WO2003047193A1

WO2003047193A1 - Mlse equalizer for correlated interference signals

Info

Publication number: WO2003047193A1
Application number: PCT/IB2002/004935
Authority: WO
Inventors: Juergen Petersen; Thomas Wagner
Original assignee: Koninklijke Philips Electronics N.V.; Philips Corporate Intellectual Property Gmbh
Priority date: 2001-11-27
Filing date: 2002-11-26
Publication date: 2003-06-05
Also published as: AU2002347491A1; DE10158056A1

Abstract

An equal performing MLSE (Viterbi algorithm) in the presence of coloured noise is proposed. For each state (Sn) in the trellis an error sequence (e) is stored. This error sequence is determined from the difference between the received and the estimated signal sequence. For each state, the error sequence is fed into a corresponding noise-prediction filter, the output of which is subtracted from the received sequence. The overall branch metric (λ) thus is calculated from the difference of the sum of the estimated signal plus predicted noise and the received signal (rn). The coefficients (pi) of the noise-prediction filter are obtained from the auto-correlation function of the noise which might be estimated within the training sequence. A simplified version is also proposed, i. E. to use only one noise-prediction filter for all states.

Description

MLSE EQUALIZER FOR CORRELATED INTERFERENCE SIGNALS

The invention relates to a method and a device for suppressing interference signals in the case of a receiver with state-based equalization for detection of digital data signals received in time steps.

In the digital transmission of data signals, additive interference signals may occur, the adjacent values of which can be correlated more or less strongly with one another. In mobile radio systems, for example, common-channel interference and adjacent-channel interference in particular exhibit correlations by virtue of the filtering in the transmitter and receiver.

For the demodulation or equalization of a received signal, which may be interfered with as a result of strong pulse distortions, state-based equalizers are frequently used. The principle of state-based equalizers of this kind is described in, for example, the publication G. David Forney, Jr.: Maximum Likelihood Sequence Estimation of Digital Sequences in the Presence of Intersymbol interference, IEEE Transactions on Information Theory, vol. IT-18: p. 363-378,1972, in the publication M. Vedat Eyuboglu, Shahid U.H. Qureshi: Reduced-State Sequence Estimation with Set Partitioning and Decision Feedback, IEEE Trans. Comm., vol. 36, No. 1, Jan. 1988, and in the textbook J. Huber: Trelliscodierung [Trellis Coding], Springer- Verlag, Berlin, 1992, in particular pp. 37, 43, 101. State-based equalizers operate here in accordance with the principle of signal reconstruction and generally undertake a sequence estimation (MLSE = Maximum Likelihood Sequence Estimation) or symbol estimation (MLSSE = Maximum Likelihood Single Symbol Estimation).

The state transitions of a state-based equalizer can be visualized in a so-called trellis diagram (see J. Huber and G.D. Forney). The principle of signal reconstruction presupposes that an estimation of the channel pulse response can be undertaken. An estimation of this kind may take place, for example, by means of a training sequence.

According to theory, it is known that state-based equalizers only operate optimally when the additive interference-signal values in the scanning grid under consideration are uncorrelated, i.e. no significant correlation exists between adjacent interference-signal values. In the literature (see Lochmann, Dietmar, "Digitale Nachrichtentechnik" [Digital Telecommunications], Verlag Technik Berlin, pp. 351 to 355), a method of decorrelating correlated interference-signal values is described. It is here not just the interference signal, but also the useful signal that is filtered in the receiver. Although the interference power of the interference signal is reduced by the decorrelation, the useful signal is additionally distorted, so that, in some circumstances, energy-rich spectral components of the useful signal are attenuated. This is disadvantageous. In order to design a decorrelation filter, the power density spectrum of the interference must be known. Determination of this power density spectrum is difficult, particularly in the case of TDMA (Time Division Multiple Access) systems, since the type of interference, for example in the known frequency jumping method, can change from time slot to time slot, as a result of which the interval of the interference evaluation is restricted to one time slot.

Since the use of a decorrelation filter is problematic, a decorrelation is frequently dispensed with in the case of receivers normally encountered in practice, so a receiving filter which operates independently of the type of interference is then used. The interfering influence of correlated interference signals on the processing of the received signal is accordingly accepted. The receiving filter is then interpreted as a compromise as regards possible interference types.

J. Huber, pp. 172 to 175, describes an interference prediction in the case of an optimum Nyquist filter and stateless detection.

It is an object of the invention to propose a method and a device of the type specified above by means of which, in the case of a state-based equalization, the bit or block error rate during data transmission is reduced by means of a reduction in the effective power of a correlated interference signal. The above object is achieved according to the invention by the features as claimed in claim 1 as regards the method, and by the features as claimed in claim 10 as regards the device.

It is achieved thereby that, in the case of state-based equalization, the effective interference signal power is reduced without changing the useful signal. It is also advantageous that, with the method, only small losses occur in the case of uncorrelated interference, and that gains in common-channel interference and adjacent-channel interference are achieved. The invention will be further described with reference to embodiments shown in the drawings, to which, however, the invention is not restricted.

Fig. 1 is a simplified block diagram of a transmission system in baseband, in which symbols have the following meanings: " : Transmitted data symbols ^a" : Estimated data symbols in the receiver r_n: Received-signal values after the receiving filter h,: i-th coefficient of total channel pulse response p,: i-th coefficient of interference prediction L: Length of total channel pulse response in symbol periods

M: Number of coefficients of interference prediction u_n: Correlated interference signal

Fig. 2 is a trellis diagram of a state-based equalizer with eight states and one binary symbol alphabet. Fig. 3 shows a trellis step of a state-based equalizer with eight states and a binary symbol alphabet, with:

S_n: State in time step n. a_n-_ι(S_n): Hypothetical data symbols of past state transitions belonging to the selected path, which ends in state S_n, which symbols are stored in the path memory of the state S_n in time step n. e_n.^S_n): Hypothetical error signal values of past state transitions belonging to the selected path, which ends in state S„, which symbols are stored in the state memory of state S_n in time step n. λ'n(S_n|S_n-ι): Branch metrics of state transition (S_njS_n-ι) with interference prediction.

Λ_n(S_n): State metrics of state S_n in time step n.

Fig. 4 shows a recursive processing of the state trellis in a state-based equalizer, with:

Z : Number of states L: Length of the total channel pulse response in symbol periods

M : Number of coefficients of interference prediction

P: Length of path memory

D: Delay operator Λ_n(i): State metrics of i-th state in time step n an(i) = a_n(i), a_n-1(i), ... , a_n-p₊ι(i): Path memory to the i-th state in time step n <?n(i) = en(i), en.ι(i), ... , en-M+ι(i): Error signal vector to the i-th state in time step n Fig. 5 is a functional diagram of a branch metrics calculation with interference prediction.

A data source 1 transmits data symbols a_n via a modulator 2 and a transmission channel 3 to a receiver 4. A received-signal value r_n occurs in the receiver 4, downstream of a receiving filter 5. This value is processed in a state-based equalizer 6 with interference prediction. From a training sequence, which is contained in the received signal, the coefficients ho... of the total channel pulse response are estimated by a block 7, and then passed on to the state-based equalizer 6 and to a block 8. Further in block 8, the coefficients Pi ... are estimated from the received signal (for example the training sequence) for interference prediction and also passed on to equalizer 6. The coefficients pi ... correspond to the coefficients of a predictor filter.

The data symbols ^a" estimated in equalizer 6 are fed to a data sink. In the example it is assumed that the transmitted data sequence contains a training sequence which renders possible both the estimation of the total channel pulse response and the estimation of the coefficients for interference prediction. The total channel pulse response thus also contains the pulse response of the receiving filter 5. The coefficients h₀... of the total channel pulse response estimated in the receiver and the coefficients pi ... for interference prediction are passed on to the state-based equalizer 6. The principle of the state-based equalizer 6 offers an opportunity of separating the hypothetical useful signal and interference signal, so that an individual processing of the hypothetical interference-signal values is possible. To calculate the coefficients pi ... for interference prediction in block 8, an initial estimate of the autocorrelation coefficients of the correlated interference signal u_n is required. An initial estimate of the autocorrelation coefficients of the correlated interference signal u_n may take place, for example, by evaluation of the interference-signal values within the training sequence. These initial-estimate values can be further improved by the inclusion of the interference-signal values estimated during detection. If the autocorrelation coefficients of the interference signal u_n are known and, from these, the coefficients of the interference prediction, the interference prediction can estimate the interference-signal value in the current time step for each state transition using the immediately preceding interference-signal values. This interference prediction value is subtracted from the received signal. The effective interference power is reduced thereby upon detection of the data symbols. It should be noted here that the corrected received-signal values are different for each state transition.

The interference prediction is an optimum when all linear statistical linkages of the interference process are taken account of by the interference prediction, which demands an interference prediction of a very high order. In practice, a first or second order interference prediction frequently suffices for achieving a large proportion of the maximum possible gain.

In a state-based equalizer, the squared error signal, normalized where applicable with the variance of the error signal, is generally used as the similarity criterion or metrics. The error signal corresponds to the difference between the particular received-signal value and the hypothetical, reconstructed signal values of the state transitions. These metrics are an optimum in the case of additive white Gaussian noise (AWGN). The use of these metrics requires a calculation of the error signal. The error signal is simultaneously used in the present case for interference prediction of correlated interference, as will be described in greater detail below.

If we consider a state transition of a state-based equalizer 6 - without interference prediction - we obtain the following as the interference and error signal e_n for this state transition:

L-\ ^en (^Sn I Vl ) = ^rn ^~ ∑ * " ^an-i i^Sn I ^Sn-\ )

^<=° (1) wherein:

S_n: State index in time step n.

(S_n|S_n-ι): State transition from S„-ι to S_n r_n: Received-signal value in time step n. hi : i-th coefficient of total channel pulse response of length L a_n-i(S_n|S_n-ι): Data symbols belonging to the state transition (S_n|S_n-ι). e_n(S_n|S_n-ι): Error signal for the state transition (S_n|S_n-ι).

As the branch metrics λ_n( S_n!S_n-ι) for the state transition (S_n|S_{n- >} ^we obtain: K i^Sn I ^Sn-l ) =

I ^Sn-l } I ^σ N^' (2)

wherein: ^N : Variance of the error signal.

When interference prediction is used for the suppression of correlated interference, the following relations arise for the error signal e'_n(S_n|S_n-ι) and the branch metrics λ'_n(S_n|S_n-ι):

L-\ M ^en' (^Sn I ^Sn-l ) = >^*„ - ∑ ^an-ι (^Sn I ^Sn-l ) ^~ ∑ ^en-ι (^Sn-, I <?„_,_] ) " Pi ;=0 /=1 0)

wherein: e_n-^S_n- S_n.,-_!): Past error-signal values in time step n-i. p, : i-th coefficient of interference prediction.

M: Number of coefficients of interference prediction.

G' : Variance of error signal with interference prediction.

The interference prediction is a filtration of the past interference-signal values with the coefficients p, of the interference prediction. The past error-signal values β_n., may be of a state-dependent or state-independent form. State-dependent means that the past error- signal values originate from a sequence of past valid state transitions, i.e. they belong to a valid, selected path in the state trellis, which leads to the state S_n under consideration in time step n. The M past error-signal values can thus be assigned to a specific state and stored in a state-related memory (buffer). With a number Z of states, M . Z error-signal values must be stored. The error-signal values belonging to a selected state transition are each copied into the state-related memory of the subsequent state.

Fig. 2 is an example of a trellis diagram of a state-based equalizer with eight states (0 to 7) and one binary symbol alphabet. All possible state transitions and the fusion of the paths into the subsequent states are shown. Fig. 3 shows, in greater detail than Fig. 2, a trellis recursion step using the example of a state-based equalizer with eight states and a binary symbol alphabet in time step n. Three buffers 10, 11, 12, which store different variables, are assigned to each state. Buffer 10 stores the state metrics Λ_n(S„).

Buffer 11 is denoted a path memory and contains the hypothetical data symbols a_n(i), a_n-ι(i), ... , a_π-p₊ι(i) of past state transitions belonging to the selected path, which ends in the state S_n = 1 • The number P of stored past data symbols depends on the equalization method.

Buffer 12 stores the hypothetical error-signal values en(i), e_n.ι(i), ... ,er,._M+i(i) of past state transitions belonging to the selected path, which ends in the state S_n = i. A trellis step consists in the branch metrics λ'_n(S_n|S_n-ι) for all state transitions

(S_n|S_n-ι) being calculated in accordance with the above equation 3. Using the example of the

Viterbi algorithm, the following equation 4 shows how the state metrics Λ_n(S_n) of the subsequent states S_n are formed from the state metrics Λ_n-1(S_n-ι) of the previous states S_n-1 and the branch metrics λ'n(Sn|S_n-ι):

Λ„(S_n) = rmn(A_n^(S_n__x) + λ_n' {S_n \ S_n__{)) ^VS"-> (4)

The branch metrics of the state transition are added to the state metrics for all previous states S_π-ι for which a transition into the subsequent state S_n exists, and the state transition with the smallest metrics is selected. Selection of the state transition implies simultaneously that the path memory and the state memory of the error-signal values of the previous state are adopted into the corresponding buffer of the subsequent state and supplemented by the data symbol a_n and the error-signal value e„ which belong to the state transition. Generally, the oldest value in each case is shifted out of the buffers.

Fig. 4 shows the recursive processing of the state trellis in a state-based equalizer. The state metrics Λ_n(i), the path memory a_n(i) and the error-signal vectors e__n(i), which are calculated in time step n, in turn form the output variables for the next time step n+1.

Fig. 5 shows the determination of the error signal e'_n with interference prediction and the branch metrics λ'_n with interference prediction from the received-signal value r_n and the data symbols a_n-i and the past error-signal values β_n-i, as corresponding to equation 3 above.

The possible use of state-independent error-signal values e_n.j, in the interference prediction according to equation 3 above means that only one error-signal value per time step is stored, i.e. a total of M error-signal values for M past time steps, which are then used for all states. One example of a possible criterion for the selection of the valid error-signal value is the selection of the error-signal value that leads to the smallest branch metrics with interference prediction. The described method for interference prediction may be used for state-based equalization with a full state quantity as well as with a reduced state quantity, for example by set partitioning.

The described method forms a combination of a state-based equalization with an interference prediction for correlated interference. The interference prediction in conjunction with the state-based equalization leads to a reduction of correlated interference without changing the useful signal, and thus also to a reduction in the bit and block error rate. Interference suppression is all the more effective in proportion as the interference-signal values are correlated. The method also has the advantage that it can be applied to any types of interference with low sensitivity to the parameters of the interference, and it causes only low loss levels in the case of uncorrelated interference. In the case of common-channel interference and adjacent-channel interference, the method brings about high gain.

Claims

CLA S:

1. A method of suppressing interference signals in a receiver with state-based equalization of data signals which are received in time steps, characterized in that, for the suppression of a correlated interference signal by means of interference prediction for all state transitions of the state-based equalizer in each time step, a state-transition-related interference-signal value is estimated by filtration of associated preceding interference-signal values by means of the coefficients of interference prediction and is subtracted from the received-signal value.

2. A method as claimed in claim 1, characterized in that the state-based equalizer is an equalizer with a reduced number of states which possesses only a partial-quantity of all possible states.

3. ^* A method as claimed in claim 1 or 2, characterized in that the interference- signal values are estimated by filtration of directly preceding interference-signal values.

4. A method as claimed in claim 2 or 3, characterized in that a state-related interference-signal value is assessed in each case for each state transition of the state-based equalizer by filtration of the preceding state-related interference-signal values with the coefficients of interference prediction and is subtracted from the received-signal value.

5. A method as claimed in claim 2 or 3, characterized in that only one interference-signal value is estimated state-independently for all state transitions of the state- based equalizer by filtration of preceding state-independent interference-signal values with the coefficients of interference prediction and is subtracted from the received-signal value.

6. A method as claimed in any one of the preceding claims, characterized in that the coefficients of interference prediction are determined, for example, from the autocorrelation coefficients of the interference signal by evaluation of the estimated interference-signal value in a training sequence and are evaluated in the state-based equalization.

7. A method as claimed in any one of the preceding claims, characterized in that the error signal of a state transition or time step (e'_n) corresponding to the interference-signal value is estimated in that a first sum of preceding data-symbol values (a_n), with inclusion of coefficients (h,) of the channel pulse response, and a second sum of preceding error signals (e_n), with inclusion of coefficients (p,) of an interference prediction, are subtracted from the relevant received signal (r_n).

8. A method as claimed in one of the preceding claims, characterized in that branch metrics (λ'_n) of the state-based equalization are determined from the square of the sum of the error signal (e'_n) relating to the variance (σ'N²) of the error signal.

9. A method as claimed in one of the preceding claims, characterized in that the initial estimated values of the coefficients of interference prediction are improved through inclusion of further interference-signal values obtained during the equalization.

10. A device for implementing the method as claimed in one of the preceding claims, characterized in that the receiver (4) is equipped with a receiving filter (5) and an equalizer (6) operating with state-based equalization of the received signal (r_n), in that a block (7) applies estimated values of the channel pulse response to the equalizer (6), and in that a block (8) applies estimated values of the coefficients of interference prediction to the equalizer (6).