BACKGROUND OF THE INVENTION Digital cellular systems require equalization to handle intersymbol interference (ISI) from time dispersion or delay spread in the channel. Typically, a non-linear equalizer (e. g., maximum likelihood sequence estimator (MLSE)) is used for such channels. An estimate of the radio channel is required for this equalization. Furthermore, if the channel changes fast enough, then the equalizer must continuously track the variations in this channel.

Our invention provides a novel method for estimation or tracking of such radio channels. This method is specially suited for those situations where the medium rays are spaced closer than one symboling interval.

A radio channel can be effectively broken into three parts. These are the transmit filter, which is known, the channel medium, which is unknown, and the receiver filter, which is known. The inputs to the radio channel are typically symbols from a discrete alphabet, such as +1 and-1, with a sampling rate equal to the baud rate. The transmit filter, the channel medium, and the receiver filter are typically modeled at a higher sampling rate, for example eight times the symbol rate. The equalizer uses a symbol-spaced model of the radio channel that relates the transmitted symbols to the output of the receive filter at the sampling

rate of one sample per symbol period (this application refers to this symbol-spaced channel as the effective baseband channel). The reference"Digital Communication" (Lee/Messerschmitt, Kulwer Academic Publishers, 1994) presents a method for directly estimating the effective baseband channel. Similarly, the reference"Licentiate Thesis, Lars Lindbum, Upsala University 1992, ISSN 03468887) presents a method for directly tracking the effective baseband channel. However, the present invention provides an improvement over these methods when the underlying rays of the channel medium are spaced closer than one symbol period. A. Khayrallah, R. Ramesh, G. Bottomley, and D. Koilpillai,"Improved Channel Estimation with Side Information""VTC-97, Vol. 2, pp. 1049-1051, presents a method for estimating fading mobile radio channels were the effective baseband channel is modeled as a convolution of a known pulse and an unknown medium. The authors only estimate the medium part of the channel explicitly, and reconstruct the effective baseband channel as a convolution of this estimated medium and the known pulse. However, as shown in the present application, the effective baseband channel cannot always be expressed as the convolution of a known pulse and an unknown medium. Specifically, the method of Khayrallah is not applicable when the channel medium has rays that are placed closer than one symbol-period (i. e., fractionally-spaced medium).

The present invention is directed to overcoming one or more of the problems discussed above in a novel and simple manner.

SUMMARY OF THE INVENTION In accordance with the invention there is provided a method for symbol-space estimation or tracking of fractionally-spaced radio channels.

Broadly, there is disclosed herein the method of estimating time varying radio

channels comprising the steps of receiving signals transmitted over the channels and converting the signals to digital samples for processing; determining positions offading rays of the channel medium; estimating impulse response of a channel medium for each of the estimated positions; and calculating estimates of the time varying radio channels using the digital samples and the estimated impulse responses.

It is a feature of the invention that the fading rays are spaced closer than a transmit symbol period.

It is another feature of the invention that the radio channel estimates are provided for a GSM equalizer.

It is a further feature of the invention that the radio channel estimates are provided for a code-division multiple access equalizer.

There is disclosed in accordance with another aspect of the invention a channel estimator adapted to track time varying radio channels including means for receiving signals transmitted over the channels, and converting the signals to digital samples for processing. Means are provided for determining positions of fading rays of the channel medium. Means estimate impulse response of a channel medium for each of the estimated positions. Means are provided for calculating estimates of the time-varying radio channel using the digital samples and the estimated impulse responses of the medium.

Further features and advantages of the invention will be readily apparent from the specification and from the drawing.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram of a digital communication system; Fig. 2 is a block diagram of an effective radio channel of the system of Fig. 1;

Fig. 3 is a block diagram ofthe symbol-spaced radio channel in accordance with the invention; Fig. 4 is a curve illustrating simulation results comparing a tracking equalizer to a non-tracking equalizer; Fig. 5 is a block diagram of a digital communication system including a baseband processor utilizing the method according to the invention; and Fig. 6 is a block diagram of the baseband processor of Fig. 5.

DETAILED DESCRIPTION OF THE INVENTION The invention relates to a method of symbol-spaced estimation and tracking of a fractionally-spaced fading radio channel. This can be used by any digital communication equalizer that needs to estimate or track a multi-ray fading channel such as a Typical Urban channel ofthe Global System for Mobile Communication (GSM). Using this tracking method, the block error rate (BLER) of current design GMS/GMSK equalizers can be improved by 4dB (Eb/NO) at 10% BLER with coding rate of 0.73 at 120 kilometers per hour.

The disclosed method can also be used in code-division multiple access (CDMA) systems to estimate or track the fading rays for use with rake fingers. For example, the method can be used by a CDMA finger searcher to find the rake fingers, assuming that the medium rays are spaced closer than one sampling period used to digitize the output of the radio channel.

Referring to Fig. 1, a block diagram illustrates a discrete-time model of a digital communication system 10. The digital communication system is modeled at baseband. The signal s (n) represents transmitted symbols. A block 12 takes the transmitted symbols and upsamples these the symbols from one sample per symbol toMsamples per symbol to produce

a signal e (n) (i. e., block 12 inserts"M-1"zeros between each consecutive pair of samples of s (n)). The block 12 is connected to a transmit filter 14 represented by impulse responsep (k).

The output of the transmit filter 14, represented by d (n), is transmitted through the channel medium blockl 6. The sequence c (k) represents the impulse response of the channel medium, which produces an output represented by f (n). A receiver includes a receive filter 18 represented by q (k) which produces an output z (n). The output of the receive filter 18 is coupled to a sample rate converter 20 which downsamples from M samples per symbol to 1 sample per symbol. The output, r (n), of the downsampler 20 is summed in a summer 22 with signal v (n) representing white Gaussian noise (AWGN) to produce an outputy (n). This output is supplied to an equalizer 24 which produces estimated symbols s (n). The AWGN should be added before the receive filter 18 and downsampling. However, if the receive filter 18 is flat over the bandwidth of the communication system, then adding the AWGN at the output of the downsampler 20 results in the same input to the equalizer.

With respect to the channel medium 16, co) represents a multipath ray with delay jT (I/T is the highest sampling frequency in the model of the digital communication system). Assuming that Tis small enough, the c (k)'s can accurately model the continuous-time medium. With a few rays, c (k) has only a few non-zero coefficients. In the model of Fig. 1, I p (k), c (k), and q (k) operate with a sampling frequency-. S (n) and r (n) operate at the lower sampling frequency of MT (i. e., they operate at the symbol rate). In the transmitter, a new symbol s (n) is modulated every MT seconds using the transmit filter 14. As a result:

Each symbol of s (n) belongs to a discrete alphabet.

The present invention utilizes several properties of the communications system 10 of Fig. 1. One property of the system 10 is that for any p (k), c (k), and q (k) the system with an input signal s (n) and output signal r (n) is linear time invariant (LTI). Specifically, where ho (k) = h (MK) with h (k) = p (k) * c (k) * q (k). Fig. 2 illustrates a block diagram of the effective baseband channel where blocks 14,16 and 18 of Fig. 1 are modeled with a block 26 representing the function h (k). It can be seen that z-transform of e (k), z (k), and r (k) are related as: S(zM)(3)E(z)= S(zM)H(z)(4)Z(z)= R (z) = [S ( ?) H (z)] #M (5) -S (z) Ho (z), (6) where in going from equation (5) to equation (6), the first Nobel Identity is used from P. P.

Vaidyanathan,"Multirate Systems and Filter Banks,"Prentice Hall, 1992. The function ho (n) is referred to herein as the effective baseband channel. An estimate of ho (n) is denoted by ho (n), which is used by the equalizer 24 to compute # (n).

The second property is that r (n) is a linear function of each tap of the medium channel. Specifically, ifri (n) is the output sequence resulting from s (k) (n) and c (n) = d (n- i), the symbol-spaced channel ho (n) can be expressed as

Using equation (7), r (n) can be written as s(n)*h0(n)(8)r(n)= where ri(n)*s(n).# Assuming that c (i) = 0 for i > P, equation (11) can be expressed in block diagram form as in Fig. 3.

As is apparent, Figs. 1 and 3 are two different representations of the same LTI system having input signal s (n) and output signal r (n). In Fig. 1, different portions of the model operate at different sampling rates, as discussed above. In the representation of Fig. 3, all parts of the model operate at the sampling frequency of 1/ (M7), which is the symbol rate (i. e., the baud rate).

For equalization, each tap ofho (k) needs to be estimated and/or tracked. Direct estimation or tracking of taps ho (k)'s is complicated. However, estimation or tracking of c (k)'s is relatively simple. Different taps of ho (k)'s are typically correlated with each other, and this correlation needs to be included in any good estimator of these taps. On the other hand, different taps of c (k) are independent; hence, the correlation among c (k)'s is not needed by an estimator of c (k)'s.

Because the statistical model for c (k)'s is relatively simple, in accordance with

the invention, ho (k)'s are estimated indirectly. This is done by first estimating each c (k), denoted by c (k), and then equation (7) is used to obtain ho (n) as: To estimate ho (n)'s, estimates are first obtained of the fading rays, c (n)'s, and then form the estimate of the baseband channel according to equation (12).

To obtain c (i)'s, first recall that y (n) = r (n) + v (n) (13) where in going from (13) to (14), equation (11) is used. Next, equation (14) is expressed in matrix form as: =Gc+v, (15) where go is an (Nx 1) vector whose n-th element is g, (n) = r, (n) * s (n). Recall that"*" represents discrete-time convolution.

From equation (15), it can be seen that the receive signal vector y is a linear function of the unknown fading rays, c, plus noise. Therefore, standard linear estimation techniques (e. g. H. Van Trees,"Detection, Estimation, and Modulation Theory, Part 1", 1968)

can be used to estimate c based on observations of y and based on knowing the transmitted sequence s (n). Note that matrix G can be formed from the knowledge of s (n)'s since, r (n)'s are known (see the line after equation (11)).

As an example, the Maximum-Likelihood estimate of c, denoted by c is givenby: #ML#,(16)(GHG)-1GH and the corresponding ML estimate of the effective baseband channel h0(n) is If the rays of the channel medium, c (k)'s, vary over time, the effective baseband channel will vary over time. If this variation is significant over the equalization window, it will become necessary to continuously track the variations of the effective baseband channel. In this section, the effective baseband channel is indirectly tracked by first tracking the fading rays of the channel medium, and then forming the tracked values of the effective baseband channel from the tracked values of the rays of the channel medium.

Let c (k, n) represent the value of the medium ray with delay"kT'at time "nit'. Let ho (7, n) represent the impulse response of the corresponding effective baseband channel at time"nMT'.

Various standard tracking algorithms (e. g., Recursive Least Squares (RLS), Least Mean Square (LMS), Normalized Least Mean Square (NLMS), Kalman Tracker, Kalman LMS Tracker, etc.) can be used to recursively estimate (or track) c (k, n). Let

c (k, n) represent the tracked value of c (k, n). The corresponding tracked value for the effective baseband channel, according to the current invention, is In order to utilize the disclosed method, the position of the medium rays (i. e., the value of les for which c (k, n)'s are non-zero) must first be determined. The present invention does not relate to how these delays are determined, but rather the estimation of the value of c (k, n) once it is known that c (k, n) is not zero for a particular k. Another way to state this problem is to say that first the position of the non-zero medium rays must be found.

In most communication systems, the transmitted symbols are grouped into blocks, and each block contains a few known symbols. These known symbols are sometimes referred to as the"preamble","midamble", or"training sequence". Given the training sequence, many methods can be used to determine the position of the non-zero delays. As an example, consider determining the best positions for placing three rays of the medium channel.

One iterative algorithm for placing these three rays is as follows. First, the best position for placing just one ray is found. Next, with the position of the first ray fixed, the best position for placing the second ray is found. During the search for the position of the second ray, the position of the first ray is fixed, but the value of the first ray is allowed to vary. Finally, with the position of the first and second rays fixed, the best position for placing the third ray is found. During the search for the position of the third ray, the position of the first ray and the position of the second ray are fixed, but the value of the first and the value of the second ray are allowed to vary. This procedure can easily be generalized to more than three rays.

Fig. 4 shows performance of a tracking equalizer in accordance with the

invention, and a non-tracking equalizer at 120 kilometers/hour based on GMSK modulation at symbol rate of 13 MHZ/48 and for a--dominated environment. The solid line represents No the tracking equalizer and the dashed line represents a non-tracking equalizer. The non- tracking equalizer estimates the channel of the midamble, and uses this channel estimate for demodulating the entire burst using a five-tap MLSE equalizer.

From Fig. 4, it can be seen that with coding rate 0.73, the non-tracking equalizer requires about 4dB more b to achieve 10% BLER compared to the tracking No equalizer.

Referring to Fig. 5, a block diagram of a hardware implementation of a radio communication system is illustrated. A signal to be transmitted is delivered on a line 101 to a digital symbol generator 102. The symbol generator 102 develops the transmitted symbols s (n) which are supplied to a digital transmitter 103. The transmitter 103 transmits the symbols via a channel medium, where they are received by an antenna 104 of a radio unit 105. The radio unit 105 filters and amplifies signals and converts them to a suitable form for processing.

The signals may be represented by analog samples that are provided to an analog-to-digital converter 106, which outputs digital samples for processing by a baseband processor 107. The baseband processor 107 produces the estimated symbols s (n). In the system 100, a transmission function 109 is represented to include the transmitter 103, the channel, the radio unit 105 and the A/D converter 106.

Referring to Fig. 6, the baseband processor 107 according to the invention is

illustrated. The baseband processor 107 receives signals from the transmission function 109 at a synchronization block 206. The output of the synchronization block 206 corresponds to the signal y (n) shown in Fig. 1, above. The synchronization block 206 is coupled to a channel tap estimator 202 and a branch metric processor 203. The channel tap estimator 202 receives the valuesy (n) and develops an estimate of ho (n) which is denoted by ho (n) that is also input to the branch metric processor 203. The branch metric processor 203 is connected to a sequence estimation processor 204. The processors 203 and 204, in the illustrated embodiment of the invention, implement a Viterbi equalizer function utilizing the digital samples and the channel tap estimates for determining the estimated symbols s (n).

As will be appreciated by one of ordinary skill in the art, the present invention may be embodied as methods or devices. Accordingly, the present invention may take the form of an entirely hardware embodiment, and entirely software embodiment, or an embodiment combining hardware and software aspects. The present invention has been described with respect to the block diagram illustrations and various equations. It should be understood that each block of the block diagram or combinations of blocks can be implemented by computer program instructions. These program instructions, which represent steps, may be provided to a processor to produce a machine. Likewise, the equations identified herein can be implemented similarly by computer program instructions.

Accordingly, blocks of the block diagram and the defined equations support combinations of means for producing the specified functions and combinations of steps for performing the specified functions. It will be understood that each block and combination of blocks can be implemented by special purpose hardware-based systems which perform the specified functions or steps, or combinations of special purpose hardware and computer instructions.

Thus, in accordance with the invention there is described a novel method for symbol-spaced tracking of a fractionally-spaced fading radio channel.