CA2114625C - A method and an arrangement of estimating transmitted symbols at a receiver in digital signal transmission - Google Patents

Abstract

In a digital time-shared radio transmission system, a radio receiver receives a signal whose symbol frequency is lower than the channel bandwidth of the system. A
correlating and sampling circuit receives a baseband signal, samples the signal eight times with each symbol time, performs channel correlation, generates a channel estimate and samples down the once sampled signal to an observed signal with two values for each symbol time. A channel equalizer executes a fractional viterbi algorithm which utilizes two delta-metric values for each state transition and generates estimated symbols. A channel estimating filter receives a symbol sequence of alternating zero-value symbols .OMEGA. and the estimated symbols and generates an estimated signal. The channel estimating filter is adapted with the aid of an error signal and the filter delivers a channel estimate to the channel equalizer. Weighting factors are generated with the aid of the error signal and the two aforementioned delta-metric values are coweighted to a common delta-metric value with the aid of the weighting factors, this common delta-metric value being used to generate the estimated symbols. The use of the weighting factors improves the metric calculation and enables the channel estimating filter and the channel equalizer to be relatively simple.
The insertion of the zero-value symbols simplifies the generation of the weighting factors and the adaptation of the channel estimating filter.

Description

WO 94/~0924 ~ ~ i ~ ~ " ~ PCT/SE93/a047~
A METHOD AND AN ARRANGEMENT OF ESTIMATING TRANSMITTED BYMHOLB AT
A RECEIVER IN DIGITAL SIGNAL. TRANSMI88ION
TECHNICAL FIELD
The present invention relates to a method of estimating in a receiver transmitted symbols from a transmitted radio signal in conjunction with the transmission of digital signals over a radio channel, wherein said symbol estimation is effected in accordance with a viterbi algorithm which has a predetermined number of states, said method comprising the following method steps:
- receiving and filtering the transmitted signal to form a baseband signal:
- sampling the baseband signal at at least two sampling time points for each symbol:
- effecting correlation to determine the estimated impulse 1~ response of the radio channel with the, aid of the sampled signal values;
- determining a symbol sampling time point at one of the sampling time points:
- selecting at least two of the sampling time points with each symbol, of which one is the symbol sampling time point, and selecting the sampel signal values at these time points:
- determining the delta-metric values in accordance with the viterbi algorithm for an indicated transmitted symbol, said determining process being effected for each of the selected sample signal values and for each state transition of the viterbi algorithm: and - generating at least preliminarily estimated symbols in accordance with the viterbi algorithm.
The invention also relates to an arrangement for carrying out the method.

94/00924 2 2 ~. ;~ ~ ~ ,~; ~ PCT/SE93/00477 BACKGROUND ART
One problem which often occurs when transmitting digital radio signals over a channel is that a transmitted signal is subjected to multipath propagation, resulting in time dispersion and noise.
For instance, in mobile telephony, the channel transmission properties will change as a result of a mutual change in the positions of transmitter and receiver. These problems have been solved in time-shared, digital radio transmission systems, by giving the signal sequences, which are transmitted in a time slot, a synchronization sequence and a data sequence. The synehroniza-tion sequence is known to the receiver and with the aid this sequence the receiver is able to make an appraisal of the channel transmission properties, a channel estimate. The receiver makes an appraisal of the symbols of the data sequence, which contains the information to be transmitted, with the aid of this channel estimate.
In certain cases, it is not sufficient to make a channel estimate only once with each time slot. In the case of long time slots, in the order of several milliseconds, transmitter and receiver are able to change their mutual positions guite considerably during the course of the time slot. This means that the channel transmis-sion properties may change considerably over the duration of the time slot, such that the appraisal of the transmitted symbols made by the receiver will be def icient and the transmitted information therefore unclear. A radio receiver in Which these disturbances are partially avoided is described in an article in IEEE Transac-tions on Information Theory, January 1973, pages 120-124, F.R. Magee, Jr. and J.G. Proakis: "Adaptive Maximum-Likelihood Sequence Estimation for Digital Signaling in the Presence of Intersymbol Interference". The article describes a channel equalizer comprising a viterbi analyzer which includes an adaptive filter as a channel estimation circuit. Received symbols are compared successively with hypothetical symbols and those hypothetical symbols which coincide closest with the received symbols are selected successively to form an estimated symbol sequence. The parameters of the adaptation filter are adjusted ~ii~~J~J

successively to the changed channel with the aid of the selected, decided symbols.
A description of the viterbi algorithm is found in an article by G. David Forney, Jr.: "The viterbi Algorithm" in Proceedings of the IEEE, Vol. 61, No. 3, March 1973. The article describes in more detail the state of the viterbi algorithm and its state tran-sitions and discloses how these state transitions are chosen so as to obtain the most probable symbol sequence.
Signal transmission between transmitter and receiver may be encumbered with certain deficiencies, despite carrying' out sequence estimation and adaptive channel estimation in the aforesaid manner. One reason for these deficiencies is that the symbol frequency of the system is lower than the channel bandwidth of the system, as in the case, for instance, of the North American mobile telephone system ADC. Such systems are also known as ''excess bandwidth systems". A solution to these symbol frequency problems is described in an article by Yongbing Van, et al, in NovAtel Communications Ltd: "A Fractionally-Spaced Maximum-Likelihood Sequence Estimation Receiver in a Multipath Fading Environment" published by IEEE 1992. According to this article, a received radio signal is sampled twice with each symbol and the channel estimation is effected through an adaptive filter which utilizes this double sampling rate. The symbol estimation is carried out in a viterbi analyzer which also utilizes the double sampling rate. The delta-metric values, i.e. deviations between the received and the hypothetical sequences, are calculated on both the sampling occasions for each symbol and the two delta-metric values are summed directly in order to determine a best state transition according to the viterbi algorithm. When adapting the filter with the aid of the estimated symbols, a fictive symbol is inserted at each alternate sampling time point. These fictive symbols. are generated by interpolation between the estimated symbols in a second filter. The proposed solution has certain drawbacks. It is necessary to sample the received symbols at a time point which has been very well established and the adaptive' channel estimation is of high complexity. The interpolation in the -~ <
~ 94/00924 4 ' PCT/SE93/00477 second filter results in delays which impair the symbol es-timation. The signalling processing filters used, for instance a transmitter filter, or receiver filter must be known. The receiver filters in particular, which may include coil's and capacitors, cause problems in this respect as a result of aging, manufacturing accuracy and temperature variations.
Another solution to the problems that occur at the aforesaid relatively low symbol frequency is given in a paper by R.A. Iltis:
"A Bayesian Maximum-Likelihood Sequence Estimation Algorithm for A-Priori Unknown Channel and Symbol Timing", Department of Electrical and Computer Engineering, University of California, Santa Barbara, August 21, 1990. This paper also states that sampling of a received signal shall be effected twice for each symbol. The symbol estimation is effected in accordance with a viterbi algorithm, which calculates two delta-metric values for each symbol, and these two values are co-weighted in the metric calculation. The channel estimation is effected in an adaptive filter with filter coefficients at a symbol time spacing, although the coefficients are adapted on each sampling occasion, thus twice with each symbol. The proposed solution further involves a relatively complicated metric calculation and fails to solve the problem of symbolsynchronizationfor complicated, rapidly varying channels. As with the aforedescribed solution proposed by Yongbing Van, a receiver filter for instance must be known witty good accuracy by the receiver.
DISCLOSURE OF THE INVENTION
The present invention relates to a method and to an arrangement for symbol estimation in digital radio transmission systems. The method readily solves those problems which occur at low sampling rates, or expressed more precisely those problems which occur when the digital radio transmission system has a symbol frequency which is below the signal bandwidth of the system. A received radio signal is sampled at least twice for each symbol time, to provide the observed signal values, and the symbol estimation is effected in accordance with a viterbi algorithm. This algorithm utilizes 2:1~:-~p~
WO 94/00924 5 PCT/SE93/OOd77 the estimated values of the radio channel impulse response, which are constant or, in the case of long time slots, are generated adaptively in a channel estimation filter which is updated at each sampling moment. The symbol values estimated in accordance with the viterbi algorithm are utilized for this adaptation of the channel estimation filter. The estimated values of the received signal are formed in the channel estimation filter with the aid of the estimated symbols, and the error signals are formed at each sampling moment as a difference between the estimated signal values and the observed sampled signal values. The coef f icients of the channel estimation filter are adapted when applicable with the aid of the error signals in accordance with a selected adaptation algorithm. When selecting the state transitions according to the viterbi algorithm, there is calculated for each state transition y5 a delta-metric value for each symbol sampling moment. Each of the delta-metric values is multiplied~by a respective weighting factor and the values are summed to form a total delta-metric value for an observed state transition. The weight factors are generated according to the inverted values of the error signals and the fact that the coefficients in the channel estimation filter are able to contribute to residual interferences to different extents is hereby taken into account in the metric calculation. This residual interference arises. because the true channeltransmissionfunction is represented by the channel estimate, Which always implies an approximation. The greater the number of coefficients possessed by the filter, the better the approximation, although a large number of coefficients will result in a complicated filter and, above all, will mean that the viterbi algorithm must have a large number of states and Will therefore become complicated and require a complicated calculating process. The accuracy of the metric calculation is relatively good when the delta-metric values are weighted, which opens up the possibility of including relatively few coefficients in the channel estimation filter.
As before mentioned, the estimated symbols are used when adapting the channel estimation filter. In order to enable this adaptation to be carried out at the sampling time points between the symbols, fictive symbols are inserted in these intermediate time points.

~~.i~~~«~ . , ..0 94/00924 6 PCT/SE93/Q0477 The fictive symbols are assigned Zero-values, therewith simpli-fying filter adaptation. In this way, the new filter coefficient values need only be generated ance_with each symbol, irrespective of the number of sampling time points per symbol. The fictive symbols are also utilized when generating the disturbance level values, even in those instances when the channel estimation filter is constant and not adapted. The insertion of the zero-value fictive symbols results in relatively small time delays when adapting the channel estimation filter and when generating the disturbance level values. This is contributory in' enabling the symbols to be estimated with good accuracy.
The invention has the characteristic features set forth in the following Claims.
BRIEF DESCRIPTION OF THE DRAWINGS
An exemplifying embodiment of the invention will now be described in more detail with reference to the accompanying drawings, in which Figure 1 is a block schematic overall view of a transmitter and a receiver in a digital radio system:
20~~~ Figure 2 illustrates time slats and a symbol sequence for time-shared radio transmission:
Figure 3 illustrates a complex plane with symbol values:
Figure 4 is a block schematic illustrating the receiver:
Figure 5 is a block schematic illustrating a channel estimation filter:
Figure 6 is a diagrammatic illustration of a radio channel impulse response:
Figure 7 illustrates certain states and state transiticans in a viterbi algorithm:
Figure 8 is a block schematic illustrating a weighting factor generating circuit:
Figure 9 is a flowsheet illustrating the inventive method: and Figure 10 is a block schematic which illustrates an alternative embodiment of the invention.

Cy ~ .~ .q l,.~a.15't~~J
1w0 94/00924 .~ ~ PC1YSE93/00477 BEST MODES OF CARRYING OUT THE INVENTTON
Figure 1 illustrates a radio transmission system for time-shared, digital signal transmission. A transmitter includes a unit 10 which receives an information carrying signal and generates corresponding digital symbols S(k). The symbols S(kj are signal processed in the unit 11 and transmitted to a radio transmitter 12, which transmits the signal analogized in the unit 11 on a selected carrier frequency as a signal R(T). This signal is transmitted over a radio channel 13 to a receiver having a radio receiver 14. Amcng other things, the channel 13 subjects the signal R(T) to multipath propagation, as indicated by double signal paths in the Figure. For instance, the signals travelling along a signal path are reflected by a building 18 prior to reaching the receiver. The radio receiver 14 demodulates the received signal to one baseband and delivers a baseband signal y(T) to a correlating and sampling circuit 15. In turn, this circuit delivers an observed sample signal designated y(k/2) . The signal y(k/2) is processed in a channel equalizer 17 in accordance with a viterbi algorithm, and the equalizer produces estimated symbols SD(k) , which shall coincide as close as possible with the symbols S(k) transmitted by the transmitter. The correlating and sampling -~circuit 15 is cot~ected to a channel estimating circuit 16 and delivers thereto the initial values of a channel estimate which includes the channel 13. The circuit 16 is adaptive and generates successively the new coefficient values for the channel estimate, this estimate thereby being adapted successively to the time varying channel 13 with the aid of the signal y(k/2) and the estimated symbols SD(k).
As before mentioned, the radio transmission system according to the illustrated embodiment is time-shared, as shown in Figure 2, in which T represents time. A carrier frequency, or actually a frequency pair for two-way communication, is divided into three time slots 19. numbered 1, 2 and 3. A symbol sequence SS comprising a synchronization sequence SY and two data sequences SDi and SD2 is transmitted in each time slot, said sequences containing the information to be transmitted. The symbol sequence SS includes ~i ,~ ~l ~ .~ J
\ y0 94/00924 a PCT/SE93/00477 binary signals, although the aforesaid symbols S (k) are modulated in accordance with QPSK-modulation for instance, as illustrated in Figure 3. In a complex plane having coordinate axes referenced I and Q, the four possible values S0, S1, S2 and S3 of the symbols S (k) are marked and corresponding binary digits 00, 01, 10 and 11 are given. The time taken to transmit one such modulated symbol is designated one symbol time TS, as illustrated schematically in Figure 2. It is these full symbol times TS that are counted by the integer symbol counter k. .
The system illustrated in Figures 1 and 2 may be comprised of a mobile telephone system in which the transmitter is a base station and the receiver is a mobile station, or vice versa. The three time slots 1, 2 and 3 the signal sequence SS comply with the standard for the American mobile telephone system ADC. In this system, the time slots have a length of 6.7 milliseconds, Which requires the channel estimation circuit 16 to be adaptive, as mentioned above.
As mentioned in the introduction, problems arise with channel equalization and symbol estimation in digital radio transmission systems, the symbol frequency R = 1/TS of which is lower than the signal bandwidth B of the system. This is the case, for instance, in the aforesaid ADC-system, the signal bandwidth of which is B =
kHz and the symbol frequency of which is R = 24. 3 kEd. According to the sampling theorem, it is not sufficient to sample the baseband signal y(T) in such systems at the symbol frequency R.
25 The symbol frequency R can only be used as the sampling frequency when so-called matched filters are used in the receiver, that is to say filters which are matched at any moment to the cascade of all transmitter and receiver filters and to the transmission func-tion of the channel 13. In other cases, it is necessary to use a 30 higher sampling frequency, particularly when wishing to use a simple channel estimation filter., which causes problems with channel estimation and channel equalization processes. It is these problems that are solved by means of the present invention for a channel equalizer which functions in accordance with the viterbi algorithm. Those problems associated with effecting channel estimation in an adaptive filter are also solved in a simple manner.
The receiver illustrated schematically in the right half of Figure 1 is shown in more detail in Figure 4. The radio receiver 14 is connected to the correlating and sampling unit 15, which includes a first sampling unit 21, a second sampling unit 22, a correlating eircuit 23, a synchronization circuit 24 and a generator 25 for generating the synchronizing sequence SY known by the receiver.
The first sampling unit 21 receives the continuous baseband signal y(T) fram the radio receiver 14 and samples this signal eight times for each symbol, i.e. with the sampling frequency of i3/TS.
This sampling frequency is used in the aforesaid mobile telephone system ADC. The thus sampled signal, designated y(k/8), is delivered to the correlating circuit 23. There is generated in this circuit by least squares estimation e.g. correlation, a first channel estimate HF for the observed symbol sequence SS with the aid of the synchronization sequence SY from the generator 25 and the transmitted, observed synchronization sequence. When genera-ting this first channel estimate, a symbol sampling time point TO
is also established in the synchronizing circuit 24. This symbol sampling time point controls the second sampling unit 22 through which,. according to the illustrated embodiment, two of the original eight sampling time points of each symbol are selected with a time spacing of TS/2. The observed signal y(k/2) is obtained in this way and is delivered by the sampling unit to the viterbi analyzer 1'7. Down-sampling is effected in the unit 22 in order to simplify signal processing in this analyzer: The original eight samples are used to establish the symbol sampling time point TO, which is a starting point for symbol counting with the aforesaid counter k. The symbol sampling time point and the channel estimate HF are sent to the channel estimating circuit 16.
The manner in which the channel estimate HF is generated in the correlating and sampling circuit 15 will now be described in brief. An impulse response which includes the transmission function of the channel 13 is generated with the aid of the signal y(k/8) and the synchronizing sequence SY. The impulse response ~~i~~~
.O 94/00934 ' O PCT/SE93/00477 extends over a time period which includes several symbol times TS, and the discrete values of the impulse response are generated at the time spacing TS/8. There is selected within this time period a second, shorter time period which contains-the first channel estimate HF. The choice is made so that the first channel estimate HF will obtain maximum energy. Furthermore, the first channel estimate NF is produced only at time points which are mutually spaced by the time spacing TS/2. A more detailed description of how the channel estimate is chosen is given in the Swedish Patent Application No. 8903842-6. It should be noted that the channel estimate, both the first channel estimate HF and. the later, adapted channel estimate, includes both the physical radio channel 13 and the transmitter filter 1l and receiver filter, for instance MF-filter, used to separate the carrier frequency.
The channel estimating circuit 16 includes an adaptive channel estimating filter 31, a delay circuit 32, a difference former 33, a circuit 34 which executes an adaptation algorithm, a quadrating and average value forming circuit 35; a signal switch 36 and a symbol generator 37. The channel estimating filter 31 receives the first transmission function HF and the symbol sampling time point TO, and also receives the symbols SD(k) estimated in the channel equalizer 17. With the aid hereof, estimated signal values y(k/2) are formed and delivered to the difference former 33. This circuit also receives the observed signal y(k/2) , which is delayed in the circuit 32, and delivers an error signal e(k/2) = y(k/2)-y/k/2).
The error signal is delivered to the circuit 34, which controls the adaptive filter 31 through its adaptation algorithm. In turn, this filter delivers to the egualizer 17 successively adapted values N(k/2) for the channel estimate. The equalizer also receives weighting factors a(k/2), which are generated in the circuit 35 with the aid of the error signal e(1/2), as will be described in more detail herebelow. The channel equalizer 17 receives from the symbol generator 37 hypothetical symbols S(k), which take the four symbol values S0, S1, S2 and S3 shown in Figure 3. The signal switch 36 is controlled from the synchronizing circuit 24 and shifts at an interval of one-half symbol time, TS/2, alternating between an estimated symbol SD(k) and a fictive ~~~ej~ 5 h'0 94/00924 a 1 PCT/SE93/00477 symbol it, which has zero-value. This zero-value shall not be confused with the binary value 00 for the complex value symbol SO
in Figure 3. The fictive zero-value symbol ft lies on the origin in the complex plane I-Q. as shown in Figure 3. The generation of the fictive symbols t1 have been illustrated schematically in the Figure, by connecting one terminal 36A of the signal switch 26 to ground potential. The reason why the zero-values are switched-in will be explained in more detail below with reference to Figure 5.
This Figure shows the channel estimating filter 31, the delay circuit 32, the difference former 33 and the circuit 34 with the adaptation algorithm. The filter 31 has delay circuits 41, coefficient circuits 42, summators 43 and a switch 44. The delay circuits 41 are connected sequentially in series and function to delay the incoming signal successively through one-half symbol time TS/2. The subsequent delayed signals are multiplied in the coefficient circuits 42 with coefficients HO (k) , H1 (k) , H2 (k) and H3(k) respectively, which are the values of the channel estimate H(k/2) at four time points with a mutual spacing of one-half symbol time TS/2. The output signals from the coefficient circuits 42 are summed in the summators 43 to the estimated signal values y(k/2). The error signals e(k/2) are formed in the difference former 33 and delivered to the adaptation algorithm in the circuit 34. This algorithm is chosen in dependence on those disturbances which the radio channel 13 is assumed to have, and in the illus-trated embodiment is a so-called LMS-algorithm (Least Mean Square). The output signal from the circuit 34 adjusts the coefficients in the coefficient circuits 42, so that the effect of the error signals a (k/2) will be minimized in accordance with the LMS-algorithm. The coefficient circuits obtain their starting values through the first channel estimate HF from the correlating and synchronization circuit 15. These starting values are applied with the aid of the switch 44, which is controlled from the synchronization circuit 24. The estimated signal values j~(k/2) are generated with the aid of the estimated symbols SD(k) , which are delayed by the viterbi algorithm through a number q symbol times TS. The observed signal values y(k/2) are therefore. delayed by the .d 'v ' \ J 94/0092.4 1 2 PCT/SE93/00477 number q symbol times in the delay circuit 32. By inserting the taro-value fictive symbols f1 between the estimated symbols SD(k) , the coefficient circuits 42 obtain a zero-value_input signal with each alternate update. Consequently, these coefficient circuits need only be updated once for each symbol time TS, which simpli-fies the updating process. This will become more evident from the following description of the channel estimating method.
The estimated signal y(k/2) has two separate values for each symbol, i.e. the value y(k) at the symbol sampling time point TO
and the value Q(k-~S) one half symbol time TS/2 earlier ori. These values are generated as follows:
Y(k-'~) = HO(k) SD(k) + H2 (k) SD(k°1) Y(k) = H1(k) SD(k) + H3(k) SD(k-1) (1) In Figure 5, the symbol values of the symbol sequence SD(k) , n at time position k-s one-half symbol time TS/2 prior to the symbol sampling time point TO are marked at the inputs of the coefficient circuits 42. The symbol values are shifted TS/2 to the right in the Figure at symbol sampling time point TO through one-half symbol time. During a symbol time, the error signals e(k/2) have two separate values during the symbol time TS:
a(k-'s = y(k-'s) ° ~'(k-$) e(k) = y(k) ' Y(k) (2) Where y(k) and y(k-'s) are the.two signal values observed during a symbol time of the observed signal y(k/2). In the case of the illustrated embodiment, the channel estimate is updated by the LMS-algorithm in accordance with the following relationships:

r n ~ ~.
'v ~ i,1 ~:'1 ( Ho(k) 1 ( Ho(k-1) 1 ( ~~(k) 1 FIl (k) til (k-1) ~0 HZ(k) = Ha(k-1) + h D(k-1) e(k) H3 (k) l ( ti3 (k-1) ( 0 l l (3) ( ~i~ (k) ( HO (k-1) ( 0 ~ ~

Hi(k) Hi(k-1) SD(k) H2(k) a H2(k-1) + ~ o e(k-~) H3 (k) J 1 H3 (k-1) ( SD(k-1)1 ~

1~ where ~e is a parameter, the step length, in the adaptation algorithm. It will be seen from the equations (3) that the values of the coefficient circuits 42 need only be calculated once for each symbol time, as a result of the insertion of the zero-value fictive symbols t1. 3t will also be seen from the equations (1) that the insertion of the zero-value symbol f1 also simplifies genera-tion of the estimated signals y(k/2). Each of the relationships (1) has only two terms instead of the four terms which would be required if values other than zero-values were inserted between the estimated symbols S~(kj and SD(k-1).
An example of the channel estimate appearance is shown in Figure 6, which is a diagram in which the coordinate axes are referenced T and H. A curve A shows a continuous channel impulse response and the chosen time points on the time spacing TS/2 denote the discrete values HO (k) , Hl (k) , H2 (k) and ~i3 (k) of the channel estimate. The aforesaid symbol sampling time point TO is given in the Figure and the symbol counter k denotes that the discrete channel estimate values relate to the transmitted symbol with number k.
The channel equalizer 17 functions in accordance with a so-called fractional viterbi algorithm, since it channel equalizes the signal y(k/2) which is sampled in fractions of the symbol time TS.
The reader is referred to the aforesaid reference "The viterbi Algorithm" by G. Forney for a more detailed description of the viterbi algorithm. The algorithm has a number of states .~,~ ~ ~ '~i '.l rd J

N = ML 1 in a known manner, where M signifies the number of values that a symbol can have, and L is the length of the channel estimate in number of symbol times TS. In the case of the illustrated embodiment, M = 4 according to Figure 3 and L = 2 according to Figure 5, so that the equali2er 1? will have N = 4 number of states. These states are illustrated in Figure 7 and are there referenced B and numbered 0, 1, 2 and 3. The algorithm is il-lustrated by node plans in columns, of which some are shown in the Figure. The node plan relates to separate time points referenced k-2, k-1 and k, where the letter k represents the aforesaid symbol counter. The viterbi algorithm compares, in a known manner, sequences of the observed signals y(k/2) with hypothetical se quences that are generated with the aid of the hypothetical symbols ~ (k) and with the aid of the ehannel estimate H (k/2) . The hypothetical symbols are given by the equation:
S(k) _ (SO(k). S1(k)~ S2(k)~ S3(k)) (4) Deviations between the two sequences are referred to as metric values Jj (k) which are calculated successively by addition of the delta-metric values. These delta-metric values are calculated for transitions between the states B, as illustrated in Figure 7 with a full line arrow for the transition from the state 3 having the metric value J3 (k-1) to the state 0 having the metric value JO (k) .
In the illustrated embodiment, the channel equalizer receives the observed signal values y(k/2) which have these two values with each symbol time y (k) and y (k-~S ) . The two delta-metric values are generated for each state transition in accordance with the following general equation for the transition i to j with the aid of these values and with the aid of the channel estimate H(k/2) and the hypothetical symbols S(k):
AJ..(k) = iY(k)-(H1(k) Sj(k) + H3(k) Silk 1))~2 AJi~(k-~S) = iY(k-~S)-(HO(k)Sj(k) + H2 (k) Silk-1) ~2 (5) These delta-metric values are generated in full accord With the viterbialgorithm and with known devices schematically illustrated ~~~ ~L
wo ~ioo9ia a 5 Pcris~3~ooa7~
with a circuit 178 in Figure 7. This circuit receives the observed signal y(k/2), the hypothetical symbols S(k) and the channel estimate Ii(k/2) . The states B are realized with the aid of memory circuits which store the metric values.
That part of the inventive symbol estimating process which relates to the viterbi algorithm is concerned with the continued proces-sing of these delta-metric values. iihen generating a total, summed delta-metric value for the state transition f to j, .the two delta-metric values are jointly weighted with the aid of weighting factors a(k/2) _ (ak,ek-~). The generation of these weighting factors will be explained in more detail below. The metric value Jj (k) in the new state j is generated in accordance with the general equation:
J~ (k) = Jl (k_1) + (ak eJij (k) + ak-~ SJij (k-5) 7 in which the expression contained within the square brackets is the total summed delta-metric value.
The metric values are generated in a metric calculating circuit 17A in Figure 7 and are there illustrated for the state tran-sitions 3 to 0. The metric calculating circuit 17A receives the delta-metric values ~J30 (k-~S) and eJ30 (k) from the circuit 178 and the weighting factors ck-~ and ak from the circuit 35 and genera-tes the total summed delta-metric value for the transition 3 to 0.
Also generated in the circuit 17A are the total delta-metric values for the remaining transitions from the state 0, 1 and 2 to the state 0, as shown by the broken arrows in the Figure. According to the viterbi algorithm, there is chosen the state transition which has the lowest of these total delta-metric values, which in the illustrated embodiment is assumed to be the transition 3 to 0.
The new metric value JO(k) for the selected state transition is then generated in accordance with the above equation (6) .
The new metric values are generated successively until the last node plane of the algorithm and the estimated symbols $D(k) are decided on the basis of the metric values thus obtained, in ~ ~ ~.~ z~
\O 94/00924 , 6 ~'CT/SE93/00477 accordance with the viterbi algorithm. Preliminarily estimated symbols Sp(k) can be decided at an earlier stage, for instance after the node plane referenced k l Figure ~._ According to one alternative, these preliminarily estimated symbols Sp(k) can be used in the symbol sequence instead of the estimated symbols SD(k) . The preliminarily estimated symbols are used in this way to update the channel estimation filter 31 according to the equation (3) and also for producing the error signals e(k-~) and e(k) in accordance with the equations (1) and (2).
The aforesaid weighting factors ak-~ and ak are generated with the aid of the error signals a (k) . and a (k-~ ) . The generation of these weighting factors is based on the observation that the statistical expectation values of the absolute value of respective error signals represent a combined interference caused by noise, intersymbolic interference and co-channel interference. The greater the expectation value, the less the estimated signal values y(k-~) and y(k) will correspond to their respective observed signals y(k-5) and y(k). The weighting factors shall be correspondingly smaller, so that the delta-metric value, eJi~ (k-or ~Ji~(k), associated with a large error signal.will give a correspondingly smaller contribution when generating the new metric value 3~(k). The powers of the two error signals e(k) and e(k-~S' can differ considerably from one another, particularly when the channel estimate H(k/2) has only a few coefficients.
The statistical expectation values are estimated by squaring (or quadrating) the value of the error signals and forming average values. The expectation values and the weighting factors are generated in the circuit 35, which is shown in more detail in Figure 8. The circuit has two quadrators 51 and 52, two lowpass filters 53 and 54, two inverters 55 and 56 and two signal switches 57 and 58. The signal switch 57 receives the error signals a (k/2 ) and delivers these signals alternately to the quadrators 51 and 52 at intervals of one-half symbol time TS/2. The signal switch 57 is controlled by signals from the synchronizing circuit 24 in Figure 4 in a manner not more closely shown. The two error signals e(k-~) and a (k) are squared in their respective quadrators 51 and 52 and 2~~.
WO 94/009?A , ' PCT/SE93/00477 the squared values are formed into average values by filtering said squared values through respective lowpass filters 53 and 54.
These filters deliver signals a2 (k-~Sj and a2 (k) which correspond to the aforesaid statistical expectation values of the error signals. The signals c2 (k-~ j and a2 (k) are inverted in respective inverters 55 and 56 to produce the aforesaid weighting factors ak and ak-~ and are delivered to the signal switch 58. This switch is controlled from the synchronizing circuit 24 in a manner not shown in detail, and applies the weighting factors to the metric calculating circuit 17A in the channel equalizer 17 at intervals of one-half symbol time TS/2. The circuit 35 in Figure 4 thus generates the weighting factors in accordance With the following equation.
ak-~ = 1/ , e~( -'~s 7 ; 2 ak = 1/ WD ; 2 .. - (7) where 'the line above a (k-': ) and a (k) denotes the formation of mean values.
According to one alternative, attention is also paid to the size of the filter coefficients when generating the weighting factors in accordance with the following equationsa ak-~ ~ (H02+H22) /a2(k-~) ak = (Hla+H32j /o2(k) (8) In order to generate these alternative weighting factors, the circuit 35 receives the channel estimate H(k/2) from the channel estimation circuit 31 via a connection 38 shown by a broken line in Figure 4. The filter coefficients HO and Ii2 and respectively H1 and H3 are squared and summed in pairs in circuits 59 and 60 included in the quadrating and mean-value-forming circuit 35, and multiplied with the inverted values of a2(k-s) and a2(k) respec-tively. The thus generated weighting factors a (k/2 ) are delivered ~~iv3 94/00924 ' $ PCT/SE93/00477 to the channel equalizer 17 and used in the metric calculation as described above with reference to equations (5) and (6).
Figure 9 is a flowsheet which presents an overall view of the inventive method. In block 70, the radio signal R(T) is received 5 and filtered to form a baseband signal y(T). This signal is sa~apled eight times with each symbol time TS according to block 71 and the sampled y(k/8) is used for channel correlation, block 72.
The channel correlation produces the sampled impulse response of the radio channel 13, this response being used to determine the channel estimate HF and also to determine the symbol time point TO. The sampled signal y(k/8) according to block 73 is down sampled on the basis of this time point TO to the observed signal y (k/2) , which has two signal values for each symbol time TS. One delta-metric value for each observed signal value for each state transition is generated in block 74 in accordance with the viterbi algorithm, in the illustrated example two delta-metric values DJ i.(k-~) and llJij(k) are produced for each transition. The estimated symbols SD(k) are decided in block 75, and the symbol sequence. of these estimated symbols and the fictive zero-value symbols ft are generated in block 76. The estimated signal values ~(k/2) are generated in block 77 with the aid of~the channel esti-mate HF and the said symbol sequence. The error signals e(k/2) are generated in block 78 with the aid of the estimated signal values and -the observed signal values y(k/2). The weighting factors a(k/2) are generated in block 79, by quadrating, lowpass filtering and inverting the error signals. The weighting factors are used in block 74 to generate the total summed delta-metric values.
According to one simplified alternative, the values of the filter coefficients are only adjusted in the channel estimating ,circuit 31 once for each symbol sequence SS, with the aid of the. first channel estimate HF. This means that the circuit 34 with the adaptation algorithm is omitted. In this case, the symbol sequence with'alternating estimated symbols SO(k) and the fictive zero-value symbols n, which are delivered to the channel estimation circuit 31, is used solely to generate the estimated signal 13(k/2). According to this simplified alternative, however, the 2~i'~u;~~7 dsa! 94/00924 ' 9 PCT/SE93/00477 insertion of the fictive symbols t1 is significant to the genera-tion of error signals e(k/2j , which according to equations (1) and (2) are generated with the aid of the estimated symbols SD(k).
A more complicated inventive alternative will be described with reference to Figure 10. In this alternative, there is used a channel equalizer 80 which includes an adaptive channel estimating circuit 81 for each states B of the viterbi algorithm. The reader is referred to Swedish Patent No. 8903526-5 for a more detailed description of this equalizer. each of, the channel estimating 10~ circuits 81 generates a respective channel estimate H0, Hl, H2 and H3, which are updated with the aid of transition vectors Sij, in the illustrated embodiment SiO. S11' Sm2 and Sn3. These vectors are used instead of the decided symbols SD(kj in the preceding embodiment, and the channel estimates are updated in accordance with the LMS-algorithm, fox instance. Each of the channel estimating circuits 81 delivers its updated channel estimate to its respective state in the viterbi algorithm. The error signal e(k/2) is generated by selecting one of the channel estimates through a switch 82 which is controlled by the viterbi algorithm.
In this case, there is selected the channel estimate which belongs to the state having the smallest metric value, in the illustrated example the channel estimate H0, and this value is delivered to a circuit 83. The zero-value symbols D are inserted by the circuit 36, so as to generate the symbol sequence Sp(kj, tt. The estimated signal values ~3(k/2) are generated in the circuit 83 with the aid of this symbol sequence. The error signal e(k/2) is generated in the difference former 33, which receives the observed signal values y(kj2). The weighting factors are generated in the aforedescribed manner.
30~ In the aforedescribed exemplifying embodiment, the observed sampled signal y(k/2) has two signal values for each symbol time TS. It lies within the scope of the invention to select, for instance, four or still more signal values for each symbol time.
It is required, however, that the channel estimating filter 31 has correspondingly more coefficient circuits 42. According to the aforedescribed example, the channel estimate H(k/2) extends over ~,~~~!~ i~~3 ~ 94/00924 2 o PCT/SE93/00477 two symbol times, although the estimate may be broader. This also requires the channel estimating filter 31 to have more coefficient circuits 42, and, above all, requires the equalizer 17 to be more complicated and to have a correspondingly largeY number of states B. In order to avoid a delay when adapting the channel estimate Ii(k/2) and generating the weighting factors ~a(k/2), the preli-minarily determined symbols $p(k) can be used.
As before mentioned, adaptation of the channel estimating filter 31 is simplified by the insertion of the fictive zero-value symbols i1 in accordance with the invention. According to known techniques, for instance the technique according to the aforesaid article in IEEE by Yongbing Wan, et al, interpolated symbol values are used between the estimated symbol values as fictive symbols.
This results in a delay when adapting the channel estimate, which always impairs the final symbol estimation. The technique defined in the article has the serious drawback that the filters in the transmission chain, transmitter and receiver filters, must be known to a high degree of accuracy. The insertion of these zero-value symbols f! has the added advantage of avoiding a delay when generating the error signals e(k-~) and e(k). This enables the weighting factors ak-~ and ak to be generated in the absence of unnecessary delay, which improves the generation of the summed delta-metric values. This is also utilized in the aforesaid simpler embodiment which lacks adaptation of the channel estimate H(k/2) . The use of the weighting factors in the generatian of the total delta-metric value affordsimportant advantages.The channel estimate H(k/2) can be short, in other words it may embrace only a few symbol times TS, and the channel estimating filter has only a few coefficient circuits.. This means that a viterbi algorithm used in the symbol estimation will have a small number of states, which is extremely significant when effecting symbol estimation in practice.

Claims (61)

The embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. In digital transmission of signals over a radio channel, a method in a receiver of estimating transmitted symbols from a transmitted-radio signal wherein symbol estimation is effected in accordance with a viterbi algorithm which has a predetermined number of states, the method comprising the following steps:
receiving and filtering the transmitted signal to form a baseband signal;
sampling the baseband signal at at least two sampling time-points for each symbol;
effecting correlation to determine an estimated impulse response of the radio channel, a channel estimate, using the sampled baseband signal, and at least one symbol sequence known to the receiver;
determining a symbol sampling time-point at one of the sampling time-points;
selecting at least two of the sampling time-points with each symbol, of which one is the symbol sampling time-point, and selecting observed sampled signal values at these time-points;
determining the delta-metric values according to the viterbi algorithm for an indicated transmitted symbol, this determining step being carried out for each of the observed sampled signal values and for each state transition of the viterbi algorithm;
generating at least preliminarily estimated symbols according to the viterbi algorithm;

generating a symbol sequence from the estimated symbols and fictive zero-value symbols, said symbol sequence having at least one fictive symbol between two consecutive estimated symbols;
generating estimated signal values at the selected sampling time-points using the channel estimate and the symbol sequence;
generating an error signal at each of the selected sampling time-points of the indicated symbol using the observed, selected sampled signal values and the estimated signal values;
determining weighting factors using the error signals;
and generating a summed delta-metric for an indicated one of the state transitions of the indicated symbol, by multiplying the delta-metric values with their respective weighting factors, and summating.

2. The method according to claim 1, further comprising the step of successively adapting the channel estimate of the radio channel using the error signals, in accordance with a chosen adaptation algorithm.

3. The method according to claim 1, in which each state in the viterbi algorithm is connected with a respective channel estimate, and further comprising the steps of:
selecting a channel estimate which is connected to the state having the smallest metric value; and generating the estimated signal values using the selected channel estimate.

4. The method according to claim 1, 2 or 3, further comprising the steps of:

quadrating the value of the error signals at the selected sampling time-points; and lowpass filtering the squared error signals.

5. A method according to claim 4, further comprising the step of inverting the squared lowpass filter error signals to form the weighting factors.

6. A method according to claim 4, further comprising the steps of:

quadrating and summating coefficient values in the channel estimate; and dividing the resultant sum by the corresponding squared and lowpass filtered error signals to form the weighting factors.

7. An arrangement, in a receiver of a digital radio transmission system, for estimating symbols from a radio signal transmitted over a radio channel, the arrangement comprising:

a radio receiver having a filter which receives the radio signal and generates a baseband signal;

a first sampling unit which samples the baseband signal and delivers the sampled signal values at at least two sampling time-points with each symbol;

a correlating circuit, which generates a channel estimate for the radio channel using the sampled signal values and at least one symbol sequence known to the receiver;

a synchronizing circuit for determining a symbol sampling time-point in one of the sampling time-points;

a second sampling unit which is connected to the first sampling unit and is controlled by the synchronizing circuit, and delivers at least two observed sampled signal values with each symbol;

a channel equalizer which, for an indicated symbol according to the viterbi algorithm, generates a delta-metric value for each observed signal value of a state transition and also generates at least preliminarily estimated symbols;

a circuit, which generates a symbol sequence of the estimated symbols and fictive zero-value symbols which lie therebetween;

a channel-estimating filter, which generates the estimated signal values using the generated symbol sequence;

a difference former, which generates error signals for the indicated symbol using the observed and the estimated signal values;

a squaring and mean-value-forming circuit, which forms weighting factors in dependence on the error signals; and a metric calculating circuit, which generates a summed delta-metric value by multiplying the delta-metric values of a state transition by the corresponding weighting factors, and summates the products obtained.

8. The arrangement according to claim 7, wherein the channel-estimating filter has coefficient circuits whose values are adapted to the channel estimate of the radio channel by an adaptation circuit using the error signals in accordance with a selected adaptation algorithm.

9. The arrangement according to claim 7 or 8, wherein the arrangement includes:

squarers which quadrate the value of the error signals at the selected sampling time-points; and lowpass filters, which filter the squared error signals.

10. The arrangement according to claim 9, further including inverters, which receive the squared, lowpass filtered error signals and deliver the weighting factors.

11. The arrangement according to claim 9, further including:

inverters, which receive the squared lowpass filtered error signals and deliver corresponding inverted values;
and circuits which are connected to the inverters, and in which the values of the coefficient circuits are squared, summed and multiplied by said inverted values.

12. A method of detecting digital symbols transmitted over a communication channel using a viterbi algorithm having a number of states, the method comprising the steps of:

(12a) sampling a received signal to generate a received signal sequence comprising a plurality F of samples per symbol;

(12b) generating at least an initial value of an estimated impulse response of the communication channel, using the received-signal sequence and at least one symbol sequence known to the receiver, the estimated impulse response having a plurality F of taps per symbol;

(12c) determining the plurality F of delta-metric values according to the viterbi algorithm for each of a selected number of transitions between said states, each of said delta-metric values being associated with a respective one of the plurality F of the samples per symbol of the received-signal sequence, and also with a respective one of the plurality F of symbol sampled subsets of the estimated impulse response;

(12d) combining the plurality F of the delta-metric values determined by step (c) into a combined delta-metric value for each of said selected number of state transitions, by forming a weighted sum of said deta-metric values; and (12e) generating an at least preliminarily detected symbol sequence according to the viterbi algorithm based on the combined delta-metric values produced by step (12d).

13. The method of claim 12, wherein said combined delta-metric values are formed by multiplying the respective delta-metric value by a respective weighting factor and summing, the weighting factors being at least initially generated by the steps of:

(13a) zero-padding the known symbol sequence by inserting a number of zero symbols between at least two successive symbols in the known sequence, thereby generating a padded known symbol sequence having the plurality F of samples per symbol;

(13b) generating an error sequence having the plurality F
of samples per symbol by determining a difference between the received-signal sequence and a convolution of the estimated impulse response and the zero-padded known symbol sequence;

(13c) squaring the error sequence to form a squared error sequence;

(13d) splitting the squared error sequence into the plurality F of partial squared error sequences having one sample per symbol;

(13e) averaging the partial squared error sequences to form a respective averaged partial squared error signal;
and (13f) transforming the plurality F of the averaged partial squared error signals into the weighting factors.

14. The method of claim 13, wherein the weighting factors are at least initially generated as inverses of the averaged partial squared error signals.

15. The method of claim 13, wherein the weighting factors are at least initially generated as ratios of energies of the plurality F of the symbol-sampled subsets of the estimated impulse response and the respective averaged partial squared error signals.

16. The method of claim 13, wherein the weighting factors are at least initially generated as ratios of an estimate of the strength of the received signal and their respective averaged partial squared error signals.

17. The method of claim 12, further comprising the steps of generating an error sequence by:

(17a) zero-padding the at least preliminarily detected symbol sequence by inserting the number F-1 of zero symbols between at least two successive detected symbols, thereby generating a zero-padded detected symbol sequence having the plurality F of samples per symbol;

(17b) generating an error sequence having the plurality F
of samples per symbol by determining a difference between the received-signal sequence and a convolution of the estimated impulse response and the zero-padded detected symbol sequence.

18. The method of claim 17, further comprising the step of adapting the estimated impulse response initially produced by step (12b), based on the error sequence of step (17b) and a chosen adaptation algorithm.

19. The method of claim 17 or 18, further comprising adapting the weighting factors by the steps of:

(a) squaring the error sequence of step (17b) to form a squared error sequence;

(b) splitting the squared error sequence into the plurality F of partial squared error sequences having one sample per symbol;

(c) averaging each partial squared error sequence to form a respective averaged partial squared error signal; and (d) transforming the plurality F of the averaged partial squared error signals into the weighting factors.

20. The method of claim 19, wherein the weighting factors are inverses of the averaged partial squared error signals.

21. The method of claim 19, wherein the weighting factors are ratios of energies of each of the F symbol-sampled subsets of the estimated impulse response and the respective averaged partial squared error signals.

22. The method of claim 19, wherein the weighting factors are ratios of an estimate of the strength of the received signal and their respective averaged partial squares error signals.

23. The method of any one of claims 12 to 22, further comprising updating an estimated impulse response for at least one indicated state, and also updating weighting factors for at least said indicated state, comprising the steps of:

(23a) zero-padding the hypothetical symbols associated with said indicated state, by inserting a number F-1 of zero-symbols in between at least two succesive symbols, thereby generating a zero-padded hypothetical symbol sequence having the plurality F of samples per symbol;

(23b) generating an error sequence having the plurality F
of samples per symbol by determining a difference between the received-signal sequence and a convolution of the estimated impulse response associated with a preceding state having the best metric into said indicated state, and the zero-padded hypothetical symbol sequence associated with said indicated state;

(23c) updating the estimated impulse response associated with said indicated state using the impulse response estimate associated with the preceding state having the best metric into said dedicated state, the error sequence, and a chosen adaptation algorithm; and (23d) updating the weighting factors associated with said indicated state, by using the error sequence.

24. In a receiver of digital symbols transmitted over a communication channel, an apparatus for estimating the transmitted symbols using a viterbi algorithm having a number of states, the apparatus comprising:

(24a) means for sampling the received signal into a received-signal sequence comprising a plurality F of samples per symbol;

(24b) means for generating at least an initial value of an estimated impulse response of the communication channel, using samples generated by the sampling means and at least one symbol sequence known to the receiver, the generating means having the plurality F of taps for the impulse response;

(24c) means for determining the plurality F of delta-metric values according to the viterbi algorithm for each of a selected number of transitions between said states, each of said delta-metric values being associated with a respective one of the plurality F of the samples per symbol received from the sampling means, and also with a respective one of the plurality F of symbol-sampled subsets of the estimated impulse response;

(24d) means for combining the plurality F of the delta-metric values into a combined delta-metric value for each of said selected number of state transitions, the combining means being connected to the determining means of (24c) and forming a weighted sum of said deta-metric values; and (24e) means for generating an at least preliminarily detected symbol sequence according to the viterbi algorithm connected to the combining means of (24d).

25. The apparatus of claim 24, further comprising means for at least initially generating weighting factors, the apparatus comprising:

(25a) means for zero-padding the at least one known symbol sequence, the zero-padding means inserting a number F-1 of zero symbols between at least two successive symbols in said sequence into at least one padded known symbol sequence having the plurality F of samples per symbol;

(25b) means for generating an error sequence having the plurality F of samples per symbol, the error generating means comprising means for determining a difference between the received-signal sequence and a convolution of the estimated impulse response and the at least one zero-padded known symbol sequence;

(25c) means for squaring the error sequence to form a squared error sequence;

(25d) means for splitting the squared error sequence into the plurality F of partial squared error sequences having one sample per symbol;

(25e) means for averaging the partial squared error sequences to form a respective averaged partial squared error signal; and (25f) means for transforming the plurality F of the averaged partial squared error signals into the weighting factors.

26. The apparatus of claim 25, wherein the means for at least initially generating the weighting factors further includes means for inverting the averaged partial squared error signals.

27. The apparatus of claim 25, wherein the means for at least initially generating the weighting factors further includes means for generating ratios of energies of the plurality F of the symbol-sampled subsets of the estimated impulse response and the respective averaged partial squared error signals.

28. The apparatus of claim 25, wherein the means for at least initially generating the weighting factors further includes means for generating ratios of an estimate of the strength of the received-signal samples and their respective averaged paratial squared error signals.

29. The apparatus of claim 24, further comprising means for generating an error sequence, comprising:

(29a) means for zero-padding the at least preliminarily detected symbol sequence, the zero-padding means inserting the number F-1 of zero symbols between at least two successive detected symbols, thereby generating a zero-padded detected symbol sequence having the plurality F of samples per symbol;

(29b) means for generating an error sequence having the plurality F of samples per symbol, the error-generating means comprising means for determining a difference between the received-signal sequence and a convolution of the estimated impulse response and the zero-padded detected symbol sequence.

30. The apparatus of claim 29, further comprising means for adapting the estimated impulse response initially produced by the means of (24b), based on the error sequence from the generating means of (29b) and a chosen adaptation algorithm.

31. The apparatus of claim 29 and 30, further comprising means for adapting the weighting factors, comprising:

(31a) means for squaring the error sequence from the generating means of (29b), the squaring means forming a squared error sequence;

(31b) means for splitting the squared error sequence into the plurality F of partial squared error sequences having one sample per symbol;
(31c) means for averaging each partial squared error sequence, the averaging means forming a respective averaged partial squared error signal; and (31d) means for transforming the plurality F of the averaged partial squared error signals into the weighting factors.

32. The apparatus of claim 31, wherein the transforming means for the weighting factors comprises means for inverting the averaged partial squared error signals.

33. The apparatus of claim 31, wherein the transforming means for the weighting factors comprises means for forming ratios of energies of each of the F symbol-sampled subsets of the estimated impulse response and the respective averaged partial squared error signals.

34. The apparatus of claim 31, the transforming means for the weighting factors comprising means for forming ratios of an estimate of the strength of the received signal and their respective averaged partial squared error signals.

35. The apparatus of any one of claims 24 to 34, further comprising means for updating an estimated impulse response for at least one indicated state, and also means for updating weighting factors for said at least one indicated state, comprising:

(35a) means for zero-padding the hypothetical symbols associated with said indicated state, the zero-padding means inserting a number F-1 of zero-symbols inbetween at least two succesive symbols, thereby generating a zero-padded hypothetical symbol sequence having the plurality F
of samples per symbol;
(35b) means for generating an error sequence having the plurality F of samples per symbol, the generating means determining a difference between the received-signal sequence and a convolution of the estimated impulse response associated with a preceding state having the best metric into said indicated state, and the zero-padded hypothetical symbol sequence associated with said indicated state;
(35c) means for updating the estimated impulse response associated with said indicated state, the updating means using:
the impulse response estimate associated with the preceding state having the best metric into said indicated state;
the error sequence; and means for generating a chosen adaptation algorithm;
and (35d) means for updating the weighting factors associated with said indicated state, the updating means using the error sequence.

36. In a method of receiving successive digital symbols transmitted over a communication channel that uses a viterbi algorithm having a number of states, a method of estimating transmitted symbols which comprises the steps of:
(a) sampling a received signal at a recurrent plurality of time-points during each received symbol;
(b) correlating samples generated by step (a) with a predetermined sequence to produce at least an initial estimated impulse response of the communication channel;
(c) designating one of the plurality of time-points as a symbol-sampling time-point based on the correlation performed in step (b);
(d) selecting at least two samples for each received symbol, the selected samples being generated at the symbol-sampling time-point and at least one other time-point;
(e) for each symbol, determining a plurality of delta-metric values according to the viterbi algorithm for each transition between states, based on the selected samples and the estimated impulse response;
(f) for each symbol, weighting delta-metric values determined by step (e) and generating a combined delta-metric value for each transition between states by combining respective weighted delta-metric values determined by step (e); and (g) generating at least preliminarily estimated symbols using the combined delta-metric values generated in step (f).

37. The method of claim 36, further comprising the step of adapting the estimated impulse response by:
(a) generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the at least preliminarily estimated symbols;
(b) generating error signals as the difference between the selected samples of the received signal and the estimated signal values; and (c) successively adapting the estimated impulse response using the error signals and a selected adaptation algorithm.

38. The method of claim 37, wherein step (f) includes the step of generating respective weighting factors from the error signals and the estimated impulse response.

39. The method of claim 37, wherein step (f) includes the step of generating respective weighting factors for the delta-metric values from the error signals.

40. The method of claim 39, wherein the weighting factors are generated by squaring, averaging, and inverting the error signals.

41. The method of claim 36, wherein step (f) includes the step of generating respective weighting factors for the delta-metric values by:
(a) generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the at least preliminarily estimated symbols;
(b) generating error signals as the difference between the selected samples of the received signal and the estimated signal values; and (c) generating the weighting factors based on the error signals.

42. The method of claim 41, wherein the weighting factors are generated by squaring, low-pass filtering, and inverting the error signals.

43. The method of claim 36, wherein step (f) includes the step of generating respective weighting factors for the delta-metric values by:
(a) generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the predetermined sequence;
(b) generating error signals as the difference between the selected samples of the portion of the received signal corresponding to the sequence known to the receiver and the estimated signal values; and (c) generating the weighting factors based on the error signals.

44. The method of claim 43, wherein the weighting factors are generated by squaring, averaging, and inverting the error signals.

45. In a receiver of successive digital symbols transmitted over a communication channel that uses a viterbi algorithm having a number of states, an apparatus for estimating transmitted symbols comprising:
means for sampling a received signal at a recurrent plurality of time-points during each received symbol;
means for correlating samples generated by the sampling means with a predetermined synchronizing sequence to produce at least an initial estimated impulse response of the communication channel;

means for designating one of the plurality of time-points as a symbol-sampling time-point;
means, connected to the sampling means, for selecting at least two samples for each received symbol, the selected samples being the samples generated at the symbol-sampling time-point and at least one other time-point; and channel equalizer means for determining, for each symbol, a plurality of delta-metric values according to the viterbi algorithm for each transition between states based on the selected samples and the estimated impulse response produced by the correlating means, for weighting the delta-metric values for the transitions between states, and for generating preliminary estimated symbols from combined delta-metric values for the transitions between states, wherein each combined delta-metric value is a combination of respective weighted delta-metric values.

46. The apparatus of claim 45, further comprising means for adapting the estimated impulse response produced by the correlating means, wherein the adapting means includes:
means for generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the preliminary estimated symbols;
means for generating error signals as the difference between the selected samples of the received signal and the estimated signal values; and means for successively adapting the estimated impulse response using the error signals and a selected adaptation algorithm.

47. The apparatus of claim 46, further comprising means for weighting the delta-metric values based on the error signals and the estimated impulse response.

48. The apparatus of claim 46, further comprising means for generating respective weighting factors for the delta-metric values from the error signals.

49. The apparatus of claim 48, wherein the weighting factor generating means includes means for squaring, averaging, and inverting the error signals.

50. The apparatus of claim 45, further comprising means for generating respective weighting factors for the delta-metric values, wherein the weighting factor generating means comprises:
means for generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the preliminary estimated symbols; and means for generating error signals as the difference between the selected samples of the received signal and the estimated signal values;
wherein the weighting factors are generated from the error signals.

51. The apparatus of claim 50, wherein the weighting factor generating means includes means for squaring, low-pass filtering, and inverting the error signals.

52. The apparatus of claim 45, further comprising means for generating respective weighting factors for the delta-metric values, comprising:

means for generating estimated signal values corresponding to the symbol-sampling time-points and the other time-points based on the estimated impulse response and the predetermined sequence; and means for generating error signals as the difference between the selected samples of the portion of the received signal corresponding to the predetermined sequence and the estimated signal values;
wherein the weighting factors are generated from the error signals.

53. The apparatus of claim 52, wherein the weighting factor generating means includes means for squaring, averaging, and inverting the error signals.

54. A method of detecting digital symbols successively transmitted over a communication channel using a viterbi algorithm having a number of states, the method comprising the steps of:
(a) sampling a received signal to generate a received-signal sequence comprising a plurality F of samples per symbol;
(b) correlating the received-signal sequence with a predetermined synchronizing sequence to produce an estimated impulse response of the communication channel, the estimated impulse response having F taps per symbol;
(c) determining F delta-metric values according to the viterbi algorithm for each transition between states, each delta-metric value being associated with a respective one of the F samples per symbol of the received-signal sequence;
(d) weighting the F delta-metric values determined by step (c) and combining the weighted F delta-metric values determined by step (c) into a combined delta-metric value for each state transition by forming a weighted sum of the respective delta-metric values; and (e) generating an estimated symbol sequence according to the viterbi algorithm based on the combined delta-metric values produced by step (d).

55. The method of claim 54, wherein each combined delta-metric value is formed by multiplying each respective delta-metric value by a respective weighting factor and summing each weighted delta-metric value, the weighting factors being generated by the steps of:
padding the estimated symbol sequence with F-1 zero symbols, a zero symbol being inserted between successive estimated symbols to generate a padded estimated symbol sequence having F samples per symbol;
generating an error sequence having F samples per symbol by determining a difference between the received-signal sequence and a convolution of the estimated impulse response and the padded estimated symbol sequence;
squaring the error sequence to form a squared error sequence;
splitting the squared error sequence into F partial error sequences having one sample per symbol;
low-pass filtering each partial error sequence to form a respective averaged partial error signal; and transforming the F averaged partial error signal into the weighting factors.

56. The method of claim 55, wherein the weighting factors are inverses of the averaged partial error signals.

57. The method of claim 55, wherein the weighting factors are ratios of energies of each of F symbol-sampled subsets of the estimated impulse response and the respective averaged partial error signals.

58. The method of claim 55, wherein the weighting factors are ratios of a strength of the received signal and their respective averaged partial error signals.

59. The method of any one of claims 55 to 58, further comprising the step of adapting the estimated impulse response produced by step (b) based on the received-signal sequence and the error sequence.

60. The method of any one of claims 55 to 59, wherein a respective estimated impulse response is produced for each state, and the estimated impulse response used in forming the weighting factors is the one associated with the state having a smallest metric value.

61. The method of any one of claims 55 to 59, wherein a respective estimated impulse response and respective weighting factors are produced for each state, and the estimated impulse response and the estimated symbol sequence used in generating the weighting factors for a particular state are those associated with the preceding state producing a smallest metric value in the transition to the particular state.