KR20000010890A - Method for decoding data signals using fixed-length decision window - Google Patents

Abstract

PURPOSE: A method for decoding is provided to decode transmitted data signals using a fixed-length decision window in which multiple transition paths have converged at a single state. CONSTITUTION: A method for decoding a transmitted data sequence using a full-length decision window, such as in a trellis decoder, in which transition paths have converged. Transition paths within the decision window and their associated soft values are compared with a stored candidate for the next best path and associated stored soft value. Based on the comparison, and whether the transition paths and the best path have merged during the decision window, the decision window is shifted or the stored candidate for the next best transition path and associated stored soft value are replaced by the possible candidate for the next best path and associated soft value. The decision window is then shifted.

Description

Data signal decoding method using fixed length decision window

A typical digital communication system is shown in FIG. When the information source 10 supplies analog or digital information, the source encoder 11 encodes the information into an information sequence x. Channel encoder 12 converts information sequence x into a new information sequence by introducing redundancy into the information sequence to improve the reliability of the transmission. Duplicates can be used, for example, for error detection by the receiver, for error correction at the receiver, and also for making automatic repetition requests, where a request for repetitive transmission of data is automatically initiated when an error is detected at the receiver. More redundancy is required to correct errors than to detect errors in encoded messages. The modulator 13 receives the encoded sequence from the channel encoder 12 and generates a channel signal for transmission over the transmission channel 14 using conventional modulation techniques such as amplitude, frequency, phase, pulse modulation. . In general, transmission channel 14 has noise or damage, such as frequency or phase distortion, and exhibits various fading characteristics. Digital demodulator 15 receives the signal transmitted over the transmission channel and demodulates the transmitted signal to produce an estimate of the encoded information sequence. The channel decoder 16 uses the estimate to reproduce the information sequence using the redundant bits provided by the channel encoder 12. Finally, the source decoder 17 transforms the reconstructed information sequence into a form suitable for the information destination 18.

Channel encoder 12 typically introduces redundancy using one of two general methods: block decoding or convolutional decoding. The (n, k) block code converts a block of k information bits into an n-bit codeword by adding n-k parity (redundant) bits according to a predetermined encoding rule. The n-bit codeword is then transmitted over the communication channel. The code rate R is defined as R = k / n. The receiver estimates the original k information bits using a reception sequence that includes redundancy with n-k parity bits introduced. The block code is "memory independence" in that each n-bit codeword output from the decoder depends only on the current k-bit information block.

Convolutional coding is generally implemented by making a stream of information bits into k-bit blocks and passing the blocks to a shift register. The shift register stage stores a group of up to v k-bit blocks, which are connected to a linear algebra function generator. The generator outputs are selectively combined to produce n bits of coded output. Convolutional coding is not memory-independent because each encoded block depends on the previous v message blocks as well as the current k-bit block being input into the shift register. Therefore, the convolutional code has about v memory. The code rate R of the convolution code is R = k / n, and the code length limit is v + 1. Typically, k and n are small integers and redundancy is added by increasing the length of the shift register. The operation of the convolutional encoder and decoder can be described by a grid diagram or state diagram, as known in the art.

Decoders, demodulators, equalizers, and other conventional digital communication system devices typically use a Viterbi algorithm to estimate the transmitted signal encoded using a convolutional code and to minimize communication channel damage. The Viterbi algorithm finds the shortest transition path through a sequence of possible states (“lattices”) to generate the most likely estimate of a single transmitted data sequence. The Viterbi algorithm describes how all possible transmission sequences are correlated with the receiving sequence and then the "survival" sequence is selected based on the best correlation, that is, the path with the best "metric" for estimating the receiving sequence. Can be.

The Viterbi algorithm can be improved by generating a confidence indication for each estimated data bit. The confidence indication may be, for example, the size of the estimated data bits. This improvement is commonly referred to as "soft" information. If concatenated code is used, additional soft information may be used by subsequent decoders to improve the performance of the system. Another variant of the Viterbi algorithm is described in European Patent Application Nos. 0 606 724 to Neill et al. And “List and Soft Symbol Output Viterbi Algorithms: Expansion and Comparison” (NEE Communications Report, Vol. 43). No. 2/3/4, February / March / April 1995). Neil's literature discloses a decoding method that uses soft information and estimated bits to generate a list of candidate sequences in reverse order of likelihood. The list of candidate sequences can be used to determine the best estimate of the transmission sequence at subsequent stages. For example, a cyclic redundancy check (CRC) such as that used in the GMS radio link protocol can be used to check candidate sequences in a list and select a first candidate sequence with correct CRC information. System performance is improved by correcting errors without the need to retransmit erroneous frames.

Neil's method can be generally described as follows. The received symbols are decoded using the Viterbi algorithm to produce estimates and soft information (ie, best predictions) of the most likely data sequences. Soft information is used to find the least likely (weakest) bits in the sequence. Instead of selecting the best incoming transition path (to shift the least significant bit), an alternative incoming transition path is determined through back-tracking in the Viterbi grating. In order to generate a list of candidate sequences ranked in the reverse order of likelihood, each of the l list elements has a candidate for the (l + 1) th list element. The (l + 1) th list component is selected as the candidate with the highest cumulative metric. This step is repeated until a sufficient number of alternatives have been created. Neil's method cannot be used in most practical applications because the entire Viterbi lattice must be preserved.

In the Viterbi algorithm, the grid tends to converge after a certain period of time. In order to use the memory more efficiently, it is desirable to preserve only a portion of the grid (crystal window). It is therefore desirable for practical decoding schemes to provide improved performance over the Viterbi algorithm by assigning estimates of multiple data sequences without having to remember the entire grid.

Summary of the Invention

The present invention overcomes the above-mentioned problems and provides advantages by providing a method of decoding a transmission data signal using a fixed length decision window in which a plurality of transition paths converge to a single state. For each state of the decision window, the possible candidates for the second best sequence and associated soft value are determined and compared with the stored candidates for the second best sequence and associated second best soft value. If the soft value (confidence indication) associated with the possible candidate sequence is less than the second best soft value associated with the stored candidate sequence, and the best candidate sequence and the best sequence generated by the Viterbi algorithm converge in a common state within the decision window, Possible candidates for the second best sequence replace the memorized candidates for the second best sequence. The method according to the invention does not need to preserve the entire grid to complete the decoding operation since the existing soft value and the best transition path are discarded from memory.

The present invention relates generally to the decoding of encoded data communication signals, and more particularly to an improved decoding method using fixed length determination windows.

The invention will be more fully understood from the following description of the preferred embodiments with reference to the accompanying drawings. Like reference numerals in the drawings indicate like elements.

1 is a block diagram of a receiver in a typical communication system in which the method of the present invention may be used;

2 illustrates a crystal window in a Viterbi grating according to an embodiment of the present invention;

3 is a flowchart illustrating a method according to an embodiment of the invention.

2, a fixed-length determination window of length w "+ w 'is shown. All back-tracking sequences proceed through state S. From this state possible candidate sequences will be generated according to the method of the present invention. Can be.

A typical Viterbi algorithm (VA) estimates the probability of the most likely transmission bit sequence. In order to generate a list of possible alternatives to this sequence, according to the invention, the following additional information is generated or preserved at the decoder.

Information on paths excluded from VA is preserved, which is done for all states between t = k and t = k + w 'with reference to FIG. Thus the required memory is w '* N * w "bits. The determination of which path is to be preserved as the second best path (a decision not normally made in VA) is made at t = k + w'. The second best path entering each state in the column of state S at t = k + w 'is stored until the fixed-length decision window column is shifted past t = k.

Reliability measurements, ie "soft" measurements, for each state between t = k and t = k + w 'are preserved. This reliability measure indicates the certainty of the bit shifted out of that state. The reliability measure can be calculated, for example, as the difference in the absolute value of the accumulation metric between two incoming paths leading to the same state at time t = k + w '. One path corresponds to a "1" symbol to be shifted and the other path corresponds to a "0" to be shifted. Each time the decision window is shifted, the next best path and soft value is determined for the new column (rightmost column in FIG. 2).

Threshold MINSOFT is generated, which is used to determine if a possible candidate for an alternative sequence needs to be generated.

To generate an alternative sequence in the decision window, a conventional VA is run until the decision window is filled with the estimated symbols. After the w '+ w "bit time (starting at 0) the decision window can be filled and a possible alternative sequence can be derived. Therefore, in order to generate a possible alternative to the best sequence at that time, shown in bold in FIG. Replace part of the VA estimation sequence with an alternative sequence, with the following conditions:

The bit to be shifted out of state S has a soft value less than the current threshold MINSOFT. In other words, the bit is less reliable than other conventional decoded bits. State S is determined from the best path at time t = k + w '.

The sequence estimated by the VA (shown in FIG. 2) and the alternative sequence must be re-merged in the decision window, for example before t = k-w, to achieve the appropriate "termination" of the alternative sequence.

If this condition is met, a possible alternative sequence is constructed, which is done by selecting another path to state S (shown in dashed lines in FIG. 2) and then back-tracking the grating. The MINSOFT is updated with the soft value of state S, and the existing candidate for the second best sequence is replaced with the newly found sequence.

When a possible alternative sequence is constructed, a new bit decision is made. The decision window is shifted (to the right in FIG. 2) to make room for the newly received bits. The length of the decision window is constant regardless of whether a possible alternative sequence is generated. 2, the decision window now includes a lattice between t = kw "+1 and t = k + w '+ 1. State S can be found in the column t = k + 1. T = k Run the scheme described earlier to check if a sequence can occur between +1 and t = kw "+1. This step is performed until the grating is terminated, i.e. all the bits of the current burst are decoded.

In order to generate two estimates of the transmission sequence, the best estimate is generated by a conventional VA and a second best estimate is generated in accordance with the present invention. To generate a list of several (L) estimated sequences, the algorithm described above can be extended to handle a list of possible candidates for alternative sequences, where the candidate sequences are preserved with their corresponding MINSOFT values. When determining whether to generate a new candidate sequence, the list is checked to see if the soft value of state S is less than any stored MINSOFT value. If it is small, it executes the generation operation and inserts the newly found candidate into the list. Insertion is done so that the list is sorted in increasing MINSOFT values. In other words, the list is preferably sorted in order of least likely. Shift the decision window and repeat the procedure.

3 is a flowchart illustrating a decoding method according to an embodiment of the present invention. This method may be implemented in a decoder, demodulator, equalizer or other similar device to determine the most probable data sequence from the received coded data sequence. The flowchart begins at step 100 where the fixed-length decision window shown in FIG. 2 is updated (shifted), and conventional VA decoding is performed to determine the best transition path. As shown in FIG. 2, the decision window includes a state in which a plurality of transition paths converge to a state at time t = k, and a second time forward from the time t = k by the first time or bit offset w ′. Or extend backward by the bit offset w ".

In step 101, it is determined whether a bit larger than w " has been received at the decoder. There is no need to preserve additional information for the first w " bit because there are no alternative sequences different from the best sequence up to the w " bit. .

In step 102, the next possible best transition path and associated soft information (reliability indication) is determined for each state at the current time t = k + w '. This is done by back-tracking the grid to determine the best sequence and the second best sequence for each state at t = k + w '. The soft value is calculated as the absolute value of (metric 0-metric 1), where metric 0 and metric 1 are accumulated amplitude magnitudes. In step 103, it is determined whether the decision window is filled. If filled, the state S (state found by back-tracking the grating from the state at t = k + w 'with the best confidence indication) is selected.

In step 105, the soft information (reliability indication) SOFT of the state S determined in step 104 is compared with the threshold reliability indication MINSOFT; If the SOFT is not less than MINSOFT, the process returns to step 100 and the fixed-length decision window is shifted to determine a new best state S. If the list of L sequences is estimated, then there is a (L-1) MINSOFT value.

If SOFT is less than MINSOFT, proceed to step 106 where the best path and the second best path merge at some point before the end of the decision window defined by time k, time offset w ', and then the length of the best path w ". If the best possible candidates for the next best path have not been merged (ie, the sequence differs by more than w " bits), the process proceeds to step 100 where the fixed-length decision window is shifted and the new state S is determined. Is determined. If the best possible candidates for the next best path have been merged, the value of MINSOFT is set equal to the value of SOFT, and the existing next best path is discarded from memory and a new next best path of length w " Become.

The above-described method may be implemented in a channel decoder of a GSM system such as the system shown in FIG. 1, for example. This method may be encoded in a suitable storage medium such as a computer disk with machine readable computer code. A typical decision window, for example, has a length of 31 bits and an offset w 'of approximately 14 bits. However, other decision window lengths and offsets w 'can also be used.

In the foregoing description, the terms "path" and "sequence" may be used interchangeably.

While the foregoing description contains many specific details, it is to be understood that this is for illustrative purposes only and is not intended to limit the invention. Those skilled in the art will readily appreciate various modifications without departing from the spirit and scope of the invention as defined by the following claims and their equivalents.

Claims (16)

A method of decoding a sequence of transmitted communication signals having encoded data symbols, the method comprising: For each state in the decision window in the decoding trellis of the decoder, a possible candidate for the next best decoding path and an associated reliability indicator are determined, and each possible candidate for the next best decoding path is stored. Comparing with the best decoding path present; Shifting the decision window if the confidence indicator is not less than a threshold or if the best decoding path and possibly the next best decoding path were not merged during the decision window; And If the reliability indicator is less than the threshold and the best decoding path and the next best decoding path merged during the decision window, define a new threshold as the reliability indicator and define the next possible best path associated with this new threshold for the next best path. Preserving in memory as a new memorized candidate, discarding all other next best path, and shifting the decision window Method comprising a. 2. The method of claim 1, wherein the reliability indicator is based on the amplitude of each decoded symbol in the next best decoding path. The method of claim 1, wherein the crystal window is a fixed-length crystal window. 2. The method of claim 1, wherein each state in the decision window is decoded using a Viterbi algorithm prior to the determining step. 2. The method of claim 1, wherein the reliability indicator is calculated as: metric0-metric1-where metric0 and metric1 are cumulative metric amplitude magnitudes of sequentially decoded symbols. The method of claim 1, wherein said determining step is performed only after a reception threshold of a data symbol is received by a decoder. The method of claim 1, wherein the determining step is executed only after all the symbols in the decision window have been received by the decoder. 2. The method of claim 1, wherein the stored best decoding path is generated by a Viterbi algorithm. A storage medium encoded with machine readable computer code, comprising: Means for determining a possible candidate for the next best decoding path and an associated second best confidence indicator for each state in the decision window of the decoder grid and comparing each possible candidate for the next best decoding path with the stored best decoding path. ; Means for shifting a decision window if the second best confidence indicator is not less than a threshold or if the best decoding path and possibly the next best decoding path were not merged during the decision window; And If the reliability indicator is less than the threshold and the best decoding path and the next best decoding path merged during the decision window, define a new threshold as the reliability indicator and define the next possible best path associated with this new threshold for the next best path. Means for saving in memory as a new memorized candidate, discarding all other next best path, and shifting the decision window And a storage medium. 10. The storage medium of claim 9, wherein the reliability indicator is based on the amplitude of each decoded symbol in the next best decoding path. 10. The storage medium of claim 9, wherein the determination window is a fixed-length determination window. 10. The storage medium of claim 9, wherein each state in the decision window is decoded in advance using a Viterbi algorithm. 10. The storage medium of claim 9, wherein the reliability indicator is calculated as: metric0-metric1, wherein metric0 and metric1 are cumulative metric amplitude magnitudes of sequentially decoded symbols. 10. The storage medium of claim 9, wherein said determining means is executed only after a threshold of data symbols is received by a decoder. 10. The storage medium of claim 9, wherein the determining means is executed only after all the symbols in the decision window have been received by the decoder. 10. The storage medium of claim 9, wherein the stored best decoding path is generated by a Viterbi algorithm.