Patent Grant
8320442
The present invention relates to the estimation of channel impulse responses for decision feedback equalizers.
Since the adoption of the Advanced Television Systems Committee (ATSC) digital television (DTV) standard in 1996, there has been an ongoing effort to improve the design of receivers built for the ATSC DTV signal. The primary obstacle that faces designers in designing receivers so that they achieve good reception is the presence of multipath interference in the broadcast television channel.
The broadcast television channel is a relatively severe multipath environment due to a variety of conditions that are encountered in the channel and at the receiver. Strong interfering signals may arrive at the receiver both before and after the largest amplitude signal. In addition, the signal transmitted through the channel is subject to time varying channel conditions due to the movement of the transmitter and signal reflectors, airplane flutter, and, for indoor reception, people walking around the room. If mobile reception is desired, movement of the receiver must also be considered. Designers add equalizers to receivers in order to cancel the effects of multipath interference and thereby improve signal reception.
Because the channel is not known a priori at the receiver, the equalizer must be able to adapt its response to the channel conditions that it encounters and to changes in those channel conditions. To aid in the convergence of an adaptive equalizer to the channel conditions, the field sync segment of the frame as defined in the ATSC standard may be used as a training sequence for the equalizer.
The frame as defined in the ATSC standard is shown in
The first segment in each field is a field sync segment. As shown in
Because the first 704 symbols of each field sync segment are known, these symbols, as discussed above, may be used as a training sequence for an adaptive equalizer. The original Grand Alliance receiver used an adaptive decision feedback equalizer (DFE) with 256 taps. The adaptive decision feedback equalizer was adapted to the channel using a standard least mean square (LMS) algorithm, and was trained with the field sync segment of the transmitted frame.
However, because the field sync segment is transmitted relatively infrequently (about every 260,000 symbols), the total convergence time of this equalizer is quite long if the equalizer adapts only on training symbols prior to convergence. Therefore, it is known to use the symbol decisions made by the receiver in order to adapt equalizers to follow channel variations that occur between training sequences.
An adaptive decision feedback equalizer in an 8 VSB receiver would be expected to use an 8 level slicer to make the symbol decisions that would be used to adapt the equalizer to the channel between transmissions of the training sequence. However, use of a symbol slicer results in many symbol decision errors being fed to the feedback filter of the decision feedback equalizer when the channel has significant multipath distortion or a low signal to noise ratio. These errors give rise to further errors resulting in what is called error propagation within the decision feedback equalizer. Error propagation greatly degrades the performance of the decision feedback equalizer.
The present invention instead relies on decoders to avoid the convergence and tracking problems of previous decision feedback equalizers.
In accordance with one aspect of the present invention, a method of operating a decision feedback equalizer comprises the following: making first symbol decisions from an output of the decision feedback equalizer, wherein the first symbol decisions are characterized by a relatively long processing delay; making second symbol decisions from the output of the decision feedback equalizer, wherein the second symbol decisions are characterized by a relatively short processing delay; and, determining tap weights for the decision feedback equalizer based on the first and second symbol decisions.
In accordance with another aspect of the present invention, a decision feedback equalizer comprises a feed forward filter, a feedback filter, a summer, first and second decoders, and a tap weight controller. The feed forward filter receives data to be equalized. The summer combines outputs from the feed forward filter and the feedback filter to provide an equalizer output. The first decoder is characterized by a relatively short processing delay, and the first decoder decodes the equalizer output to provide a first decoded equalizer output and supplies the first decoded equalizer output as an input to the feedback filter. The second decoder is characterized by a relatively long processing delay, and the second decoder decodes the equalizer output to provide a second decoded equalizer output. The tap weight controller determines tap weights based on the first and second decoded equalizer outputs and supplies the tap weights to the feed forward filter and the feedback filter.
In accordance with still another aspect of the present invention, a decision feedback equalizer comprises a feed forward filter, a feedback filter, a summer, first and second decoders, and a tap weight controller. The feed forward filter receives data to be equalized. The summer combines outputs from the feed forward filter and the feedback filter to provide an equalizer output. The first decoder decodes the equalizer output to provide a first decoded equalizer output and supplies the first decoded equalizer output as an input to the feedback filter. The second decoder characterized by a relatively long processing delay and by relatively shorter processing delays, the second decoder decodes the equalizer output to provide a second decoded equalizer output in accordance with the relatively long processing delay and third decoded equalizer outputs in accordance with the relatively shorter processing delays. The tap weight controller determines tap weights based on the second and third decoded equalizer outputs and supplies the tap weights to the feed forward filter and the feedback filter.
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:
The signal from the channel is processed by an automatic gain controller 16, which provides the equalizer input signal y. A channel impulse response and noise estimator 18 uses the transmitted training sequence as received in the equalizer input signal y and a stored version of the transmitted training sequence to provide an estimate ĥ0 of the channel impulse response. A tap weight calculator 20 calculates a set of tap weights based on the initial estimate ĥ0 of the channel impulse response using, for example, a minimum mean square error (MMSE) based algorithm, and supplies this set of tap weights to a decision feedback equalizer 22 that includes a feed forward filter 24 and a feedback filter 26.
The decision feedback equalizer 22 equalizes the data symbols contained in the equalizer input signal y based on these training sequence based tap weights and includes a summer 28 which supplies the output of the decision feedback equalizer 22 to the short traceback trellis decoder 12 and to the long traceback trellis decoder 14. The output, such as the maximum delay output, of the long traceback trellis decoder 14 forms the symbol decisions b. The feedback filter 26 filters the output of the short traceback trellis decoder 12, and the filtered output of the feedback filter 26 is subtracted by the summer 28 from the output of the feed forward filter 24 to provide the equalizer output.
The equalizer input signal y is delayed by a delay 30, and the delayed equalizer input signal y and the symbol decisions b are processed by a least squares channel impulse and noise update estimator 32 that produces an updated channel impulse estimate ĥLS. A tap weight calculator 34 uses the updated channel impulse estimate ĥLS to calculate an updated set of tap weights for the decision feedback equalizer 22. The tap weights determined by the tap weight calculator 34 are provided to the decision feedback equalizer 22 during periods when the tap weights based on the training sequence are not available from the tap weight calculator 20. The delay imposed by the delay 30 is equal to the combined processing delay of the decision feedback equalizer 22 and the long traceback trellis decoder 14.
Because the data symbols in an 8-VSB system are trellis coded, it is desirable to make use of the long traceback trellis decoder 14 as the symbol decision device to supply symbol decisions to the least square channel impulse and noise update estimator 32. By using a trellis decoder instead of a symbol slicer, the number of symbol decision errors supplied to the feedback filter 26 is reduced.
The reliability of a trellis decoder is proportional to its traceback depth. The long traceback trellis decoder 14, because of its longer traceback depth, produces more reliable decisions. However, because of its longer traceback depth, the decision process of the long traceback trellis decoder 14 incurs a longer delay.
By contrast, the symbol decisions of the short traceback trellis decoder 12 are less reliable because of its shorter traceback depth. However, while its symbol decisions are less reliable than a trellis decoder with a longer delay, the short traceback trellis decoder 12 is still significantly more reliable than an 8 level symbol slicer.
It is well known that a symbol decision device with a delay greater than zero creates a problem for a decision feedback equalizer with respect to the cancellation of short delay multipath. Therefore, decision feedback equalizers for 8 VSB receivers with a zero delay trellis decoder in the feedback loop have been used to reduce error propagation. Thus, the decision feedback equalizer 22 uses the short traceback trellis decoder 12 in the feedback loop of the feedback filter 26.
As indicated above, the output of the decision feedback equalizer 22 is the output of the summer 28. This output is fed to the long traceback trellis decoder 14. The long traceback trellis decoder 14 has a long traceback depth (e.g., traceback depth=32, delay=12×31=372 symbols). The long traceback trellis decoder 14 provides the final bit decisions for the subsequent stages of the receiver in which the decision feedback equalizer 22 is used. Also, as described below, the long traceback trellis decoder 14 provides symbol decisions used by the least squares channel impulse and noise update estimator 32 in order to calculate the updated channel impulse response estimate that is used by the tap weight calculator 34 to calculate the updated tap weights for the decision feedback equalizer 22 so that the decision feedback equalizer 22 can follow variations in the channel impulse response that occur between training sequences.
Thus, the channel impulse response estimate ĥ0 is formed by the channel impulse response and noise estimator 18 from the received training sequence and a set of tap weights are calculated by the tap weight calculator 20 from that channel impulse response estimate. Then, as the decision feedback equalizer 22 runs, reliable symbols decisions are taken from the long traceback trellis decoder 14 as relatively long pseudo training sequences, and these relatively long pseudo training sequences are used by the least squares channel impulse and noise update estimator 32 to calculate the updated channel impulse response estimates ĥLS from which updated decision feedback equalizer tap weights are calculated by the tap weight calculator 34. This process allows for the tracking of time varying channel impulse responses.
As indicated above, the channel impulse response estimate ĥ0 is based on the received training sequence. The channel impulse response estimate ĥ0 to be estimated is of length Lh=Lha+Lhc+1 where Lha is the length of the anti-causal part of the channel impulse response estimate ĥ0 and Lhc is the length of the causal part of the channel impulse response estimate ĥ0. The length of the training sequence is Ln.
It may be assumed that the Ln long vector of the a priori known training symbols as transmitted is given by the following expression:
a=[a0, - - - , aL
The vector of received symbols is given by the following equation:
y=[yL
The first received training data element is designated y0. Typically, this would mean that y0 contains a contribution from the first transmitted training symbol multiplied by the maximum magnitude tap of the channel impulse response vector h. Note that the vector y contains data elements comprised of contributions due to multipath only of a priori known training symbols. Also, the vector y does not include y0 which may contain contributions of unknown 8 level symbols due to multipath.
A convolution matrix A of size (Ln−Lha−Lhc)×(Lha+Lhc+1) may be formed from the known transmitted training symbols as given by the following equation:
Because the vector y of received symbols is given by the following equation:
y=Ah+v (4)
where h is the channel impulse response vector of length Lh and v is a noise vector, the least squares channel impulse response estimate is given by the solution of equation (4) according to the following equation:
ĥ0=(ATA)−1ATy (5)
However, this method is only effective if Ln satisfies the following inequality:
Ln≧2(Lha+Lhc)−1 (6)
If the training sequence is too short with respect to the length of the channel impulse response, then this method does not produce a good result because the system of equations given by equation (4) to be solved is underdetermined, which is often the case for 8 VSB terrestrial channels. For example, with Ln=704, the channel impulse response must be less than 352 symbols long. However, longer channel impulse responses are commonly found in practice.
A better method for finding the channel impulse response is based on a modified form of the convolution matrix A. A long vector a of length Ln of a priori known training symbols is again given by the expression (1). However, the convolution matrix A this time is an (Ln+Lha+Lhc)×Lh convolution matrix comprising training symbols and zeros and is given by the following equation:
The vector of received symbols is given by the following equation:
y=[y−Lha, - - - , y0, - - - , yLn+Lhc−1]T (8)
where y0 through yLn−1 are the received training symbols. So, the vector of equation (8) contains the known training symbols as well as contributions from random symbols before and after the training sequence due to multipath.
Again, equation (4) needs to be solved. Now, the convolution matrix A is a taller matrix because zeros have been substituted for the unknown symbols that surround the training sequence. This new convolution matrix A yields an over-determined system of equations.
The initial channel impulse response and noise estimator 18 solves equation (4) according to equation (5) using the new convolution matrix A of equation (7) and vector y of equation (8) to produce the channel impulse response estimate ĥ0. More complicated methods may be utilized to give even more accurate results, if necessary.
The tap weight calculator 20 uses the channel impulse response estimate ĥ0 to calculate a set of minimum mean square error (MMSE) tap weights for the decision feedback equalizer 22. Methods for calculating minimum mean square error tap weights from a channel impulse response are well known. Alternatively, the tap weight calculator 20 may use other methods such as the zero-forcing method to calculate the tap weights.
Accurate channel impulse response estimate updates can also be calculated between training sequences (when only a priori unknown symbols are received). For example, a least squares channel impulse response estimation may be calculated from an over determined system of equations. Dynamic changes to the channel impulse response may be accurately tracked by using receiver trellis decoder decisions on input symbols to form a long sequence of near perfectly decoded symbols. This sequence should have relatively few errors, even near threshold, and is selected to be long enough so that the underdetermined system problem of the “too short” 8 VSB training sequence is eliminated. The channel impulse response may be, for example, updated as often as once per segment (or more or less often).
The updated channel impulse response to be estimated is, as before, of length Lh=Lha+Lhc+1 where Lha is the length of the anti-causal part of the channel impulse response and Lhc is the length of the causal part of the channel impulse response. A vector b of length Lb is defined as the reliable trellis decoder decisions on the input symbols that are provided by the long traceback trellis decoder 14. Also, a Toeplitz matrix B is then defined according to the following equation:
where the elements are real and consist of the symbol decisions in the vector b. To ensure an over determined system of equations, Lb is given by the following inequality:
Lb≧2Lh−1 (10)
The Toeplitz matrix B is of dimension (Lb−Lh+1)×Lh with (Lb−Lh+1)≧Lh.
The received signal vector y has elements yi for Lhc≦i≦(Lb−Lha−1) where yi is the received symbol corresponding to the symbol decision bi. The received signal vector y is given by the following equation:
y=Bh+v (11)
where h is the Lh long channel impulse response vector and v is a noise vector. The least squares solution for h is given by the following equation:
ĥLS=(BTB)−1BTy (12)
By utilizing reliable trellis decoder input symbol decisions, there is sufficient support for calculating a channel impulse response estimate with the required delay spread. As required by inequality (10), the vector b of symbol decisions must be at least twice as long as the channel impulse response being estimated. The system of equations is sufficiently over determined in order to diminish the adverse effect of additive White Gaussian Noise (AWGN). Therefore, a vector b of symbol decisions that is longer than twice the channel impulse response length is preferred.
The tap weight calculations performed by the tap weight calculator 20 and the tap weight calculator 34 require not only a channel impulse response estimate but also a noise estimate. The noise may be estimated by calculating an estimate of the received vector y according to ŷ=Aĥ where ĥ is the latest calculated channel impulse response estimate. Then, the noise estimation is given by the following equation:
where ∥. ∥ is the 2-norm.
In order to apply the above equations to an 8 VSB receiver, the following parameters may be used as an example: Lh=512, Lha=63, Lhc=448, Lb=2496, and Ln=704. The vector b is formed from a sequence of trellis decoder decisions made by the long traceback trellis decoder 14 on the input symbols. The delay (31×12=372) of the long traceback trellis decoder 14 is not significant compared to a channel impulse response estimate update rate of once per segment. Normally, the long traceback trellis decoder 14 would just make output bit pair decisions, but it can also make equally reliable decisions on the input symbols.
The vector b, for example, may be selected as three segments (Lb=2496 symbols) long. So, three data segments may be used to produce a single channel impulse response estimate update. A new channel impulse response update can be obtained once per segment by proceeding in a sliding window manner. Optionally, several consecutive channel impulse response estimate updates can be averaged in order to further improve channel impulse response accuracy if necessary. This additional averaging can be a problem if the channel impulse response is varying rapidly.
A vector b with fewer than three segments of symbol decisions may be used provided that, as stated in inequality (10), the length of the vector b is at least twice as long as the channel impulse response to be estimated. As previously stated, however, long b vectors help to diminish the adverse effects of AWGN.
The latency time (which may be referred to as Tap Update Latency or TUL) involved in updating the decision feedback equalizer 22 with new tap weights is caused by the sum of (i) the symbol decision delay of the long traceback trellis decoder 14, (ii) the time delay resulting from the calculation by the least squares channel impulse and noise update estimator 32 of the channel impulse response estimate update, and (iii) the time delay resulting from the calculation by the tap weight calculator 34 of the MMSE tap weights.
The delay of the first item (i) may be reduced if, instead of using only decisions of the long traceback trellis decoder 14 for the channel impulse response estimate update, a combination of symbol decisions from the long traceback trellis decoder 14 and the short traceback trellis decoder 12 is used. The use of this combination of symbol decisions is illustrated in
The first row of the timing diagram in
The second row represents the delay that the processing of the decision feedback equalizer 22 imposes on these segment time periods as the corresponding equalized segments exit from the output of the decision feedback equalizer 22 and are provided to the long traceback trellis decoder 14. As shown in
The third row represents the additional delay that the processing of the long traceback trellis decoder 14 imposes on these segment time periods as the corresponding segments of symbol decisions exit from the output of the long traceback trellis decoder 14 and are provided to the least squares channel impulse and noise update estimator 32. As shown in
The fourth row represents the additional delays of making the channel impulse response and tap weight calculations by the least squares channel impulse and noise update estimator 32 and the tap weight calculator 34. For the sake of convenience (and not of necessity), it may be assumed that each of the delays given in items (i), (ii), and (iii) above is a ½ segment delay. With these assumptions, the updated tap weights calculated by the tap weight calculator 34 from the vector b that is composed of the symbol decisions in the three segment time periods 1, 2, and 3 will not be applied to the decision feedback equalizer 22 until after the second half of the equalized segment in segment time period 5 begins being output from the decision feedback equalizer 22. This corresponds to a 1.5 segment update delay. Accordingly, the tap update latency TUL is 1.5 segments.
In a channel whose channel impulse response is rapidly changing, this delay between (i) the time that segments are processed by the decision feedback equalizer 22 and (ii) the time at which the updated tap weights calculated on the basis on these segments are applied to the decision feedback equalizer 22 may degrade performance of the decision feedback equalizer 22 because the channel impulse response changes too much between the end of segment 3 and the beginning of segment 5.
Several delay assumptions are made above for the purpose of a clear explanation. However, these assumptions are not intended to be limiting.
The timing diagram of
The size of the portion of the three segment long decision vector b that is contributed by the short traceback trellis decoder 12 is chosen to be equal to the delay imposed by the processing of the long traceback trellis decoder 14. As an example, given the assumptions discussed above, this delay is 0.5 segment and removes the delay imposed by the processing of the long traceback trellis decoder 14 from the tap update latency TUL, thereby reducing it to one segment.
Accordingly, the first row of the timing diagram in
The second row represents the delay that the processing of the decision feedback equalizer 22 imposes on these segment time periods as the corresponding equalized segments exit from the output of the decision feedback equalizer 22 and are provided to the long traceback trellis decoder 14. As shown in
The third row represents the zero delay that the processing of the short traceback trellis decoder 12 imposes on these segment time periods as the corresponding segments of symbol decisions exit from the output of the short traceback trellis decoder 12 and are provided to the least squares channel impulse and noise update estimator 32.
The fourth row represents the additional delay that the processing of the long traceback trellis decoder 14 imposes on these segment time periods as the corresponding segments of symbol decisions exit from the output of the long traceback trellis decoder 14 and are provided to the least squares channel impulse and noise update estimator 32. As shown in
The fifth row represents the additional delays of making the channel impulse response and tap weight calculations by the least squares channel impulse and noise update estimator 32 and the tap weight calculator 34.
As shown in
The symbol decisions c of the short traceback trellis decoder 12 are somewhat less reliable than the symbol decisions b of the long traceback trellis decoder 14. However, when the channel impulse response is changing rapidly, as is the case with mobile receivers, the reduced the tap update latency TUL is a worthwhile tradeoff against the less accurate symbol decisions c.
The long traceback trellis decoder 14 has the capability of outputting a reliable decision after a delay Dmax equal to its maximum traceback depth minus 1. It is well known that the path memories internal to a long traceback trellis decoder simultaneously hold symbol decisions of delay zero up to delay Dmax. These symbol decisions can be output in parallel at any desired time as shown by US published patent application US2002/0154248 A1. This published application describes the use of such parallel outputs to feed decisions back into the feedback filter of a decision feedback equalizer. This operation is effectively parallel loading where, for each symbol update, a new set of decisions with delays of zero up to delay Dmax are simultaneously loaded into the feedback filter.
This concept may be applied by the least squares channel impulse and noise update estimator 32 in determining the updated channel impulse estimate ĥLS. Instead of using a combination of symbol decisions from long traceback trellis decoder 14 and the short traceback trellis decoder 12 as described in relation to
Accordingly, the first row of the timing diagram in
The second row represents the delay that the processing of the decision feedback equalizer 22 imposes on these segment time periods as the corresponding equalized segments exit from the output of the decision feedback equalizer 22 and are provided to the long traceback trellis decoder 14. As shown in
The third row represents the zero to Dmax delay that the processing of the long traceback trellis decoder 14 imposes on these segment time periods as the corresponding segments of symbol decisions are output in parallel from the long traceback trellis decoder 14 and are provided to the least squares channel impulse and noise update estimator 32.
The fourth row represents the additional delay that the processing of the long traceback trellis decoder 14 imposes on these segment time periods as the corresponding segments of symbol decisions exit from the output of the long traceback trellis decoder 14 and are provided to the least squares channel impulse and noise update estimator 32. As shown in
The fifth row represents the additional delays of making the channel impulse response and tap weight calculations by the least squares channel impulse and noise update estimator 32 and the tap weight calculator 34.
As shown in
The parallel symbol decisions from the long traceback trellis decoder 14 (see b′ in
Certain modifications of the present invention have been discussed above. Other modifications of the present invention will occur to those practicing in the art of the present invention. For example, the decoders 12 and 14 may be 12-phase trellis decoders. The use of 12-phase trellis decoders is, for the most part, specific to the digital television application in compliance with the ATSC standard. For other applications, however, decoders other than 12-phase trellis decoders may be used.
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.
This application claims the benefit of U.S. Provisional Application 60/571,447 filed on May 14, 2004.
| Number | Name | Date | Kind |
|---|---|---|---|
| 4833693 | Eyuboglu | May 1989 | A |
| 5056117 | Gitlin et al. | Oct 1991 | A |
| 5406587 | Horwitz et al. | Apr 1995 | A |
| 5453797 | Nicolas et al. | Sep 1995 | A |
| 5550596 | Strolle et al. | Aug 1996 | A |
| 6178209 | Hulyalkar et al. | Jan 2001 | B1 |
| 7218673 | Chang et al. | May 2007 | B2 |
| 20020136329 | Liu et al. | Sep 2002 | A1 |
| 20020154248 | Wittig et al. | Oct 2002 | A1 |
| 20020186762 | Xia et al. | Dec 2002 | A1 |
| 20040091039 | Xia et al. | May 2004 | A1 |
| 20050154967 | Heo et al. | Jul 2005 | A1 |
| Number | Date | Country |
|---|---|---|
| 2 300 726 | Sep 2000 | CA |
| Number | Date | Country | |
|---|---|---|---|
| 20050254570 A1 | Nov 2005 | US |
| Number | Date | Country | |
|---|---|---|---|
| 60571447 | May 2004 | US |
