This application contains subject matter related to the subject matter disclosed in U.S. Pat. application Ser. No. 10/142,108 filed May 9, 2002.
The present invention relates to equalizers and, more particularly, to equalizers that adapt to the condition of a channel through which signals are received.
Linear adaptive equalizers having a plurality of taps are widely used in digital communication receivers in order to provide correction for multipath channel distortion. Adaptive algorithms, such as the least mean squares (LMS) algorithm, are implemented in order to determine the weight values for the taps of the equalizer. Such adaptive algorithms are easy to implement and provide reasonably good performance. However, under difficult channel conditions, these algorithms may fail to provide tap weights that converge to the desired values.
It is well known that this failure may be avoided if the tap weights, instead of being initialized to values of zero as is often done, are initialized at least somewhat close to their final desired values based on a knowledge of the channel impulse response (CIR). An estimate of the channel impulse response may be derived from an a priori known training signal periodically transmitted prior to, and/or along with, the unknown data. One such system with this feature is specified in the ATSC 8VSB standard for digital terrestrial television broadcasting.
The channel impulse response is typically estimated in a receiver by cross-correlating the training signal as received with a representation of the known transmitted training signal stored in the receiver. The Z-transform of the estimated channel impulse response is derived and inverted. From the inverted Z-transform, a vector is formed having a plurality of elements, and these elements are used to initialize a corresponding number of tap weights of the equalizer.
A conventional linear adaptive equalizer 10 that utilizes a transversal filter 12 is shown in
During training, and as described above, the conventional tap weight update algorithm 18 typically estimates the channel impulse response by a-periodically cross-correlating the training signal as received with a stored version of the known training signal. If s[k] is defined as the stored known training sequence for k=0 . . . (L−1), and if u[k] is defined as received data sampled at the symbol rate, with u[o] being the first received training symbol in the received signal, the cross-correlation is given by the following equation:
where Nc is the length of the causal response of the channel (post ghosts), and Na is the length of the anti-causal channel response (pre-ghosts).
The conventional tap weight update algorithm 18 then determines the Z-transform of h[i] and inverts the Z-transform in order to determine the tap weights that are supplied to the multipliers 161 through 16n.
The present invention provides a novel technique for forming a more accurate estimate of the channel impulse response h[i]. Alternatively and/or additionally, the present invention provides a novel technique for determining the inverse of the Z-transform of the estimated channel impulse response h[i].
In accordance with one aspect of the present invention, a method for determining the initial values of the tap weights for the taps of a linear equalizer comprises the following: determining a channel impulse response of a channel; and, determining, from the channel impulse response, only values corresponding to the weights of the equalizer taps necessary to achieve optimum initialization of the equalizer.
In accordance with another aspect of the present invention, a method for determining values for the tap weights of the taps of a linear equalizer comprises the following: determining a channel impulse response of a channel; and, performing a nested summation to determine from the channel impulse response only values corresponding to the weights of the equalizer taps necessary to achieve optimum initialization of the equalizer.
In accordance with still another aspect of the present invention, a method for determining values for the tap weights of the taps of a linear equalizer comprises the following: cross-correlating data received over a channel with a representation of the data as they should have been received so as to produce a correlation vector, wherein the correlation vector is characterized by a known noise component resulting from the finiteness of the correlation; subtracting the known noise component from the correlation vector so as to produce a channel impulse response related to the channel; and, performing a nested summation to determine from the channel impulse response only values corresponding to the weights of the equalizer taps necessary to achieve optimum initialization of the equalizer.
These and other features and advantages will become more apparent from a detailed consideration of the invention when taken in conjunction with the drawings in which:
Equation (1) results in a cross-correlation vector h[i] having N=Na+Nc+1 elements. The inventors have recognized that, due to the finiteness of the correlation operation as represented by equation (1), the cross-correlation vector h[i] is characterized by an a prior known first noise component in the main path and in each reflected path in proportion to the relative gain in each respective path, a second noise component due to the additive white Gaussian noise (AWGN), and a third noise component due to neighboring unknown symbols. Regarding the first noise component, the a-periodic autocorrelation of a binary valued ({+A, −A}) PN sequence of length L (PNL) is given by the following equation:
rPNL(m)=LA2, if m=0, ±L, ±2L, . . . =−A2, otherwise (2)
However, if the PN sequence is finite and the standard linear correlation is used, then the autocorrelation values corresponding to the non-zero lags will not be constant and will not be as low as −A2. As an illustration, consider the following sequence composed of six PN511 appended back-to-back:
y=[PN511, PN511, PN511, PN511, PN511, PN511] (3)
The cross-correlation rxy[m] of x and y where x=[PN511] is shown in
The training sequence used as part of the ATSC 8VSB digital television broadcast standard is actually given by the following equation:
{tilde over (s)}=[PN511, PN63, PN63, PN63] (4)
As explained above, the correlation of this sequence with the received signal will be “noisy” because the sequences are finite in length and will not achieve their low correlation value for non-zero lag. The inventors have recognized that the channel impulse response may be more accurately estimated by compensating for this “noise” component.
In general, this compensation is accomplished by deriving a representation of the “noise” component and by iteratively subtracting it from the main and each dominant reflected path detected in the initial channel impulse response estimate.
Let S be a vector representation of the elements of the known (i.e., as transmitted) training sequence s[k], for k=0 . . . (L−1). Also, let q[k] represent the elements of a vector [Z, S, Z] of length L+2 (N−1), where the training sequence has a length L, where Z is a zero vector of length (N−1), and where q[0] coincides with s[0].
A correction vector {tilde over (r)}[m] representing the noise component due to the finiteness of the correlation may be derived according to the following equation:
Equation (5) is modified by setting the main path element of rf[m]=0 in order to derive the final correction vector {tilde over (r)}[m] according to the following equation:
where the elements of rf[m] have been normalized to the main path element rf[0].
The channel estimate h[i] may now be corrected in a series of steps. In the first step, main path correction is achieved by subtracting {tilde over (r)}[m] from h[i] with the zero elements of the respective vectors in alignment and with {tilde over (r)}[m] truncated to −Na≦m≦Nc. This operation results in a first correction of vector h[i], with no modification of the main path element h[0] because {tilde over (r)}[0]=0.
The foregoing subtraction operation is repeated in turn for each dominant reflected path to further refine h[i]. Thus, the next largest data element of h[i] is located at h[τ] and is assumed to have a level gh[0], where |g|≦1. If |g| is greater than a predetermined threshold value, the foregoing vector subtraction is repeated with h[τ] in alignment with {tilde over (r)}[0] and with {tilde over (r)}[m] truncated to −(Na−τ)≦m≦(Nc−τ). This operation, thus, results in yet a further refinement in the accuracy of correlation vector h[i]. This process is repeated in subsequent steps until there are no longer any elements of h[i] satisfying the threshold condition. The final corrected channel estimate is referred to as h′[i].
As an example of the foregoing, a linear equalizer for an ATSC transmission with Na set to 127 and Nc set to 448 may be assumed. Therefore, N is equal to 576 according to this assumption. It may further be assumed that the training signal S has 515 elements (L=515) consisting of 4 segment sync symbols and a 511 symbol pseudorandom number sequence (PN511). As indicated above, the vector used to derive {tilde over (r)}[m] is defined as [Z, S, Z]. This definition is used in the general case. In the case where the signals of interest are VSB signals such as those used in digital television, the vector used to derive {tilde over (r)}[m] is defined as [Z, S, W], where W may be assumed to consist of three 63 symbol pseudorandom number sequences (3PN63), 24 mode symbols, and 362 zeros. Vector Z is a zero vector having N−1=575 elements. Accordingly, the length of vector q is 575+515+575=1665.
In this example, rf[m] is given by the following equation:
and h[i] is initially given by the following equation:
The first and second subtraction iterations for this example are illustrated in
As shown in
This process is repeated in subsequent steps until there are no longer any elements of the vector h[i] that satisfy the threshold condition. The final corrected channel estimate is designated h′[i].
As another example, the foregoing conditions may be assumed except that the channel is a purely anti-causal channel such that Nc=0. In this case, N=Na+1=128 and rf[m] and h[i] are given by the following equations:
The first and second subtraction iterations for this example are shown in
As shown in
This process is repeated in subsequent steps until there are no longer any elements of the vector h[i] that satisfy the threshold condition. The final corrected channel estimate is designated h′[i].
The vector h′[i] represents the channel impulse response corrected for the “noise” component caused by the finiteness of the correlation process. The Z transform of this vector may be determined and inverted in order to determine the tap weights for the initialization of the taps of the equalizer.
A linear adaptive equalizer 30 as shown in
As described up to this point, the linear adaptive equalizer 30 is the same as the conventional linear adaptive equalizer 10 shown in
It should be noted that equation (5) does not depend on any received signal and instead relies only on the known training signal s[k] and the vector [Z, S, W]. Because both of these vectors are known a priori, the correction vector rf[m] may be programmed into the tap weight initializer 44 in the factory.
The inventors have developed a novel technique for forming the inverse of the Z transform of a channel impulse response vector h[i] for an anti-causal channel. Although this technique is described herein for the corrected vector h′[i], it should be recognized that this technique also may be used for uncorrected channel impulse response vectors.
For the case of a purely anti-causal (precursor only) multi-path channel, the corrected channel impulse response estimate produced by the previously described technique is h′[i]. The Z-transform of h′[i] is the channel transfer function and is given by the following equation:
If it is assumed that a level thresholding algorithm is used after the correlation in order to set the elements of h′(i) below a given threshold to zero, equation (11) can be written as a finite sequence according to the following equation:
Equation (12) indicates that there are m paths exceeding the level thresholding criteria each with a respective delay Dm.
The inverse of the Z-transform as given by equation (11) traditionally yields the desired initial tap weights of a zero forcing equalizer having N taps. As will be shown below, this inverse of H′[z], denoted as H1′[z], has an infinite number of terms and, therefore, cannot be used directly. However, the terms in this infinite number of terms are usually truncated to provide a realizable solution.
More specifically, it is known that the inverse of H′[z] as given in equation (11) can be expressed as an infinite summation according to the following equation:
where
γ(z)=βmzDm+ . . . +β2zD2+β1zD1
whenever
Since equation (13) is an infinite series, it has no practical realization and, therefore, must be truncated before it can be implemented. The simple approach for truncating equation (13) is to drop all terms of the summation exceeding └N/Dam┘, where └ ┘ denotes rounding to the largest integer smaller than N/Dam. This approach results in a finite summation sequence that provides an estimate of H1′[z] and that may be referred to as Hff. The finite summation sequence Hff is given by the following equation:
where
N′=└N/Dm┘
As an example, it is assumed that there are two dominant multipaths. Thus, m=2. It may further be assumed that the first multipath has a delay D1=19, that the second multipath has a delay D2=30, and that the equalizer has 99 taps. Therefore, in this example, N′=99/30=3. Using equation (14), Hff is given by the following equation:
The tap weight values from equation (15) are shown in Table 1 below.
Table 1 does not provide for the best initialization of the equalizer because the truncation of equation (13) given by equation (15) does not include all the available tap weights within the span of the equalizer.
Thus, while the above described truncation of Hi′[z] is straightforward, it does not provide the most accurate solution because it does not provide all of the tap weight values necessary to achieve optimum equalizer initialization. A better solution, for example, might be realized if Hff is calculated assuming N (the number of equalizer taps) to be larger than the actual number of equalizer taps, and then truncating the solution by dropping all tap weight values for taps greater than the actual number of tap values. This approach typically results in additional tap values as compared to those that would be produced using equation (14), and these additional tap weights provide a better solution.
However, it is difficult to predict how large the assumed value of N needs to be before it is assured that all relevant tap weight values necessary to achieve optimum initialization will be included in the solution.
The novel technique described herein for forming the inverse of a channel impulse response vector h[i] for an anti-causal channel avoids this uncertainty by providing a technique for calculating Hff such that the calculation includes all relevant tap weight values (i.e., all tap weight values less than the length of the physical equalizer) necessary to achieve optimum initialization. The algorithm of this novel technique simultaneously inverts and truncates H′(z) to provide an optimum inverse given the constraints of the equalizer being implemented (in particular, the number of taps available in the equalizer).
This novel technique was invented by the inventors as a result of their discovery that all necessary tap weight values to achieve optimum initialization are produced using a nested summation approach. In the case of only two dominant multi-paths, the following nested summation equation is used:
For the exemplary values given above, Dk,j=19k+11j, kmax=5, and jmax(k)=(99−19k)/11. Using these values in equation (16) produces the following equation:
Expanding equation (17) results in:
The optimum solution given by equation (18) is illustrated in Table 2, which includes four tap weight values (for taps 76, 87, 95 and 98) not included in Table 1.
It will be observed that equation (16) includes two nested summations because there are assumed to be only two dominant multipaths characterizing the Z-transform of the channel impulse response.
Equation (16) can be generalized according to the following equation in order to include m nested summations for the case of m multipaths:
and where
The outermost summation of these m nested summations is performed over limits between 0 and the number of equalizer taps N divided by the shortest delay D characterizing the dominant multi-paths. In the example above where m is 2, D1 is 19, and D2 is 30, the shortest delay is 19.
As shown in
A block 62 stores the “noise” caused by the finiteness of the cross correlation in accordance with equations (5) and (6) and iteratively subtracts this “noise” from the cross correlation as discussed above to produce the corrected cross correlation h′[i]. A block 64 determines the inverted Z transform of h′[i] in accordance with equations (11) and (16) (if m=2) or in accordance with equations (11) and (19) (in the general case) in order to determine the initial tap weights to be applied to the taps of the multipliers 361 through 36n of
Modifications of the present invention will occur to those practicing in the art of the present invention. For example, the present invention as described above relies on the use of a training signal. However, other signals, such as data signals, can be used to implement the present invention.
Accordingly, the description of the present invention is to be construed as illustrative only and is for the purpose of teaching those skilled in the art the best mode of carrying out the invention. The details may be varied substantially without departing from the spirit of the invention, and the exclusive use of all modifications which are within the scope of the appended claims is reserved.
The present application claims the benefit of Provisional Application Ser. No. 60/302,986 filed on Jul. 3, 2001.
Number | Name | Date | Kind |
---|---|---|---|
4550415 | Debus et al. | Oct 1985 | A |
5008937 | Yamamura et al. | Apr 1991 | A |
5031193 | Atkinson et al. | Jul 1991 | A |
5117291 | Fadavi-Ardekani et al. | May 1992 | A |
5297169 | Backstrom et al. | Mar 1994 | A |
5353306 | Yamamoto | Oct 1994 | A |
5524023 | Tsujimoto | Jun 1996 | A |
5596607 | Larsson et al. | Jan 1997 | A |
5604724 | Shiokawa | Feb 1997 | A |
5646964 | Ushirokawa et al. | Jul 1997 | A |
5675394 | Choi | Oct 1997 | A |
5689528 | Tsujimoto | Nov 1997 | A |
5724390 | Blaker et al. | Mar 1998 | A |
5751768 | Guglielmi et al. | May 1998 | A |
5784304 | Koike | Jul 1998 | A |
5970092 | Currivan | Oct 1999 | A |
6084907 | Nagano et al. | Jul 2000 | A |
6411253 | Cox et al. | Jun 2002 | B1 |
6678310 | Andren et al. | Jan 2004 | B1 |
6804694 | Albert et al. | Oct 2004 | B2 |
20010004384 | Takanashi et al. | Jun 2001 | A1 |
Number | Date | Country |
---|---|---|
0 637 139 | Feb 1995 | EP |
WO 9949586 | Sep 1999 | WO |
WO 9949586 | Sep 1999 | WO |
Number | Date | Country | |
---|---|---|---|
20030063665 A1 | Apr 2003 | US |
Number | Date | Country | |
---|---|---|---|
60302986 | Jul 2001 | US |