vary in time from data samples, synchronization information and data detections, and attempts; form weighting factors that vary in time using data samples, channel derivation calculations, synchronization information and tentative data detections; pre-process the data samples, the channel derivation calculations and the weighting factors to produce metric multipliers; and combining the metric multipliers with the hypothesized data sequences to generate and accumulate metric systems using a sequence calculation algorithm, producing a stream of demodulated data.