MXPA97003119A

MXPA97003119A - Method and apparatus for calculation of ca

Info

Publication number: MXPA97003119A
Application number: MXPA/A/1997/003119A
Authority: MX
Inventors: Skild Johan; Ericsson Linus; Eriksson Perolof
Original assignee: Telefon Ab L M Ericsson
Priority date: 1994-10-31
Filing date: 1997-04-29
Publication date: 1997-06-01

Abstract

The present invention relates to a method for forming a channel calculation in a digital radio communication system, comprising the steps of: forming, from a received training sequence, a first channel calculation having a predetermined number of derivations form, from the same training sequence received, at least one additional channel calculation that has fewer derivations than the first channel calculation, and combine the first channel calculation and the additional channel calculation to form a channel calculation combines

Description

"METHOD AND APPARATUS FOR CALCULATING CHANNEL" TECHNICAL FIELD The present invention relates to a method and an apparatus for forming a channel calculation in a digital radio communication system.

BACKGROUND OF THE INVENTION In TDMA radio communication systems (TDMA = Time Division Multiple Access) and other frame-based communication systems, the information is transmitted on a channel in the form of frames or signal bursts. In order to synchronize the receiver to these signal frames, each signal frame comprises a known synchronization word at a predetermined position within each signal frame. For example, in the European GSM system for mobile telephony, this synchronization word is 26 bits long. When the receiver waits ur. new signal frame from the transmitter, a training sequence that is identical to the transmitted synchronization word is generated by a training sequence generator in the receiver. The received signals are compared to the locally generated training sequence, and when the best possible correlation between this sequence and the received signals is obtained, synchronization is considered to exist between the locally generated and received signal. In addition to the synchronization itself, the training sequence is also used for channel calculation. Since the radio channel is frequently subjected to multiple path propagation, the receiver comprises some kind of equalizer to eliminate this phenomenon. The equalizer requires a time-limited calculation of the channel impulse response. The impulse response can be obtained from the correlation signal. Forney [1] and Ungerboec [2] describe two different algorithms that, given the response of the channel impulse and the Gaussian channel noise with known correlation properties, determine the most likely sequence sent. Both algorithms will work properly, but with degraded operation, if an approximate calculation of the impulse response of the channel is used instead of the true impulse response, or if the noise is not Gaussian. The equalizer makes use of the channel calculation to initiate and update, v.gr, the filter derivations. An example is the Maximum Possibility Sequence Calculation (MLSE) detector, where an FIR filter is used directly with a channel model [2]. Another example is the decision feedback equalizers (DFE) [4], where the derivations of both the forward and feedback filter are calculated from the channel calculation. When the channel is calculated from a received synchronization word, the calculation will contain noise, since the received data is noisy and the training sequence is of finite length. Even a channel calculation that is continually updated will be noisy. The noise content in the channel calculation will be very high in eg fading dives, since the level of the signal in that case is low compared to the level of noise or interference, providing a low ratio of signal to noise in the received data. The synchronization process is also altered by receiver noise. Incorrect and unstable synchronization provides an incorrect channel calculation as a side effect, thus causing a considerable loss in receiver operation. The noisy channel calculation provides incorrect settings for the filter leads in the detector or equalizer, v.gr, the MLSE will have an incorrect FIR channel model. This causes degraded demodulation operation in a manner similar to the degradation of noisy samples received. The effect is that the interference or noise in the received samples has a double impact - first the channel model deteriorates, and then the incorrectly adjusted equalizer will have to equalize and demodulate the noisy samples. A common feature of the prior art described is that the channel calculation is used directly in the detection or equalization procedure without taking into account the noise content of the calculation. Nevertheless, for example, the MLSE is optimized in the sense that it provides the most possible demodulated symbol sequence only if the receiver noise is Gaussian additive and the channel calculation is accurate. It is not optimal if there is noise in the channel calculation. A method for reducing the influence of noisy shunts in the channel calculation is proposed in U.S. Patent Number 5,251,233 (Labedz et al.). In the same it is suggested to print the derivations in the channel calculation that are below a certain threshold value, thus reducing the noise contribution of the noisy branches with low useful signal content. By completely eliminating some derivations of the channel calculation, however, you can remove the vital information, since it is very difficult to distinguish between the energy of a useful signal and the noise energy in a derivation.

Patent EP-A-0 535 403 describes a method in which a channel calculation of a current burst is combined with a channel calculation of a previous burst. Both calculations "have the same number of derivations.

COMPENDIUM OF THE INVENTION Therefore, an object of the present invention is to provide a method and apparatus for forming an improved channel calculation in a digital radio communication system. The method of the invention is characterized by the features of claim 1. The apparatus of the invention is characterized by the features of the claim 9. The present invention is based on the observation that a long channel calculation (a calculation having many derivations) will contain more noise than a short calculation (which has smaller derivations). This is shown in the annex, where the noise of the derivation is provided by equation (9). In this equation N represents the total number of samples used in calculating the channel calculation and M represents the number of derivations in the channel calculation. Increasing the number of derivations M in the channel calculation, therefore will also increase the noise of the derivation, in addition the contribution of total noise in the demodulation process in the equalizing detector is also proportional to the number of derivations in the channel calculation . From this point of view, therefore, it is desirable that there be a channel calculation that is as short as possible. On the other hand, a short channel calculation can, in a case where there is a great dispersion of time, cause the receiver to exclude the energy received outside the scope of the calculation, ignoring in this way the important information present in the signal. The short channel calculation, however, includes the part of the signal with the most energy content. The solution to this problem, according to the present invention, is to combine a long channel calculation with at least one shorter channel calculation. This combines the properties of both. 1. It contains reliable information (not so loud) in the part of the received signal that has the highest amount of energy. 2. It also contains information about the signal dispersion caused by the time dispersion over a longer time interval.

BRIEF DESCRIPTION OF THE DRAWINGS The invention, together with the objects and additional advantages thereof, can be better understood by referring to the following description taken together with the accompanying drawings, in which: Figure 1 is a diagram functional of an apparatus in accordance with the present invention; Figure 2 is a flow chart illustrating the method according to the present invention; Figure 3 illustrates the manner in which the channel calculations are combined in one embodiment of the present invention; Figure 4 illustrates the manner in which the channel calculations are combined in another embodiment in accordance with the present invention; Figure 5 illustrates the manner in which the channel calculations are combined in a further embodiment in accordance with the present invention; and Figure 6 illustrates a simplified calculation of the amplitude of a complex number that can be used in a preferred embodiment of the invention.

DETAILED DESCRIPTION OF THE PREFERRED MODALITIES The apparatus and method of the present invention will now be described with reference to Fugures 1 and 2. In Figure 1, an A / D converter receives analog samples b (nTs) and converts these samples into a sequence of digital bn samples. These digital bn samples are sent to a Maximum Possibility Sequence Calculator 12, which sends a sequence of symbols one detected. The digital bn sequence is also sent to a correlator 16 which correlates the sequence bn with a sequence u ^ of locally generated training received from the training sequence generator 17. Correlation values Cj_ of correlator 16 are used to synchronize the burst (step 110 in Figure 2). This synchronization step will be further described below. The methods for burst synchronization will be described below with reference to the European GSM system. In this system, a pairing word comprises 26 bits. The 16 central bits in this word have good correlation properties when they correlate with the whole synchronization word, that is, a maximum correlation = 16 in the central position and a correlation of 0 in the remaining ten positions (C (k) = [0 0 0 0 0 16 0 0 0 0 0]). These 16 core bits are generated as a training sequence in a training sequence generator in the receiver. This training sequence is used to form, for example, 11 correlation values Cj_ with the received signal frame. According to a burst synchronization method, the final synchronization position is selected by comparing the mutually displaced windows, each containing 5 correlation values, with respect to energy contents, and selecting the time position of the window with energy maximum as the synchronization position. Another method of burst synchronization is described in Patent Number EP-A-0 551 803. Since two channel calculations will be combined, burst synchronization is carried out for both short and long calculations. The synchronization step is carried out in the synchronizers 18 and 19, respectively, (Since the calculations are of different length, they will not necessarily be synchronized to the same burst position). The synchronization positions are sent to the channel calculators 20, 22 for the long and short channel calculations, respectively. These calculators calculate the channel calculations around the respective synchronization positions, as will be described further below.

In the ANNEX it has been shown that the derivation noise can be calculated according to the formula (9). This formula indicates that the noise of the derivation will be reduced by using as many bn samples as possible (increasing N). Due to this reason, the long hjcL calculation is recalculated using so many of the 26 samples of the sequence uj? of training as possible, and the word synchronization bn received. Therefore, the channel calculator 20 will calculate five correlation values (in GSM) of the samples of N-M + l = 25-5 + 1 = 22 (All 26 samples are actually used since 5 correlation values are formed each based on 22 samples and are displaced in a sample). These calculations are carried out in the same way as in correlator 16, but since we now know the synchronization position, the entire training sequence can be used to form the five derivations of the long channel calculation h] L. The described procedure corresponds to step 120 in Figure 2. The short channel calculation hvs is similarly formed in the channel calculator 22. However, in this case, less than five leads have to be calculated (M < 5). In a preferred embodiment of the present invention, the short channel calculation comprises only one derivation, which means that samples of N-M + l = 26-l + l = 26 can be used for the channel calculation. Therefore, in this case, the entire training sequence is used to form a single correlation value with a significantly reduced derivation noise, the procedure described corresponds to step 130 in Figure 2. The calculations h] L and h]? Calculated channel numbers are combined in averaging circuit 24 (corresponding to step 150 in Figure 2). An example of this averaging process is illustrated in Figure 3 (to simplify the illustration only the amplitudes of the calculations are shown). In this mode, calculations are combined with their original burst synchronization (since the short calculation illustrated only has one derivation assumed to have derivations with values of zero in all other time positions), this may result in a situation , where the maximum values of the two calculations are not in the same time position, as indicated in Figure 3. Since the short and long channel calculations have been synchronized separately with the burst, it may be preferable to synchronize primer mutually in the calculations of the calculated channel. This is indicated by line 26 in Figure 1, wherein the channel calculator 20 informs the synchronizer 18 of the appropriate synchronization position to be used for the short channel calculation. this corresponds to step 140 in Figure 2. There are different ways to obtain mutual synchronization between the short and long channel calculations. A method is illustrated in Figure 4. According to this method, the maximum derivation of the short calculation h ^ s is synchronized with the position of the maximum derivation of the calculation h]? L long. This modality implies that the derivations of the short calculation can be shifted as indicated by the derivation of ions in Figure 4. This ion derivation represents the burst synchronization position of the short calculation. The derivation will be recalculated in the calculator 22 and then moved to the calculation position h ^ s of short line derivation. This timing of the calculation is reasonable, since the maximum derivations of the long and short cus- toms usually have coincident positions. Another method of synchronization of mutual channel calculation is illustrated in Figure 5. Here, the shortwave synchronization position of the short calculation (which contains only one derivation in the example), h] ^ is indicated by the derivation of ions. However, this derivation will not be recalculated. Instead, the derivation corresponding to the position of the maximum derivation of the long calculation will be recalculated and averaged with the maximum derivation of the long calculation. As in the previous modality, this calculation timing is reasonable, since the maximum derivations of the long and short calculations will usually have coincident positions. After the mutual synchronization of the short channel calculation hvs and the long channel calculation hj ^, these calculations are combined in averaging circuit 24. In a simple mode currently preferred, averaging circuit 24 performs simple averaging of the corresponding derivations of the long and short calculations, as illustrated in Figures 3, 4 and 5. In a more elaborate mode, it can be formed calculating the weighting factors of the reliability measurements (noise measurements) for each of the two channel calculations. The combined hvc channel calcual is sent to the maximum possibility sequence calculator 12 to adjust the filter coefficients thereof. In case the short channel calculation contains only one derivation, or the previously described process will be simplified observing that the maximum correlation value c can be directly used to represent this derivation. In this way, if a reduction in the complexity of the calculation is desired, it may be sufficient to provide the maximum value of C_ as the short calculation (without having to recalculate the short calculation). In this mode, the position of the burst synchronization of this calculation can be used as a reference position for the largest derivation of the long calculation. In a more elaborate mode, the energy of the maximum derivation of the long calculation is compared with the total energy of the long calculation. If the maximum derivation is very dominant, this indicates that there is very little time dispersion, in this case, a single derivation model of the channel is a very good model and a short calculation of a single derivation is recalculated using the whole sequence of training. On the other hand, if the time dispersion is large, the channel calculation will be dispersed and the maximum derivation will not be the dominant one. In this case, a single derivation mode is not so good and there is not much to gain by recalculating this derivation, using the complete training sequence. Therefore in this case, the single derivation of the burst synchronization is used as the short calculation. Another simplification that can be carried out is to take into account only the amplitude of the short calculation and ignoring its phase (in case the short calculation contains only one derivation). It is a non-coherent combination, it does not imply a significant loss in operation, since the phase of the two largest derivations of both calculations usually remain close to each other. An additional simplification can be carried out by calculating the amplitude of the short calculation by adding the absolute values of its real and imaginary parts, respectively, as illustrated in Figure 6. In this approximation z is approximated by | x | + | y | , where x and y are the real and imaginary parts, respectively. In the illustrated modes, two channel calculations have been combined. However, it is evident that it is possible to combine more than two calculations. In this way a feasible modality would be a combination of three channel calculations of different lengths. It will also be appreciated that the present invention is applicable to systems other than the GSM system described. The methods described above are simple ways to improve receiver performance without dramatic increase in complexity. The improvement in receiver operation is approximately 1 dB for some propagation conditions in the case of GSM. It will be understood by those skilled in the art that various modifications and changes may be made in the present invention without deviating from the spirit and scope thereof, which is defined by the appended claims.

REFERENCES [1] G, D. Forney, Jr., "Maximum-Likelihood Sequence Estimation of Digital Sequences in the Presence of Intersymbol Interference," Volume IT-18, pages 363 to 378, May 1972. [2] Gottfried Ungerboeck "Adaptive Maximum-Likelihood Receiver for Carrier- Modulated Data- Transmission System ", IEEE Trans. on Communications, Volu in COM-22, pages 624 to 636, May of 1974. [3] Simon Haykin, "Adaptive Filter Theory", pages 307 to 316, Prentice-Hall, Englewood Cliffs, NJ, 1986. [4] S.U.H. Quereshi, "Adaptive Equalization", Proc. IEEE, Volume 73, pages 1349 to 1387, September 1985.

ANNEX Minimal-quadratic calculation of the channel The model used for the channel is a FIR filter with filter leads. { h ^} modeling the propagation and a source ej_ of additive white Gaussian noise (AWGN) modeling the noise. It is similar to Haykin's linear regression model (Simon Haykin, "Adaptive Filter Theory", Prentice Hall, 1986, pages 307 to 316). The expression for the samples b¿ of the received discrete signal is: M-l bi = L nkui-k + ei (1) k = 0 where is the input signal to the channel and M is the length of the channel calculation (filter derivations { hj ?.}.). In this way, the impulse response of the radio channel is limited to the M samples of h ^. In GSM, for example, M = 5. The impulse response is an unknown unknown parameter of the training sequence. In the GSM example, the training sequence is N = 26 symbols long. The minimum-quadratic method (see Haykin [3]) is used to calculate the derivation model M (5 derivations in GSM). The input data of the training sequence is with windows using the covariance method. Therefore, there are N-M + l samples received (22 in GSM) to be used for the calculation. The input is placed as an array A and the samples b received, the error e and the impulse response h are placed in vectors as follows: UM-1 U , • • UN-1 A = uM-2 UM-1 • • • • UN ~ 2 (2) U0 C7t. UN-M V) b = [JD0 b? . . . ^ -] T (3) e = [eg e? . . . eN-M] t (4) h = [ho h? . . . ? ^ f_2] T (5) where T represents a transposition. In vector form, the channel model can be expressed (for the training sequence) as: b = Ah + e (6) The minimum quadratic calculation h is (see Haykin) h = (AHA) -1AHb = F_1AHb (7) where F = A ^ A is the correlation matrix for determining the training sequence. Here H represents Hermitean transposition. The variance of each channel calculation element h¿ is called N¿, and it depends on the noise variance s ^ and on the training sequence used. N is a diagonal element of s ^ F- !, where F is the determinant correlation matrix defined above. If the training sequence is close to white (which is the case in e.g., GSM), the correlation matrix will be: F »(N- l) I f l * (8) W-M + l where I is the identity matrix. The conclusion is that the noise of the derivation is: Ni * s¿ N-M + l

Claims

R E I V I N D I C A C I O N E S:

1. A method for forming a channel calculation in a digital radio communication system, characterized by: forming a first channel calculation having a predetermined number of derivations; forming at least one additional channel calculation that has fewer derivations than the first channel calculation; and combining the first calculation and the additional channel calculations to form a combined channel calculation.

2. The method according to claim 1, characterized by averaging the corresponding derivations of the first calculation and the additional channel calculations, the additional channel calculation (s) is filled with values of zero in the positions of the derivation corresponding to the derivations of the first channel calculation that do not correspond in the additional channel calculation (s).

3. The method according to claim 2, characterized by considering only the amplitude of the derivation and ignoring phase of the derivation of the phases of the calculation (s) of additional channel in the averaging.

The method according to claim 3, characterized to approximate the amplitude of each derivation of the additional channel calculation (s) with the sum of the amplitudes of its real and imaginary parts, respectively.

5. The method according to claim 2, 3 or 4, characterized in that the single additional channel calculation has a single derivation.

The method according to claim 5, characterized in that the single derivation of the additional channel calculation is averaged with a maximum amplitude derivation in the first channel calculation.

The method according to claim 5, characterized by averaging a derivation of the maximum amplitude of the first channel calculation, with the single derivation of the additional channel calculation as calculated in the same time position.

8. The method according to claim 2, 3, 4, 5, 6 or 7, characterized by weighted averaging the first and the additional channel calculations.

9. An apparatus for forming a channel calculation in a digital radio communication system, characterized by: a first means (20) for forming a first channel calculation having a first predetermined number of derivations; a second means (22) for forming at least one additional channel calculation having fewer derivations than the first channel calculation; and a means (24) for combining the first and the additional channel calculation (s) to form a combined channel calculation.

10. The apparatus according to claim 9, characterized by averaging means (24) for averaging the corresponding derivations of the first and the calculation (s) of additional channel, the calculation (s) of additional channel is filled with values of zero in the derivation positions that correspond to the derivations of the first channel calculation that do not correspond to the additional channel calculation (s). The apparatus according to claim 10, characterized in that the means of averaging (24) consists only of the amplitude of the derivation and ignoring the derivation phase of the derivations of the calculation (s) of additional channel in the averaging. L2. The apparatus according to claim 11, characterized in that the averaging means (24) approaches the amplitude of each derivation and the calculation (s) of additional channel with the sum of the amplitudes of its real and imaginary parts, respectively. The apparatus according to claim 10, 11 or 12, characterized in that a single additional channel calculation has a single branch. The apparatus according to claim 13, characterized by the first means (20) by synchronizing the single derivation of the additional channel calculation with a maximum amplitude branch in the first channel calculation. The apparatus according to claim 13, characterized in that the first means (20) that synchronizes a maximum amplitude derivation of the first channel calculation with the single derivation of the additional channel calculation, as calculated in the same time position. .